{"updated_at": "2026-09-10T08:01:36.918266+00:00", "sources": {"original": {"source": "original_nsys", "kernel_events": 954123, "clock_mapping": {"systemClockNs": 2851263799648076, "formula": "trace_ns = round(host perf_counter_seconds * 1e9) - systemClockNs", "validation": "same-host monotonic alignment; windows and per-device activity retained for audit"}, "windows_ns": {"full": null, "measured": [[323197028408, 333168523651]], "steady": [[328882277477, 329745872302], [329745872302, 330592670221], [330592670221, 331431741970], [331431741970, 332290679758]]}, "scope_notes": "full includes initialization/warmup/validation/shutdown. measured is epoch1. steady is CPU-ready intervals5–8. Profile timings are not clean benchmark timings.", "scopes": {"full": {"top_time": [{"name": "nvjet_sm90_tst_128x200_64x5_2x1_v_bz_coopA_bias_TNT", "count": 115618, "total_ms": 13687.7209, "mean_us": 118.38745610545071, "max_us": 246.367}, {"name": "ncclDevKernel_SendRecv(ncclDevKernelArgsStorage<(unsigned long)4096>)", "count": 60891, "total_ms": 9205.16888, "mean_us": 151.17453942290322, "max_us": 2195111.106}, {"name": "void cutlass::device_kernel<flash::enable_sm90_or_later<flash::FlashAttnFwdSm90<flash::CollectiveMainloopFwdSm90<(int)2, cute::tuple<cute::C<(int)1>, cute::C<(int)1>, cute::C<(int)1>>, cute::tuple<cute::C<(int)128>, cute::C<(int)160>, cute::C<(int)128>>, (int)128, cutlass::bfloat16_t, float, cutlass::arch::Sm90, (bool)0, (bool)0, (bool)0, (bool)1, (bool)0, (bool)0, (bool)0, (bool)1, (bool)1, (bool)1, (bool)0, (bool)0, cutlass::bfloat16_t, (int)1>, flash::CollectiveEpilogueFwd<cute::tuple<cute::C<(int)128>, cute::C<(int)128>, cute::C<(int)160>>, cute::tuple<cute::C<(int)1>, cute::C<(int)1>, cute::C<(int)1>>, cutlass::bfloat16_t, cutlass::arch::Sm90, (int)256, (bool)1, (bool)1, (bool)0, (bool)0, (int)1>, flash::VarlenDynamicPersistentTileScheduler<(int)128, (int)160, (int)256, (int)128, (bool)0, (bool)1, (bool)1, (bool)0, (bool)0, (bool)1>>>>(T1::Params)", "count": 12848, "total_ms": 8782.203928, "mean_us": 683.5463829389788, "max_us": 1085.252}, {"name": "void at::native::elementwise_kernel<(int)128, (int)2, void at::native::gpu_kernel_impl_nocast<at::native::direct_copy_kernel_cuda(at::TensorIteratorBase &)::[lambda() (instance 3)]::operator ()() const::[lambda() (instance 7)]::operator ()() const::[lambda(float) (instance 1)]>(at::TensorIteratorBase &, const T1 &)::[lambda(int) (instance 1)]>(int, T3)", "count": 104492, "total_ms": 4873.156834, "mean_us": 46.636650021054244, "max_us": 746.336}, {"name": "nvjet_sm90_tst_256x152_64x4_1x2_h_bz_coopA_bias_TNT", "count": 12848, "total_ms": 3419.719443, "mean_us": 266.16745353362387, "max_us": 289.953}, {"name": "sm90_xmma_fprop_implicit_gemm_f32f32_tf32f32_f32_nhwckrsc_nhwc_tilesize256x32x32_warpgroupsize1x1x1_g1_execute_segment_k_off_kernel__5x_cudnn", "count": 1800, "total_ms": 2604.967243, "mean_us": 1447.204023888889, "max_us": 5629.347}, {"name": "sm80_xmma_fprop_implicit_gemm_bf16bf16_bf16f32_f32_nhwckrsc_nhwc_tilesize256x32x32_stage4_warpsize4x1x1_g1_tensor16x8x16_execute_kernel__5x_cudnn", "count": 960, "total_ms": 2528.289153, "mean_us": 2633.634534375, "max_us": 2741.478}, {"name": "sm90_xmma_fprop_implicit_gemm_f32f32_tf32f32_f32_nhwckrsc_nhwc_tilesize128x128x32_warpgroupsize1x1x1_g1_execute_segment_k_off_kernel__5x_cudnn", "count": 4059, "total_ms": 2073.42463, "mean_us": 510.82153978812516, "max_us": 1829.089}, {"name": "sm90_xmma_fprop_implicit_gemm_bf16bf16_bf16f32_f32_nhwckrsc_nhwc_tilesize256x128x64_warpgroupsize2x1x1_g1_execute_segment_k_off_kernel__5x_cudnn", "count": 3840, "total_ms": 1556.632865, "mean_us": 405.37314192708334, "max_us": 1255.844}, {"name": "void at::native::elementwise_kernel<(int)128, (int)4, void at::native::gpu_kernel_impl<at::native::BinaryFunctor<float, float, float, at::native::binary_internal::DivFunctor<float>>>(at::TensorIteratorBase &, const T1 &)::[lambda(int) (instance 1)]>(int, T3)", "count": 5280, "total_ms": 1057.238829, "mean_us": 200.23462670454546, "max_us": 723.522}, {"name": "void at::native::vectorized_gather_kernel<(int)16, long>(char *, char *, T2 *, int, long, long, long, long, bool)", "count": 25896, "total_ms": 918.162639, "mean_us": 35.455770736793326, "max_us": 159.808}, {"name": "void at::native::elementwise_kernel<(int)128, (int)4, void at::native::gpu_kernel_impl_nocast<at::native::CUDAFunctor_add<c10::BFloat16>>(at::TensorIteratorBase &, const T1 &)::[lambda(int) (instance 1)]>(int, T3)", "count": 7824, "total_ms": 903.7869, "mean_us": 115.51468558282208, "max_us": 420.192}, {"name": "void at::native::elementwise_kernel<(int)128, (int)4, void at::native::gpu_kernel_impl<at::native::BinaryFunctor<float, float, float, at::native::binary_internal::MulFunctor<float>>>(at::TensorIteratorBase &, const T1 &)::[lambda(int) (instance 1)]>(int, T3)", "count": 5672, "total_ms": 885.325854, "mean_us": 156.08706875881524, "max_us": 586.818}, {"name": "void cudnn::engines_precompiled::nchwToNhwcKernel<__nv_bfloat16, __nv_bfloat16, float, (bool)0, (bool)1, (cudnnKernelDataType_t)0>(cudnn::engines_precompiled::nchw2nhwc_params_t<T3>, const T1 *, T2 *)", "count": 12448, "total_ms": 825.883203, "mean_us": 66.34665833868894, "max_us": 435.041}, {"name": "void at::native::index_elementwise_kernel<(int)128, (int)4, void at::native::gpu_index_kernel<void at::native::index_put_kernel_impl<at::native::OpaqueType<(int)2>>(at::TensorIterator &, c10::ArrayRef<long>, c10::ArrayRef<long>)::[lambda(char *, const char *, long) (instance 1)]>(at::TensorIteratorBase &, c10::ArrayRef<long>, c10::ArrayRef<long>, const T1 &, bool)::[lambda(int) (instance 1)]>(long, T3)", "count": 25696, "total_ms": 760.071163, "mean_us": 29.57935721513076, "max_us": 33.343}, {"name": "void at::native::vectorized_elementwise_kernel<(int)4, at::native::<unnamed>::silu_kernel(at::TensorIteratorBase &)::[lambda() (instance 1)]::operator ()() const::[lambda() (instance 2)]::operator ()() const::[lambda(float) (instance 1)], std::array<char *, (unsigned long)2>>(int, T2, T3)", "count": 13740, "total_ms": 687.111505, "mean_us": 50.008115356623, "max_us": 290.368}, {"name": "void at::native::vectorized_elementwise_kernel<(int)4, at::native::AUnaryFunctor<float, float, float, at::native::binary_internal::MulFunctor<float>>, std::array<char *, (unsigned long)2>>(int, T2, T3)", "count": 15395, "total_ms": 684.614974, "mean_us": 44.46995608963949, "max_us": 288.193}, {"name": "void cudnn::engines_precompiled::nchwToNhwcKernel<float, float, float, (bool)0, (bool)1, (cudnnKernelDataType_t)0>(cudnn::engines_precompiled::nchw2nhwc_params_t<T3>, const T1 *, T2 *)", "count": 19748, "total_ms": 679.247489, "mean_us": 34.39576103909257, "max_us": 735.297}, {"name": "fmha_cutlassF_f32_aligned_32x128_gmem_sm80(PyTorchMemEffAttention::AttentionKernel<float, cutlass::arch::Sm80, (bool)1, (int)32, (int)128, (int)65536, (bool)1, (bool)1>::Params)", "count": 300, "total_ms": 576.298234, "mean_us": 1920.9941133333332, "max_us": 2235.366}, {"name": "void at::native::elementwise_kernel<(int)128, (int)2, void at::native::gpu_kernel_impl_nocast<at::native::CUDAFunctor_add<float>>(at::TensorIteratorBase &, const T1 &)::[lambda(int) (instance 1)]>(int, T3)", "count": 12561, "total_ms": 555.427831, "mean_us": 44.21844049040681, "max_us": 568.704}], "top_count": [{"name": "nvjet_sm90_tst_128x200_64x5_2x1_v_bz_coopA_bias_TNT", "count": 115618, "total_ms": 13687.7209, "mean_us": 118.38745610545071, "max_us": 246.367}, {"name": "void at::native::elementwise_kernel<(int)128, (int)2, void at::native::gpu_kernel_impl_nocast<at::native::direct_copy_kernel_cuda(at::TensorIteratorBase &)::[lambda() (instance 3)]::operator ()() const::[lambda() (instance 7)]::operator ()() const::[lambda(float) (instance 1)]>(at::TensorIteratorBase &, const T1 &)::[lambda(int) (instance 1)]>(int, T3)", "count": 104492, "total_ms": 4873.156834, "mean_us": 46.636650021054244, "max_us": 746.336}, {"name": "ncclDevKernel_SendRecv(ncclDevKernelArgsStorage<(unsigned long)4096>)", "count": 60891, "total_ms": 9205.16888, "mean_us": 151.17453942290322, "max_us": 2195111.106}, {"name": "void at::native::vectorized_elementwise_kernel<(int)4, at::native::FillFunctor<int>, std::array<char *, (unsigned long)1>>(int, T2, T3)", "count": 38595, "total_ms": 383.132614, "mean_us": 9.927001269594507, "max_us": 15.808}, {"name": "triton_poi_fused_all_to_all_single_clone_transpose_unsqueeze_view_5", "count": 26541, "total_ms": 180.682964, "mean_us": 6.80769240043706, "max_us": 20.352}, {"name": "triton_poi_fused_clone_transpose_view_8", "count": 26540, "total_ms": 166.223276, "mean_us": 6.263122682743029, "max_us": 12.832}, {"name": "void at::native::vectorized_gather_kernel<(int)16, long>(char *, char *, T2 *, int, long, long, long, long, bool)", "count": 25896, "total_ms": 918.162639, "mean_us": 35.455770736793326, "max_us": 159.808}, {"name": "void at::native::index_elementwise_kernel<(int)128, (int)4, void at::native::gpu_index_kernel<void at::native::index_put_kernel_impl<at::native::OpaqueType<(int)2>>(at::TensorIterator &, c10::ArrayRef<long>, c10::ArrayRef<long>)::[lambda(char *, const char *, long) (instance 1)]>(at::TensorIteratorBase &, c10::ArrayRef<long>, c10::ArrayRef<long>, const T1 &, bool)::[lambda(int) (instance 1)]>(long, T3)", "count": 25696, "total_ms": 760.071163, "mean_us": 29.57935721513076, "max_us": 33.343}, {"name": "rotary_kernel", "count": 25696, "total_ms": 247.216284, "mean_us": 9.62080806351183, "max_us": 11.52}, {"name": "void at::native::vectorized_elementwise_kernel<(int)4, at::native::FillFunctor<float>, std::array<char *, (unsigned long)1>>(int, T2, T3)", "count": 20469, "total_ms": 541.747607, "mean_us": 26.466735404758417, "max_us": 203.552}, {"name": "void cudnn::engines_precompiled::nchwToNhwcKernel<float, float, float, (bool)0, (bool)1, (cudnnKernelDataType_t)0>(cudnn::engines_precompiled::nchw2nhwc_params_t<T3>, const T1 *, T2 *)", "count": 19748, "total_ms": 679.247489, "mean_us": 34.39576103909257, "max_us": 735.297}, {"name": "void at::native::vectorized_elementwise_kernel<(int)4, at::native::AUnaryFunctor<float, float, float, at::native::binary_internal::MulFunctor<float>>, std::array<char *, (unsigned long)2>>(int, T2, T3)", "count": 15395, "total_ms": 684.614974, "mean_us": 44.46995608963949, "max_us": 288.193}, {"name": "triton_red_fused__to_copy_add_mul_native_layer_norm_split_squeeze_1", "count": 15074, "total_ms": 343.237799, "mean_us": 22.77018701074698, "max_us": 44.032}, {"name": "triton_red_fused__to_copy_add_mul_native_layer_norm_split_squeeze_view_9", "count": 14540, "total_ms": 352.541204, "mean_us": 24.246300137551582, "max_us": 31.04}, {"name": "void at::native::vectorized_elementwise_kernel<(int)4, at::native::<unnamed>::launch_clamp_scalar(at::TensorIteratorBase &, c10::Scalar, c10::Scalar, at::native::detail::ClampLimits)::[lambda() (instance 1)]::operator ()() const::[lambda() (instance 7)]::operator ()() const::[lambda(float) (instance 1)], std::array<char *, (unsigned long)2>>(int, T2, T3)", "count": 14322, "total_ms": 24.681315, "mean_us": 1.7233148303309596, "max_us": 100.896}, {"name": "triton_red_fused__to_copy_div_mul_pow_split_with_sizes_sqrt_sum_view_3", "count": 14120, "total_ms": 147.338168, "mean_us": 10.434714447592068, "max_us": 12.448}, {"name": "triton_red_fused__to_copy_div_mul_pow_split_with_sizes_sqrt_sum_view_2", "count": 14120, "total_ms": 134.163305, "mean_us": 9.501650495750708, "max_us": 11.168}, {"name": "void at::native::vectorized_elementwise_kernel<(int)4, at::native::<unnamed>::silu_kernel(at::TensorIteratorBase &)::[lambda() (instance 1)]::operator ()() const::[lambda() (instance 2)]::operator ()() const::[lambda(float) (instance 1)], std::array<char *, (unsigned long)2>>(int, T2, T3)", "count": 13740, "total_ms": 687.111505, "mean_us": 50.008115356623, "max_us": 290.368}, {"name": "triton_poi_fused_all_to_all_single_clone_split_with_sizes_stack_transpose_view_4", "count": 13696, "total_ms": 261.196118, "mean_us": 19.07097824182243, "max_us": 39.041}, {"name": "triton_poi_fused_add_view_7", "count": 13696, "total_ms": 141.021316, "mean_us": 10.29653300233645, "max_us": 13.023}], "groups": 167, "devices": [{"device": 4, "window_ms": 191970.226833, "non_nccl_ms": 16029.666051, "nccl_ms": 4646.488953, "memcpy_ms": 7705.775133, "non_nccl_nccl_overlap_ms": 336.142936, "kernel_memcpy_overlap_ms": 0.0, "observed_busy_ms": 28045.787201, "no_observed_kernel_or_memcpy_ms": 163924.439632}, {"device": 5, "window_ms": 191970.226833, "non_nccl_ms": 14383.528842, "nccl_ms": 1408.482742, "memcpy_ms": 6784.770732, "non_nccl_nccl_overlap_ms": 245.610209, "kernel_memcpy_overlap_ms": 0.0, "observed_busy_ms": 22331.172107, "no_observed_kernel_or_memcpy_ms": 169639.054726}, {"device": 6, "window_ms": 191970.226833, "non_nccl_ms": 14145.436708, "nccl_ms": 1790.571562, "memcpy_ms": 7100.500687, "non_nccl_nccl_overlap_ms": 363.108641, "kernel_memcpy_overlap_ms": 0.0, "observed_busy_ms": 22673.400316, "no_observed_kernel_or_memcpy_ms": 169296.826517}, {"device": 7, "window_ms": 191970.226833, "non_nccl_ms": 14315.585839, "nccl_ms": 1582.24157, "memcpy_ms": 6757.855891, "non_nccl_nccl_overlap_ms": 259.41463, "kernel_memcpy_overlap_ms": 0.0, "observed_busy_ms": 22396.26867, "no_observed_kernel_or_memcpy_ms": 169573.958163}], "copies": [{"direction": "Host-to-Device", "count": 32608, "summed_gpu_ms": 26867.101387, "bytes_full_overlapping_calls": 197522071900}, {"direction": "Device-to-Device", "count": 160803, "summed_gpu_ms": 845.80593, "bytes_full_overlapping_calls": 1452686357580}, {"direction": "Device-to-Host", "count": 648, "summed_gpu_ms": 635.995126, "bytes_full_overlapping_calls": 1985586682}], "wait_apis": [{"api": "cudaStreamSynchronize_v3020", "kind": "sync", "count": 12382, "summed_thread_ms": 22041.741648}, {"api": "cudaDeviceSynchronize_v3020", "kind": "sync", "count": 1701, "summed_thread_ms": 13849.004869}, {"api": "cudaStreamWaitEvent_v3020", "kind": "sync", "count": 66272, "summed_thread_ms": 80.561524}, {"api": "pthread_cond_timedwait", "kind": "os_wait", "count": 393488, "summed_thread_ms": 11708041.761058}, {"api": "epoll_wait", "kind": "os_wait", "count": 596470, "summed_thread_ms": 6442484.894235}, {"api": "poll", "kind": "os_wait", "count": 115505, "summed_thread_ms": 5976805.631145}, {"api": "pthread_cond_wait", "kind": "os_wait", "count": 779, "summed_thread_ms": 4799374.621223}, {"api": "futex", "kind": "os_wait", "count": 84, "summed_thread_ms": 2618409.711788}, {"api": "sem_clockwait", "kind": "os_wait", "count": 1255, "summed_thread_ms": 1171173.692554}, {"api": "sem_wait", "kind": "os_wait", "count": 1012, "summed_thread_ms": 994592.252525}, {"api": "epoll_pwait", "kind": "os_wait", "count": 1489, "summed_thread_ms": 327931.41165}, {"api": "waitpid", "kind": "os_wait", "count": 1126, "summed_thread_ms": 238430.743904}, {"api": "clock_nanosleep", "kind": "os_wait", "count": 580, "summed_thread_ms": 19552.561305}, {"api": "nanosleep", "kind": "os_wait", "count": 320, "summed_thread_ms": 3424.689692}, {"api": "sleep", "kind": "os_wait", "count": 6300, "summed_thread_ms": 796.474967}, {"api": "sem_timedwait", "kind": "os_wait", "count": 88, "summed_thread_ms": 3.077466}, {"api": "epoll_ctl", "kind": "os_wait", "count": 234, "summed_thread_ms": 0.449025}], "overlap_interpretation": "Same-device overlapping activity intervals, not proof of simultaneous SM execution or net speedup. Non-NCCL includes copies/conversions implemented as kernels. Gaps are not attributed CPU waits. Thread times overlap and must not be added as wall time."}, "measured": {"top_time": [{"name": "nvjet_sm90_tst_128x200_64x5_2x1_v_bz_coopA_bias_TNT", "count": 43618, "total_ms": 5162.769672, "mean_us": 118.36328286487229, "max_us": 246.367}, {"name": "void cutlass::device_kernel<flash::enable_sm90_or_later<flash::FlashAttnFwdSm90<flash::CollectiveMainloopFwdSm90<(int)2, cute::tuple<cute::C<(int)1>, cute::C<(int)1>, cute::C<(int)1>>, cute::tuple<cute::C<(int)128>, cute::C<(int)160>, cute::C<(int)128>>, (int)128, cutlass::bfloat16_t, float, cutlass::arch::Sm90, (bool)0, (bool)0, (bool)0, (bool)1, (bool)0, (bool)0, (bool)0, (bool)1, (bool)1, (bool)1, (bool)0, (bool)0, cutlass::bfloat16_t, (int)1>, flash::CollectiveEpilogueFwd<cute::tuple<cute::C<(int)128>, cute::C<(int)128>, cute::C<(int)160>>, cute::tuple<cute::C<(int)1>, cute::C<(int)1>, cute::C<(int)1>>, cutlass::bfloat16_t, cutlass::arch::Sm90, (int)256, (bool)1, (bool)1, (bool)0, (bool)0, (int)1>, flash::VarlenDynamicPersistentTileScheduler<(int)128, (int)160, (int)256, (int)128, (bool)0, (bool)1, (bool)1, (bool)0, (bool)0, (bool)1>>>>(T1::Params)", "count": 4848, "total_ms": 2843.140252, "mean_us": 586.4563226072607, "max_us": 1085.252}, {"name": "ncclDevKernel_SendRecv(ncclDevKernelArgsStorage<(unsigned long)4096>)", "count": 23347, "total_ms": 2168.249254, "mean_us": 92.87057240758985, "max_us": 63937.036}, {"name": "void at::native::elementwise_kernel<(int)128, (int)2, void at::native::gpu_kernel_impl_nocast<at::native::direct_copy_kernel_cuda(at::TensorIteratorBase &)::[lambda() (instance 3)]::operator ()() const::[lambda() (instance 7)]::operator ()() const::[lambda(float) (instance 1)]>(at::TensorIteratorBase &, const T1 &)::[lambda(int) (instance 1)]>(int, T3)", "count": 44685, "total_ms": 2156.448784, "mean_us": 48.25889636343292, "max_us": 703.458}, {"name": "nvjet_sm90_tst_256x152_64x4_1x2_h_bz_coopA_bias_TNT", "count": 4848, "total_ms": 1290.937434, "mean_us": 266.28247400990097, "max_us": 289.953}, {"name": "sm80_xmma_fprop_implicit_gemm_bf16bf16_bf16f32_f32_nhwckrsc_nhwc_tilesize256x32x32_stage4_warpsize4x1x1_g1_tensor16x8x16_execute_kernel__5x_cudnn", "count": 480, "total_ms": 1263.74789, "mean_us": 2632.8081041666665, "max_us": 2735.848}, {"name": "sm90_xmma_fprop_implicit_gemm_f32f32_tf32f32_f32_nhwckrsc_nhwc_tilesize256x32x32_warpgroupsize1x1x1_g1_execute_segment_k_off_kernel__5x_cudnn", "count": 720, "total_ms": 935.329666, "mean_us": 1299.0689805555555, "max_us": 1365.637}, {"name": "sm90_xmma_fprop_implicit_gemm_f32f32_tf32f32_f32_nhwckrsc_nhwc_tilesize128x128x32_warpgroupsize1x1x1_g1_execute_segment_k_off_kernel__5x_cudnn", "count": 1652, "total_ms": 810.684405, "mean_us": 490.7290587167071, "max_us": 894.049}, {"name": "sm90_xmma_fprop_implicit_gemm_bf16bf16_bf16f32_f32_nhwckrsc_nhwc_tilesize256x128x64_warpgroupsize2x1x1_g1_execute_segment_k_off_kernel__5x_cudnn", "count": 1920, "total_ms": 778.853069, "mean_us": 405.65264010416666, "max_us": 1255.844}, {"name": "void at::native::elementwise_kernel<(int)128, (int)4, void at::native::gpu_kernel_impl<at::native::BinaryFunctor<float, float, float, at::native::binary_internal::DivFunctor<float>>>(at::TensorIteratorBase &, const T1 &)::[lambda(int) (instance 1)]>(int, T3)", "count": 2640, "total_ms": 528.573897, "mean_us": 200.21738522727273, "max_us": 723.522}, {"name": "void at::native::elementwise_kernel<(int)128, (int)4, void at::native::gpu_kernel_impl_nocast<at::native::CUDAFunctor_add<c10::BFloat16>>(at::TensorIteratorBase &, const T1 &)::[lambda(int) (instance 1)]>(int, T3)", "count": 3912, "total_ms": 451.897858, "mean_us": 115.51581237218814, "max_us": 420.129}, {"name": "void at::native::elementwise_kernel<(int)128, (int)4, void at::native::gpu_kernel_impl<at::native::BinaryFunctor<float, float, float, at::native::binary_internal::MulFunctor<float>>>(at::TensorIteratorBase &, const T1 &)::[lambda(int) (instance 1)]>(int, T3)", "count": 2836, "total_ms": 442.665354, "mean_us": 156.08792454160792, "max_us": 586.21}, {"name": "void cudnn::engines_precompiled::nchwToNhwcKernel<__nv_bfloat16, __nv_bfloat16, float, (bool)0, (bool)1, (cudnnKernelDataType_t)0>(cudnn::engines_precompiled::nchw2nhwc_params_t<T3>, const T1 *, T2 *)", "count": 6224, "total_ms": 413.032835, "mean_us": 66.3613166773779, "max_us": 434.848}, {"name": "void at::native::vectorized_gather_kernel<(int)16, long>(char *, char *, T2 *, int, long, long, long, long, bool)", "count": 9796, "total_ms": 337.374465, "mean_us": 34.440022968558594, "max_us": 159.296}, {"name": "void at::native::vectorized_elementwise_kernel<(int)4, at::native::<unnamed>::silu_kernel(at::TensorIteratorBase &)::[lambda() (instance 1)]::operator ()() const::[lambda() (instance 2)]::operator ()() const::[lambda(float) (instance 1)], std::array<char *, (unsigned long)2>>(int, T2, T3)", "count": 6000, "total_ms": 307.613585, "mean_us": 51.26893083333333, "max_us": 288.257}, {"name": "void at::native::vectorized_elementwise_kernel<(int)4, at::native::AUnaryFunctor<float, float, float, at::native::binary_internal::MulFunctor<float>>, std::array<char *, (unsigned long)2>>(int, T2, T3)", "count": 6689, "total_ms": 306.940946, "mean_us": 45.88741904619525, "max_us": 288.193}, {"name": "void at::native::index_elementwise_kernel<(int)128, (int)4, void at::native::gpu_index_kernel<void at::native::index_put_kernel_impl<at::native::OpaqueType<(int)2>>(at::TensorIterator &, c10::ArrayRef<long>, c10::ArrayRef<long>)::[lambda(char *, const char *, long) (instance 1)]>(at::TensorIteratorBase &, c10::ArrayRef<long>, c10::ArrayRef<long>, const T1 &, bool)::[lambda(int) (instance 1)]>(long, T3)", "count": 9696, "total_ms": 287.102668, "mean_us": 29.61042367986799, "max_us": 33.343}, {"name": "void at::native::vectorized_elementwise_kernel<(int)8, at::native::bfloat16_copy_kernel_cuda(at::TensorIteratorBase &)::[lambda(float) (instance 1)], std::array<char *, (unsigned long)2>>(int, T2, T3)", "count": 3605, "total_ms": 262.253387, "mean_us": 72.7471253814147, "max_us": 319.201}, {"name": "void cudnn::engines_precompiled::nchwToNhwcKernel<float, float, float, (bool)0, (bool)1, (cudnnKernelDataType_t)0>(cudnn::engines_precompiled::nchw2nhwc_params_t<T3>, const T1 *, T2 *)", "count": 7904, "total_ms": 249.177577, "mean_us": 31.525503162955467, "max_us": 191.425}, {"name": "void at::native::vectorized_elementwise_kernel<(int)4, at::native::FillFunctor<float>, std::array<char *, (unsigned long)1>>(int, T2, T3)", "count": 8778, "total_ms": 241.4681, "mean_us": 27.508327637275006, "max_us": 203.488}], "top_count": [{"name": "void at::native::elementwise_kernel<(int)128, (int)2, void at::native::gpu_kernel_impl_nocast<at::native::direct_copy_kernel_cuda(at::TensorIteratorBase &)::[lambda() (instance 3)]::operator ()() const::[lambda() (instance 7)]::operator ()() const::[lambda(float) (instance 1)]>(at::TensorIteratorBase &, const T1 &)::[lambda(int) (instance 1)]>(int, T3)", "count": 44685, "total_ms": 2156.448784, "mean_us": 48.25889636343292, "max_us": 703.458}, {"name": "nvjet_sm90_tst_128x200_64x5_2x1_v_bz_coopA_bias_TNT", "count": 43618, "total_ms": 5162.769672, "mean_us": 118.36328286487229, "max_us": 246.367}, {"name": "ncclDevKernel_SendRecv(ncclDevKernelArgsStorage<(unsigned long)4096>)", "count": 23347, "total_ms": 2168.249254, "mean_us": 92.87057240758985, "max_us": 63937.036}, {"name": "void at::native::vectorized_gather_kernel<(int)16, long>(char *, char *, T2 *, int, long, long, long, long, bool)", "count": 9796, "total_ms": 337.374465, "mean_us": 34.440022968558594, "max_us": 159.296}, {"name": "void at::native::index_elementwise_kernel<(int)128, (int)4, void at::native::gpu_index_kernel<void at::native::index_put_kernel_impl<at::native::OpaqueType<(int)2>>(at::TensorIterator &, c10::ArrayRef<long>, c10::ArrayRef<long>)::[lambda(char *, const char *, long) (instance 1)]>(at::TensorIteratorBase &, c10::ArrayRef<long>, c10::ArrayRef<long>, const T1 &, bool)::[lambda(int) (instance 1)]>(long, T3)", "count": 9696, "total_ms": 287.102668, "mean_us": 29.61042367986799, "max_us": 33.343}, {"name": "rotary_kernel", "count": 9696, "total_ms": 93.406791, "mean_us": 9.633538675742574, "max_us": 11.52}, {"name": "triton_poi_fused_all_to_all_single_clone_transpose_unsqueeze_view_5", "count": 9693, "total_ms": 63.858008, "mean_us": 6.588054059630661, "max_us": 7.616}, {"name": "triton_poi_fused_clone_transpose_view_8", "count": 9692, "total_ms": 59.608564, "mean_us": 6.150285183656624, "max_us": 8.96}, {"name": "void at::native::vectorized_elementwise_kernel<(int)4, at::native::FillFunctor<float>, std::array<char *, (unsigned long)1>>(int, T2, T3)", "count": 8778, "total_ms": 241.4681, "mean_us": 27.508327637275006, "max_us": 203.488}, {"name": "void cudnn::engines_precompiled::nchwToNhwcKernel<float, float, float, (bool)0, (bool)1, (cudnnKernelDataType_t)0>(cudnn::engines_precompiled::nchw2nhwc_params_t<T3>, const T1 *, T2 *)", "count": 7904, "total_ms": 249.177577, "mean_us": 31.525503162955467, "max_us": 191.425}, {"name": "void at::native::vectorized_elementwise_kernel<(int)4, at::native::AUnaryFunctor<float, float, float, at::native::binary_internal::MulFunctor<float>>, std::array<char *, (unsigned long)2>>(int, T2, T3)", "count": 6689, "total_ms": 306.940946, "mean_us": 45.88741904619525, "max_us": 288.193}, {"name": "void at::native::vectorized_elementwise_kernel<(int)4, at::native::<unnamed>::launch_clamp_scalar(at::TensorIteratorBase &, c10::Scalar, c10::Scalar, at::native::detail::ClampLimits)::[lambda() (instance 1)]::operator ()() const::[lambda() (instance 7)]::operator ()() const::[lambda(float) (instance 1)], std::array<char *, (unsigned long)2>>(int, T2, T3)", "count": 6260, "total_ms": 10.841775, "mean_us": 1.7319129392971246, "max_us": 30.304}, {"name": "void cudnn::engines_precompiled::nchwToNhwcKernel<__nv_bfloat16, __nv_bfloat16, float, (bool)0, (bool)1, (cudnnKernelDataType_t)0>(cudnn::engines_precompiled::nchw2nhwc_params_t<T3>, const T1 *, T2 *)", "count": 6224, "total_ms": 413.032835, "mean_us": 66.3613166773779, "max_us": 434.848}, {"name": "void at::native::vectorized_elementwise_kernel<(int)4, at::native::<unnamed>::silu_kernel(at::TensorIteratorBase &)::[lambda() (instance 1)]::operator ()() const::[lambda() (instance 2)]::operator ()() const::[lambda(float) (instance 1)], std::array<char *, (unsigned long)2>>(int, T2, T3)", "count": 6000, "total_ms": 307.613585, "mean_us": 51.26893083333333, "max_us": 288.257}, {"name": "void at::native::unrolled_elementwise_kernel<at::native::direct_copy_kernel_cuda(at::TensorIteratorBase &)::[lambda() (instance 3)]::operator ()() const::[lambda() (instance 4)]::operator ()() const::[lambda(long) (instance 1)], std::array<char *, (unsigned long)2>, (int)4, TrivialOffsetCalculator<(int)1, unsigned int>, TrivialOffsetCalculator<(int)1, unsigned int>, at::native::memory::LoadWithCast<(int)1>, at::native::memory::StoreWithCast<(int)1>>(int, T1, T2, T4, T5, T6, T7)", "count": 5040, "total_ms": 7.914498, "mean_us": 1.570336904761905, "max_us": 3.456}, {"name": "void at::native::elementwise_kernel<(int)128, (int)2, void at::native::gpu_kernel_impl_nocast<at::native::CUDAFunctor_add<float>>(at::TensorIteratorBase &, const T1 &)::[lambda(int) (instance 1)]>(int, T3)", "count": 5032, "total_ms": 202.073882, "mean_us": 40.15776669316375, "max_us": 142.529}, {"name": "void cutlass::device_kernel<flash::enable_sm90_or_later<flash::FlashAttnFwdSm90<flash::CollectiveMainloopFwdSm90<(int)2, cute::tuple<cute::C<(int)1>, cute::C<(int)1>, cute::C<(int)1>>, cute::tuple<cute::C<(int)128>, cute::C<(int)160>, cute::C<(int)128>>, (int)128, cutlass::bfloat16_t, float, cutlass::arch::Sm90, (bool)0, (bool)0, (bool)0, (bool)1, (bool)0, (bool)0, (bool)0, (bool)1, (bool)1, (bool)1, (bool)0, (bool)0, cutlass::bfloat16_t, (int)1>, flash::CollectiveEpilogueFwd<cute::tuple<cute::C<(int)128>, cute::C<(int)128>, cute::C<(int)160>>, cute::tuple<cute::C<(int)1>, cute::C<(int)1>, cute::C<(int)1>>, cutlass::bfloat16_t, cutlass::arch::Sm90, (int)256, (bool)1, (bool)1, (bool)0, (bool)0, (int)1>, flash::VarlenDynamicPersistentTileScheduler<(int)128, (int)160, (int)256, (int)128, (bool)0, (bool)1, (bool)1, (bool)0, (bool)0, (bool)1>>>>(T1::Params)", "count": 4848, "total_ms": 2843.140252, "mean_us": 586.4563226072607, "max_us": 1085.252}, {"name": "nvjet_sm90_tst_256x152_64x4_1x2_h_bz_coopA_bias_TNT", "count": 4848, "total_ms": 1290.937434, "mean_us": 266.28247400990097, "max_us": 289.953}, {"name": "triton_red_fused__to_copy_add_mul_native_layer_norm_split_squeeze_1", "count": 4848, "total_ms": 107.045822, "mean_us": 22.08040882838284, "max_us": 26.304}, {"name": "triton_poi_fused_all_to_all_single_clone_split_with_sizes_stack_transpose_view_4", "count": 4848, "total_ms": 89.539498, "mean_us": 18.469368399339935, "max_us": 21.312}], "groups": 157, "devices": [{"device": 4, "window_ms": 9971.495243, "non_nccl_ms": 5940.22998, "nccl_ms": 650.12565, "memcpy_ms": 102.50232, "non_nccl_nccl_overlap_ms": 134.402546, "kernel_memcpy_overlap_ms": 0.0, "observed_busy_ms": 6558.455404, "no_observed_kernel_or_memcpy_ms": 3413.039839}, {"device": 5, "window_ms": 9971.495243, "non_nccl_ms": 5769.624493, "nccl_ms": 464.543866, "memcpy_ms": 79.664989, "non_nccl_nccl_overlap_ms": 90.596321, "kernel_memcpy_overlap_ms": 0.0, "observed_busy_ms": 6223.237027, "no_observed_kernel_or_memcpy_ms": 3748.258216}, {"device": 6, "window_ms": 9971.495243, "non_nccl_ms": 5668.634213, "nccl_ms": 607.090124, "memcpy_ms": 79.190963, "non_nccl_nccl_overlap_ms": 129.76956, "kernel_memcpy_overlap_ms": 0.0, "observed_busy_ms": 6225.14574, "no_observed_kernel_or_memcpy_ms": 3746.349503}, {"device": 7, "window_ms": 9971.495243, "non_nccl_ms": 5729.642279, "nccl_ms": 514.082193, "memcpy_ms": 79.625981, "non_nccl_nccl_overlap_ms": 95.488343, "kernel_memcpy_overlap_ms": 0.0, "observed_busy_ms": 6227.86211, "no_observed_kernel_or_memcpy_ms": 3743.633133}], "copies": [{"direction": "Host-to-Device", "count": 5546, "summed_gpu_ms": 11.201998, "bytes_full_overlapping_calls": 71829872}, {"direction": "Device-to-Device", "count": 60050, "summed_gpu_ms": 311.217889, "bytes_full_overlapping_calls": 539563182976}, {"direction": "Device-to-Host", "count": 80, "summed_gpu_ms": 18.564366, "bytes_full_overlapping_calls": 152156220}], "wait_apis": [{"api": "cudaStreamSynchronize_v3020", "kind": "sync", "count": 746, "summed_thread_ms": 8549.979875}, {"api": "cudaDeviceSynchronize_v3020", "kind": "sync", "count": 101, "summed_thread_ms": 4024.924676}, {"api": "cudaStreamWaitEvent_v3020", "kind": "sync", "count": 24646, "summed_thread_ms": 29.756961}, {"api": "pthread_cond_timedwait", "kind": "os_wait", "count": 65424, "summed_thread_ms": 623159.672447}, {"api": "poll", "kind": "os_wait", "count": 43173, "summed_thread_ms": 430741.518694}, {"api": "epoll_wait", "kind": "os_wait", "count": 40319, "summed_thread_ms": 367858.597399}, {"api": "pthread_cond_wait", "kind": "os_wait", "count": 304, "summed_thread_ms": 294162.018887}, {"api": "futex", "kind": "os_wait", "count": 12, "summed_thread_ms": 119657.942916}, {"api": "sem_clockwait", "kind": "os_wait", "count": 123, "summed_thread_ms": 71731.053247}, {"api": "sem_wait", "kind": "os_wait", "count": 29, "summed_thread_ms": 56248.829374}, {"api": "epoll_pwait", "kind": "os_wait", "count": 31, "summed_thread_ms": 9970.487947}, {"api": "epoll_ctl", "kind": "os_wait", "count": 34, "summed_thread_ms": 0.052908}], "overlap_interpretation": "Same-device overlapping activity intervals, not proof of simultaneous SM execution or net speedup. Non-NCCL includes copies/conversions implemented as kernels. Gaps are not attributed CPU waits. Thread times overlap and must not be added as wall time."}, "steady": {"top_time": [{"name": "nvjet_sm90_tst_128x200_64x5_2x1_v_bz_coopA_bias_TNT", "count": 7618, "total_ms": 917.505966, "mean_us": 120.43921843003413, "max_us": 246.367}, {"name": "void cutlass::device_kernel<flash::enable_sm90_or_later<flash::FlashAttnFwdSm90<flash::CollectiveMainloopFwdSm90<(int)2, cute::tuple<cute::C<(int)1>, cute::C<(int)1>, cute::C<(int)1>>, cute::tuple<cute::C<(int)128>, cute::C<(int)160>, cute::C<(int)128>>, (int)128, cutlass::bfloat16_t, float, cutlass::arch::Sm90, (bool)0, (bool)0, (bool)0, (bool)1, (bool)0, (bool)0, (bool)0, (bool)1, (bool)1, (bool)1, (bool)0, (bool)0, cutlass::bfloat16_t, (int)1>, flash::CollectiveEpilogueFwd<cute::tuple<cute::C<(int)128>, cute::C<(int)128>, cute::C<(int)160>>, cute::tuple<cute::C<(int)1>, cute::C<(int)1>, cute::C<(int)1>>, cutlass::bfloat16_t, cutlass::arch::Sm90, (int)256, (bool)1, (bool)1, (bool)0, (bool)0, (int)1>, flash::VarlenDynamicPersistentTileScheduler<(int)128, (int)160, (int)256, (int)128, (bool)0, (bool)1, (bool)1, (bool)0, (bool)0, (bool)1>>>>(T1::Params)", "count": 848, "total_ms": 840.687064, "mean_us": 991.3762547169811, "max_us": 1085.252}, {"name": "ncclDevKernel_SendRecv(ncclDevKernelArgsStorage<(unsigned long)4096>)", "count": 4971, "total_ms": 454.471394, "mean_us": 91.4245411386039, "max_us": 1672.71}, {"name": "void at::native::elementwise_kernel<(int)128, (int)2, void at::native::gpu_kernel_impl_nocast<at::native::direct_copy_kernel_cuda(at::TensorIteratorBase &)::[lambda() (instance 3)]::operator ()() const::[lambda() (instance 7)]::operator ()() const::[lambda(float) (instance 1)]>(at::TensorIteratorBase &, const T1 &)::[lambda(int) (instance 1)]>(int, T3)", "count": 13895, "total_ms": 384.26423, "mean_us": 27.6548564231738, "max_us": 230.145}, {"name": "sm90_xmma_fprop_implicit_gemm_f32f32_tf32f32_f32_nhwckrsc_nhwc_tilesize256x32x32_warpgroupsize1x1x1_g1_execute_segment_k_off_kernel__5x_cudnn", "count": 288, "total_ms": 383.824099, "mean_us": 1332.7225659722224, "max_us": 1365.637}, {"name": "sm90_xmma_fprop_implicit_gemm_f32f32_tf32f32_f32_nhwckrsc_nhwc_tilesize128x128x32_warpgroupsize1x1x1_g1_execute_segment_k_off_kernel__5x_cudnn", "count": 672, "total_ms": 332.943711, "mean_us": 495.45195089285716, "max_us": 894.049}, {"name": "nvjet_sm90_tst_256x152_64x4_1x2_h_bz_coopA_bias_TNT", "count": 848, "total_ms": 228.023722, "mean_us": 268.89589858490564, "max_us": 289.953}, {"name": "void cudnn::engines_precompiled::nchwToNhwcKernel<float, float, float, (bool)0, (bool)1, (cudnnKernelDataType_t)0>(cudnn::engines_precompiled::nchw2nhwc_params_t<T3>, const T1 *, T2 *)", "count": 3168, "total_ms": 101.231038, "mean_us": 31.95424179292929, "max_us": 191.425}, {"name": "fmha_cutlassF_f32_aligned_32x128_gmem_sm80(PyTorchMemEffAttention::AttentionKernel<float, cutlass::arch::Sm80, (bool)1, (int)32, (int)128, (int)65536, (bool)1, (bool)1>::Params)", "count": 48, "total_ms": 93.787351, "mean_us": 1953.9031458333332, "max_us": 2197.672}, {"name": "void at::native::elementwise_kernel<(int)128, (int)2, void at::native::gpu_kernel_impl_nocast<at::native::CUDAFunctor_add<float>>(at::TensorIteratorBase &, const T1 &)::[lambda(int) (instance 1)]>(int, T3)", "count": 1980, "total_ms": 80.488442, "mean_us": 40.65072828282828, "max_us": 142.529}, {"name": "void at::native::vectorized_gather_kernel<(int)16, long>(char *, char *, T2 *, int, long, long, long, long, bool)", "count": 1696, "total_ms": 65.722455, "mean_us": 38.7514475235849, "max_us": 41.504}, {"name": "void at::native::elementwise_kernel<(int)128, (int)2, void at::native::gpu_kernel_impl_nocast<at::native::BinaryFunctor<float, float, float, at::native::binary_internal::DivFunctor<float>>>(at::TensorIteratorBase &, const T1 &)::[lambda(int) (instance 1)]>(int, T3)", "count": 1465, "total_ms": 64.085437, "mean_us": 43.74432559726963, "max_us": 133.057}, {"name": "void at::native::elementwise_kernel<(int)128, (int)2, void at::native::gpu_kernel_impl_nocast<at::native::BinaryFunctor<float, float, float, at::native::binary_internal::MulFunctor<float>>>(at::TensorIteratorBase &, const T1 &)::[lambda(int) (instance 1)]>(int, T3)", "count": 1474, "total_ms": 63.967672, "mean_us": 43.39733514246947, "max_us": 126.817}, {"name": "void at::native::index_elementwise_kernel<(int)128, (int)4, void at::native::gpu_index_kernel<void at::native::index_put_kernel_impl<at::native::OpaqueType<(int)2>>(at::TensorIterator &, c10::ArrayRef<long>, c10::ArrayRef<long>)::[lambda(char *, const char *, long) (instance 1)]>(at::TensorIteratorBase &, c10::ArrayRef<long>, c10::ArrayRef<long>, const T1 &, bool)::[lambda(int) (instance 1)]>(long, T3)", "count": 1696, "total_ms": 50.882687, "mean_us": 30.001584316037736, "max_us": 33.343}, {"name": "void cudnn::engines_precompiled::nhwcToNchwKernel<float, float, float, (bool)1, (bool)0, (cudnnKernelDataType_t)0>(cudnn::engines_precompiled::nhwc2nchw_params_t<T3>, const T1 *, T2 *)", "count": 1632, "total_ms": 47.09927, "mean_us": 28.85984681372549, "max_us": 83.68}, {"name": "sm90_xmma_fprop_implicit_gemm_f32f32_tf32f32_f32_nhwckrsc_nhwc_tilesize64x64x32_warpgroupsize1x1x1_g1_execute_segment_k_off_kernel__5x_cudnn", "count": 480, "total_ms": 45.873365, "mean_us": 95.56951041666667, "max_us": 113.12}, {"name": "void at::native::vectorized_elementwise_kernel<(int)4, at::native::<unnamed>::silu_kernel(at::TensorIteratorBase &)::[lambda() (instance 1)]::operator ()() const::[lambda() (instance 2)]::operator ()() const::[lambda(float) (instance 1)], std::array<char *, (unsigned long)2>>(int, T2, T3)", "count": 1392, "total_ms": 38.896366, "mean_us": 27.94279166666667, "max_us": 75.04}, {"name": "void at::native::vectorized_elementwise_kernel<(int)4, at::native::AUnaryFunctor<float, float, float, at::native::binary_internal::MulFunctor<float>>, std::array<char *, (unsigned long)2>>(int, T2, T3)", "count": 1507, "total_ms": 37.886606, "mean_us": 25.140415394824153, "max_us": 73.12}, {"name": "void at::native::vectorized_elementwise_kernel<(int)4, at::native::CUDAFunctorOnSelf_add<float>, std::array<char *, (unsigned long)2>>(int, T2, T3)", "count": 1454, "total_ms": 37.843735, "mean_us": 26.0273280605227, "max_us": 73.184}, {"name": "void at::native::vectorized_elementwise_kernel<(int)4, at::native::FillFunctor<float>, std::array<char *, (unsigned long)1>>(int, T2, T3)", "count": 2450, "total_ms": 33.987061, "mean_us": 13.872269795918369, "max_us": 68.161}], "top_count": [{"name": "void at::native::elementwise_kernel<(int)128, (int)2, void at::native::gpu_kernel_impl_nocast<at::native::direct_copy_kernel_cuda(at::TensorIteratorBase &)::[lambda() (instance 3)]::operator ()() const::[lambda() (instance 7)]::operator ()() const::[lambda(float) (instance 1)]>(at::TensorIteratorBase &, const T1 &)::[lambda(int) (instance 1)]>(int, T3)", "count": 13895, "total_ms": 384.26423, "mean_us": 27.6548564231738, "max_us": 230.145}, {"name": "nvjet_sm90_tst_128x200_64x5_2x1_v_bz_coopA_bias_TNT", "count": 7618, "total_ms": 917.505966, "mean_us": 120.43921843003413, "max_us": 246.367}, {"name": "ncclDevKernel_SendRecv(ncclDevKernelArgsStorage<(unsigned long)4096>)", "count": 4971, "total_ms": 454.471394, "mean_us": 91.4245411386039, "max_us": 1672.71}, {"name": "void cudnn::engines_precompiled::nchwToNhwcKernel<float, float, float, (bool)0, (bool)1, (cudnnKernelDataType_t)0>(cudnn::engines_precompiled::nchw2nhwc_params_t<T3>, const T1 *, T2 *)", "count": 3168, "total_ms": 101.231038, "mean_us": 31.95424179292929, "max_us": 191.425}, {"name": "void at::native::vectorized_elementwise_kernel<(int)4, at::native::FillFunctor<float>, std::array<char *, (unsigned long)1>>(int, T2, T3)", "count": 2450, "total_ms": 33.987061, "mean_us": 13.872269795918369, "max_us": 68.161}, {"name": "void at::native::elementwise_kernel<(int)128, (int)2, void at::native::gpu_kernel_impl_nocast<at::native::CUDAFunctor_add<float>>(at::TensorIteratorBase &, const T1 &)::[lambda(int) (instance 1)]>(int, T3)", "count": 1980, "total_ms": 80.488442, "mean_us": 40.65072828282828, "max_us": 142.529}, {"name": "void at::native::vectorized_gather_kernel<(int)16, long>(char *, char *, T2 *, int, long, long, long, long, bool)", "count": 1696, "total_ms": 65.722455, "mean_us": 38.7514475235849, "max_us": 41.504}, {"name": "void at::native::index_elementwise_kernel<(int)128, (int)4, void at::native::gpu_index_kernel<void at::native::index_put_kernel_impl<at::native::OpaqueType<(int)2>>(at::TensorIterator &, c10::ArrayRef<long>, c10::ArrayRef<long>)::[lambda(char *, const char *, long) (instance 1)]>(at::TensorIteratorBase &, c10::ArrayRef<long>, c10::ArrayRef<long>, const T1 &, bool)::[lambda(int) (instance 1)]>(long, T3)", "count": 1696, "total_ms": 50.882687, "mean_us": 30.001584316037736, "max_us": 33.343}, {"name": "rotary_kernel", "count": 1696, "total_ms": 16.601388, "mean_us": 9.78855424528302, "max_us": 11.52}, {"name": "triton_poi_fused_all_to_all_single_clone_transpose_unsqueeze_view_5", "count": 1693, "total_ms": 11.367168, "mean_us": 6.714216184288246, "max_us": 7.616}, {"name": "triton_poi_fused_clone_transpose_view_8", "count": 1692, "total_ms": 10.415643, "mean_us": 6.155817375886525, "max_us": 7.584}, {"name": "void cudnn::engines_precompiled::nhwcToNchwKernel<float, float, float, (bool)1, (bool)0, (cudnnKernelDataType_t)0>(cudnn::engines_precompiled::nhwc2nchw_params_t<T3>, const T1 *, T2 *)", "count": 1632, "total_ms": 47.09927, "mean_us": 28.85984681372549, "max_us": 83.68}, {"name": "void at::native::vectorized_elementwise_kernel<(int)4, at::native::AUnaryFunctor<float, float, float, at::native::binary_internal::MulFunctor<float>>, std::array<char *, (unsigned long)2>>(int, T2, T3)", "count": 1507, "total_ms": 37.886606, "mean_us": 25.140415394824153, "max_us": 73.12}, {"name": "void at::native::elementwise_kernel<(int)128, (int)2, void at::native::gpu_kernel_impl_nocast<at::native::BinaryFunctor<float, float, float, at::native::binary_internal::MulFunctor<float>>>(at::TensorIteratorBase &, const T1 &)::[lambda(int) (instance 1)]>(int, T3)", "count": 1474, "total_ms": 63.967672, "mean_us": 43.39733514246947, "max_us": 126.817}, {"name": "void at::native::elementwise_kernel<(int)128, (int)2, void at::native::gpu_kernel_impl_nocast<at::native::BinaryFunctor<float, float, float, at::native::binary_internal::DivFunctor<float>>>(at::TensorIteratorBase &, const T1 &)::[lambda(int) (instance 1)]>(int, T3)", "count": 1465, "total_ms": 64.085437, "mean_us": 43.74432559726963, "max_us": 133.057}, {"name": "void at::native::vectorized_elementwise_kernel<(int)4, at::native::CUDAFunctorOnSelf_add<float>, std::array<char *, (unsigned long)2>>(int, T2, T3)", "count": 1454, "total_ms": 37.843735, "mean_us": 26.0273280605227, "max_us": 73.184}, {"name": "void at::native::vectorized_elementwise_kernel<(int)4, at::native::<unnamed>::launch_clamp_scalar(at::TensorIteratorBase &, c10::Scalar, c10::Scalar, at::native::detail::ClampLimits)::[lambda() (instance 1)]::operator ()() const::[lambda() (instance 7)]::operator ()() const::[lambda(float) (instance 1)], std::array<char *, (unsigned long)2>>(int, T2, T3)", "count": 1448, "total_ms": 2.367554, "mean_us": 1.6350511049723757, "max_us": 28.384}, {"name": "void at::native::reduce_kernel<(int)128, (int)4, at::native::ReduceOp<float, at::native::NormTwoOps<float, float, float, (bool)1>, unsigned int, float, (int)4, (int)4>>(T3)", "count": 1440, "total_ms": 31.842517, "mean_us": 22.112859027777777, "max_us": 50.048}, {"name": "void at::native::vectorized_elementwise_kernel<(int)4, at::native::<unnamed>::silu_kernel(at::TensorIteratorBase &)::[lambda() (instance 1)]::operator ()() const::[lambda() (instance 2)]::operator ()() const::[lambda(float) (instance 1)], std::array<char *, (unsigned long)2>>(int, T2, T3)", "count": 1392, "total_ms": 38.896366, "mean_us": 27.94279166666667, "max_us": 75.04}, {"name": "void cutlass::device_kernel<flash::enable_sm90_or_later<flash::FlashAttnFwdSm90<flash::CollectiveMainloopFwdSm90<(int)2, cute::tuple<cute::C<(int)1>, cute::C<(int)1>, cute::C<(int)1>>, cute::tuple<cute::C<(int)128>, cute::C<(int)160>, cute::C<(int)128>>, (int)128, cutlass::bfloat16_t, float, cutlass::arch::Sm90, (bool)0, (bool)0, (bool)0, (bool)1, (bool)0, (bool)0, (bool)0, (bool)1, (bool)1, (bool)1, (bool)0, (bool)0, cutlass::bfloat16_t, (int)1>, flash::CollectiveEpilogueFwd<cute::tuple<cute::C<(int)128>, cute::C<(int)128>, cute::C<(int)160>>, cute::tuple<cute::C<(int)1>, cute::C<(int)1>, cute::C<(int)1>>, cutlass::bfloat16_t, cutlass::arch::Sm90, (int)256, (bool)1, (bool)1, (bool)0, (bool)0, (int)1>, flash::VarlenDynamicPersistentTileScheduler<(int)128, (int)160, (int)256, (int)128, (bool)0, (bool)1, (bool)1, (bool)0, (bool)0, (bool)1>>>>(T1::Params)", "count": 848, "total_ms": 840.687064, "mean_us": 991.3762547169811, "max_us": 1085.252}], "groups": 102, "devices": [{"device": 4, "window_ms": 3408.402281, "non_nccl_ms": 1241.858338, "nccl_ms": 153.587024, "memcpy_ms": 20.660212, "non_nccl_nccl_overlap_ms": 34.919841, "kernel_memcpy_overlap_ms": 0.0, "observed_busy_ms": 1381.185733, "no_observed_kernel_or_memcpy_ms": 2027.216548}, {"device": 5, "window_ms": 3408.402281, "non_nccl_ms": 1033.99239, "nccl_ms": 83.29838, "memcpy_ms": 12.759331, "non_nccl_nccl_overlap_ms": 15.608163, "kernel_memcpy_overlap_ms": 0.0, "observed_busy_ms": 1114.441938, "no_observed_kernel_or_memcpy_ms": 2293.960343}, {"device": 6, "window_ms": 3408.402281, "non_nccl_ms": 1002.215853, "nccl_ms": 121.913302, "memcpy_ms": 12.520494, "non_nccl_nccl_overlap_ms": 26.179806, "kernel_memcpy_overlap_ms": 0.0, "observed_busy_ms": 1110.469843, "no_observed_kernel_or_memcpy_ms": 2297.932438}, {"device": 7, "window_ms": 3408.402281, "non_nccl_ms": 1014.352014, "nccl_ms": 115.954984, "memcpy_ms": 12.667178, "non_nccl_nccl_overlap_ms": 20.021247, "kernel_memcpy_overlap_ms": 0.0, "observed_busy_ms": 1122.952929, "no_observed_kernel_or_memcpy_ms": 2285.449352}], "copies": [{"direction": "Device-to-Host", "count": 22, "summed_gpu_ms": 3.635364, "bytes_full_overlapping_calls": 62300174}, {"direction": "Host-to-Device", "count": 982, "summed_gpu_ms": 1.956103, "bytes_full_overlapping_calls": 20017168}, {"direction": "Device-to-Device", "count": 10950, "summed_gpu_ms": 53.015748, "bytes_full_overlapping_calls": 90721055616}], "wait_apis": [{"api": "cudaStreamSynchronize_v3020", "kind": "sync", "count": 170, "summed_thread_ms": 1731.31351}, {"api": "cudaDeviceSynchronize_v3020", "kind": "sync", "count": 33, "summed_thread_ms": 1087.979398}, {"api": "cudaStreamWaitEvent_v3020", "kind": "sync", "count": 9800, "summed_thread_ms": 11.527508}, {"api": "pthread_cond_timedwait", "kind": "os_wait", "count": 24327, "summed_thread_ms": 212614.925718}, {"api": "poll", "kind": "os_wait", "count": 17196, "summed_thread_ms": 146507.264235}, {"api": "epoll_wait", "kind": "os_wait", "count": 13817, "summed_thread_ms": 125709.220199}, {"api": "pthread_cond_wait", "kind": "os_wait", "count": 142, "summed_thread_ms": 102252.109344}, {"api": "futex", "kind": "os_wait", "count": 12, "summed_thread_ms": 40900.827372}, {"api": "sem_clockwait", "kind": "os_wait", "count": 49, "summed_thread_ms": 24627.404524}, {"api": "sem_wait", "kind": "os_wait", "count": 14, "summed_thread_ms": 19027.120195}, {"api": "epoll_pwait", "kind": "os_wait", "count": 14, "summed_thread_ms": 3407.945175}, {"api": "epoll_ctl", "kind": "os_wait", "count": 14, "summed_thread_ms": 0.021936}], "overlap_interpretation": "Same-device overlapping activity intervals, not proof of simultaneous SM execution or net speedup. Non-NCCL includes copies/conversions implemented as kernels. Gaps are not attributed CPU waits. Thread times overlap and must not be added as wall time."}}}, "rebased": {"source": "rebased_nsys", "kernel_events": 988240, "clock_mapping": {"systemClockNs": 2851818204847309, "formula": "trace_ns = round(host perf_counter_seconds * 1e9) - systemClockNs", "validation": "same-host monotonic alignment; windows and per-device activity retained for audit"}, "windows_ns": {"full": null, "measured": [[309433753421, 319346930577]], "steady": [[315069533639, 315916112943], [315916112943, 316765250838], [316765250838, 317611000609], [317611000609, 318464083836]]}, "scope_notes": "full includes initialization/warmup/validation/shutdown. measured is epoch1. steady is CPU-ready intervals5–8. Profile timings are not clean benchmark timings.", "scopes": {"full": {"top_time": [{"name": "nvjet_sm90_tst_128x200_64x5_2x1_v_bz_coopA_bias_TNT", "count": 122560, "total_ms": 14513.075276, "mean_us": 118.41608417101827, "max_us": 246.112}, {"name": "ncclDevKernel_SendRecv(ncclDevKernelArgsStorage<(unsigned long)4096>)", "count": 63978, "total_ms": 13309.561557, "mean_us": 208.033410813092, "max_us": 3176721.962}, {"name": "void cutlass::device_kernel<flash::enable_sm90_or_later<flash::FlashAttnFwdSm90<flash::CollectiveMainloopFwdSm90<(int)2, cute::tuple<cute::C<(int)1>, cute::C<(int)1>, cute::C<(int)1>>, cute::tuple<cute::C<(int)128>, cute::C<(int)160>, cute::C<(int)128>>, (int)128, cutlass::bfloat16_t, float, cutlass::arch::Sm90, (bool)0, (bool)0, (bool)0, (bool)1, (bool)0, (bool)0, (bool)0, (bool)1, (bool)1, (bool)1, (bool)0, (bool)0, cutlass::bfloat16_t, (int)1>, flash::CollectiveEpilogueFwd<cute::tuple<cute::C<(int)128>, cute::C<(int)128>, cute::C<(int)160>>, cute::tuple<cute::C<(int)1>, cute::C<(int)1>, cute::C<(int)1>>, cutlass::bfloat16_t, cutlass::arch::Sm90, (int)256, (bool)1, (bool)1, (bool)0, (bool)0, (int)1>, flash::VarlenDynamicPersistentTileScheduler<(int)128, (int)160, (int)256, (int)128, (bool)0, (bool)1, (bool)1, (bool)0, (bool)0, (bool)1>>>>(T1::Params)", "count": 13619, "total_ms": 9516.278755, "mean_us": 698.7501839342095, "max_us": 1079.907}, {"name": "void at::native::elementwise_kernel<(int)128, (int)2, void at::native::gpu_kernel_impl_nocast<at::native::direct_copy_kernel_cuda(at::TensorIteratorBase &)::[lambda() (instance 3)]::operator ()() const::[lambda() (instance 7)]::operator ()() const::[lambda(float) (instance 1)]>(at::TensorIteratorBase &, const T1 &)::[lambda(int) (instance 1)]>(int, T3)", "count": 104520, "total_ms": 4872.91664, "mean_us": 46.62185840030616, "max_us": 747.937}, {"name": "nvjet_sm90_tst_256x152_64x4_1x2_h_bz_coopA_bias_TNT", "count": 13621, "total_ms": 3621.387804, "mean_us": 265.8679835548051, "max_us": 289.441}, {"name": "sm90_xmma_fprop_implicit_gemm_f32f32_tf32f32_f32_nhwckrsc_nhwc_tilesize256x32x32_warpgroupsize1x1x1_g1_execute_segment_k_off_kernel__5x_cudnn", "count": 1800, "total_ms": 2604.827749, "mean_us": 1447.1265272222222, "max_us": 5640.612}, {"name": "sm80_xmma_fprop_implicit_gemm_bf16bf16_bf16f32_f32_nhwckrsc_nhwc_tilesize256x32x32_stage4_warpsize4x1x1_g1_tensor16x8x16_execute_kernel__5x_cudnn", "count": 960, "total_ms": 2530.390921, "mean_us": 2635.8238760416666, "max_us": 2741.606}, {"name": "sm90_xmma_fprop_implicit_gemm_f32f32_tf32f32_f32_nhwckrsc_nhwc_tilesize128x128x32_warpgroupsize1x1x1_g1_execute_segment_k_off_kernel__5x_cudnn", "count": 4059, "total_ms": 2127.316819, "mean_us": 524.0987482138457, "max_us": 1829.857}, {"name": "sm90_xmma_fprop_implicit_gemm_bf16bf16_bf16f32_f32_nhwckrsc_nhwc_tilesize256x128x64_warpgroupsize2x1x1_g1_execute_segment_k_off_kernel__5x_cudnn", "count": 3840, "total_ms": 1556.351743, "mean_us": 405.29993307291664, "max_us": 1255.94}, {"name": "void at::native::elementwise_kernel<(int)128, (int)4, void at::native::gpu_kernel_impl<at::native::BinaryFunctor<float, float, float, at::native::binary_internal::DivFunctor<float>>>(at::TensorIteratorBase &, const T1 &)::[lambda(int) (instance 1)]>(int, T3)", "count": 5280, "total_ms": 1057.385312, "mean_us": 200.2623696969697, "max_us": 722.497}, {"name": "void at::native::vectorized_gather_kernel<(int)16, long>(char *, char *, T2 *, int, long, long, long, long, bool)", "count": 27438, "total_ms": 978.062319, "mean_us": 35.646268642029305, "max_us": 160.545}, {"name": "void at::native::elementwise_kernel<(int)128, (int)4, void at::native::gpu_kernel_impl_nocast<at::native::CUDAFunctor_add<c10::BFloat16>>(at::TensorIteratorBase &, const T1 &)::[lambda(int) (instance 1)]>(int, T3)", "count": 7824, "total_ms": 903.849974, "mean_us": 115.52274718813906, "max_us": 421.377}, {"name": "void at::native::elementwise_kernel<(int)128, (int)4, void at::native::gpu_kernel_impl<at::native::BinaryFunctor<float, float, float, at::native::binary_internal::MulFunctor<float>>>(at::TensorIteratorBase &, const T1 &)::[lambda(int) (instance 1)]>(int, T3)", "count": 5672, "total_ms": 885.404174, "mean_us": 156.1008769393512, "max_us": 587.426}, {"name": "void cudnn::engines_precompiled::nchwToNhwcKernel<__nv_bfloat16, __nv_bfloat16, float, (bool)0, (bool)1, (cudnnKernelDataType_t)0>(cudnn::engines_precompiled::nchw2nhwc_params_t<T3>, const T1 *, T2 *)", "count": 12448, "total_ms": 826.041609, "mean_us": 66.35938375642674, "max_us": 435.168}, {"name": "void at::native::index_elementwise_kernel<(int)128, (int)4, void at::native::gpu_index_kernel<void at::native::index_put_kernel_impl<at::native::OpaqueType<(int)2>>(at::TensorIterator &, c10::ArrayRef<long>, c10::ArrayRef<long>)::[lambda(char *, const char *, long) (instance 1)]>(at::TensorIteratorBase &, c10::ArrayRef<long>, c10::ArrayRef<long>, const T1 &, bool)::[lambda(int) (instance 1)]>(long, T3)", "count": 27241, "total_ms": 805.141565, "mean_us": 29.556241143864032, "max_us": 32.959}, {"name": "void at::native::vectorized_elementwise_kernel<(int)4, at::native::<unnamed>::silu_kernel(at::TensorIteratorBase &)::[lambda() (instance 1)]::operator ()() const::[lambda() (instance 2)]::operator ()() const::[lambda(float) (instance 1)], std::array<char *, (unsigned long)2>>(int, T2, T3)", "count": 13740, "total_ms": 686.971518, "mean_us": 49.997927074235804, "max_us": 290.464}, {"name": "void at::native::vectorized_elementwise_kernel<(int)4, at::native::AUnaryFunctor<float, float, float, at::native::binary_internal::MulFunctor<float>>, std::array<char *, (unsigned long)2>>(int, T2, T3)", "count": 15450, "total_ms": 684.810257, "mean_us": 44.3242884789644, "max_us": 289.728}, {"name": "void cudnn::engines_precompiled::nchwToNhwcKernel<float, float, float, (bool)0, (bool)1, (cudnnKernelDataType_t)0>(cudnn::engines_precompiled::nchw2nhwc_params_t<T3>, const T1 *, T2 *)", "count": 19748, "total_ms": 679.067487, "mean_us": 34.38664609074336, "max_us": 737.184}, {"name": "fmha_cutlassF_f32_aligned_32x128_gmem_sm80(PyTorchMemEffAttention::AttentionKernel<float, cutlass::arch::Sm80, (bool)1, (int)32, (int)128, (int)65536, (bool)1, (bool)1>::Params)", "count": 300, "total_ms": 575.16341, "mean_us": 1917.2113666666667, "max_us": 2185.543}, {"name": "void at::native::elementwise_kernel<(int)128, (int)2, void at::native::gpu_kernel_impl_nocast<at::native::CUDAFunctor_add<float>>(at::TensorIteratorBase &, const T1 &)::[lambda(int) (instance 1)]>(int, T3)", "count": 12588, "total_ms": 556.770809, "mean_us": 44.2302835239911, "max_us": 568.577}], "top_count": [{"name": "nvjet_sm90_tst_128x200_64x5_2x1_v_bz_coopA_bias_TNT", "count": 122560, "total_ms": 14513.075276, "mean_us": 118.41608417101827, "max_us": 246.112}, {"name": "void at::native::elementwise_kernel<(int)128, (int)2, void at::native::gpu_kernel_impl_nocast<at::native::direct_copy_kernel_cuda(at::TensorIteratorBase &)::[lambda() (instance 3)]::operator ()() const::[lambda() (instance 7)]::operator ()() const::[lambda(float) (instance 1)]>(at::TensorIteratorBase &, const T1 &)::[lambda(int) (instance 1)]>(int, T3)", "count": 104520, "total_ms": 4872.91664, "mean_us": 46.62185840030616, "max_us": 747.937}, {"name": "ncclDevKernel_SendRecv(ncclDevKernelArgsStorage<(unsigned long)4096>)", "count": 63978, "total_ms": 13309.561557, "mean_us": 208.033410813092, "max_us": 3176721.962}, {"name": "void at::native::vectorized_elementwise_kernel<(int)4, at::native::FillFunctor<int>, std::array<char *, (unsigned long)1>>(int, T2, T3)", "count": 39759, "total_ms": 389.673464, "mean_us": 9.800886943836616, "max_us": 16.576}, {"name": "triton_poi_fused_all_to_all_single_clone_transpose_unsqueeze_view_5", "count": 28084, "total_ms": 190.875524, "mean_us": 6.796593220338983, "max_us": 20.384}, {"name": "triton_poi_fused_clone_transpose_view_8", "count": 28083, "total_ms": 254.795417, "mean_us": 9.07294153046327, "max_us": 12.768}, {"name": "void at::native::vectorized_gather_kernel<(int)16, long>(char *, char *, T2 *, int, long, long, long, long, bool)", "count": 27438, "total_ms": 978.062319, "mean_us": 35.646268642029305, "max_us": 160.545}, {"name": "rotary_kernel", "count": 27242, "total_ms": 261.916138, "mean_us": 9.614423977681522, "max_us": 11.552}, {"name": "void at::native::index_elementwise_kernel<(int)128, (int)4, void at::native::gpu_index_kernel<void at::native::index_put_kernel_impl<at::native::OpaqueType<(int)2>>(at::TensorIterator &, c10::ArrayRef<long>, c10::ArrayRef<long>)::[lambda(char *, const char *, long) (instance 1)]>(at::TensorIteratorBase &, c10::ArrayRef<long>, c10::ArrayRef<long>, const T1 &, bool)::[lambda(int) (instance 1)]>(long, T3)", "count": 27241, "total_ms": 805.141565, "mean_us": 29.556241143864032, "max_us": 32.959}, {"name": "void at::native::vectorized_elementwise_kernel<(int)4, at::native::FillFunctor<float>, std::array<char *, (unsigned long)1>>(int, T2, T3)", "count": 20504, "total_ms": 541.62229, "mean_us": 26.41544527896996, "max_us": 203.072}, {"name": "void cudnn::engines_precompiled::nchwToNhwcKernel<float, float, float, (bool)0, (bool)1, (cudnnKernelDataType_t)0>(cudnn::engines_precompiled::nchw2nhwc_params_t<T3>, const T1 *, T2 *)", "count": 19748, "total_ms": 679.067487, "mean_us": 34.38664609074336, "max_us": 737.184}, {"name": "triton_red_fused__to_copy_add_mul_native_layer_norm_split_squeeze_1", "count": 15847, "total_ms": 360.579959, "mean_us": 22.753830945920363, "max_us": 44.064}, {"name": "void at::native::vectorized_elementwise_kernel<(int)4, at::native::AUnaryFunctor<float, float, float, at::native::binary_internal::MulFunctor<float>>, std::array<char *, (unsigned long)2>>(int, T2, T3)", "count": 15450, "total_ms": 684.810257, "mean_us": 44.3242884789644, "max_us": 289.728}, {"name": "triton_red_fused__to_copy_add_mul_native_layer_norm_split_squeeze_view_9", "count": 15313, "total_ms": 387.460614, "mean_us": 25.302724090641938, "max_us": 33.248}, {"name": "triton_red_fused__to_copy_div_mul_pow_split_with_sizes_sqrt_sum_view_2", "count": 15105, "total_ms": 143.63734, "mean_us": 9.50925786163522, "max_us": 11.04}, {"name": "triton_red_fused__to_copy_div_mul_pow_split_with_sizes_sqrt_sum_view_3", "count": 14893, "total_ms": 155.448514, "mean_us": 10.437689787148326, "max_us": 12.608}, {"name": "triton_poi_fused_all_to_all_single_clone_split_with_sizes_stack_transpose_view_4", "count": 14469, "total_ms": 275.116208, "mean_us": 19.014182597276935, "max_us": 38.912}, {"name": "triton_poi_fused_silu_view_6", "count": 14467, "total_ms": 161.532316, "mean_us": 11.165571023709132, "max_us": 20.256}, {"name": "triton_poi_fused_gelu_view_13", "count": 14465, "total_ms": 321.685942, "mean_us": 22.23891752506049, "max_us": 31.36}, {"name": "triton_poi_fused__to_copy_add_mul_split_squeeze_view_14", "count": 14465, "total_ms": 221.466937, "mean_us": 15.310538333909436, "max_us": 18.272}], "groups": 167, "devices": [{"device": 4, "window_ms": 190278.084491, "non_nccl_ms": 16036.882544, "nccl_ms": 5296.227035, "memcpy_ms": 7382.847682, "non_nccl_nccl_overlap_ms": 339.357552, "kernel_memcpy_overlap_ms": 0.0, "observed_busy_ms": 28376.599709, "no_observed_kernel_or_memcpy_ms": 161901.484782}, {"device": 5, "window_ms": 190278.084491, "non_nccl_ms": 14425.233801, "nccl_ms": 2150.925659, "memcpy_ms": 6615.012227, "non_nccl_nccl_overlap_ms": 245.069106, "kernel_memcpy_overlap_ms": 0.0, "observed_busy_ms": 22946.102581, "no_observed_kernel_or_memcpy_ms": 167331.98191}, {"device": 6, "window_ms": 190278.084491, "non_nccl_ms": 16312.994595, "nccl_ms": 3453.648545, "memcpy_ms": 7527.186286, "non_nccl_nccl_overlap_ms": 474.182126, "kernel_memcpy_overlap_ms": 0.0, "observed_busy_ms": 26819.6473, "no_observed_kernel_or_memcpy_ms": 163458.437191}, {"device": 7, "window_ms": 190278.084491, "non_nccl_ms": 14388.571025, "nccl_ms": 2636.987616, "memcpy_ms": 6773.6332, "non_nccl_nccl_overlap_ms": 256.912808, "kernel_memcpy_overlap_ms": 0.0, "observed_busy_ms": 23542.279033, "no_observed_kernel_or_memcpy_ms": 166735.805458}], "copies": [{"direction": "Host-to-Device", "count": 33481, "summed_gpu_ms": 26740.04699, "bytes_full_overlapping_calls": 197522805020}, {"direction": "Device-to-Device", "count": 169349, "summed_gpu_ms": 886.13919, "bytes_full_overlapping_calls": 1519817572796}, {"direction": "Device-to-Host", "count": 656, "summed_gpu_ms": 672.493215, "bytes_full_overlapping_calls": 1985586690}], "wait_apis": [{"api": "cudaStreamSynchronize_v3020", "kind": "sync", "count": 12471, "summed_thread_ms": 24015.792697}, {"api": "cudaDeviceSynchronize_v3020", "kind": "sync", "count": 1713, "summed_thread_ms": 17046.196328}, {"api": "cudaStreamWaitEvent_v3020", "kind": "sync", "count": 66310, "summed_thread_ms": 82.739448}, {"api": "pthread_cond_timedwait", "kind": "os_wait", "count": 291761, "summed_thread_ms": 8908799.362993}, {"api": "poll", "kind": "os_wait", "count": 84809, "summed_thread_ms": 5027632.012185}, {"api": "epoll_wait", "kind": "os_wait", "count": 456867, "summed_thread_ms": 4973512.475385}, {"api": "pthread_cond_wait", "kind": "os_wait", "count": 583, "summed_thread_ms": 3959227.485109}, {"api": "futex", "kind": "os_wait", "count": 53, "summed_thread_ms": 2605492.605929}, {"api": "sem_clockwait", "kind": "os_wait", "count": 1038, "summed_thread_ms": 789029.18214}, {"api": "sem_wait", "kind": "os_wait", "count": 741, "summed_thread_ms": 756851.760683}, {"api": "waitpid", "kind": "os_wait", "count": 898, "summed_thread_ms": 216028.666692}, {"api": "epoll_pwait", "kind": "os_wait", "count": 653, "summed_thread_ms": 214389.881784}, {"api": "clock_nanosleep", "kind": "os_wait", "count": 422, "summed_thread_ms": 14013.582432}, {"api": "nanosleep", "kind": "os_wait", "count": 270, "summed_thread_ms": 2154.96136}, {"api": "sleep", "kind": "os_wait", "count": 6300, "summed_thread_ms": 1092.012352}, {"api": "sem_timedwait", "kind": "os_wait", "count": 77, "summed_thread_ms": 3.065123}, {"api": "epoll_ctl", "kind": "os_wait", "count": 145, "summed_thread_ms": 0.278708}], "overlap_interpretation": "Same-device overlapping activity intervals, not proof of simultaneous SM execution or net speedup. Non-NCCL includes copies/conversions implemented as kernels. Gaps are not attributed CPU waits. Thread times overlap and must not be added as wall time."}, "measured": {"top_time": [{"name": "nvjet_sm90_tst_128x200_64x5_2x1_v_bz_coopA_bias_TNT", "count": 50560, "total_ms": 5992.534462, "mean_us": 118.5232290743671, "max_us": 245.856}, {"name": "void cutlass::device_kernel<flash::enable_sm90_or_later<flash::FlashAttnFwdSm90<flash::CollectiveMainloopFwdSm90<(int)2, cute::tuple<cute::C<(int)1>, cute::C<(int)1>, cute::C<(int)1>>, cute::tuple<cute::C<(int)128>, cute::C<(int)160>, cute::C<(int)128>>, (int)128, cutlass::bfloat16_t, float, cutlass::arch::Sm90, (bool)0, (bool)0, (bool)0, (bool)1, (bool)0, (bool)0, (bool)0, (bool)1, (bool)1, (bool)1, (bool)0, (bool)0, cutlass::bfloat16_t, (int)1>, flash::CollectiveEpilogueFwd<cute::tuple<cute::C<(int)128>, cute::C<(int)128>, cute::C<(int)160>>, cute::tuple<cute::C<(int)1>, cute::C<(int)1>, cute::C<(int)1>>, cutlass::bfloat16_t, cutlass::arch::Sm90, (int)256, (bool)1, (bool)1, (bool)0, (bool)0, (int)1>, flash::VarlenDynamicPersistentTileScheduler<(int)128, (int)160, (int)256, (int)128, (bool)0, (bool)1, (bool)1, (bool)0, (bool)0, (bool)1>>>>(T1::Params)", "count": 5619, "total_ms": 3587.081971, "mean_us": 638.3844048763125, "max_us": 1079.427}, {"name": "ncclDevKernel_SendRecv(ncclDevKernelArgsStorage<(unsigned long)4096>)", "count": 26434, "total_ms": 2379.381307, "mean_us": 90.01215506544601, "max_us": 87488.285}, {"name": "void at::native::elementwise_kernel<(int)128, (int)2, void at::native::gpu_kernel_impl_nocast<at::native::direct_copy_kernel_cuda(at::TensorIteratorBase &)::[lambda() (instance 3)]::operator ()() const::[lambda() (instance 7)]::operator ()() const::[lambda(float) (instance 1)]>(at::TensorIteratorBase &, const T1 &)::[lambda(int) (instance 1)]>(int, T3)", "count": 44713, "total_ms": 2157.005835, "mean_us": 48.24113423389171, "max_us": 704.354}, {"name": "nvjet_sm90_tst_256x152_64x4_1x2_h_bz_coopA_bias_TNT", "count": 5621, "total_ms": 1493.726349, "mean_us": 265.74032182885605, "max_us": 288.609}, {"name": "sm80_xmma_fprop_implicit_gemm_bf16bf16_bf16f32_f32_nhwckrsc_nhwc_tilesize256x32x32_stage4_warpsize4x1x1_g1_tensor16x8x16_execute_kernel__5x_cudnn", "count": 480, "total_ms": 1265.251833, "mean_us": 2635.94131875, "max_us": 2727.176}, {"name": "sm90_xmma_fprop_implicit_gemm_f32f32_tf32f32_f32_nhwckrsc_nhwc_tilesize256x32x32_warpgroupsize1x1x1_g1_execute_segment_k_off_kernel__5x_cudnn", "count": 720, "total_ms": 934.815189, "mean_us": 1298.3544291666667, "max_us": 1366.212}, {"name": "sm90_xmma_fprop_implicit_gemm_f32f32_tf32f32_f32_nhwckrsc_nhwc_tilesize128x128x32_warpgroupsize1x1x1_g1_execute_segment_k_off_kernel__5x_cudnn", "count": 1652, "total_ms": 825.69242, "mean_us": 499.81381355932206, "max_us": 899.009}, {"name": "sm90_xmma_fprop_implicit_gemm_bf16bf16_bf16f32_f32_nhwckrsc_nhwc_tilesize256x128x64_warpgroupsize2x1x1_g1_execute_segment_k_off_kernel__5x_cudnn", "count": 1920, "total_ms": 778.570635, "mean_us": 405.5055390625, "max_us": 1255.94}, {"name": "void at::native::elementwise_kernel<(int)128, (int)4, void at::native::gpu_kernel_impl<at::native::BinaryFunctor<float, float, float, at::native::binary_internal::DivFunctor<float>>>(at::TensorIteratorBase &, const T1 &)::[lambda(int) (instance 1)]>(int, T3)", "count": 2640, "total_ms": 528.636736, "mean_us": 200.24118787878785, "max_us": 722.497}, {"name": "void at::native::elementwise_kernel<(int)128, (int)4, void at::native::gpu_kernel_impl_nocast<at::native::CUDAFunctor_add<c10::BFloat16>>(at::TensorIteratorBase &, const T1 &)::[lambda(int) (instance 1)]>(int, T3)", "count": 3912, "total_ms": 451.885966, "mean_us": 115.51277249488753, "max_us": 420.64}, {"name": "void at::native::elementwise_kernel<(int)128, (int)4, void at::native::gpu_kernel_impl<at::native::BinaryFunctor<float, float, float, at::native::binary_internal::MulFunctor<float>>>(at::TensorIteratorBase &, const T1 &)::[lambda(int) (instance 1)]>(int, T3)", "count": 2836, "total_ms": 442.678735, "mean_us": 156.0926428067701, "max_us": 587.426}, {"name": "void cudnn::engines_precompiled::nchwToNhwcKernel<__nv_bfloat16, __nv_bfloat16, float, (bool)0, (bool)1, (cudnnKernelDataType_t)0>(cudnn::engines_precompiled::nchw2nhwc_params_t<T3>, const T1 *, T2 *)", "count": 6224, "total_ms": 413.143307, "mean_us": 66.37906603470438, "max_us": 435.168}, {"name": "void at::native::vectorized_gather_kernel<(int)16, long>(char *, char *, T2 *, int, long, long, long, long, bool)", "count": 11338, "total_ms": 397.165881, "mean_us": 35.02962436055742, "max_us": 159.264}, {"name": "void at::native::index_elementwise_kernel<(int)128, (int)4, void at::native::gpu_index_kernel<void at::native::index_put_kernel_impl<at::native::OpaqueType<(int)2>>(at::TensorIterator &, c10::ArrayRef<long>, c10::ArrayRef<long>)::[lambda(char *, const char *, long) (instance 1)]>(at::TensorIteratorBase &, c10::ArrayRef<long>, c10::ArrayRef<long>, const T1 &, bool)::[lambda(int) (instance 1)]>(long, T3)", "count": 11241, "total_ms": 332.393023, "mean_us": 29.56970225068944, "max_us": 32.767}, {"name": "void at::native::vectorized_elementwise_kernel<(int)4, at::native::<unnamed>::silu_kernel(at::TensorIteratorBase &)::[lambda() (instance 1)]::operator ()() const::[lambda() (instance 2)]::operator ()() const::[lambda(float) (instance 1)], std::array<char *, (unsigned long)2>>(int, T2, T3)", "count": 6000, "total_ms": 307.537735, "mean_us": 51.25628916666667, "max_us": 288.544}, {"name": "void at::native::vectorized_elementwise_kernel<(int)4, at::native::AUnaryFunctor<float, float, float, at::native::binary_internal::MulFunctor<float>>, std::array<char *, (unsigned long)2>>(int, T2, T3)", "count": 6744, "total_ms": 307.077113, "mean_us": 45.53337974495848, "max_us": 289.728}, {"name": "void at::native::vectorized_elementwise_kernel<(int)8, at::native::bfloat16_copy_kernel_cuda(at::TensorIteratorBase &)::[lambda(float) (instance 1)], std::array<char *, (unsigned long)2>>(int, T2, T3)", "count": 3685, "total_ms": 263.048045, "mean_us": 71.38345861601086, "max_us": 319.361}, {"name": "void cudnn::engines_precompiled::nchwToNhwcKernel<float, float, float, (bool)0, (bool)1, (cudnnKernelDataType_t)0>(cudnn::engines_precompiled::nchw2nhwc_params_t<T3>, const T1 *, T2 *)", "count": 7904, "total_ms": 249.1105, "mean_us": 31.51701670040486, "max_us": 190.657}, {"name": "void at::native::vectorized_elementwise_kernel<(int)4, at::native::FillFunctor<float>, std::array<char *, (unsigned long)1>>(int, T2, T3)", "count": 8813, "total_ms": 241.488334, "mean_us": 27.40137682968342, "max_us": 203.072}], "top_count": [{"name": "nvjet_sm90_tst_128x200_64x5_2x1_v_bz_coopA_bias_TNT", "count": 50560, "total_ms": 5992.534462, "mean_us": 118.5232290743671, "max_us": 245.856}, {"name": "void at::native::elementwise_kernel<(int)128, (int)2, void at::native::gpu_kernel_impl_nocast<at::native::direct_copy_kernel_cuda(at::TensorIteratorBase &)::[lambda() (instance 3)]::operator ()() const::[lambda() (instance 7)]::operator ()() const::[lambda(float) (instance 1)]>(at::TensorIteratorBase &, const T1 &)::[lambda(int) (instance 1)]>(int, T3)", "count": 44713, "total_ms": 2157.005835, "mean_us": 48.24113423389171, "max_us": 704.354}, {"name": "ncclDevKernel_SendRecv(ncclDevKernelArgsStorage<(unsigned long)4096>)", "count": 26434, "total_ms": 2379.381307, "mean_us": 90.01215506544601, "max_us": 87488.285}, {"name": "void at::native::vectorized_gather_kernel<(int)16, long>(char *, char *, T2 *, int, long, long, long, long, bool)", "count": 11338, "total_ms": 397.165881, "mean_us": 35.02962436055742, "max_us": 159.264}, {"name": "rotary_kernel", "count": 11242, "total_ms": 108.169155, "mean_us": 9.62187822451521, "max_us": 11.359}, {"name": "void at::native::index_elementwise_kernel<(int)128, (int)4, void at::native::gpu_index_kernel<void at::native::index_put_kernel_impl<at::native::OpaqueType<(int)2>>(at::TensorIterator &, c10::ArrayRef<long>, c10::ArrayRef<long>)::[lambda(char *, const char *, long) (instance 1)]>(at::TensorIteratorBase &, c10::ArrayRef<long>, c10::ArrayRef<long>, const T1 &, bool)::[lambda(int) (instance 1)]>(long, T3)", "count": 11241, "total_ms": 332.393023, "mean_us": 29.56970225068944, "max_us": 32.767}, {"name": "triton_poi_fused_all_to_all_single_clone_transpose_unsqueeze_view_5", "count": 11236, "total_ms": 74.082312, "mean_us": 6.59329939480242, "max_us": 7.776}, {"name": "triton_poi_fused_clone_transpose_view_8", "count": 11235, "total_ms": 102.233998, "mean_us": 9.099599287939474, "max_us": 12.16}, {"name": "void at::native::vectorized_elementwise_kernel<(int)4, at::native::FillFunctor<float>, std::array<char *, (unsigned long)1>>(int, T2, T3)", "count": 8813, "total_ms": 241.488334, "mean_us": 27.40137682968342, "max_us": 203.072}, {"name": "void cudnn::engines_precompiled::nchwToNhwcKernel<float, float, float, (bool)0, (bool)1, (cudnnKernelDataType_t)0>(cudnn::engines_precompiled::nchw2nhwc_params_t<T3>, const T1 *, T2 *)", "count": 7904, "total_ms": 249.1105, "mean_us": 31.51701670040486, "max_us": 190.657}, {"name": "void at::native::vectorized_elementwise_kernel<(int)4, at::native::AUnaryFunctor<float, float, float, at::native::binary_internal::MulFunctor<float>>, std::array<char *, (unsigned long)2>>(int, T2, T3)", "count": 6744, "total_ms": 307.077113, "mean_us": 45.53337974495848, "max_us": 289.728}, {"name": "void at::native::vectorized_elementwise_kernel<(int)4, at::native::<unnamed>::launch_clamp_scalar(at::TensorIteratorBase &, c10::Scalar, c10::Scalar, at::native::detail::ClampLimits)::[lambda() (instance 1)]::operator ()() const::[lambda() (instance 7)]::operator ()() const::[lambda(float) (instance 1)], std::array<char *, (unsigned long)2>>(int, T2, T3)", "count": 6260, "total_ms": 10.837109, "mean_us": 1.731167571884984, "max_us": 30.112}, {"name": "void cudnn::engines_precompiled::nchwToNhwcKernel<__nv_bfloat16, __nv_bfloat16, float, (bool)0, (bool)1, (cudnnKernelDataType_t)0>(cudnn::engines_precompiled::nchw2nhwc_params_t<T3>, const T1 *, T2 *)", "count": 6224, "total_ms": 413.143307, "mean_us": 66.37906603470438, "max_us": 435.168}, {"name": "void at::native::vectorized_elementwise_kernel<(int)4, at::native::<unnamed>::silu_kernel(at::TensorIteratorBase &)::[lambda() (instance 1)]::operator ()() const::[lambda() (instance 2)]::operator ()() const::[lambda(float) (instance 1)], std::array<char *, (unsigned long)2>>(int, T2, T3)", "count": 6000, "total_ms": 307.537735, "mean_us": 51.25628916666667, "max_us": 288.544}, {"name": "void at::native::unrolled_elementwise_kernel<at::native::direct_copy_kernel_cuda(at::TensorIteratorBase &)::[lambda() (instance 3)]::operator ()() const::[lambda() (instance 4)]::operator ()() const::[lambda(long) (instance 1)], std::array<char *, (unsigned long)2>, (int)4, TrivialOffsetCalculator<(int)1, unsigned int>, TrivialOffsetCalculator<(int)1, unsigned int>, at::native::memory::LoadWithCast<(int)1>, at::native::memory::StoreWithCast<(int)1>>(int, T1, T2, T4, T5, T6, T7)", "count": 5811, "total_ms": 9.088919, "mean_us": 1.564088625021511, "max_us": 3.136}, {"name": "nvjet_sm90_tst_256x152_64x4_1x2_h_bz_coopA_bias_TNT", "count": 5621, "total_ms": 1493.726349, "mean_us": 265.74032182885605, "max_us": 288.609}, {"name": "triton_red_fused__to_copy_add_mul_native_layer_norm_split_squeeze_1", "count": 5621, "total_ms": 124.531301, "mean_us": 22.154652375022238, "max_us": 26.016}, {"name": "triton_poi_fused_all_to_all_single_clone_split_with_sizes_stack_transpose_view_4", "count": 5621, "total_ms": 103.460887, "mean_us": 18.406135385162784, "max_us": 21.184}, {"name": "triton_red_fused__to_copy_div_mul_pow_split_with_sizes_sqrt_sum_view_3", "count": 5621, "total_ms": 59.157434, "mean_us": 10.524361145703612, "max_us": 12.511}, {"name": "triton_red_fused__to_copy_div_mul_pow_split_with_sizes_sqrt_sum_view_2", "count": 5621, "total_ms": 53.399818, "mean_us": 9.500056573563423, "max_us": 11.008}], "groups": 157, "devices": [{"device": 4, "window_ms": 9913.177156, "non_nccl_ms": 5962.002534, "nccl_ms": 578.495449, "memcpy_ms": 93.772609, "non_nccl_nccl_overlap_ms": 126.626182, "kernel_memcpy_overlap_ms": 0.0, "observed_busy_ms": 6507.64441, "no_observed_kernel_or_memcpy_ms": 3405.532746}, {"device": 5, "window_ms": 9913.177156, "non_nccl_ms": 5772.499506, "nccl_ms": 402.511957, "memcpy_ms": 79.866486, "non_nccl_nccl_overlap_ms": 91.469678, "kernel_memcpy_overlap_ms": 0.0, "observed_busy_ms": 6163.408271, "no_observed_kernel_or_memcpy_ms": 3749.768885}, {"device": 6, "window_ms": 9913.177156, "non_nccl_ms": 7798.248181, "nccl_ms": 918.725322, "memcpy_ms": 120.708142, "non_nccl_nccl_overlap_ms": 238.961984, "kernel_memcpy_overlap_ms": 0.0, "observed_busy_ms": 8598.719661, "no_observed_kernel_or_memcpy_ms": 1314.457495}, {"device": 7, "window_ms": 9913.177156, "non_nccl_ms": 5759.471468, "nccl_ms": 544.659735, "memcpy_ms": 79.471242, "non_nccl_nccl_overlap_ms": 91.085124, "kernel_memcpy_overlap_ms": 0.0, "observed_busy_ms": 6292.517321, "no_observed_kernel_or_memcpy_ms": 3620.659835}], "copies": [{"direction": "Host-to-Device", "count": 6419, "summed_gpu_ms": 12.543561, "bytes_full_overlapping_calls": 72562992}, {"direction": "Device-to-Device", "count": 68615, "summed_gpu_ms": 351.688025, "bytes_full_overlapping_calls": 606922033392}, {"direction": "Device-to-Host", "count": 88, "summed_gpu_ms": 9.586893, "bytes_full_overlapping_calls": 152156228}], "wait_apis": [{"api": "cudaStreamSynchronize_v3020", "kind": "sync", "count": 835, "summed_thread_ms": 10186.477322}, {"api": "cudaDeviceSynchronize_v3020", "kind": "sync", "count": 105, "summed_thread_ms": 4475.880993}, {"api": "cudaStreamWaitEvent_v3020", "kind": "sync", "count": 24684, "summed_thread_ms": 28.825794}, {"api": "pthread_cond_timedwait", "kind": "os_wait", "count": 1686, "summed_thread_ms": 332473.392385}, {"api": "poll", "kind": "os_wait", "count": 30682, "summed_thread_ms": 238008.55453}, {"api": "pthread_cond_wait", "kind": "os_wait", "count": 221, "summed_thread_ms": 216420.231876}, {"api": "epoll_wait", "kind": "os_wait", "count": 10065, "summed_thread_ms": 188243.342194}, {"api": "futex", "kind": "os_wait", "count": 12, "summed_thread_ms": 118958.125872}, {"api": "sem_wait", "kind": "os_wait", "count": 3, "summed_thread_ms": 29739.531468}, {"api": "sem_clockwait", "kind": "os_wait", "count": 4, "summed_thread_ms": 19826.049427}, {"api": "epoll_pwait", "kind": "os_wait", "count": 29, "summed_thread_ms": 9911.982228}], "overlap_interpretation": "Same-device overlapping activity intervals, not proof of simultaneous SM execution or net speedup. Non-NCCL includes copies/conversions implemented as kernels. Gaps are not attributed CPU waits. Thread times overlap and must not be added as wall time."}, "steady": {"top_time": [{"name": "nvjet_sm90_tst_128x200_64x5_2x1_v_bz_coopA_bias_TNT", "count": 13126, "total_ms": 1574.520988, "mean_us": 119.95436446746915, "max_us": 245.856}, {"name": "void cutlass::device_kernel<flash::enable_sm90_or_later<flash::FlashAttnFwdSm90<flash::CollectiveMainloopFwdSm90<(int)2, cute::tuple<cute::C<(int)1>, cute::C<(int)1>, cute::C<(int)1>>, cute::tuple<cute::C<(int)128>, cute::C<(int)160>, cute::C<(int)128>>, (int)128, cutlass::bfloat16_t, float, cutlass::arch::Sm90, (bool)0, (bool)0, (bool)0, (bool)1, (bool)0, (bool)0, (bool)0, (bool)1, (bool)1, (bool)1, (bool)0, (bool)0, cutlass::bfloat16_t, (int)1>, flash::CollectiveEpilogueFwd<cute::tuple<cute::C<(int)128>, cute::C<(int)128>, cute::C<(int)160>>, cute::tuple<cute::C<(int)1>, cute::C<(int)1>, cute::C<(int)1>>, cutlass::bfloat16_t, cutlass::arch::Sm90, (int)256, (bool)1, (bool)1, (bool)0, (bool)0, (int)1>, flash::VarlenDynamicPersistentTileScheduler<(int)128, (int)160, (int)256, (int)128, (bool)0, (bool)1, (bool)1, (bool)0, (bool)0, (bool)1>>>>(T1::Params)", "count": 1459, "total_ms": 1428.09285, "mean_us": 978.8162097326937, "max_us": 1079.427}, {"name": "ncclDevKernel_SendRecv(ncclDevKernelArgsStorage<(unsigned long)4096>)", "count": 7420, "total_ms": 695.087986, "mean_us": 93.67762614555257, "max_us": 2432.649}, {"name": "nvjet_sm90_tst_256x152_64x4_1x2_h_bz_coopA_bias_TNT", "count": 1461, "total_ms": 388.515463, "mean_us": 265.92434154688567, "max_us": 288.609}, {"name": "void at::native::elementwise_kernel<(int)128, (int)2, void at::native::gpu_kernel_impl_nocast<at::native::direct_copy_kernel_cuda(at::TensorIteratorBase &)::[lambda() (instance 3)]::operator ()() const::[lambda() (instance 7)]::operator ()() const::[lambda(float) (instance 1)]>(at::TensorIteratorBase &, const T1 &)::[lambda(int) (instance 1)]>(int, T3)", "count": 13916, "total_ms": 384.311752, "mean_us": 27.616538660534637, "max_us": 230.688}, {"name": "sm90_xmma_fprop_implicit_gemm_f32f32_tf32f32_f32_nhwckrsc_nhwc_tilesize256x32x32_warpgroupsize1x1x1_g1_execute_segment_k_off_kernel__5x_cudnn", "count": 288, "total_ms": 383.262111, "mean_us": 1330.77121875, "max_us": 1355.781}, {"name": "sm90_xmma_fprop_implicit_gemm_f32f32_tf32f32_f32_nhwckrsc_nhwc_tilesize128x128x32_warpgroupsize1x1x1_g1_execute_segment_k_off_kernel__5x_cudnn", "count": 672, "total_ms": 337.404766, "mean_us": 502.0904255952381, "max_us": 888.034}, {"name": "void at::native::vectorized_gather_kernel<(int)16, long>(char *, char *, T2 *, int, long, long, long, long, bool)", "count": 2918, "total_ms": 113.005016, "mean_us": 38.72687320082248, "max_us": 41.472}, {"name": "void cudnn::engines_precompiled::nchwToNhwcKernel<float, float, float, (bool)0, (bool)1, (cudnnKernelDataType_t)0>(cudnn::engines_precompiled::nchw2nhwc_params_t<T3>, const T1 *, T2 *)", "count": 3168, "total_ms": 101.16054, "mean_us": 31.931988636363634, "max_us": 190.657}, {"name": "fmha_cutlassF_f32_aligned_32x128_gmem_sm80(PyTorchMemEffAttention::AttentionKernel<float, cutlass::arch::Sm80, (bool)1, (int)32, (int)128, (int)65536, (bool)1, (bool)1>::Params)", "count": 48, "total_ms": 93.262298, "mean_us": 1942.9645416666667, "max_us": 2185.543}, {"name": "void at::native::index_elementwise_kernel<(int)128, (int)4, void at::native::gpu_index_kernel<void at::native::index_put_kernel_impl<at::native::OpaqueType<(int)2>>(at::TensorIterator &, c10::ArrayRef<long>, c10::ArrayRef<long>)::[lambda(char *, const char *, long) (instance 1)]>(at::TensorIteratorBase &, c10::ArrayRef<long>, c10::ArrayRef<long>, const T1 &, bool)::[lambda(int) (instance 1)]>(long, T3)", "count": 2921, "total_ms": 86.738054, "mean_us": 29.694643615200274, "max_us": 32.767}, {"name": "void at::native::elementwise_kernel<(int)128, (int)2, void at::native::gpu_kernel_impl_nocast<at::native::CUDAFunctor_add<float>>(at::TensorIteratorBase &, const T1 &)::[lambda(int) (instance 1)]>(int, T3)", "count": 2002, "total_ms": 81.826604, "mean_us": 40.872429570429574, "max_us": 142.081}, {"name": "void at::native::elementwise_kernel<(int)128, (int)2, void at::native::gpu_kernel_impl_nocast<at::native::BinaryFunctor<float, float, float, at::native::binary_internal::MulFunctor<float>>>(at::TensorIteratorBase &, const T1 &)::[lambda(int) (instance 1)]>(int, T3)", "count": 1490, "total_ms": 65.249574, "mean_us": 43.79166040268457, "max_us": 127.744}, {"name": "void at::native::elementwise_kernel<(int)128, (int)2, void at::native::gpu_kernel_impl_nocast<at::native::BinaryFunctor<float, float, float, at::native::binary_internal::DivFunctor<float>>>(at::TensorIteratorBase &, const T1 &)::[lambda(int) (instance 1)]>(int, T3)", "count": 1480, "total_ms": 64.239183, "mean_us": 43.40485337837838, "max_us": 133.057}, {"name": "void cudnn::engines_precompiled::nhwcToNchwKernel<float, float, float, (bool)1, (bool)0, (cudnnKernelDataType_t)0>(cudnn::engines_precompiled::nhwc2nchw_params_t<T3>, const T1 *, T2 *)", "count": 1632, "total_ms": 47.092331, "mean_us": 28.855594975490195, "max_us": 82.912}, {"name": "sm90_xmma_fprop_implicit_gemm_f32f32_tf32f32_f32_nhwckrsc_nhwc_tilesize64x64x32_warpgroupsize1x1x1_g1_execute_segment_k_off_kernel__5x_cudnn", "count": 480, "total_ms": 45.808385, "mean_us": 95.43413541666668, "max_us": 112.769}, {"name": "void cutlass::device_kernel<flash::enable_sm90_or_later<flash::FlashAttnFwdSm90<flash::CollectiveMainloopFwdSm90<(int)2, cute::tuple<cute::C<(int)1>, cute::C<(int)1>, cute::C<(int)1>>, cute::tuple<cute::C<(int)128>, cute::C<(int)176>, cute::C<(int)128>>, (int)128, cutlass::bfloat16_t, float, cutlass::arch::Sm90, (bool)0, (bool)0, (bool)0, (bool)0, (bool)0, (bool)0, (bool)0, (bool)1, (bool)1, (bool)0, (bool)0, (bool)0>, flash::CollectiveEpilogueFwd<cute::tuple<cute::C<(int)128>, cute::C<(int)128>, cute::C<(int)176>>, cute::tuple<cute::C<(int)1>, cute::C<(int)1>, cute::C<(int)1>>, cutlass::bfloat16_t, cutlass::arch::Sm90, (int)256, (bool)0, (bool)0, (bool)0, (bool)0>, flash::StaticPersistentTileScheduler<(bool)0>>>>(T1::Params)", "count": 1458, "total_ms": 42.067651, "mean_us": 28.852984224965706, "max_us": 32.544}, {"name": "void at::native::vectorized_elementwise_kernel<(int)4, at::native::<unnamed>::silu_kernel(at::TensorIteratorBase &)::[lambda() (instance 1)]::operator ()() const::[lambda() (instance 2)]::operator ()() const::[lambda(float) (instance 1)], std::array<char *, (unsigned long)2>>(int, T2, T3)", "count": 1392, "total_ms": 38.899033, "mean_us": 27.94470761494253, "max_us": 75.488}, {"name": "void at::native::vectorized_elementwise_kernel<(int)4, at::native::AUnaryFunctor<float, float, float, at::native::binary_internal::MulFunctor<float>>, std::array<char *, (unsigned long)2>>(int, T2, T3)", "count": 1549, "total_ms": 37.983605, "mean_us": 24.521371852808265, "max_us": 72.992}, {"name": "void at::native::vectorized_elementwise_kernel<(int)4, at::native::CUDAFunctorOnSelf_add<float>, std::array<char *, (unsigned long)2>>(int, T2, T3)", "count": 1460, "total_ms": 37.832062, "mean_us": 25.912371232876712, "max_us": 73.344}], "top_count": [{"name": "void at::native::elementwise_kernel<(int)128, (int)2, void at::native::gpu_kernel_impl_nocast<at::native::direct_copy_kernel_cuda(at::TensorIteratorBase &)::[lambda() (instance 3)]::operator ()() const::[lambda() (instance 7)]::operator ()() const::[lambda(float) (instance 1)]>(at::TensorIteratorBase &, const T1 &)::[lambda(int) (instance 1)]>(int, T3)", "count": 13916, "total_ms": 384.311752, "mean_us": 27.616538660534637, "max_us": 230.688}, {"name": "nvjet_sm90_tst_128x200_64x5_2x1_v_bz_coopA_bias_TNT", "count": 13126, "total_ms": 1574.520988, "mean_us": 119.95436446746915, "max_us": 245.856}, {"name": "ncclDevKernel_SendRecv(ncclDevKernelArgsStorage<(unsigned long)4096>)", "count": 7420, "total_ms": 695.087986, "mean_us": 93.67762614555257, "max_us": 2432.649}, {"name": "void cudnn::engines_precompiled::nchwToNhwcKernel<float, float, float, (bool)0, (bool)1, (cudnnKernelDataType_t)0>(cudnn::engines_precompiled::nchw2nhwc_params_t<T3>, const T1 *, T2 *)", "count": 3168, "total_ms": 101.16054, "mean_us": 31.931988636363634, "max_us": 190.657}, {"name": "rotary_kernel", "count": 2922, "total_ms": 28.249024, "mean_us": 9.667701574264203, "max_us": 11.359}, {"name": "void at::native::index_elementwise_kernel<(int)128, (int)4, void at::native::gpu_index_kernel<void at::native::index_put_kernel_impl<at::native::OpaqueType<(int)2>>(at::TensorIterator &, c10::ArrayRef<long>, c10::ArrayRef<long>)::[lambda(char *, const char *, long) (instance 1)]>(at::TensorIteratorBase &, c10::ArrayRef<long>, c10::ArrayRef<long>, const T1 &, bool)::[lambda(int) (instance 1)]>(long, T3)", "count": 2921, "total_ms": 86.738054, "mean_us": 29.694643615200274, "max_us": 32.767}, {"name": "void at::native::vectorized_gather_kernel<(int)16, long>(char *, char *, T2 *, int, long, long, long, long, bool)", "count": 2918, "total_ms": 113.005016, "mean_us": 38.72687320082248, "max_us": 41.472}, {"name": "triton_poi_fused_clone_transpose_view_8", "count": 2917, "total_ms": 26.988606, "mean_us": 9.252178950977031, "max_us": 12.128}, {"name": "triton_poi_fused_all_to_all_single_clone_transpose_unsqueeze_view_5", "count": 2917, "total_ms": 19.443414, "mean_us": 6.665551594103531, "max_us": 7.743}, {"name": "void at::native::vectorized_elementwise_kernel<(int)4, at::native::FillFunctor<float>, std::array<char *, (unsigned long)1>>(int, T2, T3)", "count": 2477, "total_ms": 34.015324, "mean_us": 13.732468308437626, "max_us": 67.808}, {"name": "void at::native::elementwise_kernel<(int)128, (int)2, void at::native::gpu_kernel_impl_nocast<at::native::CUDAFunctor_add<float>>(at::TensorIteratorBase &, const T1 &)::[lambda(int) (instance 1)]>(int, T3)", "count": 2002, "total_ms": 81.826604, "mean_us": 40.872429570429574, "max_us": 142.081}, {"name": "void cudnn::engines_precompiled::nhwcToNchwKernel<float, float, float, (bool)1, (bool)0, (cudnnKernelDataType_t)0>(cudnn::engines_precompiled::nhwc2nchw_params_t<T3>, const T1 *, T2 *)", "count": 1632, "total_ms": 47.092331, "mean_us": 28.855594975490195, "max_us": 82.912}, {"name": "void at::native::vectorized_elementwise_kernel<(int)4, at::native::AUnaryFunctor<float, float, float, at::native::binary_internal::MulFunctor<float>>, std::array<char *, (unsigned long)2>>(int, T2, T3)", "count": 1549, "total_ms": 37.983605, "mean_us": 24.521371852808265, "max_us": 72.992}, {"name": "void at::native::elementwise_kernel<(int)128, (int)2, void at::native::gpu_kernel_impl_nocast<at::native::BinaryFunctor<float, float, float, at::native::binary_internal::MulFunctor<float>>>(at::TensorIteratorBase &, const T1 &)::[lambda(int) (instance 1)]>(int, T3)", "count": 1490, "total_ms": 65.249574, "mean_us": 43.79166040268457, "max_us": 127.744}, {"name": "void at::native::elementwise_kernel<(int)128, (int)2, void at::native::gpu_kernel_impl_nocast<at::native::BinaryFunctor<float, float, float, at::native::binary_internal::DivFunctor<float>>>(at::TensorIteratorBase &, const T1 &)::[lambda(int) (instance 1)]>(int, T3)", "count": 1480, "total_ms": 64.239183, "mean_us": 43.40485337837838, "max_us": 133.057}, {"name": "nvjet_sm90_tst_256x152_64x4_1x2_h_bz_coopA_bias_TNT", "count": 1461, "total_ms": 388.515463, "mean_us": 265.92434154688567, "max_us": 288.609}, {"name": "triton_red_fused__to_copy_add_mul_native_layer_norm_split_squeeze_1", "count": 1461, "total_ms": 32.613571, "mean_us": 22.32277275838467, "max_us": 25.697}, {"name": "triton_poi_fused_all_to_all_single_clone_split_with_sizes_stack_transpose_view_4", "count": 1461, "total_ms": 26.772754, "mean_us": 18.32495140314853, "max_us": 21.152}, {"name": "triton_red_fused__to_copy_div_mul_pow_split_with_sizes_sqrt_sum_view_3", "count": 1461, "total_ms": 15.504394, "mean_us": 10.612179329226558, "max_us": 12.511}, {"name": "triton_red_fused__to_copy_div_mul_pow_split_with_sizes_sqrt_sum_view_2", "count": 1461, "total_ms": 13.972858, "mean_us": 9.56390006844627, "max_us": 11.008}], "groups": 102, "devices": [{"device": 4, "window_ms": 3394.550197, "non_nccl_ms": 1258.743393, "nccl_ms": 144.691819, "memcpy_ms": 21.610715, "non_nccl_nccl_overlap_ms": 30.920963, "kernel_memcpy_overlap_ms": 0.0, "observed_busy_ms": 1394.124964, "no_observed_kernel_or_memcpy_ms": 2000.425233}, {"device": 5, "window_ms": 3394.550197, "non_nccl_ms": 1023.109935, "nccl_ms": 80.731259, "memcpy_ms": 12.684056, "non_nccl_nccl_overlap_ms": 16.238738, "kernel_memcpy_overlap_ms": 0.0, "observed_busy_ms": 1100.286512, "no_observed_kernel_or_memcpy_ms": 2294.263685}, {"device": 6, "window_ms": 3394.550197, "non_nccl_ms": 2683.18842, "nccl_ms": 402.558407, "memcpy_ms": 45.465136, "non_nccl_nccl_overlap_ms": 110.643956, "kernel_memcpy_overlap_ms": 0.0, "observed_busy_ms": 3020.568007, "no_observed_kernel_or_memcpy_ms": 373.98219}, {"device": 7, "window_ms": 3394.550197, "non_nccl_ms": 1016.739071, "nccl_ms": 90.169871, "memcpy_ms": 12.556835, "non_nccl_nccl_overlap_ms": 15.111894, "kernel_memcpy_overlap_ms": 0.0, "observed_busy_ms": 1104.353883, "no_observed_kernel_or_memcpy_ms": 2290.196314}], "copies": [{"direction": "Device-to-Host", "count": 28, "summed_gpu_ms": 4.231974, "bytes_full_overlapping_calls": 62300180}, {"direction": "Host-to-Device", "count": 1673, "summed_gpu_ms": 3.041646, "bytes_full_overlapping_calls": 20595952}, {"direction": "Device-to-Device", "count": 17742, "summed_gpu_ms": 85.043122, "bytes_full_overlapping_calls": 144137728112}], "wait_apis": [{"api": "cudaStreamSynchronize_v3020", "kind": "sync", "count": 239, "summed_thread_ms": 2951.60229}, {"api": "cudaDeviceSynchronize_v3020", "kind": "sync", "count": 37, "summed_thread_ms": 1488.981018}, {"api": "cudaStreamWaitEvent_v3020", "kind": "sync", "count": 9832, "summed_thread_ms": 11.522278}, {"api": "pthread_cond_timedwait", "kind": "os_wait", "count": 601, "summed_thread_ms": 113723.236231}, {"api": "poll", "kind": "os_wait", "count": 12251, "summed_thread_ms": 81433.30987}, {"api": "pthread_cond_wait", "kind": "os_wait", "count": 102, "summed_thread_ms": 74680.343125}, {"api": "epoll_wait", "kind": "os_wait", "count": 3466, "summed_thread_ms": 64478.46314}, {"api": "futex", "kind": "os_wait", "count": 12, "summed_thread_ms": 40734.602364}, {"api": "sem_wait", "kind": "os_wait", "count": 3, "summed_thread_ms": 10183.650591}, {"api": "sem_clockwait", "kind": "os_wait", "count": 3, "summed_thread_ms": 6788.901691}, {"api": "epoll_pwait", "kind": "os_wait", "count": 7, "summed_thread_ms": 3394.305995}], "overlap_interpretation": "Same-device overlapping activity intervals, not proof of simultaneous SM execution or net speedup. Non-NCCL includes copies/conversions implemented as kernels. Gaps are not attributed CPU waits. Thread times overlap and must not be added as wall time."}}}, "worker_flush_v2": {"source": "original_nsys_worker_flush_v2", "kernel_events": 1091749, "clock_mapping": {"systemClockNs": 2951076468893971, "formula": "trace_ns = round(host perf_counter_seconds * 1e9) - systemClockNs", "validation": "same-host monotonic alignment; windows and per-device activity retained for audit"}, "windows_ns": {"full": null, "measured": [[254697730877, 264553757105]], "steady": [[260193278338, 261040772109], [261040772109, 261893917185], [261893917185, 262747644781], [262747644781, 263603828424]]}, "scope_notes": "full includes initialization/warmup/validation/shutdown. measured is epoch1. steady is CPU-ready intervals5–8. Profile timings are not clean benchmark timings.", "scopes": {"full": {"top_time": [{"name": "nvjet_sm90_tst_128x200_64x5_2x1_v_bz_coopA_bias_TNT", "count": 144000, "total_ms": 16922.111948, "mean_us": 117.51466630555555, "max_us": 239.296}, {"name": "ncclDevKernel_SendRecv(ncclDevKernelArgsStorage<(unsigned long)4096>)", "count": 73504, "total_ms": 13495.557754, "mean_us": 183.6030386645625, "max_us": 1824356.15}, {"name": "void cutlass::device_kernel<flash::enable_sm90_or_later<flash::FlashAttnFwdSm90<flash::CollectiveMainloopFwdSm90<(int)2, cute::tuple<cute::C<(int)1>, cute::C<(int)1>, cute::C<(int)1>>, cute::tuple<cute::C<(int)128>, cute::C<(int)160>, cute::C<(int)128>>, (int)128, cutlass::bfloat16_t, float, cutlass::arch::Sm90, (bool)0, (bool)0, (bool)0, (bool)1, (bool)0, (bool)0, (bool)0, (bool)1, (bool)1, (bool)1, (bool)0, (bool)0, cutlass::bfloat16_t, (int)1>, flash::CollectiveEpilogueFwd<cute::tuple<cute::C<(int)128>, cute::C<(int)128>, cute::C<(int)160>>, cute::tuple<cute::C<(int)1>, cute::C<(int)1>, cute::C<(int)1>>, cutlass::bfloat16_t, cutlass::arch::Sm90, (int)256, (bool)1, (bool)1, (bool)0, (bool)0, (int)1>, flash::VarlenDynamicPersistentTileScheduler<(int)128, (int)160, (int)256, (int)128, (bool)0, (bool)1, (bool)1, (bool)0, (bool)0, (bool)1>>>>(T1::Params)", "count": 16000, "total_ms": 11811.459247, "mean_us": 738.2162029374999, "max_us": 1052.29}, {"name": "void at::native::elementwise_kernel<(int)128, (int)2, void at::native::gpu_kernel_impl_nocast<at::native::direct_copy_kernel_cuda(at::TensorIteratorBase &)::[lambda() (instance 3)]::operator ()() const::[lambda() (instance 7)]::operator ()() const::[lambda(float) (instance 1)]>(at::TensorIteratorBase &, const T1 &)::[lambda(int) (instance 1)]>(int, T3)", "count": 104597, "total_ms": 4871.552223, "mean_us": 46.5744927961605, "max_us": 710.306}, {"name": "nvjet_sm90_tst_256x152_64x4_1x2_h_bz_coopA_bias_TNT", "count": 16000, "total_ms": 4225.027781, "mean_us": 264.06423631250004, "max_us": 282.305}, {"name": "sm90_xmma_fprop_implicit_gemm_f32f32_tf32f32_f32_nhwckrsc_nhwc_tilesize256x32x32_warpgroupsize1x1x1_g1_execute_segment_k_off_kernel__5x_cudnn", "count": 1800, "total_ms": 2603.532357, "mean_us": 1446.406865, "max_us": 5604.43}, {"name": "sm80_xmma_fprop_implicit_gemm_bf16bf16_bf16f32_f32_nhwckrsc_nhwc_tilesize256x32x32_stage4_warpsize4x1x1_g1_tensor16x8x16_execute_kernel__5x_cudnn", "count": 960, "total_ms": 2532.205398, "mean_us": 2637.7139562499997, "max_us": 2751.172}, {"name": "sm90_xmma_fprop_implicit_gemm_f32f32_tf32f32_f32_nhwckrsc_nhwc_tilesize128x128x32_warpgroupsize1x1x1_g1_execute_segment_k_off_kernel__5x_cudnn", "count": 4059, "total_ms": 2095.366604, "mean_us": 516.2272983493472, "max_us": 1838.885}, {"name": "sm90_xmma_fprop_implicit_gemm_bf16bf16_bf16f32_f32_nhwckrsc_nhwc_tilesize256x128x64_warpgroupsize2x1x1_g1_execute_segment_k_off_kernel__5x_cudnn", "count": 3840, "total_ms": 1547.538683, "mean_us": 403.00486536458334, "max_us": 1223.617}, {"name": "void at::native::vectorized_gather_kernel<(int)16, long>(char *, char *, T2 *, int, long, long, long, long, bool)", "count": 32200, "total_ms": 1161.867362, "mean_us": 36.08283732919254, "max_us": 159.936}, {"name": "void at::native::elementwise_kernel<(int)128, (int)4, void at::native::gpu_kernel_impl<at::native::BinaryFunctor<float, float, float, at::native::binary_internal::DivFunctor<float>>>(at::TensorIteratorBase &, const T1 &)::[lambda(int) (instance 1)]>(int, T3)", "count": 5280, "total_ms": 1057.141018, "mean_us": 200.2161018939394, "max_us": 722.273}, {"name": "void at::native::index_elementwise_kernel<(int)128, (int)4, void at::native::gpu_index_kernel<void at::native::index_put_kernel_impl<at::native::OpaqueType<(int)2>>(at::TensorIterator &, c10::ArrayRef<long>, c10::ArrayRef<long>)::[lambda(char *, const char *, long) (instance 1)]>(at::TensorIteratorBase &, c10::ArrayRef<long>, c10::ArrayRef<long>, const T1 &, bool)::[lambda(int) (instance 1)]>(long, T3)", "count": 32000, "total_ms": 945.35033, "mean_us": 29.5421978125, "max_us": 32.223}, {"name": "void at::native::elementwise_kernel<(int)128, (int)4, void at::native::gpu_kernel_impl_nocast<at::native::CUDAFunctor_add<c10::BFloat16>>(at::TensorIteratorBase &, const T1 &)::[lambda(int) (instance 1)]>(int, T3)", "count": 7824, "total_ms": 903.798617, "mean_us": 115.51618315439673, "max_us": 420.064}, {"name": "void at::native::elementwise_kernel<(int)128, (int)4, void at::native::gpu_kernel_impl<at::native::BinaryFunctor<float, float, float, at::native::binary_internal::MulFunctor<float>>>(at::TensorIteratorBase &, const T1 &)::[lambda(int) (instance 1)]>(int, T3)", "count": 5672, "total_ms": 885.504321, "mean_us": 156.1185333215797, "max_us": 587.744}, {"name": "void cudnn::engines_precompiled::nchwToNhwcKernel<__nv_bfloat16, __nv_bfloat16, float, (bool)0, (bool)1, (cudnnKernelDataType_t)0>(cudnn::engines_precompiled::nchw2nhwc_params_t<T3>, const T1 *, T2 *)", "count": 12448, "total_ms": 825.820638, "mean_us": 66.34163223007712, "max_us": 434.88}, {"name": "void at::native::vectorized_elementwise_kernel<(int)4, at::native::<unnamed>::silu_kernel(at::TensorIteratorBase &)::[lambda() (instance 1)]::operator ()() const::[lambda() (instance 2)]::operator ()() const::[lambda(float) (instance 1)], std::array<char *, (unsigned long)2>>(int, T2, T3)", "count": 13740, "total_ms": 686.701668, "mean_us": 49.97828733624454, "max_us": 290.529}, {"name": "void at::native::vectorized_elementwise_kernel<(int)4, at::native::AUnaryFunctor<float, float, float, at::native::binary_internal::MulFunctor<float>>, std::array<char *, (unsigned long)2>>(int, T2, T3)", "count": 15612, "total_ms": 684.695224, "mean_us": 43.85698334614399, "max_us": 288.32}, {"name": "void cudnn::engines_precompiled::nchwToNhwcKernel<float, float, float, (bool)0, (bool)1, (cudnnKernelDataType_t)0>(cudnn::engines_precompiled::nchw2nhwc_params_t<T3>, const T1 *, T2 *)", "count": 19748, "total_ms": 678.500732, "mean_us": 34.357946728782665, "max_us": 730.402}, {"name": "fmha_cutlassF_f32_aligned_32x128_gmem_sm80(PyTorchMemEffAttention::AttentionKernel<float, cutlass::arch::Sm80, (bool)1, (int)32, (int)128, (int)65536, (bool)1, (bool)1>::Params)", "count": 300, "total_ms": 572.780363, "mean_us": 1909.2678766666668, "max_us": 2126.434}, {"name": "void at::native::elementwise_kernel<(int)128, (int)2, void at::native::gpu_kernel_impl_nocast<at::native::CUDAFunctor_add<float>>(at::TensorIteratorBase &, const T1 &)::[lambda(int) (instance 1)]>(int, T3)", "count": 12673, "total_ms": 561.28389, "mean_us": 44.28974118204056, "max_us": 547.617}], "top_count": [{"name": "nvjet_sm90_tst_128x200_64x5_2x1_v_bz_coopA_bias_TNT", "count": 144000, "total_ms": 16922.111948, "mean_us": 117.51466630555555, "max_us": 239.296}, {"name": "void at::native::elementwise_kernel<(int)128, (int)2, void at::native::gpu_kernel_impl_nocast<at::native::direct_copy_kernel_cuda(at::TensorIteratorBase &)::[lambda() (instance 3)]::operator ()() const::[lambda() (instance 7)]::operator ()() const::[lambda(float) (instance 1)]>(at::TensorIteratorBase &, const T1 &)::[lambda(int) (instance 1)]>(int, T3)", "count": 104597, "total_ms": 4871.552223, "mean_us": 46.5744927961605, "max_us": 710.306}, {"name": "ncclDevKernel_SendRecv(ncclDevKernelArgsStorage<(unsigned long)4096>)", "count": 73504, "total_ms": 13495.557754, "mean_us": 183.6030386645625, "max_us": 1824356.15}, {"name": "void at::native::vectorized_elementwise_kernel<(int)4, at::native::FillFunctor<int>, std::array<char *, (unsigned long)1>>(int, T2, T3)", "count": 42162, "total_ms": 392.347451, "mean_us": 9.305712513637873, "max_us": 16.832}, {"name": "triton_poi_fused_all_to_all_single_clone_transpose_unsqueeze_view_5", "count": 32848, "total_ms": 222.355154, "mean_us": 6.769214381393083, "max_us": 20.672}, {"name": "triton_poi_fused_clone_transpose_view_8", "count": 32848, "total_ms": 206.413014, "mean_us": 6.283883767657088, "max_us": 12.832}, {"name": "void at::native::vectorized_gather_kernel<(int)16, long>(char *, char *, T2 *, int, long, long, long, long, bool)", "count": 32200, "total_ms": 1161.867362, "mean_us": 36.08283732919254, "max_us": 159.936}, {"name": "void at::native::index_elementwise_kernel<(int)128, (int)4, void at::native::gpu_index_kernel<void at::native::index_put_kernel_impl<at::native::OpaqueType<(int)2>>(at::TensorIterator &, c10::ArrayRef<long>, c10::ArrayRef<long>)::[lambda(char *, const char *, long) (instance 1)]>(at::TensorIteratorBase &, c10::ArrayRef<long>, c10::ArrayRef<long>, const T1 &, bool)::[lambda(int) (instance 1)]>(long, T3)", "count": 32000, "total_ms": 945.35033, "mean_us": 29.5421978125, "max_us": 32.223}, {"name": "rotary_kernel", "count": 32000, "total_ms": 306.911107, "mean_us": 9.59097209375, "max_us": 11.136}, {"name": "void at::native::vectorized_elementwise_kernel<(int)4, at::native::FillFunctor<float>, std::array<char *, (unsigned long)1>>(int, T2, T3)", "count": 20607, "total_ms": 541.717623, "mean_us": 26.288039161449994, "max_us": 210.209}, {"name": "void cudnn::engines_precompiled::nchwToNhwcKernel<float, float, float, (bool)0, (bool)1, (cudnnKernelDataType_t)0>(cudnn::engines_precompiled::nchw2nhwc_params_t<T3>, const T1 *, T2 *)", "count": 19748, "total_ms": 678.500732, "mean_us": 34.357946728782665, "max_us": 730.402}, {"name": "triton_red_fused__to_copy_add_mul_native_layer_norm_split_squeeze_1", "count": 18226, "total_ms": 412.511204, "mean_us": 22.633117743882366, "max_us": 43.968}, {"name": "triton_red_fused__to_copy_div_mul_pow_split_with_sizes_sqrt_sum_view_3", "count": 17696, "total_ms": 188.477531, "mean_us": 10.65085505198915, "max_us": 13.056}, {"name": "triton_red_fused__to_copy_add_mul_native_layer_norm_split_squeeze_view_9", "count": 17484, "total_ms": 450.102377, "mean_us": 25.743672900937998, "max_us": 32.416}, {"name": "triton_red_fused__to_copy_div_mul_pow_split_with_sizes_sqrt_sum_view_2", "count": 17272, "total_ms": 164.226229, "mean_us": 9.508234657248726, "max_us": 11.392}, {"name": "triton_poi_fused_gelu_view_13", "count": 16848, "total_ms": 373.206334, "mean_us": 22.15137310066477, "max_us": 31.488}, {"name": "triton_poi_fused_all_to_all_single_clone_split_with_sizes_stack_transpose_view_4", "count": 16848, "total_ms": 317.054516, "mean_us": 18.818525403608735, "max_us": 38.912}, {"name": "triton_poi_fused__to_copy_add_mul_split_squeeze_view_14", "count": 16848, "total_ms": 257.641415, "mean_us": 15.292106778252611, "max_us": 17.92}, {"name": "triton_poi_fused_silu_view_6", "count": 16848, "total_ms": 185.931526, "mean_us": 11.035821818613485, "max_us": 24.031}, {"name": "triton_poi_fused__to_copy_all_to_all_single_clone_div_mul_sqrt_transpose_view_11", "count": 16848, "total_ms": 185.724899, "mean_us": 11.023557632953468, "max_us": 20.448}], "groups": 167, "devices": [{"device": 0, "window_ms": 178346.625206, "non_nccl_ms": 17909.850181, "nccl_ms": 5577.479321, "memcpy_ms": 7462.544521, "non_nccl_nccl_overlap_ms": 381.294084, "kernel_memcpy_overlap_ms": 0.0, "observed_busy_ms": 30568.579939, "no_observed_kernel_or_memcpy_ms": 147778.045267}, {"device": 1, "window_ms": 178346.625206, "non_nccl_ms": 16414.547681, "nccl_ms": 2849.319883, "memcpy_ms": 7609.287677, "non_nccl_nccl_overlap_ms": 388.151316, "kernel_memcpy_overlap_ms": 0.0, "observed_busy_ms": 26485.003925, "no_observed_kernel_or_memcpy_ms": 151861.621281}, {"device": 2, "window_ms": 178346.625206, "non_nccl_ms": 16477.70204, "nccl_ms": 2774.220395, "memcpy_ms": 7091.280374, "non_nccl_nccl_overlap_ms": 371.118388, "kernel_memcpy_overlap_ms": 0.0, "observed_busy_ms": 25972.084421, "no_observed_kernel_or_memcpy_ms": 152374.540785}, {"device": 3, "window_ms": 178346.625206, "non_nccl_ms": 16564.144238, "nccl_ms": 2533.416596, "memcpy_ms": 6835.135561, "non_nccl_nccl_overlap_ms": 306.802134, "kernel_memcpy_overlap_ms": 0.0, "observed_busy_ms": 25625.894261, "no_observed_kernel_or_memcpy_ms": 152720.730945}], "copies": [{"direction": "Host-to-Device", "count": 36162, "summed_gpu_ms": 27567.770097, "bytes_full_overlapping_calls": 197525080492}, {"direction": "Device-to-Device", "count": 195512, "summed_gpu_ms": 1009.04179, "bytes_full_overlapping_calls": 1724851179724}, {"direction": "Device-to-Host", "count": 677, "summed_gpu_ms": 421.436246, "bytes_full_overlapping_calls": 1984388632}], "wait_apis": [{"api": "cudaStreamSynchronize_v3020", "kind": "sync", "count": 13886, "summed_thread_ms": 27231.824078}, {"api": "cudaDeviceSynchronize_v3020", "kind": "sync", "count": 1732, "summed_thread_ms": 18826.586116}, {"api": "cudaStreamWaitEvent_v3020", "kind": "sync", "count": 66435, "summed_thread_ms": 88.399712}, {"api": "pthread_cond_timedwait", "kind": "os_wait", "count": 410631, "summed_thread_ms": 10762837.771767}, {"api": "epoll_wait", "kind": "os_wait", "count": 569762, "summed_thread_ms": 6201948.425167}, {"api": "poll", "kind": "os_wait", "count": 94392, "summed_thread_ms": 4919144.452237}, {"api": "pthread_cond_wait", "kind": "os_wait", "count": 622, "summed_thread_ms": 4141299.161341}, {"api": "futex", "kind": "os_wait", "count": 59, "summed_thread_ms": 1877931.799133}, {"api": "sem_clockwait", "kind": "os_wait", "count": 1225, "summed_thread_ms": 1221093.546362}, {"api": "sem_wait", "kind": "os_wait", "count": 971, "summed_thread_ms": 1001800.423264}, {"api": "epoll_pwait", "kind": "os_wait", "count": 1329, "summed_thread_ms": 267661.790608}, {"api": "waitpid", "kind": "os_wait", "count": 1110, "summed_thread_ms": 211691.960679}, {"api": "clock_nanosleep", "kind": "os_wait", "count": 565, "summed_thread_ms": 18801.178668}, {"api": "nanosleep", "kind": "os_wait", "count": 401, "summed_thread_ms": 5099.974}, {"api": "sleep", "kind": "os_wait", "count": 6300, "summed_thread_ms": 1241.855977}, {"api": "sem_timedwait", "kind": "os_wait", "count": 77, "summed_thread_ms": 2.225573}, {"api": "epoll_ctl", "kind": "os_wait", "count": 241, "summed_thread_ms": 0.419349}], "overlap_interpretation": "Same-device overlapping activity intervals, not proof of simultaneous SM execution or net speedup. Non-NCCL includes copies/conversions implemented as kernels. Gaps are not attributed CPU waits. Thread times overlap and must not be added as wall time."}, "measured": {"top_time": [{"name": "nvjet_sm90_tst_128x200_64x5_2x1_v_bz_coopA_bias_TNT", "count": 72000, "total_ms": 8494.219328, "mean_us": 117.97526844444445, "max_us": 239.296}, {"name": "void cutlass::device_kernel<flash::enable_sm90_or_later<flash::FlashAttnFwdSm90<flash::CollectiveMainloopFwdSm90<(int)2, cute::tuple<cute::C<(int)1>, cute::C<(int)1>, cute::C<(int)1>>, cute::tuple<cute::C<(int)128>, cute::C<(int)160>, cute::C<(int)128>>, (int)128, cutlass::bfloat16_t, float, cutlass::arch::Sm90, (bool)0, (bool)0, (bool)0, (bool)1, (bool)0, (bool)0, (bool)0, (bool)1, (bool)1, (bool)1, (bool)0, (bool)0, cutlass::bfloat16_t, (int)1>, flash::CollectiveEpilogueFwd<cute::tuple<cute::C<(int)128>, cute::C<(int)128>, cute::C<(int)160>>, cute::tuple<cute::C<(int)1>, cute::C<(int)1>, cute::C<(int)1>>, cutlass::bfloat16_t, cutlass::arch::Sm90, (int)256, (bool)1, (bool)1, (bool)0, (bool)0, (int)1>, flash::VarlenDynamicPersistentTileScheduler<(int)128, (int)160, (int)256, (int)128, (bool)0, (bool)1, (bool)1, (bool)0, (bool)0, (bool)1>>>>(T1::Params)", "count": 8000, "total_ms": 5928.947366, "mean_us": 741.11842075, "max_us": 1052.29}, {"name": "ncclDevKernel_SendRecv(ncclDevKernelArgsStorage<(unsigned long)4096>)", "count": 35960, "total_ms": 3026.823537, "mean_us": 84.1719559788654, "max_us": 8694.093}, {"name": "void at::native::elementwise_kernel<(int)128, (int)2, void at::native::gpu_kernel_impl_nocast<at::native::direct_copy_kernel_cuda(at::TensorIteratorBase &)::[lambda() (instance 3)]::operator ()() const::[lambda() (instance 7)]::operator ()() const::[lambda(float) (instance 1)]>(at::TensorIteratorBase &, const T1 &)::[lambda(int) (instance 1)]>(int, T3)", "count": 44790, "total_ms": 2156.718588, "mean_us": 48.15178807769591, "max_us": 704.193}, {"name": "nvjet_sm90_tst_256x152_64x4_1x2_h_bz_coopA_bias_TNT", "count": 8000, "total_ms": 2120.151663, "mean_us": 265.018957875, "max_us": 282.305}, {"name": "sm80_xmma_fprop_implicit_gemm_bf16bf16_bf16f32_f32_nhwckrsc_nhwc_tilesize256x32x32_stage4_warpsize4x1x1_g1_tensor16x8x16_execute_kernel__5x_cudnn", "count": 480, "total_ms": 1265.434178, "mean_us": 2636.3212041666666, "max_us": 2751.172}, {"name": "sm90_xmma_fprop_implicit_gemm_f32f32_tf32f32_f32_nhwckrsc_nhwc_tilesize256x32x32_warpgroupsize1x1x1_g1_execute_segment_k_off_kernel__5x_cudnn", "count": 720, "total_ms": 935.07911, "mean_us": 1298.720986111111, "max_us": 1367.139}, {"name": "sm90_xmma_fprop_implicit_gemm_f32f32_tf32f32_f32_nhwckrsc_nhwc_tilesize128x128x32_warpgroupsize1x1x1_g1_execute_segment_k_off_kernel__5x_cudnn", "count": 1652, "total_ms": 818.842472, "mean_us": 495.6673559322034, "max_us": 916.769}, {"name": "sm90_xmma_fprop_implicit_gemm_bf16bf16_bf16f32_f32_nhwckrsc_nhwc_tilesize256x128x64_warpgroupsize2x1x1_g1_execute_segment_k_off_kernel__5x_cudnn", "count": 1920, "total_ms": 774.386561, "mean_us": 403.32633385416665, "max_us": 1223.617}, {"name": "void at::native::vectorized_gather_kernel<(int)16, long>(char *, char *, T2 *, int, long, long, long, long, bool)", "count": 16100, "total_ms": 581.281909, "mean_us": 36.10446639751553, "max_us": 159.52}, {"name": "void at::native::elementwise_kernel<(int)128, (int)4, void at::native::gpu_kernel_impl<at::native::BinaryFunctor<float, float, float, at::native::binary_internal::DivFunctor<float>>>(at::TensorIteratorBase &, const T1 &)::[lambda(int) (instance 1)]>(int, T3)", "count": 2640, "total_ms": 528.57428, "mean_us": 200.21753030303032, "max_us": 722.273}, {"name": "void at::native::index_elementwise_kernel<(int)128, (int)4, void at::native::gpu_index_kernel<void at::native::index_put_kernel_impl<at::native::OpaqueType<(int)2>>(at::TensorIterator &, c10::ArrayRef<long>, c10::ArrayRef<long>)::[lambda(char *, const char *, long) (instance 1)]>(at::TensorIteratorBase &, c10::ArrayRef<long>, c10::ArrayRef<long>, const T1 &, bool)::[lambda(int) (instance 1)]>(long, T3)", "count": 16000, "total_ms": 474.70931, "mean_us": 29.669331875, "max_us": 32.223}, {"name": "void at::native::elementwise_kernel<(int)128, (int)4, void at::native::gpu_kernel_impl_nocast<at::native::CUDAFunctor_add<c10::BFloat16>>(at::TensorIteratorBase &, const T1 &)::[lambda(int) (instance 1)]>(int, T3)", "count": 3912, "total_ms": 451.890179, "mean_us": 115.51384943762781, "max_us": 420.064}, {"name": "void at::native::elementwise_kernel<(int)128, (int)4, void at::native::gpu_kernel_impl<at::native::BinaryFunctor<float, float, float, at::native::binary_internal::MulFunctor<float>>>(at::TensorIteratorBase &, const T1 &)::[lambda(int) (instance 1)]>(int, T3)", "count": 2836, "total_ms": 442.785817, "mean_us": 156.13040091678423, "max_us": 587.744}, {"name": "void cudnn::engines_precompiled::nchwToNhwcKernel<__nv_bfloat16, __nv_bfloat16, float, (bool)0, (bool)1, (cudnnKernelDataType_t)0>(cudnn::engines_precompiled::nchw2nhwc_params_t<T3>, const T1 *, T2 *)", "count": 6224, "total_ms": 413.084788, "mean_us": 66.36966388174807, "max_us": 433.537}, {"name": "void at::native::vectorized_elementwise_kernel<(int)4, at::native::<unnamed>::silu_kernel(at::TensorIteratorBase &)::[lambda() (instance 1)]::operator ()() const::[lambda() (instance 2)]::operator ()() const::[lambda(float) (instance 1)], std::array<char *, (unsigned long)2>>(int, T2, T3)", "count": 6000, "total_ms": 307.380406, "mean_us": 51.23006766666667, "max_us": 288.416}, {"name": "void at::native::vectorized_elementwise_kernel<(int)4, at::native::AUnaryFunctor<float, float, float, at::native::binary_internal::MulFunctor<float>>, std::array<char *, (unsigned long)2>>(int, T2, T3)", "count": 6906, "total_ms": 307.075133, "mean_us": 44.46497726614538, "max_us": 288.032}, {"name": "void at::native::vectorized_elementwise_kernel<(int)8, at::native::bfloat16_copy_kernel_cuda(at::TensorIteratorBase &)::[lambda(float) (instance 1)], std::array<char *, (unsigned long)2>>(int, T2, T3)", "count": 3936, "total_ms": 265.550438, "mean_us": 67.46708282520325, "max_us": 319.648}, {"name": "void cudnn::engines_precompiled::nchwToNhwcKernel<float, float, float, (bool)0, (bool)1, (cudnnKernelDataType_t)0>(cudnn::engines_precompiled::nchw2nhwc_params_t<T3>, const T1 *, T2 *)", "count": 7904, "total_ms": 248.748276, "mean_us": 31.471188765182184, "max_us": 191.68}, {"name": "void at::native::vectorized_elementwise_kernel<(int)4, at::native::FillFunctor<float>, std::array<char *, (unsigned long)1>>(int, T2, T3)", "count": 8916, "total_ms": 241.535998, "mean_us": 27.090174742036787, "max_us": 203.168}], "top_count": [{"name": "nvjet_sm90_tst_128x200_64x5_2x1_v_bz_coopA_bias_TNT", "count": 72000, "total_ms": 8494.219328, "mean_us": 117.97526844444445, "max_us": 239.296}, {"name": "void at::native::elementwise_kernel<(int)128, (int)2, void at::native::gpu_kernel_impl_nocast<at::native::direct_copy_kernel_cuda(at::TensorIteratorBase &)::[lambda() (instance 3)]::operator ()() const::[lambda() (instance 7)]::operator ()() const::[lambda(float) (instance 1)]>(at::TensorIteratorBase &, const T1 &)::[lambda(int) (instance 1)]>(int, T3)", "count": 44790, "total_ms": 2156.718588, "mean_us": 48.15178807769591, "max_us": 704.193}, {"name": "ncclDevKernel_SendRecv(ncclDevKernelArgsStorage<(unsigned long)4096>)", "count": 35960, "total_ms": 3026.823537, "mean_us": 84.1719559788654, "max_us": 8694.093}, {"name": "void at::native::vectorized_gather_kernel<(int)16, long>(char *, char *, T2 *, int, long, long, long, long, bool)", "count": 16100, "total_ms": 581.281909, "mean_us": 36.10446639751553, "max_us": 159.52}, {"name": "void at::native::index_elementwise_kernel<(int)128, (int)4, void at::native::gpu_index_kernel<void at::native::index_put_kernel_impl<at::native::OpaqueType<(int)2>>(at::TensorIterator &, c10::ArrayRef<long>, c10::ArrayRef<long>)::[lambda(char *, const char *, long) (instance 1)]>(at::TensorIteratorBase &, c10::ArrayRef<long>, c10::ArrayRef<long>, const T1 &, bool)::[lambda(int) (instance 1)]>(long, T3)", "count": 16000, "total_ms": 474.70931, "mean_us": 29.669331875, "max_us": 32.223}, {"name": "rotary_kernel", "count": 16000, "total_ms": 154.208682, "mean_us": 9.638042625, "max_us": 11.136}, {"name": "triton_poi_fused_all_to_all_single_clone_transpose_unsqueeze_view_5", "count": 16000, "total_ms": 105.932275, "mean_us": 6.620767187499999, "max_us": 7.68}, {"name": "triton_poi_fused_clone_transpose_view_8", "count": 16000, "total_ms": 99.076819, "mean_us": 6.1923011875, "max_us": 9.728}, {"name": "void at::native::vectorized_elementwise_kernel<(int)4, at::native::FillFunctor<float>, std::array<char *, (unsigned long)1>>(int, T2, T3)", "count": 8916, "total_ms": 241.535998, "mean_us": 27.090174742036787, "max_us": 203.168}, {"name": "void at::native::unrolled_elementwise_kernel<at::native::direct_copy_kernel_cuda(at::TensorIteratorBase &)::[lambda() (instance 3)]::operator ()() const::[lambda() (instance 4)]::operator ()() const::[lambda(long) (instance 1)], std::array<char *, (unsigned long)2>, (int)4, TrivialOffsetCalculator<(int)1, unsigned int>, TrivialOffsetCalculator<(int)1, unsigned int>, at::native::memory::LoadWithCast<(int)1>, at::native::memory::StoreWithCast<(int)1>>(int, T1, T2, T4, T5, T6, T7)", "count": 8192, "total_ms": 12.795394, "mean_us": 1.561937744140625, "max_us": 3.136}, {"name": "void cutlass::device_kernel<flash::enable_sm90_or_later<flash::FlashAttnFwdSm90<flash::CollectiveMainloopFwdSm90<(int)2, cute::tuple<cute::C<(int)1>, cute::C<(int)1>, cute::C<(int)1>>, cute::tuple<cute::C<(int)128>, cute::C<(int)160>, cute::C<(int)128>>, (int)128, cutlass::bfloat16_t, float, cutlass::arch::Sm90, (bool)0, (bool)0, (bool)0, (bool)1, (bool)0, (bool)0, (bool)0, (bool)1, (bool)1, (bool)1, (bool)0, (bool)0, cutlass::bfloat16_t, (int)1>, flash::CollectiveEpilogueFwd<cute::tuple<cute::C<(int)128>, cute::C<(int)128>, cute::C<(int)160>>, cute::tuple<cute::C<(int)1>, cute::C<(int)1>, cute::C<(int)1>>, cutlass::bfloat16_t, cutlass::arch::Sm90, (int)256, (bool)1, (bool)1, (bool)0, (bool)0, (int)1>, flash::VarlenDynamicPersistentTileScheduler<(int)128, (int)160, (int)256, (int)128, (bool)0, (bool)1, (bool)1, (bool)0, (bool)0, (bool)1>>>>(T1::Params)", "count": 8000, "total_ms": 5928.947366, "mean_us": 741.11842075, "max_us": 1052.29}, {"name": "nvjet_sm90_tst_256x152_64x4_1x2_h_bz_coopA_bias_TNT", "count": 8000, "total_ms": 2120.151663, "mean_us": 265.018957875, "max_us": 282.305}, {"name": "void cutlass::device_kernel<flash::enable_sm90_or_later<flash::FlashAttnFwdSm90<flash::CollectiveMainloopFwdSm90<(int)2, cute::tuple<cute::C<(int)1>, cute::C<(int)1>, cute::C<(int)1>>, cute::tuple<cute::C<(int)128>, cute::C<(int)176>, cute::C<(int)128>>, (int)128, cutlass::bfloat16_t, float, cutlass::arch::Sm90, (bool)0, (bool)0, (bool)0, (bool)0, (bool)0, (bool)0, (bool)0, (bool)1, (bool)1, (bool)0, (bool)0, (bool)0>, flash::CollectiveEpilogueFwd<cute::tuple<cute::C<(int)128>, cute::C<(int)128>, cute::C<(int)176>>, cute::tuple<cute::C<(int)1>, cute::C<(int)1>, cute::C<(int)1>>, cutlass::bfloat16_t, cutlass::arch::Sm90, (int)256, (bool)0, (bool)0, (bool)0, (bool)0>, flash::StaticPersistentTileScheduler<(bool)0>>>>(T1::Params)", "count": 8000, "total_ms": 230.433766, "mean_us": 28.80422075, "max_us": 31.521}, {"name": "triton_red_fused__to_copy_add_mul_native_layer_norm_split_squeeze_view_9", "count": 8000, "total_ms": 206.041148, "mean_us": 25.7551435, "max_us": 32.128}, {"name": "triton_red_fused__to_copy_add_mul_native_layer_norm_split_squeeze_view_12", "count": 8000, "total_ms": 183.419965, "mean_us": 22.927495625, "max_us": 26.048}, {"name": "triton_red_fused__to_copy_add_mul_native_layer_norm_split_squeeze_1", "count": 8000, "total_ms": 177.212277, "mean_us": 22.151534625, "max_us": 25.888}, {"name": "triton_poi_fused_gelu_view_13", "count": 8000, "total_ms": 176.494579, "mean_us": 22.061822375, "max_us": 24.128}, {"name": "triton_poi_fused_all_to_all_single_clone_split_with_sizes_stack_transpose_view_4", "count": 8000, "total_ms": 146.551541, "mean_us": 18.318942625, "max_us": 21.312}, {"name": "triton_poi_fused__to_copy_add_mul_split_squeeze_view_14", "count": 8000, "total_ms": 122.343321, "mean_us": 15.292915125, "max_us": 17.056}, {"name": "triton_poi_fused_silu_view_6", "count": 8000, "total_ms": 90.317354, "mean_us": 11.289669250000001, "max_us": 24.031}], "groups": 157, "devices": [{"device": 0, "window_ms": 9856.026228, "non_nccl_ms": 7911.87902, "nccl_ms": 814.307453, "memcpy_ms": 141.874317, "non_nccl_nccl_overlap_ms": 191.792462, "kernel_memcpy_overlap_ms": 0.0, "observed_busy_ms": 8676.268328, "no_observed_kernel_or_memcpy_ms": 1179.7579}, {"device": 1, "window_ms": 9856.026228, "non_nccl_ms": 7915.894253, "nccl_ms": 820.606019, "memcpy_ms": 123.673819, "non_nccl_nccl_overlap_ms": 191.183304, "kernel_memcpy_overlap_ms": 0.0, "observed_busy_ms": 8668.990787, "no_observed_kernel_or_memcpy_ms": 1187.035441}, {"device": 2, "window_ms": 9856.026228, "non_nccl_ms": 7943.501537, "nccl_ms": 781.562549, "memcpy_ms": 123.205881, "non_nccl_nccl_overlap_ms": 186.927715, "kernel_memcpy_overlap_ms": 0.0, "observed_busy_ms": 8661.342252, "no_observed_kernel_or_memcpy_ms": 1194.683976}, {"device": 3, "window_ms": 9856.026228, "non_nccl_ms": 7994.684075, "nccl_ms": 708.835106, "memcpy_ms": 124.026528, "non_nccl_nccl_overlap_ms": 153.306601, "kernel_memcpy_overlap_ms": 0.0, "observed_busy_ms": 8674.239108, "no_observed_kernel_or_memcpy_ms": 1181.78712}], "copies": [{"direction": "Host-to-Device", "count": 9100, "summed_gpu_ms": 17.718054, "bytes_full_overlapping_calls": 74838464}, {"direction": "Device-to-Device", "count": 94968, "summed_gpu_ms": 476.381478, "bytes_full_overlapping_calls": 814231992320}, {"direction": "Device-to-Host", "count": 109, "summed_gpu_ms": 18.681013, "bytes_full_overlapping_calls": 150958170}], "wait_apis": [{"api": "cudaStreamSynchronize_v3020", "kind": "sync", "count": 1098, "summed_thread_ms": 14120.414013}, {"api": "cudaDeviceSynchronize_v3020", "kind": "sync", "count": 120, "summed_thread_ms": 5659.135055}, {"api": "cudaStreamWaitEvent_v3020", "kind": "sync", "count": 24809, "summed_thread_ms": 32.311059}, {"api": "pthread_cond_timedwait", "kind": "os_wait", "count": 62437, "summed_thread_ms": 676334.954448}, {"api": "poll", "kind": "os_wait", "count": 37138, "summed_thread_ms": 425808.320358}, {"api": "epoll_wait", "kind": "os_wait", "count": 43697, "summed_thread_ms": 422744.387744}, {"api": "pthread_cond_wait", "kind": "os_wait", "count": 260, "summed_thread_ms": 268719.684758}, {"api": "sem_clockwait", "kind": "os_wait", "count": 126, "summed_thread_ms": 90658.209275}, {"api": "futex", "kind": "os_wait", "count": 9, "summed_thread_ms": 88704.236052}, {"api": "sem_wait", "kind": "os_wait", "count": 37, "summed_thread_ms": 65356.447611}, {"api": "epoll_pwait", "kind": "os_wait", "count": 41, "summed_thread_ms": 9853.924634}, {"api": "epoll_ctl", "kind": "os_wait", "count": 46, "summed_thread_ms": 0.079941}], "overlap_interpretation": "Same-device overlapping activity intervals, not proof of simultaneous SM execution or net speedup. Non-NCCL includes copies/conversions implemented as kernels. Gaps are not attributed CPU waits. Thread times overlap and must not be added as wall time."}, "steady": {"top_time": [{"name": "nvjet_sm90_tst_128x200_64x5_2x1_v_bz_coopA_bias_TNT", "count": 28800, "total_ms": 3432.218463, "mean_us": 119.17425218749999, "max_us": 238.56}, {"name": "void cutlass::device_kernel<flash::enable_sm90_or_later<flash::FlashAttnFwdSm90<flash::CollectiveMainloopFwdSm90<(int)2, cute::tuple<cute::C<(int)1>, cute::C<(int)1>, cute::C<(int)1>>, cute::tuple<cute::C<(int)128>, cute::C<(int)160>, cute::C<(int)128>>, (int)128, cutlass::bfloat16_t, float, cutlass::arch::Sm90, (bool)0, (bool)0, (bool)0, (bool)1, (bool)0, (bool)0, (bool)0, (bool)1, (bool)1, (bool)1, (bool)0, (bool)0, cutlass::bfloat16_t, (int)1>, flash::CollectiveEpilogueFwd<cute::tuple<cute::C<(int)128>, cute::C<(int)128>, cute::C<(int)160>>, cute::tuple<cute::C<(int)1>, cute::C<(int)1>, cute::C<(int)1>>, cutlass::bfloat16_t, cutlass::arch::Sm90, (int)256, (bool)1, (bool)1, (bool)0, (bool)0, (int)1>, flash::VarlenDynamicPersistentTileScheduler<(int)128, (int)160, (int)256, (int)128, (bool)0, (bool)1, (bool)1, (bool)0, (bool)0, (bool)1>>>>(T1::Params)", "count": 3200, "total_ms": 3149.761347, "mean_us": 984.3004209375, "max_us": 1051.298}, {"name": "ncclDevKernel_SendRecv(ncclDevKernelArgsStorage<(unsigned long)4096>)", "count": 14384, "total_ms": 1232.356857, "mean_us": 85.6755323275862, "max_us": 2412.423}, {"name": "nvjet_sm90_tst_256x152_64x4_1x2_h_bz_coopA_bias_TNT", "count": 3200, "total_ms": 853.086899, "mean_us": 266.5896559375, "max_us": 282.305}, {"name": "void at::native::elementwise_kernel<(int)128, (int)2, void at::native::gpu_kernel_impl_nocast<at::native::direct_copy_kernel_cuda(at::TensorIteratorBase &)::[lambda() (instance 3)]::operator ()() const::[lambda() (instance 7)]::operator ()() const::[lambda(float) (instance 1)]>(at::TensorIteratorBase &, const T1 &)::[lambda(int) (instance 1)]>(int, T3)", "count": 13972, "total_ms": 384.421003, "mean_us": 27.51367041225308, "max_us": 234.112}, {"name": "sm90_xmma_fprop_implicit_gemm_f32f32_tf32f32_f32_nhwckrsc_nhwc_tilesize256x32x32_warpgroupsize1x1x1_g1_execute_segment_k_off_kernel__5x_cudnn", "count": 288, "total_ms": 383.406434, "mean_us": 1331.2723402777779, "max_us": 1361.925}, {"name": "sm90_xmma_fprop_implicit_gemm_f32f32_tf32f32_f32_nhwckrsc_nhwc_tilesize128x128x32_warpgroupsize1x1x1_g1_execute_segment_k_off_kernel__5x_cudnn", "count": 672, "total_ms": 337.359273, "mean_us": 502.0227276785714, "max_us": 910.689}, {"name": "void at::native::vectorized_gather_kernel<(int)16, long>(char *, char *, T2 *, int, long, long, long, long, bool)", "count": 6400, "total_ms": 247.727709, "mean_us": 38.70745453125, "max_us": 41.664}, {"name": "void at::native::index_elementwise_kernel<(int)128, (int)4, void at::native::gpu_index_kernel<void at::native::index_put_kernel_impl<at::native::OpaqueType<(int)2>>(at::TensorIterator &, c10::ArrayRef<long>, c10::ArrayRef<long>)::[lambda(char *, const char *, long) (instance 1)]>(at::TensorIteratorBase &, c10::ArrayRef<long>, c10::ArrayRef<long>, const T1 &, bool)::[lambda(int) (instance 1)]>(long, T3)", "count": 6400, "total_ms": 191.355589, "mean_us": 29.899310781249998, "max_us": 32.064}, {"name": "void cudnn::engines_precompiled::nchwToNhwcKernel<float, float, float, (bool)0, (bool)1, (cudnnKernelDataType_t)0>(cudnn::engines_precompiled::nchw2nhwc_params_t<T3>, const T1 *, T2 *)", "count": 3168, "total_ms": 101.056065, "mean_us": 31.899010416666666, "max_us": 191.68}, {"name": "void cutlass::device_kernel<flash::enable_sm90_or_later<flash::FlashAttnFwdSm90<flash::CollectiveMainloopFwdSm90<(int)2, cute::tuple<cute::C<(int)1>, cute::C<(int)1>, cute::C<(int)1>>, cute::tuple<cute::C<(int)128>, cute::C<(int)176>, cute::C<(int)128>>, (int)128, cutlass::bfloat16_t, float, cutlass::arch::Sm90, (bool)0, (bool)0, (bool)0, (bool)0, (bool)0, (bool)0, (bool)0, (bool)1, (bool)1, (bool)0, (bool)0, (bool)0>, flash::CollectiveEpilogueFwd<cute::tuple<cute::C<(int)128>, cute::C<(int)128>, cute::C<(int)176>>, cute::tuple<cute::C<(int)1>, cute::C<(int)1>, cute::C<(int)1>>, cutlass::bfloat16_t, cutlass::arch::Sm90, (int)256, (bool)0, (bool)0, (bool)0, (bool)0>, flash::StaticPersistentTileScheduler<(bool)0>>>>(T1::Params)", "count": 3200, "total_ms": 93.039508, "mean_us": 29.07484625, "max_us": 31.521}, {"name": "fmha_cutlassF_f32_aligned_32x128_gmem_sm80(PyTorchMemEffAttention::AttentionKernel<float, cutlass::arch::Sm80, (bool)1, (int)32, (int)128, (int)65536, (bool)1, (bool)1>::Params)", "count": 48, "total_ms": 92.675485, "mean_us": 1930.7392708333332, "max_us": 2104.74}, {"name": "void at::native::elementwise_kernel<(int)128, (int)2, void at::native::gpu_kernel_impl_nocast<at::native::CUDAFunctor_add<float>>(at::TensorIteratorBase &, const T1 &)::[lambda(int) (instance 1)]>(int, T3)", "count": 2064, "total_ms": 85.579279, "mean_us": 41.462828972868216, "max_us": 143.969}, {"name": "triton_red_fused__to_copy_add_mul_native_layer_norm_split_squeeze_view_9", "count": 3200, "total_ms": 82.64547, "mean_us": 25.826709375, "max_us": 31.872}, {"name": "triton_red_fused__to_copy_add_mul_native_layer_norm_split_squeeze_view_12", "count": 3200, "total_ms": 73.792145, "mean_us": 23.060045312499998, "max_us": 25.696}, {"name": "triton_red_fused__to_copy_add_mul_native_layer_norm_split_squeeze_1", "count": 3200, "total_ms": 71.370039, "mean_us": 22.3031371875, "max_us": 25.632}, {"name": "triton_poi_fused_gelu_view_13", "count": 3200, "total_ms": 71.156884, "mean_us": 22.23652625, "max_us": 24.128}, {"name": "void at::native::elementwise_kernel<(int)128, (int)2, void at::native::gpu_kernel_impl_nocast<at::native::BinaryFunctor<float, float, float, at::native::binary_internal::MulFunctor<float>>>(at::TensorIteratorBase &, const T1 &)::[lambda(int) (instance 1)]>(int, T3)", "count": 1536, "total_ms": 68.972207, "mean_us": 44.903780598958335, "max_us": 129.057}, {"name": "void at::native::elementwise_kernel<(int)128, (int)2, void at::native::gpu_kernel_impl_nocast<at::native::BinaryFunctor<float, float, float, at::native::binary_internal::DivFunctor<float>>>(at::TensorIteratorBase &, const T1 &)::[lambda(int) (instance 1)]>(int, T3)", "count": 1520, "total_ms": 64.480655, "mean_us": 42.42148355263158, "max_us": 135.424}, {"name": "rotary_kernel", "count": 6400, "total_ms": 62.212757, "mean_us": 9.72074328125, "max_us": 10.976}], "top_count": [{"name": "nvjet_sm90_tst_128x200_64x5_2x1_v_bz_coopA_bias_TNT", "count": 28800, "total_ms": 3432.218463, "mean_us": 119.17425218749999, "max_us": 238.56}, {"name": "ncclDevKernel_SendRecv(ncclDevKernelArgsStorage<(unsigned long)4096>)", "count": 14384, "total_ms": 1232.356857, "mean_us": 85.6755323275862, "max_us": 2412.423}, {"name": "void at::native::elementwise_kernel<(int)128, (int)2, void at::native::gpu_kernel_impl_nocast<at::native::direct_copy_kernel_cuda(at::TensorIteratorBase &)::[lambda() (instance 3)]::operator ()() const::[lambda() (instance 7)]::operator ()() const::[lambda(float) (instance 1)]>(at::TensorIteratorBase &, const T1 &)::[lambda(int) (instance 1)]>(int, T3)", "count": 13972, "total_ms": 384.421003, "mean_us": 27.51367041225308, "max_us": 234.112}, {"name": "void at::native::vectorized_gather_kernel<(int)16, long>(char *, char *, T2 *, int, long, long, long, long, bool)", "count": 6400, "total_ms": 247.727709, "mean_us": 38.70745453125, "max_us": 41.664}, {"name": "void at::native::index_elementwise_kernel<(int)128, (int)4, void at::native::gpu_index_kernel<void at::native::index_put_kernel_impl<at::native::OpaqueType<(int)2>>(at::TensorIterator &, c10::ArrayRef<long>, c10::ArrayRef<long>)::[lambda(char *, const char *, long) (instance 1)]>(at::TensorIteratorBase &, c10::ArrayRef<long>, c10::ArrayRef<long>, const T1 &, bool)::[lambda(int) (instance 1)]>(long, T3)", "count": 6400, "total_ms": 191.355589, "mean_us": 29.899310781249998, "max_us": 32.064}, {"name": "rotary_kernel", "count": 6400, "total_ms": 62.212757, "mean_us": 9.72074328125, "max_us": 10.976}, {"name": "triton_poi_fused_all_to_all_single_clone_transpose_unsqueeze_view_5", "count": 6400, "total_ms": 42.839876, "mean_us": 6.693730625, "max_us": 7.648}, {"name": "triton_poi_fused_clone_transpose_view_8", "count": 6400, "total_ms": 39.675898, "mean_us": 6.1993590625, "max_us": 9.728}, {"name": "void cutlass::device_kernel<flash::enable_sm90_or_later<flash::FlashAttnFwdSm90<flash::CollectiveMainloopFwdSm90<(int)2, cute::tuple<cute::C<(int)1>, cute::C<(int)1>, cute::C<(int)1>>, cute::tuple<cute::C<(int)128>, cute::C<(int)160>, cute::C<(int)128>>, (int)128, cutlass::bfloat16_t, float, cutlass::arch::Sm90, (bool)0, (bool)0, (bool)0, (bool)1, (bool)0, (bool)0, (bool)0, (bool)1, (bool)1, (bool)1, (bool)0, (bool)0, cutlass::bfloat16_t, (int)1>, flash::CollectiveEpilogueFwd<cute::tuple<cute::C<(int)128>, cute::C<(int)128>, cute::C<(int)160>>, cute::tuple<cute::C<(int)1>, cute::C<(int)1>, cute::C<(int)1>>, cutlass::bfloat16_t, cutlass::arch::Sm90, (int)256, (bool)1, (bool)1, (bool)0, (bool)0, (int)1>, flash::VarlenDynamicPersistentTileScheduler<(int)128, (int)160, (int)256, (int)128, (bool)0, (bool)1, (bool)1, (bool)0, (bool)0, (bool)1>>>>(T1::Params)", "count": 3200, "total_ms": 3149.761347, "mean_us": 984.3004209375, "max_us": 1051.298}, {"name": "nvjet_sm90_tst_256x152_64x4_1x2_h_bz_coopA_bias_TNT", "count": 3200, "total_ms": 853.086899, "mean_us": 266.5896559375, "max_us": 282.305}, {"name": "void cutlass::device_kernel<flash::enable_sm90_or_later<flash::FlashAttnFwdSm90<flash::CollectiveMainloopFwdSm90<(int)2, cute::tuple<cute::C<(int)1>, cute::C<(int)1>, cute::C<(int)1>>, cute::tuple<cute::C<(int)128>, cute::C<(int)176>, cute::C<(int)128>>, (int)128, cutlass::bfloat16_t, float, cutlass::arch::Sm90, (bool)0, (bool)0, (bool)0, (bool)0, (bool)0, (bool)0, (bool)0, (bool)1, (bool)1, (bool)0, (bool)0, (bool)0>, flash::CollectiveEpilogueFwd<cute::tuple<cute::C<(int)128>, cute::C<(int)128>, cute::C<(int)176>>, cute::tuple<cute::C<(int)1>, cute::C<(int)1>, cute::C<(int)1>>, cutlass::bfloat16_t, cutlass::arch::Sm90, (int)256, (bool)0, (bool)0, (bool)0, (bool)0>, flash::StaticPersistentTileScheduler<(bool)0>>>>(T1::Params)", "count": 3200, "total_ms": 93.039508, "mean_us": 29.07484625, "max_us": 31.521}, {"name": "triton_red_fused__to_copy_add_mul_native_layer_norm_split_squeeze_view_9", "count": 3200, "total_ms": 82.64547, "mean_us": 25.826709375, "max_us": 31.872}, {"name": "triton_red_fused__to_copy_add_mul_native_layer_norm_split_squeeze_view_12", "count": 3200, "total_ms": 73.792145, "mean_us": 23.060045312499998, "max_us": 25.696}, {"name": "triton_red_fused__to_copy_add_mul_native_layer_norm_split_squeeze_1", "count": 3200, "total_ms": 71.370039, "mean_us": 22.3031371875, "max_us": 25.632}, {"name": "triton_poi_fused_gelu_view_13", "count": 3200, "total_ms": 71.156884, "mean_us": 22.23652625, "max_us": 24.128}, {"name": "triton_poi_fused_all_to_all_single_clone_split_with_sizes_stack_transpose_view_4", "count": 3200, "total_ms": 58.729221, "mean_us": 18.3528815625, "max_us": 21.312}, {"name": "triton_poi_fused__to_copy_add_mul_split_squeeze_view_14", "count": 3200, "total_ms": 49.037935, "mean_us": 15.3243546875, "max_us": 16.832}, {"name": "triton_poi_fused_silu_view_6", "count": 3200, "total_ms": 36.534415, "mean_us": 11.4170046875, "max_us": 24.031}, {"name": "triton_poi_fused__to_copy_all_to_all_single_clone_div_mul_sqrt_transpose_view_11", "count": 3200, "total_ms": 35.005508, "mean_us": 10.939221250000001, "max_us": 12.288}, {"name": "triton_red_fused__to_copy_div_mul_pow_split_with_sizes_sqrt_sum_view_3", "count": 3200, "total_ms": 34.654719, "mean_us": 10.8295996875, "max_us": 12.672}], "groups": 102, "devices": [{"device": 0, "window_ms": 3410.550086, "non_nccl_ms": 2685.684333, "nccl_ms": 339.034096, "memcpy_ms": 53.001604, "non_nccl_nccl_overlap_ms": 82.001168, "kernel_memcpy_overlap_ms": 0.0, "observed_busy_ms": 2995.718865, "no_observed_kernel_or_memcpy_ms": 414.831221}, {"device": 1, "window_ms": 3410.550086, "non_nccl_ms": 2684.934826, "nccl_ms": 328.385994, "memcpy_ms": 45.710873, "non_nccl_nccl_overlap_ms": 82.180448, "kernel_memcpy_overlap_ms": 0.0, "observed_busy_ms": 2976.851245, "no_observed_kernel_or_memcpy_ms": 433.698841}, {"device": 2, "window_ms": 3410.550086, "non_nccl_ms": 2698.229647, "nccl_ms": 322.07389, "memcpy_ms": 45.569764, "non_nccl_nccl_overlap_ms": 79.487262, "kernel_memcpy_overlap_ms": 0.0, "observed_busy_ms": 2986.386039, "no_observed_kernel_or_memcpy_ms": 424.164047}, {"device": 3, "window_ms": 3410.550086, "non_nccl_ms": 2720.954962, "nccl_ms": 277.652996, "memcpy_ms": 45.837074, "non_nccl_nccl_overlap_ms": 61.816397, "kernel_memcpy_overlap_ms": 0.0, "observed_busy_ms": 2982.628635, "no_observed_kernel_or_memcpy_ms": 427.921451}], "copies": [{"direction": "Device-to-Host", "count": 44, "summed_gpu_ms": 7.486263, "bytes_full_overlapping_calls": 62300196}, {"direction": "Host-to-Device", "count": 3632, "summed_gpu_ms": 6.331538, "bytes_full_overlapping_calls": 22254464}, {"direction": "Device-to-Device", "count": 37008, "summed_gpu_ms": 176.301514, "bytes_full_overlapping_calls": 295678976000}], "wait_apis": [{"api": "cudaStreamSynchronize_v3020", "kind": "sync", "count": 432, "summed_thread_ms": 6147.609279}, {"api": "cudaDeviceSynchronize_v3020", "kind": "sync", "count": 48, "summed_thread_ms": 2463.959061}, {"api": "cudaStreamWaitEvent_v3020", "kind": "sync", "count": 9924, "summed_thread_ms": 12.985831}, {"api": "pthread_cond_timedwait", "kind": "os_wait", "count": 24673, "summed_thread_ms": 233739.853063}, {"api": "poll", "kind": "os_wait", "count": 14801, "summed_thread_ms": 146598.263782}, {"api": "epoll_wait", "kind": "os_wait", "count": 15155, "summed_thread_ms": 146265.653371}, {"api": "pthread_cond_wait", "kind": "os_wait", "count": 124, "summed_thread_ms": 95495.638356}, {"api": "sem_clockwait", "kind": "os_wait", "count": 49, "summed_thread_ms": 31497.721894}, {"api": "futex", "kind": "os_wait", "count": 9, "summed_thread_ms": 30694.950774}, {"api": "sem_wait", "kind": "os_wait", "count": 17, "summed_thread_ms": 22379.143418}, {"api": "epoll_pwait", "kind": "os_wait", "count": 15, "summed_thread_ms": 3409.833616}, {"api": "epoll_ctl", "kind": "os_wait", "count": 19, "summed_thread_ms": 0.031797}], "overlap_interpretation": "Same-device overlapping activity intervals, not proof of simultaneous SM execution or net speedup. Non-NCCL includes copies/conversions implemented as kernels. Gaps are not attributed CPU waits. Thread times overlap and must not be added as wall time."}}}}, "attention": [{"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_dit_only_704x1280_tp1_sp1_v2/captures/qkv_rank0_003f18b0d1c40f62.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 9, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "vllm_fa2", "status": "measured", "q_shape": [10560, 40, 128], "k_shape": [512, 40, 128], "q_len": 10560, "k_len": 512, "reference_error": {"max_abs": 0.00048828125, "rmse": 1.896029815795992e-06, "relative_l2": 0.0002471550743179307, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 6.37440089465216e-05, "rmse": 5.101539129336812e-06, "relative_l2": 0.0018193710309880842, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.32681599259376526, "p95_ms": 0.32771520018577577, "p99_ms": 0.32822784334421157, "mean_ms": 0.3269034662842751, "n": 300}, {"repeat": 1, "p50_ms": 0.3245120048522949, "p95_ms": 0.3256959915161133, "p99_ms": 0.3266240060329437, "mean_ms": 0.32514058709144594, "n": 300}, {"repeat": 2, "p50_ms": 0.32444798946380615, "p95_ms": 0.3253760039806366, "p99_ms": 0.3257628867030144, "mean_ms": 0.3250464004278183, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_003f18b0d1c40f62_vllm_fa2.json", "configuration": "704x1280_tp1_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_dit_only_704x1280_tp1_sp1_v2", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_dit_only_704x1280_tp1_sp1_v2/captures/qkv_rank0_003f18b0d1c40f62.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 9, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "vllm_fa3", "status": "measured", "q_shape": [10560, 40, 128], "k_shape": [512, 40, 128], "q_len": 10560, "k_len": 512, "reference_error": {"max_abs": 0.000244140625, "rmse": 7.112279759947695e-07, "relative_l2": 9.271141297436885e-05, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 6.37440089465216e-05, "rmse": 5.160676778595407e-06, "relative_l2": 0.0018404613966942096, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.26419198513031006, "p95_ms": 0.2656959891319275, "p99_ms": 0.2662425607442856, "mean_ms": 0.26457002719243367, "n": 300}, {"repeat": 1, "p50_ms": 0.2646400034427643, "p95_ms": 0.2659216120839119, "p99_ms": 0.2668191903829575, "mean_ms": 0.26516714721918105, "n": 300}, {"repeat": 2, "p50_ms": 0.26422399282455444, "p95_ms": 0.2654415875673294, "p99_ms": 0.2661775854229927, "mean_ms": 0.2647998934984207, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_003f18b0d1c40f62_vllm_fa3.json", "configuration": "704x1280_tp1_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_dit_only_704x1280_tp1_sp1_v2", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_dit_only_704x1280_tp1_sp1_v2/captures/qkv_rank0_003f18b0d1c40f62.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 9, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_math", "status": "measured", "q_shape": [10560, 40, 128], "k_shape": [512, 40, 128], "q_len": 10560, "k_len": 512, "reference_error": {"max_abs": 0.00048828125, "rmse": 4.682157496897501e-06, "relative_l2": 0.0006103379675114096, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 6.09328142618451e-05, "rmse": 4.570944062662757e-06, "relative_l2": 0.0016301439626430338, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 4.6535680294036865, "p95_ms": 4.659971094131469, "p99_ms": 4.66140986919403, "mean_ms": 4.653405432701111, "n": 300}, {"repeat": 1, "p50_ms": 4.654208183288574, "p95_ms": 4.659428644180298, "p99_ms": 4.662913494110107, "mean_ms": 4.6541989215215045, "n": 300}, {"repeat": 2, "p50_ms": 4.6545281410217285, "p95_ms": 4.659585642814636, "p99_ms": 4.6611234569549564, "mean_ms": 4.6545550839106244, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_003f18b0d1c40f62_sdpa_math.json", "configuration": "704x1280_tp1_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_dit_only_704x1280_tp1_sp1_v2", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_dit_only_704x1280_tp1_sp1_v2/captures/qkv_rank0_003f18b0d1c40f62.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 9, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_flash", "status": "measured", "q_shape": [10560, 40, 128], "k_shape": [512, 40, 128], "q_len": 10560, "k_len": 512, "reference_error": {"max_abs": 0.00048828125, "rmse": 1.8961924711144778e-06, "relative_l2": 0.00024717627708963487, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 6.37440089465216e-05, "rmse": 5.101539129336812e-06, "relative_l2": 0.0018193710309880842, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.34676799178123474, "p95_ms": 0.34784161001443864, "p99_ms": 0.34884225100278854, "mean_ms": 0.3469243735074997, "n": 300}, {"repeat": 1, "p50_ms": 0.3468160033226013, "p95_ms": 0.34777759611606596, "p99_ms": 0.3481622385978699, "mean_ms": 0.34707882593075434, "n": 300}, {"repeat": 2, "p50_ms": 0.34672001004219055, "p95_ms": 0.347680002450943, "p99_ms": 0.34825631201267243, "mean_ms": 0.3469708795348803, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_003f18b0d1c40f62_sdpa_flash.json", "configuration": "704x1280_tp1_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_dit_only_704x1280_tp1_sp1_v2", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_dit_only_704x1280_tp1_sp1_v2/captures/qkv_rank0_003f18b0d1c40f62.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 9, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_efficient", "status": "measured", "q_shape": [10560, 40, 128], "k_shape": [512, 40, 128], "q_len": 10560, "k_len": 512, "reference_error": {"max_abs": 0.00048828125, "rmse": 3.3147722424217164e-06, "relative_l2": 0.00043209382737412807, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 6.37440089465216e-05, "rmse": 5.786214176812004e-06, "relative_l2": 0.0020635479186754463, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.6606720089912415, "p95_ms": 0.6615999937057495, "p99_ms": 0.6636563289165497, "mean_ms": 0.6608664526542027, "n": 300}, {"repeat": 1, "p50_ms": 0.6617599725723267, "p95_ms": 0.6634608209133148, "p99_ms": 0.6641603332757949, "mean_ms": 0.6621464534600576, "n": 300}, {"repeat": 2, "p50_ms": 0.6606079936027527, "p95_ms": 0.6615695863962173, "p99_ms": 0.6621139174699784, "mean_ms": 0.6608883194128672, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_003f18b0d1c40f62_sdpa_efficient.json", "configuration": "704x1280_tp1_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_dit_only_704x1280_tp1_sp1_v2", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_dit_only_704x1280_tp1_sp1_v2/captures/qkv_rank0_003f18b0d1c40f62.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 9, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_cudnn", "status": "measured", "q_shape": [10560, 40, 128], "k_shape": [512, 40, 128], "q_len": 10560, "k_len": 512, "reference_error": {"max_abs": 0.00048828125, "rmse": 3.3132782459397346e-06, "relative_l2": 0.00043189907895379274, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 6.37440089465216e-05, "rmse": 5.786232531806798e-06, "relative_l2": 0.002063554464650093, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.1947679966688156, "p95_ms": 0.1963520050048828, "p99_ms": 0.19689792171120643, "mean_ms": 0.19517045309146244, "n": 300}, {"repeat": 1, "p50_ms": 0.1940159946680069, "p95_ms": 0.19491200149059296, "p99_ms": 0.19536064490675925, "mean_ms": 0.19445866669217746, "n": 300}, {"repeat": 2, "p50_ms": 0.1939840018749237, "p95_ms": 0.19484800100326538, "p99_ms": 0.19555264636874198, "mean_ms": 0.19444992005825043, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_003f18b0d1c40f62_sdpa_cudnn.json", "configuration": "704x1280_tp1_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_dit_only_704x1280_tp1_sp1_v2", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_dit_only_704x1280_tp1_sp1_v2/captures/qkv_rank0_003f18b0d1c40f62.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 9, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "native_dense", "status": "measured", "q_shape": [10560, 40, 128], "k_shape": [512, 40, 128], "q_len": 10560, "k_len": 512, "reference_error": {"max_abs": 0.0, "rmse": 0.0, "relative_l2": 0.0, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 6.37440089465216e-05, "rmse": 5.166876854269409e-06, "relative_l2": 0.0018426725407794024, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.18723200261592865, "p95_ms": 0.18819359540939332, "p99_ms": 0.18857695803046226, "mean_ms": 0.18761525303125381, "n": 300}, {"repeat": 1, "p50_ms": 0.1870719939470291, "p95_ms": 0.1881919950246811, "p99_ms": 0.1894780847430229, "mean_ms": 0.18776767988999685, "n": 300}, {"repeat": 2, "p50_ms": 0.18729600310325623, "p95_ms": 0.18825919553637505, "p99_ms": 0.1892169627547264, "mean_ms": 0.188309013346831, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_003f18b0d1c40f62_native_dense.json", "configuration": "704x1280_tp1_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_dit_only_704x1280_tp1_sp1_v2", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_dit_only_704x1280_tp1_sp1_v2/captures/qkv_rank0_146bc622ce0024e2.pt", "context": {"phase": "commit", "step": null, "start_frame": 27, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "vllm_fa2", "status": "measured", "q_shape": [10560, 40, 128], "k_shape": [512, 40, 128], "q_len": 10560, "k_len": 512, "reference_error": {"max_abs": 0.00048828125, "rmse": 1.5469812223031836e-06, "relative_l2": 0.00020593358604271422, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 6.248058869558204e-05, "rmse": 3.99686897246697e-06, "relative_l2": 0.0017776198736656203, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.32688000798225403, "p95_ms": 0.32793599367141724, "p99_ms": 0.32844863772392274, "mean_ms": 0.32708426743745805, "n": 300}, {"repeat": 1, "p50_ms": 0.3248000144958496, "p95_ms": 0.3279680013656616, "p99_ms": 0.32880031287670136, "mean_ms": 0.3257795202732086, "n": 300}, {"repeat": 2, "p50_ms": 0.3242560029029846, "p95_ms": 0.32521759569644926, "p99_ms": 0.32611328214406965, "mean_ms": 0.3246695460875829, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_146bc622ce0024e2_vllm_fa2.json", "configuration": "704x1280_tp1_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_dit_only_704x1280_tp1_sp1_v2", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_dit_only_704x1280_tp1_sp1_v2/captures/qkv_rank0_146bc622ce0024e2.pt", "context": {"phase": "commit", "step": null, "start_frame": 27, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "vllm_fa3", "status": "measured", "q_shape": [10560, 40, 128], "k_shape": [512, 40, 128], "q_len": 10560, "k_len": 512, "reference_error": {"max_abs": 0.000244140625, "rmse": 5.919636576743952e-07, "relative_l2": 7.880199001404455e-05, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 6.248058869558204e-05, "rmse": 4.060999388811001e-06, "relative_l2": 0.001806142075265148, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.26659199595451355, "p95_ms": 0.271620811522007, "p99_ms": 0.2727459329366684, "mean_ms": 0.2675592537720998, "n": 300}, {"repeat": 1, "p50_ms": 0.26471999287605286, "p95_ms": 0.2702464088797569, "p99_ms": 0.2712959945201874, "mean_ms": 0.266061332821846, "n": 300}, {"repeat": 2, "p50_ms": 0.2645919919013977, "p95_ms": 0.2695695966482162, "p99_ms": 0.2713283222913742, "mean_ms": 0.2655516799290975, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_146bc622ce0024e2_vllm_fa3.json", "configuration": "704x1280_tp1_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_dit_only_704x1280_tp1_sp1_v2", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_dit_only_704x1280_tp1_sp1_v2/captures/qkv_rank0_146bc622ce0024e2.pt", "context": {"phase": "commit", "step": null, "start_frame": 27, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_math", "status": "measured", "q_shape": [10560, 40, 128], "k_shape": [512, 40, 128], "q_len": 10560, "k_len": 512, "reference_error": {"max_abs": 0.0009765625, "rmse": 3.642738970299405e-06, "relative_l2": 0.00048492010656369874, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 5.958972380441796e-05, "rmse": 3.7838496823589655e-06, "relative_l2": 0.0016828788836109959, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 4.653791904449463, "p95_ms": 4.659846496582031, "p99_ms": 4.661568803787231, "mean_ms": 4.653881171544393, "n": 300}, {"repeat": 1, "p50_ms": 4.653984069824219, "p95_ms": 4.65925452709198, "p99_ms": 4.66361713886261, "mean_ms": 4.653801698684692, "n": 300}, {"repeat": 2, "p50_ms": 4.653840065002441, "p95_ms": 4.660160064697266, "p99_ms": 4.662083010673523, "mean_ms": 4.654205972353617, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_146bc622ce0024e2_sdpa_math.json", "configuration": "704x1280_tp1_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_dit_only_704x1280_tp1_sp1_v2", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_dit_only_704x1280_tp1_sp1_v2/captures/qkv_rank0_146bc622ce0024e2.pt", "context": {"phase": "commit", "step": null, "start_frame": 27, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_flash", "status": "measured", "q_shape": [10560, 40, 128], "k_shape": [512, 40, 128], "q_len": 10560, "k_len": 512, "reference_error": {"max_abs": 0.00048828125, "rmse": 1.5472345544791572e-06, "relative_l2": 0.00020596730953121307, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 6.248058869558204e-05, "rmse": 3.99686897246697e-06, "relative_l2": 0.0017776198736656203, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.34643200039863586, "p95_ms": 0.34729599952697754, "p99_ms": 0.34844832807779313, "mean_ms": 0.3465621332327525, "n": 300}, {"repeat": 1, "p50_ms": 0.3463680148124695, "p95_ms": 0.34729599952697754, "p99_ms": 0.34791039675474167, "mean_ms": 0.3467328010996183, "n": 300}, {"repeat": 2, "p50_ms": 0.3466239869594574, "p95_ms": 0.3478128030896187, "p99_ms": 0.3488332912325859, "mean_ms": 0.3469196793437004, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_146bc622ce0024e2_sdpa_flash.json", "configuration": "704x1280_tp1_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_dit_only_704x1280_tp1_sp1_v2", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_dit_only_704x1280_tp1_sp1_v2/captures/qkv_rank0_146bc622ce0024e2.pt", "context": {"phase": "commit", "step": null, "start_frame": 27, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_efficient", "status": "measured", "q_shape": [10560, 40, 128], "k_shape": [512, 40, 128], "q_len": 10560, "k_len": 512, "reference_error": {"max_abs": 0.00048828125, "rmse": 2.7076855079471806e-06, "relative_l2": 0.00036044612467711634, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 6.248058869558204e-05, "rmse": 4.753777167533007e-06, "relative_l2": 0.0021142571413264835, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.6605119705200195, "p95_ms": 0.6615359783172607, "p99_ms": 0.6621772927045823, "mean_ms": 0.6607244811455408, "n": 300}, {"repeat": 1, "p50_ms": 0.6605439782142639, "p95_ms": 0.6616320013999939, "p99_ms": 0.662081909775734, "mean_ms": 0.6609844273328781, "n": 300}, {"repeat": 2, "p50_ms": 0.6614400148391724, "p95_ms": 0.6632384270429611, "p99_ms": 0.6637759804725647, "mean_ms": 0.662027518749237, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_146bc622ce0024e2_sdpa_efficient.json", "configuration": "704x1280_tp1_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_dit_only_704x1280_tp1_sp1_v2", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_dit_only_704x1280_tp1_sp1_v2/captures/qkv_rank0_146bc622ce0024e2.pt", "context": {"phase": "commit", "step": null, "start_frame": 27, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_cudnn", "status": "measured", "q_shape": [10560, 40, 128], "k_shape": [512, 40, 128], "q_len": 10560, "k_len": 512, "reference_error": {"max_abs": 0.00048828125, "rmse": 2.705855465444674e-06, "relative_l2": 0.00036020251007487134, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 6.248058869558204e-05, "rmse": 4.753777266973011e-06, "relative_l2": 0.0021142571855527334, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.194240003824234, "p95_ms": 0.19526559710502625, "p99_ms": 0.19584192112088203, "mean_ms": 0.19466346641381582, "n": 300}, {"repeat": 1, "p50_ms": 0.19464000314474106, "p95_ms": 0.19549760594964027, "p99_ms": 0.1958745539188385, "mean_ms": 0.19521653334299724, "n": 300}, {"repeat": 2, "p50_ms": 0.19385600090026855, "p95_ms": 0.19500799477100372, "p99_ms": 0.19616000354290009, "mean_ms": 0.19473909283677737, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_146bc622ce0024e2_sdpa_cudnn.json", "configuration": "704x1280_tp1_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_dit_only_704x1280_tp1_sp1_v2", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_dit_only_704x1280_tp1_sp1_v2/captures/qkv_rank0_146bc622ce0024e2.pt", "context": {"phase": "commit", "step": null, "start_frame": 27, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "native_dense", "status": "measured", "q_shape": [10560, 40, 128], "k_shape": [512, 40, 128], "q_len": 10560, "k_len": 512, "reference_error": {"max_abs": 0.0, "rmse": 0.0, "relative_l2": 0.0, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 6.248058869558204e-05, "rmse": 4.076969633579108e-06, "relative_l2": 0.0018132448911649572, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.18571200221776962, "p95_ms": 0.18672319874167442, "p99_ms": 0.18710464164614676, "mean_ms": 0.18608682706952095, "n": 300}, {"repeat": 1, "p50_ms": 0.18681600689888, "p95_ms": 0.18873760625720024, "p99_ms": 0.18944351717829705, "mean_ms": 0.18752384006977083, "n": 300}, {"repeat": 2, "p50_ms": 0.18848000466823578, "p95_ms": 0.18944159746170045, "p99_ms": 0.19014816239476204, "mean_ms": 0.18927434722582498, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_146bc622ce0024e2_native_dense.json", "configuration": "704x1280_tp1_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_dit_only_704x1280_tp1_sp1_v2", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_dit_only_704x1280_tp1_sp1_v2/captures/qkv_rank0_15833641ab5dc2f0.pt", "context": {"phase": "commit", "step": null, "start_frame": 0, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "vllm_fa2", "status": "measured", "q_shape": [10560, 40, 128], "k_shape": [66880, 40, 128], "q_len": 10560, "k_len": 10560, "reference_error": {"max_abs": 0.03125, "rmse": 0.0003762186107970482, "relative_l2": 0.0005248836805985898, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.008178553063441285, "rmse": 0.0007568670538157323, "relative_l2": 0.001639986913325226, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 5.949167966842651, "p95_ms": 6.292315340042115, "p99_ms": 6.6228173637390135, "mean_ms": 6.003804896672567, "n": 300}, {"repeat": 1, "p50_ms": 5.9988319873809814, "p95_ms": 6.163175964355469, "p99_ms": 6.2175055646896356, "mean_ms": 6.021709980964661, "n": 300}, {"repeat": 2, "p50_ms": 5.999279975891113, "p95_ms": 6.157631874084473, "p99_ms": 6.160897107124328, "mean_ms": 6.015049165089925, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_15833641ab5dc2f0_vllm_fa2.json", "configuration": "704x1280_tp1_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_dit_only_704x1280_tp1_sp1_v2", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_dit_only_704x1280_tp1_sp1_v2/captures/qkv_rank0_15833641ab5dc2f0.pt", "context": {"phase": "commit", "step": null, "start_frame": 0, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "vllm_fa3", "status": "measured", "q_shape": [10560, 40, 128], "k_shape": [66880, 40, 128], "q_len": 10560, "k_len": 10560, "reference_error": {"max_abs": 0.0, "rmse": 0.0, "relative_l2": 0.0, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.008178553063441285, "rmse": 0.0007573742732439552, "relative_l2": 0.0016410859613288037, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 3.426640033721924, "p95_ms": 3.6025503635406495, "p99_ms": 3.7225679302215577, "mean_ms": 3.4441406893730164, "n": 300}, {"repeat": 1, "p50_ms": 3.4268959760665894, "p95_ms": 3.4904352188110352, "p99_ms": 3.4930207252502443, "mean_ms": 3.433674867947896, "n": 300}, {"repeat": 2, "p50_ms": 3.4263360500335693, "p95_ms": 3.458817648887634, "p99_ms": 3.461185624599457, "mean_ms": 3.423536747296651, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_15833641ab5dc2f0_vllm_fa3.json", "configuration": "704x1280_tp1_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_dit_only_704x1280_tp1_sp1_v2", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_dit_only_704x1280_tp1_sp1_v2/captures/qkv_rank0_15833641ab5dc2f0.pt", "context": {"phase": "commit", "step": null, "start_frame": 0, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_math", "status": "measured", "q_shape": [10560, 40, 128], "k_shape": [66880, 40, 128], "q_len": 10560, "k_len": 10560, "reference_error": {"max_abs": 0.0625, "rmse": 0.0006849708508225278, "relative_l2": 0.0009556412441181042, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.007719305713610236, "rmse": 0.0007465987096451862, "relative_l2": 0.0016177373650375622, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 88.24224090576172, "p95_ms": 88.39567031860352, "p99_ms": 88.43388839721679, "mean_ms": 88.23730499267577, "n": 300}, {"repeat": 1, "p50_ms": 88.24846267700195, "p95_ms": 88.38026313781738, "p99_ms": 88.4175919342041, "mean_ms": 88.24143575032552, "n": 300}, {"repeat": 2, "p50_ms": 88.2425765991211, "p95_ms": 88.36799812316895, "p99_ms": 88.4325814819336, "mean_ms": 88.23356267293295, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_15833641ab5dc2f0_sdpa_math.json", "configuration": "704x1280_tp1_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_dit_only_704x1280_tp1_sp1_v2", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_dit_only_704x1280_tp1_sp1_v2/captures/qkv_rank0_15833641ab5dc2f0.pt", "context": {"phase": "commit", "step": null, "start_frame": 0, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_flash", "status": "measured", "q_shape": [10560, 40, 128], "k_shape": [66880, 40, 128], "q_len": 10560, "k_len": 10560, "reference_error": {"max_abs": 0.03125, "rmse": 0.00037672846466290034, "relative_l2": 0.0005255950063171898, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.008178553063441285, "rmse": 0.0007568670538157323, "relative_l2": 0.001639986913325226, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 6.5004801750183105, "p95_ms": 6.852884650230408, "p99_ms": 6.978721146583557, "mean_ms": 6.543526191711425, "n": 300}, {"repeat": 1, "p50_ms": 6.501760005950928, "p95_ms": 6.678880214691162, "p99_ms": 6.791842956542968, "mean_ms": 6.530882245699565, "n": 300}, {"repeat": 2, "p50_ms": 6.50220799446106, "p95_ms": 6.6358413934707645, "p99_ms": 6.672968444824218, "mean_ms": 6.534068698883057, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_15833641ab5dc2f0_sdpa_flash.json", "configuration": "704x1280_tp1_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_dit_only_704x1280_tp1_sp1_v2", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_dit_only_704x1280_tp1_sp1_v2/captures/qkv_rank0_15833641ab5dc2f0.pt", "context": {"phase": "commit", "step": null, "start_frame": 0, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_efficient", "status": "measured", "q_shape": [10560, 40, 128], "k_shape": [66880, 40, 128], "q_len": 10560, "k_len": 10560, "reference_error": {"max_abs": 0.03125, "rmse": 0.0006093949787574122, "relative_l2": 0.0008502011070394384, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.007912222877458852, "rmse": 0.000753810530580441, "relative_l2": 0.0016333640089711773, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 12.5458083152771, "p95_ms": 12.746878576278686, "p99_ms": 12.808735094070435, "mean_ms": 12.559364805221557, "n": 300}, {"repeat": 1, "p50_ms": 12.538800239562988, "p95_ms": 12.739156532287598, "p99_ms": 12.750859651565552, "mean_ms": 12.54854621887207, "n": 300}, {"repeat": 2, "p50_ms": 12.53382396697998, "p95_ms": 12.567487716674805, "p99_ms": 12.575346431732179, "mean_ms": 12.537116247812907, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_15833641ab5dc2f0_sdpa_efficient.json", "configuration": "704x1280_tp1_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_dit_only_704x1280_tp1_sp1_v2", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_dit_only_704x1280_tp1_sp1_v2/captures/qkv_rank0_15833641ab5dc2f0.pt", "context": {"phase": "commit", "step": null, "start_frame": 0, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_cudnn", "status": "measured", "q_shape": [10560, 40, 128], "k_shape": [66880, 40, 128], "q_len": 10560, "k_len": 10560, "reference_error": {"max_abs": 0.03125, "rmse": 0.0006068988808677422, "relative_l2": 0.0008467186609034306, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.007912222877458852, "rmse": 0.0007538871835807566, "relative_l2": 0.0016335301014397962, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 3.992032051086426, "p95_ms": 4.354535961151123, "p99_ms": 4.739808082580566, "mean_ms": 4.040498775641123, "n": 300}, {"repeat": 1, "p50_ms": 3.99398410320282, "p95_ms": 4.220135998725891, "p99_ms": 4.267619957923889, "mean_ms": 4.031067742506663, "n": 300}, {"repeat": 2, "p50_ms": 4.02894401550293, "p95_ms": 4.145425748825073, "p99_ms": 4.310080370903015, "mean_ms": 4.025533762772878, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_15833641ab5dc2f0_sdpa_cudnn.json", "configuration": "704x1280_tp1_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_dit_only_704x1280_tp1_sp1_v2", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_dit_only_704x1280_tp1_sp1_v2/captures/qkv_rank0_15833641ab5dc2f0.pt", "context": {"phase": "commit", "step": null, "start_frame": 0, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "native_dense", "status": "measured", "q_shape": [10560, 40, 128], "k_shape": [66880, 40, 128], "q_len": 10560, "k_len": 10560, "reference_error": {"max_abs": 0.03125, "rmse": 0.0003575225369779525, "relative_l2": 0.0004987997396204449, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.008178553063441285, "rmse": 0.0007583933289138638, "relative_l2": 0.0016432940610923942, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 3.395680069923401, "p95_ms": 3.5830400466918944, "p99_ms": 3.6625145363807676, "mean_ms": 3.399575463136037, "n": 300}, {"repeat": 1, "p50_ms": 3.3973439931869507, "p95_ms": 3.4951823711395265, "p99_ms": 3.533782296180725, "mean_ms": 3.388911254405975, "n": 300}, {"repeat": 2, "p50_ms": 3.408959984779358, "p95_ms": 3.492696022987366, "p99_ms": 3.581829183101654, "mean_ms": 3.3910914103190106, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_15833641ab5dc2f0_native_dense.json", "configuration": "704x1280_tp1_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_dit_only_704x1280_tp1_sp1_v2", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_dit_only_704x1280_tp1_sp1_v2/captures/qkv_rank0_227f18fd0026d1aa.pt", "context": {"phase": "commit", "step": null, "start_frame": 9, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "vllm_fa2", "status": "measured", "q_shape": [10560, 40, 128], "k_shape": [512, 40, 128], "q_len": 10560, "k_len": 512, "reference_error": {"max_abs": 0.00048828125, "rmse": 1.5529998868913357e-06, "relative_l2": 0.00020546925532088684, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 6.307664314882458e-05, "rmse": 4.137486204813533e-06, "relative_l2": 0.00178708309216853, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.32422399520874023, "p95_ms": 0.3251855880022049, "p99_ms": 0.3257283192873001, "mean_ms": 0.3244105593363444, "n": 300}, {"repeat": 1, "p50_ms": 0.32438400387763977, "p95_ms": 0.3253439962863922, "p99_ms": 0.32579233437776567, "mean_ms": 0.32479061305522916, "n": 300}, {"repeat": 2, "p50_ms": 0.3243519961833954, "p95_ms": 0.32528320997953414, "p99_ms": 0.3258576002717018, "mean_ms": 0.3247380272547404, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_227f18fd0026d1aa_vllm_fa2.json", "configuration": "704x1280_tp1_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_dit_only_704x1280_tp1_sp1_v2", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_dit_only_704x1280_tp1_sp1_v2/captures/qkv_rank0_227f18fd0026d1aa.pt", "context": {"phase": "commit", "step": null, "start_frame": 9, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "vllm_fa3", "status": "measured", "q_shape": [10560, 40, 128], "k_shape": [512, 40, 128], "q_len": 10560, "k_len": 512, "reference_error": {"max_abs": 0.000244140625, "rmse": 5.855249762798006e-07, "relative_l2": 7.746773316823079e-05, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 6.307664314882458e-05, "rmse": 4.161822254469216e-06, "relative_l2": 0.0017975944366702345, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.2646079957485199, "p95_ms": 0.26915838569402695, "p99_ms": 0.27126686692237856, "mean_ms": 0.2653903989990552, "n": 300}, {"repeat": 1, "p50_ms": 0.26423999667167664, "p95_ms": 0.2692544087767601, "p99_ms": 0.27043423533439637, "mean_ms": 0.2652497070034345, "n": 300}, {"repeat": 2, "p50_ms": 0.2645759880542755, "p95_ms": 0.2692879870533943, "p99_ms": 0.271652469933033, "mean_ms": 0.2655080538988113, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_227f18fd0026d1aa_vllm_fa3.json", "configuration": "704x1280_tp1_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_dit_only_704x1280_tp1_sp1_v2", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_dit_only_704x1280_tp1_sp1_v2/captures/qkv_rank0_227f18fd0026d1aa.pt", "context": {"phase": "commit", "step": null, "start_frame": 9, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_math", "status": "measured", "q_shape": [10560, 40, 128], "k_shape": [512, 40, 128], "q_len": 10560, "k_len": 512, "reference_error": {"max_abs": 0.001953125, "rmse": 3.94899944506398e-06, "relative_l2": 0.0005224713678917771, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 6.095633961755026e-05, "rmse": 3.932673083918934e-06, "relative_l2": 0.001698619216451148, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 4.651872158050537, "p95_ms": 4.65718412399292, "p99_ms": 4.661158390045166, "mean_ms": 4.651537912686666, "n": 300}, {"repeat": 1, "p50_ms": 4.6519200801849365, "p95_ms": 4.6576334714889525, "p99_ms": 4.661601710319519, "mean_ms": 4.652023239135742, "n": 300}, {"repeat": 2, "p50_ms": 4.652144193649292, "p95_ms": 4.65773298740387, "p99_ms": 4.66170331954956, "mean_ms": 4.652112631797791, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_227f18fd0026d1aa_sdpa_math.json", "configuration": "704x1280_tp1_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_dit_only_704x1280_tp1_sp1_v2", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_dit_only_704x1280_tp1_sp1_v2/captures/qkv_rank0_227f18fd0026d1aa.pt", "context": {"phase": "commit", "step": null, "start_frame": 9, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_flash", "status": "measured", "q_shape": [10560, 40, 128], "k_shape": [512, 40, 128], "q_len": 10560, "k_len": 512, "reference_error": {"max_abs": 0.00048828125, "rmse": 1.5530705346192146e-06, "relative_l2": 0.00020547860235056788, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 6.307664314882458e-05, "rmse": 4.137486204813533e-06, "relative_l2": 0.00178708309216853, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.34641599655151367, "p95_ms": 0.3476159870624542, "p99_ms": 0.3480342376232147, "mean_ms": 0.34683103998502096, "n": 300}, {"repeat": 1, "p50_ms": 0.34649598598480225, "p95_ms": 0.3474879860877991, "p99_ms": 0.3480976128578186, "mean_ms": 0.3467878386378288, "n": 300}, {"repeat": 2, "p50_ms": 0.3463680148124695, "p95_ms": 0.3474256008863449, "p99_ms": 0.3479360044002533, "mean_ms": 0.34676042705774307, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_227f18fd0026d1aa_sdpa_flash.json", "configuration": "704x1280_tp1_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_dit_only_704x1280_tp1_sp1_v2", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_dit_only_704x1280_tp1_sp1_v2/captures/qkv_rank0_227f18fd0026d1aa.pt", "context": {"phase": "commit", "step": null, "start_frame": 9, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_efficient", "status": "measured", "q_shape": [10560, 40, 128], "k_shape": [512, 40, 128], "q_len": 10560, "k_len": 512, "reference_error": {"max_abs": 0.0009765625, "rmse": 2.7253691683191186e-06, "relative_l2": 0.0003605792751279293, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 6.307664314882458e-05, "rmse": 4.736439880333926e-06, "relative_l2": 0.0020457860662761937, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.659855991601944, "p95_ms": 0.6614416152238846, "p99_ms": 0.6620214146375656, "mean_ms": 0.6601703464984894, "n": 300}, {"repeat": 1, "p50_ms": 0.660319983959198, "p95_ms": 0.6617296218872071, "p99_ms": 0.6621452850103379, "mean_ms": 0.6606067206462224, "n": 300}, {"repeat": 2, "p50_ms": 0.659775972366333, "p95_ms": 0.6607040166854858, "p99_ms": 0.6611868929862976, "mean_ms": 0.6600628264745076, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_227f18fd0026d1aa_sdpa_efficient.json", "configuration": "704x1280_tp1_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_dit_only_704x1280_tp1_sp1_v2", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_dit_only_704x1280_tp1_sp1_v2/captures/qkv_rank0_227f18fd0026d1aa.pt", "context": {"phase": "commit", "step": null, "start_frame": 9, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_cudnn", "status": "measured", "q_shape": [10560, 40, 128], "k_shape": [512, 40, 128], "q_len": 10560, "k_len": 512, "reference_error": {"max_abs": 0.00048828125, "rmse": 2.7206857835417723e-06, "relative_l2": 0.0003599596410953031, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 6.307664314882458e-05, "rmse": 4.736452756624415e-06, "relative_l2": 0.002045791627865982, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.19542400538921356, "p95_ms": 0.1963231973350048, "p99_ms": 0.19708800315856934, "mean_ms": 0.19565781419475872, "n": 300}, {"repeat": 1, "p50_ms": 0.19528000056743622, "p95_ms": 0.19648000597953796, "p99_ms": 0.1971222360432148, "mean_ms": 0.19521312018235523, "n": 300}, {"repeat": 2, "p50_ms": 0.1969279944896698, "p95_ms": 0.1976960003376007, "p99_ms": 0.19798431515693665, "mean_ms": 0.19725130622585615, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_227f18fd0026d1aa_sdpa_cudnn.json", "configuration": "704x1280_tp1_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_dit_only_704x1280_tp1_sp1_v2", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_dit_only_704x1280_tp1_sp1_v2/captures/qkv_rank0_227f18fd0026d1aa.pt", "context": {"phase": "commit", "step": null, "start_frame": 9, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "native_dense", "status": "measured", "q_shape": [10560, 40, 128], "k_shape": [512, 40, 128], "q_len": 10560, "k_len": 512, "reference_error": {"max_abs": 0.0, "rmse": 0.0, "relative_l2": 0.0, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 6.307664314882458e-05, "rmse": 4.17386793468327e-06, "relative_l2": 0.0018027972652426081, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.18668800592422485, "p95_ms": 0.1879359930753708, "p99_ms": 0.18838527649641038, "mean_ms": 0.18702346622943877, "n": 300}, {"repeat": 1, "p50_ms": 0.18665599822998047, "p95_ms": 0.187937593460083, "p99_ms": 0.18893471330404282, "mean_ms": 0.1875552000105381, "n": 300}, {"repeat": 2, "p50_ms": 0.18649600446224213, "p95_ms": 0.18787359297275544, "p99_ms": 0.18864063844084739, "mean_ms": 0.1871487998465697, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_227f18fd0026d1aa_native_dense.json", "configuration": "704x1280_tp1_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_dit_only_704x1280_tp1_sp1_v2", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_dit_only_704x1280_tp1_sp1_v2/captures/qkv_rank0_4624946207ce8f34.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 0, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "vllm_fa2", "status": "measured", "q_shape": [10560, 40, 128], "k_shape": [66880, 40, 128], "q_len": 10560, "k_len": 10560, "reference_error": {"max_abs": 0.03125, "rmse": 0.00037247839774330594, "relative_l2": 0.0007601394315501981, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.008999900738183353, "rmse": 0.0006804288244503694, "relative_l2": 0.0018858608044861479, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 6.0002241134643555, "p95_ms": 6.443619132041932, "p99_ms": 6.941253380775452, "mean_ms": 6.087191254297892, "n": 300}, {"repeat": 1, "p50_ms": 6.051503896713257, "p95_ms": 6.268280172348023, "p99_ms": 6.271396989822388, "mean_ms": 6.097411193847656, "n": 300}, {"repeat": 2, "p50_ms": 6.157055854797363, "p95_ms": 6.212528014183045, "p99_ms": 6.267910528182983, "mean_ms": 6.1201882632573446, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_4624946207ce8f34_vllm_fa2.json", "configuration": "704x1280_tp1_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_dit_only_704x1280_tp1_sp1_v2", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_dit_only_704x1280_tp1_sp1_v2/captures/qkv_rank0_4624946207ce8f34.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 0, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "vllm_fa3", "status": "measured", "q_shape": [10560, 40, 128], "k_shape": [66880, 40, 128], "q_len": 10560, "k_len": 10560, "reference_error": {"max_abs": 0.0, "rmse": 0.0, "relative_l2": 0.0, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.008999900738183353, "rmse": 0.0006941992574560287, "relative_l2": 0.0019240266183567644, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 3.4953439235687256, "p95_ms": 3.85353446006775, "p99_ms": 3.932617211341858, "mean_ms": 3.546550501982371, "n": 300}, {"repeat": 1, "p50_ms": 3.5319679975509644, "p95_ms": 3.717859137058258, "p99_ms": 3.7295680046081543, "mean_ms": 3.537785495122274, "n": 300}, {"repeat": 2, "p50_ms": 3.564703941345215, "p95_ms": 3.6526031732559203, "p99_ms": 3.737866172790527, "mean_ms": 3.54920064051946, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_4624946207ce8f34_vllm_fa3.json", "configuration": "704x1280_tp1_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_dit_only_704x1280_tp1_sp1_v2", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_dit_only_704x1280_tp1_sp1_v2/captures/qkv_rank0_4624946207ce8f34.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 0, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_math", "status": "measured", "q_shape": [10560, 40, 128], "k_shape": [66880, 40, 128], "q_len": 10560, "k_len": 10560, "reference_error": {"max_abs": 0.03125, "rmse": 0.0006212168327484413, "relative_l2": 0.0012677551583548208, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.007237463011211975, "rmse": 0.0005975797342955189, "relative_l2": 0.001656238180934626, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 88.25566482543945, "p95_ms": 88.3844367980957, "p99_ms": 88.4438339996338, "mean_ms": 88.24658294677734, "n": 300}, {"repeat": 1, "p50_ms": 88.23184204101562, "p95_ms": 88.36935462951661, "p99_ms": 88.42047325134277, "mean_ms": 88.21909591674805, "n": 300}, {"repeat": 2, "p50_ms": 88.2347183227539, "p95_ms": 88.37420768737793, "p99_ms": 88.41965972900391, "mean_ms": 88.22676887512208, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_4624946207ce8f34_sdpa_math.json", "configuration": "704x1280_tp1_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_dit_only_704x1280_tp1_sp1_v2", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_dit_only_704x1280_tp1_sp1_v2/captures/qkv_rank0_4624946207ce8f34.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 0, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_flash", "status": "measured", "q_shape": [10560, 40, 128], "k_shape": [66880, 40, 128], "q_len": 10560, "k_len": 10560, "reference_error": {"max_abs": 0.03125, "rmse": 0.00037276356346990753, "relative_l2": 0.0007607213866773404, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.008999900738183353, "rmse": 0.0006803588393833245, "relative_l2": 0.001885666835491748, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 6.523999929428101, "p95_ms": 6.858020901679993, "p99_ms": 6.978235902786255, "mean_ms": 6.5860531012217205, "n": 300}, {"repeat": 1, "p50_ms": 6.55675196647644, "p95_ms": 6.792067384719848, "p99_ms": 6.854340906143189, "mean_ms": 6.58248276869456, "n": 300}, {"repeat": 2, "p50_ms": 6.558255910873413, "p95_ms": 6.792049384117126, "p99_ms": 6.794336161613464, "mean_ms": 6.583528000513713, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_4624946207ce8f34_sdpa_flash.json", "configuration": "704x1280_tp1_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_dit_only_704x1280_tp1_sp1_v2", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_dit_only_704x1280_tp1_sp1_v2/captures/qkv_rank0_4624946207ce8f34.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 0, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_efficient", "status": "measured", "q_shape": [10560, 40, 128], "k_shape": [66880, 40, 128], "q_len": 10560, "k_len": 10560, "reference_error": {"max_abs": 0.03125, "rmse": 0.0005909000010255683, "relative_l2": 0.0012058857469423787, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.007237463011211975, "rmse": 0.0006850272498120121, "relative_l2": 0.0018986056939444714, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 12.573599815368652, "p95_ms": 12.825662469863891, "p99_ms": 12.89915735244751, "mean_ms": 12.656282971700033, "n": 300}, {"repeat": 1, "p50_ms": 12.742144107818604, "p95_ms": 12.817705249786377, "p99_ms": 12.828450756072998, "mean_ms": 12.677107928593953, "n": 300}, {"repeat": 2, "p50_ms": 12.743135929107666, "p95_ms": 12.815038537979126, "p99_ms": 12.824492855072021, "mean_ms": 12.67174506187439, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_4624946207ce8f34_sdpa_efficient.json", "configuration": "704x1280_tp1_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_dit_only_704x1280_tp1_sp1_v2", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_dit_only_704x1280_tp1_sp1_v2/captures/qkv_rank0_4624946207ce8f34.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 0, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_cudnn", "status": "measured", "q_shape": [10560, 40, 128], "k_shape": [66880, 40, 128], "q_len": 10560, "k_len": 10560, "reference_error": {"max_abs": 0.03125, "rmse": 0.0005893852974096597, "relative_l2": 0.0012027945986971661, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.007237463011211975, "rmse": 0.0006857538348473918, "relative_l2": 0.0019006194802364555, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 4.073407888412476, "p95_ms": 4.271139168739319, "p99_ms": 4.322624526023865, "mean_ms": 4.098439578215281, "n": 300}, {"repeat": 1, "p50_ms": 4.143696069717407, "p95_ms": 4.223224091529846, "p99_ms": 4.233443760871888, "mean_ms": 4.110505493481954, "n": 300}, {"repeat": 2, "p50_ms": 4.092927932739258, "p95_ms": 4.193025755882263, "p99_ms": 4.198786435127258, "mean_ms": 4.1053436891237896, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_4624946207ce8f34_sdpa_cudnn.json", "configuration": "704x1280_tp1_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_dit_only_704x1280_tp1_sp1_v2", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_dit_only_704x1280_tp1_sp1_v2/captures/qkv_rank0_4624946207ce8f34.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 0, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "native_dense", "status": "measured", "q_shape": [10560, 40, 128], "k_shape": [66880, 40, 128], "q_len": 10560, "k_len": 10560, "reference_error": {"max_abs": 0.03125, "rmse": 0.0003554075732281615, "relative_l2": 0.0007253019566210408, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.008999900738183353, "rmse": 0.0006932121785677012, "relative_l2": 0.0019212908533222126, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 3.4671519994735718, "p95_ms": 3.718928110599518, "p99_ms": 3.752023959159851, "mean_ms": 3.4844092774391173, "n": 300}, {"repeat": 1, "p50_ms": 3.4583840370178223, "p95_ms": 3.577473568916321, "p99_ms": 3.620584545135498, "mean_ms": 3.4813514661788942, "n": 300}, {"repeat": 2, "p50_ms": 3.4436320066452026, "p95_ms": 3.7144848465919496, "p99_ms": 3.823781125545502, "mean_ms": 3.4603341873486837, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_4624946207ce8f34_native_dense.json", "configuration": "704x1280_tp1_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_dit_only_704x1280_tp1_sp1_v2", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_dit_only_704x1280_tp1_sp1_v2/captures/qkv_rank0_6c1a0557389da998.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 27, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "vllm_fa2", "status": "measured", "q_shape": [10560, 40, 128], "k_shape": [66880, 40, 128], "q_len": 10560, "k_len": 63360, "reference_error": {"max_abs": 0.03125, "rmse": 0.00038618004036594455, "relative_l2": 0.0008082790262040963, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.008982203813774792, "rmse": 0.000752808786776018, "relative_l2": 0.001938066427686598, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 36.2913761138916, "p95_ms": 37.041042518615725, "p99_ms": 37.38791450500487, "mean_ms": 36.29430489857992, "n": 300}, {"repeat": 1, "p50_ms": 36.35921669006348, "p95_ms": 36.717800521850585, "p99_ms": 36.94534423828125, "mean_ms": 36.29164648691813, "n": 300}, {"repeat": 2, "p50_ms": 36.28988838195801, "p95_ms": 36.71707382202148, "p99_ms": 36.90351280212403, "mean_ms": 36.23441446940104, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_6c1a0557389da998_vllm_fa2.json", "configuration": "704x1280_tp1_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_dit_only_704x1280_tp1_sp1_v2", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_dit_only_704x1280_tp1_sp1_v2/captures/qkv_rank0_6c1a0557389da998.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 27, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "vllm_fa3", "status": "measured", "q_shape": [10560, 40, 128], "k_shape": [66880, 40, 128], "q_len": 10560, "k_len": 63360, "reference_error": {"max_abs": 0.0, "rmse": 0.0, "relative_l2": 0.0, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.008982203813774792, "rmse": 0.0007603426648000307, "relative_l2": 0.0019574619984146654, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 20.022096633911133, "p95_ms": 20.845104694366455, "p99_ms": 21.77596853256225, "mean_ms": 20.10694906870524, "n": 300}, {"repeat": 1, "p50_ms": 20.022496223449707, "p95_ms": 20.230475234985352, "p99_ms": 20.411222915649414, "mean_ms": 20.030157000223795, "n": 300}, {"repeat": 2, "p50_ms": 20.023008346557617, "p95_ms": 20.219062328338623, "p99_ms": 20.336678733825682, "mean_ms": 20.032400023142497, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_6c1a0557389da998_vllm_fa3.json", "configuration": "704x1280_tp1_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_dit_only_704x1280_tp1_sp1_v2", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_dit_only_704x1280_tp1_sp1_v2/captures/qkv_rank0_6c1a0557389da998.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 27, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_math", "status": "oom", "q_shape": [10560, 40, 128], "k_shape": [66880, 40, 128], "q_len": 10560, "k_len": 63360, "reference_error": null, "oracle_error": null, "repeats": null, "error": "CUDA out of memory. Tried to allocate 99.70 GiB. GPU 0 has a total capacity of 139.81 GiB of which 34.50 GiB is free. Including non-PyTorch memory, this process has 105.30 GiB memory in use. Of the allocated memory 104.62 GiB is allocated by PyTorch, and 3.25 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://docs.pytorch.org/docs/stable/notes/cuda.html#optimizing-memory-usage-with-pytorch-cuda-alloc-conf)", "dispatch_trace": null, "configuration": "704x1280_tp1_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_dit_only_704x1280_tp1_sp1_v2", "audit_status": "recorded_failure"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_dit_only_704x1280_tp1_sp1_v2/captures/qkv_rank0_6c1a0557389da998.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 27, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_flash", "status": "measured", "q_shape": [10560, 40, 128], "k_shape": [66880, 40, 128], "q_len": 10560, "k_len": 63360, "reference_error": {"max_abs": 0.03125, "rmse": 0.00038657706619591744, "relative_l2": 0.000809110005585952, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.008982203813774792, "rmse": 0.0007528591023141345, "relative_l2": 0.001938195962379773, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 38.6678409576416, "p95_ms": 39.69907112121582, "p99_ms": 40.50859275817871, "mean_ms": 38.790742022196454, "n": 300}, {"repeat": 1, "p50_ms": 38.681135177612305, "p95_ms": 39.09899368286133, "p99_ms": 39.43498378753662, "mean_ms": 38.70646764119466, "n": 300}, {"repeat": 2, "p50_ms": 38.66814422607422, "p95_ms": 39.047236633300784, "p99_ms": 39.10098133087158, "mean_ms": 38.691976127624514, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_6c1a0557389da998_sdpa_flash.json", "configuration": "704x1280_tp1_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_dit_only_704x1280_tp1_sp1_v2", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_dit_only_704x1280_tp1_sp1_v2/captures/qkv_rank0_6c1a0557389da998.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 27, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_efficient", "status": "measured", "q_shape": [10560, 40, 128], "k_shape": [66880, 40, 128], "q_len": 10560, "k_len": 63360, "reference_error": {"max_abs": 0.03125, "rmse": 0.0006353318507256788, "relative_l2": 0.001329756476109901, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.00919098333101509, "rmse": 0.0007510854117432321, "relative_l2": 0.001933629689763201, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 75.26118469238281, "p95_ms": 75.54803237915038, "p99_ms": 75.6216088104248, "mean_ms": 75.30430956522623, "n": 300}, {"repeat": 1, "p50_ms": 75.47726058959961, "p95_ms": 75.61190681457519, "p99_ms": 75.63500801086425, "mean_ms": 75.47183670043945, "n": 300}, {"repeat": 2, "p50_ms": 75.47402954101562, "p95_ms": 75.6069839477539, "p99_ms": 75.62469596862793, "mean_ms": 75.47296773274739, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_6c1a0557389da998_sdpa_efficient.json", "configuration": "704x1280_tp1_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_dit_only_704x1280_tp1_sp1_v2", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_dit_only_704x1280_tp1_sp1_v2/captures/qkv_rank0_6c1a0557389da998.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 27, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_cudnn", "status": "measured", "q_shape": [10560, 40, 128], "k_shape": [66880, 40, 128], "q_len": 10560, "k_len": 63360, "reference_error": {"max_abs": 0.03125, "rmse": 0.0006344365122063293, "relative_l2": 0.0013278825228474358, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.00919098333101509, "rmse": 0.0007507384628611804, "relative_l2": 0.001932736488206259, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 22.224639892578125, "p95_ms": 22.934800720214845, "p99_ms": 23.939005870819084, "mean_ms": 22.339452794392905, "n": 300}, {"repeat": 1, "p50_ms": 22.320591926574707, "p95_ms": 22.63811616897583, "p99_ms": 22.652464866638184, "mean_ms": 22.306585178375244, "n": 300}, {"repeat": 2, "p50_ms": 22.33191967010498, "p95_ms": 22.628672695159914, "p99_ms": 22.635938987731933, "mean_ms": 22.302975826263427, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_6c1a0557389da998_sdpa_cudnn.json", "configuration": "704x1280_tp1_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_dit_only_704x1280_tp1_sp1_v2", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_dit_only_704x1280_tp1_sp1_v2/captures/qkv_rank0_6c1a0557389da998.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 27, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "native_dense", "status": "measured", "q_shape": [10560, 40, 128], "k_shape": [66880, 40, 128], "q_len": 10560, "k_len": 63360, "reference_error": {"max_abs": 0.03125, "rmse": 0.0003710655036097366, "relative_l2": 0.000776644136324088, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.008982203813774792, "rmse": 0.0007623522854336237, "relative_l2": 0.0019626356605062473, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 20.202655792236328, "p95_ms": 21.155917358398437, "p99_ms": 22.05991340637207, "mean_ms": 20.28225066502889, "n": 300}, {"repeat": 1, "p50_ms": 20.14036750793457, "p95_ms": 20.559696865081786, "p99_ms": 20.698397312164307, "mean_ms": 20.15699665705363, "n": 300}, {"repeat": 2, "p50_ms": 20.17967987060547, "p95_ms": 20.51612768173218, "p99_ms": 20.7120977973938, "mean_ms": 20.206677010854087, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_6c1a0557389da998_native_dense.json", "configuration": "704x1280_tp1_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_dit_only_704x1280_tp1_sp1_v2", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_dit_only_704x1280_tp1_sp1_v2/captures/qkv_rank0_6e8918fca4760b4e.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 27, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "vllm_fa2", "status": "measured", "q_shape": [10560, 40, 128], "k_shape": [512, 40, 128], "q_len": 10560, "k_len": 512, "reference_error": {"max_abs": 0.000244140625, "rmse": 1.8576455099480584e-06, "relative_l2": 0.00024310025907789328, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 7.818887076814496e-05, "rmse": 5.353783328355767e-06, "relative_l2": 0.0018571216816885075, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.3245120048522949, "p95_ms": 0.32710560262203214, "p99_ms": 0.3279059061408043, "mean_ms": 0.3251458130280177, "n": 300}, {"repeat": 1, "p50_ms": 0.32422399520874023, "p95_ms": 0.32524800300598145, "p99_ms": 0.3255683106184006, "mean_ms": 0.32475466748078663, "n": 300}, {"repeat": 2, "p50_ms": 0.3243519961833954, "p95_ms": 0.325439989566803, "p99_ms": 0.3258886477351189, "mean_ms": 0.3246887466311455, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_6e8918fca4760b4e_vllm_fa2.json", "configuration": "704x1280_tp1_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_dit_only_704x1280_tp1_sp1_v2", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_dit_only_704x1280_tp1_sp1_v2/captures/qkv_rank0_6e8918fca4760b4e.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 27, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "vllm_fa3", "status": "measured", "q_shape": [10560, 40, 128], "k_shape": [512, 40, 128], "q_len": 10560, "k_len": 512, "reference_error": {"max_abs": 0.000244140625, "rmse": 7.014523557031672e-07, "relative_l2": 9.179536595601951e-05, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 7.818887076814496e-05, "rmse": 5.395104933635903e-06, "relative_l2": 0.0018714553303218902, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.26553601026535034, "p95_ms": 0.26998400688171387, "p99_ms": 0.2714940908551216, "mean_ms": 0.2661842129627864, "n": 300}, {"repeat": 1, "p50_ms": 0.26447999477386475, "p95_ms": 0.2697312042117119, "p99_ms": 0.27056063652038576, "mean_ms": 0.265516052544117, "n": 300}, {"repeat": 2, "p50_ms": 0.2648639976978302, "p95_ms": 0.27043519616127015, "p99_ms": 0.27232288241386415, "mean_ms": 0.26605045328537624, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_6e8918fca4760b4e_vllm_fa3.json", "configuration": "704x1280_tp1_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_dit_only_704x1280_tp1_sp1_v2", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_dit_only_704x1280_tp1_sp1_v2/captures/qkv_rank0_6e8918fca4760b4e.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 27, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_math", "status": "measured", "q_shape": [10560, 40, 128], "k_shape": [512, 40, 128], "q_len": 10560, "k_len": 512, "reference_error": {"max_abs": 0.00048828125, "rmse": 4.489848934334054e-06, "relative_l2": 0.0005875628225687315, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 7.818887076814496e-05, "rmse": 4.7402353330505205e-06, "relative_l2": 0.0016442940017181583, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 4.653088092803955, "p95_ms": 4.659496188163757, "p99_ms": 4.66329430103302, "mean_ms": 4.653250872294108, "n": 300}, {"repeat": 1, "p50_ms": 4.6525280475616455, "p95_ms": 4.657793688774109, "p99_ms": 4.660128755569458, "mean_ms": 4.652471359570821, "n": 300}, {"repeat": 2, "p50_ms": 4.652416229248047, "p95_ms": 4.658656120300293, "p99_ms": 4.660448393821716, "mean_ms": 4.653031147321065, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_6e8918fca4760b4e_sdpa_math.json", "configuration": "704x1280_tp1_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_dit_only_704x1280_tp1_sp1_v2", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_dit_only_704x1280_tp1_sp1_v2/captures/qkv_rank0_6e8918fca4760b4e.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 27, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_flash", "status": "measured", "q_shape": [10560, 40, 128], "k_shape": [512, 40, 128], "q_len": 10560, "k_len": 512, "reference_error": {"max_abs": 0.000244140625, "rmse": 1.8578953239835818e-06, "relative_l2": 0.00024313295092164486, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 7.818887076814496e-05, "rmse": 5.353783328355767e-06, "relative_l2": 0.0018571216816885075, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.34668800234794617, "p95_ms": 0.34774558842182157, "p99_ms": 0.3481926462054253, "mean_ms": 0.3468639992674192, "n": 300}, {"repeat": 1, "p50_ms": 0.34668800234794617, "p95_ms": 0.3492479920387268, "p99_ms": 0.3500166380405426, "mean_ms": 0.3472033066550891, "n": 300}, {"repeat": 2, "p50_ms": 0.3466239869594574, "p95_ms": 0.34752159416675565, "p99_ms": 0.3478102433681488, "mean_ms": 0.34699125399192177, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_6e8918fca4760b4e_sdpa_flash.json", "configuration": "704x1280_tp1_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_dit_only_704x1280_tp1_sp1_v2", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_dit_only_704x1280_tp1_sp1_v2/captures/qkv_rank0_6e8918fca4760b4e.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 27, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_efficient", "status": "measured", "q_shape": [10560, 40, 128], "k_shape": [512, 40, 128], "q_len": 10560, "k_len": 512, "reference_error": {"max_abs": 0.00048828125, "rmse": 3.2413443973460267e-06, "relative_l2": 0.00042417762621326465, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 7.818887076814496e-05, "rmse": 5.891822370152756e-06, "relative_l2": 0.002043756797238275, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.6602879762649536, "p95_ms": 0.6618271887302398, "p99_ms": 0.6624982112646103, "mean_ms": 0.6606462925672532, "n": 300}, {"repeat": 1, "p50_ms": 0.659775972366333, "p95_ms": 0.6607679724693298, "p99_ms": 0.661703056693077, "mean_ms": 0.6603778125842412, "n": 300}, {"repeat": 2, "p50_ms": 0.6604480147361755, "p95_ms": 0.6616992145776749, "p99_ms": 0.6625298988819123, "mean_ms": 0.6609261864423752, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_6e8918fca4760b4e_sdpa_efficient.json", "configuration": "704x1280_tp1_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_dit_only_704x1280_tp1_sp1_v2", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_dit_only_704x1280_tp1_sp1_v2/captures/qkv_rank0_6e8918fca4760b4e.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 27, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_cudnn", "status": "measured", "q_shape": [10560, 40, 128], "k_shape": [512, 40, 128], "q_len": 10560, "k_len": 512, "reference_error": {"max_abs": 0.00048828125, "rmse": 3.239739932173183e-06, "relative_l2": 0.00042396765832805134, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 7.818887076814496e-05, "rmse": 5.891822370152756e-06, "relative_l2": 0.002043756797238275, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.19366399943828583, "p95_ms": 0.19542400538921356, "p99_ms": 0.19593727439641953, "mean_ms": 0.19426122695207595, "n": 300}, {"repeat": 1, "p50_ms": 0.1950719952583313, "p95_ms": 0.1960960030555725, "p99_ms": 0.19667264729738235, "mean_ms": 0.19590314681331317, "n": 300}, {"repeat": 2, "p50_ms": 0.19510400295257568, "p95_ms": 0.19609760269522666, "p99_ms": 0.19660832688212396, "mean_ms": 0.19555744032065073, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_6e8918fca4760b4e_sdpa_cudnn.json", "configuration": "704x1280_tp1_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_dit_only_704x1280_tp1_sp1_v2", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_dit_only_704x1280_tp1_sp1_v2/captures/qkv_rank0_6e8918fca4760b4e.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 27, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "native_dense", "status": "measured", "q_shape": [10560, 40, 128], "k_shape": [512, 40, 128], "q_len": 10560, "k_len": 512, "reference_error": {"max_abs": 0.0, "rmse": 0.0, "relative_l2": 0.0, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 7.818887076814496e-05, "rmse": 5.410283113664583e-06, "relative_l2": 0.001876720341154611, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.18691200017929077, "p95_ms": 0.18828800320625305, "p99_ms": 0.1891871951520443, "mean_ms": 0.18731146668394408, "n": 300}, {"repeat": 1, "p50_ms": 0.1865919977426529, "p95_ms": 0.1889919936656952, "p99_ms": 0.19013983279466604, "mean_ms": 0.188549013286829, "n": 300}, {"repeat": 2, "p50_ms": 0.18751999735832214, "p95_ms": 0.1894720047712326, "p99_ms": 0.19425215855240757, "mean_ms": 0.18963210692008337, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_6e8918fca4760b4e_native_dense.json", "configuration": "704x1280_tp1_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_dit_only_704x1280_tp1_sp1_v2", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_dit_only_704x1280_tp1_sp1_v2/captures/qkv_rank0_71ccfd9a9e128870.pt", "context": {"phase": "commit", "step": null, "start_frame": 9, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "vllm_fa2", "status": "measured", "q_shape": [10560, 40, 128], "k_shape": [66880, 40, 128], "q_len": 10560, "k_len": 42240, "reference_error": {"max_abs": 0.0625, "rmse": 0.0003708618079538423, "relative_l2": 0.000483445044055905, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.014540188720360447, "rmse": 0.0009034149203516021, "relative_l2": 0.001673807418418208, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 23.93889617919922, "p95_ms": 24.70022048950195, "p99_ms": 24.97909105300903, "mean_ms": 24.050463689168293, "n": 300}, {"repeat": 1, "p50_ms": 24.01307201385498, "p95_ms": 24.48407688140869, "p99_ms": 24.601215229034423, "mean_ms": 24.063959153493247, "n": 300}, {"repeat": 2, "p50_ms": 24.053071975708008, "p95_ms": 24.485508918762207, "p99_ms": 24.49066177368164, "mean_ms": 24.090085163116456, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_71ccfd9a9e128870_vllm_fa2.json", "configuration": "704x1280_tp1_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_dit_only_704x1280_tp1_sp1_v2", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_dit_only_704x1280_tp1_sp1_v2/captures/qkv_rank0_71ccfd9a9e128870.pt", "context": {"phase": "commit", "step": null, "start_frame": 9, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "vllm_fa3", "status": "measured", "q_shape": [10560, 40, 128], "k_shape": [66880, 40, 128], "q_len": 10560, "k_len": 42240, "reference_error": {"max_abs": 0.0, "rmse": 0.0, "relative_l2": 0.0, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.014540188720360447, "rmse": 0.0009039246564090026, "relative_l2": 0.0016747518349593698, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 13.383344173431396, "p95_ms": 14.101180601119996, "p99_ms": 15.194212884902953, "mean_ms": 13.469137283960977, "n": 300}, {"repeat": 1, "p50_ms": 13.397727966308594, "p95_ms": 13.657464170455933, "p99_ms": 13.798795137405396, "mean_ms": 13.434210577011108, "n": 300}, {"repeat": 2, "p50_ms": 13.390848159790039, "p95_ms": 13.608537244796754, "p99_ms": 13.649164600372314, "mean_ms": 13.426661659876505, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_71ccfd9a9e128870_vllm_fa3.json", "configuration": "704x1280_tp1_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_dit_only_704x1280_tp1_sp1_v2", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_dit_only_704x1280_tp1_sp1_v2/captures/qkv_rank0_71ccfd9a9e128870.pt", "context": {"phase": "commit", "step": null, "start_frame": 9, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_math", "status": "oom", "q_shape": [10560, 40, 128], "k_shape": [66880, 40, 128], "q_len": 10560, "k_len": 42240, "reference_error": null, "oracle_error": null, "repeats": null, "error": "CUDA out of memory. Tried to allocate 16.62 GiB. GPU 0 has a total capacity of 139.81 GiB of which 2.23 GiB is free. Including non-PyTorch memory, this process has 137.58 GiB memory in use. Of the allocated memory 136.89 GiB is allocated by PyTorch, and 6.25 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://docs.pytorch.org/docs/stable/notes/cuda.html#optimizing-memory-usage-with-pytorch-cuda-alloc-conf)", "dispatch_trace": null, "configuration": "704x1280_tp1_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_dit_only_704x1280_tp1_sp1_v2", "audit_status": "recorded_failure"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_dit_only_704x1280_tp1_sp1_v2/captures/qkv_rank0_71ccfd9a9e128870.pt", "context": {"phase": "commit", "step": null, "start_frame": 9, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_flash", "status": "measured", "q_shape": [10560, 40, 128], "k_shape": [66880, 40, 128], "q_len": 10560, "k_len": 42240, "reference_error": {"max_abs": 0.0625, "rmse": 0.00037179299959547006, "relative_l2": 0.0004846589193446413, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.014540188720360447, "rmse": 0.0009034055105428151, "relative_l2": 0.001673789984338475, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 25.718656539916992, "p95_ms": 26.54323434829712, "p99_ms": 27.511741065979002, "mean_ms": 25.807872047424315, "n": 300}, {"repeat": 1, "p50_ms": 25.77899169921875, "p95_ms": 26.077507877349852, "p99_ms": 26.357937049865722, "mean_ms": 25.773418769836425, "n": 300}, {"repeat": 2, "p50_ms": 25.80003261566162, "p95_ms": 26.076140785217284, "p99_ms": 26.478140277862547, "mean_ms": 25.790641905466714, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_71ccfd9a9e128870_sdpa_flash.json", "configuration": "704x1280_tp1_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_dit_only_704x1280_tp1_sp1_v2", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_dit_only_704x1280_tp1_sp1_v2/captures/qkv_rank0_71ccfd9a9e128870.pt", "context": {"phase": "commit", "step": null, "start_frame": 9, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_efficient", "status": "measured", "q_shape": [10560, 40, 128], "k_shape": [66880, 40, 128], "q_len": 10560, "k_len": 42240, "reference_error": {"max_abs": 0.0625, "rmse": 0.0007558482411220217, "relative_l2": 0.0009853025531124345, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.014540188720360447, "rmse": 0.0008957329515698156, "relative_l2": 0.0016595746046298252, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 50.177743911743164, "p95_ms": 50.28296451568603, "p99_ms": 50.287256736755374, "mean_ms": 50.18437391916911, "n": 300}, {"repeat": 1, "p50_ms": 50.16641426086426, "p95_ms": 50.27409439086914, "p99_ms": 50.2862202835083, "mean_ms": 50.179937795003255, "n": 300}, {"repeat": 2, "p50_ms": 50.168527603149414, "p95_ms": 50.275118827819824, "p99_ms": 50.285678749084475, "mean_ms": 50.180529747009274, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_71ccfd9a9e128870_sdpa_efficient.json", "configuration": "704x1280_tp1_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_dit_only_704x1280_tp1_sp1_v2", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_dit_only_704x1280_tp1_sp1_v2/captures/qkv_rank0_71ccfd9a9e128870.pt", "context": {"phase": "commit", "step": null, "start_frame": 9, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_cudnn", "status": "measured", "q_shape": [10560, 40, 128], "k_shape": [66880, 40, 128], "q_len": 10560, "k_len": 42240, "reference_error": {"max_abs": 0.0625, "rmse": 0.0007553607883353866, "relative_l2": 0.0009846671233408701, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.014540188720360447, "rmse": 0.0008957258433842463, "relative_l2": 0.0016595614348963292, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 14.685232162475586, "p95_ms": 15.656624221801758, "p99_ms": 16.984072895050048, "mean_ms": 14.852280095418294, "n": 300}, {"repeat": 1, "p50_ms": 14.81497573852539, "p95_ms": 14.960332822799682, "p99_ms": 15.051093015670777, "mean_ms": 14.796220483779907, "n": 300}, {"repeat": 2, "p50_ms": 14.813807964324951, "p95_ms": 14.959009790420533, "p99_ms": 15.02002049446106, "mean_ms": 14.806751251220703, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_71ccfd9a9e128870_sdpa_cudnn.json", "configuration": "704x1280_tp1_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_dit_only_704x1280_tp1_sp1_v2", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_dit_only_704x1280_tp1_sp1_v2/captures/qkv_rank0_71ccfd9a9e128870.pt", "context": {"phase": "commit", "step": null, "start_frame": 9, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "native_dense", "status": "measured", "q_shape": [10560, 40, 128], "k_shape": [66880, 40, 128], "q_len": 10560, "k_len": 42240, "reference_error": {"max_abs": 0.0625, "rmse": 0.0003518306066429341, "relative_l2": 0.0004586365041662059, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.014540188720360447, "rmse": 0.0009027966776015421, "relative_l2": 0.0016726619654506716, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 13.343695640563965, "p95_ms": 14.361319971084596, "p99_ms": 15.29698717117309, "mean_ms": 13.47507500330607, "n": 300}, {"repeat": 1, "p50_ms": 13.461151599884033, "p95_ms": 13.755194902420044, "p99_ms": 14.119501724243165, "mean_ms": 13.45603263537089, "n": 300}, {"repeat": 2, "p50_ms": 13.456431865692139, "p95_ms": 13.730931520462036, "p99_ms": 13.753128414154054, "mean_ms": 13.44608447710673, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_71ccfd9a9e128870_native_dense.json", "configuration": "704x1280_tp1_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_dit_only_704x1280_tp1_sp1_v2", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_dit_only_704x1280_tp1_sp1_v2/captures/qkv_rank0_7cb64f86f262cbae.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 9, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "vllm_fa2", "status": "measured", "q_shape": [10560, 40, 128], "k_shape": [66880, 40, 128], "q_len": 10560, "k_len": 42240, "reference_error": {"max_abs": 0.03125, "rmse": 0.00039086875743339975, "relative_l2": 0.0008193950430639397, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.008166105858370631, "rmse": 0.0007497399548139017, "relative_l2": 0.001914449886474462, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 24.138976097106934, "p95_ms": 24.922061347961424, "p99_ms": 25.634215583801264, "mean_ms": 24.24629006067912, "n": 300}, {"repeat": 1, "p50_ms": 24.119279861450195, "p95_ms": 24.487914848327637, "p99_ms": 24.70007257461548, "mean_ms": 24.16064479827881, "n": 300}, {"repeat": 2, "p50_ms": 24.264927864074707, "p95_ms": 24.486248397827147, "p99_ms": 24.49862855911255, "mean_ms": 24.162241694132486, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_7cb64f86f262cbae_vllm_fa2.json", "configuration": "704x1280_tp1_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_dit_only_704x1280_tp1_sp1_v2", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_dit_only_704x1280_tp1_sp1_v2/captures/qkv_rank0_7cb64f86f262cbae.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 9, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "vllm_fa3", "status": "measured", "q_shape": [10560, 40, 128], "k_shape": [66880, 40, 128], "q_len": 10560, "k_len": 42240, "reference_error": {"max_abs": 0.0, "rmse": 0.0, "relative_l2": 0.0, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.008166105858370631, "rmse": 0.0007599393361606038, "relative_l2": 0.0019404938559013684, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 13.402640342712402, "p95_ms": 14.034929990768433, "p99_ms": 15.349745225906371, "mean_ms": 13.569270165761312, "n": 300}, {"repeat": 1, "p50_ms": 13.5175199508667, "p95_ms": 13.815449762344361, "p99_ms": 13.934928150177003, "mean_ms": 13.530414950052897, "n": 300}, {"repeat": 2, "p50_ms": 13.516176223754883, "p95_ms": 13.656486749649048, "p99_ms": 13.75749282836914, "mean_ms": 13.50589835802714, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_7cb64f86f262cbae_vllm_fa3.json", "configuration": "704x1280_tp1_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_dit_only_704x1280_tp1_sp1_v2", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_dit_only_704x1280_tp1_sp1_v2/captures/qkv_rank0_7cb64f86f262cbae.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 9, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_math", "status": "oom", "q_shape": [10560, 40, 128], "k_shape": [66880, 40, 128], "q_len": 10560, "k_len": 42240, "reference_error": null, "oracle_error": null, "repeats": null, "error": "CUDA out of memory. Tried to allocate 16.62 GiB. GPU 0 has a total capacity of 139.81 GiB of which 2.23 GiB is free. Including non-PyTorch memory, this process has 137.58 GiB memory in use. Of the allocated memory 136.89 GiB is allocated by PyTorch, and 6.25 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://docs.pytorch.org/docs/stable/notes/cuda.html#optimizing-memory-usage-with-pytorch-cuda-alloc-conf)", "dispatch_trace": null, "configuration": "704x1280_tp1_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_dit_only_704x1280_tp1_sp1_v2", "audit_status": "recorded_failure"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_dit_only_704x1280_tp1_sp1_v2/captures/qkv_rank0_7cb64f86f262cbae.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 9, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_flash", "status": "measured", "q_shape": [10560, 40, 128], "k_shape": [66880, 40, 128], "q_len": 10560, "k_len": 42240, "reference_error": {"max_abs": 0.03125, "rmse": 0.00039125604206418876, "relative_l2": 0.000820206924547656, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.008166105858370631, "rmse": 0.000749741695940512, "relative_l2": 0.0019144543324155112, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 25.838144302368164, "p95_ms": 26.54440059661865, "p99_ms": 27.55845794677734, "mean_ms": 25.889210980733235, "n": 300}, {"repeat": 1, "p50_ms": 25.85275173187256, "p95_ms": 26.07986373901367, "p99_ms": 26.368224773406983, "mean_ms": 25.85260721206665, "n": 300}, {"repeat": 2, "p50_ms": 25.853296279907227, "p95_ms": 26.07896089553833, "p99_ms": 26.090707511901854, "mean_ms": 25.84844029744466, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_7cb64f86f262cbae_sdpa_flash.json", "configuration": "704x1280_tp1_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_dit_only_704x1280_tp1_sp1_v2", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_dit_only_704x1280_tp1_sp1_v2/captures/qkv_rank0_7cb64f86f262cbae.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 9, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_efficient", "status": "measured", "q_shape": [10560, 40, 128], "k_shape": [66880, 40, 128], "q_len": 10560, "k_len": 42240, "reference_error": {"max_abs": 0.03125, "rmse": 0.0006261769082387287, "relative_l2": 0.0013126816736672607, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.007458894141629369, "rmse": 0.0007553631967359011, "relative_l2": 0.0019288087515583728, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 50.35748863220215, "p95_ms": 50.45923385620117, "p99_ms": 50.47006416320801, "mean_ms": 50.357834536234535, "n": 300}, {"repeat": 1, "p50_ms": 50.35433578491211, "p95_ms": 50.45992431640625, "p99_ms": 50.47080341339111, "mean_ms": 50.35949110666911, "n": 300}, {"repeat": 2, "p50_ms": 50.34231948852539, "p95_ms": 50.465166091918945, "p99_ms": 50.47626308441162, "mean_ms": 50.35797121683756, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_7cb64f86f262cbae_sdpa_efficient.json", "configuration": "704x1280_tp1_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_dit_only_704x1280_tp1_sp1_v2", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_dit_only_704x1280_tp1_sp1_v2/captures/qkv_rank0_7cb64f86f262cbae.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 9, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_cudnn", "status": "measured", "q_shape": [10560, 40, 128], "k_shape": [66880, 40, 128], "q_len": 10560, "k_len": 42240, "reference_error": {"max_abs": 0.03125, "rmse": 0.0006243007566241007, "relative_l2": 0.0013087486160774007, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.007458894141629369, "rmse": 0.0007556984185515758, "relative_l2": 0.001929664735506996, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 14.81550407409668, "p95_ms": 15.455009984970093, "p99_ms": 16.456413593292233, "mean_ms": 14.942785387039185, "n": 300}, {"repeat": 1, "p50_ms": 14.957472324371338, "p95_ms": 15.114199829101562, "p99_ms": 15.148013191223145, "mean_ms": 14.924658133188883, "n": 300}, {"repeat": 2, "p50_ms": 14.869776248931885, "p95_ms": 15.090534257888795, "p99_ms": 15.11595290184021, "mean_ms": 14.885847562154135, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_7cb64f86f262cbae_sdpa_cudnn.json", "configuration": "704x1280_tp1_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_dit_only_704x1280_tp1_sp1_v2", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_dit_only_704x1280_tp1_sp1_v2/captures/qkv_rank0_7cb64f86f262cbae.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 9, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "native_dense", "status": "measured", "q_shape": [10560, 40, 128], "k_shape": [66880, 40, 128], "q_len": 10560, "k_len": 42240, "reference_error": {"max_abs": 0.03125, "rmse": 0.0003744601232531848, "relative_l2": 0.0007849969151628921, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.008166105858370631, "rmse": 0.0007626402458939546, "relative_l2": 0.001947390588960866, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 13.451168060302734, "p95_ms": 14.459568071365357, "p99_ms": 15.307854795455928, "mean_ms": 13.547866671880087, "n": 300}, {"repeat": 1, "p50_ms": 13.58407974243164, "p95_ms": 13.8295428276062, "p99_ms": 14.329408216476441, "mean_ms": 13.578878396352133, "n": 300}, {"repeat": 2, "p50_ms": 13.579024314880371, "p95_ms": 13.761367559432983, "p99_ms": 14.064159488677978, "mean_ms": 13.5761643409729, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_7cb64f86f262cbae_native_dense.json", "configuration": "704x1280_tp1_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_dit_only_704x1280_tp1_sp1_v2", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_dit_only_704x1280_tp1_sp1_v2/captures/qkv_rank0_847f3e9ceb4b37a4.pt", "context": {"phase": "commit", "step": null, "start_frame": 0, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "vllm_fa2", "status": "measured", "q_shape": [10560, 40, 128], "k_shape": [512, 40, 128], "q_len": 10560, "k_len": 512, "reference_error": {"max_abs": 0.0009765625, "rmse": 1.6778567024883627e-06, "relative_l2": 0.00020327745038212488, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.00043489744124647345, "rmse": 1.7589970025427215e-05, "relative_l2": 0.001761251841652134, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.32419198751449585, "p95_ms": 0.32678401470184326, "p99_ms": 0.32748928517103193, "mean_ms": 0.3245768530170123, "n": 300}, {"repeat": 1, "p50_ms": 0.32419198751449585, "p95_ms": 0.3251200020313263, "p99_ms": 0.32560095816850665, "mean_ms": 0.32457930624485015, "n": 300}, {"repeat": 2, "p50_ms": 0.3242560029029846, "p95_ms": 0.32528160959482194, "p99_ms": 0.32576480686664583, "mean_ms": 0.32465365370114646, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_847f3e9ceb4b37a4_vllm_fa2.json", "configuration": "704x1280_tp1_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_dit_only_704x1280_tp1_sp1_v2", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_dit_only_704x1280_tp1_sp1_v2/captures/qkv_rank0_847f3e9ceb4b37a4.pt", "context": {"phase": "commit", "step": null, "start_frame": 0, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "vllm_fa3", "status": "measured", "q_shape": [10560, 40, 128], "k_shape": [512, 40, 128], "q_len": 10560, "k_len": 512, "reference_error": {"max_abs": 0.00048828125, "rmse": 6.261262689096e-07, "relative_l2": 7.585710470533997e-05, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.00043489744124647345, "rmse": 1.7598296726068673e-05, "relative_l2": 0.0017620855791069708, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.264847993850708, "p95_ms": 0.26924960762262345, "p99_ms": 0.2709766468405724, "mean_ms": 0.265868053038915, "n": 300}, {"repeat": 1, "p50_ms": 0.26447999477386475, "p95_ms": 0.26915678679943084, "p99_ms": 0.2702425688505173, "mean_ms": 0.2654146131873131, "n": 300}, {"repeat": 2, "p50_ms": 0.264384001493454, "p95_ms": 0.26880479156970977, "p99_ms": 0.2697020772099495, "mean_ms": 0.2653918926914533, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_847f3e9ceb4b37a4_vllm_fa3.json", "configuration": "704x1280_tp1_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_dit_only_704x1280_tp1_sp1_v2", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_dit_only_704x1280_tp1_sp1_v2/captures/qkv_rank0_847f3e9ceb4b37a4.pt", "context": {"phase": "commit", "step": null, "start_frame": 0, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_math", "status": "measured", "q_shape": [10560, 40, 128], "k_shape": [512, 40, 128], "q_len": 10560, "k_len": 512, "reference_error": {"max_abs": 0.001953125, "rmse": 3.825510661476973e-06, "relative_l2": 0.0004634722694264582, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.00043489744124647345, "rmse": 1.747372544361234e-05, "relative_l2": 0.0017496124822042455, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 4.650976181030273, "p95_ms": 4.656484651565552, "p99_ms": 4.660169982910157, "mean_ms": 4.651194767951965, "n": 300}, {"repeat": 1, "p50_ms": 4.650831937789917, "p95_ms": 4.655521559715271, "p99_ms": 4.657030305862427, "mean_ms": 4.650811305046082, "n": 300}, {"repeat": 2, "p50_ms": 4.651439905166626, "p95_ms": 4.655969786643982, "p99_ms": 4.6598100757598875, "mean_ms": 4.651565534273783, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_847f3e9ceb4b37a4_sdpa_math.json", "configuration": "704x1280_tp1_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_dit_only_704x1280_tp1_sp1_v2", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_dit_only_704x1280_tp1_sp1_v2/captures/qkv_rank0_847f3e9ceb4b37a4.pt", "context": {"phase": "commit", "step": null, "start_frame": 0, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_flash", "status": "measured", "q_shape": [10560, 40, 128], "k_shape": [512, 40, 128], "q_len": 10560, "k_len": 512, "reference_error": {"max_abs": 0.0009765625, "rmse": 1.676763289620977e-06, "relative_l2": 0.0002031449800826246, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.00043489744124647345, "rmse": 1.7589970025427215e-05, "relative_l2": 0.001761251841652134, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.3463360071182251, "p95_ms": 0.3471695989370346, "p99_ms": 0.3477139300107956, "mean_ms": 0.3465598936875661, "n": 300}, {"repeat": 1, "p50_ms": 0.34617599844932556, "p95_ms": 0.347135990858078, "p99_ms": 0.3478083235025406, "mean_ms": 0.3464630402127902, "n": 300}, {"repeat": 2, "p50_ms": 0.34627199172973633, "p95_ms": 0.34726559221744535, "p99_ms": 0.347747828066349, "mean_ms": 0.3465955199797948, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_847f3e9ceb4b37a4_sdpa_flash.json", "configuration": "704x1280_tp1_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_dit_only_704x1280_tp1_sp1_v2", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_dit_only_704x1280_tp1_sp1_v2/captures/qkv_rank0_847f3e9ceb4b37a4.pt", "context": {"phase": "commit", "step": null, "start_frame": 0, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_efficient", "status": "measured", "q_shape": [10560, 40, 128], "k_shape": [512, 40, 128], "q_len": 10560, "k_len": 512, "reference_error": {"max_abs": 0.0009765625, "rmse": 2.9159285516187726e-06, "relative_l2": 0.00035327362616273096, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.00043489744124647345, "rmse": 1.7758429651179162e-05, "relative_l2": 0.0017781193988833715, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.6599680185317993, "p95_ms": 0.6613455921411514, "p99_ms": 0.6623679995536804, "mean_ms": 0.660128426750501, "n": 300}, {"repeat": 1, "p50_ms": 0.6596800088882446, "p95_ms": 0.6606416016817093, "p99_ms": 0.6609600186347961, "mean_ms": 0.6602127995093664, "n": 300}, {"repeat": 2, "p50_ms": 0.6600959897041321, "p95_ms": 0.6611872136592865, "p99_ms": 0.6620172548294068, "mean_ms": 0.660338239868482, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_847f3e9ceb4b37a4_sdpa_efficient.json", "configuration": "704x1280_tp1_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_dit_only_704x1280_tp1_sp1_v2", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_dit_only_704x1280_tp1_sp1_v2/captures/qkv_rank0_847f3e9ceb4b37a4.pt", "context": {"phase": "commit", "step": null, "start_frame": 0, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_cudnn", "status": "measured", "q_shape": [10560, 40, 128], "k_shape": [512, 40, 128], "q_len": 10560, "k_len": 512, "reference_error": {"max_abs": 0.0009765625, "rmse": 2.9124506587042092e-06, "relative_l2": 0.00035285226884221245, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.00043489744124647345, "rmse": 1.7758429651179162e-05, "relative_l2": 0.0017781193988833715, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.194815993309021, "p95_ms": 0.1957135997712612, "p99_ms": 0.19625695690512657, "mean_ms": 0.19502122610807418, "n": 300}, {"repeat": 1, "p50_ms": 0.19494399428367615, "p95_ms": 0.1961279958486557, "p99_ms": 0.19654432639479638, "mean_ms": 0.19550698652863502, "n": 300}, {"repeat": 2, "p50_ms": 0.19491200149059296, "p95_ms": 0.19597120210528374, "p99_ms": 0.1964803259074688, "mean_ms": 0.19542304009199143, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_847f3e9ceb4b37a4_sdpa_cudnn.json", "configuration": "704x1280_tp1_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_dit_only_704x1280_tp1_sp1_v2", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_dit_only_704x1280_tp1_sp1_v2/captures/qkv_rank0_847f3e9ceb4b37a4.pt", "context": {"phase": "commit", "step": null, "start_frame": 0, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "native_dense", "status": "measured", "q_shape": [10560, 40, 128], "k_shape": [512, 40, 128], "q_len": 10560, "k_len": 512, "reference_error": {"max_abs": 0.0, "rmse": 0.0, "relative_l2": 0.0, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.00043489744124647345, "rmse": 1.7601063432965562e-05, "relative_l2": 0.001762362604457805, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.18748800456523895, "p95_ms": 0.18854400515556335, "p99_ms": 0.18880096703767776, "mean_ms": 0.1878402136762937, "n": 300}, {"repeat": 1, "p50_ms": 0.187391996383667, "p95_ms": 0.18915360197424888, "p99_ms": 0.18969695910811424, "mean_ms": 0.18846560001373291, "n": 300}, {"repeat": 2, "p50_ms": 0.18719999492168427, "p95_ms": 0.18960000574588776, "p99_ms": 0.1900495943427086, "mean_ms": 0.18822314699490866, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_847f3e9ceb4b37a4_native_dense.json", "configuration": "704x1280_tp1_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_dit_only_704x1280_tp1_sp1_v2", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_dit_only_704x1280_tp1_sp1_v2/captures/qkv_rank0_883584e23546203b.pt", "context": {"phase": "commit", "step": null, "start_frame": 27, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "vllm_fa2", "status": "measured", "q_shape": [10560, 40, 128], "k_shape": [66880, 40, 128], "q_len": 10560, "k_len": 63360, "reference_error": {"max_abs": 0.0625, "rmse": 0.00036308608325687203, "relative_l2": 0.00047817375707571655, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.00992615478884451, "rmse": 0.0009261728612270919, "relative_l2": 0.0017342987872258817, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 36.00105667114258, "p95_ms": 36.77894611358643, "p99_ms": 37.86806503295898, "mean_ms": 36.10828637440999, "n": 300}, {"repeat": 1, "p50_ms": 36.071327209472656, "p95_ms": 36.70859699249267, "p99_ms": 36.72662437438965, "mean_ms": 36.116687049865725, "n": 300}, {"repeat": 2, "p50_ms": 36.07356834411621, "p95_ms": 36.70777816772461, "p99_ms": 36.71821853637695, "mean_ms": 36.09740272521972, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_883584e23546203b_vllm_fa2.json", "configuration": "704x1280_tp1_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_dit_only_704x1280_tp1_sp1_v2", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_dit_only_704x1280_tp1_sp1_v2/captures/qkv_rank0_883584e23546203b.pt", "context": {"phase": "commit", "step": null, "start_frame": 27, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "vllm_fa3", "status": "measured", "q_shape": [10560, 40, 128], "k_shape": [66880, 40, 128], "q_len": 10560, "k_len": 63360, "reference_error": {"max_abs": 0.0, "rmse": 0.0, "relative_l2": 0.0, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.010579403963054013, "rmse": 0.0009308798019245433, "relative_l2": 0.0017431127374989645, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 20.01806354522705, "p95_ms": 20.64001684188843, "p99_ms": 21.338599586486815, "mean_ms": 20.038706690470377, "n": 300}, {"repeat": 1, "p50_ms": 20.017632484436035, "p95_ms": 20.24172878265381, "p99_ms": 20.632615795135496, "mean_ms": 19.97959108352661, "n": 300}, {"repeat": 2, "p50_ms": 20.020639419555664, "p95_ms": 20.23396921157837, "p99_ms": 20.424229640960693, "mean_ms": 20.037266171773275, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_883584e23546203b_vllm_fa3.json", "configuration": "704x1280_tp1_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_dit_only_704x1280_tp1_sp1_v2", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_dit_only_704x1280_tp1_sp1_v2/captures/qkv_rank0_883584e23546203b.pt", "context": {"phase": "commit", "step": null, "start_frame": 27, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_math", "status": "oom", "q_shape": [10560, 40, 128], "k_shape": [66880, 40, 128], "q_len": 10560, "k_len": 63360, "reference_error": null, "oracle_error": null, "repeats": null, "error": "CUDA out of memory. Tried to allocate 99.70 GiB. GPU 0 has a total capacity of 139.81 GiB of which 34.50 GiB is free. Including non-PyTorch memory, this process has 105.30 GiB memory in use. Of the allocated memory 104.62 GiB is allocated by PyTorch, and 3.25 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://docs.pytorch.org/docs/stable/notes/cuda.html#optimizing-memory-usage-with-pytorch-cuda-alloc-conf)", "dispatch_trace": null, "configuration": "704x1280_tp1_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_dit_only_704x1280_tp1_sp1_v2", "audit_status": "recorded_failure"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_dit_only_704x1280_tp1_sp1_v2/captures/qkv_rank0_883584e23546203b.pt", "context": {"phase": "commit", "step": null, "start_frame": 27, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_flash", "status": "measured", "q_shape": [10560, 40, 128], "k_shape": [66880, 40, 128], "q_len": 10560, "k_len": 63360, "reference_error": {"max_abs": 0.0625, "rmse": 0.0003640728742468385, "relative_l2": 0.00047947333196134194, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.00992615478884451, "rmse": 0.0009261736710817458, "relative_l2": 0.001734300303713789, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 38.8860969543457, "p95_ms": 39.771784973144534, "p99_ms": 40.503518180847166, "mean_ms": 38.886136258443194, "n": 300}, {"repeat": 1, "p50_ms": 38.91908645629883, "p95_ms": 39.10256938934326, "p99_ms": 39.32148975372314, "mean_ms": 38.84960786183675, "n": 300}, {"repeat": 2, "p50_ms": 38.8953914642334, "p95_ms": 39.09829750061035, "p99_ms": 39.27394199371338, "mean_ms": 38.84095855712891, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_883584e23546203b_sdpa_flash.json", "configuration": "704x1280_tp1_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_dit_only_704x1280_tp1_sp1_v2", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_dit_only_704x1280_tp1_sp1_v2/captures/qkv_rank0_883584e23546203b.pt", "context": {"phase": "commit", "step": null, "start_frame": 27, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_efficient", "status": "measured", "q_shape": [10560, 40, 128], "k_shape": [66880, 40, 128], "q_len": 10560, "k_len": 63360, "reference_error": {"max_abs": 0.0625, "rmse": 0.0007647648352198849, "relative_l2": 0.0010071729306071174, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.008974668806738695, "rmse": 0.0009140189670257117, "relative_l2": 0.0017115400940532074, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 75.1886100769043, "p95_ms": 75.34743537902833, "p99_ms": 75.36686744689942, "mean_ms": 75.22650085449219, "n": 300}, {"repeat": 1, "p50_ms": 75.18736267089844, "p95_ms": 75.34989471435547, "p99_ms": 75.3596322631836, "mean_ms": 75.22587893168132, "n": 300}, {"repeat": 2, "p50_ms": 75.18315505981445, "p95_ms": 75.34827537536621, "p99_ms": 75.36599327087403, "mean_ms": 75.22287251790364, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_883584e23546203b_sdpa_efficient.json", "configuration": "704x1280_tp1_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_dit_only_704x1280_tp1_sp1_v2", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_dit_only_704x1280_tp1_sp1_v2/captures/qkv_rank0_883584e23546203b.pt", "context": {"phase": "commit", "step": null, "start_frame": 27, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_cudnn", "status": "measured", "q_shape": [10560, 40, 128], "k_shape": [66880, 40, 128], "q_len": 10560, "k_len": 63360, "reference_error": {"max_abs": 0.0625, "rmse": 0.0007651511965023901, "relative_l2": 0.0010076817571211688, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.008974668806738695, "rmse": 0.0009150847472603187, "relative_l2": 0.001713535813692281, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 22.2183198928833, "p95_ms": 23.239272022247317, "p99_ms": 24.798551559448235, "mean_ms": 22.372761166890463, "n": 300}, {"repeat": 1, "p50_ms": 22.405263900756836, "p95_ms": 22.63584566116333, "p99_ms": 22.811920680999755, "mean_ms": 22.339016952514648, "n": 300}, {"repeat": 2, "p50_ms": 22.4060640335083, "p95_ms": 22.633831405639647, "p99_ms": 22.63670509338379, "mean_ms": 22.349112758636476, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_883584e23546203b_sdpa_cudnn.json", "configuration": "704x1280_tp1_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_dit_only_704x1280_tp1_sp1_v2", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_dit_only_704x1280_tp1_sp1_v2/captures/qkv_rank0_883584e23546203b.pt", "context": {"phase": "commit", "step": null, "start_frame": 27, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "native_dense", "status": "measured", "q_shape": [10560, 40, 128], "k_shape": [66880, 40, 128], "q_len": 10560, "k_len": 63360, "reference_error": {"max_abs": 0.0625, "rmse": 0.00034756432040626225, "relative_l2": 0.00045773204916959594, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.010579403963054013, "rmse": 0.0009298519468354258, "relative_l2": 0.0017411880343370311, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 19.95235252380371, "p95_ms": 21.16257972717285, "p99_ms": 22.13080606460571, "mean_ms": 20.094276383717855, "n": 300}, {"repeat": 1, "p50_ms": 20.089632034301758, "p95_ms": 20.350982666015625, "p99_ms": 20.558177852630614, "mean_ms": 20.039921480814616, "n": 300}, {"repeat": 2, "p50_ms": 20.083279609680176, "p95_ms": 20.33346109390259, "p99_ms": 20.452449474334717, "mean_ms": 20.032189458211263, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_883584e23546203b_native_dense.json", "configuration": "704x1280_tp1_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_dit_only_704x1280_tp1_sp1_v2", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_dit_only_704x1280_tp1_sp1_v2/captures/qkv_rank0_ef7fffa69ac75786.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 0, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "vllm_fa2", "status": "measured", "q_shape": [10560, 40, 128], "k_shape": [512, 40, 128], "q_len": 10560, "k_len": 512, "reference_error": {"max_abs": 0.0009765625, "rmse": 2.358827629875496e-06, "relative_l2": 0.0002753269825014281, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.0004067786029473136, "rmse": 1.930963898854266e-05, "relative_l2": 0.0018472856840449385, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.3242560029029846, "p95_ms": 0.3266240060329437, "p99_ms": 0.32736256420612336, "mean_ms": 0.32462549328804013, "n": 300}, {"repeat": 1, "p50_ms": 0.32412800192832947, "p95_ms": 0.3250559866428375, "p99_ms": 0.32579201459884644, "mean_ms": 0.32479775995016097, "n": 300}, {"repeat": 2, "p50_ms": 0.3242720067501068, "p95_ms": 0.32524800300598145, "p99_ms": 0.3255695906281471, "mean_ms": 0.32506773273150125, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_ef7fffa69ac75786_vllm_fa2.json", "configuration": "704x1280_tp1_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_dit_only_704x1280_tp1_sp1_v2", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_dit_only_704x1280_tp1_sp1_v2/captures/qkv_rank0_ef7fffa69ac75786.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 0, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "vllm_fa3", "status": "measured", "q_shape": [10560, 40, 128], "k_shape": [512, 40, 128], "q_len": 10560, "k_len": 512, "reference_error": {"max_abs": 0.00048828125, "rmse": 8.875116763085793e-07, "relative_l2": 0.00010359210171949823, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.0004067786029473136, "rmse": 1.936247177353708e-05, "relative_l2": 0.0018523400119599413, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.2643359899520874, "p95_ms": 0.2688640058040619, "p99_ms": 0.27008096009492877, "mean_ms": 0.26497717211643856, "n": 300}, {"repeat": 1, "p50_ms": 0.26451200246810913, "p95_ms": 0.26867198944091797, "p99_ms": 0.27024064868688585, "mean_ms": 0.2654658127824465, "n": 300}, {"repeat": 2, "p50_ms": 0.2641279995441437, "p95_ms": 0.26851360946893693, "p99_ms": 0.2696364915370941, "mean_ms": 0.26510207980871203, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_ef7fffa69ac75786_vllm_fa3.json", "configuration": "704x1280_tp1_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_dit_only_704x1280_tp1_sp1_v2", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_dit_only_704x1280_tp1_sp1_v2/captures/qkv_rank0_ef7fffa69ac75786.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 0, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_math", "status": "measured", "q_shape": [10560, 40, 128], "k_shape": [512, 40, 128], "q_len": 10560, "k_len": 512, "reference_error": {"max_abs": 0.001953125, "rmse": 8.647488410216817e-06, "relative_l2": 0.001009351789866365, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.0003842328223661873, "rmse": 1.6787201932200433e-05, "relative_l2": 0.0016059729455804724, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 4.65123176574707, "p95_ms": 4.656358432769776, "p99_ms": 4.658771805763244, "mean_ms": 4.651349220275879, "n": 300}, {"repeat": 1, "p50_ms": 4.650928020477295, "p95_ms": 4.6561598777771, "p99_ms": 4.6578275585174564, "mean_ms": 4.651221555074056, "n": 300}, {"repeat": 2, "p50_ms": 4.651455879211426, "p95_ms": 4.657601523399353, "p99_ms": 4.661197276115417, "mean_ms": 4.651674779256185, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_ef7fffa69ac75786_sdpa_math.json", "configuration": "704x1280_tp1_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_dit_only_704x1280_tp1_sp1_v2", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_dit_only_704x1280_tp1_sp1_v2/captures/qkv_rank0_ef7fffa69ac75786.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 0, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_flash", "status": "measured", "q_shape": [10560, 40, 128], "k_shape": [512, 40, 128], "q_len": 10560, "k_len": 512, "reference_error": {"max_abs": 0.0009765625, "rmse": 2.360198143520804e-06, "relative_l2": 0.00027548695153929275, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.0004067786029473136, "rmse": 1.930963898854266e-05, "relative_l2": 0.0018472856840449385, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.3465439975261688, "p95_ms": 0.34940800070762634, "p99_ms": 0.34992192447185516, "mean_ms": 0.3470385068655014, "n": 300}, {"repeat": 1, "p50_ms": 0.34624001383781433, "p95_ms": 0.3471056118607521, "p99_ms": 0.34755328148603437, "mean_ms": 0.3466729595263799, "n": 300}, {"repeat": 2, "p50_ms": 0.34620800614356995, "p95_ms": 0.34723361283540727, "p99_ms": 0.347680002450943, "mean_ms": 0.34662325332562127, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_ef7fffa69ac75786_sdpa_flash.json", "configuration": "704x1280_tp1_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_dit_only_704x1280_tp1_sp1_v2", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_dit_only_704x1280_tp1_sp1_v2/captures/qkv_rank0_ef7fffa69ac75786.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 0, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_efficient", "status": "measured", "q_shape": [10560, 40, 128], "k_shape": [512, 40, 128], "q_len": 10560, "k_len": 512, "reference_error": {"max_abs": 0.0009765625, "rmse": 4.15795898420779e-06, "relative_l2": 0.0004853251191343128, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.0004067786029473136, "rmse": 1.9527557065743348e-05, "relative_l2": 0.001868133144970867, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.6599360108375549, "p95_ms": 0.6611536055803299, "p99_ms": 0.661954578757286, "mean_ms": 0.6601458116372426, "n": 300}, {"repeat": 1, "p50_ms": 0.6602240204811096, "p95_ms": 0.6615999937057495, "p99_ms": 0.6621139174699784, "mean_ms": 0.6605333292484283, "n": 300}, {"repeat": 2, "p50_ms": 0.6599200069904327, "p95_ms": 0.6608655959367752, "p99_ms": 0.6611523252725601, "mean_ms": 0.6603557324409485, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_ef7fffa69ac75786_sdpa_efficient.json", "configuration": "704x1280_tp1_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_dit_only_704x1280_tp1_sp1_v2", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_dit_only_704x1280_tp1_sp1_v2/captures/qkv_rank0_ef7fffa69ac75786.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 0, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_cudnn", "status": "measured", "q_shape": [10560, 40, 128], "k_shape": [512, 40, 128], "q_len": 10560, "k_len": 512, "reference_error": {"max_abs": 0.0009765625, "rmse": 4.127908307752139e-06, "relative_l2": 0.0004818175453015025, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.0004067786029473136, "rmse": 1.952755709178034e-05, "relative_l2": 0.0018681331474617357, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.19404800236225128, "p95_ms": 0.19638719782233238, "p99_ms": 0.19734432503581048, "mean_ms": 0.1944941864411036, "n": 300}, {"repeat": 1, "p50_ms": 0.1968960016965866, "p95_ms": 0.19782400131225586, "p99_ms": 0.19801760271191596, "mean_ms": 0.19733824014663695, "n": 300}, {"repeat": 2, "p50_ms": 0.19622400403022766, "p95_ms": 0.1974399983882904, "p99_ms": 0.19776128083467484, "mean_ms": 0.19664341419935227, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_ef7fffa69ac75786_sdpa_cudnn.json", "configuration": "704x1280_tp1_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_dit_only_704x1280_tp1_sp1_v2", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_dit_only_704x1280_tp1_sp1_v2/captures/qkv_rank0_ef7fffa69ac75786.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 0, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "native_dense", "status": "measured", "q_shape": [10560, 40, 128], "k_shape": [512, 40, 128], "q_len": 10560, "k_len": 512, "reference_error": {"max_abs": 0.0, "rmse": 0.0, "relative_l2": 0.0, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.0004067786029473136, "rmse": 1.9372534423903118e-05, "relative_l2": 0.001853302670556291, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.18809600174427032, "p95_ms": 0.18934400379657745, "p99_ms": 0.18985695287585258, "mean_ms": 0.18838581318656603, "n": 300}, {"repeat": 1, "p50_ms": 0.18908800184726715, "p95_ms": 0.19049599766731262, "p99_ms": 0.19088064059615134, "mean_ms": 0.18974442690610885, "n": 300}, {"repeat": 2, "p50_ms": 0.1884320005774498, "p95_ms": 0.1906576059758663, "p99_ms": 0.19158592596650123, "mean_ms": 0.1891430400311947, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_ef7fffa69ac75786_native_dense.json", "configuration": "704x1280_tp1_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_dit_only_704x1280_tp1_sp1_v2", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp1_sp1_v1/captures/qkv_rank0_034f783e4e41e1c0.pt", "context": {"phase": "commit", "step": null, "start_frame": 0, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "vllm_fa2", "status": "measured", "q_shape": [4680, 40, 128], "k_shape": [512, 40, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.00048828125, "rmse": 1.662243295694675e-06, "relative_l2": 0.00019691306908771765, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.00025586326656432123, "rmse": 1.8789914191711385e-05, "relative_l2": 0.0017170595682592, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.15583999454975128, "p95_ms": 0.15667360052466392, "p99_ms": 0.15718400478363037, "mean_ms": 0.1559857063492139, "n": 300}, {"repeat": 1, "p50_ms": 0.15568000078201294, "p95_ms": 0.15651680678129196, "p99_ms": 0.15695999562740326, "mean_ms": 0.156098240117232, "n": 300}, {"repeat": 2, "p50_ms": 0.15568000078201294, "p95_ms": 0.15663999319076538, "p99_ms": 0.15721727758646012, "mean_ms": 0.15603423972924552, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_034f783e4e41e1c0_vllm_fa2.json", "configuration": "480x832_tp1_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp1_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp1_sp1_v1/captures/qkv_rank0_034f783e4e41e1c0.pt", "context": {"phase": "commit", "step": null, "start_frame": 0, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "vllm_fa3", "status": "measured", "q_shape": [4680, 40, 128], "k_shape": [512, 40, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.00048828125, "rmse": 6.46612970484491e-07, "relative_l2": 7.659922278514309e-05, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.00025586326656432123, "rmse": 1.8794715046210773e-05, "relative_l2": 0.0017174982798504232, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.12836799770593643, "p95_ms": 0.12938079461455346, "p99_ms": 0.13001855909824372, "mean_ms": 0.12864960014820098, "n": 300}, {"repeat": 1, "p50_ms": 0.1284479945898056, "p95_ms": 0.12953760251402854, "p99_ms": 0.13014559999108313, "mean_ms": 0.1289168001214663, "n": 300}, {"repeat": 2, "p50_ms": 0.1284479945898056, "p95_ms": 0.1295360028743744, "p99_ms": 0.12976063713431357, "mean_ms": 0.12889802659551303, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_034f783e4e41e1c0_vllm_fa3.json", "configuration": "480x832_tp1_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp1_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp1_sp1_v1/captures/qkv_rank0_034f783e4e41e1c0.pt", "context": {"phase": "commit", "step": null, "start_frame": 0, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_math", "status": "measured", "q_shape": [4680, 40, 128], "k_shape": [512, 40, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.001953125, "rmse": 4.196141354132547e-06, "relative_l2": 0.0004970843164224159, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.00024379653738876295, "rmse": 1.860982023786444e-05, "relative_l2": 0.0017006022261189715, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 2.1432000398635864, "p95_ms": 2.147007942199707, "p99_ms": 2.14844927072525, "mean_ms": 2.143589865366618, "n": 300}, {"repeat": 1, "p50_ms": 2.1433759927749634, "p95_ms": 2.1479055285453796, "p99_ms": 2.150025987625122, "mean_ms": 2.1438634546597797, "n": 300}, {"repeat": 2, "p50_ms": 2.1431360244750977, "p95_ms": 2.14630389213562, "p99_ms": 2.1477442598342895, "mean_ms": 2.1435194714864094, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_034f783e4e41e1c0_sdpa_math.json", "configuration": "480x832_tp1_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp1_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp1_sp1_v1/captures/qkv_rank0_034f783e4e41e1c0.pt", "context": {"phase": "commit", "step": null, "start_frame": 0, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_flash", "status": "measured", "q_shape": [4680, 40, 128], "k_shape": [512, 40, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.00048828125, "rmse": 1.6583713065147254e-06, "relative_l2": 0.00019645438456489654, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.00025586326656432123, "rmse": 1.8789914191711385e-05, "relative_l2": 0.0017170595682592, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.16700799763202667, "p95_ms": 0.16812959909439087, "p99_ms": 0.1685788820683956, "mean_ms": 0.1671758935848872, "n": 300}, {"repeat": 1, "p50_ms": 0.16687999665737152, "p95_ms": 0.16774719581007957, "p99_ms": 0.1682899272441864, "mean_ms": 0.1672004270553589, "n": 300}, {"repeat": 2, "p50_ms": 0.16700799763202667, "p95_ms": 0.16841759383678437, "p99_ms": 0.16912127882242203, "mean_ms": 0.16741557319959005, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_034f783e4e41e1c0_sdpa_flash.json", "configuration": "480x832_tp1_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp1_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp1_sp1_v1/captures/qkv_rank0_034f783e4e41e1c0.pt", "context": {"phase": "commit", "step": null, "start_frame": 0, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_efficient", "status": "measured", "q_shape": [4680, 40, 128], "k_shape": [512, 40, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.0009765625, "rmse": 2.9304238840428427e-06, "relative_l2": 0.0003471445860118049, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.00025586326656432123, "rmse": 1.8933874262258647e-05, "relative_l2": 0.0017302149245880482, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.3084160089492798, "p95_ms": 0.30924960076808927, "p99_ms": 0.3096028757095337, "mean_ms": 0.30855050692955655, "n": 300}, {"repeat": 1, "p50_ms": 0.30847999453544617, "p95_ms": 0.30934399366378784, "p99_ms": 0.3098883253335953, "mean_ms": 0.3087692799170812, "n": 300}, {"repeat": 2, "p50_ms": 0.3086079955101013, "p95_ms": 0.3094735950231552, "p99_ms": 0.31030686616897585, "mean_ms": 0.3089064535498619, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_034f783e4e41e1c0_sdpa_efficient.json", "configuration": "480x832_tp1_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp1_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp1_sp1_v1/captures/qkv_rank0_034f783e4e41e1c0.pt", "context": {"phase": "commit", "step": null, "start_frame": 0, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_cudnn", "status": "measured", "q_shape": [4680, 40, 128], "k_shape": [512, 40, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.0009765625, "rmse": 2.926237675645e-06, "relative_l2": 0.0003466486783756634, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.00025586326656432123, "rmse": 1.893387426216046e-05, "relative_l2": 0.0017302149245790755, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.09510400146245956, "p95_ms": 0.09590560011565685, "p99_ms": 0.09619232229888439, "mean_ms": 0.09533248014748097, "n": 300}, {"repeat": 1, "p50_ms": 0.09504000097513199, "p95_ms": 0.09567999839782715, "p99_ms": 0.0962886430323124, "mean_ms": 0.09536405377089978, "n": 300}, {"repeat": 2, "p50_ms": 0.09494400024414062, "p95_ms": 0.0955215971916914, "p99_ms": 0.09580863937735558, "mean_ms": 0.09521013344327608, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_034f783e4e41e1c0_sdpa_cudnn.json", "configuration": "480x832_tp1_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp1_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp1_sp1_v1/captures/qkv_rank0_034f783e4e41e1c0.pt", "context": {"phase": "commit", "step": null, "start_frame": 0, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "native_dense", "status": "measured", "q_shape": [4680, 40, 128], "k_shape": [512, 40, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.0, "rmse": 0.0, "relative_l2": 0.0, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.00025586326656432123, "rmse": 1.879633017463274e-05, "relative_l2": 0.0017176458734840402, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.11883200332522392, "p95_ms": 0.12980160489678383, "p99_ms": 0.15953696593642214, "mean_ms": 0.12139872014522553, "n": 300}, {"repeat": 1, "p50_ms": 0.11838399991393089, "p95_ms": 0.13306079953908923, "p99_ms": 0.18354560554027483, "mean_ms": 0.1274739202360312, "n": 300}, {"repeat": 2, "p50_ms": 0.11846400052309036, "p95_ms": 0.13901920020580294, "p99_ms": 0.1868735997378826, "mean_ms": 0.12187722663084666, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_034f783e4e41e1c0_native_dense.json", "configuration": "480x832_tp1_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp1_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp1_sp1_v1/captures/qkv_rank0_3019e6cfd6e84b2b.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 27, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "vllm_fa2", "status": "measured", "q_shape": [4680, 40, 128], "k_shape": [29640, 40, 128], "q_len": 4680, "k_len": 28080, "reference_error": {"max_abs": 0.03125, "rmse": 0.0004348162159959158, "relative_l2": 0.0008792996434419078, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.008820491288137777, "rmse": 0.0008120587250742453, "relative_l2": 0.0019695529708380166, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 7.253151893615723, "p95_ms": 7.655353665351868, "p99_ms": 8.232817735671997, "mean_ms": 7.347301425933838, "n": 300}, {"repeat": 1, "p50_ms": 7.322448015213013, "p95_ms": 7.5710800886154175, "p99_ms": 7.635991568565369, "mean_ms": 7.36504586537679, "n": 300}, {"repeat": 2, "p50_ms": 7.377392053604126, "p95_ms": 7.569121527671814, "p99_ms": 7.571588063240052, "mean_ms": 7.392029646237691, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_3019e6cfd6e84b2b_vllm_fa2.json", "configuration": "480x832_tp1_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp1_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp1_sp1_v1/captures/qkv_rank0_3019e6cfd6e84b2b.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 27, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "vllm_fa3", "status": "measured", "q_shape": [4680, 40, 128], "k_shape": [29640, 40, 128], "q_len": 4680, "k_len": 28080, "reference_error": {"max_abs": 0.0, "rmse": 0.0, "relative_l2": 0.0, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.008820491288137777, "rmse": 0.0008167053351597663, "relative_l2": 0.0019808227773380664, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 4.101680040359497, "p95_ms": 4.517087936401367, "p99_ms": 4.778965611457824, "mean_ms": 4.152604250907898, "n": 300}, {"repeat": 1, "p50_ms": 4.143551826477051, "p95_ms": 4.358297681808471, "p99_ms": 4.36269606590271, "mean_ms": 4.1462031141916915, "n": 300}, {"repeat": 2, "p50_ms": 4.163631916046143, "p95_ms": 4.270212697982788, "p99_ms": 4.312289409637451, "mean_ms": 4.143507834275564, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_3019e6cfd6e84b2b_vllm_fa3.json", "configuration": "480x832_tp1_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp1_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp1_sp1_v1/captures/qkv_rank0_3019e6cfd6e84b2b.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 27, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_math", "status": "measured", "q_shape": [4680, 40, 128], "k_shape": [29640, 40, 128], "q_len": 4680, "k_len": 28080, "reference_error": {"max_abs": 0.03125, "rmse": 0.0007442528116819959, "relative_l2": 0.0015050524977402479, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.007724527303786921, "rmse": 0.0006858574812207812, "relative_l2": 0.0016634666902770664, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 105.15153503417969, "p95_ms": 105.66178741455079, "p99_ms": 105.69487289428712, "mean_ms": 105.21039644877116, "n": 300}, {"repeat": 1, "p50_ms": 105.61425399780273, "p95_ms": 105.7282283782959, "p99_ms": 105.87249923706055, "mean_ms": 105.47073608398438, "n": 300}, {"repeat": 2, "p50_ms": 105.63996887207031, "p95_ms": 105.7557861328125, "p99_ms": 105.90217254638672, "mean_ms": 105.51253211975097, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_3019e6cfd6e84b2b_sdpa_math.json", "configuration": "480x832_tp1_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp1_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp1_sp1_v1/captures/qkv_rank0_3019e6cfd6e84b2b.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 27, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_flash", "status": "measured", "q_shape": [4680, 40, 128], "k_shape": [29640, 40, 128], "q_len": 4680, "k_len": 28080, "reference_error": {"max_abs": 0.03125, "rmse": 0.0004353529875318782, "relative_l2": 0.0008803851204844336, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.008820491288137777, "rmse": 0.0008120515445332757, "relative_l2": 0.00196953555528005, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 7.912064075469971, "p95_ms": 8.333637237548828, "p99_ms": 8.410056495666504, "mean_ms": 7.957116921742757, "n": 300}, {"repeat": 1, "p50_ms": 7.966256141662598, "p95_ms": 8.109745454788207, "p99_ms": 8.141536922454835, "mean_ms": 7.958272422154744, "n": 300}, {"repeat": 2, "p50_ms": 7.9191038608551025, "p95_ms": 8.081180334091187, "p99_ms": 8.125607538223267, "mean_ms": 7.9526588916778564, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_3019e6cfd6e84b2b_sdpa_flash.json", "configuration": "480x832_tp1_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp1_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp1_sp1_v1/captures/qkv_rank0_3019e6cfd6e84b2b.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 27, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_efficient", "status": "measured", "q_shape": [4680, 40, 128], "k_shape": [29640, 40, 128], "q_len": 4680, "k_len": 28080, "reference_error": {"max_abs": 0.03125, "rmse": 0.0006306339979739321, "relative_l2": 0.0012752887982586919, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.008820491288137777, "rmse": 0.000811379428513884, "relative_l2": 0.001967905416889973, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 15.469472408294678, "p95_ms": 15.497402000427247, "p99_ms": 15.507747993469238, "mean_ms": 15.471780360539753, "n": 300}, {"repeat": 1, "p50_ms": 15.475247859954834, "p95_ms": 15.496871900558471, "p99_ms": 15.50135588645935, "mean_ms": 15.47369055112203, "n": 300}, {"repeat": 2, "p50_ms": 15.469568252563477, "p95_ms": 15.500657510757446, "p99_ms": 15.509058647155761, "mean_ms": 15.472233400344848, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_3019e6cfd6e84b2b_sdpa_efficient.json", "configuration": "480x832_tp1_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp1_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp1_sp1_v1/captures/qkv_rank0_3019e6cfd6e84b2b.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 27, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_cudnn", "status": "measured", "q_shape": [4680, 40, 128], "k_shape": [29640, 40, 128], "q_len": 4680, "k_len": 28080, "reference_error": {"max_abs": 0.03125, "rmse": 0.0006300878851718787, "relative_l2": 0.00127418443100085, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.008820491288137777, "rmse": 0.0008098227142425212, "relative_l2": 0.001964129789434424, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 4.8501598834991455, "p95_ms": 5.242601776123047, "p99_ms": 5.587009921073913, "mean_ms": 4.90286656220754, "n": 300}, {"repeat": 1, "p50_ms": 4.905503988265991, "p95_ms": 5.087574458122253, "p99_ms": 5.096741394996643, "mean_ms": 4.92415904045105, "n": 300}, {"repeat": 2, "p50_ms": 4.906800031661987, "p95_ms": 5.035327911376953, "p99_ms": 5.061650233268738, "mean_ms": 4.911470816930135, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_3019e6cfd6e84b2b_sdpa_cudnn.json", "configuration": "480x832_tp1_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp1_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp1_sp1_v1/captures/qkv_rank0_3019e6cfd6e84b2b.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 27, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "native_dense", "status": "measured", "q_shape": [4680, 40, 128], "k_shape": [29640, 40, 128], "q_len": 4680, "k_len": 28080, "reference_error": {"max_abs": 0.03125, "rmse": 0.0004025621359403169, "relative_l2": 0.0008140743826326347, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.007724527303786921, "rmse": 0.0008217814009871526, "relative_l2": 0.0019931341782525556, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 4.145632028579712, "p95_ms": 4.365801811218263, "p99_ms": 4.568429932594299, "mean_ms": 4.14625013589859, "n": 300}, {"repeat": 1, "p50_ms": 4.13318395614624, "p95_ms": 4.362577414512634, "p99_ms": 4.40860924243927, "mean_ms": 4.14617150704066, "n": 300}, {"repeat": 2, "p50_ms": 4.167407989501953, "p95_ms": 4.333227372169494, "p99_ms": 4.3780409574508665, "mean_ms": 4.153146243890126, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_3019e6cfd6e84b2b_native_dense.json", "configuration": "480x832_tp1_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp1_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp1_sp1_v1/captures/qkv_rank0_53c9bd15f7acb7c0.pt", "context": {"phase": "commit", "step": null, "start_frame": 27, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "vllm_fa2", "status": "measured", "q_shape": [4680, 40, 128], "k_shape": [512, 40, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.000244140625, "rmse": 1.4939722527085189e-06, "relative_l2": 0.00019791011277052472, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 6.036895268355172e-05, "rmse": 3.8066916294210143e-06, "relative_l2": 0.0017721563376145706, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.1568640023469925, "p95_ms": 0.15785600244998932, "p99_ms": 0.1583046458661556, "mean_ms": 0.15709621315201125, "n": 300}, {"repeat": 1, "p50_ms": 0.15680000185966492, "p95_ms": 0.15760160014033317, "p99_ms": 0.15776063427329062, "mean_ms": 0.1571909332772096, "n": 300}, {"repeat": 2, "p50_ms": 0.15695999562740326, "p95_ms": 0.1578303948044777, "p99_ms": 0.1583372786641121, "mean_ms": 0.15731562654177347, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_53c9bd15f7acb7c0_vllm_fa2.json", "configuration": "480x832_tp1_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp1_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp1_sp1_v1/captures/qkv_rank0_53c9bd15f7acb7c0.pt", "context": {"phase": "commit", "step": null, "start_frame": 27, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "vllm_fa3", "status": "measured", "q_shape": [4680, 40, 128], "k_shape": [512, 40, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.000244140625, "rmse": 5.704917836581556e-07, "relative_l2": 7.557442451407555e-05, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 6.036895268355172e-05, "rmse": 3.834593322207515e-06, "relative_l2": 0.0017851456119018848, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.12828800082206726, "p95_ms": 0.12931199371814728, "p99_ms": 0.13011360719799994, "mean_ms": 0.12853045329451562, "n": 300}, {"repeat": 1, "p50_ms": 0.12812800705432892, "p95_ms": 0.12918399274349213, "p99_ms": 0.12976063713431357, "mean_ms": 0.12857322658101716, "n": 300}, {"repeat": 2, "p50_ms": 0.12814400345087051, "p95_ms": 0.12905760630965232, "p99_ms": 0.12973024427890778, "mean_ms": 0.12857141335805258, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_53c9bd15f7acb7c0_vllm_fa3.json", "configuration": "480x832_tp1_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp1_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp1_sp1_v1/captures/qkv_rank0_53c9bd15f7acb7c0.pt", "context": {"phase": "commit", "step": null, "start_frame": 27, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_math", "status": "measured", "q_shape": [4680, 40, 128], "k_shape": [512, 40, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.001953125, "rmse": 3.564230695697496e-06, "relative_l2": 0.0004721622490958563, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 6.036895268355172e-05, "rmse": 3.591073235205049e-06, "relative_l2": 0.0016717779668364214, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 2.144864082336426, "p95_ms": 2.148227310180664, "p99_ms": 2.150273172855377, "mean_ms": 2.145516262849172, "n": 300}, {"repeat": 1, "p50_ms": 2.1456000804901123, "p95_ms": 2.149440050125122, "p99_ms": 2.151392624378204, "mean_ms": 2.1458315809567767, "n": 300}, {"repeat": 2, "p50_ms": 2.14465594291687, "p95_ms": 2.1485150933265684, "p99_ms": 2.1509146070480347, "mean_ms": 2.1450547258059185, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_53c9bd15f7acb7c0_sdpa_math.json", "configuration": "480x832_tp1_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp1_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp1_sp1_v1/captures/qkv_rank0_53c9bd15f7acb7c0.pt", "context": {"phase": "commit", "step": null, "start_frame": 27, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_flash", "status": "measured", "q_shape": [4680, 40, 128], "k_shape": [512, 40, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.000244140625, "rmse": 1.4940423774061753e-06, "relative_l2": 0.00019791940235859838, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 6.036895268355172e-05, "rmse": 3.8066900674542536e-06, "relative_l2": 0.0017721556104610306, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.16704000532627106, "p95_ms": 0.16787359714508057, "p99_ms": 0.1687046431005001, "mean_ms": 0.1671912533044815, "n": 300}, {"repeat": 1, "p50_ms": 0.16713599860668182, "p95_ms": 0.16812799870967865, "p99_ms": 0.1686412826180458, "mean_ms": 0.16743221337596575, "n": 300}, {"repeat": 2, "p50_ms": 0.16683200001716614, "p95_ms": 0.1677439957857132, "p99_ms": 0.16832064017653464, "mean_ms": 0.16714826653401058, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_53c9bd15f7acb7c0_sdpa_flash.json", "configuration": "480x832_tp1_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp1_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp1_sp1_v1/captures/qkv_rank0_53c9bd15f7acb7c0.pt", "context": {"phase": "commit", "step": null, "start_frame": 27, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_efficient", "status": "measured", "q_shape": [4680, 40, 128], "k_shape": [512, 40, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.00048828125, "rmse": 2.6246665498487903e-06, "relative_l2": 0.00034769591732634734, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 6.036895268355172e-05, "rmse": 4.3905271507836556e-06, "relative_l2": 0.0020439534570110473, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.3086079955101013, "p95_ms": 0.3094080090522766, "p99_ms": 0.3096646508574486, "mean_ms": 0.3087334403395653, "n": 300}, {"repeat": 1, "p50_ms": 0.3086879998445511, "p95_ms": 0.30953601002693176, "p99_ms": 0.31001632630825043, "mean_ms": 0.3089323737223943, "n": 300}, {"repeat": 2, "p50_ms": 0.308351993560791, "p95_ms": 0.3091199994087219, "p99_ms": 0.30957022815942764, "mean_ms": 0.30869248022635776, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_53c9bd15f7acb7c0_sdpa_efficient.json", "configuration": "480x832_tp1_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp1_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp1_sp1_v1/captures/qkv_rank0_53c9bd15f7acb7c0.pt", "context": {"phase": "commit", "step": null, "start_frame": 27, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_cudnn", "status": "measured", "q_shape": [4680, 40, 128], "k_shape": [512, 40, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.00048828125, "rmse": 2.6236565079294788e-06, "relative_l2": 0.00034756211463369933, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 6.036895268355172e-05, "rmse": 4.3905005043652074e-06, "relative_l2": 0.0020439410521135865, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.0944959968328476, "p95_ms": 0.09504000097513199, "p99_ms": 0.0954252764582634, "mean_ms": 0.09474357297023138, "n": 300}, {"repeat": 1, "p50_ms": 0.09446399658918381, "p95_ms": 0.09513600170612335, "p99_ms": 0.09542495645582676, "mean_ms": 0.09494741303225358, "n": 300}, {"repeat": 2, "p50_ms": 0.09436800330877304, "p95_ms": 0.09484959952533245, "p99_ms": 0.09526560261845589, "mean_ms": 0.0947074136386315, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_53c9bd15f7acb7c0_sdpa_cudnn.json", "configuration": "480x832_tp1_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp1_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp1_sp1_v1/captures/qkv_rank0_53c9bd15f7acb7c0.pt", "context": {"phase": "commit", "step": null, "start_frame": 27, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "native_dense", "status": "measured", "q_shape": [4680, 40, 128], "k_shape": [512, 40, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.0, "rmse": 0.0, "relative_l2": 0.0, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 6.036895268355172e-05, "rmse": 3.843290768525026e-06, "relative_l2": 0.0017891945961940492, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.1167840026319027, "p95_ms": 0.12769120335578918, "p99_ms": 0.16029376596212386, "mean_ms": 0.11889023996889592, "n": 300}, {"repeat": 1, "p50_ms": 0.11668800190091133, "p95_ms": 0.1291055992245674, "p99_ms": 0.15341184377670264, "mean_ms": 0.11935327999293804, "n": 300}, {"repeat": 2, "p50_ms": 0.11670400202274323, "p95_ms": 0.12368480302393436, "p99_ms": 0.15516800403594969, "mean_ms": 0.11845045310755571, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_53c9bd15f7acb7c0_native_dense.json", "configuration": "480x832_tp1_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp1_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp1_sp1_v1/captures/qkv_rank0_7c21104bed6fe513.pt", "context": {"phase": "commit", "step": null, "start_frame": 9, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "vllm_fa2", "status": "measured", "q_shape": [4680, 40, 128], "k_shape": [512, 40, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.0009765625, "rmse": 1.505958327083979e-06, "relative_l2": 0.00019927001916073125, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 8.546650151804169e-05, "rmse": 4.432017925885755e-06, "relative_l2": 0.001751077866033789, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.15673600137233734, "p95_ms": 0.1575375996530056, "p99_ms": 0.15804800391197205, "mean_ms": 0.1570402130484581, "n": 300}, {"repeat": 1, "p50_ms": 0.15673600137233734, "p95_ms": 0.15763359367847443, "p99_ms": 0.15788927525281907, "mean_ms": 0.1571825064222018, "n": 300}, {"repeat": 2, "p50_ms": 0.15668799728155136, "p95_ms": 0.15747199952602386, "p99_ms": 0.15779520198702812, "mean_ms": 0.157094506919384, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_7c21104bed6fe513_vllm_fa2.json", "configuration": "480x832_tp1_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp1_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp1_sp1_v1/captures/qkv_rank0_7c21104bed6fe513.pt", "context": {"phase": "commit", "step": null, "start_frame": 9, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "vllm_fa3", "status": "measured", "q_shape": [4680, 40, 128], "k_shape": [512, 40, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.000244140625, "rmse": 5.653098539063279e-07, "relative_l2": 7.48024054807609e-05, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 8.546650151804169e-05, "rmse": 4.471097426135494e-06, "relative_l2": 0.0017665180670996055, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.12889599800109863, "p95_ms": 0.12988799810409546, "p99_ms": 0.1302752010524273, "mean_ms": 0.12915296028057735, "n": 300}, {"repeat": 1, "p50_ms": 0.12902399897575378, "p95_ms": 0.13017759323120118, "p99_ms": 0.13084864541888236, "mean_ms": 0.1294914132853349, "n": 300}, {"repeat": 2, "p50_ms": 0.12905600666999817, "p95_ms": 0.13007999956607819, "p99_ms": 0.13056031584739686, "mean_ms": 0.12949375996987025, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_7c21104bed6fe513_vllm_fa3.json", "configuration": "480x832_tp1_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp1_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp1_sp1_v1/captures/qkv_rank0_7c21104bed6fe513.pt", "context": {"phase": "commit", "step": null, "start_frame": 9, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_math", "status": "measured", "q_shape": [4680, 40, 128], "k_shape": [512, 40, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.001953125, "rmse": 3.551328468041258e-06, "relative_l2": 0.00046991558740069227, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 8.546650151804169e-05, "rmse": 4.251983876886612e-06, "relative_l2": 0.0016799469176471491, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 2.1452479362487793, "p95_ms": 2.1510480761528017, "p99_ms": 2.155617003440857, "mean_ms": 2.145508270263672, "n": 300}, {"repeat": 1, "p50_ms": 2.143791913986206, "p95_ms": 2.147684836387634, "p99_ms": 2.1492223381996154, "mean_ms": 2.144557767709096, "n": 300}, {"repeat": 2, "p50_ms": 2.144223928451538, "p95_ms": 2.1490575551986693, "p99_ms": 2.1510091710090635, "mean_ms": 2.1448210080464682, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_7c21104bed6fe513_sdpa_math.json", "configuration": "480x832_tp1_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp1_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp1_sp1_v1/captures/qkv_rank0_7c21104bed6fe513.pt", "context": {"phase": "commit", "step": null, "start_frame": 9, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_flash", "status": "measured", "q_shape": [4680, 40, 128], "k_shape": [512, 40, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.0009765625, "rmse": 1.506214088175611e-06, "relative_l2": 0.0001993038617423708, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 8.546650151804169e-05, "rmse": 4.432017925885755e-06, "relative_l2": 0.001751077866033789, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.1669600009918213, "p95_ms": 0.1680655986070633, "p99_ms": 0.16880223646759987, "mean_ms": 0.1671631998817126, "n": 300}, {"repeat": 1, "p50_ms": 0.1666879951953888, "p95_ms": 0.16767999529838562, "p99_ms": 0.16838560059666632, "mean_ms": 0.16700074632962544, "n": 300}, {"repeat": 2, "p50_ms": 0.16683200001716614, "p95_ms": 0.16771200299263, "p99_ms": 0.16851424157619477, "mean_ms": 0.1671389864385128, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_7c21104bed6fe513_sdpa_flash.json", "configuration": "480x832_tp1_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp1_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp1_sp1_v1/captures/qkv_rank0_7c21104bed6fe513.pt", "context": {"phase": "commit", "step": null, "start_frame": 9, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_efficient", "status": "measured", "q_shape": [4680, 40, 128], "k_shape": [512, 40, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.0009765625, "rmse": 2.636319746929073e-06, "relative_l2": 0.0003488406531816835, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 8.546650151804169e-05, "rmse": 5.084117071383684e-06, "relative_l2": 0.002008720411536978, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.3107520043849945, "p95_ms": 0.3115200102329254, "p99_ms": 0.3120979195833206, "mean_ms": 0.3108887462814649, "n": 300}, {"repeat": 1, "p50_ms": 0.31091201305389404, "p95_ms": 0.31158879548311236, "p99_ms": 0.3119683188199997, "mean_ms": 0.31118794629971186, "n": 300}, {"repeat": 2, "p50_ms": 0.3107840120792389, "p95_ms": 0.3115519881248474, "p99_ms": 0.31213024735450745, "mean_ms": 0.3111872002482414, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_7c21104bed6fe513_sdpa_efficient.json", "configuration": "480x832_tp1_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp1_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp1_sp1_v1/captures/qkv_rank0_7c21104bed6fe513.pt", "context": {"phase": "commit", "step": null, "start_frame": 9, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_cudnn", "status": "measured", "q_shape": [4680, 40, 128], "k_shape": [512, 40, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.0009765625, "rmse": 2.6357006047551833e-06, "relative_l2": 0.0003487587276259524, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 8.546650151804169e-05, "rmse": 5.0836227950647876e-06, "relative_l2": 0.002008525124348118, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.09523200243711472, "p95_ms": 0.09609760157763958, "p99_ms": 0.09648063704371453, "mean_ms": 0.09548512026667595, "n": 300}, {"repeat": 1, "p50_ms": 0.09513600170612335, "p95_ms": 0.09593600034713745, "p99_ms": 0.09664383828639984, "mean_ms": 0.09554144004980723, "n": 300}, {"repeat": 2, "p50_ms": 0.09497600048780441, "p95_ms": 0.09561759792268276, "p99_ms": 0.09619584225118159, "mean_ms": 0.09534933343529702, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_7c21104bed6fe513_sdpa_cudnn.json", "configuration": "480x832_tp1_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp1_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp1_sp1_v1/captures/qkv_rank0_7c21104bed6fe513.pt", "context": {"phase": "commit", "step": null, "start_frame": 9, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "native_dense", "status": "measured", "q_shape": [4680, 40, 128], "k_shape": [512, 40, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.0, "rmse": 0.0, "relative_l2": 0.0, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 8.546650151804169e-05, "rmse": 4.48689709008337e-06, "relative_l2": 0.0017727604700619917, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.11740799993276596, "p95_ms": 0.1293823942542076, "p99_ms": 0.16653120100498192, "mean_ms": 0.11972479986647765, "n": 300}, {"repeat": 1, "p50_ms": 0.11823999881744385, "p95_ms": 0.1284928023815155, "p99_ms": 0.13987104490399335, "mean_ms": 0.12042431977887948, "n": 300}, {"repeat": 2, "p50_ms": 0.11919999867677689, "p95_ms": 0.12881120443344116, "p99_ms": 0.18761600673198628, "mean_ms": 0.12457727971176306, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_7c21104bed6fe513_native_dense.json", "configuration": "480x832_tp1_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp1_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp1_sp1_v1/captures/qkv_rank0_825f8b3d6e43b9fd.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 27, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "vllm_fa2", "status": "measured", "q_shape": [4680, 40, 128], "k_shape": [512, 40, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.00048828125, "rmse": 1.8947004539816732e-06, "relative_l2": 0.00024331551476974812, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.00018629015193599718, "rmse": 6.425491210484444e-06, "relative_l2": 0.0018716524974291156, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.15667200088500977, "p95_ms": 0.1576319932937622, "p99_ms": 0.1580163162946701, "mean_ms": 0.15687626679738362, "n": 300}, {"repeat": 1, "p50_ms": 0.15667200088500977, "p95_ms": 0.15753920003771782, "p99_ms": 0.15846687957644462, "mean_ms": 0.15707711984713874, "n": 300}, {"repeat": 2, "p50_ms": 0.1568319946527481, "p95_ms": 0.1577936016023159, "p99_ms": 0.15827263817191123, "mean_ms": 0.15722143962979318, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_825f8b3d6e43b9fd_vllm_fa2.json", "configuration": "480x832_tp1_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp1_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp1_sp1_v1/captures/qkv_rank0_825f8b3d6e43b9fd.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 27, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "vllm_fa3", "status": "measured", "q_shape": [4680, 40, 128], "k_shape": [512, 40, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.0001220703125, "rmse": 7.086400029174426e-07, "relative_l2": 9.100283199592305e-05, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.00018629015193599718, "rmse": 6.451693484237799e-06, "relative_l2": 0.001879284840156278, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.12883199751377106, "p95_ms": 0.12982719764113426, "p99_ms": 0.13062911629676818, "mean_ms": 0.1291148794690768, "n": 300}, {"repeat": 1, "p50_ms": 0.1287360042333603, "p95_ms": 0.12969599664211273, "p99_ms": 0.13024159342050554, "mean_ms": 0.12921663999557495, "n": 300}, {"repeat": 2, "p50_ms": 0.12881600111722946, "p95_ms": 0.1297920048236847, "p99_ms": 0.13021056056022645, "mean_ms": 0.12926762729883193, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_825f8b3d6e43b9fd_vllm_fa3.json", "configuration": "480x832_tp1_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp1_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp1_sp1_v1/captures/qkv_rank0_825f8b3d6e43b9fd.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 27, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_math", "status": "measured", "q_shape": [4680, 40, 128], "k_shape": [512, 40, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.0009765625, "rmse": 4.891993262289943e-06, "relative_l2": 0.000628224823804117, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.00018629015193599718, "rmse": 5.611731845242477e-06, "relative_l2": 0.0016346161840378117, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 2.144864082336426, "p95_ms": 2.1500176668167112, "p99_ms": 2.151784715652466, "mean_ms": 2.1453021788597106, "n": 300}, {"repeat": 1, "p50_ms": 2.1442400217056274, "p95_ms": 2.1483583092689513, "p99_ms": 2.1522582507133485, "mean_ms": 2.145215783119202, "n": 300}, {"repeat": 2, "p50_ms": 2.1442880630493164, "p95_ms": 2.148129630088806, "p99_ms": 2.151299421787262, "mean_ms": 2.1449693854649863, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_825f8b3d6e43b9fd_sdpa_math.json", "configuration": "480x832_tp1_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp1_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp1_sp1_v1/captures/qkv_rank0_825f8b3d6e43b9fd.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 27, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_flash", "status": "measured", "q_shape": [4680, 40, 128], "k_shape": [512, 40, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.00048828125, "rmse": 1.8972653771256816e-06, "relative_l2": 0.0002436448996040726, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.00018629015193599718, "rmse": 6.425491210484444e-06, "relative_l2": 0.0018716524974291156, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.16678400337696075, "p95_ms": 0.16764960214495658, "p99_ms": 0.16806431829929352, "mean_ms": 0.16690410643815995, "n": 300}, {"repeat": 1, "p50_ms": 0.16676799952983856, "p95_ms": 0.16784000396728516, "p99_ms": 0.1682899272441864, "mean_ms": 0.1671444266537825, "n": 300}, {"repeat": 2, "p50_ms": 0.16697600483894348, "p95_ms": 0.16800159811973572, "p99_ms": 0.16838400065898895, "mean_ms": 0.1672979203859965, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_825f8b3d6e43b9fd_sdpa_flash.json", "configuration": "480x832_tp1_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp1_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp1_sp1_v1/captures/qkv_rank0_825f8b3d6e43b9fd.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 27, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_efficient", "status": "measured", "q_shape": [4680, 40, 128], "k_shape": [512, 40, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.00048828125, "rmse": 3.3053897176802623e-06, "relative_l2": 0.00042447480232660824, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.00018629015193599718, "rmse": 6.716977998482818e-06, "relative_l2": 0.0019565583757275027, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.3081279993057251, "p95_ms": 0.30880161076784135, "p99_ms": 0.30940832883119584, "mean_ms": 0.3082713602979978, "n": 300}, {"repeat": 1, "p50_ms": 0.3086720108985901, "p95_ms": 0.3094112083315849, "p99_ms": 0.30992289274930956, "mean_ms": 0.3090608008702596, "n": 300}, {"repeat": 2, "p50_ms": 0.3086079955101013, "p95_ms": 0.3095376089215279, "p99_ms": 0.3101817747950554, "mean_ms": 0.30887519945700964, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_825f8b3d6e43b9fd_sdpa_efficient.json", "configuration": "480x832_tp1_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp1_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp1_sp1_v1/captures/qkv_rank0_825f8b3d6e43b9fd.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 27, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_cudnn", "status": "measured", "q_shape": [4680, 40, 128], "k_shape": [512, 40, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.00048828125, "rmse": 3.3046751039086524e-06, "relative_l2": 0.00042438303234928245, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.00018629015193599718, "rmse": 6.716978124601895e-06, "relative_l2": 0.0019565584124641624, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.09463999792933464, "p95_ms": 0.09551999717950821, "p99_ms": 0.09580831937491893, "mean_ms": 0.09497376014788945, "n": 300}, {"repeat": 1, "p50_ms": 0.09523200243711472, "p95_ms": 0.09584159962832928, "p99_ms": 0.09622464254498482, "mean_ms": 0.09560746684670449, "n": 300}, {"repeat": 2, "p50_ms": 0.09532800316810608, "p95_ms": 0.09597120061516762, "p99_ms": 0.09664479829370975, "mean_ms": 0.09582229360938072, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_825f8b3d6e43b9fd_sdpa_cudnn.json", "configuration": "480x832_tp1_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp1_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp1_sp1_v1/captures/qkv_rank0_825f8b3d6e43b9fd.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 27, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "native_dense", "status": "measured", "q_shape": [4680, 40, 128], "k_shape": [512, 40, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.0, "rmse": 0.0, "relative_l2": 0.0, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.00018629015193599718, "rmse": 6.461484411018777e-06, "relative_l2": 0.001882136795277135, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.11971200257539749, "p95_ms": 0.13828000202775, "p99_ms": 0.16992447555065132, "mean_ms": 0.12312213328977425, "n": 300}, {"repeat": 1, "p50_ms": 0.11881600320339203, "p95_ms": 0.14582239985466006, "p99_ms": 0.20413408190011945, "mean_ms": 0.12403552023073038, "n": 300}, {"repeat": 2, "p50_ms": 0.11807999759912491, "p95_ms": 0.1273456007242203, "p99_ms": 0.16425055831670754, "mean_ms": 0.12040330698092779, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_825f8b3d6e43b9fd_native_dense.json", "configuration": "480x832_tp1_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp1_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp1_sp1_v1/captures/qkv_rank0_8b7d4bb8fe8219a6.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 0, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "vllm_fa2", "status": "measured", "q_shape": [4680, 40, 128], "k_shape": [512, 40, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.0009765625, "rmse": 2.3935024065536843e-06, "relative_l2": 0.0002581170658084147, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.0006142374604239942, "rmse": 1.9996831763657214e-05, "relative_l2": 0.0020143029066056256, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.15670399367809296, "p95_ms": 0.15753600001335144, "p99_ms": 0.15792416289448738, "mean_ms": 0.15691829348603883, "n": 300}, {"repeat": 1, "p50_ms": 0.15673600137233734, "p95_ms": 0.15760000050067902, "p99_ms": 0.15817632481455804, "mean_ms": 0.15708789293964703, "n": 300}, {"repeat": 2, "p50_ms": 0.15676799416542053, "p95_ms": 0.15766560062766075, "p99_ms": 0.1579843233525753, "mean_ms": 0.1572183466454347, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_8b7d4bb8fe8219a6_vllm_fa2.json", "configuration": "480x832_tp1_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp1_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp1_sp1_v1/captures/qkv_rank0_8b7d4bb8fe8219a6.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 0, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "vllm_fa3", "status": "measured", "q_shape": [4680, 40, 128], "k_shape": [512, 40, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.00048828125, "rmse": 8.901622595868577e-07, "relative_l2": 9.599575496929614e-05, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.0006142374604239942, "rmse": 2.0004182011344022e-05, "relative_l2": 0.0020150433051574963, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.1289920061826706, "p95_ms": 0.1302096001803875, "p99_ms": 0.130880958288908, "mean_ms": 0.12924522618452708, "n": 300}, {"repeat": 1, "p50_ms": 0.12910400331020355, "p95_ms": 0.1303039938211441, "p99_ms": 0.1309123258292675, "mean_ms": 0.1295681064327558, "n": 300}, {"repeat": 2, "p50_ms": 0.12905600666999817, "p95_ms": 0.1300815999507904, "p99_ms": 0.1306918367743492, "mean_ms": 0.12951669365167617, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_8b7d4bb8fe8219a6_vllm_fa3.json", "configuration": "480x832_tp1_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp1_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp1_sp1_v1/captures/qkv_rank0_8b7d4bb8fe8219a6.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 0, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_math", "status": "measured", "q_shape": [4680, 40, 128], "k_shape": [512, 40, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.001953125, "rmse": 1.0804165804234078e-05, "relative_l2": 0.001165129213265306, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.0004727375432675984, "rmse": 1.7092660525431627e-05, "relative_l2": 0.001721762536432087, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 2.1430399417877197, "p95_ms": 2.146657633781433, "p99_ms": 2.1480335497856142, "mean_ms": 2.14346240123113, "n": 300}, {"repeat": 1, "p50_ms": 2.142688035964966, "p95_ms": 2.145953583717346, "p99_ms": 2.147424302101135, "mean_ms": 2.142916052341461, "n": 300}, {"repeat": 2, "p50_ms": 2.143199920654297, "p95_ms": 2.147265601158142, "p99_ms": 2.149165799617767, "mean_ms": 2.143641495704651, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_8b7d4bb8fe8219a6_sdpa_math.json", "configuration": "480x832_tp1_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp1_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp1_sp1_v1/captures/qkv_rank0_8b7d4bb8fe8219a6.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 0, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_flash", "status": "measured", "q_shape": [4680, 40, 128], "k_shape": [512, 40, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.0009765625, "rmse": 2.394257205139189e-06, "relative_l2": 0.0002581984638448793, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.0006142374604239942, "rmse": 1.9996831763657214e-05, "relative_l2": 0.0020143029066056256, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.1669120043516159, "p95_ms": 0.16787199676036835, "p99_ms": 0.16857632204890252, "mean_ms": 0.16707797328631083, "n": 300}, {"repeat": 1, "p50_ms": 0.1669439971446991, "p95_ms": 0.16803520545363426, "p99_ms": 0.16848031401634217, "mean_ms": 0.1672806399067243, "n": 300}, {"repeat": 2, "p50_ms": 0.16679999977350235, "p95_ms": 0.1678416036069393, "p99_ms": 0.16822720676660538, "mean_ms": 0.16723050648967425, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_8b7d4bb8fe8219a6_sdpa_flash.json", "configuration": "480x832_tp1_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp1_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp1_sp1_v1/captures/qkv_rank0_8b7d4bb8fe8219a6.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 0, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_efficient", "status": "measured", "q_shape": [4680, 40, 128], "k_shape": [512, 40, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.0009765625, "rmse": 4.183506491393435e-06, "relative_l2": 0.0004511524272514699, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.0006142374604239942, "rmse": 2.0105374323793734e-05, "relative_l2": 0.0020252365183375982, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.3086400032043457, "p95_ms": 0.3094415873289108, "p99_ms": 0.3098249500989914, "mean_ms": 0.3087645875414213, "n": 300}, {"repeat": 1, "p50_ms": 0.3086720108985901, "p95_ms": 0.3093776017427444, "p99_ms": 0.309857277572155, "mean_ms": 0.3089486938714981, "n": 300}, {"repeat": 2, "p50_ms": 0.3084479868412018, "p95_ms": 0.3092175930738449, "p99_ms": 0.3096009555459023, "mean_ms": 0.3088306130965551, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_8b7d4bb8fe8219a6_sdpa_efficient.json", "configuration": "480x832_tp1_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp1_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp1_sp1_v1/captures/qkv_rank0_8b7d4bb8fe8219a6.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 0, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_cudnn", "status": "measured", "q_shape": [4680, 40, 128], "k_shape": [512, 40, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.0009765625, "rmse": 4.1799841347098204e-06, "relative_l2": 0.000450772573707385, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.0006142374604239942, "rmse": 2.0105374115138647e-05, "relative_l2": 0.0020252364973195416, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.0944959968328476, "p95_ms": 0.09507200121879578, "p99_ms": 0.09552095718681812, "mean_ms": 0.09472373299300671, "n": 300}, {"repeat": 1, "p50_ms": 0.09452799707651138, "p95_ms": 0.09526720270514488, "p99_ms": 0.09574623890221119, "mean_ms": 0.09497823946177959, "n": 300}, {"repeat": 2, "p50_ms": 0.0944959968328476, "p95_ms": 0.09516800194978714, "p99_ms": 0.095551997423172, "mean_ms": 0.09490207952757676, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_8b7d4bb8fe8219a6_sdpa_cudnn.json", "configuration": "480x832_tp1_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp1_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp1_sp1_v1/captures/qkv_rank0_8b7d4bb8fe8219a6.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 0, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "native_dense", "status": "measured", "q_shape": [4680, 40, 128], "k_shape": [512, 40, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.0, "rmse": 0.0, "relative_l2": 0.0, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.0006142374604239942, "rmse": 2.001643072823305e-05, "relative_l2": 0.002016277131911822, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.11711999773979187, "p95_ms": 0.1247871968895197, "p99_ms": 0.14301312044262876, "mean_ms": 0.11853642669816812, "n": 300}, {"repeat": 1, "p50_ms": 0.11799999698996544, "p95_ms": 0.1307344041764736, "p99_ms": 0.18645920202136013, "mean_ms": 0.12265002672870953, "n": 300}, {"repeat": 2, "p50_ms": 0.11734399944543839, "p95_ms": 0.12656480222940444, "p99_ms": 0.17408351957797918, "mean_ms": 0.12132501343886058, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_8b7d4bb8fe8219a6_native_dense.json", "configuration": "480x832_tp1_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp1_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp1_sp1_v1/captures/qkv_rank0_a2b4332ea3af9d3b.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 0, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "vllm_fa2", "status": "measured", "q_shape": [4680, 40, 128], "k_shape": [29640, 40, 128], "q_len": 4680, "k_len": 4680, "reference_error": {"max_abs": 0.03125, "rmse": 0.00043242308729223185, "relative_l2": 0.0008511296867427952, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.006950112781240136, "rmse": 0.0007533224446526795, "relative_l2": 0.0019169673843472507, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 1.2559680342674255, "p95_ms": 1.2663072168827056, "p99_ms": 1.267393958568573, "mean_ms": 1.2503018701076507, "n": 300}, {"repeat": 1, "p50_ms": 1.256111979484558, "p95_ms": 1.2770256280899048, "p99_ms": 1.2777279615402222, "mean_ms": 1.2544754139582317, "n": 300}, {"repeat": 2, "p50_ms": 1.266864001750946, "p95_ms": 1.3572800159454346, "p99_ms": 1.3579532790184021, "mean_ms": 1.28492693344752, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_a2b4332ea3af9d3b_vllm_fa2.json", "configuration": "480x832_tp1_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp1_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp1_sp1_v1/captures/qkv_rank0_a2b4332ea3af9d3b.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 0, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "vllm_fa3", "status": "measured", "q_shape": [4680, 40, 128], "k_shape": [29640, 40, 128], "q_len": 4680, "k_len": 4680, "reference_error": {"max_abs": 0.0, "rmse": 0.0, "relative_l2": 0.0, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.006950112781240136, "rmse": 0.0007627068873890609, "relative_l2": 0.0019408478232929027, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.7404960095882416, "p95_ms": 0.7475519776344299, "p99_ms": 0.7480383741855622, "mean_ms": 0.7280946117639542, "n": 300}, {"repeat": 1, "p50_ms": 0.7364960014820099, "p95_ms": 0.7440975815057754, "p99_ms": 0.7456006252765656, "mean_ms": 0.7347877327601114, "n": 300}, {"repeat": 2, "p50_ms": 0.74030402302742, "p95_ms": 0.742659193277359, "p99_ms": 0.7433628630638123, "mean_ms": 0.7336599463224411, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_a2b4332ea3af9d3b_vllm_fa3.json", "configuration": "480x832_tp1_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp1_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp1_sp1_v1/captures/qkv_rank0_a2b4332ea3af9d3b.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 0, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_math", "status": "measured", "q_shape": [4680, 40, 128], "k_shape": [29640, 40, 128], "q_len": 4680, "k_len": 4680, "reference_error": {"max_abs": 0.03125, "rmse": 0.0006944301932668124, "relative_l2": 0.0013668330166203373, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.006950112781240136, "rmse": 0.0006602534478209683, "relative_l2": 0.0016801362203659833, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 17.97551918029785, "p95_ms": 17.98509473800659, "p99_ms": 18.023530292510987, "mean_ms": 17.976907030741373, "n": 300}, {"repeat": 1, "p50_ms": 17.975199699401855, "p95_ms": 17.98353490829468, "p99_ms": 17.985865993499758, "mean_ms": 17.97617727279663, "n": 300}, {"repeat": 2, "p50_ms": 17.97551918029785, "p95_ms": 17.985151290893555, "p99_ms": 18.262670860290523, "mean_ms": 17.984538529713948, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_a2b4332ea3af9d3b_sdpa_math.json", "configuration": "480x832_tp1_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp1_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp1_sp1_v1/captures/qkv_rank0_a2b4332ea3af9d3b.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 0, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_flash", "status": "measured", "q_shape": [4680, 40, 128], "k_shape": [29640, 40, 128], "q_len": 4680, "k_len": 4680, "reference_error": {"max_abs": 0.03125, "rmse": 0.0004326789214216551, "relative_l2": 0.0008516332399267792, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.006950112781240136, "rmse": 0.0007533268077478835, "relative_l2": 0.0019169784870447213, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 1.3482720255851746, "p95_ms": 1.350625568628311, "p99_ms": 1.3523910117149354, "mean_ms": 1.3420354123910267, "n": 300}, {"repeat": 1, "p50_ms": 1.3472319841384888, "p95_ms": 1.3501759767532349, "p99_ms": 1.3510738742351531, "mean_ms": 1.344140477180481, "n": 300}, {"repeat": 2, "p50_ms": 1.3692479729652405, "p95_ms": 1.4480687618255617, "p99_ms": 1.4549481749534607, "mean_ms": 1.3775728003184, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_a2b4332ea3af9d3b_sdpa_flash.json", "configuration": "480x832_tp1_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp1_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp1_sp1_v1/captures/qkv_rank0_a2b4332ea3af9d3b.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 0, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_efficient", "status": "measured", "q_shape": [4680, 40, 128], "k_shape": [29640, 40, 128], "q_len": 4680, "k_len": 4680, "reference_error": {"max_abs": 0.03125, "rmse": 0.0006332863205797866, "relative_l2": 0.0012464847587781746, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.006950112781240136, "rmse": 0.0007546674581946931, "relative_l2": 0.001920390018452802, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 2.6236159801483154, "p95_ms": 2.6450608134269715, "p99_ms": 2.6475523328781128, "mean_ms": 2.6276213343938193, "n": 300}, {"repeat": 1, "p50_ms": 2.6218559741973877, "p95_ms": 2.626879930496216, "p99_ms": 2.6296371150016786, "mean_ms": 2.6252295422554015, "n": 300}, {"repeat": 2, "p50_ms": 2.621616005897522, "p95_ms": 2.627721607685089, "p99_ms": 2.631016550064087, "mean_ms": 2.6227756730715432, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_a2b4332ea3af9d3b_sdpa_efficient.json", "configuration": "480x832_tp1_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp1_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp1_sp1_v1/captures/qkv_rank0_a2b4332ea3af9d3b.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 0, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_cudnn", "status": "measured", "q_shape": [4680, 40, 128], "k_shape": [29640, 40, 128], "q_len": 4680, "k_len": 4680, "reference_error": {"max_abs": 0.03125, "rmse": 0.0006327277900670061, "relative_l2": 0.0012453854143760154, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.006950112781240136, "rmse": 0.0007544565760142951, "relative_l2": 0.0019198533899949183, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.8421760201454163, "p95_ms": 0.8583248227834701, "p99_ms": 0.8603522998094558, "mean_ms": 0.8204788259665171, "n": 300}, {"repeat": 1, "p50_ms": 0.8429759740829468, "p95_ms": 0.8565472155809403, "p99_ms": 0.857375979423523, "mean_ms": 0.8426257054011027, "n": 300}, {"repeat": 2, "p50_ms": 0.8345280289649963, "p95_ms": 0.867668804526329, "p99_ms": 0.8705961579084396, "mean_ms": 0.8325176537036896, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_a2b4332ea3af9d3b_sdpa_cudnn.json", "configuration": "480x832_tp1_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp1_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp1_sp1_v1/captures/qkv_rank0_a2b4332ea3af9d3b.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 0, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "native_dense", "status": "measured", "q_shape": [4680, 40, 128], "k_shape": [29640, 40, 128], "q_len": 4680, "k_len": 4680, "reference_error": {"max_abs": 0.03125, "rmse": 0.0004027831411821665, "relative_l2": 0.0007927899754991063, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.006950112781240136, "rmse": 0.0007607101265716498, "relative_l2": 0.0019357666984857367, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.7181439995765686, "p95_ms": 0.7785327762365342, "p99_ms": 0.7812486135959625, "mean_ms": 0.7213957323630651, "n": 300}, {"repeat": 1, "p50_ms": 0.7113759815692902, "p95_ms": 0.7196303963661194, "p99_ms": 0.7248335981369018, "mean_ms": 0.7015460254748662, "n": 300}, {"repeat": 2, "p50_ms": 0.7063040137290955, "p95_ms": 0.717275220155716, "p99_ms": 0.7192662113904953, "mean_ms": 0.7015077338616054, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_a2b4332ea3af9d3b_native_dense.json", "configuration": "480x832_tp1_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp1_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp1_sp1_v1/captures/qkv_rank0_a43a6e1d65520b5b.pt", "context": {"phase": "commit", "step": null, "start_frame": 9, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "vllm_fa2", "status": "measured", "q_shape": [4680, 40, 128], "k_shape": [29640, 40, 128], "q_len": 4680, "k_len": 18720, "reference_error": {"max_abs": 0.0625, "rmse": 0.00039391477402921305, "relative_l2": 0.0005205380094523126, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.012678517544874701, "rmse": 0.0009683098051771197, "relative_l2": 0.0016927232205301553, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 4.888623952865601, "p95_ms": 5.221404790878296, "p99_ms": 5.426115036010741, "mean_ms": 4.924604924519857, "n": 300}, {"repeat": 1, "p50_ms": 4.931152105331421, "p95_ms": 5.104095935821533, "p99_ms": 5.104864120483398, "mean_ms": 4.950297182401021, "n": 300}, {"repeat": 2, "p50_ms": 4.97105598449707, "p95_ms": 5.016936182975769, "p99_ms": 5.103044695854187, "mean_ms": 4.946988600095113, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_a43a6e1d65520b5b_vllm_fa2.json", "configuration": "480x832_tp1_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp1_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp1_sp1_v1/captures/qkv_rank0_a43a6e1d65520b5b.pt", "context": {"phase": "commit", "step": null, "start_frame": 9, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "vllm_fa3", "status": "measured", "q_shape": [4680, 40, 128], "k_shape": [29640, 40, 128], "q_len": 4680, "k_len": 18720, "reference_error": {"max_abs": 0.0, "rmse": 0.0, "relative_l2": 0.0, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.012678517544874701, "rmse": 0.0009692782411752042, "relative_l2": 0.001694416164351224, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 2.7189279794692993, "p95_ms": 2.755199909210205, "p99_ms": 2.8618361949920654, "mean_ms": 2.7216342306137085, "n": 300}, {"repeat": 1, "p50_ms": 2.7476959228515625, "p95_ms": 2.949993669986725, "p99_ms": 3.1076518392562864, "mean_ms": 2.77427659034729, "n": 300}, {"repeat": 2, "p50_ms": 2.753119945526123, "p95_ms": 2.888460803031921, "p99_ms": 2.894259743690491, "mean_ms": 2.7717079456647236, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_a43a6e1d65520b5b_vllm_fa3.json", "configuration": "480x832_tp1_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp1_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp1_sp1_v1/captures/qkv_rank0_a43a6e1d65520b5b.pt", "context": {"phase": "commit", "step": null, "start_frame": 9, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_math", "status": "measured", "q_shape": [4680, 40, 128], "k_shape": [29640, 40, 128], "q_len": 4680, "k_len": 18720, "reference_error": {"max_abs": 0.0625, "rmse": 0.0008261064498701522, "relative_l2": 0.0010916569658269153, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.012678517544874701, "rmse": 0.00094767651387136, "relative_l2": 0.001656653719712869, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 70.10358428955078, "p95_ms": 70.15436363220215, "p99_ms": 70.17025573730469, "mean_ms": 70.11026229858399, "n": 300}, {"repeat": 1, "p50_ms": 70.10441589355469, "p95_ms": 70.25004043579102, "p99_ms": 70.45162658691406, "mean_ms": 70.12212130228679, "n": 300}, {"repeat": 2, "p50_ms": 70.1032829284668, "p95_ms": 70.13204727172851, "p99_ms": 70.14065307617187, "mean_ms": 70.10309237162272, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_a43a6e1d65520b5b_sdpa_math.json", "configuration": "480x832_tp1_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp1_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp1_sp1_v1/captures/qkv_rank0_a43a6e1d65520b5b.pt", "context": {"phase": "commit", "step": null, "start_frame": 9, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_flash", "status": "measured", "q_shape": [4680, 40, 128], "k_shape": [29640, 40, 128], "q_len": 4680, "k_len": 18720, "reference_error": {"max_abs": 0.0625, "rmse": 0.00039463054221809164, "relative_l2": 0.0005214838601104565, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.012678517544874701, "rmse": 0.0009683237781535962, "relative_l2": 0.0016927476469912064, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 5.246079921722412, "p95_ms": 5.627675127983093, "p99_ms": 5.861166944503784, "mean_ms": 5.298614185651143, "n": 300}, {"repeat": 1, "p50_ms": 5.334991931915283, "p95_ms": 5.477249383926392, "p99_ms": 5.5163454341888425, "mean_ms": 5.340477013587952, "n": 300}, {"repeat": 2, "p50_ms": 5.3344480991363525, "p95_ms": 5.382531046867371, "p99_ms": 5.4318173265457155, "mean_ms": 5.306980155309041, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_a43a6e1d65520b5b_sdpa_flash.json", "configuration": "480x832_tp1_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp1_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp1_sp1_v1/captures/qkv_rank0_a43a6e1d65520b5b.pt", "context": {"phase": "commit", "step": null, "start_frame": 9, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_efficient", "status": "measured", "q_shape": [4680, 40, 128], "k_shape": [29640, 40, 128], "q_len": 4680, "k_len": 18720, "reference_error": {"max_abs": 0.0625, "rmse": 0.0007411087561173678, "relative_l2": 0.000979336907704824, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.012678517544874701, "rmse": 0.0009547840566502768, "relative_l2": 0.0016690785682876319, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 10.335968017578125, "p95_ms": 10.355147123336792, "p99_ms": 10.361407175064087, "mean_ms": 10.336888958613077, "n": 300}, {"repeat": 1, "p50_ms": 10.336351871490479, "p95_ms": 10.355715131759643, "p99_ms": 10.362242603302002, "mean_ms": 10.33746454556783, "n": 300}, {"repeat": 2, "p50_ms": 10.337119579315186, "p95_ms": 10.354916620254517, "p99_ms": 10.359564218521118, "mean_ms": 10.337490011850992, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_a43a6e1d65520b5b_sdpa_efficient.json", "configuration": "480x832_tp1_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp1_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp1_sp1_v1/captures/qkv_rank0_a43a6e1d65520b5b.pt", "context": {"phase": "commit", "step": null, "start_frame": 9, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_cudnn", "status": "measured", "q_shape": [4680, 40, 128], "k_shape": [29640, 40, 128], "q_len": 4680, "k_len": 18720, "reference_error": {"max_abs": 0.0625, "rmse": 0.0007408573956537501, "relative_l2": 0.0009790047478468727, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.012678517544874701, "rmse": 0.0009547032318614938, "relative_l2": 0.0016689372767340032, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 3.248207926750183, "p95_ms": 3.585548770427704, "p99_ms": 3.7446413898468016, "mean_ms": 3.2751423009236653, "n": 300}, {"repeat": 1, "p50_ms": 3.2508000135421753, "p95_ms": 3.4439457058906555, "p99_ms": 3.4465004897117613, "mean_ms": 3.2776380785306296, "n": 300}, {"repeat": 2, "p50_ms": 3.2263680696487427, "p95_ms": 3.414711928367615, "p99_ms": 3.4445214915275573, "mean_ms": 3.2673211773236592, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_a43a6e1d65520b5b_sdpa_cudnn.json", "configuration": "480x832_tp1_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp1_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp1_sp1_v1/captures/qkv_rank0_a43a6e1d65520b5b.pt", "context": {"phase": "commit", "step": null, "start_frame": 9, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "native_dense", "status": "measured", "q_shape": [4680, 40, 128], "k_shape": [29640, 40, 128], "q_len": 4680, "k_len": 18720, "reference_error": {"max_abs": 0.0625, "rmse": 0.00035034596568066397, "relative_l2": 0.00046296408162019444, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.012678517544874701, "rmse": 0.0009680282200371425, "relative_l2": 0.0016922309754837381, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 2.6856000423431396, "p95_ms": 2.753312110900879, "p99_ms": 2.758281214237213, "mean_ms": 2.6927443162600198, "n": 300}, {"repeat": 1, "p50_ms": 2.7341920137405396, "p95_ms": 3.0691951394081114, "p99_ms": 3.1469158029556272, "mean_ms": 2.779280003706614, "n": 300}, {"repeat": 2, "p50_ms": 2.742959976196289, "p95_ms": 2.91696640253067, "p99_ms": 2.936512632369995, "mean_ms": 2.773186244169871, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_a43a6e1d65520b5b_native_dense.json", "configuration": "480x832_tp1_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp1_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp1_sp1_v1/captures/qkv_rank0_b9e7999e0b6e56f3.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 9, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "vllm_fa2", "status": "measured", "q_shape": [4680, 40, 128], "k_shape": [512, 40, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.00048828125, "rmse": 1.9123103215654033e-06, "relative_l2": 0.0002448618943983648, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 7.629575161298766e-05, "rmse": 4.942429805964212e-06, "relative_l2": 0.0018663346713588256, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.15700799971818924, "p95_ms": 0.15798560306429862, "p99_ms": 0.15840095922350883, "mean_ms": 0.15727935959895453, "n": 300}, {"repeat": 1, "p50_ms": 0.15689599514007568, "p95_ms": 0.15766560062766075, "p99_ms": 0.15836896628141403, "mean_ms": 0.15727392012874286, "n": 300}, {"repeat": 2, "p50_ms": 0.156031996011734, "p95_ms": 0.15769599378108978, "p99_ms": 0.1582115176320076, "mean_ms": 0.15653631950418154, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_b9e7999e0b6e56f3_vllm_fa2.json", "configuration": "480x832_tp1_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp1_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp1_sp1_v1/captures/qkv_rank0_b9e7999e0b6e56f3.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 9, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "vllm_fa3", "status": "measured", "q_shape": [4680, 40, 128], "k_shape": [512, 40, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.000244140625, "rmse": 7.242414660602147e-07, "relative_l2": 9.273554369365357e-05, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 7.629575161298766e-05, "rmse": 4.97898511900548e-06, "relative_l2": 0.0018801384987938588, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.12796799838542938, "p95_ms": 0.12905600666999817, "p99_ms": 0.12950527518987656, "mean_ms": 0.12821568017204601, "n": 300}, {"repeat": 1, "p50_ms": 0.12808000296354294, "p95_ms": 0.1291535995900631, "p99_ms": 0.13021216049790382, "mean_ms": 0.12860223973790805, "n": 300}, {"repeat": 2, "p50_ms": 0.12809599936008453, "p95_ms": 0.1293135941028595, "p99_ms": 0.13023999333381653, "mean_ms": 0.12861599986751873, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_b9e7999e0b6e56f3_vllm_fa3.json", "configuration": "480x832_tp1_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp1_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp1_sp1_v1/captures/qkv_rank0_b9e7999e0b6e56f3.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 9, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_math", "status": "measured", "q_shape": [4680, 40, 128], "k_shape": [512, 40, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.0009765625, "rmse": 5.20960873006185e-06, "relative_l2": 0.0006670646747714977, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 5.955835214807581e-05, "rmse": 4.4437602027578725e-06, "relative_l2": 0.0016780296459857492, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 2.1440799236297607, "p95_ms": 2.147689640522003, "p99_ms": 2.149154920578003, "mean_ms": 2.1442261338233948, "n": 300}, {"repeat": 1, "p50_ms": 2.144831895828247, "p95_ms": 2.1492207407951356, "p99_ms": 2.1521001935005186, "mean_ms": 2.1458244291941324, "n": 300}, {"repeat": 2, "p50_ms": 2.1441760063171387, "p95_ms": 2.147270405292511, "p99_ms": 2.149029769897461, "mean_ms": 2.1448286883036296, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_b9e7999e0b6e56f3_sdpa_math.json", "configuration": "480x832_tp1_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp1_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp1_sp1_v1/captures/qkv_rank0_b9e7999e0b6e56f3.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 9, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_flash", "status": "measured", "q_shape": [4680, 40, 128], "k_shape": [512, 40, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.00048828125, "rmse": 1.9123471560559023e-06, "relative_l2": 0.0002448666108729982, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 7.629575161298766e-05, "rmse": 4.942430280889429e-06, "relative_l2": 0.001866334850697619, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.16684800386428833, "p95_ms": 0.1679072044789791, "p99_ms": 0.16860831499099732, "mean_ms": 0.16706058646241823, "n": 300}, {"repeat": 1, "p50_ms": 0.16684800386428833, "p95_ms": 0.16771360263228416, "p99_ms": 0.1682915273308754, "mean_ms": 0.16713589320580163, "n": 300}, {"repeat": 2, "p50_ms": 0.16708800196647644, "p95_ms": 0.16812959909439087, "p99_ms": 0.1687366358935833, "mean_ms": 0.16747413312395415, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_b9e7999e0b6e56f3_sdpa_flash.json", "configuration": "480x832_tp1_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp1_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp1_sp1_v1/captures/qkv_rank0_b9e7999e0b6e56f3.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 9, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_efficient", "status": "measured", "q_shape": [4680, 40, 128], "k_shape": [512, 40, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.0009765625, "rmse": 3.3335274439817913e-06, "relative_l2": 0.0004268417294815104, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 7.629575161298766e-05, "rmse": 5.409689151586598e-06, "relative_l2": 0.002042778718414212, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.30825600028038025, "p95_ms": 0.3088655978441238, "p99_ms": 0.30931358605623244, "mean_ms": 0.30834847966829937, "n": 300}, {"repeat": 1, "p50_ms": 0.30857598781585693, "p95_ms": 0.30934399366378784, "p99_ms": 0.30982398986816406, "mean_ms": 0.30884170631567637, "n": 300}, {"repeat": 2, "p50_ms": 0.3086720108985901, "p95_ms": 0.3094399869441986, "p99_ms": 0.3100179260969162, "mean_ms": 0.30892181346813835, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_b9e7999e0b6e56f3_sdpa_efficient.json", "configuration": "480x832_tp1_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp1_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp1_sp1_v1/captures/qkv_rank0_b9e7999e0b6e56f3.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 9, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_cudnn", "status": "measured", "q_shape": [4680, 40, 128], "k_shape": [512, 40, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.0009765625, "rmse": 3.3330751407732457e-06, "relative_l2": 0.0004267838142889615, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 7.629575161298766e-05, "rmse": 5.409689151586598e-06, "relative_l2": 0.002042778718414212, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.09446399658918381, "p95_ms": 0.09507680125534534, "p99_ms": 0.09568095840513706, "mean_ms": 0.09472149310012658, "n": 300}, {"repeat": 1, "p50_ms": 0.09443199634552002, "p95_ms": 0.09491360001266003, "p99_ms": 0.09568031840026378, "mean_ms": 0.09482389318446319, "n": 300}, {"repeat": 2, "p50_ms": 0.09443199634552002, "p95_ms": 0.09507360123097897, "p99_ms": 0.09545759670436382, "mean_ms": 0.09476181328296661, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_b9e7999e0b6e56f3_sdpa_cudnn.json", "configuration": "480x832_tp1_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp1_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp1_sp1_v1/captures/qkv_rank0_b9e7999e0b6e56f3.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 9, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "native_dense", "status": "measured", "q_shape": [4680, 40, 128], "k_shape": [512, 40, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.0, "rmse": 0.0, "relative_l2": 0.0, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 7.629575161298766e-05, "rmse": 4.99106817785475e-06, "relative_l2": 0.0018847012407146944, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.1178240031003952, "p95_ms": 0.12999999895691872, "p99_ms": 0.17559647575020787, "mean_ms": 0.1201876267294089, "n": 300}, {"repeat": 1, "p50_ms": 0.12030399963259697, "p95_ms": 0.17623519375920296, "p99_ms": 0.18800031363964081, "mean_ms": 0.12924469312032064, "n": 300}, {"repeat": 2, "p50_ms": 0.11868800222873688, "p95_ms": 0.1762287937104702, "p99_ms": 0.19896191895008045, "mean_ms": 0.1287707734356324, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_b9e7999e0b6e56f3_native_dense.json", "configuration": "480x832_tp1_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp1_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp1_sp1_v1/captures/qkv_rank0_cccc73c58e3e0a6e.pt", "context": {"phase": "commit", "step": null, "start_frame": 0, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "vllm_fa2", "status": "measured", "q_shape": [4680, 40, 128], "k_shape": [29640, 40, 128], "q_len": 4680, "k_len": 4680, "reference_error": {"max_abs": 0.03125, "rmse": 0.00040443792937809326, "relative_l2": 0.0005510746438495259, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.013453558575000635, "rmse": 0.000855206902521633, "relative_l2": 0.0017184019899498433, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 1.2458560466766357, "p95_ms": 1.2584336280822754, "p99_ms": 1.2626578938961028, "mean_ms": 1.2454877845446268, "n": 300}, {"repeat": 1, "p50_ms": 1.2454559803009033, "p95_ms": 1.2683087885379791, "p99_ms": 1.2774409127235413, "mean_ms": 1.2463196794191995, "n": 300}, {"repeat": 2, "p50_ms": 1.2564480304718018, "p95_ms": 1.3241263747215273, "p99_ms": 1.3344979405403137, "mean_ms": 1.2710978122552237, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_cccc73c58e3e0a6e_vllm_fa2.json", "configuration": "480x832_tp1_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp1_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp1_sp1_v1/captures/qkv_rank0_cccc73c58e3e0a6e.pt", "context": {"phase": "commit", "step": null, "start_frame": 0, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "vllm_fa3", "status": "measured", "q_shape": [4680, 40, 128], "k_shape": [29640, 40, 128], "q_len": 4680, "k_len": 4680, "reference_error": {"max_abs": 0.0, "rmse": 0.0, "relative_l2": 0.0, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.013453558575000635, "rmse": 0.0008552870221296227, "relative_l2": 0.001718562977534716, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.7259999811649323, "p95_ms": 0.7325151979923248, "p99_ms": 0.7335049575567245, "mean_ms": 0.713398934006691, "n": 300}, {"repeat": 1, "p50_ms": 0.7277439832687378, "p95_ms": 0.7346623778343201, "p99_ms": 0.7364271825551987, "mean_ms": 0.723376426299413, "n": 300}, {"repeat": 2, "p50_ms": 0.7266240119934082, "p95_ms": 0.7350432097911834, "p99_ms": 0.7367737674713135, "mean_ms": 0.7255735502640406, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_cccc73c58e3e0a6e_vllm_fa3.json", "configuration": "480x832_tp1_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp1_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp1_sp1_v1/captures/qkv_rank0_cccc73c58e3e0a6e.pt", "context": {"phase": "commit", "step": null, "start_frame": 0, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_math", "status": "measured", "q_shape": [4680, 40, 128], "k_shape": [29640, 40, 128], "q_len": 4680, "k_len": 4680, "reference_error": {"max_abs": 0.0625, "rmse": 0.0007421757478452632, "relative_l2": 0.0010112657745689109, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.013453558575000635, "rmse": 0.000840414736449667, "relative_l2": 0.0016886794894195205, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 17.977632522583008, "p95_ms": 17.98909282684326, "p99_ms": 17.991267223358154, "mean_ms": 17.978369515736897, "n": 300}, {"repeat": 1, "p50_ms": 17.978623390197754, "p95_ms": 17.989485836029054, "p99_ms": 17.991606521606446, "mean_ms": 17.978613808949788, "n": 300}, {"repeat": 2, "p50_ms": 17.977279663085938, "p95_ms": 17.98951187133789, "p99_ms": 17.99216495513916, "mean_ms": 17.978438402811687, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_cccc73c58e3e0a6e_sdpa_math.json", "configuration": "480x832_tp1_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp1_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp1_sp1_v1/captures/qkv_rank0_cccc73c58e3e0a6e.pt", "context": {"phase": "commit", "step": null, "start_frame": 0, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_flash", "status": "measured", "q_shape": [4680, 40, 128], "k_shape": [29640, 40, 128], "q_len": 4680, "k_len": 4680, "reference_error": {"max_abs": 0.03125, "rmse": 0.00040470011182650316, "relative_l2": 0.0005514318855642267, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.013453558575000635, "rmse": 0.000855206902521633, "relative_l2": 0.0017184019899498433, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 1.3257759809494019, "p95_ms": 1.3468799591064453, "p99_ms": 1.3522393524646759, "mean_ms": 1.3296024513244629, "n": 300}, {"repeat": 1, "p50_ms": 1.3280479907989502, "p95_ms": 1.3375040292739868, "p99_ms": 1.338627200126648, "mean_ms": 1.3310248517990113, "n": 300}, {"repeat": 2, "p50_ms": 1.3593599796295166, "p95_ms": 1.4558096408843995, "p99_ms": 1.4572800397872925, "mean_ms": 1.3701280001799265, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_cccc73c58e3e0a6e_sdpa_flash.json", "configuration": "480x832_tp1_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp1_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp1_sp1_v1/captures/qkv_rank0_cccc73c58e3e0a6e.pt", "context": {"phase": "commit", "step": null, "start_frame": 0, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_efficient", "status": "measured", "q_shape": [4680, 40, 128], "k_shape": [29640, 40, 128], "q_len": 4680, "k_len": 4680, "reference_error": {"max_abs": 0.03125, "rmse": 0.0006469104533169621, "relative_l2": 0.0008814602236055512, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.013453558575000635, "rmse": 0.0008510924395452369, "relative_l2": 0.0017101346322550366, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 2.6196800470352173, "p95_ms": 2.6382848501205443, "p99_ms": 2.6412272381782533, "mean_ms": 2.6215670402844746, "n": 300}, {"repeat": 1, "p50_ms": 2.617807984352112, "p95_ms": 2.6247808218002318, "p99_ms": 2.6275558710098266, "mean_ms": 2.6189094400405883, "n": 300}, {"repeat": 2, "p50_ms": 2.6180959939956665, "p95_ms": 2.6243855357170105, "p99_ms": 2.6273304796218873, "mean_ms": 2.6191371726989745, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_cccc73c58e3e0a6e_sdpa_efficient.json", "configuration": "480x832_tp1_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp1_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp1_sp1_v1/captures/qkv_rank0_cccc73c58e3e0a6e.pt", "context": {"phase": "commit", "step": null, "start_frame": 0, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_cudnn", "status": "measured", "q_shape": [4680, 40, 128], "k_shape": [29640, 40, 128], "q_len": 4680, "k_len": 4680, "reference_error": {"max_abs": 0.03125, "rmse": 0.0006457688016529154, "relative_l2": 0.0008799046442731871, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.013453558575000635, "rmse": 0.000851012888406189, "relative_l2": 0.0017099747869179142, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.8324159979820251, "p95_ms": 0.8490896135568619, "p99_ms": 0.8496972608566284, "mean_ms": 0.8195316276947657, "n": 300}, {"repeat": 1, "p50_ms": 0.8124960064888, "p95_ms": 0.8404208123683929, "p99_ms": 0.8408966505527496, "mean_ms": 0.8116229341427486, "n": 300}, {"repeat": 2, "p50_ms": 0.8141280114650726, "p95_ms": 0.8488960266113281, "p99_ms": 0.8497599959373474, "mean_ms": 0.8146669880549113, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_cccc73c58e3e0a6e_sdpa_cudnn.json", "configuration": "480x832_tp1_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp1_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp1_sp1_v1/captures/qkv_rank0_cccc73c58e3e0a6e.pt", "context": {"phase": "commit", "step": null, "start_frame": 0, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "native_dense", "status": "measured", "q_shape": [4680, 40, 128], "k_shape": [29640, 40, 128], "q_len": 4680, "k_len": 4680, "reference_error": {"max_abs": 0.03125, "rmse": 0.00036680578464484427, "relative_l2": 0.0004997982445561704, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.013453558575000635, "rmse": 0.0008563707767124402, "relative_l2": 0.0017207406096682262, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.6884960234165192, "p95_ms": 0.6945599913597107, "p99_ms": 0.6956483346223831, "mean_ms": 0.674008641242981, "n": 300}, {"repeat": 1, "p50_ms": 0.6854720115661621, "p95_ms": 0.6994256258010865, "p99_ms": 0.7003209656476974, "mean_ms": 0.6859724791844686, "n": 300}, {"repeat": 2, "p50_ms": 0.6908319890499115, "p95_ms": 0.6995199918746948, "p99_ms": 0.7003856205940246, "mean_ms": 0.6865547726551692, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_cccc73c58e3e0a6e_native_dense.json", "configuration": "480x832_tp1_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp1_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp1_sp1_v1/captures/qkv_rank0_eb19931361a68319.pt", "context": {"phase": "commit", "step": null, "start_frame": 27, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "vllm_fa2", "status": "measured", "q_shape": [4680, 40, 128], "k_shape": [29640, 40, 128], "q_len": 4680, "k_len": 28080, "reference_error": {"max_abs": 0.0625, "rmse": 0.0003908808806894715, "relative_l2": 0.0005186956147375676, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.010101945123846434, "rmse": 0.0009676761183591323, "relative_l2": 0.0016959539687613312, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 7.3132641315460205, "p95_ms": 7.772835159301758, "p99_ms": 8.095514993667603, "mean_ms": 7.378034873008728, "n": 300}, {"repeat": 1, "p50_ms": 7.375375986099243, "p95_ms": 7.5175742387771605, "p99_ms": 7.636034665107727, "mean_ms": 7.388339618047079, "n": 300}, {"repeat": 2, "p50_ms": 7.376224040985107, "p95_ms": 7.504696106910705, "p99_ms": 7.634660115242005, "mean_ms": 7.38732182820638, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_eb19931361a68319_vllm_fa2.json", "configuration": "480x832_tp1_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp1_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp1_sp1_v1/captures/qkv_rank0_eb19931361a68319.pt", "context": {"phase": "commit", "step": null, "start_frame": 27, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "vllm_fa3", "status": "measured", "q_shape": [4680, 40, 128], "k_shape": [29640, 40, 128], "q_len": 4680, "k_len": 28080, "reference_error": {"max_abs": 0.0, "rmse": 0.0, "relative_l2": 0.0, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.010101945123846434, "rmse": 0.0009681409449194487, "relative_l2": 0.0016967686260984276, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 4.112223863601685, "p95_ms": 4.5030031442642215, "p99_ms": 4.645281167030334, "mean_ms": 4.156075309912364, "n": 300}, {"repeat": 1, "p50_ms": 4.105823993682861, "p95_ms": 4.316750621795654, "p99_ms": 4.322374544143677, "mean_ms": 4.126370355288188, "n": 300}, {"repeat": 2, "p50_ms": 4.144063949584961, "p95_ms": 4.268580579757691, "p99_ms": 4.281957983970642, "mean_ms": 4.136651938756307, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_eb19931361a68319_vllm_fa3.json", "configuration": "480x832_tp1_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp1_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp1_sp1_v1/captures/qkv_rank0_eb19931361a68319.pt", "context": {"phase": "commit", "step": null, "start_frame": 27, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_math", "status": "measured", "q_shape": [4680, 40, 128], "k_shape": [29640, 40, 128], "q_len": 4680, "k_len": 28080, "reference_error": {"max_abs": 0.0625, "rmse": 0.0008318596936749729, "relative_l2": 0.0011038707608953843, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.010101945123846434, "rmse": 0.0009455398630386555, "relative_l2": 0.0016571578577981548, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 105.64184188842773, "p95_ms": 105.79609413146973, "p99_ms": 105.92597671508788, "mean_ms": 105.49984329223633, "n": 300}, {"repeat": 1, "p50_ms": 105.45116806030273, "p95_ms": 105.72242889404296, "p99_ms": 105.94999450683594, "mean_ms": 105.4462124379476, "n": 300}, {"repeat": 2, "p50_ms": 105.47745513916016, "p95_ms": 105.71800117492675, "p99_ms": 105.8995319366455, "mean_ms": 105.45995165506999, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_eb19931361a68319_sdpa_math.json", "configuration": "480x832_tp1_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp1_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp1_sp1_v1/captures/qkv_rank0_eb19931361a68319.pt", "context": {"phase": "commit", "step": null, "start_frame": 27, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_flash", "status": "measured", "q_shape": [4680, 40, 128], "k_shape": [29640, 40, 128], "q_len": 4680, "k_len": 28080, "reference_error": {"max_abs": 0.0625, "rmse": 0.00039200314501339385, "relative_l2": 0.0005201848499807142, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.010101945123846434, "rmse": 0.0009676761183591323, "relative_l2": 0.0016959539687613312, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 7.906432151794434, "p95_ms": 8.37379493713379, "p99_ms": 8.564643564224243, "mean_ms": 7.942565242449443, "n": 300}, {"repeat": 1, "p50_ms": 7.977391958236694, "p95_ms": 8.118565130233765, "p99_ms": 8.126896429061889, "mean_ms": 7.96415766398112, "n": 300}, {"repeat": 2, "p50_ms": 7.977312088012695, "p95_ms": 8.047103881835938, "p99_ms": 8.04989598274231, "mean_ms": 7.955438825289408, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_eb19931361a68319_sdpa_flash.json", "configuration": "480x832_tp1_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp1_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp1_sp1_v1/captures/qkv_rank0_eb19931361a68319.pt", "context": {"phase": "commit", "step": null, "start_frame": 27, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_efficient", "status": "measured", "q_shape": [4680, 40, 128], "k_shape": [29640, 40, 128], "q_len": 4680, "k_len": 28080, "reference_error": {"max_abs": 0.0625, "rmse": 0.0007462171936346741, "relative_l2": 0.0009902238894298147, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.010101945123846434, "rmse": 0.0009517787171752579, "relative_l2": 0.001668092104528802, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 15.329360008239746, "p95_ms": 15.359657382965088, "p99_ms": 15.363375120162964, "mean_ms": 15.33052501042684, "n": 300}, {"repeat": 1, "p50_ms": 15.325216293334961, "p95_ms": 15.359079742431641, "p99_ms": 15.363593168258667, "mean_ms": 15.328961725234985, "n": 300}, {"repeat": 2, "p50_ms": 15.325808048248291, "p95_ms": 15.359361362457275, "p99_ms": 15.366720514297485, "mean_ms": 15.32967316309611, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_eb19931361a68319_sdpa_efficient.json", "configuration": "480x832_tp1_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp1_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp1_sp1_v1/captures/qkv_rank0_eb19931361a68319.pt", "context": {"phase": "commit", "step": null, "start_frame": 27, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_cudnn", "status": "measured", "q_shape": [4680, 40, 128], "k_shape": [29640, 40, 128], "q_len": 4680, "k_len": 28080, "reference_error": {"max_abs": 0.0625, "rmse": 0.0007464779039461085, "relative_l2": 0.0009905698498027533, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.010101945123846434, "rmse": 0.0009516249798776634, "relative_l2": 0.0016678226637778573, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 4.8471999168396, "p95_ms": 5.19002573490143, "p99_ms": 5.585825614929199, "mean_ms": 4.882370985349019, "n": 300}, {"repeat": 1, "p50_ms": 4.901455879211426, "p95_ms": 5.134256148338318, "p99_ms": 5.13819420337677, "mean_ms": 4.92119562625885, "n": 300}, {"repeat": 2, "p50_ms": 4.890912055969238, "p95_ms": 5.039907360076905, "p99_ms": 5.13177649974823, "mean_ms": 4.8941934903462725, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_eb19931361a68319_sdpa_cudnn.json", "configuration": "480x832_tp1_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp1_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp1_sp1_v1/captures/qkv_rank0_eb19931361a68319.pt", "context": {"phase": "commit", "step": null, "start_frame": 27, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "native_dense", "status": "measured", "q_shape": [4680, 40, 128], "k_shape": [29640, 40, 128], "q_len": 4680, "k_len": 28080, "reference_error": {"max_abs": 0.0625, "rmse": 0.00034822456520980567, "relative_l2": 0.0004620910457416688, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.010101945123846434, "rmse": 0.0009673749392795276, "relative_l2": 0.0016954261208113068, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 4.084735870361328, "p95_ms": 4.354294300079346, "p99_ms": 4.56269139289856, "mean_ms": 4.1256507674853005, "n": 300}, {"repeat": 1, "p50_ms": 4.131919860839844, "p95_ms": 4.268772840499878, "p99_ms": 4.32990258693695, "mean_ms": 4.125397028923035, "n": 300}, {"repeat": 2, "p50_ms": 4.143167972564697, "p95_ms": 4.260166478157044, "p99_ms": 4.368906030654907, "mean_ms": 4.1167209482192995, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_eb19931361a68319_native_dense.json", "configuration": "480x832_tp1_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp1_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp1_sp1_v1/captures/qkv_rank0_f38973a7f296b1fc.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 9, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "vllm_fa2", "status": "measured", "q_shape": [4680, 40, 128], "k_shape": [29640, 40, 128], "q_len": 4680, "k_len": 18720, "reference_error": {"max_abs": 0.03125, "rmse": 0.0004405214737242676, "relative_l2": 0.0008934696641482836, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.008430754864764634, "rmse": 0.0008224519263401907, "relative_l2": 0.0019414505560311024, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 4.871887922286987, "p95_ms": 5.205516982078552, "p99_ms": 5.3908162117004395, "mean_ms": 4.915285763740539, "n": 300}, {"repeat": 1, "p50_ms": 4.931087970733643, "p95_ms": 5.105142569541931, "p99_ms": 5.149409785270691, "mean_ms": 4.950626664161682, "n": 300}, {"repeat": 2, "p50_ms": 4.930431842803955, "p95_ms": 5.06499674320221, "p99_ms": 5.104130444526672, "mean_ms": 4.936451845169067, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_f38973a7f296b1fc_vllm_fa2.json", "configuration": "480x832_tp1_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp1_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp1_sp1_v1/captures/qkv_rank0_f38973a7f296b1fc.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 9, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "vllm_fa3", "status": "measured", "q_shape": [4680, 40, 128], "k_shape": [29640, 40, 128], "q_len": 4680, "k_len": 18720, "reference_error": {"max_abs": 0.0, "rmse": 0.0, "relative_l2": 0.0, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.008430754864764634, "rmse": 0.0008316797479339394, "relative_l2": 0.001963233421132848, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 2.7458399534225464, "p95_ms": 2.7830352902412416, "p99_ms": 2.949954128265381, "mean_ms": 2.741959678332011, "n": 300}, {"repeat": 1, "p50_ms": 2.749071955680847, "p95_ms": 2.827316868305206, "p99_ms": 2.8595888924598696, "mean_ms": 2.757614399592082, "n": 300}, {"repeat": 2, "p50_ms": 2.7736639976501465, "p95_ms": 2.8283567786216737, "p99_ms": 2.829890344142914, "mean_ms": 2.767970460255941, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_f38973a7f296b1fc_vllm_fa3.json", "configuration": "480x832_tp1_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp1_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp1_sp1_v1/captures/qkv_rank0_f38973a7f296b1fc.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 9, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_math", "status": "measured", "q_shape": [4680, 40, 128], "k_shape": [29640, 40, 128], "q_len": 4680, "k_len": 18720, "reference_error": {"max_abs": 0.03125, "rmse": 0.0007453099537764154, "relative_l2": 0.00151164443462249, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.007760694409859159, "rmse": 0.0007096702226792513, "relative_l2": 0.0016752221063550086, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 70.1009292602539, "p95_ms": 70.13139495849609, "p99_ms": 70.14645721435546, "mean_ms": 70.10386980692546, "n": 300}, {"repeat": 1, "p50_ms": 70.10084915161133, "p95_ms": 70.13056640625, "p99_ms": 70.14666732788086, "mean_ms": 70.10177625020346, "n": 300}, {"repeat": 2, "p50_ms": 70.10065460205078, "p95_ms": 70.23451766967773, "p99_ms": 70.46172172546386, "mean_ms": 70.1154857635498, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_f38973a7f296b1fc_sdpa_math.json", "configuration": "480x832_tp1_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp1_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp1_sp1_v1/captures/qkv_rank0_f38973a7f296b1fc.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 9, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_flash", "status": "measured", "q_shape": [4680, 40, 128], "k_shape": [29640, 40, 128], "q_len": 4680, "k_len": 18720, "reference_error": {"max_abs": 0.03125, "rmse": 0.0004408417076032355, "relative_l2": 0.000894119165417476, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.008430754864764634, "rmse": 0.0008226065287387949, "relative_l2": 0.0019418155049151903, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 5.2872960567474365, "p95_ms": 5.724512147903442, "p99_ms": 5.864774904251099, "mean_ms": 5.307738339106242, "n": 300}, {"repeat": 1, "p50_ms": 5.302047967910767, "p95_ms": 5.434761643409729, "p99_ms": 5.443297877311706, "mean_ms": 5.326239995956421, "n": 300}, {"repeat": 2, "p50_ms": 5.335616111755371, "p95_ms": 5.380900979042053, "p99_ms": 5.432387213706971, "mean_ms": 5.318365441958109, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_f38973a7f296b1fc_sdpa_flash.json", "configuration": "480x832_tp1_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp1_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp1_sp1_v1/captures/qkv_rank0_f38973a7f296b1fc.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 9, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_efficient", "status": "measured", "q_shape": [4680, 40, 128], "k_shape": [29640, 40, 128], "q_len": 4680, "k_len": 18720, "reference_error": {"max_abs": 0.03125, "rmse": 0.0006322732450179911, "relative_l2": 0.0012823823526699727, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.00928316173610133, "rmse": 0.000829930267228196, "relative_l2": 0.001959103659647523, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 10.251296043395996, "p95_ms": 10.27426037788391, "p99_ms": 10.426821107864379, "mean_ms": 10.256685460408528, "n": 300}, {"repeat": 1, "p50_ms": 10.248559951782227, "p95_ms": 10.270663642883301, "p99_ms": 10.273103494644165, "mean_ms": 10.250827741622924, "n": 300}, {"repeat": 2, "p50_ms": 10.24724817276001, "p95_ms": 10.271470165252685, "p99_ms": 10.273516731262207, "mean_ms": 10.250528237024943, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_f38973a7f296b1fc_sdpa_efficient.json", "configuration": "480x832_tp1_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp1_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp1_sp1_v1/captures/qkv_rank0_f38973a7f296b1fc.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 9, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_cudnn", "status": "measured", "q_shape": [4680, 40, 128], "k_shape": [29640, 40, 128], "q_len": 4680, "k_len": 18720, "reference_error": {"max_abs": 0.03125, "rmse": 0.0006311120955355268, "relative_l2": 0.001280027298716865, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.00928316173610133, "rmse": 0.000830095690624396, "relative_l2": 0.0019594941521909135, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 3.2804479598999023, "p95_ms": 3.549971282482147, "p99_ms": 3.6618300676345825, "mean_ms": 3.3083159470558168, "n": 300}, {"repeat": 1, "p50_ms": 3.2860480546951294, "p95_ms": 3.4107856512069703, "p99_ms": 3.44370436668396, "mean_ms": 3.292021543979645, "n": 300}, {"repeat": 2, "p50_ms": 3.311232089996338, "p95_ms": 3.3772512435913087, "p99_ms": 3.4132515811920165, "mean_ms": 3.2926362625757855, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_f38973a7f296b1fc_sdpa_cudnn.json", "configuration": "480x832_tp1_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp1_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp1_sp1_v1/captures/qkv_rank0_f38973a7f296b1fc.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 9, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "native_dense", "status": "measured", "q_shape": [4680, 40, 128], "k_shape": [29640, 40, 128], "q_len": 4680, "k_len": 18720, "reference_error": {"max_abs": 0.03125, "rmse": 0.00039870006828546516, "relative_l2": 0.0008086471088351132, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.008430754864764634, "rmse": 0.0008339159662758888, "relative_l2": 0.0019685121580466317, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 2.7363040447235107, "p95_ms": 2.9207696199417112, "p99_ms": 2.9502877616882324, "mean_ms": 2.770487360159556, "n": 300}, {"repeat": 1, "p50_ms": 2.741215944290161, "p95_ms": 2.8754303097724914, "p99_ms": 2.926623649597167, "mean_ms": 2.7626280585924783, "n": 300}, {"repeat": 2, "p50_ms": 2.777936100959778, "p95_ms": 2.9366527438163756, "p99_ms": 2.9498963308334347, "mean_ms": 2.787818666299184, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_f38973a7f296b1fc_native_dense.json", "configuration": "480x832_tp1_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp1_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_cross_480x832_tp1_sp4_v1/captures/qkv_rank0_13f5e4136c72875f.pt", "context": {"phase": "commit", "step": null, "start_frame": 27, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "vllm_fa2", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [29640, 10, 128], "q_len": 4680, "k_len": 28080, "reference_error": {"max_abs": 0.0625, "rmse": 0.0004278408197932095, "relative_l2": 0.0005383218211689015, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.010101945123846434, "rmse": 0.0009676761183591323, "relative_l2": 0.0016959539687613312, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 1.906816005706787, "p95_ms": 1.935745644569397, "p99_ms": 1.9442966556549073, "mean_ms": 1.911606613000234, "n": 300}, {"repeat": 1, "p50_ms": 1.9334880113601685, "p95_ms": 2.0470862984657288, "p99_ms": 2.0648980259895326, "mean_ms": 1.947720852692922, "n": 300}, {"repeat": 2, "p50_ms": 1.9130079746246338, "p95_ms": 2.01724317073822, "p99_ms": 2.021417999267578, "mean_ms": 1.9328985611597698, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_13f5e4136c72875f_vllm_fa2.json", "configuration": "480x832_tp1_sp4", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp1_sp4_v1", "audit_status": "pending"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_cross_480x832_tp1_sp4_v1/captures/qkv_rank0_13f5e4136c72875f.pt", "context": {"phase": "commit", "step": null, "start_frame": 27, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "vllm_fa3", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [29640, 10, 128], "q_len": 4680, "k_len": 28080, "reference_error": {"max_abs": 0.0, "rmse": 0.0, "relative_l2": 0.0, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.010101945123846434, "rmse": 0.0009681409449194487, "relative_l2": 0.0016967686260984276, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 1.0458719730377197, "p95_ms": 1.061700850725174, "p99_ms": 1.0631366217136382, "mean_ms": 1.0378650637467701, "n": 300}, {"repeat": 1, "p50_ms": 1.05759996175766, "p95_ms": 1.0651855945587159, "p99_ms": 1.0702429008483887, "mean_ms": 1.0484679490327835, "n": 300}, {"repeat": 2, "p50_ms": 1.0579839944839478, "p95_ms": 1.1456751763820647, "p99_ms": 1.1574726545810698, "mean_ms": 1.0708190912008286, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_13f5e4136c72875f_vllm_fa3.json", "configuration": "480x832_tp1_sp4", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp1_sp4_v1", "audit_status": "pending"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_cross_480x832_tp1_sp4_v1/captures/qkv_rank0_13f5e4136c72875f.pt", "context": {"phase": "commit", "step": null, "start_frame": 27, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_math", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [29640, 10, 128], "q_len": 4680, "k_len": 28080, "reference_error": {"max_abs": 0.0625, "rmse": 0.0007962888877749233, "relative_l2": 0.0010019139465251164, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.010101945123846434, "rmse": 0.0009455398630386555, "relative_l2": 0.0016571578577981548, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 28.538111686706543, "p95_ms": 29.331832504272462, "p99_ms": 30.47968481063842, "mean_ms": 28.596927789052327, "n": 300}, {"repeat": 1, "p50_ms": 28.537039756774902, "p95_ms": 28.89883918762207, "p99_ms": 29.177503776550292, "mean_ms": 28.556398537953694, "n": 300}, {"repeat": 2, "p50_ms": 28.582639694213867, "p95_ms": 28.879370975494385, "p99_ms": 28.941119079589843, "mean_ms": 28.582187716166178, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_13f5e4136c72875f_sdpa_math.json", "configuration": "480x832_tp1_sp4", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp1_sp4_v1", "audit_status": "pending"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_cross_480x832_tp1_sp4_v1/captures/qkv_rank0_13f5e4136c72875f.pt", "context": {"phase": "commit", "step": null, "start_frame": 27, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_flash", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [29640, 10, 128], "q_len": 4680, "k_len": 28080, "reference_error": {"max_abs": 0.0625, "rmse": 0.00042904017884032254, "relative_l2": 0.0005398308897678007, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.010101945123846434, "rmse": 0.0009676761183591323, "relative_l2": 0.0016959539687613312, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 2.082304000854492, "p95_ms": 2.090646493434906, "p99_ms": 2.0945321202278135, "mean_ms": 2.0827004718780517, "n": 300}, {"repeat": 1, "p50_ms": 2.0916640758514404, "p95_ms": 2.1697935342788695, "p99_ms": 2.17571754693985, "mean_ms": 2.1022252758344013, "n": 300}, {"repeat": 2, "p50_ms": 2.0975199937820435, "p95_ms": 2.119748854637146, "p99_ms": 2.1205415511131287, "mean_ms": 2.092682549158732, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_13f5e4136c72875f_sdpa_flash.json", "configuration": "480x832_tp1_sp4", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp1_sp4_v1", "audit_status": "pending"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_cross_480x832_tp1_sp4_v1/captures/qkv_rank0_13f5e4136c72875f.pt", "context": {"phase": "commit", "step": null, "start_frame": 27, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_efficient", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [29640, 10, 128], "q_len": 4680, "k_len": 28080, "reference_error": {"max_abs": 0.0625, "rmse": 0.0006798250225789598, "relative_l2": 0.0008553757082079703, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.010101945123846434, "rmse": 0.0009517787171752579, "relative_l2": 0.001668092104528802, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 4.071968078613281, "p95_ms": 4.108132719993591, "p99_ms": 4.132014741897583, "mean_ms": 4.077283302942912, "n": 300}, {"repeat": 1, "p50_ms": 4.068559885025024, "p95_ms": 4.109140706062317, "p99_ms": 4.118469486236572, "mean_ms": 4.0742146031061806, "n": 300}, {"repeat": 2, "p50_ms": 4.0671679973602295, "p95_ms": 4.106771183013916, "p99_ms": 4.113568148612976, "mean_ms": 4.075562246640524, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_13f5e4136c72875f_sdpa_efficient.json", "configuration": "480x832_tp1_sp4", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp1_sp4_v1", "audit_status": "pending"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_cross_480x832_tp1_sp4_v1/captures/qkv_rank0_13f5e4136c72875f.pt", "context": {"phase": "commit", "step": null, "start_frame": 27, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_cudnn", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [29640, 10, 128], "q_len": 4680, "k_len": 28080, "reference_error": {"max_abs": 0.0625, "rmse": 0.0006799944528355498, "relative_l2": 0.0008555888903076417, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.010101945123846434, "rmse": 0.0009516249798776634, "relative_l2": 0.0016678226637778573, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 1.2372479438781738, "p95_ms": 1.2517183661460876, "p99_ms": 1.2544335627555847, "mean_ms": 1.2250023476282756, "n": 300}, {"repeat": 1, "p50_ms": 1.2377599477767944, "p95_ms": 1.254083251953125, "p99_ms": 1.2561369144916534, "mean_ms": 1.2334963218371073, "n": 300}, {"repeat": 2, "p50_ms": 1.2672799825668335, "p95_ms": 1.4051760554313661, "p99_ms": 1.4184709858894349, "mean_ms": 1.2828482107321422, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_13f5e4136c72875f_sdpa_cudnn.json", "configuration": "480x832_tp1_sp4", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp1_sp4_v1", "audit_status": "pending"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_cross_480x832_tp1_sp4_v1/captures/qkv_rank0_2419ddf76b22ab88.pt", "context": {"phase": "commit", "step": null, "start_frame": 9, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "vllm_fa2", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [512, 10, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.000244140625, "rmse": 1.1651005849709376e-06, "relative_l2": 0.000735961118544523, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 8.546650151804169e-05, "rmse": 4.432017925885755e-06, "relative_l2": 0.001751077866033789, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.04926399886608124, "p95_ms": 0.056480001285672196, "p99_ms": 0.07518335871398447, "mean_ms": 0.050486613251268866, "n": 300}, {"repeat": 1, "p50_ms": 0.049215998500585556, "p95_ms": 0.056300800107419495, "p99_ms": 0.07019423790276048, "mean_ms": 0.050747413262724875, "n": 300}, {"repeat": 2, "p50_ms": 0.04912000149488449, "p95_ms": 0.05484960041940212, "p99_ms": 0.07086016297340392, "mean_ms": 0.050384426588813466, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_2419ddf76b22ab88_vllm_fa2.json", "configuration": "480x832_tp1_sp4", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp1_sp4_v1", "audit_status": "pending"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_cross_480x832_tp1_sp4_v1/captures/qkv_rank0_2419ddf76b22ab88.pt", "context": {"phase": "commit", "step": null, "start_frame": 9, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "vllm_fa3", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [512, 10, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.0001220703125, "rmse": 4.452483364756388e-07, "relative_l2": 0.00028125079325307616, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 8.546650151804169e-05, "rmse": 4.471097426135494e-06, "relative_l2": 0.0017665180670996055, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.06015999987721443, "p95_ms": 0.06939199939370155, "p99_ms": 0.08705855965614318, "mean_ms": 0.06181119989603758, "n": 300}, {"repeat": 1, "p50_ms": 0.05934400111436844, "p95_ms": 0.06701280325651168, "p99_ms": 0.09005343995988366, "mean_ms": 0.06080426663160324, "n": 300}, {"repeat": 2, "p50_ms": 0.05974400043487549, "p95_ms": 0.06924639753997328, "p99_ms": 0.09847391724586486, "mean_ms": 0.06175701333830754, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_2419ddf76b22ab88_vllm_fa3.json", "configuration": "480x832_tp1_sp4", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp1_sp4_v1", "audit_status": "pending"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_cross_480x832_tp1_sp4_v1/captures/qkv_rank0_2419ddf76b22ab88.pt", "context": {"phase": "commit", "step": null, "start_frame": 9, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_math", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [512, 10, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.00048828125, "rmse": 2.337693456847005e-06, "relative_l2": 0.001476654903025604, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 8.546650151804169e-05, "rmse": 4.251983876886612e-06, "relative_l2": 0.0016799469176471491, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.614544004201889, "p95_ms": 0.6171871840953826, "p99_ms": 0.6180835396051407, "mean_ms": 0.6148807497819265, "n": 300}, {"repeat": 1, "p50_ms": 0.6152639985084534, "p95_ms": 0.6175680160522461, "p99_ms": 0.6189766371250153, "mean_ms": 0.6161168013016383, "n": 300}, {"repeat": 2, "p50_ms": 0.6153280138969421, "p95_ms": 0.6180176049470901, "p99_ms": 0.6192652863264084, "mean_ms": 0.6158316797018051, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_2419ddf76b22ab88_sdpa_math.json", "configuration": "480x832_tp1_sp4", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp1_sp4_v1", "audit_status": "pending"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_cross_480x832_tp1_sp4_v1/captures/qkv_rank0_2419ddf76b22ab88.pt", "context": {"phase": "commit", "step": null, "start_frame": 9, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_flash", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [512, 10, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.000244140625, "rmse": 1.1646446035996768e-06, "relative_l2": 0.0007356730879964678, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 8.546650151804169e-05, "rmse": 4.432017925885755e-06, "relative_l2": 0.001751077866033789, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.04899200052022934, "p95_ms": 0.04924959875643253, "p99_ms": 0.04954112067818638, "mean_ms": 0.049098560387889546, "n": 300}, {"repeat": 1, "p50_ms": 0.048928000032901764, "p95_ms": 0.049247998744249344, "p99_ms": 0.05197984009981146, "mean_ms": 0.049331306827565036, "n": 300}, {"repeat": 2, "p50_ms": 0.048928000032901764, "p95_ms": 0.04924959875643253, "p99_ms": 0.04954399999231088, "mean_ms": 0.04928384017199278, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_2419ddf76b22ab88_sdpa_flash.json", "configuration": "480x832_tp1_sp4", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp1_sp4_v1", "audit_status": "pending"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_cross_480x832_tp1_sp4_v1/captures/qkv_rank0_2419ddf76b22ab88.pt", "context": {"phase": "commit", "step": null, "start_frame": 9, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_efficient", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [512, 10, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.000244140625, "rmse": 2.0393986634254526e-06, "relative_l2": 0.001288230510613159, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 8.546650151804169e-05, "rmse": 5.084117071383684e-06, "relative_l2": 0.002008720411536978, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.08620800077915192, "p95_ms": 0.0865280032157898, "p99_ms": 0.08681631796061992, "mean_ms": 0.08630773392816385, "n": 300}, {"repeat": 1, "p50_ms": 0.08620800077915192, "p95_ms": 0.08656000345945358, "p99_ms": 0.08684831820428371, "mean_ms": 0.08644650734961033, "n": 300}, {"repeat": 2, "p50_ms": 0.08640000224113464, "p95_ms": 0.08668959699571133, "p99_ms": 0.08691327869892121, "mean_ms": 0.08658794785539309, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_2419ddf76b22ab88_sdpa_efficient.json", "configuration": "480x832_tp1_sp4", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp1_sp4_v1", "audit_status": "pending"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_cross_480x832_tp1_sp4_v1/captures/qkv_rank0_2419ddf76b22ab88.pt", "context": {"phase": "commit", "step": null, "start_frame": 9, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_cudnn", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [512, 10, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.000244140625, "rmse": 2.0372705397818046e-06, "relative_l2": 0.0012868862350395446, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 8.546650151804169e-05, "rmse": 5.0836227950647876e-06, "relative_l2": 0.002008525124348118, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.04681599885225296, "p95_ms": 0.0518800001591444, "p99_ms": 0.07385984256863581, "mean_ms": 0.05120810675124327, "n": 300}, {"repeat": 1, "p50_ms": 0.046879999339580536, "p95_ms": 0.05306560136377812, "p99_ms": 0.061144638769328456, "mean_ms": 0.04800511995951335, "n": 300}, {"repeat": 2, "p50_ms": 0.04668800160288811, "p95_ms": 0.0514576006680727, "p99_ms": 0.0558428804948926, "mean_ms": 0.04746197325487932, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_2419ddf76b22ab88_sdpa_cudnn.json", "configuration": "480x832_tp1_sp4", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp1_sp4_v1", "audit_status": "pending"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_cross_480x832_tp1_sp4_v1/captures/qkv_rank0_5d233cbdc47812ff.pt", "context": {"phase": "commit", "step": null, "start_frame": 0, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "vllm_fa2", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [29640, 10, 128], "q_len": 4680, "k_len": 4680, "reference_error": {"max_abs": 0.03125, "rmse": 0.0004155959679669699, "relative_l2": 0.0005516502103045647, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.013453558575000635, "rmse": 0.000855206902521633, "relative_l2": 0.0017184019899498433, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.33087998628616333, "p95_ms": 0.333220811188221, "p99_ms": 0.334018235206604, "mean_ms": 0.3312133322159449, "n": 300}, {"repeat": 1, "p50_ms": 0.3333440124988556, "p95_ms": 0.33430400490760803, "p99_ms": 0.33504000306129456, "mean_ms": 0.33396928012371063, "n": 300}, {"repeat": 2, "p50_ms": 0.3333120048046112, "p95_ms": 0.33427359759807584, "p99_ms": 0.3353024050593376, "mean_ms": 0.3339667197068532, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_5d233cbdc47812ff_vllm_fa2.json", "configuration": "480x832_tp1_sp4", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp1_sp4_v1", "audit_status": "pending"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_cross_480x832_tp1_sp4_v1/captures/qkv_rank0_5d233cbdc47812ff.pt", "context": {"phase": "commit", "step": null, "start_frame": 0, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "vllm_fa3", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [29640, 10, 128], "q_len": 4680, "k_len": 4680, "reference_error": {"max_abs": 0.0, "rmse": 0.0, "relative_l2": 0.0, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.013453558575000635, "rmse": 0.0008552870221296227, "relative_l2": 0.001718562977534716, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.17871999740600586, "p95_ms": 0.1799071989953518, "p99_ms": 0.18041696280241012, "mean_ms": 0.1776359465221564, "n": 300}, {"repeat": 1, "p50_ms": 0.18140800297260284, "p95_ms": 0.18726719543337822, "p99_ms": 0.18822496265172958, "mean_ms": 0.18208650703231494, "n": 300}, {"repeat": 2, "p50_ms": 0.18646400421857834, "p95_ms": 0.19072000682353973, "p99_ms": 0.19129631638526917, "mean_ms": 0.18792479972044626, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_5d233cbdc47812ff_vllm_fa3.json", "configuration": "480x832_tp1_sp4", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp1_sp4_v1", "audit_status": "pending"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_cross_480x832_tp1_sp4_v1/captures/qkv_rank0_5d233cbdc47812ff.pt", "context": {"phase": "commit", "step": null, "start_frame": 0, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_math", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [29640, 10, 128], "q_len": 4680, "k_len": 4680, "reference_error": {"max_abs": 0.03125, "rmse": 0.0007357542109918866, "relative_l2": 0.0009766191120949496, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.013453558575000635, "rmse": 0.000840414736449667, "relative_l2": 0.0016886794894195205, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 4.834480047225952, "p95_ms": 5.126006531715394, "p99_ms": 5.321837649345397, "mean_ms": 4.873423999150594, "n": 300}, {"repeat": 1, "p50_ms": 4.881824016571045, "p95_ms": 4.948004794120789, "p99_ms": 4.968353123664856, "mean_ms": 4.881787948608398, "n": 300}, {"repeat": 2, "p50_ms": 4.8788001537323, "p95_ms": 4.940875291824341, "p99_ms": 4.963937187194825, "mean_ms": 4.885460381507873, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_5d233cbdc47812ff_sdpa_math.json", "configuration": "480x832_tp1_sp4", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp1_sp4_v1", "audit_status": "pending"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_cross_480x832_tp1_sp4_v1/captures/qkv_rank0_5d233cbdc47812ff.pt", "context": {"phase": "commit", "step": null, "start_frame": 0, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_flash", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [29640, 10, 128], "q_len": 4680, "k_len": 4680, "reference_error": {"max_abs": 0.03125, "rmse": 0.00041651617644221817, "relative_l2": 0.0005528716687354007, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.013453558575000635, "rmse": 0.000855206902521633, "relative_l2": 0.0017184019899498433, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.359360009431839, "p95_ms": 0.36137758791446684, "p99_ms": 0.3620166325569153, "mean_ms": 0.35966581334670383, "n": 300}, {"repeat": 1, "p50_ms": 0.358271986246109, "p95_ms": 0.3596799969673157, "p99_ms": 0.3602883142232895, "mean_ms": 0.3585048519571622, "n": 300}, {"repeat": 2, "p50_ms": 0.35628798604011536, "p95_ms": 0.3589487984776497, "p99_ms": 0.35990431129932404, "mean_ms": 0.3564076793193817, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_5d233cbdc47812ff_sdpa_flash.json", "configuration": "480x832_tp1_sp4", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp1_sp4_v1", "audit_status": "pending"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_cross_480x832_tp1_sp4_v1/captures/qkv_rank0_5d233cbdc47812ff.pt", "context": {"phase": "commit", "step": null, "start_frame": 0, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_efficient", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [29640, 10, 128], "q_len": 4680, "k_len": 4680, "reference_error": {"max_abs": 0.03125, "rmse": 0.0005937389373871991, "relative_l2": 0.0007881120966546252, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.013453558575000635, "rmse": 0.0008510924395452369, "relative_l2": 0.0017101346322550366, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.7001919746398926, "p95_ms": 0.703044804930687, "p99_ms": 0.7042297506332398, "mean_ms": 0.7003812253475189, "n": 300}, {"repeat": 1, "p50_ms": 0.6943359971046448, "p95_ms": 0.6969120144844055, "p99_ms": 0.6999056243896484, "mean_ms": 0.6950215472777684, "n": 300}, {"repeat": 2, "p50_ms": 0.6940319836139679, "p95_ms": 0.6998800158500671, "p99_ms": 0.7025987350940704, "mean_ms": 0.6951927465200424, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_5d233cbdc47812ff_sdpa_efficient.json", "configuration": "480x832_tp1_sp4", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp1_sp4_v1", "audit_status": "pending"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_cross_480x832_tp1_sp4_v1/captures/qkv_rank0_5d233cbdc47812ff.pt", "context": {"phase": "commit", "step": null, "start_frame": 0, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_cudnn", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [29640, 10, 128], "q_len": 4680, "k_len": 4680, "reference_error": {"max_abs": 0.03125, "rmse": 0.0005926419217863406, "relative_l2": 0.0007866559494983345, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.013453558575000635, "rmse": 0.000851012888406189, "relative_l2": 0.0017099747869179142, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.1974720060825348, "p95_ms": 0.20848000049591064, "p99_ms": 0.2088915233314037, "mean_ms": 0.200685440103213, "n": 300}, {"repeat": 1, "p50_ms": 0.2120320051908493, "p95_ms": 0.215488001704216, "p99_ms": 0.21606624603271485, "mean_ms": 0.21288437336683275, "n": 300}, {"repeat": 2, "p50_ms": 0.21401600539684296, "p95_ms": 0.21548960134387016, "p99_ms": 0.216320953220129, "mean_ms": 0.21417952006061872, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_5d233cbdc47812ff_sdpa_cudnn.json", "configuration": "480x832_tp1_sp4", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp1_sp4_v1", "audit_status": "pending"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_cross_480x832_tp1_sp4_v1/captures/qkv_rank0_6b4ac7dbc60c0c5a.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 0, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "vllm_fa2", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [29640, 10, 128], "q_len": 4680, "k_len": 4680, "reference_error": {"max_abs": 0.03125, "rmse": 0.0005266509856320316, "relative_l2": 0.0009067576358098154, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.006950112781240136, "rmse": 0.0007533224446526795, "relative_l2": 0.0019169673843472507, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.33299198746681213, "p95_ms": 0.3367359936237335, "p99_ms": 0.33744001388549805, "mean_ms": 0.33379679997762046, "n": 300}, {"repeat": 1, "p50_ms": 0.33423998951911926, "p95_ms": 0.33542400598526, "p99_ms": 0.33587615460157394, "mean_ms": 0.3347480524579684, "n": 300}, {"repeat": 2, "p50_ms": 0.33395200967788696, "p95_ms": 0.3354256063699722, "p99_ms": 0.3363523107767105, "mean_ms": 0.33437098771333695, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_6b4ac7dbc60c0c5a_vllm_fa2.json", "configuration": "480x832_tp1_sp4", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp1_sp4_v1", "audit_status": "pending"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_cross_480x832_tp1_sp4_v1/captures/qkv_rank0_6b4ac7dbc60c0c5a.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 0, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "vllm_fa3", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [29640, 10, 128], "q_len": 4680, "k_len": 4680, "reference_error": {"max_abs": 0.0, "rmse": 0.0, "relative_l2": 0.0, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.006950112781240136, "rmse": 0.0007627068873890609, "relative_l2": 0.0019408478232929027, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.17723199725151062, "p95_ms": 0.19148799777030945, "p99_ms": 0.19193664118647574, "mean_ms": 0.18153866643706956, "n": 300}, {"repeat": 1, "p50_ms": 0.19145600497722626, "p95_ms": 0.19385600090026855, "p99_ms": 0.19411264285445212, "mean_ms": 0.19204949309428532, "n": 300}, {"repeat": 2, "p50_ms": 0.1916159987449646, "p95_ms": 0.1936336062848568, "p99_ms": 0.1940169547498226, "mean_ms": 0.1905245865881443, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_6b4ac7dbc60c0c5a_vllm_fa3.json", "configuration": "480x832_tp1_sp4", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp1_sp4_v1", "audit_status": "pending"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_cross_480x832_tp1_sp4_v1/captures/qkv_rank0_6b4ac7dbc60c0c5a.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 0, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_math", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [29640, 10, 128], "q_len": 4680, "k_len": 4680, "reference_error": {"max_abs": 0.03125, "rmse": 0.0008676879616324845, "relative_l2": 0.0014939356541150204, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.006950112781240136, "rmse": 0.0006602534478209683, "relative_l2": 0.0016801362203659833, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 4.832671880722046, "p95_ms": 5.215555214881897, "p99_ms": 5.3324750852584835, "mean_ms": 4.878459742863973, "n": 300}, {"repeat": 1, "p50_ms": 4.8815999031066895, "p95_ms": 4.96436493396759, "p99_ms": 4.971744699478149, "mean_ms": 4.887894085248312, "n": 300}, {"repeat": 2, "p50_ms": 4.8824639320373535, "p95_ms": 4.963060688972473, "p99_ms": 4.995023031234741, "mean_ms": 4.889163839022318, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_6b4ac7dbc60c0c5a_sdpa_math.json", "configuration": "480x832_tp1_sp4", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp1_sp4_v1", "audit_status": "pending"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_cross_480x832_tp1_sp4_v1/captures/qkv_rank0_6b4ac7dbc60c0c5a.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 0, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_flash", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [29640, 10, 128], "q_len": 4680, "k_len": 4680, "reference_error": {"max_abs": 0.03125, "rmse": 0.0005271479915826146, "relative_l2": 0.0009076133523147284, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.006950112781240136, "rmse": 0.0007533268077478835, "relative_l2": 0.0019169784870447213, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.36006399989128113, "p95_ms": 0.36182400584220886, "p99_ms": 0.36275327056646345, "mean_ms": 0.3602216532826424, "n": 300}, {"repeat": 1, "p50_ms": 0.36055999994277954, "p95_ms": 0.36153759658336637, "p99_ms": 0.3623078420758247, "mean_ms": 0.36099029233058294, "n": 300}, {"repeat": 2, "p50_ms": 0.359824001789093, "p95_ms": 0.3608959913253784, "p99_ms": 0.3612831944227219, "mean_ms": 0.3602162131667137, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_6b4ac7dbc60c0c5a_sdpa_flash.json", "configuration": "480x832_tp1_sp4", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp1_sp4_v1", "audit_status": "pending"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_cross_480x832_tp1_sp4_v1/captures/qkv_rank0_6b4ac7dbc60c0c5a.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 0, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_efficient", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [29640, 10, 128], "q_len": 4680, "k_len": 4680, "reference_error": {"max_abs": 0.03125, "rmse": 0.0007438730221660497, "relative_l2": 0.0012807581516485914, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.006950112781240136, "rmse": 0.0007546674581946931, "relative_l2": 0.001920390018452802, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.7051839828491211, "p95_ms": 0.7085472136735916, "p99_ms": 0.7099231892824173, "mean_ms": 0.7049430394172669, "n": 300}, {"repeat": 1, "p50_ms": 0.693807989358902, "p95_ms": 0.6968960165977478, "p99_ms": 0.6992646300792694, "mean_ms": 0.6943878400325775, "n": 300}, {"repeat": 2, "p50_ms": 0.6956160068511963, "p95_ms": 0.7006783723831177, "p99_ms": 0.7028857845067978, "mean_ms": 0.6964623995621999, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_6b4ac7dbc60c0c5a_sdpa_efficient.json", "configuration": "480x832_tp1_sp4", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp1_sp4_v1", "audit_status": "pending"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_cross_480x832_tp1_sp4_v1/captures/qkv_rank0_6b4ac7dbc60c0c5a.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 0, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_cudnn", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [29640, 10, 128], "q_len": 4680, "k_len": 4680, "reference_error": {"max_abs": 0.03125, "rmse": 0.0007431919192459065, "relative_l2": 0.0012795854674792623, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.006950112781240136, "rmse": 0.0007544565760142951, "relative_l2": 0.0019198533899949183, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.20148800313472748, "p95_ms": 0.21443680077791213, "p99_ms": 0.22035263404250144, "mean_ms": 0.20469962696234384, "n": 300}, {"repeat": 1, "p50_ms": 0.21615999937057495, "p95_ms": 0.22009600698947906, "p99_ms": 0.22047999501228333, "mean_ms": 0.21636245340108873, "n": 300}, {"repeat": 2, "p50_ms": 0.2130720019340515, "p95_ms": 0.21478399634361267, "p99_ms": 0.21529632017016412, "mean_ms": 0.21266421323021253, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_6b4ac7dbc60c0c5a_sdpa_cudnn.json", "configuration": "480x832_tp1_sp4", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp1_sp4_v1", "audit_status": "pending"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_cross_480x832_tp1_sp4_v1/captures/qkv_rank0_6bef1f209a2b6a6e.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 0, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "vllm_fa2", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [512, 10, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.00048828125, "rmse": 1.8211860529862304e-06, "relative_l2": 0.0002644321561886288, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.0006142374604239942, "rmse": 1.9996831763657214e-05, "relative_l2": 0.0020143029066056256, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.04931199923157692, "p95_ms": 0.05629600025713444, "p99_ms": 0.07346719942986951, "mean_ms": 0.05182794669022163, "n": 300}, {"repeat": 1, "p50_ms": 0.04944000020623207, "p95_ms": 0.057155199348926544, "p99_ms": 0.0644979192316532, "mean_ms": 0.05104874659329653, "n": 300}, {"repeat": 2, "p50_ms": 0.04944000020623207, "p95_ms": 0.058131199143826966, "p99_ms": 0.065463038906455, "mean_ms": 0.051117653300364815, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_6bef1f209a2b6a6e_vllm_fa2.json", "configuration": "480x832_tp1_sp4", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp1_sp4_v1", "audit_status": "pending"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_cross_480x832_tp1_sp4_v1/captures/qkv_rank0_6bef1f209a2b6a6e.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 0, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "vllm_fa3", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [512, 10, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.00048828125, "rmse": 7.027818468172491e-07, "relative_l2": 0.0001020423579344844, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.0006142374604239942, "rmse": 2.0004182011344022e-05, "relative_l2": 0.0020150433051574963, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.0597120001912117, "p95_ms": 0.06691360287368298, "p99_ms": 0.07802783727645864, "mean_ms": 0.06104778673499823, "n": 300}, {"repeat": 1, "p50_ms": 0.05942400172352791, "p95_ms": 0.06742719933390617, "p99_ms": 0.07286720320582386, "mean_ms": 0.06067477328081926, "n": 300}, {"repeat": 2, "p50_ms": 0.05923200026154518, "p95_ms": 0.06701600328087808, "p99_ms": 0.07679840333759783, "mean_ms": 0.06042240003744761, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_6bef1f209a2b6a6e_vllm_fa3.json", "configuration": "480x832_tp1_sp4", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp1_sp4_v1", "audit_status": "pending"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_cross_480x832_tp1_sp4_v1/captures/qkv_rank0_6bef1f209a2b6a6e.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 0, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_math", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [512, 10, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.001953125, "rmse": 1.475276971591166e-05, "relative_l2": 0.002142069284648931, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.0004727375432675984, "rmse": 1.7092660525431627e-05, "relative_l2": 0.001721762536432087, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.6156799793243408, "p95_ms": 0.6180816203355789, "p99_ms": 0.6194246256351471, "mean_ms": 0.6161258675654729, "n": 300}, {"repeat": 1, "p50_ms": 0.6164000034332275, "p95_ms": 0.6188160181045532, "p99_ms": 0.6196831876039505, "mean_ms": 0.6169649064540863, "n": 300}, {"repeat": 2, "p50_ms": 0.6163360178470612, "p95_ms": 0.6185952216386795, "p99_ms": 0.6196178925037384, "mean_ms": 0.6167882647116979, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_6bef1f209a2b6a6e_sdpa_math.json", "configuration": "480x832_tp1_sp4", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp1_sp4_v1", "audit_status": "pending"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_cross_480x832_tp1_sp4_v1/captures/qkv_rank0_6bef1f209a2b6a6e.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 0, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_flash", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [512, 10, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.00048828125, "rmse": 1.8211270710366762e-06, "relative_l2": 0.000264423592141001, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.0006142374604239942, "rmse": 1.9996831763657214e-05, "relative_l2": 0.0020143029066056256, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.048895999789237976, "p95_ms": 0.049375999718904495, "p99_ms": 0.04958496093749996, "mean_ms": 0.04901461345454057, "n": 300}, {"repeat": 1, "p50_ms": 0.0488319993019104, "p95_ms": 0.04915200173854828, "p99_ms": 0.051759361140429755, "mean_ms": 0.0492541866004467, "n": 300}, {"repeat": 2, "p50_ms": 0.048767998814582825, "p95_ms": 0.0490880012512207, "p99_ms": 0.04926431875675915, "mean_ms": 0.04905493326485157, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_6bef1f209a2b6a6e_sdpa_flash.json", "configuration": "480x832_tp1_sp4", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp1_sp4_v1", "audit_status": "pending"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_cross_480x832_tp1_sp4_v1/captures/qkv_rank0_6bef1f209a2b6a6e.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 0, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_efficient", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [512, 10, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.0009765625, "rmse": 3.4240173283262414e-06, "relative_l2": 0.0004971596852896511, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.0006142374604239942, "rmse": 2.0105374323793734e-05, "relative_l2": 0.0020252365183375982, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.0862400010228157, "p95_ms": 0.08659360334277152, "p99_ms": 0.08697791919112205, "mean_ms": 0.08633280078570048, "n": 300}, {"repeat": 1, "p50_ms": 0.08640000224113464, "p95_ms": 0.08675199747085571, "p99_ms": 0.08700831942260265, "mean_ms": 0.08660533477862677, "n": 300}, {"repeat": 2, "p50_ms": 0.08640000224113464, "p95_ms": 0.0867839977145195, "p99_ms": 0.0870399996638298, "mean_ms": 0.08661600137750308, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_6bef1f209a2b6a6e_sdpa_efficient.json", "configuration": "480x832_tp1_sp4", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp1_sp4_v1", "audit_status": "pending"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_cross_480x832_tp1_sp4_v1/captures/qkv_rank0_6bef1f209a2b6a6e.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 0, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_cudnn", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [512, 10, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.0009765625, "rmse": 3.402406925503428e-06, "relative_l2": 0.0004940219029608378, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.0006142374604239942, "rmse": 2.0105374115138647e-05, "relative_l2": 0.0020252364973195416, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.046720001846551895, "p95_ms": 0.050913600251078606, "p99_ms": 0.05877215839922403, "mean_ms": 0.04992352006336053, "n": 300}, {"repeat": 1, "p50_ms": 0.046911999583244324, "p95_ms": 0.05133439972996712, "p99_ms": 0.05483647990971799, "mean_ms": 0.047624960094690326, "n": 300}, {"repeat": 2, "p50_ms": 0.046911999583244324, "p95_ms": 0.05117279849946499, "p99_ms": 0.05837984111160039, "mean_ms": 0.047774933303395904, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_6bef1f209a2b6a6e_sdpa_cudnn.json", "configuration": "480x832_tp1_sp4", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp1_sp4_v1", "audit_status": "pending"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_cross_480x832_tp1_sp4_v1/captures/qkv_rank0_6c601440672149ef.pt", "context": {"phase": "commit", "step": null, "start_frame": 0, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "vllm_fa2", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [512, 10, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.00048828125, "rmse": 1.9783499656596346e-06, "relative_l2": 0.00028311028721459693, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.00025586326656432123, "rmse": 1.8789914191711385e-05, "relative_l2": 0.0017170595682592, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.04966399818658829, "p95_ms": 0.05790080111473799, "p99_ms": 0.06532735764980309, "mean_ms": 0.0509196800366044, "n": 300}, {"repeat": 1, "p50_ms": 0.04960000142455101, "p95_ms": 0.0596815999597311, "p99_ms": 0.07603679783642292, "mean_ms": 0.05138410662611326, "n": 300}, {"repeat": 2, "p50_ms": 0.04971199855208397, "p95_ms": 0.06462080031633377, "p99_ms": 0.08394815839827059, "mean_ms": 0.05205333340913057, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_6c601440672149ef_vllm_fa2.json", "configuration": "480x832_tp1_sp4", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp1_sp4_v1", "audit_status": "pending"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_cross_480x832_tp1_sp4_v1/captures/qkv_rank0_6c601440672149ef.pt", "context": {"phase": "commit", "step": null, "start_frame": 0, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "vllm_fa3", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [512, 10, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.00048828125, "rmse": 8.137099459926773e-07, "relative_l2": 0.00011644535118565316, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.00025586326656432123, "rmse": 1.8794715046210773e-05, "relative_l2": 0.0017174982798504232, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.06059199944138527, "p95_ms": 0.07075840160250665, "p99_ms": 0.08016671918332573, "mean_ms": 0.06263039994984865, "n": 300}, {"repeat": 1, "p50_ms": 0.059776000678539276, "p95_ms": 0.0684047993272543, "p99_ms": 0.08871391743421554, "mean_ms": 0.0613637334232529, "n": 300}, {"repeat": 2, "p50_ms": 0.05976000055670738, "p95_ms": 0.0649456024169922, "p99_ms": 0.0720195166766643, "mean_ms": 0.06077610671520233, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_6c601440672149ef_vllm_fa3.json", "configuration": "480x832_tp1_sp4", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp1_sp4_v1", "audit_status": "pending"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_cross_480x832_tp1_sp4_v1/captures/qkv_rank0_6c601440672149ef.pt", "context": {"phase": "commit", "step": null, "start_frame": 0, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_math", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [512, 10, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.0009765625, "rmse": 5.799299816110414e-06, "relative_l2": 0.0008299044481925851, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.00024379653738876295, "rmse": 1.860982023786444e-05, "relative_l2": 0.0017006022261189715, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.6147519946098328, "p95_ms": 0.6173760294914246, "p99_ms": 0.6183366614580155, "mean_ms": 0.6150759476423263, "n": 300}, {"repeat": 1, "p50_ms": 0.6148000061511993, "p95_ms": 0.6173824280500412, "p99_ms": 0.6185600161552429, "mean_ms": 0.6153272539377213, "n": 300}, {"repeat": 2, "p50_ms": 0.6146559715270996, "p95_ms": 0.6174095779657364, "p99_ms": 0.6180838596820831, "mean_ms": 0.6177608533700307, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_6c601440672149ef_sdpa_math.json", "configuration": "480x832_tp1_sp4", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp1_sp4_v1", "audit_status": "pending"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_cross_480x832_tp1_sp4_v1/captures/qkv_rank0_6c601440672149ef.pt", "context": {"phase": "commit", "step": null, "start_frame": 0, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_flash", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [512, 10, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.00048828125, "rmse": 1.9676565329363196e-06, "relative_l2": 0.00028158001154944263, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.00025586326656432123, "rmse": 1.8789914191711385e-05, "relative_l2": 0.0017170595682592, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.04867200180888176, "p95_ms": 0.04899200052022934, "p99_ms": 0.04934847898781294, "mean_ms": 0.048848213367164134, "n": 300}, {"repeat": 1, "p50_ms": 0.04864000156521797, "p95_ms": 0.04896000027656555, "p99_ms": 0.04921184148639432, "mean_ms": 0.048972800063590206, "n": 300}, {"repeat": 2, "p50_ms": 0.04864000156521797, "p95_ms": 0.04899200052022934, "p99_ms": 0.05659903842955823, "mean_ms": 0.04896288006256024, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_6c601440672149ef_sdpa_flash.json", "configuration": "480x832_tp1_sp4", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp1_sp4_v1", "audit_status": "pending"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_cross_480x832_tp1_sp4_v1/captures/qkv_rank0_6c601440672149ef.pt", "context": {"phase": "commit", "step": null, "start_frame": 0, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_efficient", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [512, 10, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.0009765625, "rmse": 3.547975701944155e-06, "relative_l2": 0.0005077304003050362, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.00025586326656432123, "rmse": 1.8933874262258647e-05, "relative_l2": 0.0017302149245880482, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.0862400010228157, "p95_ms": 0.08656000345945358, "p99_ms": 0.08684863820672035, "mean_ms": 0.08637194760143757, "n": 300}, {"repeat": 1, "p50_ms": 0.08627200126647949, "p95_ms": 0.0867839977145195, "p99_ms": 0.08736096210777759, "mean_ms": 0.08653365390996139, "n": 300}, {"repeat": 2, "p50_ms": 0.08627200126647949, "p95_ms": 0.08656160347163677, "p99_ms": 0.0876169566065073, "mean_ms": 0.0866028809050719, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_6c601440672149ef_sdpa_efficient.json", "configuration": "480x832_tp1_sp4", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp1_sp4_v1", "audit_status": "pending"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_cross_480x832_tp1_sp4_v1/captures/qkv_rank0_6c601440672149ef.pt", "context": {"phase": "commit", "step": null, "start_frame": 0, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_cudnn", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [512, 10, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.0009765625, "rmse": 3.53635267141311e-06, "relative_l2": 0.0005060670952432085, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.00025586326656432123, "rmse": 1.893387426216046e-05, "relative_l2": 0.0017302149245790755, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.046112000942230225, "p95_ms": 0.049961600452661514, "p99_ms": 0.05337824035435915, "mean_ms": 0.046565653247137864, "n": 300}, {"repeat": 1, "p50_ms": 0.04636799916625023, "p95_ms": 0.05003200080245734, "p99_ms": 0.05453888166695832, "mean_ms": 0.04704202655702829, "n": 300}, {"repeat": 2, "p50_ms": 0.04633599892258644, "p95_ms": 0.049350399523973465, "p99_ms": 0.05242432050406932, "mean_ms": 0.046830186570684114, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_6c601440672149ef_sdpa_cudnn.json", "configuration": "480x832_tp1_sp4", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp1_sp4_v1", "audit_status": "pending"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_cross_480x832_tp1_sp4_v1/captures/qkv_rank0_7baca46400faa6d9.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 27, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "vllm_fa2", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [512, 10, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.0001220703125, "rmse": 1.2177752203517502e-06, "relative_l2": 0.0007455754449189619, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.00018629015193599718, "rmse": 6.425491210484444e-06, "relative_l2": 0.0018716524974291156, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.04947200044989586, "p95_ms": 0.07763520255684853, "p99_ms": 0.08330976091325282, "mean_ms": 0.05322730672856172, "n": 300}, {"repeat": 1, "p50_ms": 0.04918399825692177, "p95_ms": 0.05686720088124275, "p99_ms": 0.06322719670832151, "mean_ms": 0.05044330670187871, "n": 300}, {"repeat": 2, "p50_ms": 0.049215998500585556, "p95_ms": 0.05673599988222122, "p99_ms": 0.07390976354479785, "mean_ms": 0.05041333326448997, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_7baca46400faa6d9_vllm_fa2.json", "configuration": "480x832_tp1_sp4", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp1_sp4_v1", "audit_status": "pending"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_cross_480x832_tp1_sp4_v1/captures/qkv_rank0_7baca46400faa6d9.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 27, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "vllm_fa3", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [512, 10, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 6.103515625e-05, "rmse": 4.5961343682608566e-07, "relative_l2": 0.0002813955210497394, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.00018629015193599718, "rmse": 6.451693484237799e-06, "relative_l2": 0.001879284840156278, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.05987200140953064, "p95_ms": 0.06708319671452045, "p99_ms": 0.07737600006163114, "mean_ms": 0.06112138661245505, "n": 300}, {"repeat": 1, "p50_ms": 0.05966399982571602, "p95_ms": 0.07758240029215818, "p99_ms": 0.18426111415028562, "mean_ms": 0.06518016008039315, "n": 300}, {"repeat": 2, "p50_ms": 0.05950399860739708, "p95_ms": 0.06631519831717014, "p99_ms": 0.07089696273207657, "mean_ms": 0.06083104003220797, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_7baca46400faa6d9_vllm_fa3.json", "configuration": "480x832_tp1_sp4", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp1_sp4_v1", "audit_status": "pending"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_cross_480x832_tp1_sp4_v1/captures/qkv_rank0_7baca46400faa6d9.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 27, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_math", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [512, 10, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.000244140625, "rmse": 3.1722574616132708e-06, "relative_l2": 0.0019421952662632134, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.00018629015193599718, "rmse": 5.611731845242477e-06, "relative_l2": 0.0016346161840378117, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.6158400177955627, "p95_ms": 0.6180512130260467, "p99_ms": 0.6188182580471039, "mean_ms": 0.6161055999994278, "n": 300}, {"repeat": 1, "p50_ms": 0.6161280274391174, "p95_ms": 0.6182768076658249, "p99_ms": 0.6189481496810914, "mean_ms": 0.6166986654202143, "n": 300}, {"repeat": 2, "p50_ms": 0.6162559986114502, "p95_ms": 0.6187520027160645, "p99_ms": 0.6193280220031738, "mean_ms": 0.6167815470695496, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_7baca46400faa6d9_sdpa_math.json", "configuration": "480x832_tp1_sp4", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp1_sp4_v1", "audit_status": "pending"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_cross_480x832_tp1_sp4_v1/captures/qkv_rank0_7baca46400faa6d9.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 27, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_flash", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [512, 10, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.0001220703125, "rmse": 1.217766212863948e-06, "relative_l2": 0.0007455699301395399, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.00018629015193599718, "rmse": 6.425491210484444e-06, "relative_l2": 0.0018716524974291156, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.04873599857091904, "p95_ms": 0.0490880012512207, "p99_ms": 0.05055680055171243, "mean_ms": 0.04887285330643257, "n": 300}, {"repeat": 1, "p50_ms": 0.048767998814582825, "p95_ms": 0.0490880012512207, "p99_ms": 0.04938079852610811, "mean_ms": 0.049108799869815506, "n": 300}, {"repeat": 2, "p50_ms": 0.04879999905824661, "p95_ms": 0.0490880012512207, "p99_ms": 0.04941311899572599, "mean_ms": 0.04907200002421935, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_7baca46400faa6d9_sdpa_flash.json", "configuration": "480x832_tp1_sp4", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp1_sp4_v1", "audit_status": "pending"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_cross_480x832_tp1_sp4_v1/captures/qkv_rank0_7baca46400faa6d9.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 27, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_efficient", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [512, 10, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.0001220703125, "rmse": 2.16051630927381e-06, "relative_l2": 0.0013227629217781355, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.00018629015193599718, "rmse": 6.716977998482818e-06, "relative_l2": 0.0019565583757275027, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.08633600175380707, "p95_ms": 0.08668799698352814, "p99_ms": 0.086945278942585, "mean_ms": 0.08641344117621581, "n": 300}, {"repeat": 1, "p50_ms": 0.08630400151014328, "p95_ms": 0.08662559650838375, "p99_ms": 0.08720000088214874, "mean_ms": 0.08658026789625485, "n": 300}, {"repeat": 2, "p50_ms": 0.08633600175380707, "p95_ms": 0.08668959699571133, "p99_ms": 0.08691487871110438, "mean_ms": 0.08654965430498124, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_7baca46400faa6d9_sdpa_efficient.json", "configuration": "480x832_tp1_sp4", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp1_sp4_v1", "audit_status": "pending"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_cross_480x832_tp1_sp4_v1/captures/qkv_rank0_7baca46400faa6d9.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 27, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_cudnn", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [512, 10, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.0001220703125, "rmse": 2.1591474379475984e-06, "relative_l2": 0.0013219248386647496, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.00018629015193599718, "rmse": 6.716978124601895e-06, "relative_l2": 0.0019565584124641624, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.046512000262737274, "p95_ms": 0.05165119841694832, "p99_ms": 0.053864960297942154, "mean_ms": 0.04704864005247752, "n": 300}, {"repeat": 1, "p50_ms": 0.046720001846551895, "p95_ms": 0.05230079963803291, "p99_ms": 0.060952638275921324, "mean_ms": 0.04763520001123349, "n": 300}, {"repeat": 2, "p50_ms": 0.046879999339580536, "p95_ms": 0.051367999985814095, "p99_ms": 0.05940288145095109, "mean_ms": 0.04766474656760693, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_7baca46400faa6d9_sdpa_cudnn.json", "configuration": "480x832_tp1_sp4", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp1_sp4_v1", "audit_status": "pending"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_cross_480x832_tp1_sp4_v1/captures/qkv_rank0_8ec3a76708f5e241.pt", "context": {"phase": "commit", "step": null, "start_frame": 27, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "vllm_fa2", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [512, 10, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.000244140625, "rmse": 1.1765090791105027e-06, "relative_l2": 0.0008027447067766402, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 6.036895268355172e-05, "rmse": 3.8066916294210143e-06, "relative_l2": 0.0017721563376145706, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.04926399886608124, "p95_ms": 0.05485120024532081, "p99_ms": 0.06497888162732109, "mean_ms": 0.05029418667157491, "n": 300}, {"repeat": 1, "p50_ms": 0.04944000020623207, "p95_ms": 0.05419039912521839, "p99_ms": 0.061218880005180776, "mean_ms": 0.05034944000343482, "n": 300}, {"repeat": 2, "p50_ms": 0.049375999718904495, "p95_ms": 0.05413119867444039, "p99_ms": 0.060526078566908795, "mean_ms": 0.05019914671778679, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_8ec3a76708f5e241_vllm_fa2.json", "configuration": "480x832_tp1_sp4", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp1_sp4_v1", "audit_status": "pending"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_cross_480x832_tp1_sp4_v1/captures/qkv_rank0_8ec3a76708f5e241.pt", "context": {"phase": "commit", "step": null, "start_frame": 27, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "vllm_fa3", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [512, 10, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.0001220703125, "rmse": 4.4691778534592407e-07, "relative_l2": 0.00030493677687725085, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 6.036895268355172e-05, "rmse": 3.834593322207515e-06, "relative_l2": 0.0017851456119018848, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.05929600074887276, "p95_ms": 0.06601440273225309, "p99_ms": 0.07689087644219393, "mean_ms": 0.06061471997449795, "n": 300}, {"repeat": 1, "p50_ms": 0.05881600081920624, "p95_ms": 0.06621919758617878, "p99_ms": 0.0740441573411226, "mean_ms": 0.06012309332688649, "n": 300}, {"repeat": 2, "p50_ms": 0.058848001062870026, "p95_ms": 0.06518559716641903, "p99_ms": 0.07301631651818745, "mean_ms": 0.05999914662291606, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_8ec3a76708f5e241_vllm_fa3.json", "configuration": "480x832_tp1_sp4", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp1_sp4_v1", "audit_status": "pending"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_cross_480x832_tp1_sp4_v1/captures/qkv_rank0_8ec3a76708f5e241.pt", "context": {"phase": "commit", "step": null, "start_frame": 27, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_math", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [512, 10, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.000244140625, "rmse": 2.2249764824750405e-06, "relative_l2": 0.001518125211035103, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 6.036895268355172e-05, "rmse": 3.591073235205049e-06, "relative_l2": 0.0016717779668364214, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.6146240234375, "p95_ms": 0.6170576125383377, "p99_ms": 0.6183376216888428, "mean_ms": 0.615204799969991, "n": 300}, {"repeat": 1, "p50_ms": 0.6151360273361206, "p95_ms": 0.6174095779657364, "p99_ms": 0.6180163246393203, "mean_ms": 0.6157668262720108, "n": 300}, {"repeat": 2, "p50_ms": 0.6153920292854309, "p95_ms": 0.6177600026130676, "p99_ms": 0.6192684864997864, "mean_ms": 0.6159633064270019, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_8ec3a76708f5e241_sdpa_math.json", "configuration": "480x832_tp1_sp4", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp1_sp4_v1", "audit_status": "pending"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_cross_480x832_tp1_sp4_v1/captures/qkv_rank0_8ec3a76708f5e241.pt", "context": {"phase": "commit", "step": null, "start_frame": 27, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_flash", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [512, 10, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.000244140625, "rmse": 1.1767946218868126e-06, "relative_l2": 0.0008029395356617809, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 6.036895268355172e-05, "rmse": 3.8066900674542536e-06, "relative_l2": 0.0017721556104610306, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.048928000032901764, "p95_ms": 0.04927999898791313, "p99_ms": 0.05058368101716036, "mean_ms": 0.04908640002210935, "n": 300}, {"repeat": 1, "p50_ms": 0.048928000032901764, "p95_ms": 0.04931199923157692, "p99_ms": 0.05269152145832764, "mean_ms": 0.04931616002072891, "n": 300}, {"repeat": 2, "p50_ms": 0.048928000032901764, "p95_ms": 0.04924959875643253, "p99_ms": 0.05297536093741654, "mean_ms": 0.049543786719441416, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_8ec3a76708f5e241_sdpa_flash.json", "configuration": "480x832_tp1_sp4", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp1_sp4_v1", "audit_status": "pending"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_cross_480x832_tp1_sp4_v1/captures/qkv_rank0_8ec3a76708f5e241.pt", "context": {"phase": "commit", "step": null, "start_frame": 27, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_efficient", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [512, 10, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.0001220703125, "rmse": 2.0326950438507145e-06, "relative_l2": 0.0013869295323891096, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 6.036895268355172e-05, "rmse": 4.3905271507836556e-06, "relative_l2": 0.0020439534570110473, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.08617600053548813, "p95_ms": 0.0865280032157898, "p99_ms": 0.0867206372320652, "mean_ms": 0.08630517375965913, "n": 300}, {"repeat": 1, "p50_ms": 0.08617600053548813, "p95_ms": 0.08649600297212601, "p99_ms": 0.08659264363348484, "mean_ms": 0.08642282731831073, "n": 300}, {"repeat": 2, "p50_ms": 0.08640000224113464, "p95_ms": 0.0867535974830389, "p99_ms": 0.08720096088945865, "mean_ms": 0.08667104117572308, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_8ec3a76708f5e241_sdpa_efficient.json", "configuration": "480x832_tp1_sp4", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp1_sp4_v1", "audit_status": "pending"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_cross_480x832_tp1_sp4_v1/captures/qkv_rank0_8ec3a76708f5e241.pt", "context": {"phase": "commit", "step": null, "start_frame": 27, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_cudnn", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [512, 10, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.0001220703125, "rmse": 2.0331514618785623e-06, "relative_l2": 0.0013872409512828844, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 6.036895268355172e-05, "rmse": 4.3905005043652074e-06, "relative_l2": 0.0020439410521135865, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.046271998435258865, "p95_ms": 0.05127519927918911, "p99_ms": 0.06155199915170668, "mean_ms": 0.04721173339833816, "n": 300}, {"repeat": 1, "p50_ms": 0.04633599892258644, "p95_ms": 0.0515920015051961, "p99_ms": 0.05659711845219131, "mean_ms": 0.04755786664783954, "n": 300}, {"repeat": 2, "p50_ms": 0.04636799916625023, "p95_ms": 0.0507631991058588, "p99_ms": 0.05416831877082584, "mean_ms": 0.04710325330495834, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_8ec3a76708f5e241_sdpa_cudnn.json", "configuration": "480x832_tp1_sp4", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp1_sp4_v1", "audit_status": "pending"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_cross_480x832_tp1_sp4_v1/captures/qkv_rank0_a7ce4999e4f2ba20.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 9, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "vllm_fa2", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [512, 10, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.0001220703125, "rmse": 1.2212824683851846e-06, "relative_l2": 0.0007223260174700013, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 7.629575161298766e-05, "rmse": 4.942429805964212e-06, "relative_l2": 0.0018663346713588256, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.049136001616716385, "p95_ms": 0.05538240056484938, "p99_ms": 0.05990112088620654, "mean_ms": 0.049964266754686834, "n": 300}, {"repeat": 1, "p50_ms": 0.049375999718904495, "p95_ms": 0.06369600035250188, "p99_ms": 0.0864726395159956, "mean_ms": 0.05302666662881772, "n": 300}, {"repeat": 2, "p50_ms": 0.04902400076389313, "p95_ms": 0.05912159886211159, "p99_ms": 0.07792959727346896, "mean_ms": 0.05096917318801085, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_a7ce4999e4f2ba20_vllm_fa2.json", "configuration": "480x832_tp1_sp4", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp1_sp4_v1", "audit_status": "pending"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_cross_480x832_tp1_sp4_v1/captures/qkv_rank0_a7ce4999e4f2ba20.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 9, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "vllm_fa3", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [512, 10, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 6.103515625e-05, "rmse": 4.6673282897397813e-07, "relative_l2": 0.00027604855903732764, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 7.629575161298766e-05, "rmse": 4.97898511900548e-06, "relative_l2": 0.0018801384987938588, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.05984000116586685, "p95_ms": 0.07549920119345188, "p99_ms": 0.08772223733365535, "mean_ms": 0.0619064533089598, "n": 300}, {"repeat": 1, "p50_ms": 0.05959999933838844, "p95_ms": 0.07402079738676548, "p99_ms": 0.09351391650736328, "mean_ms": 0.061979626528918745, "n": 300}, {"repeat": 2, "p50_ms": 0.059696000069379807, "p95_ms": 0.07064960077404978, "p99_ms": 0.08502111874520774, "mean_ms": 0.061630613394081595, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_a7ce4999e4f2ba20_vllm_fa3.json", "configuration": "480x832_tp1_sp4", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp1_sp4_v1", "audit_status": "pending"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_cross_480x832_tp1_sp4_v1/captures/qkv_rank0_a7ce4999e4f2ba20.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 9, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_math", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [512, 10, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.000244140625, "rmse": 3.1736303075170077e-06, "relative_l2": 0.0018770397514851157, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 5.955835214807581e-05, "rmse": 4.4437602027578725e-06, "relative_l2": 0.0016780296459857492, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.6153280138969421, "p95_ms": 0.6176000237464905, "p99_ms": 0.6182083112001419, "mean_ms": 0.6157153069972991, "n": 300}, {"repeat": 1, "p50_ms": 0.6159679889678955, "p95_ms": 0.618241599202156, "p99_ms": 0.6192323189973831, "mean_ms": 0.6163217089573543, "n": 300}, {"repeat": 2, "p50_ms": 0.6159999966621399, "p95_ms": 0.6186239719390869, "p99_ms": 0.619745602607727, "mean_ms": 0.6166224004824956, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_a7ce4999e4f2ba20_sdpa_math.json", "configuration": "480x832_tp1_sp4", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp1_sp4_v1", "audit_status": "pending"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_cross_480x832_tp1_sp4_v1/captures/qkv_rank0_a7ce4999e4f2ba20.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 9, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_flash", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [512, 10, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.0001220703125, "rmse": 1.2210431593591066e-06, "relative_l2": 0.0007221844784401483, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 7.629575161298766e-05, "rmse": 4.942430280889429e-06, "relative_l2": 0.001866334850697619, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.04879999905824661, "p95_ms": 0.04912000149488449, "p99_ms": 0.049247998744249344, "mean_ms": 0.04892821333060662, "n": 300}, {"repeat": 1, "p50_ms": 0.0488319993019104, "p95_ms": 0.04912000149488449, "p99_ms": 0.05049631934612972, "mean_ms": 0.049202239972849685, "n": 300}, {"repeat": 2, "p50_ms": 0.04886399954557419, "p95_ms": 0.049247998744249344, "p99_ms": 0.04947296019643542, "mean_ms": 0.049204053307573, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_a7ce4999e4f2ba20_sdpa_flash.json", "configuration": "480x832_tp1_sp4", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp1_sp4_v1", "audit_status": "pending"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_cross_480x832_tp1_sp4_v1/captures/qkv_rank0_a7ce4999e4f2ba20.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 9, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_efficient", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [512, 10, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.000244140625, "rmse": 2.168452619391683e-06, "relative_l2": 0.0012825286411493597, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 7.629575161298766e-05, "rmse": 5.409689151586598e-06, "relative_l2": 0.002042778718414212, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.08636800199747086, "p95_ms": 0.08675199747085571, "p99_ms": 0.0868803184479475, "mean_ms": 0.08647594794631004, "n": 300}, {"repeat": 1, "p50_ms": 0.08636800199747086, "p95_ms": 0.08671999722719193, "p99_ms": 0.08707199990749359, "mean_ms": 0.08671082807083924, "n": 300}, {"repeat": 2, "p50_ms": 0.08636800199747086, "p95_ms": 0.08671999722719193, "p99_ms": 0.08750336304306983, "mean_ms": 0.0866055478900671, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_a7ce4999e4f2ba20_sdpa_efficient.json", "configuration": "480x832_tp1_sp4", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp1_sp4_v1", "audit_status": "pending"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_cross_480x832_tp1_sp4_v1/captures/qkv_rank0_a7ce4999e4f2ba20.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 9, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_cudnn", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [512, 10, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.000244140625, "rmse": 2.1674583666215477e-06, "relative_l2": 0.001281940591568365, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 7.629575161298766e-05, "rmse": 5.409689151586598e-06, "relative_l2": 0.002042778718414212, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.046271998435258865, "p95_ms": 0.05060000140219927, "p99_ms": 0.05328671965748071, "mean_ms": 0.04682858670751254, "n": 300}, {"repeat": 1, "p50_ms": 0.04636799916625023, "p95_ms": 0.09732000306248666, "p99_ms": 0.18641728207468963, "mean_ms": 0.053230613234142465, "n": 300}, {"repeat": 2, "p50_ms": 0.04639999940991402, "p95_ms": 0.05097920075058937, "p99_ms": 0.05425023950636386, "mean_ms": 0.047121173441410064, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_a7ce4999e4f2ba20_sdpa_cudnn.json", "configuration": "480x832_tp1_sp4", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp1_sp4_v1", "audit_status": "pending"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_cross_480x832_tp1_sp4_v1/captures/qkv_rank0_c6a5fcb1e5f12d0d.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 9, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "vllm_fa2", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [29640, 10, 128], "q_len": 4680, "k_len": 18720, "reference_error": {"max_abs": 0.03125, "rmse": 0.0005655613731987039, "relative_l2": 0.0010091701042483979, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.008430754864764634, "rmse": 0.0008224519263401907, "relative_l2": 0.0019414505560311024, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 1.2979040145874023, "p95_ms": 1.306913548707962, "p99_ms": 1.308575987815857, "mean_ms": 1.2942225058873496, "n": 300}, {"repeat": 1, "p50_ms": 1.2898880243301392, "p95_ms": 1.2982383966445923, "p99_ms": 1.3004537010192871, "mean_ms": 1.2917638409137726, "n": 300}, {"repeat": 2, "p50_ms": 1.3096799850463867, "p95_ms": 1.3955536365509034, "p99_ms": 1.3965510714054108, "mean_ms": 1.3304891741275788, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_c6a5fcb1e5f12d0d_vllm_fa2.json", "configuration": "480x832_tp1_sp4", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp1_sp4_v1", "audit_status": "pending"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_cross_480x832_tp1_sp4_v1/captures/qkv_rank0_c6a5fcb1e5f12d0d.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 9, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "vllm_fa3", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [29640, 10, 128], "q_len": 4680, "k_len": 18720, "reference_error": {"max_abs": 0.0, "rmse": 0.0, "relative_l2": 0.0, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.008430754864764634, "rmse": 0.0008316797479339394, "relative_l2": 0.001963233421132848, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.7107840180397034, "p95_ms": 0.7171295911073685, "p99_ms": 0.7183042985200881, "mean_ms": 0.6992292282978694, "n": 300}, {"repeat": 1, "p50_ms": 0.7047199904918671, "p95_ms": 0.7141135841608047, "p99_ms": 0.7148194944858551, "mean_ms": 0.7005172272523245, "n": 300}, {"repeat": 2, "p50_ms": 0.7047359943389893, "p95_ms": 0.7104047775268555, "p99_ms": 0.7157766658067704, "mean_ms": 0.7008794671297074, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_c6a5fcb1e5f12d0d_vllm_fa3.json", "configuration": "480x832_tp1_sp4", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp1_sp4_v1", "audit_status": "pending"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_cross_480x832_tp1_sp4_v1/captures/qkv_rank0_c6a5fcb1e5f12d0d.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 9, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_math", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [29640, 10, 128], "q_len": 4680, "k_len": 18720, "reference_error": {"max_abs": 0.03125, "rmse": 0.0009919727936492876, "relative_l2": 0.001770045365575048, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.007760694409859159, "rmse": 0.0007096702226792513, "relative_l2": 0.0016752221063550086, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 19.049903869628906, "p95_ms": 19.678307056427002, "p99_ms": 21.642721672058105, "mean_ms": 19.11841812769572, "n": 300}, {"repeat": 1, "p50_ms": 19.083792686462402, "p95_ms": 19.379282760620118, "p99_ms": 19.49982099533081, "mean_ms": 19.098702691396078, "n": 300}, {"repeat": 2, "p50_ms": 19.10190486907959, "p95_ms": 19.409460163116456, "p99_ms": 19.470301818847656, "mean_ms": 19.107757670084634, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_c6a5fcb1e5f12d0d_sdpa_math.json", "configuration": "480x832_tp1_sp4", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp1_sp4_v1", "audit_status": "pending"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_cross_480x832_tp1_sp4_v1/captures/qkv_rank0_c6a5fcb1e5f12d0d.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 9, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_flash", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [29640, 10, 128], "q_len": 4680, "k_len": 18720, "reference_error": {"max_abs": 0.03125, "rmse": 0.0005659922767473519, "relative_l2": 0.001009938995123407, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.008430754864764634, "rmse": 0.0008226065287387949, "relative_l2": 0.0019418155049151903, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 1.3851680159568787, "p95_ms": 1.4061440229415894, "p99_ms": 1.4079362976551055, "mean_ms": 1.3880483210086823, "n": 300}, {"repeat": 1, "p50_ms": 1.3865920305252075, "p95_ms": 1.449958455562592, "p99_ms": 1.5396207892894744, "mean_ms": 1.3998511985937754, "n": 300}, {"repeat": 2, "p50_ms": 1.4230079650878906, "p95_ms": 1.47542724609375, "p99_ms": 1.4779315745830537, "mean_ms": 1.4266396792729696, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_c6a5fcb1e5f12d0d_sdpa_flash.json", "configuration": "480x832_tp1_sp4", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp1_sp4_v1", "audit_status": "pending"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_cross_480x832_tp1_sp4_v1/captures/qkv_rank0_c6a5fcb1e5f12d0d.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 9, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_efficient", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [29640, 10, 128], "q_len": 4680, "k_len": 18720, "reference_error": {"max_abs": 0.03125, "rmse": 0.0007752434725243829, "relative_l2": 0.0013833203133383903, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.00928316173610133, "rmse": 0.000829930267228196, "relative_l2": 0.001959103659647523, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 2.729599952697754, "p95_ms": 2.768363285064697, "p99_ms": 2.7753583216667175, "mean_ms": 2.7336693374315897, "n": 300}, {"repeat": 1, "p50_ms": 2.7240320444107056, "p95_ms": 2.749124801158905, "p99_ms": 2.7537930059432982, "mean_ms": 2.7275400535265604, "n": 300}, {"repeat": 2, "p50_ms": 2.7246240377426147, "p95_ms": 2.7507567763328553, "p99_ms": 2.7573139786720278, "mean_ms": 2.7285752503077187, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_c6a5fcb1e5f12d0d_sdpa_efficient.json", "configuration": "480x832_tp1_sp4", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp1_sp4_v1", "audit_status": "pending"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_cross_480x832_tp1_sp4_v1/captures/qkv_rank0_c6a5fcb1e5f12d0d.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 9, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_cudnn", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [29640, 10, 128], "q_len": 4680, "k_len": 18720, "reference_error": {"max_abs": 0.03125, "rmse": 0.0007742347555068339, "relative_l2": 0.0013815203901010595, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.00928316173610133, "rmse": 0.000830095690624396, "relative_l2": 0.0019594941521909135, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.832175999879837, "p95_ms": 0.8510079979896545, "p99_ms": 0.8519788706302642, "mean_ms": 0.8282544006903966, "n": 300}, {"repeat": 1, "p50_ms": 0.832368016242981, "p95_ms": 0.8599696069955826, "p99_ms": 0.8603875076770783, "mean_ms": 0.8282003192106883, "n": 300}, {"repeat": 2, "p50_ms": 0.8352000117301941, "p95_ms": 0.8449599742889404, "p99_ms": 0.8459570944309235, "mean_ms": 0.8314054389794667, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_c6a5fcb1e5f12d0d_sdpa_cudnn.json", "configuration": "480x832_tp1_sp4", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp1_sp4_v1", "audit_status": "pending"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_cross_480x832_tp1_sp4_v1/captures/qkv_rank0_cf2e5e4d322f9347.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 27, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "vllm_fa2", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [29640, 10, 128], "q_len": 4680, "k_len": 28080, "reference_error": {"max_abs": 0.03125, "rmse": 0.0005552815252655393, "relative_l2": 0.000991735346034173, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.008820491288137777, "rmse": 0.0008120587250742453, "relative_l2": 0.0019695529708380166, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 1.9332000017166138, "p95_ms": 1.9448351681232452, "p99_ms": 1.9456985569000245, "mean_ms": 1.9306499155362447, "n": 300}, {"repeat": 1, "p50_ms": 1.9490240216255188, "p95_ms": 2.1567696809768675, "p99_ms": 2.159915204048157, "mean_ms": 1.9769331208864849, "n": 300}, {"repeat": 2, "p50_ms": 1.9423680305480957, "p95_ms": 1.9519232571125031, "p99_ms": 1.9883430325984954, "mean_ms": 1.9413448532422384, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_cf2e5e4d322f9347_vllm_fa2.json", "configuration": "480x832_tp1_sp4", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp1_sp4_v1", "audit_status": "pending"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_cross_480x832_tp1_sp4_v1/captures/qkv_rank0_cf2e5e4d322f9347.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 27, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "vllm_fa3", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [29640, 10, 128], "q_len": 4680, "k_len": 28080, "reference_error": {"max_abs": 0.0, "rmse": 0.0, "relative_l2": 0.0, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.008820491288137777, "rmse": 0.0008167053351597663, "relative_l2": 0.0019808227773380664, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 1.0751360058784485, "p95_ms": 1.0816320180892944, "p99_ms": 1.084964463710785, "mean_ms": 1.0628122637669246, "n": 300}, {"repeat": 1, "p50_ms": 1.0481600165367126, "p95_ms": 1.0629135906696319, "p99_ms": 1.0638111650943756, "mean_ms": 1.0441610664129257, "n": 300}, {"repeat": 2, "p50_ms": 1.0755680203437805, "p95_ms": 1.145059210062027, "p99_ms": 1.145637148618698, "mean_ms": 1.0827430411179861, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_cf2e5e4d322f9347_vllm_fa3.json", "configuration": "480x832_tp1_sp4", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp1_sp4_v1", "audit_status": "pending"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_cross_480x832_tp1_sp4_v1/captures/qkv_rank0_cf2e5e4d322f9347.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 27, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_math", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [29640, 10, 128], "q_len": 4680, "k_len": 28080, "reference_error": {"max_abs": 0.03125, "rmse": 0.0009780852179610475, "relative_l2": 0.0017468646766550496, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.007724527303786921, "rmse": 0.0006858574812207812, "relative_l2": 0.0016634666902770664, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 28.57316780090332, "p95_ms": 29.544556903839112, "p99_ms": 30.897956428527827, "mean_ms": 28.655725644429523, "n": 300}, {"repeat": 1, "p50_ms": 28.56856060028076, "p95_ms": 28.86453046798706, "p99_ms": 29.03346351623535, "mean_ms": 28.576649201711017, "n": 300}, {"repeat": 2, "p50_ms": 28.561552047729492, "p95_ms": 28.886335182189942, "p99_ms": 29.026325664520265, "mean_ms": 28.58124595006307, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_cf2e5e4d322f9347_sdpa_math.json", "configuration": "480x832_tp1_sp4", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp1_sp4_v1", "audit_status": "pending"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_cross_480x832_tp1_sp4_v1/captures/qkv_rank0_cf2e5e4d322f9347.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 27, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_flash", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [29640, 10, 128], "q_len": 4680, "k_len": 28080, "reference_error": {"max_abs": 0.03125, "rmse": 0.0005560456884091784, "relative_l2": 0.000993100144906819, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.008820491288137777, "rmse": 0.0008120515445332757, "relative_l2": 0.00196953555528005, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 2.082495927810669, "p95_ms": 2.0923535227775574, "p99_ms": 2.0953335309028627, "mean_ms": 2.081720538139343, "n": 300}, {"repeat": 1, "p50_ms": 2.0994399785995483, "p95_ms": 2.2414193034172056, "p99_ms": 2.2464994835853576, "mean_ms": 2.118797012170156, "n": 300}, {"repeat": 2, "p50_ms": 2.0988800525665283, "p95_ms": 2.15005441904068, "p99_ms": 2.1692944526672364, "mean_ms": 2.1038309343655905, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_cf2e5e4d322f9347_sdpa_flash.json", "configuration": "480x832_tp1_sp4", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp1_sp4_v1", "audit_status": "pending"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_cross_480x832_tp1_sp4_v1/captures/qkv_rank0_cf2e5e4d322f9347.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 27, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_efficient", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [29640, 10, 128], "q_len": 4680, "k_len": 28080, "reference_error": {"max_abs": 0.03125, "rmse": 0.0007705672939716462, "relative_l2": 0.0013762367144560445, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.008820491288137777, "rmse": 0.000811379428513884, "relative_l2": 0.001967905416889973, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 4.073871850967407, "p95_ms": 4.116089677810669, "p99_ms": 4.137291507720947, "mean_ms": 4.08060139020284, "n": 300}, {"repeat": 1, "p50_ms": 4.071104049682617, "p95_ms": 4.112889766693115, "p99_ms": 4.131513767242431, "mean_ms": 4.077270828882853, "n": 300}, {"repeat": 2, "p50_ms": 4.072319984436035, "p95_ms": 4.113875102996826, "p99_ms": 4.126843638420105, "mean_ms": 4.076357014973959, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_cf2e5e4d322f9347_sdpa_efficient.json", "configuration": "480x832_tp1_sp4", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp1_sp4_v1", "audit_status": "pending"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_cross_480x832_tp1_sp4_v1/captures/qkv_rank0_cf2e5e4d322f9347.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 27, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_cudnn", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [29640, 10, 128], "q_len": 4680, "k_len": 28080, "reference_error": {"max_abs": 0.03125, "rmse": 0.0007701305226022741, "relative_l2": 0.0013754566387909396, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.008820491288137777, "rmse": 0.0008098227142425212, "relative_l2": 0.001964129789434424, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 1.2372959852218628, "p95_ms": 1.254431962966919, "p99_ms": 1.255279403924942, "mean_ms": 1.2280280510584514, "n": 300}, {"repeat": 1, "p50_ms": 1.24617600440979, "p95_ms": 1.2783151805400848, "p99_ms": 1.2796166384220122, "mean_ms": 1.2490914126237234, "n": 300}, {"repeat": 2, "p50_ms": 1.2619680166244507, "p95_ms": 1.4066128492355348, "p99_ms": 1.4270409405231477, "mean_ms": 1.2868122673034668, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_cf2e5e4d322f9347_sdpa_cudnn.json", "configuration": "480x832_tp1_sp4", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp1_sp4_v1", "audit_status": "pending"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_cross_480x832_tp1_sp4_v1/captures/qkv_rank0_df3a5d6865cf4937.pt", "context": {"phase": "commit", "step": null, "start_frame": 9, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "vllm_fa2", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [29640, 10, 128], "q_len": 4680, "k_len": 18720, "reference_error": {"max_abs": 0.0625, "rmse": 0.0004193072287393182, "relative_l2": 0.0005270094373548114, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.012678517544874701, "rmse": 0.0009683098051771197, "relative_l2": 0.0016927232205301553, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 1.2881119847297668, "p95_ms": 1.2932207882404327, "p99_ms": 1.2957183945178985, "mean_ms": 1.2860267718633016, "n": 300}, {"repeat": 1, "p50_ms": 1.2954879999160767, "p95_ms": 1.3011856555938721, "p99_ms": 1.3027970862388611, "mean_ms": 1.2933501867453256, "n": 300}, {"repeat": 2, "p50_ms": 1.3104479908943176, "p95_ms": 1.4070447981357574, "p99_ms": 1.4085760116577148, "mean_ms": 1.3274883194764455, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_df3a5d6865cf4937_vllm_fa2.json", "configuration": "480x832_tp1_sp4", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp1_sp4_v1", "audit_status": "pending"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_cross_480x832_tp1_sp4_v1/captures/qkv_rank0_df3a5d6865cf4937.pt", "context": {"phase": "commit", "step": null, "start_frame": 9, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "vllm_fa3", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [29640, 10, 128], "q_len": 4680, "k_len": 18720, "reference_error": {"max_abs": 0.0, "rmse": 0.0, "relative_l2": 0.0, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.012678517544874701, "rmse": 0.0009692782411752042, "relative_l2": 0.001694416164351224, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.7008320093154907, "p95_ms": 0.7074912011623382, "p99_ms": 0.7103145813941956, "mean_ms": 0.68625290731589, "n": 300}, {"repeat": 1, "p50_ms": 0.7038080096244812, "p95_ms": 0.7091040253639221, "p99_ms": 0.7136054199934005, "mean_ms": 0.6990899195273718, "n": 300}, {"repeat": 2, "p50_ms": 0.7041120231151581, "p95_ms": 0.7092816144227981, "p99_ms": 0.7134748888015747, "mean_ms": 0.7032470419009527, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_df3a5d6865cf4937_vllm_fa3.json", "configuration": "480x832_tp1_sp4", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp1_sp4_v1", "audit_status": "pending"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_cross_480x832_tp1_sp4_v1/captures/qkv_rank0_df3a5d6865cf4937.pt", "context": {"phase": "commit", "step": null, "start_frame": 9, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_math", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [29640, 10, 128], "q_len": 4680, "k_len": 18720, "reference_error": {"max_abs": 0.0625, "rmse": 0.0007898468096038272, "relative_l2": 0.000992724890475454, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.012678517544874701, "rmse": 0.00094767651387136, "relative_l2": 0.001656653719712869, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 19.030559539794922, "p95_ms": 19.599042415618896, "p99_ms": 20.284132518768306, "mean_ms": 19.076566734313964, "n": 300}, {"repeat": 1, "p50_ms": 19.063055992126465, "p95_ms": 19.34866361618042, "p99_ms": 19.447714824676513, "mean_ms": 19.067182801564535, "n": 300}, {"repeat": 2, "p50_ms": 19.068431854248047, "p95_ms": 19.32064332962036, "p99_ms": 19.38201732635498, "mean_ms": 19.05924919128418, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_df3a5d6865cf4937_sdpa_math.json", "configuration": "480x832_tp1_sp4", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp1_sp4_v1", "audit_status": "pending"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_cross_480x832_tp1_sp4_v1/captures/qkv_rank0_df3a5d6865cf4937.pt", "context": {"phase": "commit", "step": null, "start_frame": 9, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_flash", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [29640, 10, 128], "q_len": 4680, "k_len": 18720, "reference_error": {"max_abs": 0.0625, "rmse": 0.0004200155646248864, "relative_l2": 0.0005278997146286704, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.012678517544874701, "rmse": 0.0009683237781535962, "relative_l2": 0.0016927476469912064, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 1.3933600187301636, "p95_ms": 1.3968000411987305, "p99_ms": 1.398499231338501, "mean_ms": 1.3896048033237458, "n": 300}, {"repeat": 1, "p50_ms": 1.394271969795227, "p95_ms": 1.3978655993938447, "p99_ms": 1.4183071672916412, "mean_ms": 1.3908550425370534, "n": 300}, {"repeat": 2, "p50_ms": 1.4071840047836304, "p95_ms": 1.4537999749183654, "p99_ms": 1.4566176569461822, "mean_ms": 1.4167628828684489, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_df3a5d6865cf4937_sdpa_flash.json", "configuration": "480x832_tp1_sp4", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp1_sp4_v1", "audit_status": "pending"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_cross_480x832_tp1_sp4_v1/captures/qkv_rank0_df3a5d6865cf4937.pt", "context": {"phase": "commit", "step": null, "start_frame": 9, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_efficient", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [29640, 10, 128], "q_len": 4680, "k_len": 18720, "reference_error": {"max_abs": 0.0625, "rmse": 0.0006686263996730479, "relative_l2": 0.0008403681084910054, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.012678517544874701, "rmse": 0.0009547840566502768, "relative_l2": 0.0016690785682876319, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 2.720639944076538, "p95_ms": 2.7612529039382934, "p99_ms": 2.767498562335968, "mean_ms": 2.7275101868311564, "n": 300}, {"repeat": 1, "p50_ms": 2.7206718921661377, "p95_ms": 2.7431407809257506, "p99_ms": 2.758498656749725, "mean_ms": 2.724587409496307, "n": 300}, {"repeat": 2, "p50_ms": 2.7212640047073364, "p95_ms": 2.7444463968276978, "p99_ms": 2.752480401992798, "mean_ms": 2.7249325927098593, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_df3a5d6865cf4937_sdpa_efficient.json", "configuration": "480x832_tp1_sp4", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp1_sp4_v1", "audit_status": "pending"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_cross_480x832_tp1_sp4_v1/captures/qkv_rank0_df3a5d6865cf4937.pt", "context": {"phase": "commit", "step": null, "start_frame": 9, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_cudnn", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [29640, 10, 128], "q_len": 4680, "k_len": 18720, "reference_error": {"max_abs": 0.0625, "rmse": 0.000667980162633798, "relative_l2": 0.0008395558806181951, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.012678517544874701, "rmse": 0.0009547032318614938, "relative_l2": 0.0016689372767340032, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.8315680027008057, "p95_ms": 0.8520047992467881, "p99_ms": 0.852832019329071, "mean_ms": 0.8211341877778371, "n": 300}, {"repeat": 1, "p50_ms": 0.8321920037269592, "p95_ms": 0.8385391771793366, "p99_ms": 0.8429772543907166, "mean_ms": 0.8283171204725901, "n": 300}, {"repeat": 2, "p50_ms": 0.8319039940834045, "p95_ms": 0.8362256199121475, "p99_ms": 0.8419545859098434, "mean_ms": 0.8290972795089085, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_df3a5d6865cf4937_sdpa_cudnn.json", "configuration": "480x832_tp1_sp4", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp1_sp4_v1", "audit_status": "pending"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp2_sp1_v1/captures/qkv_rank0_0c1b0b9bae2b8b48.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 0, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "vllm_fa2", "status": "measured", "q_shape": [4680, 20, 128], "k_shape": [512, 20, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.0009765625, "rmse": 2.1592777033861512e-06, "relative_l2": 0.00026356157618343125, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.0005221612900013495, "rmse": 1.9680953528593928e-05, "relative_l2": 0.0019826252521852918, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.08367999643087387, "p95_ms": 0.08457600325345993, "p99_ms": 0.08483199775218964, "mean_ms": 0.08385162681341171, "n": 300}, {"repeat": 1, "p50_ms": 0.08371199667453766, "p95_ms": 0.08461120314896106, "p99_ms": 0.08489727824926377, "mean_ms": 0.08413877343138058, "n": 300}, {"repeat": 2, "p50_ms": 0.08367999643087387, "p95_ms": 0.08460800349712372, "p99_ms": 0.08512127995491028, "mean_ms": 0.08404949386914572, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_0c1b0b9bae2b8b48_vllm_fa2.json", "configuration": "480x832_tp2_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp2_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp2_sp1_v1/captures/qkv_rank0_0c1b0b9bae2b8b48.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 0, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "vllm_fa3", "status": "measured", "q_shape": [4680, 20, 128], "k_shape": [512, 20, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.0009765625, "rmse": 8.498341224170999e-07, "relative_l2": 0.00010373080796762069, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.0005221612900013495, "rmse": 1.9692264907610615e-05, "relative_l2": 0.0019837647409627506, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.07318399846553802, "p95_ms": 0.07407999783754349, "p99_ms": 0.07446560077369213, "mean_ms": 0.07334656002620855, "n": 300}, {"repeat": 1, "p50_ms": 0.07308799773454666, "p95_ms": 0.07411199808120728, "p99_ms": 0.07459264174103737, "mean_ms": 0.07363455985983212, "n": 300}, {"repeat": 2, "p50_ms": 0.07302399724721909, "p95_ms": 0.07388959638774395, "p99_ms": 0.07443616054952144, "mean_ms": 0.0734310399244229, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_0c1b0b9bae2b8b48_vllm_fa3.json", "configuration": "480x832_tp2_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp2_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp2_sp1_v1/captures/qkv_rank0_0c1b0b9bae2b8b48.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 0, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_math", "status": "measured", "q_shape": [4680, 20, 128], "k_shape": [512, 20, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.001953125, "rmse": 1.1453516390045069e-05, "relative_l2": 0.00139801695162653, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.0004544012099986505, "rmse": 1.6037217321311435e-05, "relative_l2": 0.0016155615625950317, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 1.1243039965629578, "p95_ms": 1.127616047859192, "p99_ms": 1.1288393425941468, "mean_ms": 1.1243529641628265, "n": 300}, {"repeat": 1, "p50_ms": 1.1239680051803589, "p95_ms": 1.126822429895401, "p99_ms": 1.1286108684539795, "mean_ms": 1.1243945614496866, "n": 300}, {"repeat": 2, "p50_ms": 1.124176025390625, "p95_ms": 1.1277631640434265, "p99_ms": 1.1296661853790284, "mean_ms": 1.124366830587387, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_0c1b0b9bae2b8b48_sdpa_math.json", "configuration": "480x832_tp2_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp2_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp2_sp1_v1/captures/qkv_rank0_0c1b0b9bae2b8b48.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 0, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_flash", "status": "measured", "q_shape": [4680, 20, 128], "k_shape": [512, 20, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.0009765625, "rmse": 2.160761994797744e-06, "relative_l2": 0.0002637427488891659, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.0005221612900013495, "rmse": 1.9680953528593928e-05, "relative_l2": 0.0019826252521852918, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.0888959988951683, "p95_ms": 0.0899519994854927, "p99_ms": 0.0907849583774805, "mean_ms": 0.08913311953345934, "n": 300}, {"repeat": 1, "p50_ms": 0.08892799913883209, "p95_ms": 0.08982399851083755, "p99_ms": 0.09027200192213058, "mean_ms": 0.0891833595186472, "n": 300}, {"repeat": 2, "p50_ms": 0.08876799792051315, "p95_ms": 0.0896959975361824, "p99_ms": 0.09014400094747543, "mean_ms": 0.08901162629326184, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_0c1b0b9bae2b8b48_sdpa_flash.json", "configuration": "480x832_tp2_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp2_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp2_sp1_v1/captures/qkv_rank0_0c1b0b9bae2b8b48.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 0, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_efficient", "status": "measured", "q_shape": [4680, 20, 128], "k_shape": [512, 20, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.0009765625, "rmse": 3.7735380020768966e-06, "relative_l2": 0.0004605982926863958, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.0005221612900013495, "rmse": 1.97311176669295e-05, "relative_l2": 0.0019876787008037094, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.16579200327396393, "p95_ms": 0.1664975941181183, "p99_ms": 0.1668512038886547, "mean_ms": 0.16591914688547452, "n": 300}, {"repeat": 1, "p50_ms": 0.16592000424861908, "p95_ms": 0.1664975941181183, "p99_ms": 0.16688031673431397, "mean_ms": 0.16610922674338022, "n": 300}, {"repeat": 2, "p50_ms": 0.16592000424861908, "p95_ms": 0.16655999422073364, "p99_ms": 0.16713727861642838, "mean_ms": 0.16609408001104992, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_0c1b0b9bae2b8b48_sdpa_efficient.json", "configuration": "480x832_tp2_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp2_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp2_sp1_v1/captures/qkv_rank0_0c1b0b9bae2b8b48.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 0, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_cudnn", "status": "measured", "q_shape": [4680, 20, 128], "k_shape": [512, 20, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.0009765625, "rmse": 3.7687607682297338e-06, "relative_l2": 0.00046001518321391776, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.0005221612900013495, "rmse": 1.9731270952981146e-05, "relative_l2": 0.001987694142575693, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.052191998809576035, "p95_ms": 0.05315359868109226, "p99_ms": 0.06738847881555554, "mean_ms": 0.052807146633664766, "n": 300}, {"repeat": 1, "p50_ms": 0.052480001002550125, "p95_ms": 0.08897439911961558, "p99_ms": 0.18926687568426132, "mean_ms": 0.0603775999819239, "n": 300}, {"repeat": 2, "p50_ms": 0.05225599929690361, "p95_ms": 0.0533135998994112, "p99_ms": 0.06286176145076748, "mean_ms": 0.05308735975374778, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_0c1b0b9bae2b8b48_sdpa_cudnn.json", "configuration": "480x832_tp2_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp2_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp2_sp1_v1/captures/qkv_rank0_0c1b0b9bae2b8b48.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 0, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "native_dense", "status": "measured", "q_shape": [4680, 20, 128], "k_shape": [512, 20, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.0, "rmse": 0.0, "relative_l2": 0.0, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.0005221612900013495, "rmse": 1.9693656912287394e-05, "relative_l2": 0.001983904969108675, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.11347199976444244, "p95_ms": 0.15694720298051834, "p99_ms": 0.17059999510645865, "mean_ms": 0.1204038400699695, "n": 300}, {"repeat": 1, "p50_ms": 0.11294400319457054, "p95_ms": 0.1578288018703461, "p99_ms": 0.16982688412070274, "mean_ms": 0.11991839987536271, "n": 300}, {"repeat": 2, "p50_ms": 0.11321599781513214, "p95_ms": 0.15976959466934204, "p99_ms": 0.18304287314414977, "mean_ms": 0.11900127989550431, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_0c1b0b9bae2b8b48_native_dense.json", "configuration": "480x832_tp2_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp2_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp2_sp1_v1/captures/qkv_rank0_2e0b4394096e26d5.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 9, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "vllm_fa2", "status": "measured", "q_shape": [4680, 20, 128], "k_shape": [29640, 20, 128], "q_len": 4680, "k_len": 18720, "reference_error": {"max_abs": 0.03125, "rmse": 0.00048482297673525744, "relative_l2": 0.000928875510902381, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.008338139423881952, "rmse": 0.0008250061956286321, "relative_l2": 0.0019467300166385222, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 2.4072799682617188, "p95_ms": 2.4440367579460145, "p99_ms": 2.4454742431640626, "mean_ms": 2.415208637714386, "n": 300}, {"repeat": 1, "p50_ms": 2.4252960681915283, "p95_ms": 2.5940192699432374, "p99_ms": 2.617798612117767, "mean_ms": 2.455643417040507, "n": 300}, {"repeat": 2, "p50_ms": 2.445456027984619, "p95_ms": 2.5495423913002013, "p99_ms": 2.553216300010681, "mean_ms": 2.4623561533292135, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_2e0b4394096e26d5_vllm_fa2.json", "configuration": "480x832_tp2_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp2_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp2_sp1_v1/captures/qkv_rank0_2e0b4394096e26d5.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 9, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "vllm_fa3", "status": "measured", "q_shape": [4680, 20, 128], "k_shape": [29640, 20, 128], "q_len": 4680, "k_len": 18720, "reference_error": {"max_abs": 0.0, "rmse": 0.0, "relative_l2": 0.0, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.008338139423881952, "rmse": 0.0008372650837925386, "relative_l2": 0.0019756567637172, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 1.3594080209732056, "p95_ms": 1.3783999681472778, "p99_ms": 1.3790416026115417, "mean_ms": 1.357194667259852, "n": 300}, {"repeat": 1, "p50_ms": 1.392080008983612, "p95_ms": 1.4094927906990051, "p99_ms": 1.4201324677467346, "mean_ms": 1.3757801612218221, "n": 300}, {"repeat": 2, "p50_ms": 1.3785279989242554, "p95_ms": 1.3976016521453858, "p99_ms": 1.3988851249217986, "mean_ms": 1.3837217052777608, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_2e0b4394096e26d5_vllm_fa3.json", "configuration": "480x832_tp2_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp2_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp2_sp1_v1/captures/qkv_rank0_2e0b4394096e26d5.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 9, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_math", "status": "measured", "q_shape": [4680, 20, 128], "k_shape": [29640, 20, 128], "q_len": 4680, "k_len": 18720, "reference_error": {"max_abs": 0.03125, "rmse": 0.0008285694457314331, "relative_l2": 0.001587461618268468, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.007445773795270227, "rmse": 0.0007114648130294856, "relative_l2": 0.0016788115224410634, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 35.09817695617676, "p95_ms": 35.12536735534668, "p99_ms": 35.128933601379394, "mean_ms": 35.102921384175616, "n": 300}, {"repeat": 1, "p50_ms": 35.09872055053711, "p95_ms": 35.14108657836914, "p99_ms": 35.14982082366944, "mean_ms": 35.105457928975426, "n": 300}, {"repeat": 2, "p50_ms": 35.09856033325195, "p95_ms": 35.136512756347656, "p99_ms": 35.14490280151367, "mean_ms": 35.10339228312174, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_2e0b4394096e26d5_sdpa_math.json", "configuration": "480x832_tp2_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp2_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp2_sp1_v1/captures/qkv_rank0_2e0b4394096e26d5.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 9, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_flash", "status": "measured", "q_shape": [4680, 20, 128], "k_shape": [29640, 20, 128], "q_len": 4680, "k_len": 18720, "reference_error": {"max_abs": 0.03125, "rmse": 0.000485181757920156, "relative_l2": 0.0009295629021202416, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.008338139423881952, "rmse": 0.000825045503408997, "relative_l2": 0.0019468227694400518, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 2.6184799671173096, "p95_ms": 2.6627535700798033, "p99_ms": 2.745675082206726, "mean_ms": 2.62400906085968, "n": 300}, {"repeat": 1, "p50_ms": 2.6381280422210693, "p95_ms": 2.8284641027450563, "p99_ms": 2.832963092327118, "mean_ms": 2.661311779022217, "n": 300}, {"repeat": 2, "p50_ms": 2.642975926399231, "p95_ms": 2.7856351137161255, "p99_ms": 2.811303987503052, "mean_ms": 2.671328422228495, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_2e0b4394096e26d5_sdpa_flash.json", "configuration": "480x832_tp2_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp2_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp2_sp1_v1/captures/qkv_rank0_2e0b4394096e26d5.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 9, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_efficient", "status": "measured", "q_shape": [4680, 20, 128], "k_shape": [29640, 20, 128], "q_len": 4680, "k_len": 18720, "reference_error": {"max_abs": 0.03125, "rmse": 0.0006785907971350899, "relative_l2": 0.0013001165448615008, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.009269161331428855, "rmse": 0.0008322708750293184, "relative_l2": 0.0019638721539042858, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 5.44376015663147, "p95_ms": 5.455091118812561, "p99_ms": 5.4880239152908326, "mean_ms": 5.444030407269795, "n": 300}, {"repeat": 1, "p50_ms": 5.443552017211914, "p95_ms": 5.452379250526429, "p99_ms": 5.457418832778931, "mean_ms": 5.4426936483383175, "n": 300}, {"repeat": 2, "p50_ms": 5.445248126983643, "p95_ms": 5.456598281860352, "p99_ms": 5.462562093734741, "mean_ms": 5.444183580080668, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_2e0b4394096e26d5_sdpa_efficient.json", "configuration": "480x832_tp2_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp2_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp2_sp1_v1/captures/qkv_rank0_2e0b4394096e26d5.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 9, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_cudnn", "status": "measured", "q_shape": [4680, 20, 128], "k_shape": [29640, 20, 128], "q_len": 4680, "k_len": 18720, "reference_error": {"max_abs": 0.03125, "rmse": 0.0006772172874122102, "relative_l2": 0.0012974850286034212, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.009269161331428855, "rmse": 0.0008319531279590026, "relative_l2": 0.001963122380432571, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 1.614080011844635, "p95_ms": 1.665027242898941, "p99_ms": 1.6707251012325286, "mean_ms": 1.6113681066036225, "n": 300}, {"repeat": 1, "p50_ms": 1.618831992149353, "p95_ms": 1.8780495464801787, "p99_ms": 1.9348485910892486, "mean_ms": 1.6634151474634806, "n": 300}, {"repeat": 2, "p50_ms": 1.6286240220069885, "p95_ms": 1.6760015964508057, "p99_ms": 1.679296656847, "mean_ms": 1.620306239128113, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_2e0b4394096e26d5_sdpa_cudnn.json", "configuration": "480x832_tp2_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp2_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp2_sp1_v1/captures/qkv_rank0_2e0b4394096e26d5.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 9, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "native_dense", "status": "measured", "q_shape": [4680, 20, 128], "k_shape": [29640, 20, 128], "q_len": 4680, "k_len": 18720, "reference_error": {"max_abs": 0.03125, "rmse": 0.00043674344646728774, "relative_l2": 0.0008367596245177434, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.008338139423881952, "rmse": 0.0008357267575398385, "relative_l2": 0.0019720268444421906, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 1.4044640064239502, "p95_ms": 1.4897696375846863, "p99_ms": 1.5000665748119355, "mean_ms": 1.3929896541436513, "n": 300}, {"repeat": 1, "p50_ms": 1.36080002784729, "p95_ms": 1.3808720231056213, "p99_ms": 1.4152780234813689, "mean_ms": 1.3597438943386078, "n": 300}, {"repeat": 2, "p50_ms": 1.3645280003547668, "p95_ms": 1.4030095875263213, "p99_ms": 1.4156182754039763, "mean_ms": 1.3645372811953227, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_2e0b4394096e26d5_native_dense.json", "configuration": "480x832_tp2_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp2_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp2_sp1_v1/captures/qkv_rank0_325560d6739aa74a.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 9, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "vllm_fa2", "status": "measured", "q_shape": [4680, 20, 128], "k_shape": [512, 20, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.00048828125, "rmse": 1.9005887193494505e-06, "relative_l2": 0.0005731343305329263, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 8.797855827336054e-05, "rmse": 5.130591479296537e-06, "relative_l2": 0.0019279313287139288, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.08356800302863121, "p95_ms": 0.08444800227880478, "p99_ms": 0.08486527800559998, "mean_ms": 0.08379786727329096, "n": 300}, {"repeat": 1, "p50_ms": 0.08352000266313553, "p95_ms": 0.0843552015721798, "p99_ms": 0.08470431677997112, "mean_ms": 0.08393440062801043, "n": 300}, {"repeat": 2, "p50_ms": 0.0835840031504631, "p95_ms": 0.08457760326564312, "p99_ms": 0.08483231775462627, "mean_ms": 0.0839610670755307, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_325560d6739aa74a_vllm_fa2.json", "configuration": "480x832_tp2_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp2_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp2_sp1_v1/captures/qkv_rank0_325560d6739aa74a.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 9, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "vllm_fa3", "status": "measured", "q_shape": [4680, 20, 128], "k_shape": [512, 20, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.00048828125, "rmse": 7.357741288347569e-07, "relative_l2": 0.00022187725753602155, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 8.797855827336054e-05, "rmse": 5.1625492166297935e-06, "relative_l2": 0.0019399401435354208, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.07276800274848938, "p95_ms": 0.07379200309515, "p99_ms": 0.07459616169333458, "mean_ms": 0.07297056031723817, "n": 300}, {"repeat": 1, "p50_ms": 0.07278400287032127, "p95_ms": 0.07366560213267803, "p99_ms": 0.07876576356589793, "mean_ms": 0.07345994681119919, "n": 300}, {"repeat": 2, "p50_ms": 0.07248000055551529, "p95_ms": 0.07360160164535046, "p99_ms": 0.07405471764504909, "mean_ms": 0.07304693341255188, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_325560d6739aa74a_vllm_fa3.json", "configuration": "480x832_tp2_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp2_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp2_sp1_v1/captures/qkv_rank0_325560d6739aa74a.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 9, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_math", "status": "measured", "q_shape": [4680, 20, 128], "k_shape": [512, 20, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.0009765625, "rmse": 5.550971830922063e-06, "relative_l2": 0.001673930025856211, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 5.9410297260108214e-05, "rmse": 4.423149753197181e-06, "relative_l2": 0.0016620947146529293, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 1.1242560148239136, "p95_ms": 1.128225612640381, "p99_ms": 1.1298886966705322, "mean_ms": 1.1248364794254302, "n": 300}, {"repeat": 1, "p50_ms": 1.124015986919403, "p95_ms": 1.1274912178516387, "p99_ms": 1.1291875278949737, "mean_ms": 1.1252183485031129, "n": 300}, {"repeat": 2, "p50_ms": 1.1240639686584473, "p95_ms": 1.1275871813297271, "p99_ms": 1.1296992111206055, "mean_ms": 1.1246238907178243, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_325560d6739aa74a_sdpa_math.json", "configuration": "480x832_tp2_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp2_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp2_sp1_v1/captures/qkv_rank0_325560d6739aa74a.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 9, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_flash", "status": "measured", "q_shape": [4680, 20, 128], "k_shape": [512, 20, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.00048828125, "rmse": 1.9008311664502724e-06, "relative_l2": 0.0005732074419617191, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 8.797855827336054e-05, "rmse": 5.130591479296537e-06, "relative_l2": 0.0019279313287139288, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.08876799792051315, "p95_ms": 0.08956799656152725, "p99_ms": 0.09020832143723964, "mean_ms": 0.0889214929441611, "n": 300}, {"repeat": 1, "p50_ms": 0.08867199718952179, "p95_ms": 0.08963359706103802, "p99_ms": 0.08992127925157548, "mean_ms": 0.08898207982381184, "n": 300}, {"repeat": 2, "p50_ms": 0.088639996945858, "p95_ms": 0.08956799656152725, "p99_ms": 0.08985663875937462, "mean_ms": 0.08891722659269968, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_325560d6739aa74a_sdpa_flash.json", "configuration": "480x832_tp2_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp2_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp2_sp1_v1/captures/qkv_rank0_325560d6739aa74a.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 9, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_efficient", "status": "measured", "q_shape": [4680, 20, 128], "k_shape": [512, 20, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.00048828125, "rmse": 3.326001439660204e-06, "relative_l2": 0.001002976387823498, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 8.797855827336054e-05, "rmse": 5.5850438234574005e-06, "relative_l2": 0.002098701680485406, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.16566400229930878, "p95_ms": 0.16636959314346314, "p99_ms": 0.16672032281756402, "mean_ms": 0.1660413863758246, "n": 300}, {"repeat": 1, "p50_ms": 0.16572800278663635, "p95_ms": 0.16640160009264945, "p99_ms": 0.16672064289450644, "mean_ms": 0.16598325336972872, "n": 300}, {"repeat": 2, "p50_ms": 0.1656000018119812, "p95_ms": 0.16633759960532188, "p99_ms": 0.16672000288963318, "mean_ms": 0.16590303947528204, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_325560d6739aa74a_sdpa_efficient.json", "configuration": "480x832_tp2_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp2_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp2_sp1_v1/captures/qkv_rank0_325560d6739aa74a.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 9, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_cudnn", "status": "measured", "q_shape": [4680, 20, 128], "k_shape": [512, 20, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.00048828125, "rmse": 3.325790277589948e-06, "relative_l2": 0.001002912710589945, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 8.797855827336054e-05, "rmse": 5.5850438234574005e-06, "relative_l2": 0.002098701680485406, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.051872000098228455, "p95_ms": 0.05283359996974468, "p99_ms": 0.05344480086117982, "mean_ms": 0.05211200008789698, "n": 300}, {"repeat": 1, "p50_ms": 0.052000001072883606, "p95_ms": 0.05299360118806362, "p99_ms": 0.0538636802881956, "mean_ms": 0.05252725366503, "n": 300}, {"repeat": 2, "p50_ms": 0.05198400095105171, "p95_ms": 0.05289760045707226, "p99_ms": 0.05341344062238931, "mean_ms": 0.052484480179846284, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_325560d6739aa74a_sdpa_cudnn.json", "configuration": "480x832_tp2_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp2_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp2_sp1_v1/captures/qkv_rank0_325560d6739aa74a.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 9, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "native_dense", "status": "measured", "q_shape": [4680, 20, 128], "k_shape": [512, 20, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.0, "rmse": 0.0, "relative_l2": 0.0, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 8.797855827336054e-05, "rmse": 5.168994684931184e-06, "relative_l2": 0.0019423621684260455, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.11203200370073318, "p95_ms": 0.12029119953513145, "p99_ms": 0.15978656113147724, "mean_ms": 0.11439370637138685, "n": 300}, {"repeat": 1, "p50_ms": 0.11243199929594994, "p95_ms": 0.11841920018196106, "p99_ms": 0.1263334468007086, "mean_ms": 0.11382688018182914, "n": 300}, {"repeat": 2, "p50_ms": 0.11201600357890129, "p95_ms": 0.12006719782948494, "p99_ms": 0.1362179134786128, "mean_ms": 0.11362399995326995, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_325560d6739aa74a_native_dense.json", "configuration": "480x832_tp2_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp2_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp2_sp1_v1/captures/qkv_rank0_60db4e364fe9a358.pt", "context": {"phase": "commit", "step": null, "start_frame": 9, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "vllm_fa2", "status": "measured", "q_shape": [4680, 20, 128], "k_shape": [512, 20, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.000244140625, "rmse": 1.6376318745477493e-06, "relative_l2": 0.0005724447641917581, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.0001101750380461744, "rmse": 4.55905245454031e-06, "relative_l2": 0.0017790650100901059, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.08291199803352356, "p95_ms": 0.0835856031626463, "p99_ms": 0.08406591936945915, "mean_ms": 0.08311775940159957, "n": 300}, {"repeat": 1, "p50_ms": 0.08291199803352356, "p95_ms": 0.08368159644305706, "p99_ms": 0.08435296155512333, "mean_ms": 0.08330357273419699, "n": 300}, {"repeat": 2, "p50_ms": 0.0830719992518425, "p95_ms": 0.08390719816088676, "p99_ms": 0.08451488271355628, "mean_ms": 0.08342026621103286, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_60db4e364fe9a358_vllm_fa2.json", "configuration": "480x832_tp2_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp2_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp2_sp1_v1/captures/qkv_rank0_60db4e364fe9a358.pt", "context": {"phase": "commit", "step": null, "start_frame": 9, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "vllm_fa3", "status": "measured", "q_shape": [4680, 20, 128], "k_shape": [512, 20, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.0001220703125, "rmse": 6.145058380888728e-07, "relative_l2": 0.00021480447165599348, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.0001101750380461744, "rmse": 4.606789203010599e-06, "relative_l2": 0.0017976931745487944, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.07222399860620499, "p95_ms": 0.07308799773454666, "p99_ms": 0.0734726406633854, "mean_ms": 0.07242698676884174, "n": 300}, {"repeat": 1, "p50_ms": 0.07212799787521362, "p95_ms": 0.07308959774672985, "p99_ms": 0.07395775690674782, "mean_ms": 0.07270069340864817, "n": 300}, {"repeat": 2, "p50_ms": 0.07203199714422226, "p95_ms": 0.0729919970035553, "p99_ms": 0.0734726406633854, "mean_ms": 0.07257024005055428, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_60db4e364fe9a358_vllm_fa3.json", "configuration": "480x832_tp2_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp2_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp2_sp1_v1/captures/qkv_rank0_60db4e364fe9a358.pt", "context": {"phase": "commit", "step": null, "start_frame": 9, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_math", "status": "measured", "q_shape": [4680, 20, 128], "k_shape": [512, 20, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.001953125, "rmse": 4.581489700584646e-06, "relative_l2": 0.0016014892187063842, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.0001101750380461744, "rmse": 4.345600527834753e-06, "relative_l2": 0.0016957703215721024, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 1.1235840320587158, "p95_ms": 1.126047968864441, "p99_ms": 1.1270764565467835, "mean_ms": 1.1260099216302235, "n": 300}, {"repeat": 1, "p50_ms": 1.1235840320587158, "p95_ms": 1.1278416395187378, "p99_ms": 1.128801953792572, "mean_ms": 1.1244426667690277, "n": 300}, {"repeat": 2, "p50_ms": 1.1234560012817383, "p95_ms": 1.1265328228473663, "p99_ms": 1.1278092551231385, "mean_ms": 1.1239518960316977, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_60db4e364fe9a358_sdpa_math.json", "configuration": "480x832_tp2_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp2_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp2_sp1_v1/captures/qkv_rank0_60db4e364fe9a358.pt", "context": {"phase": "commit", "step": null, "start_frame": 9, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_flash", "status": "measured", "q_shape": [4680, 20, 128], "k_shape": [512, 20, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.000244140625, "rmse": 1.6372100003539122e-06, "relative_l2": 0.0005722972953514387, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.0001101750380461744, "rmse": 4.55905245454031e-06, "relative_l2": 0.0017790650100901059, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.088639996945858, "p95_ms": 0.08963359706103802, "p99_ms": 0.0900483202189207, "mean_ms": 0.08879530653357506, "n": 300}, {"repeat": 1, "p50_ms": 0.08867199718952179, "p95_ms": 0.08963199704885483, "p99_ms": 0.09056223668158055, "mean_ms": 0.08898879972596964, "n": 300}, {"repeat": 2, "p50_ms": 0.08883199840784073, "p95_ms": 0.08988959901034832, "p99_ms": 0.09020832143723964, "mean_ms": 0.08914698652923107, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_60db4e364fe9a358_sdpa_flash.json", "configuration": "480x832_tp2_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp2_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp2_sp1_v1/captures/qkv_rank0_60db4e364fe9a358.pt", "context": {"phase": "commit", "step": null, "start_frame": 9, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_efficient", "status": "measured", "q_shape": [4680, 20, 128], "k_shape": [512, 20, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.00048828125, "rmse": 2.864474438521178e-06, "relative_l2": 0.0010012954803688167, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.0001101750380461744, "rmse": 5.1918242283224114e-06, "relative_l2": 0.002025989592189896, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.16582399606704712, "p95_ms": 0.16640160009264945, "p99_ms": 0.16685088381171226, "mean_ms": 0.16589983974893888, "n": 300}, {"repeat": 1, "p50_ms": 0.16547200083732605, "p95_ms": 0.166143998503685, "p99_ms": 0.16646528095006943, "mean_ms": 0.16570815960566201, "n": 300}, {"repeat": 2, "p50_ms": 0.165583997964859, "p95_ms": 0.16652800142765045, "p99_ms": 0.1667888033390045, "mean_ms": 0.16596138606468835, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_60db4e364fe9a358_sdpa_efficient.json", "configuration": "480x832_tp2_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp2_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp2_sp1_v1/captures/qkv_rank0_60db4e364fe9a358.pt", "context": {"phase": "commit", "step": null, "start_frame": 9, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_cudnn", "status": "measured", "q_shape": [4680, 20, 128], "k_shape": [512, 20, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.00048828125, "rmse": 2.8631005852361296e-06, "relative_l2": 0.0010008152411086894, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.0001101750380461744, "rmse": 5.192060864126505e-06, "relative_l2": 0.002026081933851545, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.052032001316547394, "p95_ms": 0.052960000932216644, "p99_ms": 0.0533135998994112, "mean_ms": 0.05233098675807317, "n": 300}, {"repeat": 1, "p50_ms": 0.05206400156021118, "p95_ms": 0.052931200712919235, "p99_ms": 0.053516161330044255, "mean_ms": 0.0525772799924016, "n": 300}, {"repeat": 2, "p50_ms": 0.05206400156021118, "p95_ms": 0.0530879981815815, "p99_ms": 0.05344576086848974, "mean_ms": 0.05253610669324795, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_60db4e364fe9a358_sdpa_cudnn.json", "configuration": "480x832_tp2_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp2_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp2_sp1_v1/captures/qkv_rank0_60db4e364fe9a358.pt", "context": {"phase": "commit", "step": null, "start_frame": 9, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "native_dense", "status": "measured", "q_shape": [4680, 20, 128], "k_shape": [512, 20, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.0, "rmse": 0.0, "relative_l2": 0.0, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.0001101750380461744, "rmse": 4.618265562926498e-06, "relative_l2": 0.0018021715591633961, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.11296000331640244, "p95_ms": 0.11852800101041794, "p99_ms": 0.12586079716682433, "mean_ms": 0.11382186683515708, "n": 300}, {"repeat": 1, "p50_ms": 0.11190400272607803, "p95_ms": 0.11850720085203648, "p99_ms": 0.13841888278722755, "mean_ms": 0.11339071991542975, "n": 300}, {"repeat": 2, "p50_ms": 0.11215999722480774, "p95_ms": 0.11862880177795887, "p99_ms": 0.12572479575872417, "mean_ms": 0.11509482674300671, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_60db4e364fe9a358_native_dense.json", "configuration": "480x832_tp2_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp2_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp2_sp1_v1/captures/qkv_rank0_745464bfa49527f7.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 27, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "vllm_fa2", "status": "measured", "q_shape": [4680, 20, 128], "k_shape": [29640, 20, 128], "q_len": 4680, "k_len": 28080, "reference_error": {"max_abs": 0.03125, "rmse": 0.00047688118179451696, "relative_l2": 0.0009191721890233742, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.0088576065358561, "rmse": 0.0008021409422662321, "relative_l2": 0.0019439283468281384, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 3.683487892150879, "p95_ms": 3.951864111423493, "p99_ms": 4.058688802719116, "mean_ms": 3.703865170478821, "n": 300}, {"repeat": 1, "p50_ms": 3.6848959922790527, "p95_ms": 3.8426959037780763, "p99_ms": 3.8449989628791807, "mean_ms": 3.708083732922872, "n": 300}, {"repeat": 2, "p50_ms": 3.687407970428467, "p95_ms": 3.7828848600387572, "p99_ms": 3.7916233229637144, "mean_ms": 3.7063327995936075, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_745464bfa49527f7_vllm_fa2.json", "configuration": "480x832_tp2_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp2_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp2_sp1_v1/captures/qkv_rank0_745464bfa49527f7.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 27, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "vllm_fa3", "status": "measured", "q_shape": [4680, 20, 128], "k_shape": [29640, 20, 128], "q_len": 4680, "k_len": 28080, "reference_error": {"max_abs": 0.0, "rmse": 0.0, "relative_l2": 0.0, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.0088576065358561, "rmse": 0.0008091472718205208, "relative_l2": 0.0019609076604501594, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 2.039712071418762, "p95_ms": 2.0893807768821717, "p99_ms": 2.0907840728759766, "mean_ms": 2.046982722679774, "n": 300}, {"repeat": 1, "p50_ms": 2.0610400438308716, "p95_ms": 2.2694782853126525, "p99_ms": 2.272219572067261, "mean_ms": 2.0823498646418255, "n": 300}, {"repeat": 2, "p50_ms": 2.0592960119247437, "p95_ms": 2.122023952007294, "p99_ms": 2.1269485449790952, "mean_ms": 2.069998505115509, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_745464bfa49527f7_vllm_fa3.json", "configuration": "480x832_tp2_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp2_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp2_sp1_v1/captures/qkv_rank0_745464bfa49527f7.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 27, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_math", "status": "measured", "q_shape": [4680, 20, 128], "k_shape": [29640, 20, 128], "q_len": 4680, "k_len": 28080, "reference_error": {"max_abs": 0.03125, "rmse": 0.0008194400055485837, "relative_l2": 0.0015794426209880647, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.007788278483080724, "rmse": 0.0006752488626088266, "relative_l2": 0.0016364149191540594, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 52.61552047729492, "p95_ms": 52.653671455383304, "p99_ms": 52.66102638244629, "mean_ms": 52.62278915405273, "n": 300}, {"repeat": 1, "p50_ms": 52.61463928222656, "p95_ms": 52.65769577026367, "p99_ms": 52.6642361831665, "mean_ms": 52.622242062886556, "n": 300}, {"repeat": 2, "p50_ms": 52.62583923339844, "p95_ms": 52.886840438842775, "p99_ms": 52.92323844909668, "mean_ms": 52.669296951293944, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_745464bfa49527f7_sdpa_math.json", "configuration": "480x832_tp2_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp2_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp2_sp1_v1/captures/qkv_rank0_745464bfa49527f7.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 27, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_flash", "status": "measured", "q_shape": [4680, 20, 128], "k_shape": [29640, 20, 128], "q_len": 4680, "k_len": 28080, "reference_error": {"max_abs": 0.03125, "rmse": 0.0004775709988836661, "relative_l2": 0.0009205017879005473, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.0088576065358561, "rmse": 0.0008020998230974734, "relative_l2": 0.0019438286976099822, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 3.9715681076049805, "p95_ms": 4.303052926063538, "p99_ms": 4.459296765327454, "mean_ms": 3.985702082316081, "n": 300}, {"repeat": 1, "p50_ms": 3.9727840423583984, "p95_ms": 4.145724964141846, "p99_ms": 4.183398127555847, "mean_ms": 3.9810509936014813, "n": 300}, {"repeat": 2, "p50_ms": 3.979103922843933, "p95_ms": 4.043201613426208, "p99_ms": 4.085409488677978, "mean_ms": 3.980985063711802, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_745464bfa49527f7_sdpa_flash.json", "configuration": "480x832_tp2_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp2_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp2_sp1_v1/captures/qkv_rank0_745464bfa49527f7.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 27, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_efficient", "status": "measured", "q_shape": [4680, 20, 128], "k_shape": [29640, 20, 128], "q_len": 4680, "k_len": 28080, "reference_error": {"max_abs": 0.03125, "rmse": 0.0006717455177098236, "relative_l2": 0.0012947665404965192, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.0088576065358561, "rmse": 0.0008029680037132037, "relative_l2": 0.001945932668147026, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 8.14844799041748, "p95_ms": 8.170011377334594, "p99_ms": 8.222510328292847, "mean_ms": 8.149075412750244, "n": 300}, {"repeat": 1, "p50_ms": 8.14731216430664, "p95_ms": 8.15991473197937, "p99_ms": 8.163901414871216, "mean_ms": 8.144778572718302, "n": 300}, {"repeat": 2, "p50_ms": 8.146512031555176, "p95_ms": 8.159924364089965, "p99_ms": 8.166024589538575, "mean_ms": 8.144559758504231, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_745464bfa49527f7_sdpa_efficient.json", "configuration": "480x832_tp2_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp2_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp2_sp1_v1/captures/qkv_rank0_745464bfa49527f7.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 27, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_cudnn", "status": "measured", "q_shape": [4680, 20, 128], "k_shape": [29640, 20, 128], "q_len": 4680, "k_len": 28080, "reference_error": {"max_abs": 0.03125, "rmse": 0.000671208792025958, "relative_l2": 0.0012937320200738696, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.0088576065358561, "rmse": 0.0008024639160963256, "relative_l2": 0.0019447110496556847, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 2.407920002937317, "p95_ms": 2.4540977001190187, "p99_ms": 2.4623779320716856, "mean_ms": 2.4093746185302733, "n": 300}, {"repeat": 1, "p50_ms": 2.414031982421875, "p95_ms": 2.750542414188385, "p99_ms": 2.8847047519683837, "mean_ms": 2.468691841761271, "n": 300}, {"repeat": 2, "p50_ms": 2.480623960494995, "p95_ms": 2.562872087955475, "p99_ms": 2.564393584728241, "mean_ms": 2.4723930668830874, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_745464bfa49527f7_sdpa_cudnn.json", "configuration": "480x832_tp2_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp2_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp2_sp1_v1/captures/qkv_rank0_745464bfa49527f7.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 27, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "native_dense", "status": "measured", "q_shape": [4680, 20, 128], "k_shape": [29640, 20, 128], "q_len": 4680, "k_len": 28080, "reference_error": {"max_abs": 0.03125, "rmse": 0.00044280664811300534, "relative_l2": 0.0008534946892400396, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.007788278483080724, "rmse": 0.000815752079321817, "relative_l2": 0.0019769139155240353, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 2.005919933319092, "p95_ms": 2.1274783492088316, "p99_ms": 2.144417052268982, "mean_ms": 2.0288838358720143, "n": 300}, {"repeat": 1, "p50_ms": 2.04966402053833, "p95_ms": 2.2613200068473818, "p99_ms": 2.2622083044052124, "mean_ms": 2.0705624544620513, "n": 300}, {"repeat": 2, "p50_ms": 2.0849599838256836, "p95_ms": 2.1738943696022033, "p99_ms": 2.184089193344116, "mean_ms": 2.0708233559131624, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_745464bfa49527f7_native_dense.json", "configuration": "480x832_tp2_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp2_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp2_sp1_v1/captures/qkv_rank0_7c5bdfbbb4c3e603.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 0, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "vllm_fa2", "status": "measured", "q_shape": [4680, 20, 128], "k_shape": [29640, 20, 128], "q_len": 4680, "k_len": 4680, "reference_error": {"max_abs": 0.03125, "rmse": 0.0004677193320152461, "relative_l2": 0.0009102273782300016, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.006950112781240136, "rmse": 0.0007533182743655885, "relative_l2": 0.001916956768692717, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.6298879981040955, "p95_ms": 0.631073585152626, "p99_ms": 0.6314893049001694, "mean_ms": 0.6243603189786275, "n": 300}, {"repeat": 1, "p50_ms": 0.6313119828701019, "p95_ms": 0.634177616238594, "p99_ms": 0.634880006313324, "mean_ms": 0.6294614412387212, "n": 300}, {"repeat": 2, "p50_ms": 0.6309760212898254, "p95_ms": 0.6377615839242935, "p99_ms": 0.6385923248529434, "mean_ms": 0.630646826227506, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_7c5bdfbbb4c3e603_vllm_fa2.json", "configuration": "480x832_tp2_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp2_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp2_sp1_v1/captures/qkv_rank0_7c5bdfbbb4c3e603.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 0, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "vllm_fa3", "status": "measured", "q_shape": [4680, 20, 128], "k_shape": [29640, 20, 128], "q_len": 4680, "k_len": 4680, "reference_error": {"max_abs": 0.0, "rmse": 0.0, "relative_l2": 0.0, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.006950112781240136, "rmse": 0.0007627035300087233, "relative_l2": 0.001940839276184215, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.3614879995584488, "p95_ms": 0.3733760118484497, "p99_ms": 0.3739872089028358, "mean_ms": 0.358654719889164, "n": 300}, {"repeat": 1, "p50_ms": 0.37299200892448425, "p95_ms": 0.37436801195144653, "p99_ms": 0.3747852727770805, "mean_ms": 0.36884202698866525, "n": 300}, {"repeat": 2, "p50_ms": 0.37991999089717865, "p95_ms": 0.3883199989795685, "p99_ms": 0.38889598846435547, "mean_ms": 0.37564544012149176, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_7c5bdfbbb4c3e603_vllm_fa3.json", "configuration": "480x832_tp2_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp2_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp2_sp1_v1/captures/qkv_rank0_7c5bdfbbb4c3e603.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 0, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_math", "status": "measured", "q_shape": [4680, 20, 128], "k_shape": [29640, 20, 128], "q_len": 4680, "k_len": 4680, "reference_error": {"max_abs": 0.03125, "rmse": 0.0007545501971907759, "relative_l2": 0.0014684281805771135, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.006950112781240136, "rmse": 0.0006602537827850868, "relative_l2": 0.0016801370695948418, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 9.039135932922363, "p95_ms": 9.048095703125, "p99_ms": 9.066498460769653, "mean_ms": 9.040435517628987, "n": 300}, {"repeat": 1, "p50_ms": 9.03932809829712, "p95_ms": 9.046787595748901, "p99_ms": 9.049611806869507, "mean_ms": 9.040523643493652, "n": 300}, {"repeat": 2, "p50_ms": 9.039072036743164, "p95_ms": 9.045667028427124, "p99_ms": 9.047431030273437, "mean_ms": 9.039760869344075, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_7c5bdfbbb4c3e603_sdpa_math.json", "configuration": "480x832_tp2_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp2_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp2_sp1_v1/captures/qkv_rank0_7c5bdfbbb4c3e603.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 0, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_flash", "status": "measured", "q_shape": [4680, 20, 128], "k_shape": [29640, 20, 128], "q_len": 4680, "k_len": 4680, "reference_error": {"max_abs": 0.03125, "rmse": 0.00046791042606234386, "relative_l2": 0.0009105992658591396, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.006950112781240136, "rmse": 0.0007533226374873345, "relative_l2": 0.0019169678714577073, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.6726080179214478, "p95_ms": 0.676639974117279, "p99_ms": 0.6771199703216553, "mean_ms": 0.6723560533920924, "n": 300}, {"repeat": 1, "p50_ms": 0.6747359931468964, "p95_ms": 0.6784383893013001, "p99_ms": 0.6792339211702347, "mean_ms": 0.6746559989452362, "n": 300}, {"repeat": 2, "p50_ms": 0.6754719913005829, "p95_ms": 0.6792672097682952, "p99_ms": 0.6800387346744537, "mean_ms": 0.6751666130622228, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_7c5bdfbbb4c3e603_sdpa_flash.json", "configuration": "480x832_tp2_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp2_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp2_sp1_v1/captures/qkv_rank0_7c5bdfbbb4c3e603.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 0, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_efficient", "status": "measured", "q_shape": [4680, 20, 128], "k_shape": [29640, 20, 128], "q_len": 4680, "k_len": 4680, "reference_error": {"max_abs": 0.03125, "rmse": 0.000665590970819228, "relative_l2": 0.0012953048609985595, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.006950112781240136, "rmse": 0.0007546643271149742, "relative_l2": 0.0019203820472459621, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 1.3925920128822327, "p95_ms": 1.398436814546585, "p99_ms": 1.4003861892223357, "mean_ms": 1.3854689085483551, "n": 300}, {"repeat": 1, "p50_ms": 1.3722879886627197, "p95_ms": 1.3758112013339996, "p99_ms": 1.3770924317836761, "mean_ms": 1.3726932255427042, "n": 300}, {"repeat": 2, "p50_ms": 1.3720799684524536, "p95_ms": 1.3758415520191192, "p99_ms": 1.3771843528747558, "mean_ms": 1.3725143480300903, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_7c5bdfbbb4c3e603_sdpa_efficient.json", "configuration": "480x832_tp2_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp2_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp2_sp1_v1/captures/qkv_rank0_7c5bdfbbb4c3e603.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 0, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_cudnn", "status": "measured", "q_shape": [4680, 20, 128], "k_shape": [29640, 20, 128], "q_len": 4680, "k_len": 4680, "reference_error": {"max_abs": 0.03125, "rmse": 0.0006651649442133812, "relative_l2": 0.0012944757717265267, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.006950112781240136, "rmse": 0.0007544534356409433, "relative_l2": 0.001919845395139728, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.4184640049934387, "p95_ms": 0.42416160404682157, "p99_ms": 0.4248335865139961, "mean_ms": 0.40856597423553465, "n": 300}, {"repeat": 1, "p50_ms": 0.4145440012216568, "p95_ms": 0.4171200096607208, "p99_ms": 0.41747424483299256, "mean_ms": 0.41079647998015084, "n": 300}, {"repeat": 2, "p50_ms": 0.41169600188732147, "p95_ms": 0.4305296137928963, "p99_ms": 0.4310406291484833, "mean_ms": 0.416175253589948, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_7c5bdfbbb4c3e603_sdpa_cudnn.json", "configuration": "480x832_tp2_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp2_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp2_sp1_v1/captures/qkv_rank0_7c5bdfbbb4c3e603.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 0, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "native_dense", "status": "measured", "q_shape": [4680, 20, 128], "k_shape": [29640, 20, 128], "q_len": 4680, "k_len": 4680, "reference_error": {"max_abs": 0.03125, "rmse": 0.0004351058489027083, "relative_l2": 0.0008467583633817848, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.006950112781240136, "rmse": 0.0007607068211752154, "relative_l2": 0.0019357582836691655, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.32655999064445496, "p95_ms": 0.35811520665884017, "p99_ms": 0.3589452788233757, "mean_ms": 0.3318495996793111, "n": 300}, {"repeat": 1, "p50_ms": 0.3519839942455292, "p95_ms": 0.3582448080182076, "p99_ms": 0.3588512054085731, "mean_ms": 0.3545878396431605, "n": 300}, {"repeat": 2, "p50_ms": 0.34968000650405884, "p95_ms": 0.35168481022119524, "p99_ms": 0.35238432109355927, "mean_ms": 0.34783050686120986, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_7c5bdfbbb4c3e603_native_dense.json", "configuration": "480x832_tp2_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp2_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp2_sp1_v1/captures/qkv_rank0_89e6d37b3d6f79ec.pt", "context": {"phase": "commit", "step": null, "start_frame": 27, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "vllm_fa2", "status": "measured", "q_shape": [4680, 20, 128], "k_shape": [512, 20, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.000244140625, "rmse": 1.6462628001877723e-06, "relative_l2": 0.0005708060987016288, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 6.742393421056356e-05, "rmse": 3.832209818314218e-06, "relative_l2": 0.0017802310561338699, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.08352000266313553, "p95_ms": 0.08428800106048584, "p99_ms": 0.08454432301223277, "mean_ms": 0.08373781390488148, "n": 300}, {"repeat": 1, "p50_ms": 0.08348800241947174, "p95_ms": 0.08441600203514099, "p99_ms": 0.08489727824926377, "mean_ms": 0.08392373392979303, "n": 300}, {"repeat": 2, "p50_ms": 0.08348800241947174, "p95_ms": 0.08435200154781342, "p99_ms": 0.08480351753532886, "mean_ms": 0.08390304061273733, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_89e6d37b3d6f79ec_vllm_fa2.json", "configuration": "480x832_tp2_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp2_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp2_sp1_v1/captures/qkv_rank0_89e6d37b3d6f79ec.pt", "context": {"phase": "commit", "step": null, "start_frame": 27, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "vllm_fa3", "status": "measured", "q_shape": [4680, 20, 128], "k_shape": [512, 20, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.000244140625, "rmse": 6.059949094549967e-07, "relative_l2": 0.00021011565714757058, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 6.742393421056356e-05, "rmse": 3.8849170393543405e-06, "relative_l2": 0.0018047158928799472, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.07212799787521362, "p95_ms": 0.07321599870920181, "p99_ms": 0.08634687840938526, "mean_ms": 0.07303274683654308, "n": 300}, {"repeat": 1, "p50_ms": 0.07212799787521362, "p95_ms": 0.07305599749088287, "p99_ms": 0.07344000041484833, "mean_ms": 0.07260896007219951, "n": 300}, {"repeat": 2, "p50_ms": 0.07209599763154984, "p95_ms": 0.07299359701573849, "p99_ms": 0.07353792116045951, "mean_ms": 0.07257429338991642, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_89e6d37b3d6f79ec_vllm_fa3.json", "configuration": "480x832_tp2_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp2_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp2_sp1_v1/captures/qkv_rank0_89e6d37b3d6f79ec.pt", "context": {"phase": "commit", "step": null, "start_frame": 27, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_math", "status": "measured", "q_shape": [4680, 20, 128], "k_shape": [512, 20, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.0009765625, "rmse": 4.517329437453498e-06, "relative_l2": 0.0015662864959645258, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 6.553299429325893e-05, "rmse": 3.594119650901572e-06, "relative_l2": 0.0016696276366231506, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 1.1241600513458252, "p95_ms": 1.1286159873008728, "p99_ms": 1.1320018899440765, "mean_ms": 1.1246182409922283, "n": 300}, {"repeat": 1, "p50_ms": 1.1239200234413147, "p95_ms": 1.127728009223938, "p99_ms": 1.128928906917572, "mean_ms": 1.1246451222896576, "n": 300}, {"repeat": 2, "p50_ms": 1.1243680119514465, "p95_ms": 1.127692747116089, "p99_ms": 1.1295039653778076, "mean_ms": 1.1249438937505085, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_89e6d37b3d6f79ec_sdpa_math.json", "configuration": "480x832_tp2_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp2_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp2_sp1_v1/captures/qkv_rank0_89e6d37b3d6f79ec.pt", "context": {"phase": "commit", "step": null, "start_frame": 27, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_flash", "status": "measured", "q_shape": [4680, 20, 128], "k_shape": [512, 20, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.00048828125, "rmse": 1.652267168468868e-06, "relative_l2": 0.0005728879838255039, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 6.742393421056356e-05, "rmse": 3.832209818314218e-06, "relative_l2": 0.0017802310561338699, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.088639996945858, "p95_ms": 0.08953759633004665, "p99_ms": 0.08998431973159313, "mean_ms": 0.08880074655016264, "n": 300}, {"repeat": 1, "p50_ms": 0.08857599645853043, "p95_ms": 0.08937760256230831, "p99_ms": 0.08982431851327419, "mean_ms": 0.0888683735827605, "n": 300}, {"repeat": 2, "p50_ms": 0.088639996945858, "p95_ms": 0.08959999680519104, "p99_ms": 0.09014432094991207, "mean_ms": 0.08890826667348543, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_89e6d37b3d6f79ec_sdpa_flash.json", "configuration": "480x832_tp2_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp2_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp2_sp1_v1/captures/qkv_rank0_89e6d37b3d6f79ec.pt", "context": {"phase": "commit", "step": null, "start_frame": 27, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_efficient", "status": "measured", "q_shape": [4680, 20, 128], "k_shape": [512, 20, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.00048828125, "rmse": 2.885866301979539e-06, "relative_l2": 0.0010006118616174464, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 6.742393421056356e-05, "rmse": 4.5033755183310445e-06, "relative_l2": 0.002092017226419135, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.16569599509239197, "p95_ms": 0.1664319932460785, "p99_ms": 0.16691264435648917, "mean_ms": 0.16580309296647708, "n": 300}, {"repeat": 1, "p50_ms": 0.16569599509239197, "p95_ms": 0.1664319932460785, "p99_ms": 0.16687999665737152, "mean_ms": 0.16607445274790128, "n": 300}, {"repeat": 2, "p50_ms": 0.16540800034999847, "p95_ms": 0.16604800522327423, "p99_ms": 0.1664329533278942, "mean_ms": 0.16561695943276086, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_89e6d37b3d6f79ec_sdpa_efficient.json", "configuration": "480x832_tp2_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp2_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp2_sp1_v1/captures/qkv_rank0_89e6d37b3d6f79ec.pt", "context": {"phase": "commit", "step": null, "start_frame": 27, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_cudnn", "status": "measured", "q_shape": [4680, 20, 128], "k_shape": [512, 20, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.00048828125, "rmse": 2.88526433764197e-06, "relative_l2": 0.0010004031434741182, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 6.742393421056356e-05, "rmse": 4.503302457158564e-06, "relative_l2": 0.002091983286271175, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.052352000027894974, "p95_ms": 0.053281599655747414, "p99_ms": 0.053544321581721295, "mean_ms": 0.052660266856352486, "n": 300}, {"repeat": 1, "p50_ms": 0.052319999784231186, "p95_ms": 0.053279999643564224, "p99_ms": 0.05360063835978508, "mean_ms": 0.05289429331819216, "n": 300}, {"repeat": 2, "p50_ms": 0.05225599929690361, "p95_ms": 0.053092798218131065, "p99_ms": 0.05337760038673878, "mean_ms": 0.05270026661455631, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_89e6d37b3d6f79ec_sdpa_cudnn.json", "configuration": "480x832_tp2_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp2_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp2_sp1_v1/captures/qkv_rank0_89e6d37b3d6f79ec.pt", "context": {"phase": "commit", "step": null, "start_frame": 27, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "native_dense", "status": "measured", "q_shape": [4680, 20, 128], "k_shape": [512, 20, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.0, "rmse": 0.0, "relative_l2": 0.0, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 6.742393421056356e-05, "rmse": 3.906451067226311e-06, "relative_l2": 0.0018147194018209578, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.11283200234174728, "p95_ms": 0.12004159763455391, "p99_ms": 0.14053215682506556, "mean_ms": 0.11410293338199456, "n": 300}, {"repeat": 1, "p50_ms": 0.11203200370073318, "p95_ms": 0.11818559840321541, "p99_ms": 0.1273155206441879, "mean_ms": 0.11363637345532576, "n": 300}, {"repeat": 2, "p50_ms": 0.11260800063610077, "p95_ms": 0.11755520068109036, "p99_ms": 0.1652220845222471, "mean_ms": 0.11533760026097298, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_89e6d37b3d6f79ec_native_dense.json", "configuration": "480x832_tp2_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp2_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp2_sp1_v1/captures/qkv_rank0_994f8af95ec0756f.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 27, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "vllm_fa2", "status": "measured", "q_shape": [4680, 20, 128], "k_shape": [512, 20, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.00048828125, "rmse": 1.9061654231743241e-06, "relative_l2": 0.0005755305491237531, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.00013185228070013155, "rmse": 5.9774578814411576e-06, "relative_l2": 0.0017269006831286404, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.08355200290679932, "p95_ms": 0.08444800227880478, "p99_ms": 0.08528096117079258, "mean_ms": 0.08445834711194039, "n": 300}, {"repeat": 1, "p50_ms": 0.08361600339412689, "p95_ms": 0.08448320254683495, "p99_ms": 0.08496287874877452, "mean_ms": 0.08525194687147936, "n": 300}, {"repeat": 2, "p50_ms": 0.0835840031504631, "p95_ms": 0.08457600325345993, "p99_ms": 0.08493023850023747, "mean_ms": 0.08401696061094602, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_994f8af95ec0756f_vllm_fa2.json", "configuration": "480x832_tp2_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp2_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp2_sp1_v1/captures/qkv_rank0_994f8af95ec0756f.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 27, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "vllm_fa3", "status": "measured", "q_shape": [4680, 20, 128], "k_shape": [512, 20, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.00048828125, "rmse": 7.293154807230148e-07, "relative_l2": 0.00022020299707565516, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.00013185228070013155, "rmse": 6.004525352024777e-06, "relative_l2": 0.0017347205347057705, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.07260800153017044, "p95_ms": 0.07350880093872547, "p99_ms": 0.0856329631060361, "mean_ms": 0.07306720013419787, "n": 300}, {"repeat": 1, "p50_ms": 0.07254400104284286, "p95_ms": 0.07331199944019318, "p99_ms": 0.07379360303282738, "mean_ms": 0.07317930658658346, "n": 300}, {"repeat": 2, "p50_ms": 0.07241600006818771, "p95_ms": 0.07331359945237637, "p99_ms": 0.07386272355914116, "mean_ms": 0.07290592022240162, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_994f8af95ec0756f_vllm_fa3.json", "configuration": "480x832_tp2_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp2_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp2_sp1_v1/captures/qkv_rank0_994f8af95ec0756f.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 27, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_math", "status": "measured", "q_shape": [4680, 20, 128], "k_shape": [512, 20, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.0009765625, "rmse": 5.111359837301689e-06, "relative_l2": 0.001543278300071393, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.00013185228070013155, "rmse": 5.295261799173165e-06, "relative_l2": 0.0015298127397482273, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 1.1247680187225342, "p95_ms": 1.1288975954055787, "p99_ms": 1.1301507198810576, "mean_ms": 1.1281936021645864, "n": 300}, {"repeat": 1, "p50_ms": 1.1236480474472046, "p95_ms": 1.1278560400009154, "p99_ms": 1.1312025165557862, "mean_ms": 1.124547732671102, "n": 300}, {"repeat": 2, "p50_ms": 1.1239200234413147, "p95_ms": 1.126913595199585, "p99_ms": 1.1280080211162566, "mean_ms": 1.1243336542447409, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_994f8af95ec0756f_sdpa_math.json", "configuration": "480x832_tp2_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp2_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp2_sp1_v1/captures/qkv_rank0_994f8af95ec0756f.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 27, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_flash", "status": "measured", "q_shape": [4680, 20, 128], "k_shape": [512, 20, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.00048828125, "rmse": 1.9061790221370605e-06, "relative_l2": 0.0005755346550730038, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.00013185228070013155, "rmse": 5.9774578814411576e-06, "relative_l2": 0.0017269006831286404, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.08947200328111649, "p95_ms": 0.09036800265312195, "p99_ms": 0.09091263934969902, "mean_ms": 0.08969504011174043, "n": 300}, {"repeat": 1, "p50_ms": 0.08950400352478027, "p95_ms": 0.09046400338411331, "p99_ms": 0.09120032154023647, "mean_ms": 0.08982794699569543, "n": 300}, {"repeat": 2, "p50_ms": 0.08928000181913376, "p95_ms": 0.09030560217797756, "p99_ms": 0.09081759862601757, "mean_ms": 0.08969312034547329, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_994f8af95ec0756f_sdpa_flash.json", "configuration": "480x832_tp2_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp2_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp2_sp1_v1/captures/qkv_rank0_994f8af95ec0756f.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 27, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_efficient", "status": "measured", "q_shape": [4680, 20, 128], "k_shape": [512, 20, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.00048828125, "rmse": 3.309557932188758e-06, "relative_l2": 0.0009992583387109701, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.00013185228070013155, "rmse": 6.454064602858091e-06, "relative_l2": 0.0018645934095557091, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.16704000532627106, "p95_ms": 0.16784000396728516, "p99_ms": 0.16825663968920707, "mean_ms": 0.16725045363108318, "n": 300}, {"repeat": 1, "p50_ms": 0.16697600483894348, "p95_ms": 0.16778240352869034, "p99_ms": 0.16822496682405472, "mean_ms": 0.16730506733059883, "n": 300}, {"repeat": 2, "p50_ms": 0.16582399606704712, "p95_ms": 0.16752160117030143, "p99_ms": 0.16800031781196595, "mean_ms": 0.16645653322339057, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_994f8af95ec0756f_sdpa_efficient.json", "configuration": "480x832_tp2_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp2_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp2_sp1_v1/captures/qkv_rank0_994f8af95ec0756f.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 27, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_cudnn", "status": "measured", "q_shape": [4680, 20, 128], "k_shape": [512, 20, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.00048828125, "rmse": 3.304497454572498e-06, "relative_l2": 0.0009977304233338964, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.00013185228070013155, "rmse": 6.454065552399304e-06, "relative_l2": 0.0018645936838802468, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.05209600180387497, "p95_ms": 0.053279999643564224, "p99_ms": 0.0717443224042654, "mean_ms": 0.05297493353486061, "n": 300}, {"repeat": 1, "p50_ms": 0.05193600058555603, "p95_ms": 0.052929600700736046, "p99_ms": 0.05350848134607076, "mean_ms": 0.05247850683828195, "n": 300}, {"repeat": 2, "p50_ms": 0.05193600058555603, "p95_ms": 0.05289600044488907, "p99_ms": 0.05366431884467602, "mean_ms": 0.05246133353561163, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_994f8af95ec0756f_sdpa_cudnn.json", "configuration": "480x832_tp2_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp2_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp2_sp1_v1/captures/qkv_rank0_994f8af95ec0756f.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 27, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "native_dense", "status": "measured", "q_shape": [4680, 20, 128], "k_shape": [512, 20, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.0, "rmse": 0.0, "relative_l2": 0.0, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.00013185228070013155, "rmse": 6.015429620007314e-06, "relative_l2": 0.0017378708016255132, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.11239999905228615, "p95_ms": 0.1201023980975151, "p99_ms": 0.14942272484302516, "mean_ms": 0.11373514677087466, "n": 300}, {"repeat": 1, "p50_ms": 0.11289600282907486, "p95_ms": 0.16036160588264464, "p99_ms": 0.1853558404743659, "mean_ms": 0.12717557332168022, "n": 300}, {"repeat": 2, "p50_ms": 0.11168000102043152, "p95_ms": 0.15680959448218346, "p99_ms": 0.1675945620238781, "mean_ms": 0.11776469337443511, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_994f8af95ec0756f_native_dense.json", "configuration": "480x832_tp2_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp2_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp2_sp1_v1/captures/qkv_rank0_b29cb74704615250.pt", "context": {"phase": "commit", "step": null, "start_frame": 0, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "vllm_fa2", "status": "measured", "q_shape": [4680, 20, 128], "k_shape": [512, 20, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.00048828125, "rmse": 1.9668137912064055e-06, "relative_l2": 0.000240273150956526, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.0002462965742430351, "rmse": 1.8319828422198023e-05, "relative_l2": 0.001666153736434623, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.0833280012011528, "p95_ms": 0.08416160009801388, "p99_ms": 0.08538464136421674, "mean_ms": 0.0836716802418232, "n": 300}, {"repeat": 1, "p50_ms": 0.08329600095748901, "p95_ms": 0.08422560058534145, "p99_ms": 0.08467263653874398, "mean_ms": 0.08380213355024656, "n": 300}, {"repeat": 2, "p50_ms": 0.08329600095748901, "p95_ms": 0.08422400057315826, "p99_ms": 0.08448416247963905, "mean_ms": 0.08377845366795858, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_b29cb74704615250_vllm_fa2.json", "configuration": "480x832_tp2_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp2_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp2_sp1_v1/captures/qkv_rank0_b29cb74704615250.pt", "context": {"phase": "commit", "step": null, "start_frame": 0, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "vllm_fa3", "status": "measured", "q_shape": [4680, 20, 128], "k_shape": [512, 20, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.00048828125, "rmse": 7.429210654257491e-07, "relative_l2": 9.075794876968666e-05, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.0002465302375597306, "rmse": 1.8325521411840758e-05, "relative_l2": 0.0016666715030722865, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.07235199958086014, "p95_ms": 0.07334719970822334, "p99_ms": 0.07382496327161789, "mean_ms": 0.0725381334622701, "n": 300}, {"repeat": 1, "p50_ms": 0.07228799909353256, "p95_ms": 0.07318559847772121, "p99_ms": 0.07392991662025451, "mean_ms": 0.07280309334397315, "n": 300}, {"repeat": 2, "p50_ms": 0.07235199958086014, "p95_ms": 0.07353920117020607, "p99_ms": 0.11702400252222991, "mean_ms": 0.07371541341145833, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_b29cb74704615250_vllm_fa3.json", "configuration": "480x832_tp2_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp2_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp2_sp1_v1/captures/qkv_rank0_b29cb74704615250.pt", "context": {"phase": "commit", "step": null, "start_frame": 0, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_math", "status": "measured", "q_shape": [4680, 20, 128], "k_shape": [512, 20, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.001953125, "rmse": 5.66533881506616e-06, "relative_l2": 0.000692098466269807, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.00024413624259587985, "rmse": 1.8135371052777925e-05, "relative_l2": 0.0016493776876535218, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 1.1239999532699585, "p95_ms": 1.1285487771034242, "p99_ms": 1.1309468936920166, "mean_ms": 1.1246233614285788, "n": 300}, {"repeat": 1, "p50_ms": 1.1239039897918701, "p95_ms": 1.1274575531482696, "p99_ms": 1.130218575000763, "mean_ms": 1.1246163233121236, "n": 300}, {"repeat": 2, "p50_ms": 1.124015986919403, "p95_ms": 1.1276495933532715, "p99_ms": 1.1299840211868286, "mean_ms": 1.1244130142529805, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_b29cb74704615250_sdpa_math.json", "configuration": "480x832_tp2_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp2_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp2_sp1_v1/captures/qkv_rank0_b29cb74704615250.pt", "context": {"phase": "commit", "step": null, "start_frame": 0, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_flash", "status": "measured", "q_shape": [4680, 20, 128], "k_shape": [512, 20, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.00048828125, "rmse": 1.967781654204519e-06, "relative_l2": 0.0002403913886327568, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.0002462965742430351, "rmse": 1.8319828422198023e-05, "relative_l2": 0.001666153736434623, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.08879999816417694, "p95_ms": 0.0896655973047018, "p99_ms": 0.09008096046745777, "mean_ms": 0.08918581279615562, "n": 300}, {"repeat": 1, "p50_ms": 0.08876799792051315, "p95_ms": 0.08973119780421257, "p99_ms": 0.09046528331935405, "mean_ms": 0.089139839609464, "n": 300}, {"repeat": 2, "p50_ms": 0.08892799913883209, "p95_ms": 0.08985759876668453, "p99_ms": 0.090304322168231, "mean_ms": 0.08923285298049449, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_b29cb74704615250_sdpa_flash.json", "configuration": "480x832_tp2_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp2_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp2_sp1_v1/captures/qkv_rank0_b29cb74704615250.pt", "context": {"phase": "commit", "step": null, "start_frame": 0, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_efficient", "status": "measured", "q_shape": [4680, 20, 128], "k_shape": [512, 20, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.0009765625, "rmse": 3.461132278287019e-06, "relative_l2": 0.0004228245511087574, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.0002465302375597306, "rmse": 1.846391370947778e-05, "relative_l2": 0.0016792580207234179, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.1656000018119812, "p95_ms": 0.1664000004529953, "p99_ms": 0.16697728484869004, "mean_ms": 0.16583231985569, "n": 300}, {"repeat": 1, "p50_ms": 0.16540800034999847, "p95_ms": 0.16608159840106965, "p99_ms": 0.16646784096956252, "mean_ms": 0.16570154642065366, "n": 300}, {"repeat": 2, "p50_ms": 0.16566400229930878, "p95_ms": 0.1663056068122387, "p99_ms": 0.16669279530644415, "mean_ms": 0.16851189300417901, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_b29cb74704615250_sdpa_efficient.json", "configuration": "480x832_tp2_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp2_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp2_sp1_v1/captures/qkv_rank0_b29cb74704615250.pt", "context": {"phase": "commit", "step": null, "start_frame": 0, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_cudnn", "status": "measured", "q_shape": [4680, 20, 128], "k_shape": [512, 20, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.0009765625, "rmse": 3.4567691697529983e-06, "relative_l2": 0.0004222915378463318, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.0002465302375597306, "rmse": 1.846391370947778e-05, "relative_l2": 0.0016792580207234179, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.05193600058555603, "p95_ms": 0.05289600044488907, "p99_ms": 0.053409280627965926, "mean_ms": 0.052263893658916157, "n": 300}, {"repeat": 1, "p50_ms": 0.05196800082921982, "p95_ms": 0.05299360118806362, "p99_ms": 0.053701439090073105, "mean_ms": 0.05251029352347056, "n": 300}, {"repeat": 2, "p50_ms": 0.0522879995405674, "p95_ms": 0.07104959711432457, "p99_ms": 0.07501247726380822, "mean_ms": 0.05488906676570574, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_b29cb74704615250_sdpa_cudnn.json", "configuration": "480x832_tp2_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp2_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp2_sp1_v1/captures/qkv_rank0_b29cb74704615250.pt", "context": {"phase": "commit", "step": null, "start_frame": 0, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "native_dense", "status": "measured", "q_shape": [4680, 20, 128], "k_shape": [512, 20, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.0, "rmse": 0.0, "relative_l2": 0.0, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.0002465302375597306, "rmse": 1.832763914830437e-05, "relative_l2": 0.0016668641071971892, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.112271998077631, "p95_ms": 0.14264479801058771, "p99_ms": 0.16139487951993942, "mean_ms": 0.11606922701001167, "n": 300}, {"repeat": 1, "p50_ms": 0.11168000102043152, "p95_ms": 0.11947680041193963, "p99_ms": 0.13619360014796242, "mean_ms": 0.11387829343477884, "n": 300}, {"repeat": 2, "p50_ms": 0.11283200234174728, "p95_ms": 0.12243200093507767, "p99_ms": 0.13797664657235134, "mean_ms": 0.11590752022961776, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_b29cb74704615250_native_dense.json", "configuration": "480x832_tp2_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp2_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp2_sp1_v1/captures/qkv_rank0_e8b549dc16c13a1e.pt", "context": {"phase": "commit", "step": null, "start_frame": 27, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "vllm_fa2", "status": "measured", "q_shape": [4680, 20, 128], "k_shape": [29640, 20, 128], "q_len": 4680, "k_len": 28080, "reference_error": {"max_abs": 0.0625, "rmse": 0.0004314840913423551, "relative_l2": 0.0005906502406311045, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.01046887368105942, "rmse": 0.0009489566155549423, "relative_l2": 0.0017095622012029148, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 3.6572959423065186, "p95_ms": 3.94692485332489, "p99_ms": 4.095808029174805, "mean_ms": 3.6950892758369447, "n": 300}, {"repeat": 1, "p50_ms": 3.683648109436035, "p95_ms": 3.7817280292510986, "p99_ms": 3.8461164927482603, "mean_ms": 3.6982298588752744, "n": 300}, {"repeat": 2, "p50_ms": 3.684688091278076, "p95_ms": 3.780790317058563, "p99_ms": 3.7859631061553953, "mean_ms": 3.6978577097256977, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_e8b549dc16c13a1e_vllm_fa2.json", "configuration": "480x832_tp2_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp2_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp2_sp1_v1/captures/qkv_rank0_e8b549dc16c13a1e.pt", "context": {"phase": "commit", "step": null, "start_frame": 27, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "vllm_fa3", "status": "measured", "q_shape": [4680, 20, 128], "k_shape": [29640, 20, 128], "q_len": 4680, "k_len": 28080, "reference_error": {"max_abs": 0.0, "rmse": 0.0, "relative_l2": 0.0, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.01046887368105942, "rmse": 0.0009491314136562684, "relative_l2": 0.001709877103087754, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 2.036463975906372, "p95_ms": 2.06642564535141, "p99_ms": 2.0733772206306456, "mean_ms": 2.030082986752192, "n": 300}, {"repeat": 1, "p50_ms": 2.0546239614486694, "p95_ms": 2.2580177068710325, "p99_ms": 2.327105691432953, "mean_ms": 2.0718965367476145, "n": 300}, {"repeat": 2, "p50_ms": 2.059936046600342, "p95_ms": 2.120358431339264, "p99_ms": 2.126915316581726, "mean_ms": 2.063999364376068, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_e8b549dc16c13a1e_vllm_fa3.json", "configuration": "480x832_tp2_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp2_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp2_sp1_v1/captures/qkv_rank0_e8b549dc16c13a1e.pt", "context": {"phase": "commit", "step": null, "start_frame": 27, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_math", "status": "measured", "q_shape": [4680, 20, 128], "k_shape": [29640, 20, 128], "q_len": 4680, "k_len": 28080, "reference_error": {"max_abs": 0.0625, "rmse": 0.0008228335476695758, "relative_l2": 0.0011263609543944112, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.01046887368105942, "rmse": 0.0009309913174739112, "relative_l2": 0.0016771973975551596, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 52.61899185180664, "p95_ms": 52.659303855896, "p99_ms": 52.663046875, "mean_ms": 52.63025455474853, "n": 300}, {"repeat": 1, "p50_ms": 52.621055603027344, "p95_ms": 52.65802268981933, "p99_ms": 52.66653114318848, "mean_ms": 52.627656834920245, "n": 300}, {"repeat": 2, "p50_ms": 52.62286376953125, "p95_ms": 52.77194023132324, "p99_ms": 52.92196430206299, "mean_ms": 52.641733957926434, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_e8b549dc16c13a1e_sdpa_math.json", "configuration": "480x832_tp2_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp2_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp2_sp1_v1/captures/qkv_rank0_e8b549dc16c13a1e.pt", "context": {"phase": "commit", "step": null, "start_frame": 27, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_flash", "status": "measured", "q_shape": [4680, 20, 128], "k_shape": [29640, 20, 128], "q_len": 4680, "k_len": 28080, "reference_error": {"max_abs": 0.0625, "rmse": 0.0004325234887188258, "relative_l2": 0.0005920730516288723, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.01046887368105942, "rmse": 0.0009489907279307189, "relative_l2": 0.001709623655253885, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 3.933824062347412, "p95_ms": 4.243583965301514, "p99_ms": 4.42038553237915, "mean_ms": 3.979002030690511, "n": 300}, {"repeat": 1, "p50_ms": 3.940559983253479, "p95_ms": 4.086313772201538, "p99_ms": 4.094573583602905, "mean_ms": 3.967817811965942, "n": 300}, {"repeat": 2, "p50_ms": 4.003744125366211, "p95_ms": 4.043992185592652, "p99_ms": 4.087394733428955, "mean_ms": 3.9728336056073505, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_e8b549dc16c13a1e_sdpa_flash.json", "configuration": "480x832_tp2_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp2_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp2_sp1_v1/captures/qkv_rank0_e8b549dc16c13a1e.pt", "context": {"phase": "commit", "step": null, "start_frame": 27, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_efficient", "status": "measured", "q_shape": [4680, 20, 128], "k_shape": [29640, 20, 128], "q_len": 4680, "k_len": 28080, "reference_error": {"max_abs": 0.0625, "rmse": 0.0007037804127120671, "relative_l2": 0.0009633914168808103, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.01046887368105942, "rmse": 0.000939906305892011, "relative_l2": 0.0016932579075657594, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 8.075088024139404, "p95_ms": 8.146521520614623, "p99_ms": 8.212602796554565, "mean_ms": 8.079882599512736, "n": 300}, {"repeat": 1, "p50_ms": 8.074527740478516, "p95_ms": 8.086688041687012, "p99_ms": 8.089779348373414, "mean_ms": 8.073246822357177, "n": 300}, {"repeat": 2, "p50_ms": 8.074016094207764, "p95_ms": 8.086916971206666, "p99_ms": 8.090726890563964, "mean_ms": 8.073033504486084, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_e8b549dc16c13a1e_sdpa_efficient.json", "configuration": "480x832_tp2_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp2_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp2_sp1_v1/captures/qkv_rank0_e8b549dc16c13a1e.pt", "context": {"phase": "commit", "step": null, "start_frame": 27, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_cudnn", "status": "measured", "q_shape": [4680, 20, 128], "k_shape": [29640, 20, 128], "q_len": 4680, "k_len": 28080, "reference_error": {"max_abs": 0.0625, "rmse": 0.000702367475309236, "relative_l2": 0.0009614572741540587, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.01046887368105942, "rmse": 0.0009398300104325858, "relative_l2": 0.0016931204599402107, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 2.4074079990386963, "p95_ms": 2.431905674934387, "p99_ms": 2.4344045615196226, "mean_ms": 2.4072511982917786, "n": 300}, {"repeat": 1, "p50_ms": 2.427839994430542, "p95_ms": 2.733510458469391, "p99_ms": 2.8382533502578737, "mean_ms": 2.4726325329144796, "n": 300}, {"repeat": 2, "p50_ms": 2.4532960653305054, "p95_ms": 2.556996774673462, "p99_ms": 2.5717811369895935, "mean_ms": 2.4560812822977702, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_e8b549dc16c13a1e_sdpa_cudnn.json", "configuration": "480x832_tp2_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp2_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp2_sp1_v1/captures/qkv_rank0_e8b549dc16c13a1e.pt", "context": {"phase": "commit", "step": null, "start_frame": 27, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "native_dense", "status": "measured", "q_shape": [4680, 20, 128], "k_shape": [29640, 20, 128], "q_len": 4680, "k_len": 28080, "reference_error": {"max_abs": 0.03125, "rmse": 0.0003780992332199474, "relative_l2": 0.0005175727392150928, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.01046887368105942, "rmse": 0.0009499572957040254, "relative_l2": 0.0017113649442686407, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 2.0147520303726196, "p95_ms": 2.0750784873962402, "p99_ms": 2.0903392839431763, "mean_ms": 2.003637860616048, "n": 300}, {"repeat": 1, "p50_ms": 2.0409120321273804, "p95_ms": 2.316777539253235, "p99_ms": 2.3191680908203125, "mean_ms": 2.0766244264443716, "n": 300}, {"repeat": 2, "p50_ms": 2.0623199939727783, "p95_ms": 2.1705920696258545, "p99_ms": 2.214616320133209, "mean_ms": 2.057416640520096, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_e8b549dc16c13a1e_native_dense.json", "configuration": "480x832_tp2_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp2_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp2_sp1_v1/captures/qkv_rank0_fb84ccff3c768374.pt", "context": {"phase": "commit", "step": null, "start_frame": 0, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "vllm_fa2", "status": "measured", "q_shape": [4680, 20, 128], "k_shape": [29640, 20, 128], "q_len": 4680, "k_len": 4680, "reference_error": {"max_abs": 0.03125, "rmse": 0.00041943200478150507, "relative_l2": 0.0005994209879942685, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.008621929282112362, "rmse": 0.0008376367598461469, "relative_l2": 0.0016792810532232504, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.6206559836864471, "p95_ms": 0.622497609257698, "p99_ms": 0.623328930735588, "mean_ms": 0.6202292253573736, "n": 300}, {"repeat": 1, "p50_ms": 0.6219519972801208, "p95_ms": 0.6231679916381836, "p99_ms": 0.6236803156137466, "mean_ms": 0.622438079714775, "n": 300}, {"repeat": 2, "p50_ms": 0.6224319934844971, "p95_ms": 0.6266255885362625, "p99_ms": 0.6273286187648773, "mean_ms": 0.6234763717651367, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_fb84ccff3c768374_vllm_fa2.json", "configuration": "480x832_tp2_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp2_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp2_sp1_v1/captures/qkv_rank0_fb84ccff3c768374.pt", "context": {"phase": "commit", "step": null, "start_frame": 0, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "vllm_fa3", "status": "measured", "q_shape": [4680, 20, 128], "k_shape": [29640, 20, 128], "q_len": 4680, "k_len": 4680, "reference_error": {"max_abs": 0.0, "rmse": 0.0, "relative_l2": 0.0, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.008621929282112362, "rmse": 0.0008352826699057865, "relative_l2": 0.0016745616105913906, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.36720000207424164, "p95_ms": 0.36982559263706205, "p99_ms": 0.3703052684664726, "mean_ms": 0.3551859188079834, "n": 300}, {"repeat": 1, "p50_ms": 0.36403200030326843, "p95_ms": 0.369667212665081, "p99_ms": 0.3702729704976082, "mean_ms": 0.3634714670976003, "n": 300}, {"repeat": 2, "p50_ms": 0.3683200031518936, "p95_ms": 0.37529919743537904, "p99_ms": 0.3757455852627754, "mean_ms": 0.36695114622513453, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_fb84ccff3c768374_vllm_fa3.json", "configuration": "480x832_tp2_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp2_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp2_sp1_v1/captures/qkv_rank0_fb84ccff3c768374.pt", "context": {"phase": "commit", "step": null, "start_frame": 0, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_math", "status": "measured", "q_shape": [4680, 20, 128], "k_shape": [29640, 20, 128], "q_len": 4680, "k_len": 4680, "reference_error": {"max_abs": 0.03125, "rmse": 0.0007293702619823779, "relative_l2": 0.0010423616654596161, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.008621929282112362, "rmse": 0.0008212104263772126, "relative_l2": 0.0016463497972294529, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 9.039360046386719, "p95_ms": 9.052737665176391, "p99_ms": 9.093263721466064, "mean_ms": 9.042668476104737, "n": 300}, {"repeat": 1, "p50_ms": 9.039535999298096, "p95_ms": 9.045473718643189, "p99_ms": 9.047206602096558, "mean_ms": 9.040022004445394, "n": 300}, {"repeat": 2, "p50_ms": 9.039216041564941, "p95_ms": 9.045379018783569, "p99_ms": 9.046671981811523, "mean_ms": 9.040310824712117, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_fb84ccff3c768374_sdpa_math.json", "configuration": "480x832_tp2_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp2_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp2_sp1_v1/captures/qkv_rank0_fb84ccff3c768374.pt", "context": {"phase": "commit", "step": null, "start_frame": 0, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_flash", "status": "measured", "q_shape": [4680, 20, 128], "k_shape": [29640, 20, 128], "q_len": 4680, "k_len": 4680, "reference_error": {"max_abs": 0.03125, "rmse": 0.0004196803269144641, "relative_l2": 0.0005997758714952465, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.008621929282112362, "rmse": 0.0008376367598461469, "relative_l2": 0.0016792810532232504, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.6677759885787964, "p95_ms": 0.6719359755516052, "p99_ms": 0.672449579834938, "mean_ms": 0.6681937068700791, "n": 300}, {"repeat": 1, "p50_ms": 0.6705600023269653, "p95_ms": 0.67261121571064, "p99_ms": 0.6731216216087341, "mean_ms": 0.6709852802753449, "n": 300}, {"repeat": 2, "p50_ms": 0.6709439754486084, "p95_ms": 0.6727039813995361, "p99_ms": 0.6730239987373352, "mean_ms": 0.6711466672023138, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_fb84ccff3c768374_sdpa_flash.json", "configuration": "480x832_tp2_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp2_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp2_sp1_v1/captures/qkv_rank0_fb84ccff3c768374.pt", "context": {"phase": "commit", "step": null, "start_frame": 0, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_efficient", "status": "measured", "q_shape": [4680, 20, 128], "k_shape": [29640, 20, 128], "q_len": 4680, "k_len": 4680, "reference_error": {"max_abs": 0.03125, "rmse": 0.0005859263189171446, "relative_l2": 0.0008373622636644502, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.008764376048810973, "rmse": 0.0008320545195320431, "relative_l2": 0.001668089865296236, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 1.383504033088684, "p95_ms": 1.3890976309776306, "p99_ms": 1.3984079492092132, "mean_ms": 1.3839672549565634, "n": 300}, {"repeat": 1, "p50_ms": 1.38372802734375, "p95_ms": 1.3884815514087676, "p99_ms": 1.3897033619880677, "mean_ms": 1.3846387187639873, "n": 300}, {"repeat": 2, "p50_ms": 1.3836640119552612, "p95_ms": 1.3875903725624084, "p99_ms": 1.389730830192566, "mean_ms": 1.3844395720958709, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_fb84ccff3c768374_sdpa_efficient.json", "configuration": "480x832_tp2_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp2_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp2_sp1_v1/captures/qkv_rank0_fb84ccff3c768374.pt", "context": {"phase": "commit", "step": null, "start_frame": 0, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_cudnn", "status": "measured", "q_shape": [4680, 20, 128], "k_shape": [29640, 20, 128], "q_len": 4680, "k_len": 4680, "reference_error": {"max_abs": 0.03125, "rmse": 0.0005839901205190188, "relative_l2": 0.0008345951930939483, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.008764376048810973, "rmse": 0.0008320674169439367, "relative_l2": 0.0016681157218255416, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.41254401206970215, "p95_ms": 0.4159375935792923, "p99_ms": 0.41686656773090364, "mean_ms": 0.40427861322959263, "n": 300}, {"repeat": 1, "p50_ms": 0.4115680009126663, "p95_ms": 0.41654399037361145, "p99_ms": 0.4173766514658928, "mean_ms": 0.41002517292896906, "n": 300}, {"repeat": 2, "p50_ms": 0.41595199704170227, "p95_ms": 0.4204480051994324, "p99_ms": 0.42112190812826156, "mean_ms": 0.4129162672162056, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_fb84ccff3c768374_sdpa_cudnn.json", "configuration": "480x832_tp2_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp2_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp2_sp1_v1/captures/qkv_rank0_fb84ccff3c768374.pt", "context": {"phase": "commit", "step": null, "start_frame": 0, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "native_dense", "status": "measured", "q_shape": [4680, 20, 128], "k_shape": [29640, 20, 128], "q_len": 4680, "k_len": 4680, "reference_error": {"max_abs": 0.03125, "rmse": 0.00038093749297515796, "relative_l2": 0.0005444074982360468, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.008621929282112362, "rmse": 0.0008364944270648977, "relative_l2": 0.0016769909223598675, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.3136799931526184, "p95_ms": 0.3510127976536751, "p99_ms": 0.3520003181695938, "mean_ms": 0.32983370711406074, "n": 300}, {"repeat": 1, "p50_ms": 0.33851200342178345, "p95_ms": 0.3473935931921005, "p99_ms": 0.34819328635931013, "mean_ms": 0.33750986665487287, "n": 300}, {"repeat": 2, "p50_ms": 0.3445119857788086, "p95_ms": 0.3470736056566238, "p99_ms": 0.34809633284807207, "mean_ms": 0.34295861373345055, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_fb84ccff3c768374_native_dense.json", "configuration": "480x832_tp2_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp2_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp2_sp1_v1/captures/qkv_rank0_fe264be86166ca15.pt", "context": {"phase": "commit", "step": null, "start_frame": 9, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "vllm_fa2", "status": "measured", "q_shape": [4680, 20, 128], "k_shape": [29640, 20, 128], "q_len": 4680, "k_len": 18720, "reference_error": {"max_abs": 0.03125, "rmse": 0.0004336853328677868, "relative_l2": 0.0005866649436142257, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.015577997885023898, "rmse": 0.000975910621199614, "relative_l2": 0.0017008123037051729, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 2.464416027069092, "p95_ms": 2.4687758922576903, "p99_ms": 2.473824985027313, "mean_ms": 2.457382934888204, "n": 300}, {"repeat": 1, "p50_ms": 2.4645919799804688, "p95_ms": 2.7156511425971983, "p99_ms": 2.783544406890869, "mean_ms": 2.490158612728119, "n": 300}, {"repeat": 2, "p50_ms": 2.4647679328918457, "p95_ms": 2.5725456953048704, "p99_ms": 2.5746611285209657, "mean_ms": 2.4789523243904115, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_fe264be86166ca15_vllm_fa2.json", "configuration": "480x832_tp2_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp2_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp2_sp1_v1/captures/qkv_rank0_fe264be86166ca15.pt", "context": {"phase": "commit", "step": null, "start_frame": 9, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "vllm_fa3", "status": "measured", "q_shape": [4680, 20, 128], "k_shape": [29640, 20, 128], "q_len": 4680, "k_len": 18720, "reference_error": {"max_abs": 0.0, "rmse": 0.0, "relative_l2": 0.0, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.015672002114976102, "rmse": 0.000975161952523733, "relative_l2": 0.0016995075275630996, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 1.3545439839363098, "p95_ms": 1.4048048555850983, "p99_ms": 1.4058572936058045, "mean_ms": 1.346677120923996, "n": 300}, {"repeat": 1, "p50_ms": 1.3565599918365479, "p95_ms": 1.368673610687256, "p99_ms": 1.5592966520786284, "mean_ms": 1.3579630931218465, "n": 300}, {"repeat": 2, "p50_ms": 1.3941280245780945, "p95_ms": 1.5580207467079163, "p99_ms": 1.5595264065265655, "mean_ms": 1.4100257047017415, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_fe264be86166ca15_vllm_fa3.json", "configuration": "480x832_tp2_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp2_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp2_sp1_v1/captures/qkv_rank0_fe264be86166ca15.pt", "context": {"phase": "commit", "step": null, "start_frame": 9, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_math", "status": "measured", "q_shape": [4680, 20, 128], "k_shape": [29640, 20, 128], "q_len": 4680, "k_len": 18720, "reference_error": {"max_abs": 0.0625, "rmse": 0.0008278077619502244, "relative_l2": 0.0011198114328113576, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.015577997885023898, "rmse": 0.0009582482838116969, "relative_l2": 0.0016700304676548234, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 35.10014533996582, "p95_ms": 35.13671455383301, "p99_ms": 35.14656352996826, "mean_ms": 35.10652441660563, "n": 300}, {"repeat": 1, "p50_ms": 35.10152053833008, "p95_ms": 35.12822513580322, "p99_ms": 35.13677249908447, "mean_ms": 35.10401830037435, "n": 300}, {"repeat": 2, "p50_ms": 35.10014343261719, "p95_ms": 35.12651767730713, "p99_ms": 35.13109188079834, "mean_ms": 35.1035151163737, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_fe264be86166ca15_sdpa_math.json", "configuration": "480x832_tp2_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp2_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp2_sp1_v1/captures/qkv_rank0_fe264be86166ca15.pt", "context": {"phase": "commit", "step": null, "start_frame": 9, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_flash", "status": "measured", "q_shape": [4680, 20, 128], "k_shape": [29640, 20, 128], "q_len": 4680, "k_len": 18720, "reference_error": {"max_abs": 0.03125, "rmse": 0.0004345181315028092, "relative_l2": 0.0005877915064173223, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.015577997885023898, "rmse": 0.0009759106202033164, "relative_l2": 0.0017008123019688303, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 2.61572802066803, "p95_ms": 2.6804161071777344, "p99_ms": 2.6835577869415284, "mean_ms": 2.627089601357778, "n": 300}, {"repeat": 1, "p50_ms": 2.6374560594558716, "p95_ms": 2.8325472831726075, "p99_ms": 2.8753932404518125, "mean_ms": 2.6665964841842653, "n": 300}, {"repeat": 2, "p50_ms": 2.649456024169922, "p95_ms": 2.7290896654129027, "p99_ms": 2.7353302001953126, "mean_ms": 2.6619254398345946, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_fe264be86166ca15_sdpa_flash.json", "configuration": "480x832_tp2_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp2_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp2_sp1_v1/captures/qkv_rank0_fe264be86166ca15.pt", "context": {"phase": "commit", "step": null, "start_frame": 9, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_efficient", "status": "measured", "q_shape": [4680, 20, 128], "k_shape": [29640, 20, 128], "q_len": 4680, "k_len": 18720, "reference_error": {"max_abs": 0.0625, "rmse": 0.0007143151591941385, "relative_l2": 0.0009662850708376887, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.015672002114976102, "rmse": 0.0009673590580544767, "relative_l2": 0.001685908680876186, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 5.396895885467529, "p95_ms": 5.481937527656555, "p99_ms": 5.490148897171021, "mean_ms": 5.402018888791402, "n": 300}, {"repeat": 1, "p50_ms": 5.396239995956421, "p95_ms": 5.405417513847351, "p99_ms": 5.408349709510803, "mean_ms": 5.395238200823466, "n": 300}, {"repeat": 2, "p50_ms": 5.3961920738220215, "p95_ms": 5.405099153518677, "p99_ms": 5.408113117218018, "mean_ms": 5.395137267112732, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_fe264be86166ca15_sdpa_efficient.json", "configuration": "480x832_tp2_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp2_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp2_sp1_v1/captures/qkv_rank0_fe264be86166ca15.pt", "context": {"phase": "commit", "step": null, "start_frame": 9, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_cudnn", "status": "measured", "q_shape": [4680, 20, 128], "k_shape": [29640, 20, 128], "q_len": 4680, "k_len": 18720, "reference_error": {"max_abs": 0.0625, "rmse": 0.0007139687634544011, "relative_l2": 0.0009658164863094145, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.015672002114976102, "rmse": 0.0009673291098942182, "relative_l2": 0.0016858564873675447, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 1.6454880237579346, "p95_ms": 1.65976802110672, "p99_ms": 1.6623750984668733, "mean_ms": 1.6339396262168884, "n": 300}, {"repeat": 1, "p50_ms": 1.6460800170898438, "p95_ms": 1.7838143944740295, "p99_ms": 1.8012482750415801, "mean_ms": 1.6578198357423146, "n": 300}, {"repeat": 2, "p50_ms": 1.6463199853897095, "p95_ms": 1.6798399686813354, "p99_ms": 1.6838102090358733, "mean_ms": 1.6434947204589845, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_fe264be86166ca15_sdpa_cudnn.json", "configuration": "480x832_tp2_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp2_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp2_sp1_v1/captures/qkv_rank0_fe264be86166ca15.pt", "context": {"phase": "commit", "step": null, "start_frame": 9, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "native_dense", "status": "measured", "q_shape": [4680, 20, 128], "k_shape": [29640, 20, 128], "q_len": 4680, "k_len": 18720, "reference_error": {"max_abs": 0.03125, "rmse": 0.00038632072143635, "relative_l2": 0.0005225927811756478, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.015672002114976102, "rmse": 0.0009753463475053411, "relative_l2": 0.0016998288902439107, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 1.3968000411987305, "p95_ms": 1.4331519603729248, "p99_ms": 1.440964208841324, "mean_ms": 1.373753816684087, "n": 300}, {"repeat": 1, "p50_ms": 1.3646720051765442, "p95_ms": 1.380587214231491, "p99_ms": 1.3915596365928649, "mean_ms": 1.3623326992988587, "n": 300}, {"repeat": 2, "p50_ms": 1.3584319949150085, "p95_ms": 1.41212797164917, "p99_ms": 1.4174115097522735, "mean_ms": 1.3687993613878886, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_fe264be86166ca15_native_dense.json", "configuration": "480x832_tp2_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp2_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp2_sp2_v1/captures/qkv_rank0_13f5e4136c72875f.pt", "context": {"phase": "commit", "step": null, "start_frame": 27, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "vllm_fa2", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [29640, 10, 128], "q_len": 4680, "k_len": 28080, "reference_error": {"max_abs": 0.0625, "rmse": 0.00042389160377658794, "relative_l2": 0.0005353895925681103, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.01046887368105942, "rmse": 0.0009489566155549423, "relative_l2": 0.0017095622012029148, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 1.9061279892921448, "p95_ms": 1.9190527677536011, "p99_ms": 1.922144694328308, "mean_ms": 1.9062417085965475, "n": 300}, {"repeat": 1, "p50_ms": 1.9183040261268616, "p95_ms": 1.9845616340637207, "p99_ms": 1.9940898680686951, "mean_ms": 1.9288920537630716, "n": 300}, {"repeat": 2, "p50_ms": 1.9149439930915833, "p95_ms": 1.9472432076930999, "p99_ms": 1.950752660036087, "mean_ms": 1.9206749860445658, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_13f5e4136c72875f_vllm_fa2.json", "configuration": "480x832_tp2_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp2_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp2_sp2_v1/captures/qkv_rank0_13f5e4136c72875f.pt", "context": {"phase": "commit", "step": null, "start_frame": 27, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "vllm_fa3", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [29640, 10, 128], "q_len": 4680, "k_len": 28080, "reference_error": {"max_abs": 0.0, "rmse": 0.0, "relative_l2": 0.0, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.01046887368105942, "rmse": 0.0009491314136562684, "relative_l2": 0.001709877103087754, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 1.0349760055541992, "p95_ms": 1.0392432153224944, "p99_ms": 1.0433612751960755, "mean_ms": 1.0207828275362651, "n": 300}, {"repeat": 1, "p50_ms": 1.0383679866790771, "p95_ms": 1.0476304054260255, "p99_ms": 1.049062772989273, "mean_ms": 1.0288913071155548, "n": 300}, {"repeat": 2, "p50_ms": 1.0402560234069824, "p95_ms": 1.15174560546875, "p99_ms": 1.1614464151859283, "mean_ms": 1.0636275200049083, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_13f5e4136c72875f_vllm_fa3.json", "configuration": "480x832_tp2_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp2_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp2_sp2_v1/captures/qkv_rank0_13f5e4136c72875f.pt", "context": {"phase": "commit", "step": null, "start_frame": 27, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_math", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [29640, 10, 128], "q_len": 4680, "k_len": 28080, "reference_error": {"max_abs": 0.0625, "rmse": 0.0007930776952824828, "relative_l2": 0.0010016842522220203, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.01046887368105942, "rmse": 0.0009309913174739112, "relative_l2": 0.0016771973975551596, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 28.389711380004883, "p95_ms": 28.89846134185791, "p99_ms": 29.57629194259643, "mean_ms": 28.423042227427164, "n": 300}, {"repeat": 1, "p50_ms": 28.355584144592285, "p95_ms": 28.66195116043091, "p99_ms": 28.729172191619874, "mean_ms": 28.370911261240643, "n": 300}, {"repeat": 2, "p50_ms": 28.376863479614258, "p95_ms": 28.690626430511475, "p99_ms": 28.75428949356079, "mean_ms": 28.39397227605184, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_13f5e4136c72875f_sdpa_math.json", "configuration": "480x832_tp2_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp2_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp2_sp2_v1/captures/qkv_rank0_13f5e4136c72875f.pt", "context": {"phase": "commit", "step": null, "start_frame": 27, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_flash", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [29640, 10, 128], "q_len": 4680, "k_len": 28080, "reference_error": {"max_abs": 0.0625, "rmse": 0.00042463452751166614, "relative_l2": 0.00053632793065334, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.01046887368105942, "rmse": 0.0009489907279307189, "relative_l2": 0.001709623655253885, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 2.0659520626068115, "p95_ms": 2.0728384971618654, "p99_ms": 2.074915406703949, "mean_ms": 2.0656604838371275, "n": 300}, {"repeat": 1, "p50_ms": 2.0835040807724, "p95_ms": 2.1701470971107484, "p99_ms": 2.185216269493103, "mean_ms": 2.0906123717625937, "n": 300}, {"repeat": 2, "p50_ms": 2.066943883895874, "p95_ms": 2.1160334944725037, "p99_ms": 2.1205468368530274, "mean_ms": 2.0733719420433045, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_13f5e4136c72875f_sdpa_flash.json", "configuration": "480x832_tp2_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp2_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp2_sp2_v1/captures/qkv_rank0_13f5e4136c72875f.pt", "context": {"phase": "commit", "step": null, "start_frame": 27, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_efficient", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [29640, 10, 128], "q_len": 4680, "k_len": 28080, "reference_error": {"max_abs": 0.0625, "rmse": 0.0006728342426313618, "relative_l2": 0.0008498126592243998, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.01046887368105942, "rmse": 0.000939906305892011, "relative_l2": 0.0016932579075657594, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 4.100528001785278, "p95_ms": 4.131371259689331, "p99_ms": 4.14356686592102, "mean_ms": 4.104144965807597, "n": 300}, {"repeat": 1, "p50_ms": 4.097423791885376, "p95_ms": 4.122318410873413, "p99_ms": 4.1303637504577635, "mean_ms": 4.100277547836304, "n": 300}, {"repeat": 2, "p50_ms": 4.098927974700928, "p95_ms": 4.12713942527771, "p99_ms": 4.136194725036621, "mean_ms": 4.103120443026225, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_13f5e4136c72875f_sdpa_efficient.json", "configuration": "480x832_tp2_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp2_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp2_sp2_v1/captures/qkv_rank0_13f5e4136c72875f.pt", "context": {"phase": "commit", "step": null, "start_frame": 27, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_cudnn", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [29640, 10, 128], "q_len": 4680, "k_len": 28080, "reference_error": {"max_abs": 0.0625, "rmse": 0.0006727400293804261, "relative_l2": 0.0008496936646663958, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.01046887368105942, "rmse": 0.0009398300104325858, "relative_l2": 0.0016931204599402107, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 1.221824049949646, "p95_ms": 1.2319088280200958, "p99_ms": 1.2339286184310914, "mean_ms": 1.2062185593446095, "n": 300}, {"repeat": 1, "p50_ms": 1.2163200378417969, "p95_ms": 1.232259249687195, "p99_ms": 1.234880931377411, "mean_ms": 1.2140055497487385, "n": 300}, {"repeat": 2, "p50_ms": 1.2450560331344604, "p95_ms": 1.4165471613407135, "p99_ms": 1.448415994644165, "mean_ms": 1.2781971196333568, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_13f5e4136c72875f_sdpa_cudnn.json", "configuration": "480x832_tp2_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp2_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp2_sp2_v1/captures/qkv_rank0_13f5e4136c72875f.pt", "context": {"phase": "commit", "step": null, "start_frame": 27, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "native_dense", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [29640, 10, 128], "q_len": 4680, "k_len": 28080, "reference_error": {"max_abs": 0.03125, "rmse": 0.00037624307013925173, "relative_l2": 0.0004752078650149346, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.01046887368105942, "rmse": 0.0009499572957040254, "relative_l2": 0.0017113649442686407, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 1.0217600464820862, "p95_ms": 1.0259903848171235, "p99_ms": 1.0284483063220977, "mean_ms": 1.0091190411647162, "n": 300}, {"repeat": 1, "p50_ms": 1.0245440006256104, "p95_ms": 1.0284224331378937, "p99_ms": 1.0297612833976746, "mean_ms": 1.0151226669549942, "n": 300}, {"repeat": 2, "p50_ms": 1.01636803150177, "p95_ms": 1.1449311792850494, "p99_ms": 1.1476191616058349, "mean_ms": 1.03413045446078, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_13f5e4136c72875f_native_dense.json", "configuration": "480x832_tp2_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp2_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp2_sp2_v1/captures/qkv_rank0_2419ddf76b22ab88.pt", "context": {"phase": "commit", "step": null, "start_frame": 9, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "vllm_fa2", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [512, 10, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.0001220703125, "rmse": 1.1608422729453961e-06, "relative_l2": 0.0007154581945605979, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.0001101750380461744, "rmse": 4.55905245454031e-06, "relative_l2": 0.0017790650100901059, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.04870399832725525, "p95_ms": 0.05382400006055832, "p99_ms": 0.06090432174503802, "mean_ms": 0.04930090653399626, "n": 300}, {"repeat": 1, "p50_ms": 0.048576001077890396, "p95_ms": 0.05395840089768172, "p99_ms": 0.07525951676070658, "mean_ms": 0.05177450669308503, "n": 300}, {"repeat": 2, "p50_ms": 0.04864000156521797, "p95_ms": 0.05392160080373287, "p99_ms": 0.0572070393338799, "mean_ms": 0.04949109334498644, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_2419ddf76b22ab88_vllm_fa2.json", "configuration": "480x832_tp2_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp2_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp2_sp2_v1/captures/qkv_rank0_2419ddf76b22ab88.pt", "context": {"phase": "commit", "step": null, "start_frame": 9, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "vllm_fa3", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [512, 10, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.0001220703125, "rmse": 4.4541355872466756e-07, "relative_l2": 0.00027452030993787847, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.0001101750380461744, "rmse": 4.606789203010599e-06, "relative_l2": 0.0017976931745487944, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.05923200026154518, "p95_ms": 0.06747839972376823, "p99_ms": 0.07744512103497982, "mean_ms": 0.060343040078878404, "n": 300}, {"repeat": 1, "p50_ms": 0.0594559982419014, "p95_ms": 0.06746239960193634, "p99_ms": 0.08373152360320084, "mean_ms": 0.060798399994770685, "n": 300}, {"repeat": 2, "p50_ms": 0.059328000992536545, "p95_ms": 0.06444799900054932, "p99_ms": 0.07013695724308486, "mean_ms": 0.06024469332148631, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_2419ddf76b22ab88_vllm_fa3.json", "configuration": "480x832_tp2_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp2_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp2_sp2_v1/captures/qkv_rank0_2419ddf76b22ab88.pt", "context": {"phase": "commit", "step": null, "start_frame": 9, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_math", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [512, 10, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.000244140625, "rmse": 2.3820388541869664e-06, "relative_l2": 0.0014681143663605768, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.0001101750380461744, "rmse": 4.345600527834753e-06, "relative_l2": 0.0016957703215721024, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.6177279949188232, "p95_ms": 0.6198128163814545, "p99_ms": 0.6209926450252533, "mean_ms": 0.6181908277670543, "n": 300}, {"repeat": 1, "p50_ms": 0.6180160045623779, "p95_ms": 0.6208639740943909, "p99_ms": 0.6219225496053695, "mean_ms": 0.6187925322850545, "n": 300}, {"repeat": 2, "p50_ms": 0.6180160045623779, "p95_ms": 0.6203216224908829, "p99_ms": 0.6216066926717758, "mean_ms": 0.618996798992157, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_2419ddf76b22ab88_sdpa_math.json", "configuration": "480x832_tp2_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp2_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp2_sp2_v1/captures/qkv_rank0_2419ddf76b22ab88.pt", "context": {"phase": "commit", "step": null, "start_frame": 9, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_flash", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [512, 10, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.0001220703125, "rmse": 1.160652165824931e-06, "relative_l2": 0.0007153410264488302, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.0001101750380461744, "rmse": 4.55905245454031e-06, "relative_l2": 0.0017790650100901059, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.0490880012512207, "p95_ms": 0.049375999718904495, "p99_ms": 0.05135455969721076, "mean_ms": 0.04919349351276954, "n": 300}, {"repeat": 1, "p50_ms": 0.04899200052022934, "p95_ms": 0.04931199923157692, "p99_ms": 0.04954720091074704, "mean_ms": 0.04930048003792763, "n": 300}, {"repeat": 2, "p50_ms": 0.048815999180078506, "p95_ms": 0.04915200173854828, "p99_ms": 0.04934047922492025, "mean_ms": 0.04916629309455554, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_2419ddf76b22ab88_sdpa_flash.json", "configuration": "480x832_tp2_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp2_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp2_sp2_v1/captures/qkv_rank0_2419ddf76b22ab88.pt", "context": {"phase": "commit", "step": null, "start_frame": 9, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_efficient", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [512, 10, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.000244140625, "rmse": 2.051716090308326e-06, "relative_l2": 0.0012645275968443054, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.0001101750380461744, "rmse": 5.1918242283224114e-06, "relative_l2": 0.002025989592189896, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.08646400272846222, "p95_ms": 0.08694399893283844, "p99_ms": 0.08726400136947632, "mean_ms": 0.0866014943520228, "n": 300}, {"repeat": 1, "p50_ms": 0.08675199747085571, "p95_ms": 0.0870399996638298, "p99_ms": 0.08716832064092159, "mean_ms": 0.08697119901577631, "n": 300}, {"repeat": 2, "p50_ms": 0.08671999722719193, "p95_ms": 0.08710400015115738, "p99_ms": 0.08723200112581253, "mean_ms": 0.08697087946037452, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_2419ddf76b22ab88_sdpa_efficient.json", "configuration": "480x832_tp2_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp2_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp2_sp2_v1/captures/qkv_rank0_2419ddf76b22ab88.pt", "context": {"phase": "commit", "step": null, "start_frame": 9, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_cudnn", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [512, 10, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.000244140625, "rmse": 2.0515569933390577e-06, "relative_l2": 0.0012644295411195566, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.0001101750380461744, "rmse": 5.192060864126505e-06, "relative_l2": 0.002026081933851545, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.046911999583244324, "p95_ms": 0.053435200639069085, "p99_ms": 0.07288192316889758, "mean_ms": 0.047981760017573837, "n": 300}, {"repeat": 1, "p50_ms": 0.04694399982690811, "p95_ms": 0.05286400020122528, "p99_ms": 0.06466079995036117, "mean_ms": 0.04819274687518676, "n": 300}, {"repeat": 2, "p50_ms": 0.047520000487565994, "p95_ms": 0.0598768012598157, "p99_ms": 0.068936323300004, "mean_ms": 0.05019882656633854, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_2419ddf76b22ab88_sdpa_cudnn.json", "configuration": "480x832_tp2_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp2_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp2_sp2_v1/captures/qkv_rank0_2419ddf76b22ab88.pt", "context": {"phase": "commit", "step": null, "start_frame": 9, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "native_dense", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [512, 10, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.0, "rmse": 0.0, "relative_l2": 0.0, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.0001101750380461744, "rmse": 4.618265562926498e-06, "relative_l2": 0.0018021715591633961, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.10734399780631065, "p95_ms": 0.11863200180232525, "p99_ms": 0.12946335479617116, "mean_ms": 0.10923946663737297, "n": 300}, {"repeat": 1, "p50_ms": 0.10750399902462959, "p95_ms": 0.12372160181403163, "p99_ms": 0.1707635232806205, "mean_ms": 0.11150090684493383, "n": 300}, {"repeat": 2, "p50_ms": 0.10708800330758095, "p95_ms": 0.11780320256948472, "p99_ms": 0.13881536573171607, "mean_ms": 0.10965600006282329, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_2419ddf76b22ab88_native_dense.json", "configuration": "480x832_tp2_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp2_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp2_sp2_v1/captures/qkv_rank0_5d233cbdc47812ff.pt", "context": {"phase": "commit", "step": null, "start_frame": 0, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "vllm_fa2", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [29640, 10, 128], "q_len": 4680, "k_len": 4680, "reference_error": {"max_abs": 0.03125, "rmse": 0.0004145917324844293, "relative_l2": 0.0005495130869674437, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.008621929282112362, "rmse": 0.0008376367598461469, "relative_l2": 0.0016792810532232504, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.3356800079345703, "p95_ms": 0.3370240032672882, "p99_ms": 0.33766400814056396, "mean_ms": 0.33583989322185515, "n": 300}, {"repeat": 1, "p50_ms": 0.3314880132675171, "p95_ms": 0.3327695935964584, "p99_ms": 0.33331232488155366, "mean_ms": 0.3318628261486689, "n": 300}, {"repeat": 2, "p50_ms": 0.3317759931087494, "p95_ms": 0.3333440124988556, "p99_ms": 0.33366560012102126, "mean_ms": 0.33220864047606785, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_5d233cbdc47812ff_vllm_fa2.json", "configuration": "480x832_tp2_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp2_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp2_sp2_v1/captures/qkv_rank0_5d233cbdc47812ff.pt", "context": {"phase": "commit", "step": null, "start_frame": 0, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "vllm_fa3", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [29640, 10, 128], "q_len": 4680, "k_len": 4680, "reference_error": {"max_abs": 0.0, "rmse": 0.0, "relative_l2": 0.0, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.008621929282112362, "rmse": 0.0008352826699057865, "relative_l2": 0.0016745616105913906, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.17372800409793854, "p95_ms": 0.17443199455738068, "p99_ms": 0.17504032656550408, "mean_ms": 0.17392618656158448, "n": 300}, {"repeat": 1, "p50_ms": 0.1810080036520958, "p95_ms": 0.18947360441088676, "p99_ms": 0.18995296090841293, "mean_ms": 0.18364480043450992, "n": 300}, {"repeat": 2, "p50_ms": 0.18432000279426575, "p95_ms": 0.18742560371756553, "p99_ms": 0.18796832069754602, "mean_ms": 0.18349130680163703, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_5d233cbdc47812ff_vllm_fa3.json", "configuration": "480x832_tp2_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp2_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp2_sp2_v1/captures/qkv_rank0_5d233cbdc47812ff.pt", "context": {"phase": "commit", "step": null, "start_frame": 0, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_math", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [29640, 10, 128], "q_len": 4680, "k_len": 4680, "reference_error": {"max_abs": 0.03125, "rmse": 0.0007400938236523078, "relative_l2": 0.0009809439258318846, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.008621929282112362, "rmse": 0.0008212104263772126, "relative_l2": 0.0016463497972294529, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 4.813695907592773, "p95_ms": 4.887220692634583, "p99_ms": 4.937508935928345, "mean_ms": 4.825016749699911, "n": 300}, {"repeat": 1, "p50_ms": 4.821167945861816, "p95_ms": 4.899361777305603, "p99_ms": 4.920202326774597, "mean_ms": 4.829630513191223, "n": 300}, {"repeat": 2, "p50_ms": 4.826240062713623, "p95_ms": 4.8557072401046755, "p99_ms": 4.861812567710876, "mean_ms": 4.825392525990804, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_5d233cbdc47812ff_sdpa_math.json", "configuration": "480x832_tp2_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp2_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp2_sp2_v1/captures/qkv_rank0_5d233cbdc47812ff.pt", "context": {"phase": "commit", "step": null, "start_frame": 0, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_flash", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [29640, 10, 128], "q_len": 4680, "k_len": 4680, "reference_error": {"max_abs": 0.03125, "rmse": 0.0004149869009238966, "relative_l2": 0.000550036855803215, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.008621929282112362, "rmse": 0.0008376367598461469, "relative_l2": 0.0016792810532232504, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.36294400691986084, "p95_ms": 0.365119993686676, "p99_ms": 0.3656022301316261, "mean_ms": 0.36201632032791775, "n": 300}, {"repeat": 1, "p50_ms": 0.35715198516845703, "p95_ms": 0.35926398634910583, "p99_ms": 0.36016031324863435, "mean_ms": 0.3575894385576248, "n": 300}, {"repeat": 2, "p50_ms": 0.3569920063018799, "p95_ms": 0.35827518701553346, "p99_ms": 0.3587846302986145, "mean_ms": 0.3573873057961464, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_5d233cbdc47812ff_sdpa_flash.json", "configuration": "480x832_tp2_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp2_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp2_sp2_v1/captures/qkv_rank0_5d233cbdc47812ff.pt", "context": {"phase": "commit", "step": null, "start_frame": 0, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_efficient", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [29640, 10, 128], "q_len": 4680, "k_len": 4680, "reference_error": {"max_abs": 0.03125, "rmse": 0.0005951016181840873, "relative_l2": 0.0007887666387074916, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.008764376048810973, "rmse": 0.0008320545195320431, "relative_l2": 0.001668089865296236, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.699999988079071, "p95_ms": 0.7042928069829941, "p99_ms": 0.7058879733085632, "mean_ms": 0.7006786131858825, "n": 300}, {"repeat": 1, "p50_ms": 0.6986240148544312, "p95_ms": 0.7009935885667801, "p99_ms": 0.7025654476881027, "mean_ms": 0.699138559103012, "n": 300}, {"repeat": 2, "p50_ms": 0.6984959840774536, "p95_ms": 0.7004815846681595, "p99_ms": 0.7017663943767548, "mean_ms": 0.6991767460107803, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_5d233cbdc47812ff_sdpa_efficient.json", "configuration": "480x832_tp2_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp2_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp2_sp2_v1/captures/qkv_rank0_5d233cbdc47812ff.pt", "context": {"phase": "commit", "step": null, "start_frame": 0, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_cudnn", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [29640, 10, 128], "q_len": 4680, "k_len": 4680, "reference_error": {"max_abs": 0.03125, "rmse": 0.0005930465143925504, "relative_l2": 0.0007860427386872028, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.008764376048810973, "rmse": 0.0008320674169439367, "relative_l2": 0.0016681157218255416, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.19731199741363525, "p95_ms": 0.19798559546470643, "p99_ms": 0.19836863800883292, "mean_ms": 0.19753119960427284, "n": 300}, {"repeat": 1, "p50_ms": 0.20414400100708008, "p95_ms": 0.21315520629286766, "p99_ms": 0.21420864686369895, "mean_ms": 0.20514837329586347, "n": 300}, {"repeat": 2, "p50_ms": 0.20630399882793427, "p95_ms": 0.2133471928536892, "p99_ms": 0.21420928671956063, "mean_ms": 0.20704074665904046, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_5d233cbdc47812ff_sdpa_cudnn.json", "configuration": "480x832_tp2_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp2_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp2_sp2_v1/captures/qkv_rank0_5d233cbdc47812ff.pt", "context": {"phase": "commit", "step": null, "start_frame": 0, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "native_dense", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [29640, 10, 128], "q_len": 4680, "k_len": 4680, "reference_error": {"max_abs": 0.03125, "rmse": 0.0003709067414037167, "relative_l2": 0.0004916116084235869, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.008621929282112362, "rmse": 0.0008364944270648977, "relative_l2": 0.0016769909223598675, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.16435199975967407, "p95_ms": 0.17455999553203583, "p99_ms": 0.17510464698076247, "mean_ms": 0.1665677869319916, "n": 300}, {"repeat": 1, "p50_ms": 0.1703680008649826, "p95_ms": 0.17466240376234055, "p99_ms": 0.17612992003560066, "mean_ms": 0.17079295992851257, "n": 300}, {"repeat": 2, "p50_ms": 0.16662399470806122, "p95_ms": 0.17424159348011017, "p99_ms": 0.17513855993747712, "mean_ms": 0.1694986661275228, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_5d233cbdc47812ff_native_dense.json", "configuration": "480x832_tp2_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp2_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp2_sp2_v1/captures/qkv_rank0_6b4ac7dbc60c0c5a.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 0, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "vllm_fa2", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [29640, 10, 128], "q_len": 4680, "k_len": 4680, "reference_error": {"max_abs": 0.03125, "rmse": 0.0005265060329404634, "relative_l2": 0.0009065080334159581, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.006950112781240136, "rmse": 0.0007533182743655885, "relative_l2": 0.001916956768692717, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.33315199613571167, "p95_ms": 0.3352335900068283, "p99_ms": 0.3358105689287186, "mean_ms": 0.3334204800923665, "n": 300}, {"repeat": 1, "p50_ms": 0.3330880105495453, "p95_ms": 0.3343679904937744, "p99_ms": 0.3348847943544388, "mean_ms": 0.33369749277830124, "n": 300}, {"repeat": 2, "p50_ms": 0.33209601044654846, "p95_ms": 0.33395200967788696, "p99_ms": 0.3346595200896263, "mean_ms": 0.3328717866539955, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_6b4ac7dbc60c0c5a_vllm_fa2.json", "configuration": "480x832_tp2_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp2_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp2_sp2_v1/captures/qkv_rank0_6b4ac7dbc60c0c5a.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 0, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "vllm_fa3", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [29640, 10, 128], "q_len": 4680, "k_len": 4680, "reference_error": {"max_abs": 0.0, "rmse": 0.0, "relative_l2": 0.0, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.006950112781240136, "rmse": 0.0007627035300087233, "relative_l2": 0.001940839276184215, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.17628799378871918, "p95_ms": 0.19331520423293114, "p99_ms": 0.19398592188954353, "mean_ms": 0.18099989359577498, "n": 300}, {"repeat": 1, "p50_ms": 0.1927039995789528, "p95_ms": 0.19385600090026855, "p99_ms": 0.19433631718158723, "mean_ms": 0.19186656003197033, "n": 300}, {"repeat": 2, "p50_ms": 0.18598400056362152, "p95_ms": 0.19219200313091278, "p99_ms": 0.19286432072520257, "mean_ms": 0.18862197359402974, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_6b4ac7dbc60c0c5a_vllm_fa3.json", "configuration": "480x832_tp2_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp2_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp2_sp2_v1/captures/qkv_rank0_6b4ac7dbc60c0c5a.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 0, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_math", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [29640, 10, 128], "q_len": 4680, "k_len": 4680, "reference_error": {"max_abs": 0.03125, "rmse": 0.0008676671731278176, "relative_l2": 0.0014938998103761964, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.006950112781240136, "rmse": 0.0006602537827850868, "relative_l2": 0.0016801370695948418, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 4.829535961151123, "p95_ms": 4.933225393295288, "p99_ms": 5.003757562637329, "mean_ms": 4.840736513137817, "n": 300}, {"repeat": 1, "p50_ms": 4.834144115447998, "p95_ms": 4.924372839927674, "p99_ms": 4.997337026596069, "mean_ms": 4.842174617449443, "n": 300}, {"repeat": 2, "p50_ms": 4.831023931503296, "p95_ms": 4.935087943077088, "p99_ms": 4.944208254814148, "mean_ms": 4.8497800588607785, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_6b4ac7dbc60c0c5a_sdpa_math.json", "configuration": "480x832_tp2_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp2_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp2_sp2_v1/captures/qkv_rank0_6b4ac7dbc60c0c5a.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 0, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_flash", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [29640, 10, 128], "q_len": 4680, "k_len": 4680, "reference_error": {"max_abs": 0.03125, "rmse": 0.0005269216644494476, "relative_l2": 0.0009072236440229834, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.006950112781240136, "rmse": 0.0007533226374873345, "relative_l2": 0.0019169678714577073, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.3605920076370239, "p95_ms": 0.361952006816864, "p99_ms": 0.36377855837345124, "mean_ms": 0.3605937059720357, "n": 300}, {"repeat": 1, "p50_ms": 0.35708799958229065, "p95_ms": 0.36051198840141296, "p99_ms": 0.3617340925335884, "mean_ms": 0.3576919466257095, "n": 300}, {"repeat": 2, "p50_ms": 0.35760000348091125, "p95_ms": 0.35897760689258573, "p99_ms": 0.35984320551156995, "mean_ms": 0.3578877859314283, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_6b4ac7dbc60c0c5a_sdpa_flash.json", "configuration": "480x832_tp2_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp2_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp2_sp2_v1/captures/qkv_rank0_6b4ac7dbc60c0c5a.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 0, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_efficient", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [29640, 10, 128], "q_len": 4680, "k_len": 4680, "reference_error": {"max_abs": 0.03125, "rmse": 0.0007439896163023662, "relative_l2": 0.001280958852816061, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.006950112781240136, "rmse": 0.0007546643271149742, "relative_l2": 0.0019203820472459621, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.7022079825401306, "p95_ms": 0.7077247947454453, "p99_ms": 0.7100486594438553, "mean_ms": 0.7027954159180323, "n": 300}, {"repeat": 1, "p50_ms": 0.7002559900283813, "p95_ms": 0.7054463863372803, "p99_ms": 0.7079753565788269, "mean_ms": 0.7013243732849757, "n": 300}, {"repeat": 2, "p50_ms": 0.699535995721817, "p95_ms": 0.702239990234375, "p99_ms": 0.7044166171550751, "mean_ms": 0.7001642676194509, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_6b4ac7dbc60c0c5a_sdpa_efficient.json", "configuration": "480x832_tp2_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp2_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp2_sp2_v1/captures/qkv_rank0_6b4ac7dbc60c0c5a.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 0, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_cudnn", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [29640, 10, 128], "q_len": 4680, "k_len": 4680, "reference_error": {"max_abs": 0.03125, "rmse": 0.0007433016303524855, "relative_l2": 0.0012797743178792796, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.006950112781240136, "rmse": 0.0007544534356409433, "relative_l2": 0.001919845395139728, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.19883199781179428, "p95_ms": 0.2126079946756363, "p99_ms": 0.21360031485557557, "mean_ms": 0.20048981308937072, "n": 300}, {"repeat": 1, "p50_ms": 0.2046400010585785, "p95_ms": 0.21140000000596046, "p99_ms": 0.21306367814540864, "mean_ms": 0.20537973349293073, "n": 300}, {"repeat": 2, "p50_ms": 0.2035519927740097, "p95_ms": 0.21273599565029144, "p99_ms": 0.21360479488968848, "mean_ms": 0.2058133333424727, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_6b4ac7dbc60c0c5a_sdpa_cudnn.json", "configuration": "480x832_tp2_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp2_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp2_sp2_v1/captures/qkv_rank0_6b4ac7dbc60c0c5a.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 0, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "native_dense", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [29640, 10, 128], "q_len": 4680, "k_len": 4680, "reference_error": {"max_abs": 0.03125, "rmse": 0.00048797092880426484, "relative_l2": 0.0008401604907811815, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.006950112781240136, "rmse": 0.0007607068211752154, "relative_l2": 0.0019357582836691655, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.17212800681591034, "p95_ms": 0.173089599609375, "p99_ms": 0.17357183545827864, "mean_ms": 0.1715512532989184, "n": 300}, {"repeat": 1, "p50_ms": 0.17296000570058823, "p95_ms": 0.18121760115027427, "p99_ms": 0.18185632631182672, "mean_ms": 0.17523114636540413, "n": 300}, {"repeat": 2, "p50_ms": 0.17158400267362595, "p95_ms": 0.18124959468841553, "p99_ms": 0.18211647346615792, "mean_ms": 0.1749981865286827, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_6b4ac7dbc60c0c5a_native_dense.json", "configuration": "480x832_tp2_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp2_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp2_sp2_v1/captures/qkv_rank0_6bef1f209a2b6a6e.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 0, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "vllm_fa2", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [512, 10, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.0009765625, "rmse": 1.9486393292294105e-06, "relative_l2": 0.0002828888309403964, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.0005221612900013495, "rmse": 1.9680953528593928e-05, "relative_l2": 0.0019826252521852918, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.048928000032901764, "p95_ms": 0.05458559822291136, "p99_ms": 0.06004703868180509, "mean_ms": 0.04965290660659472, "n": 300}, {"repeat": 1, "p50_ms": 0.04864000156521797, "p95_ms": 0.053249599412083626, "p99_ms": 0.05866079937666651, "mean_ms": 0.049427306540310384, "n": 300}, {"repeat": 2, "p50_ms": 0.04854400083422661, "p95_ms": 0.05369599908590317, "p99_ms": 0.05987232118844983, "mean_ms": 0.04931914668530226, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_6bef1f209a2b6a6e_vllm_fa2.json", "configuration": "480x832_tp2_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp2_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp2_sp2_v1/captures/qkv_rank0_6bef1f209a2b6a6e.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 0, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "vllm_fa3", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [512, 10, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.0009765625, "rmse": 8.063106499517188e-07, "relative_l2": 0.00011705412783074311, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.0005221612900013495, "rmse": 1.9692264907610615e-05, "relative_l2": 0.0019837647409627506, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.058720000088214874, "p95_ms": 0.06643199920654297, "p99_ms": 0.07419967785477627, "mean_ms": 0.05992597340295712, "n": 300}, {"repeat": 1, "p50_ms": 0.05878400057554245, "p95_ms": 0.06683520190417767, "p99_ms": 0.095324802622199, "mean_ms": 0.06457226677487293, "n": 300}, {"repeat": 2, "p50_ms": 0.05820799991488457, "p95_ms": 0.06320319697260857, "p99_ms": 0.0717939224094152, "mean_ms": 0.05917247988283634, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_6bef1f209a2b6a6e_vllm_fa3.json", "configuration": "480x832_tp2_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp2_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp2_sp2_v1/captures/qkv_rank0_6bef1f209a2b6a6e.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 0, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_math", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [512, 10, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.001953125, "rmse": 1.491327178070884e-05, "relative_l2": 0.0021649968551181054, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.0004544012099986505, "rmse": 1.6037217321311435e-05, "relative_l2": 0.0016155615625950317, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.6180160045623779, "p95_ms": 0.6203855782747268, "p99_ms": 0.6217923384904861, "mean_ms": 0.6181523219744365, "n": 300}, {"repeat": 1, "p50_ms": 0.6182079911231995, "p95_ms": 0.6203232198953629, "p99_ms": 0.6214172655344009, "mean_ms": 0.6186985596021016, "n": 300}, {"repeat": 2, "p50_ms": 0.6182720065116882, "p95_ms": 0.6207728087902069, "p99_ms": 0.6219852846860886, "mean_ms": 0.6191541343927384, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_6bef1f209a2b6a6e_sdpa_math.json", "configuration": "480x832_tp2_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp2_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp2_sp2_v1/captures/qkv_rank0_6bef1f209a2b6a6e.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 0, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_flash", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [512, 10, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.0009765625, "rmse": 1.952067439787916e-06, "relative_l2": 0.00028338649829919574, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.0005221612900013495, "rmse": 1.9680953528593928e-05, "relative_l2": 0.0019826252521852918, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.048895999789237976, "p95_ms": 0.04918399825692177, "p99_ms": 0.04955712068825956, "mean_ms": 0.049009066559374335, "n": 300}, {"repeat": 1, "p50_ms": 0.04886399954557419, "p95_ms": 0.04918399825692177, "p99_ms": 0.0502860790863633, "mean_ms": 0.049191999907294906, "n": 300}, {"repeat": 2, "p50_ms": 0.048895999789237976, "p95_ms": 0.04918559826910496, "p99_ms": 0.04955072093755006, "mean_ms": 0.049189759890238445, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_6bef1f209a2b6a6e_sdpa_flash.json", "configuration": "480x832_tp2_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp2_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp2_sp2_v1/captures/qkv_rank0_6bef1f209a2b6a6e.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 0, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_efficient", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [512, 10, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.0009765625, "rmse": 3.3775602631232615e-06, "relative_l2": 0.0004903288463563544, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.0005221612900013495, "rmse": 1.97311176669295e-05, "relative_l2": 0.0019876787008037094, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.08675199747085571, "p95_ms": 0.08710400015115738, "p99_ms": 0.08720160089433193, "mean_ms": 0.08688170582056046, "n": 300}, {"repeat": 1, "p50_ms": 0.08671999722719193, "p95_ms": 0.08707199990749359, "p99_ms": 0.0873280018568039, "mean_ms": 0.08696031908194224, "n": 300}, {"repeat": 2, "p50_ms": 0.08671999722719193, "p95_ms": 0.0872960016131401, "p99_ms": 0.08828927755355785, "mean_ms": 0.08879775923987229, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_6bef1f209a2b6a6e_sdpa_efficient.json", "configuration": "480x832_tp2_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp2_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp2_sp2_v1/captures/qkv_rank0_6bef1f209a2b6a6e.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 0, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_cudnn", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [512, 10, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.0009765625, "rmse": 3.368176479236469e-06, "relative_l2": 0.0004889665790482314, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.0005221612900013495, "rmse": 1.9731270952981146e-05, "relative_l2": 0.001987694142575693, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.04639999940991402, "p95_ms": 0.0510784013196826, "p99_ms": 0.055557438321411606, "mean_ms": 0.04711274666090806, "n": 300}, {"repeat": 1, "p50_ms": 0.04596799984574318, "p95_ms": 0.05100960098206997, "p99_ms": 0.05697920158505438, "mean_ms": 0.046807146519422534, "n": 300}, {"repeat": 2, "p50_ms": 0.04614400118589401, "p95_ms": 0.052905600331723694, "p99_ms": 0.0643228780478239, "mean_ms": 0.04735967999945084, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_6bef1f209a2b6a6e_sdpa_cudnn.json", "configuration": "480x832_tp2_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp2_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp2_sp2_v1/captures/qkv_rank0_6bef1f209a2b6a6e.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 0, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "native_dense", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [512, 10, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.0, "rmse": 0.0, "relative_l2": 0.0, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.0005221612900013495, "rmse": 1.9693656912287394e-05, "relative_l2": 0.001983904969108675, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.10796800255775452, "p95_ms": 0.11774400211870671, "p99_ms": 0.15289503425359724, "mean_ms": 0.11002581348021825, "n": 300}, {"repeat": 1, "p50_ms": 0.1074719987809658, "p95_ms": 0.1144735999405384, "p99_ms": 0.15389215767383524, "mean_ms": 0.11111146678527196, "n": 300}, {"repeat": 2, "p50_ms": 0.10743999853730202, "p95_ms": 0.1137056015431881, "p99_ms": 0.13027488052844996, "mean_ms": 0.10894261337816716, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_6bef1f209a2b6a6e_native_dense.json", "configuration": "480x832_tp2_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp2_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp2_sp2_v1/captures/qkv_rank0_6c601440672149ef.pt", "context": {"phase": "commit", "step": null, "start_frame": 0, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "vllm_fa2", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [512, 10, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.00048828125, "rmse": 2.0471240563407464e-06, "relative_l2": 0.00029158538431995045, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.0002462965742430351, "rmse": 1.8319828422198023e-05, "relative_l2": 0.001666153736434623, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.0490880012512207, "p95_ms": 0.06924639716744427, "p99_ms": 0.1369040064513683, "mean_ms": 0.053087359915177025, "n": 300}, {"repeat": 1, "p50_ms": 0.048847999423742294, "p95_ms": 0.05418719910085201, "p99_ms": 0.06426303729414937, "mean_ms": 0.04974378670255343, "n": 300}, {"repeat": 2, "p50_ms": 0.048656001687049866, "p95_ms": 0.05320319905877114, "p99_ms": 0.0599555213749408, "mean_ms": 0.04947274663796027, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_6c601440672149ef_vllm_fa2.json", "configuration": "480x832_tp2_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp2_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp2_sp2_v1/captures/qkv_rank0_6c601440672149ef.pt", "context": {"phase": "commit", "step": null, "start_frame": 0, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "vllm_fa3", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [512, 10, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.00048828125, "rmse": 7.797658202540029e-07, "relative_l2": 0.00011106718993119931, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.0002465302375597306, "rmse": 1.8325521411840758e-05, "relative_l2": 0.0016666715030722865, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.059487998485565186, "p95_ms": 0.06528319790959358, "p99_ms": 0.07580480329692361, "mean_ms": 0.06038080005596081, "n": 300}, {"repeat": 1, "p50_ms": 0.05942400172352791, "p95_ms": 0.06583200208842754, "p99_ms": 0.07022335790097708, "mean_ms": 0.06044234669456879, "n": 300}, {"repeat": 2, "p50_ms": 0.059408001601696014, "p95_ms": 0.06477280147373676, "p99_ms": 0.07308767728507513, "mean_ms": 0.06040725339204073, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_6c601440672149ef_vllm_fa3.json", "configuration": "480x832_tp2_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp2_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp2_sp2_v1/captures/qkv_rank0_6c601440672149ef.pt", "context": {"phase": "commit", "step": null, "start_frame": 0, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_math", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [512, 10, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.0009765625, "rmse": 5.985184778306253e-06, "relative_l2": 0.0008525093525244846, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.00024413624259587985, "rmse": 1.8135371052777925e-05, "relative_l2": 0.0016493776876535218, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.6177439987659454, "p95_ms": 0.6203216224908829, "p99_ms": 0.6216329401731491, "mean_ms": 0.6179509333769481, "n": 300}, {"repeat": 1, "p50_ms": 0.6182399988174438, "p95_ms": 0.6203552275896073, "p99_ms": 0.6216009324789047, "mean_ms": 0.6186803205808004, "n": 300}, {"repeat": 2, "p50_ms": 0.6181759834289551, "p95_ms": 0.6211855918169021, "p99_ms": 0.6221135765314102, "mean_ms": 0.618819518884023, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_6c601440672149ef_sdpa_math.json", "configuration": "480x832_tp2_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp2_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp2_sp2_v1/captures/qkv_rank0_6c601440672149ef.pt", "context": {"phase": "commit", "step": null, "start_frame": 0, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_flash", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [512, 10, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.00048828125, "rmse": 2.048968191193961e-06, "relative_l2": 0.00029184805661293964, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.0002462965742430351, "rmse": 1.8319828422198023e-05, "relative_l2": 0.001666153736434623, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.04896000027656555, "p95_ms": 0.04927999898791313, "p99_ms": 0.04968192163854833, "mean_ms": 0.049086506801346935, "n": 300}, {"repeat": 1, "p50_ms": 0.048767998814582825, "p95_ms": 0.0490880012512207, "p99_ms": 0.05111680097877969, "mean_ms": 0.04914143986999989, "n": 300}, {"repeat": 2, "p50_ms": 0.04879999905824661, "p95_ms": 0.0490880012512207, "p99_ms": 0.04934847924858329, "mean_ms": 0.04911061324179172, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_6c601440672149ef_sdpa_flash.json", "configuration": "480x832_tp2_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp2_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp2_sp2_v1/captures/qkv_rank0_6c601440672149ef.pt", "context": {"phase": "commit", "step": null, "start_frame": 0, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_efficient", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [512, 10, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.00048828125, "rmse": 3.6082398380383226e-06, "relative_l2": 0.0005139454038626349, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.0002465302375597306, "rmse": 1.846391370947778e-05, "relative_l2": 0.0016792580207234179, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.08656000345945358, "p95_ms": 0.08697599917650223, "p99_ms": 0.08710432015359401, "mean_ms": 0.08669621353348096, "n": 300}, {"repeat": 1, "p50_ms": 0.08675199747085571, "p95_ms": 0.08726400136947632, "p99_ms": 0.08761631660163402, "mean_ms": 0.08717994588116805, "n": 300}, {"repeat": 2, "p50_ms": 0.08671999722719193, "p95_ms": 0.08704159967601299, "p99_ms": 0.08739424228668213, "mean_ms": 0.08695114582777023, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_6c601440672149ef_sdpa_efficient.json", "configuration": "480x832_tp2_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp2_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp2_sp2_v1/captures/qkv_rank0_6c601440672149ef.pt", "context": {"phase": "commit", "step": null, "start_frame": 0, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_cudnn", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [512, 10, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.00048828125, "rmse": 3.605571072150661e-06, "relative_l2": 0.0005135652739312789, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.0002465302375597306, "rmse": 1.846391370947778e-05, "relative_l2": 0.0016792580207234179, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.046112000942230225, "p95_ms": 0.05037599988281727, "p99_ms": 0.05331583987921476, "mean_ms": 0.0466465067739288, "n": 300}, {"repeat": 1, "p50_ms": 0.04623999819159508, "p95_ms": 0.05031679943203926, "p99_ms": 0.05379679981619119, "mean_ms": 0.0469202134013176, "n": 300}, {"repeat": 2, "p50_ms": 0.04646399989724159, "p95_ms": 0.051236798986792564, "p99_ms": 0.06107967827469097, "mean_ms": 0.04742197345942259, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_6c601440672149ef_sdpa_cudnn.json", "configuration": "480x832_tp2_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp2_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp2_sp2_v1/captures/qkv_rank0_6c601440672149ef.pt", "context": {"phase": "commit", "step": null, "start_frame": 0, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "native_dense", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [512, 10, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.0, "rmse": 0.0, "relative_l2": 0.0, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.0002465302375597306, "rmse": 1.832763914830437e-05, "relative_l2": 0.0016668641071971892, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.10595200210809708, "p95_ms": 0.11337919905781746, "p99_ms": 0.1261110457777976, "mean_ms": 0.10744736015796662, "n": 300}, {"repeat": 1, "p50_ms": 0.10627199709415436, "p95_ms": 0.11520639806985855, "p99_ms": 0.12806752651929856, "mean_ms": 0.10824458686014017, "n": 300}, {"repeat": 2, "p50_ms": 0.10595200210809708, "p95_ms": 0.11722239851951599, "p99_ms": 0.134466234743595, "mean_ms": 0.10822325331469376, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_6c601440672149ef_native_dense.json", "configuration": "480x832_tp2_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp2_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp2_sp2_v1/captures/qkv_rank0_7baca46400faa6d9.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 27, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "vllm_fa2", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [512, 10, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.0001220703125, "rmse": 1.2276917065212108e-06, "relative_l2": 0.0007500414656998413, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.00013185228070013155, "rmse": 5.9774578814411576e-06, "relative_l2": 0.0017269006831286404, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.04831999912858009, "p95_ms": 0.05552160162478685, "p99_ms": 0.060982718281447836, "mean_ms": 0.04912394660214583, "n": 300}, {"repeat": 1, "p50_ms": 0.04838399961590767, "p95_ms": 0.05591040086001158, "p99_ms": 0.07398527689278123, "mean_ms": 0.049812799928088985, "n": 300}, {"repeat": 2, "p50_ms": 0.04835199937224388, "p95_ms": 0.05369759909808636, "p99_ms": 0.05888960130512714, "mean_ms": 0.049264213318626085, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_7baca46400faa6d9_vllm_fa2.json", "configuration": "480x832_tp2_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp2_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp2_sp2_v1/captures/qkv_rank0_7baca46400faa6d9.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 27, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "vllm_fa3", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [512, 10, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.0001220703125, "rmse": 4.693940939742722e-07, "relative_l2": 0.0002867699051522668, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.00013185228070013155, "rmse": 6.004525352024777e-06, "relative_l2": 0.0017347205347057705, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.058687999844551086, "p95_ms": 0.06339199841022491, "p99_ms": 0.06688448257744312, "mean_ms": 0.05925493317345778, "n": 300}, {"repeat": 1, "p50_ms": 0.058559998869895935, "p95_ms": 0.06576000154018402, "p99_ms": 0.07577536232769477, "mean_ms": 0.05985653322190046, "n": 300}, {"repeat": 2, "p50_ms": 0.05833600088953972, "p95_ms": 0.06386560164391995, "p99_ms": 0.07014527723193163, "mean_ms": 0.059257493341962494, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_7baca46400faa6d9_vllm_fa3.json", "configuration": "480x832_tp2_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp2_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp2_sp2_v1/captures/qkv_rank0_7baca46400faa6d9.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 27, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_math", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [512, 10, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.000244140625, "rmse": 3.1820071129995033e-06, "relative_l2": 0.0019440037480290939, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.00013185228070013155, "rmse": 5.295261799173165e-06, "relative_l2": 0.0015298127397482273, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.6176639795303345, "p95_ms": 0.6199071824550628, "p99_ms": 0.6217942583560944, "mean_ms": 0.6181570136547089, "n": 300}, {"repeat": 1, "p50_ms": 0.6180800199508667, "p95_ms": 0.6205472141504288, "p99_ms": 0.62108866751194, "mean_ms": 0.6186273076136907, "n": 300}, {"repeat": 2, "p50_ms": 0.6180800199508667, "p95_ms": 0.6208639740943909, "p99_ms": 0.6219872051477432, "mean_ms": 0.6190057603518168, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_7baca46400faa6d9_sdpa_math.json", "configuration": "480x832_tp2_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp2_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp2_sp2_v1/captures/qkv_rank0_7baca46400faa6d9.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 27, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_flash", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [512, 10, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.0001220703125, "rmse": 1.2277308858900616e-06, "relative_l2": 0.0007500654017996635, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.00013185228070013155, "rmse": 5.9774578814411576e-06, "relative_l2": 0.0017269006831286404, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.04918399825692177, "p95_ms": 0.04950400069355965, "p99_ms": 0.05242240026593205, "mean_ms": 0.04931093321492275, "n": 300}, {"repeat": 1, "p50_ms": 0.04902400076389313, "p95_ms": 0.04940799996256828, "p99_ms": 0.05130623944103717, "mean_ms": 0.04942858661214511, "n": 300}, {"repeat": 2, "p50_ms": 0.04896000027656555, "p95_ms": 0.04927999898791313, "p99_ms": 0.05008544135838747, "mean_ms": 0.04926656010250251, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_7baca46400faa6d9_sdpa_flash.json", "configuration": "480x832_tp2_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp2_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp2_sp2_v1/captures/qkv_rank0_7baca46400faa6d9.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 27, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_efficient", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [512, 10, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.0001220703125, "rmse": 2.168118837749083e-06, "relative_l2": 0.0013245825659967212, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.00013185228070013155, "rmse": 6.454064602858091e-06, "relative_l2": 0.0018645934095557091, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.08636800199747086, "p95_ms": 0.08675199747085571, "p99_ms": 0.08704063966870308, "mean_ms": 0.08647914782166481, "n": 300}, {"repeat": 1, "p50_ms": 0.08668799698352814, "p95_ms": 0.0870399996638298, "p99_ms": 0.08727392137050628, "mean_ms": 0.08694047950208188, "n": 300}, {"repeat": 2, "p50_ms": 0.08668799698352814, "p95_ms": 0.08710400015115738, "p99_ms": 0.08742496259510517, "mean_ms": 0.0868966395407915, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_7baca46400faa6d9_sdpa_efficient.json", "configuration": "480x832_tp2_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp2_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp2_sp2_v1/captures/qkv_rank0_7baca46400faa6d9.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 27, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_cudnn", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [512, 10, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.0001220703125, "rmse": 2.1660147069629438e-06, "relative_l2": 0.0013232970760561464, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.00013185228070013155, "rmse": 6.454065552399304e-06, "relative_l2": 0.0018645936838802468, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.046431999653577805, "p95_ms": 0.05310559831559659, "p99_ms": 0.06263520177453732, "mean_ms": 0.04954122663786014, "n": 300}, {"repeat": 1, "p50_ms": 0.04700800031423569, "p95_ms": 0.052164798602461815, "p99_ms": 0.06196000169962591, "mean_ms": 0.049139626659452915, "n": 300}, {"repeat": 2, "p50_ms": 0.046592000871896744, "p95_ms": 0.050785599276423454, "p99_ms": 0.051743998900055856, "mean_ms": 0.0472038401166598, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_7baca46400faa6d9_sdpa_cudnn.json", "configuration": "480x832_tp2_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp2_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp2_sp2_v1/captures/qkv_rank0_7baca46400faa6d9.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 27, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "native_dense", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [512, 10, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.0, "rmse": 0.0, "relative_l2": 0.0, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.00013185228070013155, "rmse": 6.015429620007314e-06, "relative_l2": 0.0017378708016255132, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.10700800269842148, "p95_ms": 0.11552320048213005, "p99_ms": 0.15260704666376107, "mean_ms": 0.10873066671192647, "n": 300}, {"repeat": 1, "p50_ms": 0.1061599999666214, "p95_ms": 0.1131919976323843, "p99_ms": 0.11999007657170288, "mean_ms": 0.10797834681967894, "n": 300}, {"repeat": 2, "p50_ms": 0.1069440022110939, "p95_ms": 0.11590720303356647, "p99_ms": 0.17163039609789765, "mean_ms": 0.11317088022828102, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_7baca46400faa6d9_native_dense.json", "configuration": "480x832_tp2_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp2_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp2_sp2_v1/captures/qkv_rank0_8ec3a76708f5e241.pt", "context": {"phase": "commit", "step": null, "start_frame": 27, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "vllm_fa2", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [512, 10, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.0001220703125, "rmse": 1.1503981904098513e-06, "relative_l2": 0.0007977497247412276, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 6.742393421056356e-05, "rmse": 3.832209818314218e-06, "relative_l2": 0.0017802310561338699, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.04870399832725525, "p95_ms": 0.05389760043472052, "p99_ms": 0.05984480116516351, "mean_ms": 0.04942922670394182, "n": 300}, {"repeat": 1, "p50_ms": 0.04864000156521797, "p95_ms": 0.05532800033688545, "p99_ms": 0.0754412806034088, "mean_ms": 0.050058026586969694, "n": 300}, {"repeat": 2, "p50_ms": 0.04838399961590767, "p95_ms": 0.052767999283969406, "p99_ms": 0.05709407862275835, "mean_ms": 0.049260159904758134, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_8ec3a76708f5e241_vllm_fa2.json", "configuration": "480x832_tp2_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp2_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp2_sp2_v1/captures/qkv_rank0_8ec3a76708f5e241.pt", "context": {"phase": "commit", "step": null, "start_frame": 27, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "vllm_fa3", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [512, 10, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 6.103515625e-05, "rmse": 4.335213801258875e-07, "relative_l2": 0.0003006276996503717, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 6.742393421056356e-05, "rmse": 3.8849170393543405e-06, "relative_l2": 0.0018047158928799472, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.0589120015501976, "p95_ms": 0.06613279692828655, "p99_ms": 0.07556544087827195, "mean_ms": 0.06011253335823615, "n": 300}, {"repeat": 1, "p50_ms": 0.05862399935722351, "p95_ms": 0.06647359952330589, "p99_ms": 0.0827743966132402, "mean_ms": 0.060073600026468436, "n": 300}, {"repeat": 2, "p50_ms": 0.058720000088214874, "p95_ms": 0.06363999992609025, "p99_ms": 0.07168640166521069, "mean_ms": 0.05969194666792949, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_8ec3a76708f5e241_vllm_fa3.json", "configuration": "480x832_tp2_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp2_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp2_sp2_v1/captures/qkv_rank0_8ec3a76708f5e241.pt", "context": {"phase": "commit", "step": null, "start_frame": 27, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_math", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [512, 10, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.000244140625, "rmse": 2.2029477187506933e-06, "relative_l2": 0.0015276457759609073, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 6.553299429325893e-05, "rmse": 3.594119650901572e-06, "relative_l2": 0.0016696276366231506, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.6176479756832123, "p95_ms": 0.6196175724267959, "p99_ms": 0.6207360029220581, "mean_ms": 0.6179576547940572, "n": 300}, {"repeat": 1, "p50_ms": 0.6185600161552429, "p95_ms": 0.620998403429985, "p99_ms": 0.6217942583560944, "mean_ms": 0.6193477328618368, "n": 300}, {"repeat": 2, "p50_ms": 0.6185920238494873, "p95_ms": 0.6208352237939835, "p99_ms": 0.6215427362918854, "mean_ms": 0.6189716279506683, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_8ec3a76708f5e241_sdpa_math.json", "configuration": "480x832_tp2_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp2_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp2_sp2_v1/captures/qkv_rank0_8ec3a76708f5e241.pt", "context": {"phase": "commit", "step": null, "start_frame": 27, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_flash", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [512, 10, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.0001220703125, "rmse": 1.1502895370753902e-06, "relative_l2": 0.0007976743785103476, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 6.742393421056356e-05, "rmse": 3.832209818314218e-06, "relative_l2": 0.0017802310561338699, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.04902400076389313, "p95_ms": 0.04931359924376011, "p99_ms": 0.04957280091941353, "mean_ms": 0.049151360156635446, "n": 300}, {"repeat": 1, "p50_ms": 0.04902400076389313, "p95_ms": 0.0493455994874239, "p99_ms": 0.07399263665080064, "mean_ms": 0.04988629348576069, "n": 300}, {"repeat": 2, "p50_ms": 0.04899200052022934, "p95_ms": 0.049247998744249344, "p99_ms": 0.04966016065329299, "mean_ms": 0.049320853513975936, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_8ec3a76708f5e241_sdpa_flash.json", "configuration": "480x832_tp2_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp2_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp2_sp2_v1/captures/qkv_rank0_8ec3a76708f5e241.pt", "context": {"phase": "commit", "step": null, "start_frame": 27, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_efficient", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [512, 10, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.000244140625, "rmse": 2.0174774858491032e-06, "relative_l2": 0.0013990304595614424, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 6.742393421056356e-05, "rmse": 4.5033755183310445e-06, "relative_l2": 0.002092017226419135, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.08643200248479843, "p95_ms": 0.0867839977145195, "p99_ms": 0.08694431893527507, "mean_ms": 0.08653546807666619, "n": 300}, {"repeat": 1, "p50_ms": 0.08643200248479843, "p95_ms": 0.08688159845769405, "p99_ms": 0.08736096210777759, "mean_ms": 0.08670538788040479, "n": 300}, {"repeat": 2, "p50_ms": 0.08671999722719193, "p95_ms": 0.08707359991967678, "p99_ms": 0.0872646413743496, "mean_ms": 0.08698549245794614, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_8ec3a76708f5e241_sdpa_efficient.json", "configuration": "480x832_tp2_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp2_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp2_sp2_v1/captures/qkv_rank0_8ec3a76708f5e241.pt", "context": {"phase": "commit", "step": null, "start_frame": 27, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_cudnn", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [512, 10, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.000244140625, "rmse": 2.016519120806084e-06, "relative_l2": 0.0013983658762409506, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 6.742393421056356e-05, "rmse": 4.503302457158564e-06, "relative_l2": 0.002091983286271175, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.046160001307725906, "p95_ms": 0.05149120092391968, "p99_ms": 0.056336639933288024, "mean_ms": 0.046801493366559345, "n": 300}, {"repeat": 1, "p50_ms": 0.04608000069856644, "p95_ms": 0.050756799057126045, "p99_ms": 0.05539648082107305, "mean_ms": 0.04684490662068128, "n": 300}, {"repeat": 2, "p50_ms": 0.046112000942230225, "p95_ms": 0.049668798223137856, "p99_ms": 0.052807679735124104, "mean_ms": 0.04669503991802534, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_8ec3a76708f5e241_sdpa_cudnn.json", "configuration": "480x832_tp2_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp2_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp2_sp2_v1/captures/qkv_rank0_8ec3a76708f5e241.pt", "context": {"phase": "commit", "step": null, "start_frame": 27, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "native_dense", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [512, 10, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.0, "rmse": 0.0, "relative_l2": 0.0, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 6.742393421056356e-05, "rmse": 3.906451067226311e-06, "relative_l2": 0.0018147194018209578, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.10601600259542465, "p95_ms": 0.11577760241925716, "p99_ms": 0.15223168417811392, "mean_ms": 0.10848309325675169, "n": 300}, {"repeat": 1, "p50_ms": 0.10556799918413162, "p95_ms": 0.11232799850404263, "p99_ms": 0.12303391784429547, "mean_ms": 0.10712703997890155, "n": 300}, {"repeat": 2, "p50_ms": 0.10595200210809708, "p95_ms": 0.11213119700551033, "p99_ms": 0.1205257607251405, "mean_ms": 0.10743231996893883, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_8ec3a76708f5e241_native_dense.json", "configuration": "480x832_tp2_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp2_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp2_sp2_v1/captures/qkv_rank0_a7ce4999e4f2ba20.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 9, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "vllm_fa2", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [512, 10, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.0001220703125, "rmse": 1.2242118156476884e-06, "relative_l2": 0.0007236570600972542, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 8.797855827336054e-05, "rmse": 5.130591479296537e-06, "relative_l2": 0.0019279313287139288, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.04819199815392494, "p95_ms": 0.05249600093811751, "p99_ms": 0.06617087677121157, "mean_ms": 0.04897258665412665, "n": 300}, {"repeat": 1, "p50_ms": 0.04790399968624115, "p95_ms": 0.05290080048143864, "p99_ms": 0.05801984146237364, "mean_ms": 0.04879637330770493, "n": 300}, {"repeat": 2, "p50_ms": 0.04800000041723251, "p95_ms": 0.05347680114209652, "p99_ms": 0.05627199921756973, "mean_ms": 0.04882453333586454, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_a7ce4999e4f2ba20_vllm_fa2.json", "configuration": "480x832_tp2_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp2_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp2_sp2_v1/captures/qkv_rank0_a7ce4999e4f2ba20.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 9, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "vllm_fa3", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [512, 10, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.0001220703125, "rmse": 4.6145071985153234e-07, "relative_l2": 0.000272773115762528, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 8.797855827336054e-05, "rmse": 5.1625492166297935e-06, "relative_l2": 0.0019399401435354208, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.05852799862623215, "p95_ms": 0.06477440148591995, "p99_ms": 0.07677088268101206, "mean_ms": 0.059727146724859875, "n": 300}, {"repeat": 1, "p50_ms": 0.05862399935722351, "p95_ms": 0.06659200042486191, "p99_ms": 0.07663360141217697, "mean_ms": 0.05992010666678349, "n": 300}, {"repeat": 2, "p50_ms": 0.05852799862623215, "p95_ms": 0.06416159681975842, "p99_ms": 0.08014975868165485, "mean_ms": 0.060398186556994915, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_a7ce4999e4f2ba20_vllm_fa3.json", "configuration": "480x832_tp2_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp2_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp2_sp2_v1/captures/qkv_rank0_a7ce4999e4f2ba20.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 9, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_math", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [512, 10, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.000244140625, "rmse": 3.1705202951548342e-06, "relative_l2": 0.0018741604732482947, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 5.9410297260108214e-05, "rmse": 4.423149753197181e-06, "relative_l2": 0.0016620947146529293, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.6178080141544342, "p95_ms": 0.6197808086872101, "p99_ms": 0.6210595399141312, "mean_ms": 0.6179553063710531, "n": 300}, {"repeat": 1, "p50_ms": 0.618399977684021, "p95_ms": 0.6206079721450806, "p99_ms": 0.6217926579713822, "mean_ms": 0.6189029333988826, "n": 300}, {"repeat": 2, "p50_ms": 0.6182239949703217, "p95_ms": 0.62048160135746, "p99_ms": 0.621667827963829, "mean_ms": 0.618678294022878, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_a7ce4999e4f2ba20_sdpa_math.json", "configuration": "480x832_tp2_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp2_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp2_sp2_v1/captures/qkv_rank0_a7ce4999e4f2ba20.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 9, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_flash", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [512, 10, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.0001220703125, "rmse": 1.2242692059675821e-06, "relative_l2": 0.0007236909847086992, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 8.797855827336054e-05, "rmse": 5.130591479296537e-06, "relative_l2": 0.0019279313287139288, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.04912000149488449, "p95_ms": 0.04944000020623207, "p99_ms": 0.049668478220701216, "mean_ms": 0.04934538676093022, "n": 300}, {"repeat": 1, "p50_ms": 0.04931199923157692, "p95_ms": 0.05283519998192787, "p99_ms": 0.05839264038950194, "mean_ms": 0.05071765342106422, "n": 300}, {"repeat": 2, "p50_ms": 0.048895999789237976, "p95_ms": 0.04915360156446695, "p99_ms": 0.04945887997746463, "mean_ms": 0.0494820266465346, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_a7ce4999e4f2ba20_sdpa_flash.json", "configuration": "480x832_tp2_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp2_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp2_sp2_v1/captures/qkv_rank0_a7ce4999e4f2ba20.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 9, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_efficient", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [512, 10, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.0001220703125, "rmse": 2.171853321126983e-06, "relative_l2": 0.0012838276589396335, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 8.797855827336054e-05, "rmse": 5.5850438234574005e-06, "relative_l2": 0.002098701680485406, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.08640000224113464, "p95_ms": 0.08672159723937511, "p99_ms": 0.08707263991236687, "mean_ms": 0.08654986808697383, "n": 300}, {"repeat": 1, "p50_ms": 0.08665599673986435, "p95_ms": 0.0870399996638298, "p99_ms": 0.08726592138409614, "mean_ms": 0.08689386633535226, "n": 300}, {"repeat": 2, "p50_ms": 0.08671999722719193, "p95_ms": 0.08707199990749359, "p99_ms": 0.08729664161801338, "mean_ms": 0.08691221252083778, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_a7ce4999e4f2ba20_sdpa_efficient.json", "configuration": "480x832_tp2_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp2_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp2_sp2_v1/captures/qkv_rank0_a7ce4999e4f2ba20.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 9, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_cudnn", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [512, 10, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.0001220703125, "rmse": 2.1703749938667826e-06, "relative_l2": 0.001282953788956266, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 8.797855827336054e-05, "rmse": 5.5850438234574005e-06, "relative_l2": 0.002098701680485406, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.04636799916625023, "p95_ms": 0.05058240126818419, "p99_ms": 0.055562238320708265, "mean_ms": 0.046917120056847735, "n": 300}, {"repeat": 1, "p50_ms": 0.04636799916625023, "p95_ms": 0.051364799961447716, "p99_ms": 0.0557135995477438, "mean_ms": 0.04724479998151461, "n": 300}, {"repeat": 2, "p50_ms": 0.04630399867892265, "p95_ms": 0.05127519927918911, "p99_ms": 0.055806399770080985, "mean_ms": 0.047539840005338195, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_a7ce4999e4f2ba20_sdpa_cudnn.json", "configuration": "480x832_tp2_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp2_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp2_sp2_v1/captures/qkv_rank0_a7ce4999e4f2ba20.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 9, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "native_dense", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [512, 10, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.0, "rmse": 0.0, "relative_l2": 0.0, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 8.797855827336054e-05, "rmse": 5.168994684931184e-06, "relative_l2": 0.0019423621684260455, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.10619199648499489, "p95_ms": 0.113569600507617, "p99_ms": 0.1368505583703517, "mean_ms": 0.10771039995054404, "n": 300}, {"repeat": 1, "p50_ms": 0.107744000852108, "p95_ms": 0.11558400094509125, "p99_ms": 0.12920064657926547, "mean_ms": 0.10944373354315758, "n": 300}, {"repeat": 2, "p50_ms": 0.10604800283908844, "p95_ms": 0.11350720003247261, "p99_ms": 0.16544479936361298, "mean_ms": 0.1125210665166378, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_a7ce4999e4f2ba20_native_dense.json", "configuration": "480x832_tp2_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp2_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp2_sp2_v1/captures/qkv_rank0_c6a5fcb1e5f12d0d.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 9, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "vllm_fa2", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [29640, 10, 128], "q_len": 4680, "k_len": 18720, "reference_error": {"max_abs": 0.03125, "rmse": 0.0005667357625975778, "relative_l2": 0.0010102590926331137, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.008338139423881952, "rmse": 0.0008250061956286321, "relative_l2": 0.0019467300166385222, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 1.2773120403289795, "p95_ms": 1.2832975924015044, "p99_ms": 1.284814451932907, "mean_ms": 1.2769700296719868, "n": 300}, {"repeat": 1, "p50_ms": 1.2791199684143066, "p95_ms": 1.286785650253296, "p99_ms": 1.2878137016296387, "mean_ms": 1.2806459744771321, "n": 300}, {"repeat": 2, "p50_ms": 1.2912800312042236, "p95_ms": 1.3203615605831147, "p99_ms": 1.3248006546497344, "mean_ms": 1.2952134414513905, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_c6a5fcb1e5f12d0d_vllm_fa2.json", "configuration": "480x832_tp2_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp2_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp2_sp2_v1/captures/qkv_rank0_c6a5fcb1e5f12d0d.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 9, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "vllm_fa3", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [29640, 10, 128], "q_len": 4680, "k_len": 18720, "reference_error": {"max_abs": 0.0, "rmse": 0.0, "relative_l2": 0.0, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.008338139423881952, "rmse": 0.0008372650837925386, "relative_l2": 0.0019756567637172, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.6954560279846191, "p95_ms": 0.699176025390625, "p99_ms": 0.7020175957679748, "mean_ms": 0.678092586795489, "n": 300}, {"repeat": 1, "p50_ms": 0.6956640183925629, "p95_ms": 0.7055424064397812, "p99_ms": 0.7064649528264999, "mean_ms": 0.692375892996788, "n": 300}, {"repeat": 2, "p50_ms": 0.6920159757137299, "p95_ms": 0.6968383997678756, "p99_ms": 0.7010569638013839, "mean_ms": 0.6921579736471176, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_c6a5fcb1e5f12d0d_vllm_fa3.json", "configuration": "480x832_tp2_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp2_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp2_sp2_v1/captures/qkv_rank0_c6a5fcb1e5f12d0d.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 9, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_math", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [29640, 10, 128], "q_len": 4680, "k_len": 18720, "reference_error": {"max_abs": 0.03125, "rmse": 0.0009933852488920604, "relative_l2": 0.0017708013970761582, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.007445773795270227, "rmse": 0.0007114648130294856, "relative_l2": 0.0016788115224410634, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 18.878047943115234, "p95_ms": 19.163749408721923, "p99_ms": 19.54822685241699, "mean_ms": 18.907795906066895, "n": 300}, {"repeat": 1, "p50_ms": 18.89521598815918, "p95_ms": 19.12330503463745, "p99_ms": 19.250985527038573, "mean_ms": 18.92063756942749, "n": 300}, {"repeat": 2, "p50_ms": 18.90532875061035, "p95_ms": 19.19017162322998, "p99_ms": 19.276229991912842, "mean_ms": 18.94955993652344, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_c6a5fcb1e5f12d0d_sdpa_math.json", "configuration": "480x832_tp2_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp2_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp2_sp2_v1/captures/qkv_rank0_c6a5fcb1e5f12d0d.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 9, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_flash", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [29640, 10, 128], "q_len": 4680, "k_len": 18720, "reference_error": {"max_abs": 0.03125, "rmse": 0.0005672101489435932, "relative_l2": 0.001011104730320219, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.008338139423881952, "rmse": 0.000825045503408997, "relative_l2": 0.0019468227694400518, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 1.3843200206756592, "p95_ms": 1.3911920011043548, "p99_ms": 1.394466896057129, "mean_ms": 1.3851513608296713, "n": 300}, {"repeat": 1, "p50_ms": 1.3816959857940674, "p95_ms": 1.3861775934696197, "p99_ms": 1.3876163578033447, "mean_ms": 1.3800169599056245, "n": 300}, {"repeat": 2, "p50_ms": 1.4065119624137878, "p95_ms": 1.4554672181606292, "p99_ms": 1.4752690732479095, "mean_ms": 1.4162214398384094, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_c6a5fcb1e5f12d0d_sdpa_flash.json", "configuration": "480x832_tp2_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp2_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp2_sp2_v1/captures/qkv_rank0_c6a5fcb1e5f12d0d.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 9, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_efficient", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [29640, 10, 128], "q_len": 4680, "k_len": 18720, "reference_error": {"max_abs": 0.03125, "rmse": 0.0007760670397883893, "relative_l2": 0.0013834115211744634, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.009269161331428855, "rmse": 0.0008322708750293184, "relative_l2": 0.0019638721539042858, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 2.7518399953842163, "p95_ms": 2.7706271767616273, "p99_ms": 2.785780725479126, "mean_ms": 2.7533267227808635, "n": 300}, {"repeat": 1, "p50_ms": 2.7486560344696045, "p95_ms": 2.7699233055114747, "p99_ms": 2.774863984584808, "mean_ms": 2.7505443207422893, "n": 300}, {"repeat": 2, "p50_ms": 2.74833607673645, "p95_ms": 2.771873641014099, "p99_ms": 2.777476818561554, "mean_ms": 2.751130243142446, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_c6a5fcb1e5f12d0d_sdpa_efficient.json", "configuration": "480x832_tp2_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp2_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp2_sp2_v1/captures/qkv_rank0_c6a5fcb1e5f12d0d.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 9, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_cudnn", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [29640, 10, 128], "q_len": 4680, "k_len": 18720, "reference_error": {"max_abs": 0.03125, "rmse": 0.0007752974384960745, "relative_l2": 0.001382039635448213, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.009269161331428855, "rmse": 0.0008319531279590026, "relative_l2": 0.001963122380432571, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.827919989824295, "p95_ms": 0.842564794421196, "p99_ms": 0.8430723172426223, "mean_ms": 0.8113216006755829, "n": 300}, {"repeat": 1, "p50_ms": 0.8271679878234863, "p95_ms": 0.8430719971656799, "p99_ms": 0.844578241109848, "mean_ms": 0.8230017085870107, "n": 300}, {"repeat": 2, "p50_ms": 0.8226720094680786, "p95_ms": 0.837260827422142, "p99_ms": 0.8434240221977234, "mean_ms": 0.8209932802120844, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_c6a5fcb1e5f12d0d_sdpa_cudnn.json", "configuration": "480x832_tp2_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp2_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp2_sp2_v1/captures/qkv_rank0_c6a5fcb1e5f12d0d.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 9, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "native_dense", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [29640, 10, 128], "q_len": 4680, "k_len": 18720, "reference_error": {"max_abs": 0.03125, "rmse": 0.0005027492059587385, "relative_l2": 0.0008961971171643583, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.008338139423881952, "rmse": 0.0008357267575398385, "relative_l2": 0.0019720268444421906, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.7022880017757416, "p95_ms": 0.7320640087127686, "p99_ms": 0.7329282981157302, "mean_ms": 0.6945548808574676, "n": 300}, {"repeat": 1, "p50_ms": 0.6907519996166229, "p95_ms": 0.6980479955673218, "p99_ms": 0.6987535864114761, "mean_ms": 0.6846967484553655, "n": 300}, {"repeat": 2, "p50_ms": 0.6926880180835724, "p95_ms": 0.7029775887727737, "p99_ms": 0.7035215997695923, "mean_ms": 0.6897617067893346, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_c6a5fcb1e5f12d0d_native_dense.json", "configuration": "480x832_tp2_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp2_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp2_sp2_v1/captures/qkv_rank0_cf2e5e4d322f9347.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 27, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "vllm_fa2", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [29640, 10, 128], "q_len": 4680, "k_len": 28080, "reference_error": {"max_abs": 0.03125, "rmse": 0.0005561887266841604, "relative_l2": 0.00099234785380834, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.0088576065358561, "rmse": 0.0008021409422662321, "relative_l2": 0.0019439283468281384, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 1.9207839965820312, "p95_ms": 1.9415776133537292, "p99_ms": 1.9446242928504944, "mean_ms": 1.9204760551452638, "n": 300}, {"repeat": 1, "p50_ms": 1.9363519549369812, "p95_ms": 2.0157344222068785, "p99_ms": 2.021920926570892, "mean_ms": 1.9448246403535208, "n": 300}, {"repeat": 2, "p50_ms": 1.921999990940094, "p95_ms": 1.9358416080474854, "p99_ms": 1.947019853591919, "mean_ms": 1.9246950368086497, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_cf2e5e4d322f9347_vllm_fa2.json", "configuration": "480x832_tp2_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp2_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp2_sp2_v1/captures/qkv_rank0_cf2e5e4d322f9347.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 27, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "vllm_fa3", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [29640, 10, 128], "q_len": 4680, "k_len": 28080, "reference_error": {"max_abs": 0.0, "rmse": 0.0, "relative_l2": 0.0, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.0088576065358561, "rmse": 0.0008091472718205208, "relative_l2": 0.0019609076604501594, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 1.0473599433898926, "p95_ms": 1.0801295578479766, "p99_ms": 1.0829478752613069, "mean_ms": 1.0446151441335678, "n": 300}, {"repeat": 1, "p50_ms": 1.0521600246429443, "p95_ms": 1.069283229112625, "p99_ms": 1.0748163092136382, "mean_ms": 1.0456963183482488, "n": 300}, {"repeat": 2, "p50_ms": 1.0497440099716187, "p95_ms": 1.1219856023788453, "p99_ms": 1.1252809822559358, "mean_ms": 1.0625361061096192, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_cf2e5e4d322f9347_vllm_fa3.json", "configuration": "480x832_tp2_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp2_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp2_sp2_v1/captures/qkv_rank0_cf2e5e4d322f9347.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 27, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_math", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [29640, 10, 128], "q_len": 4680, "k_len": 28080, "reference_error": {"max_abs": 0.03125, "rmse": 0.000977651834629706, "relative_l2": 0.0017443192451067065, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.007788278483080724, "rmse": 0.0006752488626088266, "relative_l2": 0.0016364149191540594, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 28.362319946289062, "p95_ms": 28.930531215667724, "p99_ms": 30.265896739959704, "mean_ms": 28.43719944636027, "n": 300}, {"repeat": 1, "p50_ms": 28.363248825073242, "p95_ms": 28.65727071762085, "p99_ms": 28.766960353851317, "mean_ms": 28.3917581876119, "n": 300}, {"repeat": 2, "p50_ms": 28.398128509521484, "p95_ms": 28.64749708175659, "p99_ms": 28.758988780975344, "mean_ms": 28.396267026265463, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_cf2e5e4d322f9347_sdpa_math.json", "configuration": "480x832_tp2_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp2_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp2_sp2_v1/captures/qkv_rank0_cf2e5e4d322f9347.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 27, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_flash", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [29640, 10, 128], "q_len": 4680, "k_len": 28080, "reference_error": {"max_abs": 0.03125, "rmse": 0.0005571223338881936, "relative_l2": 0.0009940135889460256, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.0088576065358561, "rmse": 0.0008020998230974734, "relative_l2": 0.0019438286976099822, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 2.0672800540924072, "p95_ms": 2.0881103038787843, "p99_ms": 2.0910848808288574, "mean_ms": 2.068050669829051, "n": 300}, {"repeat": 1, "p50_ms": 2.081152081489563, "p95_ms": 2.201889681816101, "p99_ms": 2.2043647027015685, "mean_ms": 2.1036934423446656, "n": 300}, {"repeat": 2, "p50_ms": 2.0683839321136475, "p95_ms": 2.1339567065238954, "p99_ms": 2.1383021783828733, "mean_ms": 2.0809929617245992, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_cf2e5e4d322f9347_sdpa_flash.json", "configuration": "480x832_tp2_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp2_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp2_sp2_v1/captures/qkv_rank0_cf2e5e4d322f9347.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 27, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_efficient", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [29640, 10, 128], "q_len": 4680, "k_len": 28080, "reference_error": {"max_abs": 0.03125, "rmse": 0.0007707981797681338, "relative_l2": 0.0013752524686582541, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.0088576065358561, "rmse": 0.0008029680037132037, "relative_l2": 0.001945932668147026, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 4.103104114532471, "p95_ms": 4.13745265007019, "p99_ms": 4.145247077941894, "mean_ms": 4.106611526807149, "n": 300}, {"repeat": 1, "p50_ms": 4.1010401248931885, "p95_ms": 4.130611085891724, "p99_ms": 4.141672029495239, "mean_ms": 4.103814717928569, "n": 300}, {"repeat": 2, "p50_ms": 4.101263999938965, "p95_ms": 4.13132336139679, "p99_ms": 4.140304517745972, "mean_ms": 4.1043150806427, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_cf2e5e4d322f9347_sdpa_efficient.json", "configuration": "480x832_tp2_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp2_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp2_sp2_v1/captures/qkv_rank0_cf2e5e4d322f9347.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 27, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_cudnn", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [29640, 10, 128], "q_len": 4680, "k_len": 28080, "reference_error": {"max_abs": 0.03125, "rmse": 0.0007705408137178546, "relative_l2": 0.0013747932780357468, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.0088576065358561, "rmse": 0.0008024639160963256, "relative_l2": 0.0019447110496556847, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 1.2231680154800415, "p95_ms": 1.2428223729133605, "p99_ms": 1.2447359561920166, "mean_ms": 1.211253760655721, "n": 300}, {"repeat": 1, "p50_ms": 1.226319968700409, "p95_ms": 1.2354575932025909, "p99_ms": 1.2383049786090852, "mean_ms": 1.2237531773249308, "n": 300}, {"repeat": 2, "p50_ms": 1.2430559992790222, "p95_ms": 1.466531252861023, "p99_ms": 1.4686402904987335, "mean_ms": 1.2846151494979858, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_cf2e5e4d322f9347_sdpa_cudnn.json", "configuration": "480x832_tp2_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp2_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp2_sp2_v1/captures/qkv_rank0_cf2e5e4d322f9347.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 27, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "native_dense", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [29640, 10, 128], "q_len": 4680, "k_len": 28080, "reference_error": {"max_abs": 0.03125, "rmse": 0.0005152115266434364, "relative_l2": 0.0009192366335254109, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.007788278483080724, "rmse": 0.000815752079321817, "relative_l2": 0.0019769139155240353, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 1.0210720300674438, "p95_ms": 1.0342480063438415, "p99_ms": 1.0352322685718536, "mean_ms": 1.005796478788058, "n": 300}, {"repeat": 1, "p50_ms": 1.0283839702606201, "p95_ms": 1.0343375742435454, "p99_ms": 1.0362928223609924, "mean_ms": 1.0237431488434474, "n": 300}, {"repeat": 2, "p50_ms": 1.0288479924201965, "p95_ms": 1.1885423719882966, "p99_ms": 1.190369656085968, "mean_ms": 1.0610951463381448, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_cf2e5e4d322f9347_native_dense.json", "configuration": "480x832_tp2_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp2_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp2_sp2_v1/captures/qkv_rank0_df3a5d6865cf4937.pt", "context": {"phase": "commit", "step": null, "start_frame": 9, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "vllm_fa2", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [29640, 10, 128], "q_len": 4680, "k_len": 18720, "reference_error": {"max_abs": 0.03125, "rmse": 0.0004294886557431734, "relative_l2": 0.0005372592844016082, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.015577997885023898, "rmse": 0.000975910621199614, "relative_l2": 0.0017008123037051729, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 1.2771199941635132, "p95_ms": 1.2822416305541993, "p99_ms": 1.2832006657123565, "mean_ms": 1.2752098127206166, "n": 300}, {"repeat": 1, "p50_ms": 1.276639997959137, "p95_ms": 1.2952352523803712, "p99_ms": 1.2971302151679993, "mean_ms": 1.2801086954275767, "n": 300}, {"repeat": 2, "p50_ms": 1.29694402217865, "p95_ms": 1.3607408106327057, "p99_ms": 1.363930239677429, "mean_ms": 1.3092647465070089, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_df3a5d6865cf4937_vllm_fa2.json", "configuration": "480x832_tp2_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp2_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp2_sp2_v1/captures/qkv_rank0_df3a5d6865cf4937.pt", "context": {"phase": "commit", "step": null, "start_frame": 9, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "vllm_fa3", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [29640, 10, 128], "q_len": 4680, "k_len": 18720, "reference_error": {"max_abs": 0.0, "rmse": 0.0, "relative_l2": 0.0, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.015672002114976102, "rmse": 0.000975161952523733, "relative_l2": 0.0016995075275630996, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.6948480010032654, "p95_ms": 0.7016016155481338, "p99_ms": 0.7027881616353988, "mean_ms": 0.6710149323940278, "n": 300}, {"repeat": 1, "p50_ms": 0.6974560022354126, "p95_ms": 0.7026879787445068, "p99_ms": 0.7033308935165405, "mean_ms": 0.6903265041112899, "n": 300}, {"repeat": 2, "p50_ms": 0.6901600062847137, "p95_ms": 0.6995487928390502, "p99_ms": 0.7031718152761459, "mean_ms": 0.6898210148016611, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_df3a5d6865cf4937_vllm_fa3.json", "configuration": "480x832_tp2_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp2_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp2_sp2_v1/captures/qkv_rank0_df3a5d6865cf4937.pt", "context": {"phase": "commit", "step": null, "start_frame": 9, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_math", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [29640, 10, 128], "q_len": 4680, "k_len": 18720, "reference_error": {"max_abs": 0.0625, "rmse": 0.0008008211678955669, "relative_l2": 0.001001769434055817, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.015577997885023898, "rmse": 0.0009582482838116969, "relative_l2": 0.0016700304676548234, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 18.887871742248535, "p95_ms": 19.280480670928956, "p99_ms": 19.83688112258911, "mean_ms": 18.943059940338134, "n": 300}, {"repeat": 1, "p50_ms": 18.894800186157227, "p95_ms": 19.161413288116456, "p99_ms": 19.272791061401367, "mean_ms": 18.930174922943117, "n": 300}, {"repeat": 2, "p50_ms": 18.90011215209961, "p95_ms": 19.15927267074585, "p99_ms": 19.26681505203247, "mean_ms": 18.931224740346273, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_df3a5d6865cf4937_sdpa_math.json", "configuration": "480x832_tp2_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp2_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp2_sp2_v1/captures/qkv_rank0_df3a5d6865cf4937.pt", "context": {"phase": "commit", "step": null, "start_frame": 9, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_flash", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [29640, 10, 128], "q_len": 4680, "k_len": 18720, "reference_error": {"max_abs": 0.03125, "rmse": 0.0004304051738835414, "relative_l2": 0.0005384057823909046, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.015577997885023898, "rmse": 0.0009759106202033164, "relative_l2": 0.0017008123019688303, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 1.3823519945144653, "p95_ms": 1.3859519958496094, "p99_ms": 1.3878774213790894, "mean_ms": 1.379632102648417, "n": 300}, {"repeat": 1, "p50_ms": 1.381488025188446, "p95_ms": 1.3839071810245513, "p99_ms": 1.3847362625598907, "mean_ms": 1.3789747222264608, "n": 300}, {"repeat": 2, "p50_ms": 1.4054719805717468, "p95_ms": 1.4426752388477326, "p99_ms": 1.4491893815994263, "mean_ms": 1.4087209618091583, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_df3a5d6865cf4937_sdpa_flash.json", "configuration": "480x832_tp2_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp2_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp2_sp2_v1/captures/qkv_rank0_df3a5d6865cf4937.pt", "context": {"phase": "commit", "step": null, "start_frame": 9, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_efficient", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [29640, 10, 128], "q_len": 4680, "k_len": 18720, "reference_error": {"max_abs": 0.0625, "rmse": 0.0006796999657987322, "relative_l2": 0.0008502555593719632, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.015672002114976102, "rmse": 0.0009673590580544767, "relative_l2": 0.001685908680876186, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 2.74236798286438, "p95_ms": 2.7587279796600344, "p99_ms": 2.7701718258857726, "mean_ms": 2.745163519382477, "n": 300}, {"repeat": 1, "p50_ms": 2.741935968399048, "p95_ms": 2.757016098499298, "p99_ms": 2.769228219985962, "mean_ms": 2.7439575497309368, "n": 300}, {"repeat": 2, "p50_ms": 2.741536021232605, "p95_ms": 2.7586448788642883, "p99_ms": 2.7734124159812925, "mean_ms": 2.744137601852417, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_df3a5d6865cf4937_sdpa_efficient.json", "configuration": "480x832_tp2_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp2_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp2_sp2_v1/captures/qkv_rank0_df3a5d6865cf4937.pt", "context": {"phase": "commit", "step": null, "start_frame": 9, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_cudnn", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [29640, 10, 128], "q_len": 4680, "k_len": 18720, "reference_error": {"max_abs": 0.0625, "rmse": 0.0006798219712006014, "relative_l2": 0.0008504081793166926, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.015672002114976102, "rmse": 0.0009673291098942182, "relative_l2": 0.0016858564873675447, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.818336009979248, "p95_ms": 0.8294944047927857, "p99_ms": 0.8356515204906464, "mean_ms": 0.8082406387726466, "n": 300}, {"repeat": 1, "p50_ms": 0.8068479895591736, "p95_ms": 0.8275040060281753, "p99_ms": 0.8362256193161011, "mean_ms": 0.8088043741385142, "n": 300}, {"repeat": 2, "p50_ms": 0.8277760148048401, "p95_ms": 0.8374912202358246, "p99_ms": 0.843232616186142, "mean_ms": 0.8201496527592341, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_df3a5d6865cf4937_sdpa_cudnn.json", "configuration": "480x832_tp2_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp2_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp2_sp2_v1/captures/qkv_rank0_df3a5d6865cf4937.pt", "context": {"phase": "commit", "step": null, "start_frame": 9, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "native_dense", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [29640, 10, 128], "q_len": 4680, "k_len": 18720, "reference_error": {"max_abs": 0.03125, "rmse": 0.0003845231073475975, "relative_l2": 0.00048101063142629155, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.015672002114976102, "rmse": 0.0009753463475053411, "relative_l2": 0.0016998288902439107, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.6940319836139679, "p95_ms": 0.7165807813405991, "p99_ms": 0.7170246678590775, "mean_ms": 0.6823570150136947, "n": 300}, {"repeat": 1, "p50_ms": 0.6868000030517578, "p95_ms": 0.6937631785869598, "p99_ms": 0.694977252483368, "mean_ms": 0.6784045859177907, "n": 300}, {"repeat": 2, "p50_ms": 0.6928159892559052, "p95_ms": 0.6988496094942093, "p99_ms": 0.7005833673477173, "mean_ms": 0.6895074133078257, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_df3a5d6865cf4937_native_dense.json", "configuration": "480x832_tp2_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp2_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp4_sp1_v1/captures/qkv_rank0_2419ddf76b22ab88.pt", "context": {"phase": "commit", "step": null, "start_frame": 9, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "vllm_fa2", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [512, 10, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.0001220703125, "rmse": 1.1657889261333973e-06, "relative_l2": 0.0007322554551675079, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 8.805995746343637e-05, "rmse": 4.6015956510343155e-06, "relative_l2": 0.0018109745909983298, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.0459199994802475, "p95_ms": 0.04620960149914026, "p99_ms": 0.046540160365402684, "mean_ms": 0.04613439987103144, "n": 300}, {"repeat": 1, "p50_ms": 0.0459199994802475, "p95_ms": 0.04620800167322159, "p99_ms": 0.04640127941966057, "mean_ms": 0.046367253301044305, "n": 300}, {"repeat": 2, "p50_ms": 0.045951999723911285, "p95_ms": 0.04623999819159508, "p99_ms": 0.047147201262414444, "mean_ms": 0.04628426663577557, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_2419ddf76b22ab88_vllm_fa2.json", "configuration": "480x832_tp4_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp4_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp4_sp1_v1/captures/qkv_rank0_2419ddf76b22ab88.pt", "context": {"phase": "commit", "step": null, "start_frame": 9, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "vllm_fa3", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [512, 10, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.0001220703125, "rmse": 4.43024566576941e-07, "relative_l2": 0.0002782726344168971, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 8.805995746343637e-05, "rmse": 4.623869014542002e-06, "relative_l2": 0.0018197403536657697, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.06284800171852112, "p95_ms": 0.06992800310254098, "p99_ms": 0.07640416048467158, "mean_ms": 0.06384661334256331, "n": 300}, {"repeat": 1, "p50_ms": 0.062431998550891876, "p95_ms": 0.07031519897282124, "p99_ms": 0.09211648091673849, "mean_ms": 0.06403040000547966, "n": 300}, {"repeat": 2, "p50_ms": 0.06259199976921082, "p95_ms": 0.06917919777333736, "p99_ms": 0.07934431985020628, "mean_ms": 0.06385600009312233, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_2419ddf76b22ab88_vllm_fa3.json", "configuration": "480x832_tp4_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp4_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp4_sp1_v1/captures/qkv_rank0_2419ddf76b22ab88.pt", "context": {"phase": "commit", "step": null, "start_frame": 9, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_math", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [512, 10, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.000244140625, "rmse": 2.334816730587177e-06, "relative_l2": 0.0014665453149047953, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 8.805995746343637e-05, "rmse": 4.312875575037246e-06, "relative_l2": 0.0016973477621342436, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.6175040006637573, "p95_ms": 0.6203248202800751, "p99_ms": 0.6220880275964736, "mean_ms": 0.6176667739947637, "n": 300}, {"repeat": 1, "p50_ms": 0.6176799833774567, "p95_ms": 0.6198783755302429, "p99_ms": 0.6209932845830918, "mean_ms": 0.6181083746751149, "n": 300}, {"repeat": 2, "p50_ms": 0.6174399852752686, "p95_ms": 0.6198416233062745, "p99_ms": 0.6208333057165146, "mean_ms": 0.6179621344804764, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_2419ddf76b22ab88_sdpa_math.json", "configuration": "480x832_tp4_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp4_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp4_sp1_v1/captures/qkv_rank0_2419ddf76b22ab88.pt", "context": {"phase": "commit", "step": null, "start_frame": 9, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_flash", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [512, 10, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.0001220703125, "rmse": 1.1655149026280147e-06, "relative_l2": 0.000732083335496303, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 8.805995746343637e-05, "rmse": 4.6015956510343155e-06, "relative_l2": 0.0018109745909983298, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.04879999905824661, "p95_ms": 0.04912000149488449, "p99_ms": 0.08579071670770634, "mean_ms": 0.050350186539192994, "n": 300}, {"repeat": 1, "p50_ms": 0.048767998814582825, "p95_ms": 0.049057601019740105, "p99_ms": 0.05571327954530716, "mean_ms": 0.04925642643123865, "n": 300}, {"repeat": 2, "p50_ms": 0.04879999905824661, "p95_ms": 0.0504143999889493, "p99_ms": 0.0661599968373775, "mean_ms": 0.04979071990897258, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_2419ddf76b22ab88_sdpa_flash.json", "configuration": "480x832_tp4_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp4_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp4_sp1_v1/captures/qkv_rank0_2419ddf76b22ab88.pt", "context": {"phase": "commit", "step": null, "start_frame": 9, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_efficient", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [512, 10, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.000244140625, "rmse": 2.019103127181165e-06, "relative_l2": 0.0012682392552208913, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 8.805995746343637e-05, "rmse": 5.094872965232186e-06, "relative_l2": 0.0020051056598868904, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.08649600297212601, "p95_ms": 0.08687999844551086, "p99_ms": 0.08710464015603066, "mean_ms": 0.08662784099578857, "n": 300}, {"repeat": 1, "p50_ms": 0.0865280032157898, "p95_ms": 0.08691359870135784, "p99_ms": 0.0871363203972578, "mean_ms": 0.08675861443082492, "n": 300}, {"repeat": 2, "p50_ms": 0.0865280032157898, "p95_ms": 0.08687999844551086, "p99_ms": 0.08717280060052872, "mean_ms": 0.08677920078237851, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_2419ddf76b22ab88_sdpa_efficient.json", "configuration": "480x832_tp4_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp4_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp4_sp1_v1/captures/qkv_rank0_2419ddf76b22ab88.pt", "context": {"phase": "commit", "step": null, "start_frame": 9, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_cudnn", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [512, 10, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.000244140625, "rmse": 2.018854664673764e-06, "relative_l2": 0.0012680831909262574, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 8.805995746343637e-05, "rmse": 5.095378225942453e-06, "relative_l2": 0.0020053045070645903, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.04859200119972229, "p95_ms": 0.05171999894082546, "p99_ms": 0.05475455969572067, "mean_ms": 0.04903637328495582, "n": 300}, {"repeat": 1, "p50_ms": 0.048496000468730927, "p95_ms": 0.05161759816110134, "p99_ms": 0.05429311975836752, "mean_ms": 0.04912021326522032, "n": 300}, {"repeat": 2, "p50_ms": 0.04864000156521797, "p95_ms": 0.05498400125652552, "p99_ms": 0.06382528156042097, "mean_ms": 0.04959498651325703, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_2419ddf76b22ab88_sdpa_cudnn.json", "configuration": "480x832_tp4_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp4_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp4_sp1_v1/captures/qkv_rank0_2419ddf76b22ab88.pt", "context": {"phase": "commit", "step": null, "start_frame": 9, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "native_dense", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [512, 10, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.0, "rmse": 0.0, "relative_l2": 0.0, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 8.805995746343637e-05, "rmse": 4.632311077919211e-06, "relative_l2": 0.001823062758203484, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.10814399644732475, "p95_ms": 0.1151999980211258, "p99_ms": 0.1340854461491106, "mean_ms": 0.11035093322396278, "n": 300}, {"repeat": 1, "p50_ms": 0.10732799768447876, "p95_ms": 0.11443359963595867, "p99_ms": 0.135712644457817, "mean_ms": 0.1090238935003678, "n": 300}, {"repeat": 2, "p50_ms": 0.10966400057077408, "p95_ms": 0.1297647975385189, "p99_ms": 0.1403366401791569, "mean_ms": 0.11416479979952177, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_2419ddf76b22ab88_native_dense.json", "configuration": "480x832_tp4_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp4_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp4_sp1_v1/captures/qkv_rank0_3c2cc055c01fd729.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 27, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "vllm_fa2", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [29640, 10, 128], "q_len": 4680, "k_len": 28080, "reference_error": {"max_abs": 0.03125, "rmse": 0.0005543818963847688, "relative_l2": 0.00098866153892788, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.008747478513556395, "rmse": 0.0008088118590230983, "relative_l2": 0.001962001449875581, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 1.8912639617919922, "p95_ms": 1.9047279596328734, "p99_ms": 1.9142780315876007, "mean_ms": 1.893231143951416, "n": 300}, {"repeat": 1, "p50_ms": 1.8907840251922607, "p95_ms": 1.8987328231334686, "p99_ms": 1.9009933149814606, "mean_ms": 1.892348158756892, "n": 300}, {"repeat": 2, "p50_ms": 1.8913600444793701, "p95_ms": 1.9039039611816406, "p99_ms": 1.9048725938796998, "mean_ms": 1.8945703458786012, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_3c2cc055c01fd729_vllm_fa2.json", "configuration": "480x832_tp4_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp4_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp4_sp1_v1/captures/qkv_rank0_3c2cc055c01fd729.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 27, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "vllm_fa3", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [29640, 10, 128], "q_len": 4680, "k_len": 28080, "reference_error": {"max_abs": 0.0, "rmse": 0.0, "relative_l2": 0.0, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.008747478513556395, "rmse": 0.0008127195606636765, "relative_l2": 0.001971480683147155, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 1.0253119468688965, "p95_ms": 1.036387187242508, "p99_ms": 1.0389155769348144, "mean_ms": 1.0124933338165283, "n": 300}, {"repeat": 1, "p50_ms": 1.028719961643219, "p95_ms": 1.0366719961166382, "p99_ms": 1.0394415426254273, "mean_ms": 1.025128747622172, "n": 300}, {"repeat": 2, "p50_ms": 1.0268479585647583, "p95_ms": 1.0671743988990783, "p99_ms": 1.0724223482608795, "mean_ms": 1.0297577595710754, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_3c2cc055c01fd729_vllm_fa3.json", "configuration": "480x832_tp4_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp4_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp4_sp1_v1/captures/qkv_rank0_3c2cc055c01fd729.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 27, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_math", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [29640, 10, 128], "q_len": 4680, "k_len": 28080, "reference_error": {"max_abs": 0.03125, "rmse": 0.000978205889105756, "relative_l2": 0.0017444915608145798, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.007718829491974777, "rmse": 0.0006850467181283888, "relative_l2": 0.0016617741681282927, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 27.925888061523438, "p95_ms": 28.07226209640503, "p99_ms": 28.10127914428711, "mean_ms": 27.95511613845825, "n": 300}, {"repeat": 1, "p50_ms": 28.053135871887207, "p95_ms": 28.08144474029541, "p99_ms": 28.148796043395997, "mean_ms": 28.022646630605063, "n": 300}, {"repeat": 2, "p50_ms": 28.065872192382812, "p95_ms": 28.290551567077635, "p99_ms": 28.429820671081544, "mean_ms": 28.10470184326172, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_3c2cc055c01fd729_sdpa_math.json", "configuration": "480x832_tp4_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp4_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp4_sp1_v1/captures/qkv_rank0_3c2cc055c01fd729.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 27, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_flash", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [29640, 10, 128], "q_len": 4680, "k_len": 28080, "reference_error": {"max_abs": 0.03125, "rmse": 0.0005551435483549316, "relative_l2": 0.0009900198372667314, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.008747478513556395, "rmse": 0.0008088108642645339, "relative_l2": 0.0019619990368079138, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 2.0505599975585938, "p95_ms": 2.071105647087097, "p99_ms": 2.075455904006958, "mean_ms": 2.0551954118410745, "n": 300}, {"repeat": 1, "p50_ms": 2.0638400316238403, "p95_ms": 2.079750490188599, "p99_ms": 2.08361665725708, "mean_ms": 2.0610984547932945, "n": 300}, {"repeat": 2, "p50_ms": 2.0498080253601074, "p95_ms": 2.056547260284424, "p99_ms": 2.06893696308136, "mean_ms": 2.051571096579234, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_3c2cc055c01fd729_sdpa_flash.json", "configuration": "480x832_tp4_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp4_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp4_sp1_v1/captures/qkv_rank0_3c2cc055c01fd729.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 27, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_efficient", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [29640, 10, 128], "q_len": 4680, "k_len": 28080, "reference_error": {"max_abs": 0.03125, "rmse": 0.0007702753997930319, "relative_l2": 0.0013736770033867033, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.008747478513556395, "rmse": 0.0008122163293225577, "relative_l2": 0.001970259953493054, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 4.1219518184661865, "p95_ms": 4.150755262374878, "p99_ms": 4.157055854797363, "mean_ms": 4.119360860188802, "n": 300}, {"repeat": 1, "p50_ms": 4.121551990509033, "p95_ms": 4.151153516769409, "p99_ms": 4.1572167110443115, "mean_ms": 4.120078830718994, "n": 300}, {"repeat": 2, "p50_ms": 4.117583990097046, "p95_ms": 4.153633403778076, "p99_ms": 4.161459417343139, "mean_ms": 4.119071882565816, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_3c2cc055c01fd729_sdpa_efficient.json", "configuration": "480x832_tp4_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp4_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp4_sp1_v1/captures/qkv_rank0_3c2cc055c01fd729.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 27, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_cudnn", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [29640, 10, 128], "q_len": 4680, "k_len": 28080, "reference_error": {"max_abs": 0.03125, "rmse": 0.0007704233346640188, "relative_l2": 0.001373940824262105, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.008747478513556395, "rmse": 0.0008108240198697908, "relative_l2": 0.001966882513938318, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 1.1981279850006104, "p95_ms": 1.2096528351306914, "p99_ms": 1.2112978827953338, "mean_ms": 1.1942095983028411, "n": 300}, {"repeat": 1, "p50_ms": 1.1977760195732117, "p95_ms": 1.203164768218994, "p99_ms": 1.209124834537506, "mean_ms": 1.1940574955940246, "n": 300}, {"repeat": 2, "p50_ms": 1.2341279983520508, "p95_ms": 1.4119999885559085, "p99_ms": 1.448574105501175, "mean_ms": 1.2651662921905518, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_3c2cc055c01fd729_sdpa_cudnn.json", "configuration": "480x832_tp4_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp4_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp4_sp1_v1/captures/qkv_rank0_3c2cc055c01fd729.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 27, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "native_dense", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [29640, 10, 128], "q_len": 4680, "k_len": 28080, "reference_error": {"max_abs": 0.03125, "rmse": 0.0005155979394430323, "relative_l2": 0.0009194958486234532, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.007718829491974777, "rmse": 0.0008154557272397831, "relative_l2": 0.001978118027456011, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 1.0063999891281128, "p95_ms": 1.0235231935977935, "p99_ms": 1.0244911849498748, "mean_ms": 0.9962788261969884, "n": 300}, {"repeat": 1, "p50_ms": 1.0083200335502625, "p95_ms": 1.0170623898506164, "p99_ms": 1.0213506674766542, "mean_ms": 1.0042486411333085, "n": 300}, {"repeat": 2, "p50_ms": 1.0542400479316711, "p95_ms": 1.080524843931198, "p99_ms": 1.0906889498233796, "mean_ms": 1.0474461819728216, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_3c2cc055c01fd729_native_dense.json", "configuration": "480x832_tp4_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp4_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp4_sp1_v1/captures/qkv_rank0_59302977417c4056.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 0, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "vllm_fa2", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [29640, 10, 128], "q_len": 4680, "k_len": 4680, "reference_error": {"max_abs": 0.03125, "rmse": 0.000526472702266815, "relative_l2": 0.0009064506683881809, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.00695011314270122, "rmse": 0.0007533507498643245, "relative_l2": 0.0019170394086784786, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.3309440016746521, "p95_ms": 0.3319680094718933, "p99_ms": 0.33232032477855683, "mean_ms": 0.33118389318386715, "n": 300}, {"repeat": 1, "p50_ms": 0.33105599880218506, "p95_ms": 0.3321936041116714, "p99_ms": 0.33251391142606734, "mean_ms": 0.3315531732638677, "n": 300}, {"repeat": 2, "p50_ms": 0.33132800459861755, "p95_ms": 0.3328032001852989, "p99_ms": 0.33331296473741534, "mean_ms": 0.3317954125006994, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_59302977417c4056_vllm_fa2.json", "configuration": "480x832_tp4_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp4_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp4_sp1_v1/captures/qkv_rank0_59302977417c4056.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 0, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "vllm_fa3", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [29640, 10, 128], "q_len": 4680, "k_len": 4680, "reference_error": {"max_abs": 0.0, "rmse": 0.0, "relative_l2": 0.0, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.00695011314270122, "rmse": 0.0007627110307544137, "relative_l2": 0.0019408583633232206, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.17390400171279907, "p95_ms": 0.184225594997406, "p99_ms": 0.18483264669775962, "mean_ms": 0.17719018667936326, "n": 300}, {"repeat": 1, "p50_ms": 0.1820319965481758, "p95_ms": 0.1900479942560196, "p99_ms": 0.19037055671215058, "mean_ms": 0.18483861292401949, "n": 300}, {"repeat": 2, "p50_ms": 0.18041600286960602, "p95_ms": 0.1868479996919632, "p99_ms": 0.18726559549570085, "mean_ms": 0.18202730630834899, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_59302977417c4056_vllm_fa3.json", "configuration": "480x832_tp4_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp4_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp4_sp1_v1/captures/qkv_rank0_59302977417c4056.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 0, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_math", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [29640, 10, 128], "q_len": 4680, "k_len": 4680, "reference_error": {"max_abs": 0.03125, "rmse": 0.000867680198247666, "relative_l2": 0.0014939222722514215, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.00695011314270122, "rmse": 0.0006602535380158555, "relative_l2": 0.0016801364468326902, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 4.800191879272461, "p95_ms": 4.827630615234375, "p99_ms": 4.830411190986633, "mean_ms": 4.804299834569295, "n": 300}, {"repeat": 1, "p50_ms": 4.799680233001709, "p95_ms": 4.804903888702393, "p99_ms": 4.806277151107788, "mean_ms": 4.800761493047078, "n": 300}, {"repeat": 2, "p50_ms": 4.799615859985352, "p95_ms": 4.804774475097656, "p99_ms": 4.808065905570984, "mean_ms": 4.80061323483785, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_59302977417c4056_sdpa_math.json", "configuration": "480x832_tp4_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp4_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp4_sp1_v1/captures/qkv_rank0_59302977417c4056.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 0, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_flash", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [29640, 10, 128], "q_len": 4680, "k_len": 4680, "reference_error": {"max_abs": 0.03125, "rmse": 0.0005269034847708496, "relative_l2": 0.0009071923651314894, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.00695011314270122, "rmse": 0.0007533551128110181, "relative_l2": 0.001917050510998017, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.3569279909133911, "p95_ms": 0.3579519987106323, "p99_ms": 0.358246728181839, "mean_ms": 0.35689418703317644, "n": 300}, {"repeat": 1, "p50_ms": 0.3564479947090149, "p95_ms": 0.3574720025062561, "p99_ms": 0.357856325507164, "mean_ms": 0.35676245371500653, "n": 300}, {"repeat": 2, "p50_ms": 0.35654398798942566, "p95_ms": 0.3573808088898659, "p99_ms": 0.35782559782266615, "mean_ms": 0.35692629287640254, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_59302977417c4056_sdpa_flash.json", "configuration": "480x832_tp4_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp4_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp4_sp1_v1/captures/qkv_rank0_59302977417c4056.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 0, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_efficient", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [29640, 10, 128], "q_len": 4680, "k_len": 4680, "reference_error": {"max_abs": 0.03125, "rmse": 0.0007440478283794388, "relative_l2": 0.001281059109890015, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.00695011314270122, "rmse": 0.0007546915234369167, "relative_l2": 0.0019204512533965421, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.6986879706382751, "p95_ms": 0.7019855886697769, "p99_ms": 0.704482552409172, "mean_ms": 0.6992696529626846, "n": 300}, {"repeat": 1, "p50_ms": 0.7005119919776917, "p95_ms": 0.7072335988283157, "p99_ms": 0.7091241359710694, "mean_ms": 0.7018374399344126, "n": 300}, {"repeat": 2, "p50_ms": 0.7004320025444031, "p95_ms": 0.7078767746686936, "p99_ms": 0.7100822675228119, "mean_ms": 0.7018209058046341, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_59302977417c4056_sdpa_efficient.json", "configuration": "480x832_tp4_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp4_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp4_sp1_v1/captures/qkv_rank0_59302977417c4056.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 0, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_cudnn", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [29640, 10, 128], "q_len": 4680, "k_len": 4680, "reference_error": {"max_abs": 0.03125, "rmse": 0.0007433142273059395, "relative_l2": 0.0012797960374067874, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.00695011314270122, "rmse": 0.0007544806469925304, "relative_l2": 0.0019199146395359687, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.19808000326156616, "p95_ms": 0.21379999667406083, "p99_ms": 0.21452959448099138, "mean_ms": 0.20095669344067574, "n": 300}, {"repeat": 1, "p50_ms": 0.2048639953136444, "p95_ms": 0.21385759711265565, "p99_ms": 0.21494592502713203, "mean_ms": 0.2070668796201547, "n": 300}, {"repeat": 2, "p50_ms": 0.2093760073184967, "p95_ms": 0.21040000021457672, "p99_ms": 0.2116483221948147, "mean_ms": 0.2097646934290727, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_59302977417c4056_sdpa_cudnn.json", "configuration": "480x832_tp4_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp4_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp4_sp1_v1/captures/qkv_rank0_59302977417c4056.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 0, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "native_dense", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [29640, 10, 128], "q_len": 4680, "k_len": 4680, "reference_error": {"max_abs": 0.03125, "rmse": 0.00048785716738888207, "relative_l2": 0.0008399646430927424, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.00695011314270122, "rmse": 0.0007607064604762229, "relative_l2": 0.0019357573659173654, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.16228800266981125, "p95_ms": 0.17292800545692444, "p99_ms": 0.17334432110190393, "mean_ms": 0.16557994683583577, "n": 300}, {"repeat": 1, "p50_ms": 0.1716800034046173, "p95_ms": 0.17715359330177308, "p99_ms": 0.17820991352200508, "mean_ms": 0.174168960104386, "n": 300}, {"repeat": 2, "p50_ms": 0.17393600195646286, "p95_ms": 0.175233593583107, "p99_ms": 0.17590432569384576, "mean_ms": 0.1736502399543921, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_59302977417c4056_native_dense.json", "configuration": "480x832_tp4_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp4_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp4_sp1_v1/captures/qkv_rank0_6bef1f209a2b6a6e.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 0, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "vllm_fa2", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [512, 10, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.0009765625, "rmse": 1.945037827297859e-06, "relative_l2": 0.0002824420512182635, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.0006212401684354552, "rmse": 2.205267103985341e-05, "relative_l2": 0.002220479845167307, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.0504320003092289, "p95_ms": 0.054180799052119255, "p99_ms": 0.05626207996159791, "mean_ms": 0.05093717340379953, "n": 300}, {"repeat": 1, "p50_ms": 0.05052800104022026, "p95_ms": 0.056006401591002945, "p99_ms": 0.21452799439430237, "mean_ms": 0.05449898647765319, "n": 300}, {"repeat": 2, "p50_ms": 0.050255998969078064, "p95_ms": 0.05532800033688545, "p99_ms": 0.06569568030536166, "mean_ms": 0.05127722668151061, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_6bef1f209a2b6a6e_vllm_fa2.json", "configuration": "480x832_tp4_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp4_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp4_sp1_v1/captures/qkv_rank0_6bef1f209a2b6a6e.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 0, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "vllm_fa3", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [512, 10, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.00048828125, "rmse": 6.847262505770303e-07, "relative_l2": 9.943019309019894e-05, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.0006212401684354552, "rmse": 2.2052666034902135e-05, "relative_l2": 0.0022204793412195804, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.060912000015378, "p95_ms": 0.06720479764044285, "p99_ms": 0.07574816301465034, "mean_ms": 0.06172597323854764, "n": 300}, {"repeat": 1, "p50_ms": 0.061103999614715576, "p95_ms": 0.06823039799928665, "p99_ms": 0.07950720153748984, "mean_ms": 0.062441066714624564, "n": 300}, {"repeat": 2, "p50_ms": 0.06083200126886368, "p95_ms": 0.06624319776892662, "p99_ms": 0.07565312229096889, "mean_ms": 0.0619287467127045, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_6bef1f209a2b6a6e_vllm_fa3.json", "configuration": "480x832_tp4_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp4_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp4_sp1_v1/captures/qkv_rank0_6bef1f209a2b6a6e.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 0, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_math", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [512, 10, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.001953125, "rmse": 1.4708645946835768e-05, "relative_l2": 0.002135865983459511, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.0004448218038644869, "rmse": 1.7369049942780413e-05, "relative_l2": 0.0017488868018730773, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.6168000102043152, "p95_ms": 0.6190784186124801, "p99_ms": 0.6202249592542648, "mean_ms": 0.6173402663071951, "n": 300}, {"repeat": 1, "p50_ms": 0.6168160140514374, "p95_ms": 0.6193616271018982, "p99_ms": 0.6201635038852692, "mean_ms": 0.6173204267024994, "n": 300}, {"repeat": 2, "p50_ms": 0.616864025592804, "p95_ms": 0.6189119815826416, "p99_ms": 0.6201951915025711, "mean_ms": 0.6173470934232076, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_6bef1f209a2b6a6e_sdpa_math.json", "configuration": "480x832_tp4_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp4_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp4_sp1_v1/captures/qkv_rank0_6bef1f209a2b6a6e.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 0, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_flash", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [512, 10, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.0009765625, "rmse": 1.9453377348402356e-06, "relative_l2": 0.0002824856012717668, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.0006212401684354552, "rmse": 2.205267103985341e-05, "relative_l2": 0.002220479845167307, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.049056001007556915, "p95_ms": 0.049375999718904495, "p99_ms": 0.04995583988726133, "mean_ms": 0.04919989344974359, "n": 300}, {"repeat": 1, "p50_ms": 0.04902400076389313, "p95_ms": 0.04934399947524071, "p99_ms": 0.05000095963478075, "mean_ms": 0.04938720007737478, "n": 300}, {"repeat": 2, "p50_ms": 0.04879999905824661, "p95_ms": 0.04915200173854828, "p99_ms": 0.054044160991907036, "mean_ms": 0.04917909312993288, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_6bef1f209a2b6a6e_sdpa_flash.json", "configuration": "480x832_tp4_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp4_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp4_sp1_v1/captures/qkv_rank0_6bef1f209a2b6a6e.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 0, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_efficient", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [512, 10, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.0009765625, "rmse": 3.158492816002431e-06, "relative_l2": 0.00045864978932014526, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.0006212401684354552, "rmse": 2.2159043356248427e-05, "relative_l2": 0.002231190456331462, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.08640000224113464, "p95_ms": 0.08678559772670269, "p99_ms": 0.08691295869648456, "mean_ms": 0.08653280121584733, "n": 300}, {"repeat": 1, "p50_ms": 0.08649600297212601, "p95_ms": 0.08691359870135784, "p99_ms": 0.08716800063848495, "mean_ms": 0.08673408101002375, "n": 300}, {"repeat": 2, "p50_ms": 0.08656000345945358, "p95_ms": 0.08687999844551086, "p99_ms": 0.08710464015603066, "mean_ms": 0.08686634744207064, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_6bef1f209a2b6a6e_sdpa_efficient.json", "configuration": "480x832_tp4_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp4_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp4_sp1_v1/captures/qkv_rank0_6bef1f209a2b6a6e.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 0, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_cudnn", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [512, 10, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.0009765625, "rmse": 3.149411513524098e-06, "relative_l2": 0.000457331078874664, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.0006212401684354552, "rmse": 2.215904335471974e-05, "relative_l2": 0.0022311904561775385, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.048239998519420624, "p95_ms": 0.05238560028374195, "p99_ms": 0.06071968004107471, "mean_ms": 0.04905855999638637, "n": 300}, {"repeat": 1, "p50_ms": 0.048128001391887665, "p95_ms": 0.05223039910197258, "p99_ms": 0.05613311897963285, "mean_ms": 0.048925226715703804, "n": 300}, {"repeat": 2, "p50_ms": 0.04816000163555145, "p95_ms": 0.05169919878244401, "p99_ms": 0.05552864152938124, "mean_ms": 0.04895861320197582, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_6bef1f209a2b6a6e_sdpa_cudnn.json", "configuration": "480x832_tp4_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp4_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp4_sp1_v1/captures/qkv_rank0_6bef1f209a2b6a6e.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 0, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "native_dense", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [512, 10, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.0, "rmse": 0.0, "relative_l2": 0.0, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.0006212401684354552, "rmse": 2.2056216380079165e-05, "relative_l2": 0.002220836824895582, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.10844799876213074, "p95_ms": 0.11693279631435871, "p99_ms": 0.15387615725398007, "mean_ms": 0.1113432532052199, "n": 300}, {"repeat": 1, "p50_ms": 0.10743999853730202, "p95_ms": 0.11590560339391232, "p99_ms": 0.1249353566765784, "mean_ms": 0.10927936000128587, "n": 300}, {"repeat": 2, "p50_ms": 0.10734399780631065, "p95_ms": 0.11475520208477974, "p99_ms": 0.12004447735846038, "mean_ms": 0.10877546680470308, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_6bef1f209a2b6a6e_native_dense.json", "configuration": "480x832_tp4_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp4_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp4_sp1_v1/captures/qkv_rank0_6c601440672149ef.pt", "context": {"phase": "commit", "step": null, "start_frame": 0, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "vllm_fa2", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [512, 10, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.0009765625, "rmse": 2.015030516804163e-06, "relative_l2": 0.0002887277325577258, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.00024057685809458773, "rmse": 1.7912717224201285e-05, "relative_l2": 0.0016349277449488405, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.05027199909090996, "p95_ms": 0.057284800335764885, "p99_ms": 0.062142079807817906, "mean_ms": 0.051162773445248604, "n": 300}, {"repeat": 1, "p50_ms": 0.05020799860358238, "p95_ms": 0.05692800134420395, "p99_ms": 0.06370016038417811, "mean_ms": 0.05138528001805147, "n": 300}, {"repeat": 2, "p50_ms": 0.0504320003092289, "p95_ms": 0.06435359828174116, "p99_ms": 0.07768480248749252, "mean_ms": 0.052753813328842325, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_6c601440672149ef_vllm_fa2.json", "configuration": "480x832_tp4_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp4_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp4_sp1_v1/captures/qkv_rank0_6c601440672149ef.pt", "context": {"phase": "commit", "step": null, "start_frame": 0, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "vllm_fa3", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [512, 10, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.0009765625, "rmse": 8.940885961950505e-07, "relative_l2": 0.00012811129704106908, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.00024057685809458773, "rmse": 1.792405049922453e-05, "relative_l2": 0.0016359621544995925, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.0618400014936924, "p95_ms": 0.07358400113880635, "p99_ms": 0.08896383866667738, "mean_ms": 0.06671264001478751, "n": 300}, {"repeat": 1, "p50_ms": 0.061535999178886414, "p95_ms": 0.07047679945826534, "p99_ms": 0.09146912343800066, "mean_ms": 0.06335040009270111, "n": 300}, {"repeat": 2, "p50_ms": 0.06088000163435936, "p95_ms": 0.06826079823076725, "p99_ms": 0.07749472133815287, "mean_ms": 0.062360640031596026, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_6c601440672149ef_vllm_fa3.json", "configuration": "480x832_tp4_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp4_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp4_sp1_v1/captures/qkv_rank0_6c601440672149ef.pt", "context": {"phase": "commit", "step": null, "start_frame": 0, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_math", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [512, 10, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.0009765625, "rmse": 5.874370631016668e-06, "relative_l2": 0.0008417211046446796, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.00024057685809458773, "rmse": 1.7798095759714057e-05, "relative_l2": 0.0016244660260420264, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.6167359948158264, "p95_ms": 0.6189471900463104, "p99_ms": 0.6200006449222565, "mean_ms": 0.6171113608280817, "n": 300}, {"repeat": 1, "p50_ms": 0.6152960062026978, "p95_ms": 0.6176959872245789, "p99_ms": 0.6188806140422821, "mean_ms": 0.6159454931815466, "n": 300}, {"repeat": 2, "p50_ms": 0.6161120235919952, "p95_ms": 0.6182143926620484, "p99_ms": 0.619202231168747, "mean_ms": 0.6166377592086792, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_6c601440672149ef_sdpa_math.json", "configuration": "480x832_tp4_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp4_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp4_sp1_v1/captures/qkv_rank0_6c601440672149ef.pt", "context": {"phase": "commit", "step": null, "start_frame": 0, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_flash", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [512, 10, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.0009765625, "rmse": 2.013502667932976e-06, "relative_l2": 0.0002885088116348962, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.00024057685809458773, "rmse": 1.7912717224201285e-05, "relative_l2": 0.0016349277449488405, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.0488319993019104, "p95_ms": 0.04915360156446695, "p99_ms": 0.049587520919740155, "mean_ms": 0.04902805327127377, "n": 300}, {"repeat": 1, "p50_ms": 0.0490880012512207, "p95_ms": 0.05749760158360005, "p99_ms": 0.0783670404553413, "mean_ms": 0.051576426550745964, "n": 300}, {"repeat": 2, "p50_ms": 0.0488319993019104, "p95_ms": 0.04915200173854828, "p99_ms": 0.052501760981976967, "mean_ms": 0.04919765330851078, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_6c601440672149ef_sdpa_flash.json", "configuration": "480x832_tp4_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp4_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp4_sp1_v1/captures/qkv_rank0_6c601440672149ef.pt", "context": {"phase": "commit", "step": null, "start_frame": 0, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_efficient", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [512, 10, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.0009765625, "rmse": 3.4885260618872947e-06, "relative_l2": 0.0004998605288691727, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.00024057685809458773, "rmse": 1.8041896123651594e-05, "relative_l2": 0.001646718148611771, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.08656000345945358, "p95_ms": 0.08691199868917465, "p99_ms": 0.08707231990993022, "mean_ms": 0.08667669395605723, "n": 300}, {"repeat": 1, "p50_ms": 0.0865280032157898, "p95_ms": 0.08691199868917465, "p99_ms": 0.08710528016090394, "mean_ms": 0.08679722746213277, "n": 300}, {"repeat": 2, "p50_ms": 0.08646400272846222, "p95_ms": 0.08684799820184708, "p99_ms": 0.08694463893771172, "mean_ms": 0.08669066749513149, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_6c601440672149ef_sdpa_efficient.json", "configuration": "480x832_tp4_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp4_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp4_sp1_v1/captures/qkv_rank0_6c601440672149ef.pt", "context": {"phase": "commit", "step": null, "start_frame": 0, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_cudnn", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [512, 10, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.0009765625, "rmse": 3.4767848529602677e-06, "relative_l2": 0.0004981781659457746, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.00024057685809458773, "rmse": 1.804193588379445e-05, "relative_l2": 0.0016467217775955755, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.04904000088572502, "p95_ms": 0.05206560157239437, "p99_ms": 0.05756895877420902, "mean_ms": 0.04964202661067248, "n": 300}, {"repeat": 1, "p50_ms": 0.04918399825692177, "p95_ms": 0.06251039914786816, "p99_ms": 0.09369727820158005, "mean_ms": 0.05117205334206422, "n": 300}, {"repeat": 2, "p50_ms": 0.04862400144338608, "p95_ms": 0.051873600110411644, "p99_ms": 0.05517343912273645, "mean_ms": 0.04930698651820421, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_6c601440672149ef_sdpa_cudnn.json", "configuration": "480x832_tp4_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp4_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp4_sp1_v1/captures/qkv_rank0_6c601440672149ef.pt", "context": {"phase": "commit", "step": null, "start_frame": 0, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "native_dense", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [512, 10, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.0, "rmse": 0.0, "relative_l2": 0.0, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.00024057685809458773, "rmse": 1.7926271422383447e-05, "relative_l2": 0.0016361648623772678, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.10927999764680862, "p95_ms": 0.11923519894480705, "p99_ms": 0.13402687951922412, "mean_ms": 0.11114218672116598, "n": 300}, {"repeat": 1, "p50_ms": 0.10847999900579453, "p95_ms": 0.11517439782619478, "p99_ms": 0.14759200364351244, "mean_ms": 0.11195253344873587, "n": 300}, {"repeat": 2, "p50_ms": 0.10792000219225883, "p95_ms": 0.11485120244324208, "p99_ms": 0.12612352579832067, "mean_ms": 0.10969184006253878, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_6c601440672149ef_native_dense.json", "configuration": "480x832_tp4_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp4_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp4_sp1_v1/captures/qkv_rank0_7baca46400faa6d9.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 27, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "vllm_fa2", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [512, 10, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.0001220703125, "rmse": 1.2188489496307559e-06, "relative_l2": 0.00074601842769162, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.00016451503224817338, "rmse": 6.681481814425281e-06, "relative_l2": 0.0019350013039422253, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.050592001527547836, "p95_ms": 0.05615359898656607, "p99_ms": 0.07119871728122223, "mean_ms": 0.05163424002627532, "n": 300}, {"repeat": 1, "p50_ms": 0.050912000238895416, "p95_ms": 0.0561055988073349, "p99_ms": 0.06886560276150704, "mean_ms": 0.05192416009803613, "n": 300}, {"repeat": 2, "p50_ms": 0.050464000552892685, "p95_ms": 0.055369600467383866, "p99_ms": 0.06539615765213953, "mean_ms": 0.051424746699631214, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_7baca46400faa6d9_vllm_fa2.json", "configuration": "480x832_tp4_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp4_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp4_sp1_v1/captures/qkv_rank0_7baca46400faa6d9.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 27, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "vllm_fa3", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [512, 10, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 6.103515625e-05, "rmse": 4.5075680516825706e-07, "relative_l2": 0.0002758938120796618, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.00016451503224817338, "rmse": 6.699012323812996e-06, "relative_l2": 0.0019400782553530285, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.061039999127388, "p95_ms": 0.06714239716529846, "p99_ms": 0.0782217587530612, "mean_ms": 0.062298666598896185, "n": 300}, {"repeat": 1, "p50_ms": 0.06054399907588959, "p95_ms": 0.06823679804801941, "p99_ms": 0.07769600205123413, "mean_ms": 0.062042773465315504, "n": 300}, {"repeat": 2, "p50_ms": 0.060864001512527466, "p95_ms": 0.06694400310516357, "p99_ms": 0.079287679567933, "mean_ms": 0.06229568000882864, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_7baca46400faa6d9_vllm_fa3.json", "configuration": "480x832_tp4_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp4_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp4_sp1_v1/captures/qkv_rank0_7baca46400faa6d9.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 27, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_math", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [512, 10, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.00048828125, "rmse": 3.184050713576812e-06, "relative_l2": 0.0019488555228706207, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.00012157623192320804, "rmse": 5.806839647996872e-06, "relative_l2": 0.001681699150388822, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.617792010307312, "p95_ms": 0.6200960278511047, "p99_ms": 0.621255686879158, "mean_ms": 0.6182322150468826, "n": 300}, {"repeat": 1, "p50_ms": 0.617247998714447, "p95_ms": 0.6196479797363281, "p99_ms": 0.6218902218341827, "mean_ms": 0.6178693332274755, "n": 300}, {"repeat": 2, "p50_ms": 0.6176319718360901, "p95_ms": 0.6201615840196609, "p99_ms": 0.6207686507701874, "mean_ms": 0.6180513081947963, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_7baca46400faa6d9_sdpa_math.json", "configuration": "480x832_tp4_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp4_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp4_sp1_v1/captures/qkv_rank0_7baca46400faa6d9.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 27, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_flash", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [512, 10, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.0001220703125, "rmse": 1.2202856731540554e-06, "relative_l2": 0.0007468978001718637, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.00016451503224817338, "rmse": 6.681481814425281e-06, "relative_l2": 0.0019350013039422253, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.04879999905824661, "p95_ms": 0.049057601019740105, "p99_ms": 0.050992640703916535, "mean_ms": 0.04899210662891467, "n": 300}, {"repeat": 1, "p50_ms": 0.04879999905824661, "p95_ms": 0.04902400076389313, "p99_ms": 0.05401120126247404, "mean_ms": 0.04920757322261731, "n": 300}, {"repeat": 2, "p50_ms": 0.04879999905824661, "p95_ms": 0.04912000149488449, "p99_ms": 0.06633663803338999, "mean_ms": 0.0492293332144618, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_7baca46400faa6d9_sdpa_flash.json", "configuration": "480x832_tp4_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp4_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp4_sp1_v1/captures/qkv_rank0_7baca46400faa6d9.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 27, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_efficient", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [512, 10, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.000244140625, "rmse": 2.1510412794390267e-06, "relative_l2": 0.0013165835140383996, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.00016451503224817338, "rmse": 7.003250049623255e-06, "relative_l2": 0.002028187512027239, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.08656000345945358, "p95_ms": 0.08684959821403027, "p99_ms": 0.08704095967113971, "mean_ms": 0.08666933397452037, "n": 300}, {"repeat": 1, "p50_ms": 0.0865280032157898, "p95_ms": 0.08681599795818329, "p99_ms": 0.0870729599148035, "mean_ms": 0.0867828276505073, "n": 300}, {"repeat": 2, "p50_ms": 0.0862400010228157, "p95_ms": 0.08662399649620056, "p99_ms": 0.08678431771695613, "mean_ms": 0.08652234750489395, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_7baca46400faa6d9_sdpa_efficient.json", "configuration": "480x832_tp4_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp4_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp4_sp1_v1/captures/qkv_rank0_7baca46400faa6d9.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 27, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_cudnn", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [512, 10, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.000244140625, "rmse": 2.151259588908548e-06, "relative_l2": 0.0013167171342767824, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.00016451503224817338, "rmse": 7.0029824500603725e-06, "relative_l2": 0.002028110013424759, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.04831999912858009, "p95_ms": 0.05113759823143482, "p99_ms": 0.054951040893793074, "mean_ms": 0.04870325338095426, "n": 300}, {"repeat": 1, "p50_ms": 0.048448000103235245, "p95_ms": 0.05167999863624573, "p99_ms": 0.055021121613681306, "mean_ms": 0.0490454400703311, "n": 300}, {"repeat": 2, "p50_ms": 0.04833599925041199, "p95_ms": 0.0515920015051961, "p99_ms": 0.05443360093981028, "mean_ms": 0.048894506531457105, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_7baca46400faa6d9_sdpa_cudnn.json", "configuration": "480x832_tp4_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp4_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp4_sp1_v1/captures/qkv_rank0_7baca46400faa6d9.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 27, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "native_dense", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [512, 10, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.0, "rmse": 0.0, "relative_l2": 0.0, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.00016451503224817338, "rmse": 6.709461001777485e-06, "relative_l2": 0.0019431042615665165, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.10787200182676315, "p95_ms": 0.11641919985413551, "p99_ms": 0.12487392298877222, "mean_ms": 0.10930591995517412, "n": 300}, {"repeat": 1, "p50_ms": 0.10718399658799171, "p95_ms": 0.11735359951853752, "p99_ms": 0.13176511585712414, "mean_ms": 0.10979584018389384, "n": 300}, {"repeat": 2, "p50_ms": 0.10684800148010254, "p95_ms": 0.11465440094470979, "p99_ms": 0.11858368098735804, "mean_ms": 0.10843178674578667, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_7baca46400faa6d9_native_dense.json", "configuration": "480x832_tp4_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp4_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp4_sp1_v1/captures/qkv_rank0_8e3efe1e7972bcf2.pt", "context": {"phase": "commit", "step": null, "start_frame": 0, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "vllm_fa2", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [29640, 10, 128], "q_len": 4680, "k_len": 4680, "reference_error": {"max_abs": 0.03125, "rmse": 0.0004175296662179905, "relative_l2": 0.0005544075530967276, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.013541072826506628, "rmse": 0.0008567744749459827, "relative_l2": 0.0017210073280176771, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.330704003572464, "p95_ms": 0.3327423855662346, "p99_ms": 0.33385630667209626, "mean_ms": 0.3310051193833351, "n": 300}, {"repeat": 1, "p50_ms": 0.3301439881324768, "p95_ms": 0.331167995929718, "p99_ms": 0.33171232730150224, "mean_ms": 0.33065397262573243, "n": 300}, {"repeat": 2, "p50_ms": 0.3298560082912445, "p95_ms": 0.3312320113182068, "p99_ms": 0.33158752650022505, "mean_ms": 0.33043615927298864, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_8e3efe1e7972bcf2_vllm_fa2.json", "configuration": "480x832_tp4_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp4_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp4_sp1_v1/captures/qkv_rank0_8e3efe1e7972bcf2.pt", "context": {"phase": "commit", "step": null, "start_frame": 0, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "vllm_fa3", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [29640, 10, 128], "q_len": 4680, "k_len": 4680, "reference_error": {"max_abs": 0.0, "rmse": 0.0, "relative_l2": 0.0, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.013541072826506628, "rmse": 0.0008569466434402922, "relative_l2": 0.0017213531637644518, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.17209599912166595, "p95_ms": 0.18223999440670013, "p99_ms": 0.18256063684821128, "mean_ms": 0.17510389377673466, "n": 300}, {"repeat": 1, "p50_ms": 0.18135999888181686, "p95_ms": 0.18365120515227318, "p99_ms": 0.18422463461756705, "mean_ms": 0.1816379728416602, "n": 300}, {"repeat": 2, "p50_ms": 0.18164800107479095, "p95_ms": 0.18352000415325165, "p99_ms": 0.18390752702951432, "mean_ms": 0.1818431999286016, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_8e3efe1e7972bcf2_vllm_fa3.json", "configuration": "480x832_tp4_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp4_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp4_sp1_v1/captures/qkv_rank0_8e3efe1e7972bcf2.pt", "context": {"phase": "commit", "step": null, "start_frame": 0, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_math", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [29640, 10, 128], "q_len": 4680, "k_len": 4680, "reference_error": {"max_abs": 0.03125, "rmse": 0.0007314641225412281, "relative_l2": 0.0009712584929100634, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.013541072826506628, "rmse": 0.0008422989462288157, "relative_l2": 0.0016919302584647525, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 4.800127983093262, "p95_ms": 4.8274911403656, "p99_ms": 4.828932795524597, "mean_ms": 4.804021552403768, "n": 300}, {"repeat": 1, "p50_ms": 4.799504041671753, "p95_ms": 4.80394561290741, "p99_ms": 4.806687245368957, "mean_ms": 4.8001970179875695, "n": 300}, {"repeat": 2, "p50_ms": 4.799488067626953, "p95_ms": 4.804001498222351, "p99_ms": 4.806017889976501, "mean_ms": 4.799935560226441, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_8e3efe1e7972bcf2_sdpa_math.json", "configuration": "480x832_tp4_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp4_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp4_sp1_v1/captures/qkv_rank0_8e3efe1e7972bcf2.pt", "context": {"phase": "commit", "step": null, "start_frame": 0, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_flash", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [29640, 10, 128], "q_len": 4680, "k_len": 4680, "reference_error": {"max_abs": 0.03125, "rmse": 0.00041805821885231275, "relative_l2": 0.0005551093800479279, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.013541072826506628, "rmse": 0.0008567729107790241, "relative_l2": 0.0017210041860674012, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.3567200005054474, "p95_ms": 0.3579231917858124, "p99_ms": 0.3586892768740654, "mean_ms": 0.3568609063824018, "n": 300}, {"repeat": 1, "p50_ms": 0.3566400110721588, "p95_ms": 0.3577600121498108, "p99_ms": 0.3581775930523872, "mean_ms": 0.3567362131675084, "n": 300}, {"repeat": 2, "p50_ms": 0.3569279909133911, "p95_ms": 0.35814398527145386, "p99_ms": 0.3584643226861954, "mean_ms": 0.35730677356322604, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_8e3efe1e7972bcf2_sdpa_flash.json", "configuration": "480x832_tp4_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp4_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp4_sp1_v1/captures/qkv_rank0_8e3efe1e7972bcf2.pt", "context": {"phase": "commit", "step": null, "start_frame": 0, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_efficient", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [29640, 10, 128], "q_len": 4680, "k_len": 4680, "reference_error": {"max_abs": 0.03125, "rmse": 0.0005927677961845221, "relative_l2": 0.0007870936367563921, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.013541072826506628, "rmse": 0.0008525460457428071, "relative_l2": 0.0017125136603636216, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.700543999671936, "p95_ms": 0.7089359909296036, "p99_ms": 0.7108163452148437, "mean_ms": 0.7020328547557195, "n": 300}, {"repeat": 1, "p50_ms": 0.7008000016212463, "p95_ms": 0.7087728083133698, "p99_ms": 0.7104739207029342, "mean_ms": 0.7023751465479533, "n": 300}, {"repeat": 2, "p50_ms": 0.7001280188560486, "p95_ms": 0.706600022315979, "p99_ms": 0.7097974580526352, "mean_ms": 0.7015584001938502, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_8e3efe1e7972bcf2_sdpa_efficient.json", "configuration": "480x832_tp4_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp4_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp4_sp1_v1/captures/qkv_rank0_8e3efe1e7972bcf2.pt", "context": {"phase": "commit", "step": null, "start_frame": 0, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_cudnn", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [29640, 10, 128], "q_len": 4680, "k_len": 4680, "reference_error": {"max_abs": 0.03125, "rmse": 0.0005919354486733103, "relative_l2": 0.000785988422482166, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.013541072826506628, "rmse": 0.0008525376406746021, "relative_l2": 0.0017124967770595593, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.19737599790096283, "p95_ms": 0.20454399287700653, "p99_ms": 0.20489664301276206, "mean_ms": 0.19916789347926775, "n": 300}, {"repeat": 1, "p50_ms": 0.20467199385166168, "p95_ms": 0.2085472010076046, "p99_ms": 0.20892832383513452, "mean_ms": 0.20539157321055732, "n": 300}, {"repeat": 2, "p50_ms": 0.20297600328922272, "p95_ms": 0.20876799523830414, "p99_ms": 0.20941216006875038, "mean_ms": 0.20379328002532324, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_8e3efe1e7972bcf2_sdpa_cudnn.json", "configuration": "480x832_tp4_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp4_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp4_sp1_v1/captures/qkv_rank0_8e3efe1e7972bcf2.pt", "context": {"phase": "commit", "step": null, "start_frame": 0, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "native_dense", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [29640, 10, 128], "q_len": 4680, "k_len": 4680, "reference_error": {"max_abs": 0.03125, "rmse": 0.00037096169339030194, "relative_l2": 0.000492573298055807, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.013541072826506628, "rmse": 0.0008567202360163901, "relative_l2": 0.0017208983779987137, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.16096000373363495, "p95_ms": 0.16163360103964805, "p99_ms": 0.16278431534767152, "mean_ms": 0.16127317324280738, "n": 300}, {"repeat": 1, "p50_ms": 0.16246400028467178, "p95_ms": 0.17088160440325736, "p99_ms": 0.1712009671330452, "mean_ms": 0.16513066704074542, "n": 300}, {"repeat": 2, "p50_ms": 0.16790400445461273, "p95_ms": 0.1719679981470108, "p99_ms": 0.1724486343562603, "mean_ms": 0.16933077350258827, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_8e3efe1e7972bcf2_native_dense.json", "configuration": "480x832_tp4_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp4_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp4_sp1_v1/captures/qkv_rank0_8ec3a76708f5e241.pt", "context": {"phase": "commit", "step": null, "start_frame": 27, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "vllm_fa2", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [512, 10, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.0001220703125, "rmse": 1.1569382891652851e-06, "relative_l2": 0.0007760415004802649, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 6.247908239926173e-05, "rmse": 3.834695061417089e-06, "relative_l2": 0.0017870305491370995, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.0506879985332489, "p95_ms": 0.057792000472545624, "p99_ms": 0.07232319839298713, "mean_ms": 0.05297877326607704, "n": 300}, {"repeat": 1, "p50_ms": 0.050335999578237534, "p95_ms": 0.05597600135952235, "p99_ms": 0.06489952191710466, "mean_ms": 0.051352746672928336, "n": 300}, {"repeat": 2, "p50_ms": 0.05027199909090996, "p95_ms": 0.05532159991562368, "p99_ms": 0.05921183954924338, "mean_ms": 0.051097920015454296, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_8ec3a76708f5e241_vllm_fa2.json", "configuration": "480x832_tp4_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp4_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp4_sp1_v1/captures/qkv_rank0_8ec3a76708f5e241.pt", "context": {"phase": "commit", "step": null, "start_frame": 27, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "vllm_fa3", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [512, 10, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.0001220703125, "rmse": 4.281896385506505e-07, "relative_l2": 0.00028721750563782776, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 6.247908239926173e-05, "rmse": 3.895248297740976e-06, "relative_l2": 0.0018152493465712599, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.06159999966621399, "p95_ms": 0.07018079794943333, "p99_ms": 0.09161919727921486, "mean_ms": 0.0630199467142423, "n": 300}, {"repeat": 1, "p50_ms": 0.061503998935222626, "p95_ms": 0.07112159766256809, "p99_ms": 0.08425920031964773, "mean_ms": 0.06326698676993449, "n": 300}, {"repeat": 2, "p50_ms": 0.06115199998021126, "p95_ms": 0.0689919963479042, "p99_ms": 0.08081247687339782, "mean_ms": 0.062470933397610985, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_8ec3a76708f5e241_vllm_fa3.json", "configuration": "480x832_tp4_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp4_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp4_sp1_v1/captures/qkv_rank0_8ec3a76708f5e241.pt", "context": {"phase": "commit", "step": null, "start_frame": 27, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_math", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [512, 10, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.000244140625, "rmse": 2.277165845170775e-06, "relative_l2": 0.0015274584788819907, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 6.0727240677817956e-05, "rmse": 3.617202932177795e-06, "relative_l2": 0.0016856756635666513, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.6164799928665161, "p95_ms": 0.6188480257987976, "p99_ms": 0.6199062216281891, "mean_ms": 0.6167800535758337, "n": 300}, {"repeat": 1, "p50_ms": 0.6169599890708923, "p95_ms": 0.6192319989204407, "p99_ms": 0.6203229016065598, "mean_ms": 0.6174695483843485, "n": 300}, {"repeat": 2, "p50_ms": 0.6172800064086914, "p95_ms": 0.6197119951248169, "p99_ms": 0.6200992274284363, "mean_ms": 0.6178547211488088, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_8ec3a76708f5e241_sdpa_math.json", "configuration": "480x832_tp4_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp4_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp4_sp1_v1/captures/qkv_rank0_8ec3a76708f5e241.pt", "context": {"phase": "commit", "step": null, "start_frame": 27, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_flash", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [512, 10, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.0001220703125, "rmse": 1.157019779599493e-06, "relative_l2": 0.000776096161959991, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 6.247908239926173e-05, "rmse": 3.834695061417089e-06, "relative_l2": 0.0017870305491370995, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.0488319993019104, "p95_ms": 0.04915360156446695, "p99_ms": 0.06078016076236962, "mean_ms": 0.049170133136212824, "n": 300}, {"repeat": 1, "p50_ms": 0.0488319993019104, "p95_ms": 0.04915360156446695, "p99_ms": 0.067466879338026, "mean_ms": 0.04939775973558426, "n": 300}, {"repeat": 2, "p50_ms": 0.0488319993019104, "p95_ms": 0.04912000149488449, "p99_ms": 0.0511977584660053, "mean_ms": 0.04919541316727797, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_8ec3a76708f5e241_sdpa_flash.json", "configuration": "480x832_tp4_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp4_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp4_sp1_v1/captures/qkv_rank0_8ec3a76708f5e241.pt", "context": {"phase": "commit", "step": null, "start_frame": 27, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_efficient", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [512, 10, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.0001220703125, "rmse": 2.0169412794621137e-06, "relative_l2": 0.001352907196133733, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 6.247908239926173e-05, "rmse": 4.4803732518934385e-06, "relative_l2": 0.0020879271348662404, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.08659200370311737, "p95_ms": 0.08691359870135784, "p99_ms": 0.08716960065066814, "mean_ms": 0.08671114690601826, "n": 300}, {"repeat": 1, "p50_ms": 0.0865280032157898, "p95_ms": 0.08691199868917465, "p99_ms": 0.08707327991724015, "mean_ms": 0.0867632007598877, "n": 300}, {"repeat": 2, "p50_ms": 0.0865280032157898, "p95_ms": 0.08687999844551086, "p99_ms": 0.08716800063848495, "mean_ms": 0.0867852805306514, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_8ec3a76708f5e241_sdpa_efficient.json", "configuration": "480x832_tp4_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp4_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp4_sp1_v1/captures/qkv_rank0_8ec3a76708f5e241.pt", "context": {"phase": "commit", "step": null, "start_frame": 27, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_cudnn", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [512, 10, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.0001220703125, "rmse": 2.0167798249559296e-06, "relative_l2": 0.0013527988970149197, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 6.247908239926173e-05, "rmse": 4.4803732518996515e-06, "relative_l2": 0.0020879271348691357, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.04851200059056282, "p95_ms": 0.05228959955275059, "p99_ms": 0.059833280928432916, "mean_ms": 0.049269973300397396, "n": 300}, {"repeat": 1, "p50_ms": 0.04867200180888176, "p95_ms": 0.052875199913978585, "p99_ms": 0.05763743925839662, "mean_ms": 0.04944245326022307, "n": 300}, {"repeat": 2, "p50_ms": 0.04871999844908714, "p95_ms": 0.053281599655747414, "p99_ms": 0.05981280092149972, "mean_ms": 0.0494633599370718, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_8ec3a76708f5e241_sdpa_cudnn.json", "configuration": "480x832_tp4_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp4_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp4_sp1_v1/captures/qkv_rank0_8ec3a76708f5e241.pt", "context": {"phase": "commit", "step": null, "start_frame": 27, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "native_dense", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [512, 10, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.0, "rmse": 0.0, "relative_l2": 0.0, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 6.247908239926173e-05, "rmse": 3.909016073014658e-06, "relative_l2": 0.0018216653547840833, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.10875200107693672, "p95_ms": 0.12120959833264353, "p99_ms": 0.1410006387531755, "mean_ms": 0.11133994673689207, "n": 300}, {"repeat": 1, "p50_ms": 0.10931199789047241, "p95_ms": 0.11840320006012917, "p99_ms": 0.1267244763672352, "mean_ms": 0.11143989344437917, "n": 300}, {"repeat": 2, "p50_ms": 0.10892800241708755, "p95_ms": 0.11734399944543839, "p99_ms": 0.12834655374288556, "mean_ms": 0.1106065068145593, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_8ec3a76708f5e241_native_dense.json", "configuration": "480x832_tp4_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp4_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp4_sp1_v1/captures/qkv_rank0_97529d7d3b67351f.pt", "context": {"phase": "commit", "step": null, "start_frame": 9, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "vllm_fa2", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [29640, 10, 128], "q_len": 4680, "k_len": 18720, "reference_error": {"max_abs": 0.03125, "rmse": 0.00041900370590555264, "relative_l2": 0.0005262333793292784, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.015203616879200688, "rmse": 0.0009662067833630117, "relative_l2": 0.0016903244748967818, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 1.266864001750946, "p95_ms": 1.2750191628932952, "p99_ms": 1.2767778897285462, "mean_ms": 1.2678863970438639, "n": 300}, {"repeat": 1, "p50_ms": 1.2673439979553223, "p95_ms": 1.2743824481964112, "p99_ms": 1.2761600017547607, "mean_ms": 1.2685106110572815, "n": 300}, {"repeat": 2, "p50_ms": 1.2771199941635132, "p95_ms": 1.2858592092990875, "p99_ms": 1.2870745038986207, "mean_ms": 1.2771292781829835, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_97529d7d3b67351f_vllm_fa2.json", "configuration": "480x832_tp4_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp4_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp4_sp1_v1/captures/qkv_rank0_97529d7d3b67351f.pt", "context": {"phase": "commit", "step": null, "start_frame": 9, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "vllm_fa3", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [29640, 10, 128], "q_len": 4680, "k_len": 18720, "reference_error": {"max_abs": 0.0, "rmse": 0.0, "relative_l2": 0.0, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.015203616879200688, "rmse": 0.0009679160367916269, "relative_l2": 0.0016933147177246508, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.6808000206947327, "p95_ms": 0.6954336196184159, "p99_ms": 0.6966726619005204, "mean_ms": 0.6618923731644948, "n": 300}, {"repeat": 1, "p50_ms": 0.6756480038166046, "p95_ms": 0.6814768046140671, "p99_ms": 0.6829145520925521, "mean_ms": 0.6715546679496766, "n": 300}, {"repeat": 2, "p50_ms": 0.6776160001754761, "p95_ms": 0.6824047893285752, "p99_ms": 0.6840028953552246, "mean_ms": 0.6747394128640493, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_97529d7d3b67351f_vllm_fa3.json", "configuration": "480x832_tp4_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp4_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp4_sp1_v1/captures/qkv_rank0_97529d7d3b67351f.pt", "context": {"phase": "commit", "step": null, "start_frame": 9, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_math", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [29640, 10, 128], "q_len": 4680, "k_len": 18720, "reference_error": {"max_abs": 0.0625, "rmse": 0.0007924681041203298, "relative_l2": 0.0009952732220843614, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.015203616879200688, "rmse": 0.0009519723862964815, "relative_l2": 0.0016654221970808381, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 18.716368675231934, "p95_ms": 18.80713806152344, "p99_ms": 18.846028404235838, "mean_ms": 18.748345565795898, "n": 300}, {"repeat": 1, "p50_ms": 18.710848808288574, "p95_ms": 18.805152893066406, "p99_ms": 18.862507038116455, "mean_ms": 18.7322021484375, "n": 300}, {"repeat": 2, "p50_ms": 18.71356773376465, "p95_ms": 18.804884052276613, "p99_ms": 18.82685836791992, "mean_ms": 18.739165547688803, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_97529d7d3b67351f_sdpa_math.json", "configuration": "480x832_tp4_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp4_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp4_sp1_v1/captures/qkv_rank0_97529d7d3b67351f.pt", "context": {"phase": "commit", "step": null, "start_frame": 9, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_flash", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [29640, 10, 128], "q_len": 4680, "k_len": 18720, "reference_error": {"max_abs": 0.03125, "rmse": 0.0004194986690667778, "relative_l2": 0.0005268550113895774, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.015203616879200688, "rmse": 0.000966359118105626, "relative_l2": 0.0016905909759690697, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 1.3735679984092712, "p95_ms": 1.3783696174621582, "p99_ms": 1.3799075520038604, "mean_ms": 1.3741954131921132, "n": 300}, {"repeat": 1, "p50_ms": 1.3731199502944946, "p95_ms": 1.3780207931995392, "p99_ms": 1.3793318521976472, "mean_ms": 1.3741215980052948, "n": 300}, {"repeat": 2, "p50_ms": 1.3731679916381836, "p95_ms": 1.378209638595581, "p99_ms": 1.3791680335998535, "mean_ms": 1.3742079961299896, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_97529d7d3b67351f_sdpa_flash.json", "configuration": "480x832_tp4_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp4_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp4_sp1_v1/captures/qkv_rank0_97529d7d3b67351f.pt", "context": {"phase": "commit", "step": null, "start_frame": 9, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_efficient", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [29640, 10, 128], "q_len": 4680, "k_len": 18720, "reference_error": {"max_abs": 0.0625, "rmse": 0.0006689491211432061, "relative_l2": 0.0008401437783414009, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.015203616879200688, "rmse": 0.0009622684242832102, "relative_l2": 0.0016834345369889257, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 2.7454559803009033, "p95_ms": 2.771489608287811, "p99_ms": 2.785059313774109, "mean_ms": 2.7488168501853942, "n": 300}, {"repeat": 1, "p50_ms": 2.739423990249634, "p95_ms": 2.767705535888672, "p99_ms": 2.771778202056885, "mean_ms": 2.74478133281072, "n": 300}, {"repeat": 2, "p50_ms": 2.7423359155654907, "p95_ms": 2.766696107387543, "p99_ms": 2.773731415271759, "mean_ms": 2.74633589665095, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_97529d7d3b67351f_sdpa_efficient.json", "configuration": "480x832_tp4_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp4_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp4_sp1_v1/captures/qkv_rank0_97529d7d3b67351f.pt", "context": {"phase": "commit", "step": null, "start_frame": 9, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_cudnn", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [29640, 10, 128], "q_len": 4680, "k_len": 18720, "reference_error": {"max_abs": 0.0625, "rmse": 0.0006687379203653653, "relative_l2": 0.0008398785279451066, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.015203616879200688, "rmse": 0.0009624051866869884, "relative_l2": 0.0016836737951294528, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.8009920120239258, "p95_ms": 0.8144336223602295, "p99_ms": 0.8152972745895386, "mean_ms": 0.7878456526994705, "n": 300}, {"repeat": 1, "p50_ms": 0.805296003818512, "p95_ms": 0.819937577843666, "p99_ms": 0.8210563284158706, "mean_ms": 0.8022598401705424, "n": 300}, {"repeat": 2, "p50_ms": 0.8064320087432861, "p95_ms": 0.8282560110092163, "p99_ms": 0.830441597700119, "mean_ms": 0.8055635193983713, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_97529d7d3b67351f_sdpa_cudnn.json", "configuration": "480x832_tp4_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp4_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp4_sp1_v1/captures/qkv_rank0_97529d7d3b67351f.pt", "context": {"phase": "commit", "step": null, "start_frame": 9, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "native_dense", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [29640, 10, 128], "q_len": 4680, "k_len": 18720, "reference_error": {"max_abs": 0.03125, "rmse": 0.0003695395739223262, "relative_l2": 0.000464110593868784, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.015203616879200688, "rmse": 0.0009682983192628016, "relative_l2": 0.00169398349942695, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.6745920181274414, "p95_ms": 0.6963231980800628, "p99_ms": 0.6981769859790802, "mean_ms": 0.6694493863979976, "n": 300}, {"repeat": 1, "p50_ms": 0.675167977809906, "p95_ms": 0.6797808110713959, "p99_ms": 0.6805452984571457, "mean_ms": 0.6693049595753352, "n": 300}, {"repeat": 2, "p50_ms": 0.6706719994544983, "p95_ms": 0.6796479821205139, "p99_ms": 0.6805449783802032, "mean_ms": 0.6655379223823548, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_97529d7d3b67351f_native_dense.json", "configuration": "480x832_tp4_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp4_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp4_sp1_v1/captures/qkv_rank0_9bc0adfc73179efd.pt", "context": {"phase": "commit", "step": null, "start_frame": 27, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "vllm_fa2", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [29640, 10, 128], "q_len": 4680, "k_len": 28080, "reference_error": {"max_abs": 0.03125, "rmse": 0.000424195657182054, "relative_l2": 0.0005311621360561927, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.016575376154744248, "rmse": 0.000946485639961768, "relative_l2": 0.0016896618022174945, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 1.8901760578155518, "p95_ms": 1.8980496406555176, "p99_ms": 1.8998725855350493, "mean_ms": 1.891054185628891, "n": 300}, {"repeat": 1, "p50_ms": 1.9047839641571045, "p95_ms": 1.9491007804870606, "p99_ms": 1.9588805878162383, "mean_ms": 1.9126413838068643, "n": 300}, {"repeat": 2, "p50_ms": 1.9017760157585144, "p95_ms": 1.9119679927825928, "p99_ms": 1.9160444366931915, "mean_ms": 1.9001374932130177, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_9bc0adfc73179efd_vllm_fa2.json", "configuration": "480x832_tp4_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp4_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp4_sp1_v1/captures/qkv_rank0_9bc0adfc73179efd.pt", "context": {"phase": "commit", "step": null, "start_frame": 27, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "vllm_fa3", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [29640, 10, 128], "q_len": 4680, "k_len": 28080, "reference_error": {"max_abs": 0.0, "rmse": 0.0, "relative_l2": 0.0, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.016575376154744248, "rmse": 0.0009479051607445326, "relative_l2": 0.0016921959241764796, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 1.0140800476074219, "p95_ms": 1.0186895668506621, "p99_ms": 1.0194940567016602, "mean_ms": 1.0014964292446773, "n": 300}, {"repeat": 1, "p50_ms": 1.0180000066757202, "p95_ms": 1.0369344532489777, "p99_ms": 1.046949782371521, "mean_ms": 1.019128535191218, "n": 300}, {"repeat": 2, "p50_ms": 1.0159039497375488, "p95_ms": 1.1111104130744933, "p99_ms": 1.1155660808086394, "mean_ms": 1.0220771223306655, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_9bc0adfc73179efd_vllm_fa3.json", "configuration": "480x832_tp4_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp4_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp4_sp1_v1/captures/qkv_rank0_9bc0adfc73179efd.pt", "context": {"phase": "commit", "step": null, "start_frame": 27, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_math", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [29640, 10, 128], "q_len": 4680, "k_len": 28080, "reference_error": {"max_abs": 0.0625, "rmse": 0.000803455675276601, "relative_l2": 0.0010060575243542235, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.014674623845255752, "rmse": 0.0009237150241809148, "relative_l2": 0.0016490118038726357, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 28.114848136901855, "p95_ms": 28.44688653945923, "p99_ms": 28.777763633728025, "mean_ms": 28.157398306528727, "n": 300}, {"repeat": 1, "p50_ms": 28.12454319000244, "p95_ms": 28.42513418197632, "p99_ms": 28.64854108810425, "mean_ms": 28.164739640553794, "n": 300}, {"repeat": 2, "p50_ms": 28.13241672515869, "p95_ms": 28.446806716918946, "p99_ms": 28.549840812683104, "mean_ms": 28.175674222310384, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_9bc0adfc73179efd_sdpa_math.json", "configuration": "480x832_tp4_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp4_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp4_sp1_v1/captures/qkv_rank0_9bc0adfc73179efd.pt", "context": {"phase": "commit", "step": null, "start_frame": 27, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_flash", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [29640, 10, 128], "q_len": 4680, "k_len": 28080, "reference_error": {"max_abs": 0.03125, "rmse": 0.00042479136059178093, "relative_l2": 0.0005319080538660742, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.016575376154744248, "rmse": 0.0009464889372720555, "relative_l2": 0.001689667688560623, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 2.049615979194641, "p95_ms": 2.0644976615905763, "p99_ms": 2.068546471595764, "mean_ms": 2.051428372859955, "n": 300}, {"repeat": 1, "p50_ms": 2.0630240440368652, "p95_ms": 2.0672318935394287, "p99_ms": 2.070567018985748, "mean_ms": 2.058408325513204, "n": 300}, {"repeat": 2, "p50_ms": 2.063632011413574, "p95_ms": 2.0681425094604493, "p99_ms": 2.0712704277038574, "mean_ms": 2.061227524280548, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_9bc0adfc73179efd_sdpa_flash.json", "configuration": "480x832_tp4_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp4_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp4_sp1_v1/captures/qkv_rank0_9bc0adfc73179efd.pt", "context": {"phase": "commit", "step": null, "start_frame": 27, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_efficient", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [29640, 10, 128], "q_len": 4680, "k_len": 28080, "reference_error": {"max_abs": 0.0625, "rmse": 0.0006825143821691953, "relative_l2": 0.0008546193035787624, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.014674623845255752, "rmse": 0.0009329665457116266, "relative_l2": 0.0016655275774700756, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 4.103360176086426, "p95_ms": 4.134164643287659, "p99_ms": 4.1385614204406735, "mean_ms": 4.1069624487559, "n": 300}, {"repeat": 1, "p50_ms": 4.113855838775635, "p95_ms": 4.1551824569702145, "p99_ms": 4.161320104598999, "mean_ms": 4.116643940607707, "n": 300}, {"repeat": 2, "p50_ms": 4.112720012664795, "p95_ms": 4.14518723487854, "p99_ms": 4.155647540092469, "mean_ms": 4.115284584363302, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_9bc0adfc73179efd_sdpa_efficient.json", "configuration": "480x832_tp4_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp4_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp4_sp1_v1/captures/qkv_rank0_9bc0adfc73179efd.pt", "context": {"phase": "commit", "step": null, "start_frame": 27, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_cudnn", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [29640, 10, 128], "q_len": 4680, "k_len": 28080, "reference_error": {"max_abs": 0.0625, "rmse": 0.0006823629337958054, "relative_l2": 0.0008544296655187067, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.014674623845255752, "rmse": 0.0009329419340941724, "relative_l2": 0.0016654836409240284, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 1.2022560238838196, "p95_ms": 1.2196176528930665, "p99_ms": 1.2215695881843567, "mean_ms": 1.1873091197013854, "n": 300}, {"repeat": 1, "p50_ms": 1.2017760276794434, "p95_ms": 1.2123855888843535, "p99_ms": 1.21599649310112, "mean_ms": 1.1979383464654287, "n": 300}, {"repeat": 2, "p50_ms": 1.2245439887046814, "p95_ms": 1.4083055555820465, "p99_ms": 1.448352619409561, "mean_ms": 1.256222404638926, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_9bc0adfc73179efd_sdpa_cudnn.json", "configuration": "480x832_tp4_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp4_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp4_sp1_v1/captures/qkv_rank0_9bc0adfc73179efd.pt", "context": {"phase": "commit", "step": null, "start_frame": 27, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "native_dense", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [29640, 10, 128], "q_len": 4680, "k_len": 28080, "reference_error": {"max_abs": 0.0625, "rmse": 0.00037584647978582727, "relative_l2": 0.00047062108169240925, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.016575376154744248, "rmse": 0.0009487364479612307, "relative_l2": 0.001693679934284423, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 1.0061759948730469, "p95_ms": 1.0110399723052979, "p99_ms": 1.0152399671077728, "mean_ms": 0.9950651729106903, "n": 300}, {"repeat": 1, "p50_ms": 1.00217604637146, "p95_ms": 1.0115920424461364, "p99_ms": 1.01664186835289, "mean_ms": 1.0006252815326055, "n": 300}, {"repeat": 2, "p50_ms": 1.02811199426651, "p95_ms": 1.1389807760715485, "p99_ms": 1.1398083782196045, "mean_ms": 1.0440246377388636, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_9bc0adfc73179efd_native_dense.json", "configuration": "480x832_tp4_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp4_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp4_sp1_v1/captures/qkv_rank0_a7ce4999e4f2ba20.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 9, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "vllm_fa2", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [512, 10, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.0001220703125, "rmse": 1.2252090711542444e-06, "relative_l2": 0.0007246898777295257, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 5.639846381974567e-05, "rmse": 4.843710769714728e-06, "relative_l2": 0.0018255362253831512, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.0504320003092289, "p95_ms": 0.0559392012655735, "p99_ms": 0.07391872309148301, "mean_ms": 0.052082453370094296, "n": 300}, {"repeat": 1, "p50_ms": 0.05008000135421753, "p95_ms": 0.05451200138777495, "p99_ms": 0.06425407730042931, "mean_ms": 0.05094133336097002, "n": 300}, {"repeat": 2, "p50_ms": 0.050383999943733215, "p95_ms": 0.05531200002878905, "p99_ms": 0.06317503646016118, "mean_ms": 0.05120170667767525, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_a7ce4999e4f2ba20_vllm_fa2.json", "configuration": "480x832_tp4_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp4_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp4_sp1_v1/captures/qkv_rank0_a7ce4999e4f2ba20.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 9, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "vllm_fa3", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [512, 10, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.0001220703125, "rmse": 4.6154809244275377e-07, "relative_l2": 0.0002729976773380636, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 5.639846381974567e-05, "rmse": 4.870710199118303e-06, "relative_l2": 0.001835711984998914, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.06124800071120262, "p95_ms": 0.06996800340712071, "p99_ms": 0.07807647824287411, "mean_ms": 0.0627017600958546, "n": 300}, {"repeat": 1, "p50_ms": 0.060864001512527466, "p95_ms": 0.0699232030659914, "p99_ms": 0.09646527633070938, "mean_ms": 0.06698805355777343, "n": 300}, {"repeat": 2, "p50_ms": 0.06092799827456474, "p95_ms": 0.06896160319447517, "p99_ms": 0.07868832312524318, "mean_ms": 0.062432853269080324, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_a7ce4999e4f2ba20_vllm_fa3.json", "configuration": "480x832_tp4_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp4_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp4_sp1_v1/captures/qkv_rank0_a7ce4999e4f2ba20.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 9, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_math", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [512, 10, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.000244140625, "rmse": 3.1625517125741483e-06, "relative_l2": 0.001870594389037243, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 5.639846381974567e-05, "rmse": 4.370260654146473e-06, "relative_l2": 0.0016470985816068708, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.6160800158977509, "p95_ms": 0.6187247961759568, "p99_ms": 0.6202892941236496, "mean_ms": 0.6161752518018087, "n": 300}, {"repeat": 1, "p50_ms": 0.6166399717330933, "p95_ms": 0.6193616271018982, "p99_ms": 0.6201324564218521, "mean_ms": 0.6175765335559845, "n": 300}, {"repeat": 2, "p50_ms": 0.6161440014839172, "p95_ms": 0.6188496232032776, "p99_ms": 0.6195545762777328, "mean_ms": 0.6166249610980352, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_a7ce4999e4f2ba20_sdpa_math.json", "configuration": "480x832_tp4_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp4_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp4_sp1_v1/captures/qkv_rank0_a7ce4999e4f2ba20.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 9, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_flash", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [512, 10, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.0001220703125, "rmse": 1.2251524867308096e-06, "relative_l2": 0.0007246564090262115, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 5.639846381974567e-05, "rmse": 4.843707402967178e-06, "relative_l2": 0.0018255349564965666, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.048928000032901764, "p95_ms": 0.049215998500585556, "p99_ms": 0.05021856136620032, "mean_ms": 0.04922655999660492, "n": 300}, {"repeat": 1, "p50_ms": 0.048928000032901764, "p95_ms": 0.049217598512768745, "p99_ms": 0.05045120008289809, "mean_ms": 0.049259626641869544, "n": 300}, {"repeat": 2, "p50_ms": 0.04896000027656555, "p95_ms": 0.04927999898791313, "p99_ms": 0.049459519945085004, "mean_ms": 0.04933823990325133, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_a7ce4999e4f2ba20_sdpa_flash.json", "configuration": "480x832_tp4_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp4_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp4_sp1_v1/captures/qkv_rank0_a7ce4999e4f2ba20.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 9, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_efficient", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [512, 10, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.0001220703125, "rmse": 2.158768948448774e-06, "relative_l2": 0.00127687432466021, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 5.639846381974567e-05, "rmse": 5.333178611650721e-06, "relative_l2": 0.0020100107572237247, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.08684799820184708, "p95_ms": 0.08726400136947632, "p99_ms": 0.08748800307512283, "mean_ms": 0.08699445222814878, "n": 300}, {"repeat": 1, "p50_ms": 0.08684799820184708, "p95_ms": 0.08726400136947632, "p99_ms": 0.08758591637015342, "mean_ms": 0.08710421236852804, "n": 300}, {"repeat": 2, "p50_ms": 0.08659200370311737, "p95_ms": 0.08697759918868542, "p99_ms": 0.08723232112824916, "mean_ms": 0.08682581370075544, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_a7ce4999e4f2ba20_sdpa_efficient.json", "configuration": "480x832_tp4_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp4_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp4_sp1_v1/captures/qkv_rank0_a7ce4999e4f2ba20.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 9, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_cudnn", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [512, 10, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.0001220703125, "rmse": 2.153875945349e-06, "relative_l2": 0.0012739801983420258, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 5.639846381974567e-05, "rmse": 5.333178611650721e-06, "relative_l2": 0.0020100107572237247, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.047807998955249786, "p95_ms": 0.05111360158771277, "p99_ms": 0.055593918561935414, "mean_ms": 0.04824736002832651, "n": 300}, {"repeat": 1, "p50_ms": 0.04790399968624115, "p95_ms": 0.0514272004365921, "p99_ms": 0.053835840076208105, "mean_ms": 0.048513280029098195, "n": 300}, {"repeat": 2, "p50_ms": 0.048528000712394714, "p95_ms": 0.060511998459696785, "p99_ms": 0.06761760078370571, "mean_ms": 0.05051551998903354, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_a7ce4999e4f2ba20_sdpa_cudnn.json", "configuration": "480x832_tp4_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp4_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp4_sp1_v1/captures/qkv_rank0_a7ce4999e4f2ba20.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 9, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "native_dense", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [512, 10, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.0, "rmse": 0.0, "relative_l2": 0.0, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 5.639846381974567e-05, "rmse": 4.882930400621915e-06, "relative_l2": 0.0018403176316997471, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.16369599848985672, "p95_ms": 0.17727200090885162, "p99_ms": 0.18413439363241194, "mean_ms": 0.165471679965655, "n": 300}, {"repeat": 1, "p50_ms": 0.15512000024318695, "p95_ms": 0.17048480138182642, "p99_ms": 0.20966400444507538, "mean_ms": 0.1509335463742415, "n": 300}, {"repeat": 2, "p50_ms": 0.10688000172376633, "p95_ms": 0.16792640388011934, "p99_ms": 0.18547424286603914, "mean_ms": 0.11304992000261943, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_a7ce4999e4f2ba20_native_dense.json", "configuration": "480x832_tp4_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp4_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp4_sp1_v1/captures/qkv_rank0_bf28d4d0099672a9.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 9, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "vllm_fa2", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [29640, 10, 128], "q_len": 4680, "k_len": 18720, "reference_error": {"max_abs": 0.03125, "rmse": 0.0005648215900519224, "relative_l2": 0.001007988856126751, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.008336925131513073, "rmse": 0.00082262480363599, "relative_l2": 0.0019419552084823568, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 1.268239974975586, "p95_ms": 1.2749760150909424, "p99_ms": 1.2763033926486969, "mean_ms": 1.2689609626928966, "n": 300}, {"repeat": 1, "p50_ms": 1.2690719962120056, "p95_ms": 1.2754912197589874, "p99_ms": 1.277344628572464, "mean_ms": 1.2705005892117818, "n": 300}, {"repeat": 2, "p50_ms": 1.2797279953956604, "p95_ms": 1.3094223499298097, "p99_ms": 1.3130998992919922, "mean_ms": 1.2862354091803232, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_bf28d4d0099672a9_vllm_fa2.json", "configuration": "480x832_tp4_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp4_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp4_sp1_v1/captures/qkv_rank0_bf28d4d0099672a9.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 9, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "vllm_fa3", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [29640, 10, 128], "q_len": 4680, "k_len": 18720, "reference_error": {"max_abs": 0.0, "rmse": 0.0, "relative_l2": 0.0, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.008336925131513073, "rmse": 0.0008346620757216445, "relative_l2": 0.001970371374782394, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.6813759803771973, "p95_ms": 0.6834560036659241, "p99_ms": 0.6851843219995498, "mean_ms": 0.6644339207808176, "n": 300}, {"repeat": 1, "p50_ms": 0.6825440227985382, "p95_ms": 0.6859520077705383, "p99_ms": 0.6864688116312027, "mean_ms": 0.6804402126868566, "n": 300}, {"repeat": 2, "p50_ms": 0.6832000017166138, "p95_ms": 0.6861119866371155, "p99_ms": 0.6872643446922302, "mean_ms": 0.6790842670202255, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_bf28d4d0099672a9_vllm_fa3.json", "configuration": "480x832_tp4_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp4_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp4_sp1_v1/captures/qkv_rank0_bf28d4d0099672a9.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 9, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_math", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [29640, 10, 128], "q_len": 4680, "k_len": 18720, "reference_error": {"max_abs": 0.03125, "rmse": 0.0009919706983384075, "relative_l2": 0.001770285391246231, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.007523701702825125, "rmse": 0.0007105348336103595, "relative_l2": 0.0016773464826722552, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 18.756112098693848, "p95_ms": 18.814216709136964, "p99_ms": 18.88889030456543, "mean_ms": 18.75619389216105, "n": 300}, {"repeat": 1, "p50_ms": 18.79038429260254, "p95_ms": 18.847648525238036, "p99_ms": 18.94211420059204, "mean_ms": 18.768264350891112, "n": 300}, {"repeat": 2, "p50_ms": 18.79491138458252, "p95_ms": 18.836523818969727, "p99_ms": 18.98126708984375, "mean_ms": 18.7800199953715, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_bf28d4d0099672a9_sdpa_math.json", "configuration": "480x832_tp4_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp4_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp4_sp1_v1/captures/qkv_rank0_bf28d4d0099672a9.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 9, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_flash", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [29640, 10, 128], "q_len": 4680, "k_len": 18720, "reference_error": {"max_abs": 0.03125, "rmse": 0.0005653377756525876, "relative_l2": 0.001008910048486122, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.008336925131513073, "rmse": 0.0008226377995782296, "relative_l2": 0.0019419858877636157, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 1.3823360204696655, "p95_ms": 1.3881663918495177, "p99_ms": 1.3895698916912078, "mean_ms": 1.3796112000942231, "n": 300}, {"repeat": 1, "p50_ms": 1.373744010925293, "p95_ms": 1.3788224101066588, "p99_ms": 1.3801325058937073, "mean_ms": 1.374571836789449, "n": 300}, {"repeat": 2, "p50_ms": 1.38264000415802, "p95_ms": 1.395840048789978, "p99_ms": 1.3976384031772613, "mean_ms": 1.3827437897523245, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_bf28d4d0099672a9_sdpa_flash.json", "configuration": "480x832_tp4_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp4_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp4_sp1_v1/captures/qkv_rank0_bf28d4d0099672a9.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 9, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_efficient", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [29640, 10, 128], "q_len": 4680, "k_len": 18720, "reference_error": {"max_abs": 0.03125, "rmse": 0.000774696594767337, "relative_l2": 0.0013825348537633497, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.009271584692595258, "rmse": 0.0008368465886884668, "relative_l2": 0.0019755283142706853, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 2.747968077659607, "p95_ms": 2.7758176922798157, "p99_ms": 2.7908269476890566, "mean_ms": 2.752675836086273, "n": 300}, {"repeat": 1, "p50_ms": 2.7482240200042725, "p95_ms": 2.7720863699913023, "p99_ms": 2.7892505502700806, "mean_ms": 2.751597553094228, "n": 300}, {"repeat": 2, "p50_ms": 2.7455519437789917, "p95_ms": 2.7683680772781374, "p99_ms": 2.7737307739257813, "mean_ms": 2.7494572814305624, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_bf28d4d0099672a9_sdpa_efficient.json", "configuration": "480x832_tp4_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp4_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp4_sp1_v1/captures/qkv_rank0_bf28d4d0099672a9.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 9, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_cudnn", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [29640, 10, 128], "q_len": 4680, "k_len": 18720, "reference_error": {"max_abs": 0.03125, "rmse": 0.0007741805411958278, "relative_l2": 0.0013816138970251378, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.009271584692595258, "rmse": 0.0008361158375863229, "relative_l2": 0.001973803243615582, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.8125599920749664, "p95_ms": 0.8286415845155716, "p99_ms": 0.8297011196613312, "mean_ms": 0.8033014412720998, "n": 300}, {"repeat": 1, "p50_ms": 0.8048160076141357, "p95_ms": 0.8285136133432388, "p99_ms": 0.8301791960000992, "mean_ms": 0.8076135484377543, "n": 300}, {"repeat": 2, "p50_ms": 0.805791974067688, "p95_ms": 0.8284800052642822, "p99_ms": 0.8290025693178177, "mean_ms": 0.80541951974233, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_bf28d4d0099672a9_sdpa_cudnn.json", "configuration": "480x832_tp4_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp4_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_480x832_tp4_sp1_v1/captures/qkv_rank0_bf28d4d0099672a9.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 9, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "native_dense", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [29640, 10, 128], "q_len": 4680, "k_len": 18720, "reference_error": {"max_abs": 0.03125, "rmse": 0.0005040425990420321, "relative_l2": 0.0008995217813837945, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.008336925131513073, "rmse": 0.0008360521343382134, "relative_l2": 0.001973652860532168, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.6902720034122467, "p95_ms": 0.7322896033525467, "p99_ms": 0.73257697224617, "mean_ms": 0.6728010668357214, "n": 300}, {"repeat": 1, "p50_ms": 0.6824800074100494, "p95_ms": 0.6930895954370498, "p99_ms": 0.6937948656082153, "mean_ms": 0.6760780807336172, "n": 300}, {"repeat": 2, "p50_ms": 0.6836479902267456, "p95_ms": 0.6878431856632232, "p99_ms": 0.6886793661117554, "mean_ms": 0.6770841602484385, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_bf28d4d0099672a9_native_dense.json", "configuration": "480x832_tp4_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_480x832_tp4_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp1_sp2_v1/captures/qkv_rank0_021f565f50432429.pt", "context": {"phase": "commit", "step": null, "start_frame": 9, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "vllm_fa2", "status": "measured", "q_shape": [10560, 20, 128], "k_shape": [512, 20, 128], "q_len": 10560, "k_len": 512, "reference_error": {"max_abs": 0.00048828125, "rmse": 1.7050505477474122e-06, "relative_l2": 0.0006082610797856962, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 6.307664314882458e-05, "rmse": 4.137486204813533e-06, "relative_l2": 0.00178708309216853, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.16988800466060638, "p95_ms": 0.17078399658203125, "p99_ms": 0.1710726462304592, "mean_ms": 0.1700660264492035, "n": 300}, {"repeat": 1, "p50_ms": 0.17025599628686905, "p95_ms": 0.17104159891605378, "p99_ms": 0.17162112295627593, "mean_ms": 0.170676373342673, "n": 300}, {"repeat": 2, "p50_ms": 0.17030400037765503, "p95_ms": 0.1712000072002411, "p99_ms": 0.17213088676333427, "mean_ms": 0.17066037386655808, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_021f565f50432429_vllm_fa2.json", "configuration": "704x1280_tp1_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp1_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp1_sp2_v1/captures/qkv_rank0_021f565f50432429.pt", "context": {"phase": "commit", "step": null, "start_frame": 9, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "vllm_fa3", "status": "measured", "q_shape": [10560, 20, 128], "k_shape": [512, 20, 128], "q_len": 10560, "k_len": 512, "reference_error": {"max_abs": 0.000244140625, "rmse": 6.435260545588146e-07, "relative_l2": 0.0002295719932369773, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 6.307664314882458e-05, "rmse": 4.161822254469216e-06, "relative_l2": 0.0017975944366702345, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.1419840008020401, "p95_ms": 0.1435552053153515, "p99_ms": 0.1447388868033886, "mean_ms": 0.14227253297964731, "n": 300}, {"repeat": 1, "p50_ms": 0.14195199310779572, "p95_ms": 0.1434912048280239, "p99_ms": 0.14480000734329224, "mean_ms": 0.14251488015055658, "n": 300}, {"repeat": 2, "p50_ms": 0.14195199310779572, "p95_ms": 0.14342400431632996, "p99_ms": 0.14426303580403327, "mean_ms": 0.14241983994841576, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_021f565f50432429_vllm_fa3.json", "configuration": "704x1280_tp1_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp1_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp1_sp2_v1/captures/qkv_rank0_021f565f50432429.pt", "context": {"phase": "commit", "step": null, "start_frame": 9, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_math", "status": "measured", "q_shape": [10560, 20, 128], "k_shape": [512, 20, 128], "q_len": 10560, "k_len": 512, "reference_error": {"max_abs": 0.001953125, "rmse": 4.97330809988091e-06, "relative_l2": 0.00177418186160933, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 6.095633961755026e-05, "rmse": 3.932673083918934e-06, "relative_l2": 0.001698619216451148, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 2.4257280826568604, "p95_ms": 2.430695903301239, "p99_ms": 2.4322569441795348, "mean_ms": 2.4262420241038005, "n": 300}, {"repeat": 1, "p50_ms": 2.425312042236328, "p95_ms": 2.429926359653473, "p99_ms": 2.432643914222717, "mean_ms": 2.42543199857076, "n": 300}, {"repeat": 2, "p50_ms": 2.425600051879883, "p95_ms": 2.4303136467933655, "p99_ms": 2.4322904920578003, "mean_ms": 2.4256832067171734, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_021f565f50432429_sdpa_math.json", "configuration": "704x1280_tp1_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp1_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp1_sp2_v1/captures/qkv_rank0_021f565f50432429.pt", "context": {"phase": "commit", "step": null, "start_frame": 9, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_flash", "status": "measured", "q_shape": [10560, 20, 128], "k_shape": [512, 20, 128], "q_len": 10560, "k_len": 512, "reference_error": {"max_abs": 0.00048828125, "rmse": 1.705088933075421e-06, "relative_l2": 0.0006082747733979432, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 6.307664314882458e-05, "rmse": 4.137486204813533e-06, "relative_l2": 0.00178708309216853, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.18223999440670013, "p95_ms": 0.18332960233092307, "p99_ms": 0.1837158454954624, "mean_ms": 0.18245578666528067, "n": 300}, {"repeat": 1, "p50_ms": 0.18227200210094452, "p95_ms": 0.18348799645900726, "p99_ms": 0.18387391939759254, "mean_ms": 0.18268458674351373, "n": 300}, {"repeat": 2, "p50_ms": 0.18185600638389587, "p95_ms": 0.18291200697422028, "p99_ms": 0.1835212841629982, "mean_ms": 0.1822279474635919, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_021f565f50432429_sdpa_flash.json", "configuration": "704x1280_tp1_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp1_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp1_sp2_v1/captures/qkv_rank0_021f565f50432429.pt", "context": {"phase": "commit", "step": null, "start_frame": 9, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_efficient", "status": "measured", "q_shape": [10560, 20, 128], "k_shape": [512, 20, 128], "q_len": 10560, "k_len": 512, "reference_error": {"max_abs": 0.0009765625, "rmse": 2.991261377138909e-06, "relative_l2": 0.001067104947465346, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 6.307664314882458e-05, "rmse": 4.736439880333926e-06, "relative_l2": 0.0020457860662761937, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.33590400218963623, "p95_ms": 0.33744161278009416, "p99_ms": 0.339364158809185, "mean_ms": 0.3362113071481387, "n": 300}, {"repeat": 1, "p50_ms": 0.33659200370311737, "p95_ms": 0.33760319352149964, "p99_ms": 0.3384019261598587, "mean_ms": 0.33689919968446097, "n": 300}, {"repeat": 2, "p50_ms": 0.33612799644470215, "p95_ms": 0.3372816056013107, "p99_ms": 0.33744961380958555, "mean_ms": 0.3364279468854268, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_021f565f50432429_sdpa_efficient.json", "configuration": "704x1280_tp1_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp1_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp1_sp2_v1/captures/qkv_rank0_021f565f50432429.pt", "context": {"phase": "commit", "step": null, "start_frame": 9, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_cudnn", "status": "measured", "q_shape": [10560, 20, 128], "k_shape": [512, 20, 128], "q_len": 10560, "k_len": 512, "reference_error": {"max_abs": 0.00048828125, "rmse": 2.9831965145853658e-06, "relative_l2": 0.0010642278820248976, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 6.307664314882458e-05, "rmse": 4.736452756624415e-06, "relative_l2": 0.002045791627865982, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.1029760017991066, "p95_ms": 0.10384320095181465, "p99_ms": 0.10429087691009044, "mean_ms": 0.10324096019069354, "n": 300}, {"repeat": 1, "p50_ms": 0.1029760017991066, "p95_ms": 0.10387360118329525, "p99_ms": 0.10435359738767147, "mean_ms": 0.10346634695927302, "n": 300}, {"repeat": 2, "p50_ms": 0.10300800204277039, "p95_ms": 0.10381120070815086, "p99_ms": 0.1047686405479908, "mean_ms": 0.10338666704793771, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_021f565f50432429_sdpa_cudnn.json", "configuration": "704x1280_tp1_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp1_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp1_sp2_v1/captures/qkv_rank0_021f565f50432429.pt", "context": {"phase": "commit", "step": null, "start_frame": 9, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "native_dense", "status": "measured", "q_shape": [10560, 20, 128], "k_shape": [512, 20, 128], "q_len": 10560, "k_len": 512, "reference_error": {"max_abs": 0.0, "rmse": 0.0, "relative_l2": 0.0, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 6.307664314882458e-05, "rmse": 4.17386793468327e-06, "relative_l2": 0.0018027972652426081, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.11819199845194817, "p95_ms": 0.1296784035861492, "p99_ms": 0.20919455572962728, "mean_ms": 0.12725717278818288, "n": 300}, {"repeat": 1, "p50_ms": 0.118367999792099, "p95_ms": 0.16051039919257168, "p99_ms": 0.22136095628142205, "mean_ms": 0.12565994672477246, "n": 300}, {"repeat": 2, "p50_ms": 0.1194240003824234, "p95_ms": 0.15110719576478004, "p99_ms": 0.4693977624177929, "mean_ms": 0.13028810657560824, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_021f565f50432429_native_dense.json", "configuration": "704x1280_tp1_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp1_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp1_sp2_v1/captures/qkv_rank0_03f37ea3bc112c48.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 9, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "vllm_fa2", "status": "measured", "q_shape": [10560, 20, 128], "k_shape": [66880, 20, 128], "q_len": 10560, "k_len": 42240, "reference_error": {"max_abs": 0.03125, "rmse": 0.0004340133282233664, "relative_l2": 0.0008822907784943165, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.008166105858370631, "rmse": 0.0007497399548139017, "relative_l2": 0.001914449886474462, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 12.233376026153564, "p95_ms": 12.607424402236939, "p99_ms": 13.342585830688476, "mean_ms": 12.27048578898112, "n": 300}, {"repeat": 1, "p50_ms": 12.31708812713623, "p95_ms": 12.348172616958617, "p99_ms": 12.426135749816895, "mean_ms": 12.3179572168986, "n": 300}, {"repeat": 2, "p50_ms": 12.319600105285645, "p95_ms": 12.353370761871338, "p99_ms": 12.503290224075316, "mean_ms": 12.316953268051147, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_03f37ea3bc112c48_vllm_fa2.json", "configuration": "704x1280_tp1_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp1_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp1_sp2_v1/captures/qkv_rank0_03f37ea3bc112c48.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 9, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "vllm_fa3", "status": "measured", "q_shape": [10560, 20, 128], "k_shape": [66880, 20, 128], "q_len": 10560, "k_len": 42240, "reference_error": {"max_abs": 0.0, "rmse": 0.0, "relative_l2": 0.0, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.008166105858370631, "rmse": 0.0007599393361606038, "relative_l2": 0.0019404938559013684, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 6.921839952468872, "p95_ms": 7.056446552276611, "p99_ms": 7.432116146087646, "mean_ms": 6.935972046852112, "n": 300}, {"repeat": 1, "p50_ms": 6.899807929992676, "p95_ms": 6.978583979606628, "p99_ms": 6.997702965736389, "mean_ms": 6.884411938985189, "n": 300}, {"repeat": 2, "p50_ms": 6.874383926391602, "p95_ms": 6.984819126129151, "p99_ms": 7.04326455116272, "mean_ms": 6.883646933237712, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_03f37ea3bc112c48_vllm_fa3.json", "configuration": "704x1280_tp1_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp1_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp1_sp2_v1/captures/qkv_rank0_03f37ea3bc112c48.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 9, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_math", "status": "measured", "q_shape": [10560, 20, 128], "k_shape": [66880, 20, 128], "q_len": 10560, "k_len": 42240, "reference_error": {"max_abs": 0.03125, "rmse": 0.0007660935830363456, "relative_l2": 0.0015573653153544969, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.007458894141629369, "rmse": 0.0006564985344018477, "relative_l2": 0.0016763592984293276, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 187.43006134033203, "p95_ms": 188.44970169067383, "p99_ms": 188.62073776245117, "mean_ms": 187.52850870768228, "n": 300}, {"repeat": 1, "p50_ms": 187.69387817382812, "p95_ms": 188.54813919067382, "p99_ms": 188.713475189209, "mean_ms": 187.77689397176107, "n": 300}, {"repeat": 2, "p50_ms": 187.6359634399414, "p95_ms": 188.51999053955078, "p99_ms": 188.67074127197264, "mean_ms": 187.74952473958334, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_03f37ea3bc112c48_sdpa_math.json", "configuration": "704x1280_tp1_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp1_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp1_sp2_v1/captures/qkv_rank0_03f37ea3bc112c48.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 9, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_flash", "status": "measured", "q_shape": [10560, 20, 128], "k_shape": [66880, 20, 128], "q_len": 10560, "k_len": 42240, "reference_error": {"max_abs": 0.03125, "rmse": 0.00043436468039301755, "relative_l2": 0.0008830050302445017, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.008166105858370631, "rmse": 0.000749741695940512, "relative_l2": 0.0019144543324155112, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 13.254000186920166, "p95_ms": 13.535495805740357, "p99_ms": 13.916029777526855, "mean_ms": 13.26688808441162, "n": 300}, {"repeat": 1, "p50_ms": 13.259312152862549, "p95_ms": 13.47923665046692, "p99_ms": 13.481920881271362, "mean_ms": 13.268322779337565, "n": 300}, {"repeat": 2, "p50_ms": 13.259072303771973, "p95_ms": 13.47753939628601, "p99_ms": 13.481569290161133, "mean_ms": 13.258787813186645, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_03f37ea3bc112c48_sdpa_flash.json", "configuration": "704x1280_tp1_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp1_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp1_sp2_v1/captures/qkv_rank0_03f37ea3bc112c48.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 9, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_efficient", "status": "measured", "q_shape": [10560, 20, 128], "k_shape": [66880, 20, 128], "q_len": 10560, "k_len": 42240, "reference_error": {"max_abs": 0.03125, "rmse": 0.0006787940296594905, "relative_l2": 0.0013798970536622404, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.007458894141629369, "rmse": 0.0007553631967359011, "relative_l2": 0.0019288087515583728, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 25.37764835357666, "p95_ms": 25.389102077484132, "p99_ms": 25.395304222106933, "mean_ms": 25.31437873840332, "n": 300}, {"repeat": 1, "p50_ms": 25.233872413635254, "p95_ms": 25.38787384033203, "p99_ms": 25.411109962463378, "mean_ms": 25.282974491119386, "n": 300}, {"repeat": 2, "p50_ms": 25.22606372833252, "p95_ms": 25.386113834381103, "p99_ms": 25.39298225402832, "mean_ms": 25.248815720876056, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_03f37ea3bc112c48_sdpa_efficient.json", "configuration": "704x1280_tp1_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp1_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp1_sp2_v1/captures/qkv_rank0_03f37ea3bc112c48.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 9, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_cudnn", "status": "measured", "q_shape": [10560, 20, 128], "k_shape": [66880, 20, 128], "q_len": 10560, "k_len": 42240, "reference_error": {"max_abs": 0.03125, "rmse": 0.0006763949128257963, "relative_l2": 0.0013750199715054235, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.007458894141629369, "rmse": 0.0007556984185515758, "relative_l2": 0.001929664735506996, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 7.5589280128479, "p95_ms": 8.069840288162231, "p99_ms": 8.725948734283447, "mean_ms": 7.629634558359782, "n": 300}, {"repeat": 1, "p50_ms": 7.590480089187622, "p95_ms": 7.797760009765625, "p99_ms": 7.910273079872131, "mean_ms": 7.599728326797486, "n": 300}, {"repeat": 2, "p50_ms": 7.602927923202515, "p95_ms": 7.667531251907349, "p99_ms": 7.724739680290222, "mean_ms": 7.609390408198038, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_03f37ea3bc112c48_sdpa_cudnn.json", "configuration": "704x1280_tp1_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp1_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp1_sp2_v1/captures/qkv_rank0_03f37ea3bc112c48.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 9, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "native_dense", "status": "measured", "q_shape": [10560, 20, 128], "k_shape": [66880, 20, 128], "q_len": 10560, "k_len": 42240, "reference_error": {"max_abs": 0.03125, "rmse": 0.0004118266094391264, "relative_l2": 0.0008371881604053459, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.008166105858370631, "rmse": 0.0007626402458939546, "relative_l2": 0.001947390588960866, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 6.931936025619507, "p95_ms": 7.274524974822998, "p99_ms": 7.486858439445496, "mean_ms": 6.928212682406108, "n": 300}, {"repeat": 1, "p50_ms": 6.961888074874878, "p95_ms": 7.038926219940185, "p99_ms": 7.1086221694946286, "mean_ms": 6.928581328392029, "n": 300}, {"repeat": 2, "p50_ms": 6.962735891342163, "p95_ms": 7.031583786010742, "p99_ms": 7.041222295761108, "mean_ms": 6.9321138048172, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_03f37ea3bc112c48_native_dense.json", "configuration": "704x1280_tp1_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp1_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp1_sp2_v1/captures/qkv_rank0_2da4c70927569963.pt", "context": {"phase": "commit", "step": null, "start_frame": 0, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "vllm_fa2", "status": "measured", "q_shape": [10560, 20, 128], "k_shape": [66880, 20, 128], "q_len": 10560, "k_len": 10560, "reference_error": {"max_abs": 0.03125, "rmse": 0.00039832404302559603, "relative_l2": 0.0005835441920482825, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.008178553063441285, "rmse": 0.0007568670538157323, "relative_l2": 0.001639986913325226, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 3.0461440086364746, "p95_ms": 3.3337392687797545, "p99_ms": 3.4026365470886226, "mean_ms": 3.0657912508646645, "n": 300}, {"repeat": 1, "p50_ms": 3.0503519773483276, "p95_ms": 3.1503935337066653, "p99_ms": 3.2559381484985352, "mean_ms": 3.0548326365152993, "n": 300}, {"repeat": 2, "p50_ms": 3.0601760149002075, "p95_ms": 3.10792795419693, "p99_ms": 3.14522070646286, "mean_ms": 3.0726430932680766, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_2da4c70927569963_vllm_fa2.json", "configuration": "704x1280_tp1_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp1_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp1_sp2_v1/captures/qkv_rank0_2da4c70927569963.pt", "context": {"phase": "commit", "step": null, "start_frame": 0, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "vllm_fa3", "status": "measured", "q_shape": [10560, 20, 128], "k_shape": [66880, 20, 128], "q_len": 10560, "k_len": 10560, "reference_error": {"max_abs": 0.0, "rmse": 0.0, "relative_l2": 0.0, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.008178553063441285, "rmse": 0.0007573742732439552, "relative_l2": 0.0016410859613288037, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 1.7242079973220825, "p95_ms": 1.736705595254898, "p99_ms": 1.740832291841507, "mean_ms": 1.7176165342330934, "n": 300}, {"repeat": 1, "p50_ms": 1.7321439981460571, "p95_ms": 2.01747670173645, "p99_ms": 2.039938623905182, "mean_ms": 1.7825609616438547, "n": 300}, {"repeat": 2, "p50_ms": 1.7241920232772827, "p95_ms": 1.7324192345142364, "p99_ms": 1.735821135044098, "mean_ms": 1.7203141311804453, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_2da4c70927569963_vllm_fa3.json", "configuration": "704x1280_tp1_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp1_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp1_sp2_v1/captures/qkv_rank0_2da4c70927569963.pt", "context": {"phase": "commit", "step": null, "start_frame": 0, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_math", "status": "measured", "q_shape": [10560, 20, 128], "k_shape": [66880, 20, 128], "q_len": 10560, "k_len": 10560, "reference_error": {"max_abs": 0.03125, "rmse": 0.0006830270991789939, "relative_l2": 0.0010006337897907806, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.007719305713610236, "rmse": 0.0007465987096451862, "relative_l2": 0.0016177373650375622, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 44.935489654541016, "p95_ms": 45.00536823272705, "p99_ms": 45.04443565368652, "mean_ms": 44.93828165690104, "n": 300}, {"repeat": 1, "p50_ms": 44.93764686584473, "p95_ms": 45.00893611907959, "p99_ms": 45.02891231536865, "mean_ms": 44.938587239583335, "n": 300}, {"repeat": 2, "p50_ms": 44.937679290771484, "p95_ms": 45.003840637207034, "p99_ms": 45.01679824829102, "mean_ms": 44.938468030293784, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_2da4c70927569963_sdpa_math.json", "configuration": "704x1280_tp1_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp1_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp1_sp2_v1/captures/qkv_rank0_2da4c70927569963.pt", "context": {"phase": "commit", "step": null, "start_frame": 0, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_flash", "status": "measured", "q_shape": [10560, 20, 128], "k_shape": [66880, 20, 128], "q_len": 10560, "k_len": 10560, "reference_error": {"max_abs": 0.03125, "rmse": 0.0003985985876723303, "relative_l2": 0.0005839463995897678, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.008178553063441285, "rmse": 0.0007568670538157323, "relative_l2": 0.001639986913325226, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 3.283184051513672, "p95_ms": 3.5324656009674076, "p99_ms": 3.5830777072906494, "mean_ms": 3.3167426109313967, "n": 300}, {"repeat": 1, "p50_ms": 3.284559965133667, "p95_ms": 3.438715136051178, "p99_ms": 3.4843862986564638, "mean_ms": 3.3165267149607343, "n": 300}, {"repeat": 2, "p50_ms": 3.2883039712905884, "p95_ms": 3.425252711772919, "p99_ms": 3.4323364377021788, "mean_ms": 3.3169387698173525, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_2da4c70927569963_sdpa_flash.json", "configuration": "704x1280_tp1_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp1_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp1_sp2_v1/captures/qkv_rank0_2da4c70927569963.pt", "context": {"phase": "commit", "step": null, "start_frame": 0, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_efficient", "status": "measured", "q_shape": [10560, 20, 128], "k_shape": [66880, 20, 128], "q_len": 10560, "k_len": 10560, "reference_error": {"max_abs": 0.03125, "rmse": 0.0005782278982022229, "relative_l2": 0.0008471031000619522, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.007912222877458852, "rmse": 0.000753810530580441, "relative_l2": 0.0016333640089711773, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 6.3546881675720215, "p95_ms": 6.366257667541504, "p99_ms": 6.391973214149475, "mean_ms": 6.353276901245117, "n": 300}, {"repeat": 1, "p50_ms": 6.3526880741119385, "p95_ms": 6.363249731063843, "p99_ms": 6.369093608856201, "mean_ms": 6.352450876235962, "n": 300}, {"repeat": 2, "p50_ms": 6.353551864624023, "p95_ms": 6.363617587089538, "p99_ms": 6.3699867105484005, "mean_ms": 6.352746979395548, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_2da4c70927569963_sdpa_efficient.json", "configuration": "704x1280_tp1_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp1_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp1_sp2_v1/captures/qkv_rank0_2da4c70927569963.pt", "context": {"phase": "commit", "step": null, "start_frame": 0, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_cudnn", "status": "measured", "q_shape": [10560, 20, 128], "k_shape": [66880, 20, 128], "q_len": 10560, "k_len": 10560, "reference_error": {"max_abs": 0.03125, "rmse": 0.0005754040230860933, "relative_l2": 0.0008429661267811773, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.007912222877458852, "rmse": 0.0007538871835807566, "relative_l2": 0.0016335301014397962, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 2.0377919673919678, "p95_ms": 2.043171262741089, "p99_ms": 2.045889868736267, "mean_ms": 2.0241033609708152, "n": 300}, {"repeat": 1, "p50_ms": 2.03985595703125, "p95_ms": 2.33423513174057, "p99_ms": 2.404707932472229, "mean_ms": 2.0830652821063995, "n": 300}, {"repeat": 2, "p50_ms": 2.026975989341736, "p95_ms": 2.14352171421051, "p99_ms": 2.145275137424469, "mean_ms": 2.038333226442337, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_2da4c70927569963_sdpa_cudnn.json", "configuration": "704x1280_tp1_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp1_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp1_sp2_v1/captures/qkv_rank0_2da4c70927569963.pt", "context": {"phase": "commit", "step": null, "start_frame": 0, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "native_dense", "status": "measured", "q_shape": [10560, 20, 128], "k_shape": [66880, 20, 128], "q_len": 10560, "k_len": 10560, "reference_error": {"max_abs": 0.03125, "rmse": 0.00037952754166402716, "relative_l2": 0.0005560073426101825, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.008178553063441285, "rmse": 0.0007583933289138638, "relative_l2": 0.0016432940610923942, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 1.6921119689941406, "p95_ms": 1.718177604675293, "p99_ms": 1.7209285938739776, "mean_ms": 1.6880451174577078, "n": 300}, {"repeat": 1, "p50_ms": 1.719760000705719, "p95_ms": 1.917931193113327, "p99_ms": 1.919361641407013, "mean_ms": 1.741928639014562, "n": 300}, {"repeat": 2, "p50_ms": 1.6863840222358704, "p95_ms": 1.7018768191337585, "p99_ms": 1.7347574985027314, "mean_ms": 1.6815086913108825, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_2da4c70927569963_native_dense.json", "configuration": "704x1280_tp1_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp1_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp1_sp2_v1/captures/qkv_rank0_3682f84de105b397.pt", "context": {"phase": "commit", "step": null, "start_frame": 27, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "vllm_fa2", "status": "measured", "q_shape": [10560, 20, 128], "k_shape": [66880, 20, 128], "q_len": 10560, "k_len": 63360, "reference_error": {"max_abs": 0.0625, "rmse": 0.00040847303055149484, "relative_l2": 0.0005489455616370788, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.00992615478884451, "rmse": 0.0009261728612270919, "relative_l2": 0.0017342987872258817, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 18.459728240966797, "p95_ms": 18.882421493530273, "p99_ms": 19.849180469512934, "mean_ms": 18.40857624689738, "n": 300}, {"repeat": 1, "p50_ms": 18.461023330688477, "p95_ms": 18.654288482666015, "p99_ms": 18.728085842132568, "mean_ms": 18.422458744049074, "n": 300}, {"repeat": 2, "p50_ms": 18.460944175720215, "p95_ms": 18.64409112930298, "p99_ms": 18.70751834869385, "mean_ms": 18.423856964111327, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_3682f84de105b397_vllm_fa2.json", "configuration": "704x1280_tp1_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp1_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp1_sp2_v1/captures/qkv_rank0_3682f84de105b397.pt", "context": {"phase": "commit", "step": null, "start_frame": 27, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "vllm_fa3", "status": "measured", "q_shape": [10560, 20, 128], "k_shape": [66880, 20, 128], "q_len": 10560, "k_len": 63360, "reference_error": {"max_abs": 0.0, "rmse": 0.0, "relative_l2": 0.0, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.010579403963054013, "rmse": 0.0009308798019245433, "relative_l2": 0.0017431127374989645, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 10.245696067810059, "p95_ms": 10.7887939453125, "p99_ms": 11.356843109130857, "mean_ms": 10.293122666676839, "n": 300}, {"repeat": 1, "p50_ms": 10.253215789794922, "p95_ms": 10.422563457489014, "p99_ms": 10.443972940444946, "mean_ms": 10.262020371754964, "n": 300}, {"repeat": 2, "p50_ms": 10.247632026672363, "p95_ms": 10.40871877670288, "p99_ms": 10.700505199432373, "mean_ms": 10.258898547490437, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_3682f84de105b397_vllm_fa3.json", "configuration": "704x1280_tp1_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp1_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp1_sp2_v1/captures/qkv_rank0_3682f84de105b397.pt", "context": {"phase": "commit", "step": null, "start_frame": 27, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_math", "status": "measured", "q_shape": [10560, 20, 128], "k_shape": [66880, 20, 128], "q_len": 10560, "k_len": 63360, "reference_error": {"max_abs": 0.0625, "rmse": 0.0008374986829189911, "relative_l2": 0.0011255117241022385, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.007813834310592682, "rmse": 0.0009001338955114059, "relative_l2": 0.0016855396963996849, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 280.7530822753906, "p95_ms": 281.823974609375, "p99_ms": 282.4561080932617, "mean_ms": 280.8850690714518, "n": 300}, {"repeat": 1, "p50_ms": 281.0276184082031, "p95_ms": 281.9728775024414, "p99_ms": 282.56541900634767, "mean_ms": 281.0851729329427, "n": 300}, {"repeat": 2, "p50_ms": 280.94532775878906, "p95_ms": 281.9963409423828, "p99_ms": 282.28583251953125, "mean_ms": 281.02482828776044, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_3682f84de105b397_sdpa_math.json", "configuration": "704x1280_tp1_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp1_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp1_sp2_v1/captures/qkv_rank0_3682f84de105b397.pt", "context": {"phase": "commit", "step": null, "start_frame": 27, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_flash", "status": "measured", "q_shape": [10560, 20, 128], "k_shape": [66880, 20, 128], "q_len": 10560, "k_len": 63360, "reference_error": {"max_abs": 0.0625, "rmse": 0.00040918132224603395, "relative_l2": 0.0005498974325146657, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.00992615478884451, "rmse": 0.0009261736710817458, "relative_l2": 0.001734300303713789, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 19.873647689819336, "p95_ms": 20.3908127784729, "p99_ms": 20.712855300903318, "mean_ms": 19.925184078216553, "n": 300}, {"repeat": 1, "p50_ms": 19.88601589202881, "p95_ms": 20.08617753982544, "p99_ms": 20.093803157806395, "mean_ms": 19.906418221791586, "n": 300}, {"repeat": 2, "p50_ms": 19.873583793640137, "p95_ms": 20.050492763519287, "p99_ms": 20.090955810546873, "mean_ms": 19.866901518503823, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_3682f84de105b397_sdpa_flash.json", "configuration": "704x1280_tp1_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp1_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp1_sp2_v1/captures/qkv_rank0_3682f84de105b397.pt", "context": {"phase": "commit", "step": null, "start_frame": 27, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_efficient", "status": "measured", "q_shape": [10560, 20, 128], "k_shape": [66880, 20, 128], "q_len": 10560, "k_len": 63360, "reference_error": {"max_abs": 0.0625, "rmse": 0.000724017075915385, "relative_l2": 0.0009730041658725907, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.008974668806738695, "rmse": 0.0009140189670257117, "relative_l2": 0.0017115400940532074, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 37.81545639038086, "p95_ms": 38.06339740753174, "p99_ms": 38.09622211456299, "mean_ms": 37.865717595418296, "n": 300}, {"repeat": 1, "p50_ms": 37.77217483520508, "p95_ms": 38.06138744354248, "p99_ms": 38.09702129364014, "mean_ms": 37.82964242299398, "n": 300}, {"repeat": 2, "p50_ms": 37.77048110961914, "p95_ms": 37.828484535217285, "p99_ms": 37.83245059967041, "mean_ms": 37.78504997253418, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_3682f84de105b397_sdpa_efficient.json", "configuration": "704x1280_tp1_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp1_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp1_sp2_v1/captures/qkv_rank0_3682f84de105b397.pt", "context": {"phase": "commit", "step": null, "start_frame": 27, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_cudnn", "status": "measured", "q_shape": [10560, 20, 128], "k_shape": [66880, 20, 128], "q_len": 10560, "k_len": 63360, "reference_error": {"max_abs": 0.0625, "rmse": 0.0007238168811113188, "relative_l2": 0.0009727351247341671, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.008974668806738695, "rmse": 0.0009150847472603187, "relative_l2": 0.001713535813692281, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 11.22603178024292, "p95_ms": 11.939892435073853, "p99_ms": 12.897612600326537, "mean_ms": 11.347279380162558, "n": 300}, {"repeat": 1, "p50_ms": 11.365087985992432, "p95_ms": 11.492441558837891, "p99_ms": 11.524550485610963, "mean_ms": 11.356047557195028, "n": 300}, {"repeat": 2, "p50_ms": 11.363279819488525, "p95_ms": 11.492825651168824, "p99_ms": 11.752637071609497, "mean_ms": 11.350723625818889, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_3682f84de105b397_sdpa_cudnn.json", "configuration": "704x1280_tp1_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp1_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp1_sp2_v1/captures/qkv_rank0_3682f84de105b397.pt", "context": {"phase": "commit", "step": null, "start_frame": 27, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "native_dense", "status": "measured", "q_shape": [10560, 20, 128], "k_shape": [66880, 20, 128], "q_len": 10560, "k_len": 63360, "reference_error": {"max_abs": 0.0625, "rmse": 0.0003933845741301415, "relative_l2": 0.0005286682347024865, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.010579403963054013, "rmse": 0.0009298519468354258, "relative_l2": 0.0017411880343370311, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 10.312576293945312, "p95_ms": 10.524300956726075, "p99_ms": 10.57480721473694, "mean_ms": 10.291134955088298, "n": 300}, {"repeat": 1, "p50_ms": 10.184768199920654, "p95_ms": 10.527008152008056, "p99_ms": 10.610217151641846, "mean_ms": 10.264321711858114, "n": 300}, {"repeat": 2, "p50_ms": 10.295392036437988, "p95_ms": 10.56257905960083, "p99_ms": 10.611023578643799, "mean_ms": 10.287673508326213, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_3682f84de105b397_native_dense.json", "configuration": "704x1280_tp1_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp1_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp1_sp2_v1/captures/qkv_rank0_468f321174193448.pt", "context": {"phase": "commit", "step": null, "start_frame": 9, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "vllm_fa2", "status": "measured", "q_shape": [10560, 20, 128], "k_shape": [66880, 20, 128], "q_len": 10560, "k_len": 42240, "reference_error": {"max_abs": 0.0625, "rmse": 0.0004168516215768642, "relative_l2": 0.0005550842714834343, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.014540188720360447, "rmse": 0.0009034149203516021, "relative_l2": 0.001673807418418208, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 12.21662425994873, "p95_ms": 12.444293022155762, "p99_ms": 13.167820653915404, "mean_ms": 12.236574726104736, "n": 300}, {"repeat": 1, "p50_ms": 12.310431957244873, "p95_ms": 12.317227125167847, "p99_ms": 12.321604442596435, "mean_ms": 12.25840542793274, "n": 300}, {"repeat": 2, "p50_ms": 12.249824047088623, "p95_ms": 12.317627429962158, "p99_ms": 12.330353536605834, "mean_ms": 12.25492340405782, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_468f321174193448_vllm_fa2.json", "configuration": "704x1280_tp1_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp1_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp1_sp2_v1/captures/qkv_rank0_468f321174193448.pt", "context": {"phase": "commit", "step": null, "start_frame": 9, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "vllm_fa3", "status": "measured", "q_shape": [10560, 20, 128], "k_shape": [66880, 20, 128], "q_len": 10560, "k_len": 42240, "reference_error": {"max_abs": 0.0, "rmse": 0.0, "relative_l2": 0.0, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.014540188720360447, "rmse": 0.0009039246564090026, "relative_l2": 0.0016747518349593698, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 6.834991931915283, "p95_ms": 7.121860599517823, "p99_ms": 7.35628813266754, "mean_ms": 6.85991583665212, "n": 300}, {"repeat": 1, "p50_ms": 6.8539838790893555, "p95_ms": 6.9616429805755615, "p99_ms": 7.022035722732544, "mean_ms": 6.853254612286886, "n": 300}, {"repeat": 2, "p50_ms": 6.855664014816284, "p95_ms": 6.943481659889221, "p99_ms": 7.0114317941665645, "mean_ms": 6.846350502967835, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_468f321174193448_vllm_fa3.json", "configuration": "704x1280_tp1_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp1_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp1_sp2_v1/captures/qkv_rank0_468f321174193448.pt", "context": {"phase": "commit", "step": null, "start_frame": 9, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_math", "status": "measured", "q_shape": [10560, 20, 128], "k_shape": [66880, 20, 128], "q_len": 10560, "k_len": 42240, "reference_error": {"max_abs": 0.0625, "rmse": 0.0008228731767159028, "relative_l2": 0.0010957471056314036, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.014540188720360447, "rmse": 0.0008877789123615756, "relative_l2": 0.0016448377107251286, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 187.17011260986328, "p95_ms": 188.1791244506836, "p99_ms": 188.36002197265626, "mean_ms": 187.29889979044597, "n": 300}, {"repeat": 1, "p50_ms": 187.5138702392578, "p95_ms": 188.66383361816406, "p99_ms": 189.17776596069336, "mean_ms": 187.66612075805665, "n": 300}, {"repeat": 2, "p50_ms": 187.46004486083984, "p95_ms": 188.41709899902344, "p99_ms": 188.84702880859376, "mean_ms": 187.562449798584, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_468f321174193448_sdpa_math.json", "configuration": "704x1280_tp1_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp1_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp1_sp2_v1/captures/qkv_rank0_468f321174193448.pt", "context": {"phase": "commit", "step": null, "start_frame": 9, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_flash", "status": "measured", "q_shape": [10560, 20, 128], "k_shape": [66880, 20, 128], "q_len": 10560, "k_len": 42240, "reference_error": {"max_abs": 0.0625, "rmse": 0.00041747109442033254, "relative_l2": 0.0005559091684353033, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.014540188720360447, "rmse": 0.0009034055105428151, "relative_l2": 0.001673789984338475, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 13.203423976898193, "p95_ms": 13.599419450759887, "p99_ms": 13.957542753219604, "mean_ms": 13.215591150919597, "n": 300}, {"repeat": 1, "p50_ms": 13.255856037139893, "p95_ms": 13.37065601348877, "p99_ms": 13.381009855270385, "mean_ms": 13.222981967926025, "n": 300}, {"repeat": 2, "p50_ms": 13.225008010864258, "p95_ms": 13.307778024673462, "p99_ms": 13.536196937561035, "mean_ms": 13.209014914830526, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_468f321174193448_sdpa_flash.json", "configuration": "704x1280_tp1_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp1_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp1_sp2_v1/captures/qkv_rank0_468f321174193448.pt", "context": {"phase": "commit", "step": null, "start_frame": 9, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_efficient", "status": "measured", "q_shape": [10560, 20, 128], "k_shape": [66880, 20, 128], "q_len": 10560, "k_len": 42240, "reference_error": {"max_abs": 0.0625, "rmse": 0.0007182619182184227, "relative_l2": 0.0009564455863224964, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.014540188720360447, "rmse": 0.0008957329515698156, "relative_l2": 0.0016595746046298252, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 25.211135864257812, "p95_ms": 25.370758724212646, "p99_ms": 25.374755249023437, "mean_ms": 25.23661003112793, "n": 300}, {"repeat": 1, "p50_ms": 25.19711971282959, "p95_ms": 25.36743278503418, "p99_ms": 25.373855590820312, "mean_ms": 25.20850673039754, "n": 300}, {"repeat": 2, "p50_ms": 25.186256408691406, "p95_ms": 25.223046016693115, "p99_ms": 25.227680530548096, "mean_ms": 25.194311917622883, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_468f321174193448_sdpa_efficient.json", "configuration": "704x1280_tp1_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp1_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp1_sp2_v1/captures/qkv_rank0_468f321174193448.pt", "context": {"phase": "commit", "step": null, "start_frame": 9, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_cudnn", "status": "measured", "q_shape": [10560, 20, 128], "k_shape": [66880, 20, 128], "q_len": 10560, "k_len": 42240, "reference_error": {"max_abs": 0.0625, "rmse": 0.0007165245323870014, "relative_l2": 0.0009541320639596182, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.014540188720360447, "rmse": 0.0008957258433842463, "relative_l2": 0.0016595614348963292, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 7.58568000793457, "p95_ms": 7.722448182106018, "p99_ms": 8.052259559631347, "mean_ms": 7.5863881651560465, "n": 300}, {"repeat": 1, "p50_ms": 7.5862720012664795, "p95_ms": 7.610492944717407, "p99_ms": 7.6294063329696655, "mean_ms": 7.581057279904684, "n": 300}, {"repeat": 2, "p50_ms": 7.579920053482056, "p95_ms": 7.596451139450073, "p99_ms": 7.60873923778534, "mean_ms": 7.573326616287232, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_468f321174193448_sdpa_cudnn.json", "configuration": "704x1280_tp1_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp1_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp1_sp2_v1/captures/qkv_rank0_468f321174193448.pt", "context": {"phase": "commit", "step": null, "start_frame": 9, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "native_dense", "status": "measured", "q_shape": [10560, 20, 128], "k_shape": [66880, 20, 128], "q_len": 10560, "k_len": 42240, "reference_error": {"max_abs": 0.0625, "rmse": 0.0003974722911852968, "relative_l2": 0.0005292785388547646, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.014540188720360447, "rmse": 0.0009027966776015421, "relative_l2": 0.0016726619654506716, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 6.883792161941528, "p95_ms": 7.111944079399109, "p99_ms": 7.524785838127136, "mean_ms": 6.867473910649617, "n": 300}, {"repeat": 1, "p50_ms": 6.916704177856445, "p95_ms": 7.080790543556214, "p99_ms": 7.107847099304199, "mean_ms": 6.86066645304362, "n": 300}, {"repeat": 2, "p50_ms": 6.890463829040527, "p95_ms": 7.110657405853272, "p99_ms": 7.340555419921875, "mean_ms": 6.8519306659698485, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_468f321174193448_native_dense.json", "configuration": "704x1280_tp1_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp1_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp1_sp2_v1/captures/qkv_rank0_61c120430b1af4e0.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 27, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "vllm_fa2", "status": "measured", "q_shape": [10560, 20, 128], "k_shape": [512, 20, 128], "q_len": 10560, "k_len": 512, "reference_error": {"max_abs": 0.000244140625, "rmse": 1.8666221877000994e-06, "relative_l2": 0.0006996105974784194, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 7.818887076814496e-05, "rmse": 5.353783328355767e-06, "relative_l2": 0.0018571216816885075, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.16995200514793396, "p95_ms": 0.1715536020696163, "p99_ms": 0.17193952530622483, "mean_ms": 0.1701809066037337, "n": 300}, {"repeat": 1, "p50_ms": 0.16940799355506897, "p95_ms": 0.17084959745407105, "p99_ms": 0.1715852752327919, "mean_ms": 0.16988202651341755, "n": 300}, {"repeat": 2, "p50_ms": 0.16921600699424744, "p95_ms": 0.17040159404277802, "p99_ms": 0.1710726462304592, "mean_ms": 0.16970485304792723, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_61c120430b1af4e0_vllm_fa2.json", "configuration": "704x1280_tp1_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp1_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp1_sp2_v1/captures/qkv_rank0_61c120430b1af4e0.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 27, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "vllm_fa3", "status": "measured", "q_shape": [10560, 20, 128], "k_shape": [512, 20, 128], "q_len": 10560, "k_len": 512, "reference_error": {"max_abs": 0.000244140625, "rmse": 7.068606837289954e-07, "relative_l2": 0.00026493161205105083, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 7.818887076814496e-05, "rmse": 5.395104933635903e-06, "relative_l2": 0.0018714553303218902, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.1422400027513504, "p95_ms": 0.14422720298171043, "p99_ms": 0.14499615401029586, "mean_ms": 0.1425274667640527, "n": 300}, {"repeat": 1, "p50_ms": 0.14188799262046814, "p95_ms": 0.1441935956478119, "p99_ms": 0.14518591538071632, "mean_ms": 0.14249194661776224, "n": 300}, {"repeat": 2, "p50_ms": 0.14192000031471252, "p95_ms": 0.14406559467315674, "p99_ms": 0.1452188830077648, "mean_ms": 0.1424498133858045, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_61c120430b1af4e0_vllm_fa3.json", "configuration": "704x1280_tp1_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp1_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp1_sp2_v1/captures/qkv_rank0_61c120430b1af4e0.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 27, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_math", "status": "measured", "q_shape": [10560, 20, 128], "k_shape": [512, 20, 128], "q_len": 10560, "k_len": 512, "reference_error": {"max_abs": 0.00048828125, "rmse": 4.549718704673612e-06, "relative_l2": 0.0017052360366814883, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 7.818887076814496e-05, "rmse": 4.7402353330505205e-06, "relative_l2": 0.0016442940017181583, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 2.425328016281128, "p95_ms": 2.4307568311691283, "p99_ms": 2.4335731196403505, "mean_ms": 2.4252964234352112, "n": 300}, {"repeat": 1, "p50_ms": 2.425823926925659, "p95_ms": 2.430948770046234, "p99_ms": 2.433612108230591, "mean_ms": 2.4261080479621886, "n": 300}, {"repeat": 2, "p50_ms": 2.4256319999694824, "p95_ms": 2.431915271282196, "p99_ms": 2.4338579297065737, "mean_ms": 2.4261416482925413, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_61c120430b1af4e0_sdpa_math.json", "configuration": "704x1280_tp1_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp1_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp1_sp2_v1/captures/qkv_rank0_61c120430b1af4e0.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 27, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_flash", "status": "measured", "q_shape": [10560, 20, 128], "k_shape": [512, 20, 128], "q_len": 10560, "k_len": 512, "reference_error": {"max_abs": 0.000244140625, "rmse": 1.8667253138685104e-06, "relative_l2": 0.0006996492491995728, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 7.818887076814496e-05, "rmse": 5.353783328355767e-06, "relative_l2": 0.0018571216816885075, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.18089599907398224, "p95_ms": 0.18201600015163422, "p99_ms": 0.18320224165916443, "mean_ms": 0.1811004799604416, "n": 300}, {"repeat": 1, "p50_ms": 0.18060800433158875, "p95_ms": 0.18143999576568604, "p99_ms": 0.18195487961173057, "mean_ms": 0.18095018669962884, "n": 300}, {"repeat": 2, "p50_ms": 0.1812479943037033, "p95_ms": 0.1822096012532711, "p99_ms": 0.18288031935691834, "mean_ms": 0.18159072016676267, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_61c120430b1af4e0_sdpa_flash.json", "configuration": "704x1280_tp1_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp1_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp1_sp2_v1/captures/qkv_rank0_61c120430b1af4e0.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 27, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_efficient", "status": "measured", "q_shape": [10560, 20, 128], "k_shape": [512, 20, 128], "q_len": 10560, "k_len": 512, "reference_error": {"max_abs": 0.00048828125, "rmse": 3.243521426819271e-06, "relative_l2": 0.0012156728760130148, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 7.818887076814496e-05, "rmse": 5.891822370152756e-06, "relative_l2": 0.002043756797238275, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.33583998680114746, "p95_ms": 0.3370576098561287, "p99_ms": 0.33773119390010836, "mean_ms": 0.3361292804280917, "n": 300}, {"repeat": 1, "p50_ms": 0.3367359936237335, "p95_ms": 0.33772799372673035, "p99_ms": 0.33827456533908845, "mean_ms": 0.3370979209740957, "n": 300}, {"repeat": 2, "p50_ms": 0.33583998680114746, "p95_ms": 0.3367712005972862, "p99_ms": 0.3372483175992966, "mean_ms": 0.33636917332808175, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_61c120430b1af4e0_sdpa_efficient.json", "configuration": "704x1280_tp1_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp1_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp1_sp2_v1/captures/qkv_rank0_61c120430b1af4e0.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 27, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_cudnn", "status": "measured", "q_shape": [10560, 20, 128], "k_shape": [512, 20, 128], "q_len": 10560, "k_len": 512, "reference_error": {"max_abs": 0.00048828125, "rmse": 3.24162616053367e-06, "relative_l2": 0.0012149625295984129, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 7.818887076814496e-05, "rmse": 5.891822370152756e-06, "relative_l2": 0.002043756797238275, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.10260799899697304, "p95_ms": 0.10355679877102375, "p99_ms": 0.10467711985111236, "mean_ms": 0.10289717296759288, "n": 300}, {"repeat": 1, "p50_ms": 0.10288000106811523, "p95_ms": 0.10377760045230389, "p99_ms": 0.10435487739741801, "mean_ms": 0.10329706698656083, "n": 300}, {"repeat": 2, "p50_ms": 0.10288000106811523, "p95_ms": 0.10371359996497631, "p99_ms": 0.10428831689059734, "mean_ms": 0.10333536013960838, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_61c120430b1af4e0_sdpa_cudnn.json", "configuration": "704x1280_tp1_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp1_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp1_sp2_v1/captures/qkv_rank0_61c120430b1af4e0.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 27, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "native_dense", "status": "measured", "q_shape": [10560, 20, 128], "k_shape": [512, 20, 128], "q_len": 10560, "k_len": 512, "reference_error": {"max_abs": 0.0, "rmse": 0.0, "relative_l2": 0.0, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 7.818887076814496e-05, "rmse": 5.410283113664583e-06, "relative_l2": 0.001876720341154611, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.12062400206923485, "p95_ms": 0.17353599369525916, "p99_ms": 0.18544640332460396, "mean_ms": 0.12527754647036393, "n": 300}, {"repeat": 1, "p50_ms": 0.12095999717712402, "p95_ms": 0.13063839673995972, "p99_ms": 0.2011663968861103, "mean_ms": 0.12535733334720134, "n": 300}, {"repeat": 2, "p50_ms": 0.1194240003824234, "p95_ms": 0.12624799981713294, "p99_ms": 0.13183647692203507, "mean_ms": 0.12111402648190657, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_61c120430b1af4e0_native_dense.json", "configuration": "704x1280_tp1_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp1_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp1_sp2_v1/captures/qkv_rank0_c0f6ec677589277b.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 0, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "vllm_fa2", "status": "measured", "q_shape": [10560, 20, 128], "k_shape": [512, 20, 128], "q_len": 10560, "k_len": 512, "reference_error": {"max_abs": 0.00048828125, "rmse": 2.0771213408040955e-06, "relative_l2": 0.0002714058565373477, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.0004067786029473136, "rmse": 1.930963898854266e-05, "relative_l2": 0.0018472856840449385, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.1693599969148636, "p95_ms": 0.1704655945301056, "p99_ms": 0.17091327756643296, "mean_ms": 0.16956042612592379, "n": 300}, {"repeat": 1, "p50_ms": 0.1695999950170517, "p95_ms": 0.1707215964794159, "p99_ms": 0.1711360067129135, "mean_ms": 0.16994720002015432, "n": 300}, {"repeat": 2, "p50_ms": 0.16918399930000305, "p95_ms": 0.17030400037765503, "p99_ms": 0.1708166442811489, "mean_ms": 0.16952864001194637, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_c0f6ec677589277b_vllm_fa2.json", "configuration": "704x1280_tp1_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp1_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp1_sp2_v1/captures/qkv_rank0_c0f6ec677589277b.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 0, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "vllm_fa3", "status": "measured", "q_shape": [10560, 20, 128], "k_shape": [512, 20, 128], "q_len": 10560, "k_len": 512, "reference_error": {"max_abs": 0.00048828125, "rmse": 7.738411245237725e-07, "relative_l2": 0.00010111350218177139, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.0004067786029473136, "rmse": 1.936247177353708e-05, "relative_l2": 0.0018523400119599413, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.1422239989042282, "p95_ms": 0.1444495975971222, "p99_ms": 0.14601600170135498, "mean_ms": 0.14265365342299144, "n": 300}, {"repeat": 1, "p50_ms": 0.14243200421333313, "p95_ms": 0.1451520025730133, "p99_ms": 0.14598527401685715, "mean_ms": 0.14302677298585573, "n": 300}, {"repeat": 2, "p50_ms": 0.14220799505710602, "p95_ms": 0.14473920688033104, "p99_ms": 0.14585823312401772, "mean_ms": 0.14299178640047708, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_c0f6ec677589277b_vllm_fa3.json", "configuration": "704x1280_tp1_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp1_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp1_sp2_v1/captures/qkv_rank0_c0f6ec677589277b.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 0, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_math", "status": "measured", "q_shape": [10560, 20, 128], "k_shape": [512, 20, 128], "q_len": 10560, "k_len": 512, "reference_error": {"max_abs": 0.001953125, "rmse": 9.6886168616582e-06, "relative_l2": 0.0012659575087619018, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.0003842328223661873, "rmse": 1.6787201932200433e-05, "relative_l2": 0.0016059729455804724, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 2.425600051879883, "p95_ms": 2.4313616037368773, "p99_ms": 2.4334217810630796, "mean_ms": 2.425790402889252, "n": 300}, {"repeat": 1, "p50_ms": 2.4260159730911255, "p95_ms": 2.4320720553398134, "p99_ms": 2.434850950241089, "mean_ms": 2.4265400473276775, "n": 300}, {"repeat": 2, "p50_ms": 2.426016092300415, "p95_ms": 2.4317808389663695, "p99_ms": 2.4341481804847716, "mean_ms": 2.426635844707489, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_c0f6ec677589277b_sdpa_math.json", "configuration": "704x1280_tp1_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp1_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp1_sp2_v1/captures/qkv_rank0_c0f6ec677589277b.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 0, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_flash", "status": "measured", "q_shape": [10560, 20, 128], "k_shape": [512, 20, 128], "q_len": 10560, "k_len": 512, "reference_error": {"max_abs": 0.00048828125, "rmse": 2.080157914594089e-06, "relative_l2": 0.00027180262869226337, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.0004067786029473136, "rmse": 1.930963898854266e-05, "relative_l2": 0.0018472856840449385, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.18060800433158875, "p95_ms": 0.18166880458593368, "p99_ms": 0.18288383930921553, "mean_ms": 0.1808996261159579, "n": 300}, {"repeat": 1, "p50_ms": 0.18111999332904816, "p95_ms": 0.18230559527873993, "p99_ms": 0.1832137617468834, "mean_ms": 0.18145280018448828, "n": 300}, {"repeat": 2, "p50_ms": 0.18111999332904816, "p95_ms": 0.1820800006389618, "p99_ms": 0.1829535998404026, "mean_ms": 0.18141631955901782, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_c0f6ec677589277b_sdpa_flash.json", "configuration": "704x1280_tp1_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp1_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp1_sp2_v1/captures/qkv_rank0_c0f6ec677589277b.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 0, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_efficient", "status": "measured", "q_shape": [10560, 20, 128], "k_shape": [512, 20, 128], "q_len": 10560, "k_len": 512, "reference_error": {"max_abs": 0.0009765625, "rmse": 3.643696487662091e-06, "relative_l2": 0.00047610149044697794, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.0004067786029473136, "rmse": 1.9527557065743348e-05, "relative_l2": 0.001868133144970867, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.33638399839401245, "p95_ms": 0.3372816056013107, "p99_ms": 0.3376355203986168, "mean_ms": 0.33649696052074435, "n": 300}, {"repeat": 1, "p50_ms": 0.3360320031642914, "p95_ms": 0.3375072002410889, "p99_ms": 0.33942433387041093, "mean_ms": 0.3364748803774516, "n": 300}, {"repeat": 2, "p50_ms": 0.3364480137825012, "p95_ms": 0.3373775988817215, "p99_ms": 0.3376649680733681, "mean_ms": 0.33674005419015884, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_c0f6ec677589277b_sdpa_efficient.json", "configuration": "704x1280_tp1_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp1_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp1_sp2_v1/captures/qkv_rank0_c0f6ec677589277b.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 0, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_cudnn", "status": "measured", "q_shape": [10560, 20, 128], "k_shape": [512, 20, 128], "q_len": 10560, "k_len": 512, "reference_error": {"max_abs": 0.0009765625, "rmse": 3.6389953194434763e-06, "relative_l2": 0.0004754872150254924, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.0004067786029473136, "rmse": 1.952755709178034e-05, "relative_l2": 0.0018681331474617357, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.10239999741315842, "p95_ms": 0.1034895982593298, "p99_ms": 0.20083615347743022, "mean_ms": 0.10448234674831232, "n": 300}, {"repeat": 1, "p50_ms": 0.10252799838781357, "p95_ms": 0.10335999727249146, "p99_ms": 0.10380864068865776, "mean_ms": 0.10299690653880438, "n": 300}, {"repeat": 2, "p50_ms": 0.10246399790048599, "p95_ms": 0.10332959704101086, "p99_ms": 0.10368127971887589, "mean_ms": 0.10284362643957139, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_c0f6ec677589277b_sdpa_cudnn.json", "configuration": "704x1280_tp1_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp1_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp1_sp2_v1/captures/qkv_rank0_c0f6ec677589277b.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 0, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "native_dense", "status": "measured", "q_shape": [10560, 20, 128], "k_shape": [512, 20, 128], "q_len": 10560, "k_len": 512, "reference_error": {"max_abs": 0.0, "rmse": 0.0, "relative_l2": 0.0, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.0004067786029473136, "rmse": 1.9372534423903118e-05, "relative_l2": 0.001853302670556291, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.1207360029220581, "p95_ms": 0.13282240480184557, "p99_ms": 0.1617107218503952, "mean_ms": 0.12289866680900256, "n": 300}, {"repeat": 1, "p50_ms": 0.12015999853610992, "p95_ms": 0.1290527991950512, "p99_ms": 0.16437216490507098, "mean_ms": 0.1260874665280183, "n": 300}, {"repeat": 2, "p50_ms": 0.12028799951076508, "p95_ms": 0.13474559709429745, "p99_ms": 0.40579935789108273, "mean_ms": 0.1291038935134808, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_c0f6ec677589277b_native_dense.json", "configuration": "704x1280_tp1_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp1_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp1_sp2_v1/captures/qkv_rank0_c334b2b13ebc634a.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 27, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "vllm_fa2", "status": "measured", "q_shape": [10560, 20, 128], "k_shape": [66880, 20, 128], "q_len": 10560, "k_len": 63360, "reference_error": {"max_abs": 0.03125, "rmse": 0.0004269503501298375, "relative_l2": 0.0008745090468842347, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.008982203813774792, "rmse": 0.000752808786776018, "relative_l2": 0.001938066427686598, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 18.460159301757812, "p95_ms": 18.695541191101075, "p99_ms": 18.952612686157224, "mean_ms": 18.402705554962157, "n": 300}, {"repeat": 1, "p50_ms": 18.47158432006836, "p95_ms": 18.655148124694826, "p99_ms": 18.73745674133301, "mean_ms": 18.459552205403646, "n": 300}, {"repeat": 2, "p50_ms": 18.50727939605713, "p95_ms": 18.650563526153565, "p99_ms": 18.707338600158693, "mean_ms": 18.466394163767497, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_c334b2b13ebc634a_vllm_fa2.json", "configuration": "704x1280_tp1_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp1_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp1_sp2_v1/captures/qkv_rank0_c334b2b13ebc634a.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 27, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "vllm_fa3", "status": "measured", "q_shape": [10560, 20, 128], "k_shape": [66880, 20, 128], "q_len": 10560, "k_len": 63360, "reference_error": {"max_abs": 0.0, "rmse": 0.0, "relative_l2": 0.0, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.008982203813774792, "rmse": 0.0007603426648000307, "relative_l2": 0.0019574619984146654, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 10.258960247039795, "p95_ms": 10.733911752700806, "p99_ms": 11.096925745010374, "mean_ms": 10.329833574295044, "n": 300}, {"repeat": 1, "p50_ms": 10.273184299468994, "p95_ms": 10.443777322769165, "p99_ms": 10.545044403076172, "mean_ms": 10.299845457077026, "n": 300}, {"repeat": 2, "p50_ms": 10.270383834838867, "p95_ms": 10.441476535797118, "p99_ms": 10.455860586166382, "mean_ms": 10.296408246358236, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_c334b2b13ebc634a_vllm_fa3.json", "configuration": "704x1280_tp1_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp1_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp1_sp2_v1/captures/qkv_rank0_c334b2b13ebc634a.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 27, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_math", "status": "measured", "q_shape": [10560, 20, 128], "k_shape": [66880, 20, 128], "q_len": 10560, "k_len": 63360, "reference_error": {"max_abs": 0.03125, "rmse": 0.0007713343664216744, "relative_l2": 0.0015799000548971168, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.007363246840188076, "rmse": 0.0006496499745258247, "relative_l2": 0.0016724895185775267, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 280.80604553222656, "p95_ms": 281.449169921875, "p99_ms": 281.83655822753906, "mean_ms": 280.8530523681641, "n": 300}, {"repeat": 1, "p50_ms": 280.85791015625, "p95_ms": 281.4873275756836, "p99_ms": 281.7094110107422, "mean_ms": 280.9086302693685, "n": 300}, {"repeat": 2, "p50_ms": 280.86878967285156, "p95_ms": 281.43027648925784, "p99_ms": 281.8044790649414, "mean_ms": 280.89962107340494, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_c334b2b13ebc634a_sdpa_math.json", "configuration": "704x1280_tp1_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp1_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp1_sp2_v1/captures/qkv_rank0_c334b2b13ebc634a.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 27, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_flash", "status": "measured", "q_shape": [10560, 20, 128], "k_shape": [66880, 20, 128], "q_len": 10560, "k_len": 63360, "reference_error": {"max_abs": 0.03125, "rmse": 0.00042736699286795754, "relative_l2": 0.0008753624431718681, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.008982203813774792, "rmse": 0.0007528591023141345, "relative_l2": 0.001938195962379773, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 19.8853120803833, "p95_ms": 20.3163480758667, "p99_ms": 20.94367980957031, "mean_ms": 19.981972179412843, "n": 300}, {"repeat": 1, "p50_ms": 19.92799949645996, "p95_ms": 20.211969089508056, "p99_ms": 20.214976634979248, "mean_ms": 19.97180617014567, "n": 300}, {"repeat": 2, "p50_ms": 19.907039642333984, "p95_ms": 20.2128023147583, "p99_ms": 20.214752826690674, "mean_ms": 19.96295574824015, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_c334b2b13ebc634a_sdpa_flash.json", "configuration": "704x1280_tp1_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp1_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp1_sp2_v1/captures/qkv_rank0_c334b2b13ebc634a.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 27, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_efficient", "status": "measured", "q_shape": [10560, 20, 128], "k_shape": [66880, 20, 128], "q_len": 10560, "k_len": 63360, "reference_error": {"max_abs": 0.03125, "rmse": 0.000677791661218603, "relative_l2": 0.0013882994579067784, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.00919098333101509, "rmse": 0.0007510854117432321, "relative_l2": 0.001933629689763201, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 37.84350395202637, "p95_ms": 38.11561908721924, "p99_ms": 38.121603660583496, "mean_ms": 37.917093658447264, "n": 300}, {"repeat": 1, "p50_ms": 37.83764839172363, "p95_ms": 38.07996101379395, "p99_ms": 38.10872375488281, "mean_ms": 37.886311798095704, "n": 300}, {"repeat": 2, "p50_ms": 37.82414436340332, "p95_ms": 38.08106117248535, "p99_ms": 38.11431385040283, "mean_ms": 37.876677106221514, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_c334b2b13ebc634a_sdpa_efficient.json", "configuration": "704x1280_tp1_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp1_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp1_sp2_v1/captures/qkv_rank0_c334b2b13ebc634a.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 27, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_cudnn", "status": "measured", "q_shape": [10560, 20, 128], "k_shape": [66880, 20, 128], "q_len": 10560, "k_len": 63360, "reference_error": {"max_abs": 0.03125, "rmse": 0.0006756003474177435, "relative_l2": 0.0013838110583942094, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.00919098333101509, "rmse": 0.0007507384628611804, "relative_l2": 0.001932736488206259, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 11.332335948944092, "p95_ms": 12.033541202545166, "p99_ms": 13.226400260925292, "mean_ms": 11.412516892751057, "n": 300}, {"repeat": 1, "p50_ms": 11.416143894195557, "p95_ms": 11.522656011581422, "p99_ms": 11.584039306640625, "mean_ms": 11.397764469782512, "n": 300}, {"repeat": 2, "p50_ms": 11.367103576660156, "p95_ms": 11.48100438117981, "p99_ms": 11.523070421218872, "mean_ms": 11.379797639846801, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_c334b2b13ebc634a_sdpa_cudnn.json", "configuration": "704x1280_tp1_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp1_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp1_sp2_v1/captures/qkv_rank0_c334b2b13ebc634a.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 27, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "native_dense", "status": "measured", "q_shape": [10560, 20, 128], "k_shape": [66880, 20, 128], "q_len": 10560, "k_len": 63360, "reference_error": {"max_abs": 0.03125, "rmse": 0.0004072238253154325, "relative_l2": 0.000834103823165511, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.008982203813774792, "rmse": 0.0007623522854336237, "relative_l2": 0.0019626356605062473, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 10.340208053588867, "p95_ms": 10.649710321426392, "p99_ms": 11.284195261001587, "mean_ms": 10.341876910527548, "n": 300}, {"repeat": 1, "p50_ms": 10.39680004119873, "p95_ms": 10.59515209197998, "p99_ms": 10.67559278488159, "mean_ms": 10.346855341593425, "n": 300}, {"repeat": 2, "p50_ms": 10.360064029693604, "p95_ms": 10.551106023788453, "p99_ms": 10.640726337432861, "mean_ms": 10.35986131032308, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_c334b2b13ebc634a_native_dense.json", "configuration": "704x1280_tp1_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp1_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp1_sp2_v1/captures/qkv_rank0_cbcd7690956bb027.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 0, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "vllm_fa2", "status": "measured", "q_shape": [10560, 20, 128], "k_shape": [66880, 20, 128], "q_len": 10560, "k_len": 10560, "reference_error": {"max_abs": 0.03125, "rmse": 0.00040687587603831303, "relative_l2": 0.000830508655080701, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.008999900738183353, "rmse": 0.0006804288244503694, "relative_l2": 0.0018858608044861479, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 3.074064016342163, "p95_ms": 3.30574723482132, "p99_ms": 3.4319095158576967, "mean_ms": 3.099271784623464, "n": 300}, {"repeat": 1, "p50_ms": 3.080319881439209, "p95_ms": 3.2546254992485046, "p99_ms": 3.2645794796943663, "mean_ms": 3.099412478605906, "n": 300}, {"repeat": 2, "p50_ms": 3.099423885345459, "p95_ms": 3.1587056636810305, "p99_ms": 3.1889997029304507, "mean_ms": 3.101641167799632, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_cbcd7690956bb027_vllm_fa2.json", "configuration": "704x1280_tp1_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp1_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp1_sp2_v1/captures/qkv_rank0_cbcd7690956bb027.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 0, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "vllm_fa3", "status": "measured", "q_shape": [10560, 20, 128], "k_shape": [66880, 20, 128], "q_len": 10560, "k_len": 10560, "reference_error": {"max_abs": 0.0, "rmse": 0.0, "relative_l2": 0.0, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.008999900738183353, "rmse": 0.0006941992574560287, "relative_l2": 0.0019240266183567644, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 1.778272032737732, "p95_ms": 1.8241936445236206, "p99_ms": 1.8250240087509155, "mean_ms": 1.7584524818261464, "n": 300}, {"repeat": 1, "p50_ms": 1.7801759839057922, "p95_ms": 2.0171584725379943, "p99_ms": 2.0225400948524475, "mean_ms": 1.8139735456307728, "n": 300}, {"repeat": 2, "p50_ms": 1.7729920148849487, "p95_ms": 1.7836191475391387, "p99_ms": 1.7903398525714875, "mean_ms": 1.765483413139979, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_cbcd7690956bb027_vllm_fa3.json", "configuration": "704x1280_tp1_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp1_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp1_sp2_v1/captures/qkv_rank0_cbcd7690956bb027.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 0, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_math", "status": "measured", "q_shape": [10560, 20, 128], "k_shape": [66880, 20, 128], "q_len": 10560, "k_len": 10560, "reference_error": {"max_abs": 0.03125, "rmse": 0.0006770455840095358, "relative_l2": 0.0013819748245559288, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.007237463011211975, "rmse": 0.0005975797342955189, "relative_l2": 0.001656238180934626, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 44.939537048339844, "p95_ms": 45.014467811584474, "p99_ms": 45.052265853881835, "mean_ms": 44.941683565775556, "n": 300}, {"repeat": 1, "p50_ms": 44.932289123535156, "p95_ms": 44.9954174041748, "p99_ms": 45.06025691986084, "mean_ms": 44.93324872334798, "n": 300}, {"repeat": 2, "p50_ms": 44.93011283874512, "p95_ms": 44.98937911987305, "p99_ms": 45.01571201324463, "mean_ms": 44.92791982014974, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_cbcd7690956bb027_sdpa_math.json", "configuration": "704x1280_tp1_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp1_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp1_sp2_v1/captures/qkv_rank0_cbcd7690956bb027.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 0, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_flash", "status": "measured", "q_shape": [10560, 20, 128], "k_shape": [66880, 20, 128], "q_len": 10560, "k_len": 10560, "reference_error": {"max_abs": 0.03125, "rmse": 0.00040714720620537646, "relative_l2": 0.0008310624899610699, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.008999900738183353, "rmse": 0.0006803588393833245, "relative_l2": 0.001885666835491748, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 3.3220800161361694, "p95_ms": 3.5823919057846068, "p99_ms": 3.6175042033195495, "mean_ms": 3.3568761626879375, "n": 300}, {"repeat": 1, "p50_ms": 3.333343982696533, "p95_ms": 3.45878871679306, "p99_ms": 3.4682439398765563, "mean_ms": 3.3527371668815613, "n": 300}, {"repeat": 2, "p50_ms": 3.33788800239563, "p95_ms": 3.4335536479949953, "p99_ms": 3.48428989648819, "mean_ms": 3.359396260579427, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_cbcd7690956bb027_sdpa_flash.json", "configuration": "704x1280_tp1_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp1_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp1_sp2_v1/captures/qkv_rank0_cbcd7690956bb027.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 0, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_efficient", "status": "measured", "q_shape": [10560, 20, 128], "k_shape": [66880, 20, 128], "q_len": 10560, "k_len": 10560, "reference_error": {"max_abs": 0.03125, "rmse": 0.0006288939928471144, "relative_l2": 0.0012836885520797184, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.007237463011211975, "rmse": 0.0006850272498120121, "relative_l2": 0.0018986056939444714, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 6.3976640701293945, "p95_ms": 6.547999954223633, "p99_ms": 6.632480144500732, "mean_ms": 6.409339303970337, "n": 300}, {"repeat": 1, "p50_ms": 6.4074718952178955, "p95_ms": 6.452161478996277, "p99_ms": 6.531973876953125, "mean_ms": 6.414366928736369, "n": 300}, {"repeat": 2, "p50_ms": 6.405264139175415, "p95_ms": 6.539769530296326, "p99_ms": 6.608355650901794, "mean_ms": 6.409651521046956, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_cbcd7690956bb027_sdpa_efficient.json", "configuration": "704x1280_tp1_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp1_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp1_sp2_v1/captures/qkv_rank0_cbcd7690956bb027.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 0, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_cudnn", "status": "measured", "q_shape": [10560, 20, 128], "k_shape": [66880, 20, 128], "q_len": 10560, "k_len": 10560, "reference_error": {"max_abs": 0.03125, "rmse": 0.0006274993634227908, "relative_l2": 0.001280841856377804, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.007237463011211975, "rmse": 0.0006857538348473918, "relative_l2": 0.0019006194802364555, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 2.0658719539642334, "p95_ms": 2.086305546760559, "p99_ms": 2.0958105897903443, "mean_ms": 2.064460912148158, "n": 300}, {"repeat": 1, "p50_ms": 2.0860159397125244, "p95_ms": 2.385119915008545, "p99_ms": 2.3870402216911315, "mean_ms": 2.1164071575800576, "n": 300}, {"repeat": 2, "p50_ms": 2.0795199871063232, "p95_ms": 2.1702415347099304, "p99_ms": 2.1709309577941895, "mean_ms": 2.088467100461324, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_cbcd7690956bb027_sdpa_cudnn.json", "configuration": "704x1280_tp1_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp1_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp1_sp2_v1/captures/qkv_rank0_cbcd7690956bb027.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 0, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "native_dense", "status": "measured", "q_shape": [10560, 20, 128], "k_shape": [66880, 20, 128], "q_len": 10560, "k_len": 10560, "reference_error": {"max_abs": 0.03125, "rmse": 0.00038511026327172455, "relative_l2": 0.0007860810277615414, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.008999900738183353, "rmse": 0.0006932121785677012, "relative_l2": 0.0019212908533222126, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 1.7579840421676636, "p95_ms": 1.7758111774921417, "p99_ms": 1.781413712501526, "mean_ms": 1.7442648522059123, "n": 300}, {"repeat": 1, "p50_ms": 1.7580320239067078, "p95_ms": 1.9055071890354156, "p99_ms": 1.910322241783142, "mean_ms": 1.7759907229741414, "n": 300}, {"repeat": 2, "p50_ms": 1.7654240131378174, "p95_ms": 1.7839695751667022, "p99_ms": 1.792530938386917, "mean_ms": 1.7614090653260548, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_cbcd7690956bb027_native_dense.json", "configuration": "704x1280_tp1_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp1_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp1_sp2_v1/captures/qkv_rank0_dfcab6f0a5779c76.pt", "context": {"phase": "commit", "step": null, "start_frame": 0, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "vllm_fa2", "status": "measured", "q_shape": [10560, 20, 128], "k_shape": [512, 20, 128], "q_len": 10560, "k_len": 512, "reference_error": {"max_abs": 0.0009765625, "rmse": 1.8949066567348784e-06, "relative_l2": 0.00024318943814230962, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.00043489744124647345, "rmse": 1.7589970025427215e-05, "relative_l2": 0.001761251841652134, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.17049600183963776, "p95_ms": 0.17152319476008415, "p99_ms": 0.1721615995466709, "mean_ms": 0.17053653314709663, "n": 300}, {"repeat": 1, "p50_ms": 0.16915200650691986, "p95_ms": 0.1701440066099167, "p99_ms": 0.17107392609119415, "mean_ms": 0.16957706660032273, "n": 300}, {"repeat": 2, "p50_ms": 0.16915200650691986, "p95_ms": 0.1700800061225891, "p99_ms": 0.17146463423967362, "mean_ms": 0.16953151961167653, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_dfcab6f0a5779c76_vllm_fa2.json", "configuration": "704x1280_tp1_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp1_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp1_sp2_v1/captures/qkv_rank0_dfcab6f0a5779c76.pt", "context": {"phase": "commit", "step": null, "start_frame": 0, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "vllm_fa3", "status": "measured", "q_shape": [10560, 20, 128], "k_shape": [512, 20, 128], "q_len": 10560, "k_len": 512, "reference_error": {"max_abs": 0.00048828125, "rmse": 7.010055343405709e-07, "relative_l2": 8.996598403674556e-05, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.00043489744124647345, "rmse": 1.7598296726068673e-05, "relative_l2": 0.0017620855791069708, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.14153599739074707, "p95_ms": 0.14368640631437302, "p99_ms": 0.14432095631957054, "mean_ms": 0.14180234695474306, "n": 300}, {"repeat": 1, "p50_ms": 0.14161600172519684, "p95_ms": 0.14339519664645195, "p99_ms": 0.1446415987610817, "mean_ms": 0.1421477335691452, "n": 300}, {"repeat": 2, "p50_ms": 0.14159999787807465, "p95_ms": 0.14336000382900238, "p99_ms": 0.1449939139187336, "mean_ms": 0.1421386665602525, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_dfcab6f0a5779c76_vllm_fa3.json", "configuration": "704x1280_tp1_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp1_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp1_sp2_v1/captures/qkv_rank0_dfcab6f0a5779c76.pt", "context": {"phase": "commit", "step": null, "start_frame": 0, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_math", "status": "measured", "q_shape": [10560, 20, 128], "k_shape": [512, 20, 128], "q_len": 10560, "k_len": 512, "reference_error": {"max_abs": 0.001953125, "rmse": 4.708450836802924e-06, "relative_l2": 0.0006042754187669717, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.00043489744124647345, "rmse": 1.747372544361234e-05, "relative_l2": 0.0017496124822042455, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 2.425600051879883, "p95_ms": 2.4302143692970275, "p99_ms": 2.4321627807617188, "mean_ms": 2.4255665055910747, "n": 300}, {"repeat": 1, "p50_ms": 2.42574405670166, "p95_ms": 2.4305007815361024, "p99_ms": 2.4323575043678285, "mean_ms": 2.425555734634399, "n": 300}, {"repeat": 2, "p50_ms": 2.42574405670166, "p95_ms": 2.4304656267166136, "p99_ms": 2.4332218551635743, "mean_ms": 2.425732593536377, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_dfcab6f0a5779c76_sdpa_math.json", "configuration": "704x1280_tp1_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp1_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp1_sp2_v1/captures/qkv_rank0_dfcab6f0a5779c76.pt", "context": {"phase": "commit", "step": null, "start_frame": 0, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_flash", "status": "measured", "q_shape": [10560, 20, 128], "k_shape": [512, 20, 128], "q_len": 10560, "k_len": 512, "reference_error": {"max_abs": 0.0009765625, "rmse": 1.8922107054847972e-06, "relative_l2": 0.00024284344385945878, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.00043489744124647345, "rmse": 1.7589970025427215e-05, "relative_l2": 0.001761251841652134, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.18067200481891632, "p95_ms": 0.18163199722766876, "p99_ms": 0.18265696495771408, "mean_ms": 0.18085930705070496, "n": 300}, {"repeat": 1, "p50_ms": 0.1804479956626892, "p95_ms": 0.18150559663772584, "p99_ms": 0.18173056542873384, "mean_ms": 0.18077888001998266, "n": 300}, {"repeat": 2, "p50_ms": 0.18051199615001678, "p95_ms": 0.181536003947258, "p99_ms": 0.18185696631669998, "mean_ms": 0.18086816012859344, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_dfcab6f0a5779c76_sdpa_flash.json", "configuration": "704x1280_tp1_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp1_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp1_sp2_v1/captures/qkv_rank0_dfcab6f0a5779c76.pt", "context": {"phase": "commit", "step": null, "start_frame": 0, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_efficient", "status": "measured", "q_shape": [10560, 20, 128], "k_shape": [512, 20, 128], "q_len": 10560, "k_len": 512, "reference_error": {"max_abs": 0.0009765625, "rmse": 3.2812566507897963e-06, "relative_l2": 0.00042111148771901383, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.00043489744124647345, "rmse": 1.7758429651179162e-05, "relative_l2": 0.0017781193988833715, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.33542400598526, "p95_ms": 0.3362559974193573, "p99_ms": 0.33657665461301806, "mean_ms": 0.33556885341803233, "n": 300}, {"repeat": 1, "p50_ms": 0.33555200695991516, "p95_ms": 0.33632001280784607, "p99_ms": 0.3370569708943367, "mean_ms": 0.3358679462472598, "n": 300}, {"repeat": 2, "p50_ms": 0.3353919982910156, "p95_ms": 0.33638559877872465, "p99_ms": 0.3369286498427391, "mean_ms": 0.3356454407175382, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_dfcab6f0a5779c76_sdpa_efficient.json", "configuration": "704x1280_tp1_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp1_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp1_sp2_v1/captures/qkv_rank0_dfcab6f0a5779c76.pt", "context": {"phase": "commit", "step": null, "start_frame": 0, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_cudnn", "status": "measured", "q_shape": [10560, 20, 128], "k_shape": [512, 20, 128], "q_len": 10560, "k_len": 512, "reference_error": {"max_abs": 0.0009765625, "rmse": 3.274146084048062e-06, "relative_l2": 0.0004201989283986643, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.00043489744124647345, "rmse": 1.7758429651179162e-05, "relative_l2": 0.0017781193988833715, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.10252799838781357, "p95_ms": 0.10342399775981903, "p99_ms": 0.10371199995279312, "mean_ms": 0.10307647975782554, "n": 300}, {"repeat": 1, "p50_ms": 0.10300800204277039, "p95_ms": 0.10393600165843964, "p99_ms": 0.10445215813815593, "mean_ms": 0.10350026679535707, "n": 300}, {"repeat": 2, "p50_ms": 0.10265599936246872, "p95_ms": 0.10349119827151299, "p99_ms": 0.10448287837207317, "mean_ms": 0.10307957326372465, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_dfcab6f0a5779c76_sdpa_cudnn.json", "configuration": "704x1280_tp1_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp1_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp1_sp2_v1/captures/qkv_rank0_dfcab6f0a5779c76.pt", "context": {"phase": "commit", "step": null, "start_frame": 0, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "native_dense", "status": "measured", "q_shape": [10560, 20, 128], "k_shape": [512, 20, 128], "q_len": 10560, "k_len": 512, "reference_error": {"max_abs": 0.0, "rmse": 0.0, "relative_l2": 0.0, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.00043489744124647345, "rmse": 1.7601063432965562e-05, "relative_l2": 0.001762362604457805, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.12043200060725212, "p95_ms": 0.13290239870548248, "p99_ms": 0.15284992709755876, "mean_ms": 0.12236906665066878, "n": 300}, {"repeat": 1, "p50_ms": 0.12028799951076508, "p95_ms": 0.13034240156412125, "p99_ms": 0.1581270429491995, "mean_ms": 0.12259776016076405, "n": 300}, {"repeat": 2, "p50_ms": 0.12103999778628349, "p95_ms": 0.13235200196504593, "p99_ms": 0.17713216006755828, "mean_ms": 0.12345440037548543, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_dfcab6f0a5779c76_native_dense.json", "configuration": "704x1280_tp1_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp1_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp1_sp2_v1/captures/qkv_rank0_f5a94ff018466c49.pt", "context": {"phase": "commit", "step": null, "start_frame": 27, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "vllm_fa2", "status": "measured", "q_shape": [10560, 20, 128], "k_shape": [512, 20, 128], "q_len": 10560, "k_len": 512, "reference_error": {"max_abs": 0.00048828125, "rmse": 1.6936418731255048e-06, "relative_l2": 0.000667124739076655, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 6.248058869558204e-05, "rmse": 3.99686897246697e-06, "relative_l2": 0.0017776198736656203, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.16979199647903442, "p95_ms": 0.17072319611907005, "p99_ms": 0.17123231992125512, "mean_ms": 0.16993258679906528, "n": 300}, {"repeat": 1, "p50_ms": 0.16963200271129608, "p95_ms": 0.17062560245394706, "p99_ms": 0.1713286332786083, "mean_ms": 0.1699796264866988, "n": 300}, {"repeat": 2, "p50_ms": 0.16910400241613388, "p95_ms": 0.17027359306812287, "p99_ms": 0.17091231763362885, "mean_ms": 0.1696020263930162, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_f5a94ff018466c49_vllm_fa2.json", "configuration": "704x1280_tp1_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp1_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp1_sp2_v1/captures/qkv_rank0_f5a94ff018466c49.pt", "context": {"phase": "commit", "step": null, "start_frame": 27, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "vllm_fa3", "status": "measured", "q_shape": [10560, 20, 128], "k_shape": [512, 20, 128], "q_len": 10560, "k_len": 512, "reference_error": {"max_abs": 0.000244140625, "rmse": 6.490455127070695e-07, "relative_l2": 0.0002556587228883961, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 6.248058869558204e-05, "rmse": 4.060999388811001e-06, "relative_l2": 0.001806142075265148, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.1414719969034195, "p95_ms": 0.14406879469752312, "p99_ms": 0.14492927342653275, "mean_ms": 0.14182517324884733, "n": 300}, {"repeat": 1, "p50_ms": 0.14140799641609192, "p95_ms": 0.1438400000333786, "p99_ms": 0.1443520039319992, "mean_ms": 0.14196394711732865, "n": 300}, {"repeat": 2, "p50_ms": 0.141184002161026, "p95_ms": 0.143745606392622, "p99_ms": 0.144674886316061, "mean_ms": 0.1418315732975801, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_f5a94ff018466c49_vllm_fa3.json", "configuration": "704x1280_tp1_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp1_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp1_sp2_v1/captures/qkv_rank0_f5a94ff018466c49.pt", "context": {"phase": "commit", "step": null, "start_frame": 27, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_math", "status": "measured", "q_shape": [10560, 20, 128], "k_shape": [512, 20, 128], "q_len": 10560, "k_len": 512, "reference_error": {"max_abs": 0.0009765625, "rmse": 4.455521917371253e-06, "relative_l2": 0.0017550279924830076, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 5.958972380441796e-05, "rmse": 3.7838496823589655e-06, "relative_l2": 0.0016828788836109959, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 2.426735997200012, "p95_ms": 2.432070457935333, "p99_ms": 2.4346575045585634, "mean_ms": 2.4271850593884787, "n": 300}, {"repeat": 1, "p50_ms": 2.4271039962768555, "p95_ms": 2.4321936845779417, "p99_ms": 2.4336775231361387, "mean_ms": 2.4276087403297426, "n": 300}, {"repeat": 2, "p50_ms": 2.427248001098633, "p95_ms": 2.4333438873291016, "p99_ms": 2.435559883117676, "mean_ms": 2.4273852729797363, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_f5a94ff018466c49_sdpa_math.json", "configuration": "704x1280_tp1_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp1_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp1_sp2_v1/captures/qkv_rank0_f5a94ff018466c49.pt", "context": {"phase": "commit", "step": null, "start_frame": 27, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_flash", "status": "measured", "q_shape": [10560, 20, 128], "k_shape": [512, 20, 128], "q_len": 10560, "k_len": 512, "reference_error": {"max_abs": 0.00048828125, "rmse": 1.6940296827095136e-06, "relative_l2": 0.000667277497089815, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 6.248058869558204e-05, "rmse": 3.99686897246697e-06, "relative_l2": 0.0017776198736656203, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.17936000227928162, "p95_ms": 0.1804511956870556, "p99_ms": 0.18115200102329254, "mean_ms": 0.17960223938028017, "n": 300}, {"repeat": 1, "p50_ms": 0.17980800569057465, "p95_ms": 0.1808335989713669, "p99_ms": 0.18112223342061043, "mean_ms": 0.18010954707860946, "n": 300}, {"repeat": 2, "p50_ms": 0.17982400208711624, "p95_ms": 0.18118399381637573, "p99_ms": 0.18221728160977363, "mean_ms": 0.18033226683735848, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_f5a94ff018466c49_sdpa_flash.json", "configuration": "704x1280_tp1_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp1_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp1_sp2_v1/captures/qkv_rank0_f5a94ff018466c49.pt", "context": {"phase": "commit", "step": null, "start_frame": 27, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_efficient", "status": "measured", "q_shape": [10560, 20, 128], "k_shape": [512, 20, 128], "q_len": 10560, "k_len": 512, "reference_error": {"max_abs": 0.00048828125, "rmse": 2.9498381981983565e-06, "relative_l2": 0.0011619398820482516, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 6.248058869558204e-05, "rmse": 4.753777167533007e-06, "relative_l2": 0.0021142571413264835, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.3357760012149811, "p95_ms": 0.3370240032672882, "p99_ms": 0.3375999927520752, "mean_ms": 0.33596330682436626, "n": 300}, {"repeat": 1, "p50_ms": 0.3357439935207367, "p95_ms": 0.336544007062912, "p99_ms": 0.3369283297657967, "mean_ms": 0.3360435202717781, "n": 300}, {"repeat": 2, "p50_ms": 0.3353919982910156, "p95_ms": 0.33641920536756514, "p99_ms": 0.33737951844930647, "mean_ms": 0.3360544006029765, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_f5a94ff018466c49_sdpa_efficient.json", "configuration": "704x1280_tp1_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp1_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp1_sp2_v1/captures/qkv_rank0_f5a94ff018466c49.pt", "context": {"phase": "commit", "step": null, "start_frame": 27, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_cudnn", "status": "measured", "q_shape": [10560, 20, 128], "k_shape": [512, 20, 128], "q_len": 10560, "k_len": 512, "reference_error": {"max_abs": 0.00048828125, "rmse": 2.9471226355762685e-06, "relative_l2": 0.0011608702232056985, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 6.248058869558204e-05, "rmse": 4.753777266973011e-06, "relative_l2": 0.0021142571855527334, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.10227199643850327, "p95_ms": 0.10313760302960873, "p99_ms": 0.10361599922180176, "mean_ms": 0.1025696001201868, "n": 300}, {"repeat": 1, "p50_ms": 0.10233599692583084, "p95_ms": 0.10316960327327251, "p99_ms": 0.10377984046936035, "mean_ms": 0.103097173422575, "n": 300}, {"repeat": 2, "p50_ms": 0.10224000364542007, "p95_ms": 0.10313760302960873, "p99_ms": 0.10352031849324703, "mean_ms": 0.10325066678225994, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_f5a94ff018466c49_sdpa_cudnn.json", "configuration": "704x1280_tp1_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp1_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp1_sp2_v1/captures/qkv_rank0_f5a94ff018466c49.pt", "context": {"phase": "commit", "step": null, "start_frame": 27, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "native_dense", "status": "measured", "q_shape": [10560, 20, 128], "k_shape": [512, 20, 128], "q_len": 10560, "k_len": 512, "reference_error": {"max_abs": 0.0, "rmse": 0.0, "relative_l2": 0.0, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 6.248058869558204e-05, "rmse": 4.076969633579108e-06, "relative_l2": 0.0018132448911649572, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.12089599668979645, "p95_ms": 0.13413120061159134, "p99_ms": 0.16192031621932984, "mean_ms": 0.12324864027400811, "n": 300}, {"repeat": 1, "p50_ms": 0.12081600353121758, "p95_ms": 0.13201439902186393, "p99_ms": 0.17962336078286129, "mean_ms": 0.12513407997787, "n": 300}, {"repeat": 2, "p50_ms": 0.1207360029220581, "p95_ms": 0.12819519266486168, "p99_ms": 0.15576703369617456, "mean_ms": 0.12258261335392794, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_f5a94ff018466c49_native_dense.json", "configuration": "704x1280_tp1_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp1_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp1_sp2_v1/captures/qkv_rank0_ff91739f1c7bab2a.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 9, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "vllm_fa2", "status": "measured", "q_shape": [10560, 20, 128], "k_shape": [512, 20, 128], "q_len": 10560, "k_len": 512, "reference_error": {"max_abs": 0.000244140625, "rmse": 1.8963839082334455e-06, "relative_l2": 0.0007060465351562547, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 6.37440089465216e-05, "rmse": 5.101539129336812e-06, "relative_l2": 0.0018193710309880842, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.16937600076198578, "p95_ms": 0.17043360099196433, "p99_ms": 0.17094624519348145, "mean_ms": 0.16957535997033119, "n": 300}, {"repeat": 1, "p50_ms": 0.1695999950170517, "p95_ms": 0.17059199512004852, "p99_ms": 0.1710083256661892, "mean_ms": 0.1699980797370275, "n": 300}, {"repeat": 2, "p50_ms": 0.16870400309562683, "p95_ms": 0.1705999955534935, "p99_ms": 0.1713616009056568, "mean_ms": 0.16933493345975875, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_ff91739f1c7bab2a_vllm_fa2.json", "configuration": "704x1280_tp1_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp1_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp1_sp2_v1/captures/qkv_rank0_ff91739f1c7bab2a.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 9, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "vllm_fa3", "status": "measured", "q_shape": [10560, 20, 128], "k_shape": [512, 20, 128], "q_len": 10560, "k_len": 512, "reference_error": {"max_abs": 0.000244140625, "rmse": 7.056971398700347e-07, "relative_l2": 0.00026273953196484387, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 6.37440089465216e-05, "rmse": 5.160676778595407e-06, "relative_l2": 0.0018404613966942096, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.14166399836540222, "p95_ms": 0.14371519908308983, "p99_ms": 0.14477023959159852, "mean_ms": 0.14192298675576845, "n": 300}, {"repeat": 1, "p50_ms": 0.14161600172519684, "p95_ms": 0.1440352015197277, "p99_ms": 0.1447692796587944, "mean_ms": 0.142257386893034, "n": 300}, {"repeat": 2, "p50_ms": 0.14124800264835358, "p95_ms": 0.14358559846878052, "p99_ms": 0.14432447627186776, "mean_ms": 0.14183327987790106, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_ff91739f1c7bab2a_vllm_fa3.json", "configuration": "704x1280_tp1_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp1_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp1_sp2_v1/captures/qkv_rank0_ff91739f1c7bab2a.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 9, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_math", "status": "measured", "q_shape": [10560, 20, 128], "k_shape": [512, 20, 128], "q_len": 10560, "k_len": 512, "reference_error": {"max_abs": 0.00048828125, "rmse": 4.573574597457332e-06, "relative_l2": 0.0017027968249432657, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 6.09328142618451e-05, "rmse": 4.570944062662757e-06, "relative_l2": 0.0016301439626430338, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 2.4266560077667236, "p95_ms": 2.432096004486084, "p99_ms": 2.4343413877487183, "mean_ms": 2.4268354161580405, "n": 300}, {"repeat": 1, "p50_ms": 2.427120089530945, "p95_ms": 2.4324239611625673, "p99_ms": 2.434530670642853, "mean_ms": 2.427630611260732, "n": 300}, {"repeat": 2, "p50_ms": 2.4267520904541016, "p95_ms": 2.431905674934387, "p99_ms": 2.433133132457733, "mean_ms": 2.4270580267906188, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_ff91739f1c7bab2a_sdpa_math.json", "configuration": "704x1280_tp1_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp1_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp1_sp2_v1/captures/qkv_rank0_ff91739f1c7bab2a.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 9, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_flash", "status": "measured", "q_shape": [10560, 20, 128], "k_shape": [512, 20, 128], "q_len": 10560, "k_len": 512, "reference_error": {"max_abs": 0.000244140625, "rmse": 1.8967133149094004e-06, "relative_l2": 0.0007061691772231935, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 6.37440089465216e-05, "rmse": 5.101539129336812e-06, "relative_l2": 0.0018193710309880842, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.1804800033569336, "p95_ms": 0.18147360309958457, "p99_ms": 0.18223999440670013, "mean_ms": 0.18051253353555996, "n": 300}, {"repeat": 1, "p50_ms": 0.18067200481891632, "p95_ms": 0.18208320066332817, "p99_ms": 0.18288415923714638, "mean_ms": 0.18105152000983557, "n": 300}, {"repeat": 2, "p50_ms": 0.1804800033569336, "p95_ms": 0.1813135951757431, "p99_ms": 0.18198976725339888, "mean_ms": 0.1808220793803533, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_ff91739f1c7bab2a_sdpa_flash.json", "configuration": "704x1280_tp1_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp1_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp1_sp2_v1/captures/qkv_rank0_ff91739f1c7bab2a.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 9, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_efficient", "status": "measured", "q_shape": [10560, 20, 128], "k_shape": [512, 20, 128], "q_len": 10560, "k_len": 512, "reference_error": {"max_abs": 0.00048828125, "rmse": 3.3039338400295354e-06, "relative_l2": 0.0012300943021138486, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 6.37440089465216e-05, "rmse": 5.786214176812004e-06, "relative_l2": 0.0020635479186754463, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.33657601475715637, "p95_ms": 0.337537607550621, "p99_ms": 0.33830433279275895, "mean_ms": 0.3366730665167173, "n": 300}, {"repeat": 1, "p50_ms": 0.3360320031642914, "p95_ms": 0.3370879888534546, "p99_ms": 0.33760095268487933, "mean_ms": 0.33634869327147804, "n": 300}, {"repeat": 2, "p50_ms": 0.33558401465415955, "p95_ms": 0.3365119993686676, "p99_ms": 0.3370566508173943, "mean_ms": 0.3359575463334719, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_ff91739f1c7bab2a_sdpa_efficient.json", "configuration": "704x1280_tp1_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp1_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp1_sp2_v1/captures/qkv_rank0_ff91739f1c7bab2a.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 9, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_cudnn", "status": "measured", "q_shape": [10560, 20, 128], "k_shape": [512, 20, 128], "q_len": 10560, "k_len": 512, "reference_error": {"max_abs": 0.00048828125, "rmse": 3.3020913380833707e-06, "relative_l2": 0.0012294083164811611, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 6.37440089465216e-05, "rmse": 5.786232531806798e-06, "relative_l2": 0.002063554464650093, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.10233599692583084, "p95_ms": 0.10320000350475311, "p99_ms": 0.10361663922667504, "mean_ms": 0.10261258656779924, "n": 300}, {"repeat": 1, "p50_ms": 0.10255999863147736, "p95_ms": 0.10335999727249146, "p99_ms": 0.10377984046936035, "mean_ms": 0.10302591947217782, "n": 300}, {"repeat": 2, "p50_ms": 0.10252799838781357, "p95_ms": 0.10342399775981903, "p99_ms": 0.10387392118573188, "mean_ms": 0.10319989316165447, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_ff91739f1c7bab2a_sdpa_cudnn.json", "configuration": "704x1280_tp1_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp1_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp1_sp2_v1/captures/qkv_rank0_ff91739f1c7bab2a.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 9, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "native_dense", "status": "measured", "q_shape": [10560, 20, 128], "k_shape": [512, 20, 128], "q_len": 10560, "k_len": 512, "reference_error": {"max_abs": 0.0, "rmse": 0.0, "relative_l2": 0.0, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 6.37440089465216e-05, "rmse": 5.166876854269409e-06, "relative_l2": 0.0018426725407794024, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.12171200290322304, "p95_ms": 0.16247040629386905, "p99_ms": 0.19130176305770866, "mean_ms": 0.1275009067853292, "n": 300}, {"repeat": 1, "p50_ms": 0.1202399991452694, "p95_ms": 0.12714720666408538, "p99_ms": 0.1643936064839362, "mean_ms": 0.12214751998583476, "n": 300}, {"repeat": 2, "p50_ms": 0.12095999717712402, "p95_ms": 0.13499200493097313, "p99_ms": 0.22915359303355204, "mean_ms": 0.12637834635873635, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_ff91739f1c7bab2a_native_dense.json", "configuration": "704x1280_tp1_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp1_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp2_sp1_v1/captures/qkv_rank0_021f565f50432429.pt", "context": {"phase": "commit", "step": null, "start_frame": 9, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "vllm_fa2", "status": "measured", "q_shape": [10560, 20, 128], "k_shape": [512, 20, 128], "q_len": 10560, "k_len": 512, "reference_error": {"max_abs": 0.0009765625, "rmse": 1.7280877346338352e-06, "relative_l2": 0.0005529092172008231, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.00012640277610309614, "rmse": 4.8645140986324285e-06, "relative_l2": 0.0017539323332368927, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.16860799491405487, "p95_ms": 0.16944000124931335, "p99_ms": 0.16976096361875534, "mean_ms": 0.16874976004163425, "n": 300}, {"repeat": 1, "p50_ms": 0.16857600212097168, "p95_ms": 0.16940799355506897, "p99_ms": 0.1698899246752262, "mean_ms": 0.16966581354538599, "n": 300}, {"repeat": 2, "p50_ms": 0.1685279980301857, "p95_ms": 0.16940799355506897, "p99_ms": 0.1696025550365448, "mean_ms": 0.16887018645803134, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_021f565f50432429_vllm_fa2.json", "configuration": "704x1280_tp2_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp2_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp2_sp1_v1/captures/qkv_rank0_021f565f50432429.pt", "context": {"phase": "commit", "step": null, "start_frame": 9, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "vllm_fa3", "status": "measured", "q_shape": [10560, 20, 128], "k_shape": [512, 20, 128], "q_len": 10560, "k_len": 512, "reference_error": {"max_abs": 0.0009765625, "rmse": 6.77654186299376e-07, "relative_l2": 0.00021681841620098066, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.00012640277610309614, "rmse": 4.8949696271402184e-06, "relative_l2": 0.0017649132729756936, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.14076800644397736, "p95_ms": 0.14224160239100456, "p99_ms": 0.14307519420981407, "mean_ms": 0.14105322708686194, "n": 300}, {"repeat": 1, "p50_ms": 0.1408960074186325, "p95_ms": 0.14201759397983552, "p99_ms": 0.14266207844018935, "mean_ms": 0.14187008008360863, "n": 300}, {"repeat": 2, "p50_ms": 0.1409599930047989, "p95_ms": 0.14233919605612755, "p99_ms": 0.1432003152370453, "mean_ms": 0.14160202672084174, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_021f565f50432429_vllm_fa3.json", "configuration": "704x1280_tp2_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp2_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp2_sp1_v1/captures/qkv_rank0_021f565f50432429.pt", "context": {"phase": "commit", "step": null, "start_frame": 9, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_math", "status": "measured", "q_shape": [10560, 20, 128], "k_shape": [512, 20, 128], "q_len": 10560, "k_len": 512, "reference_error": {"max_abs": 0.0009765625, "rmse": 5.389324976312201e-06, "relative_l2": 0.001724338061183575, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.00011773784889690386, "rmse": 4.642962646072207e-06, "relative_l2": 0.001674050509843634, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 2.4241440296173096, "p95_ms": 2.4286414980888367, "p99_ms": 2.435014820098877, "mean_ms": 2.4241043281555177, "n": 300}, {"repeat": 1, "p50_ms": 2.423583984375, "p95_ms": 2.4272655725479124, "p99_ms": 2.429376208782196, "mean_ms": 2.423979624907176, "n": 300}, {"repeat": 2, "p50_ms": 2.423856019973755, "p95_ms": 2.4283888578414916, "p99_ms": 2.430148870944977, "mean_ms": 2.423761388460795, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_021f565f50432429_sdpa_math.json", "configuration": "704x1280_tp2_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp2_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp2_sp1_v1/captures/qkv_rank0_021f565f50432429.pt", "context": {"phase": "commit", "step": null, "start_frame": 9, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_flash", "status": "measured", "q_shape": [10560, 20, 128], "k_shape": [512, 20, 128], "q_len": 10560, "k_len": 512, "reference_error": {"max_abs": 0.0009765625, "rmse": 1.7453262657288557e-06, "relative_l2": 0.0005584247605047974, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.00012640277610309614, "rmse": 4.8645140986324285e-06, "relative_l2": 0.0017539323332368927, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.1804479956626892, "p95_ms": 0.181536003947258, "p99_ms": 0.18192064687609671, "mean_ms": 0.18074389378229777, "n": 300}, {"repeat": 1, "p50_ms": 0.1804479956626892, "p95_ms": 0.18134400248527527, "p99_ms": 0.18150623634457588, "mean_ms": 0.18074143985907237, "n": 300}, {"repeat": 2, "p50_ms": 0.18035200238227844, "p95_ms": 0.18144479617476464, "p99_ms": 0.18198784723877906, "mean_ms": 0.18068511972824733, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_021f565f50432429_sdpa_flash.json", "configuration": "704x1280_tp2_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp2_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp2_sp1_v1/captures/qkv_rank0_021f565f50432429.pt", "context": {"phase": "commit", "step": null, "start_frame": 9, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_efficient", "status": "measured", "q_shape": [10560, 20, 128], "k_shape": [512, 20, 128], "q_len": 10560, "k_len": 512, "reference_error": {"max_abs": 0.0009765625, "rmse": 3.0000739136387933e-06, "relative_l2": 0.0009598867498970541, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.00012640277610309614, "rmse": 5.380012688015438e-06, "relative_l2": 0.0019397987168724244, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.3356480002403259, "p95_ms": 0.3365456074476242, "p99_ms": 0.33712159663438795, "mean_ms": 0.3358619740605354, "n": 300}, {"repeat": 1, "p50_ms": 0.3361920118331909, "p95_ms": 0.33715200424194336, "p99_ms": 0.33769598603248596, "mean_ms": 0.3364988805850347, "n": 300}, {"repeat": 2, "p50_ms": 0.33657601475715637, "p95_ms": 0.33798399567604065, "p99_ms": 0.33859231293201447, "mean_ms": 0.33689909319082895, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_021f565f50432429_sdpa_efficient.json", "configuration": "704x1280_tp2_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp2_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp2_sp1_v1/captures/qkv_rank0_021f565f50432429.pt", "context": {"phase": "commit", "step": null, "start_frame": 9, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_cudnn", "status": "measured", "q_shape": [10560, 20, 128], "k_shape": [512, 20, 128], "q_len": 10560, "k_len": 512, "reference_error": {"max_abs": 0.0009765625, "rmse": 3.0146050086653557e-06, "relative_l2": 0.0009645360372076371, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.00012640277610309614, "rmse": 5.380203378410327e-06, "relative_l2": 0.0019398674715398903, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.10172799974679947, "p95_ms": 0.10259359888732433, "p99_ms": 0.10304416224360466, "mean_ms": 0.10198293310900529, "n": 300}, {"repeat": 1, "p50_ms": 0.1019200012087822, "p95_ms": 0.1026895996183157, "p99_ms": 0.10310464277863503, "mean_ms": 0.1022937602798144, "n": 300}, {"repeat": 2, "p50_ms": 0.10201600193977356, "p95_ms": 0.10272159986197948, "p99_ms": 0.10326527655124665, "mean_ms": 0.10241162727276484, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_021f565f50432429_sdpa_cudnn.json", "configuration": "704x1280_tp2_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp2_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp2_sp1_v1/captures/qkv_rank0_021f565f50432429.pt", "context": {"phase": "commit", "step": null, "start_frame": 9, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "native_dense", "status": "measured", "q_shape": [10560, 20, 128], "k_shape": [512, 20, 128], "q_len": 10560, "k_len": 512, "reference_error": {"max_abs": 0.0, "rmse": 0.0, "relative_l2": 0.0, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.00012640277610309614, "rmse": 4.907322717693605e-06, "relative_l2": 0.0017693672604650258, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.11976000294089317, "p95_ms": 0.17800479903817176, "p99_ms": 0.2044016063213348, "mean_ms": 0.12980319981773694, "n": 300}, {"repeat": 1, "p50_ms": 0.12092799693346024, "p95_ms": 0.18563520535826683, "p99_ms": 0.19406400233507154, "mean_ms": 0.13485344022512435, "n": 300}, {"repeat": 2, "p50_ms": 0.11950400099158287, "p95_ms": 0.18843360394239425, "p99_ms": 0.21966144576668678, "mean_ms": 0.13284063977499802, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_021f565f50432429_native_dense.json", "configuration": "704x1280_tp2_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp2_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp2_sp1_v1/captures/qkv_rank0_260748fc9c594def.pt", "context": {"phase": "commit", "step": null, "start_frame": 0, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "vllm_fa2", "status": "measured", "q_shape": [10560, 20, 128], "k_shape": [66880, 20, 128], "q_len": 10560, "k_len": 10560, "reference_error": {"max_abs": 0.03125, "rmse": 0.00039271686131757454, "relative_l2": 0.0005742793049821513, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.00800240782572681, "rmse": 0.000782955795343756, "relative_l2": 0.0016773537871330796, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 2.946255922317505, "p95_ms": 3.242099177837372, "p99_ms": 3.3990338134765623, "mean_ms": 2.9749081548055014, "n": 300}, {"repeat": 1, "p50_ms": 2.970400094985962, "p95_ms": 3.10360164642334, "p99_ms": 3.1548181533813477, "mean_ms": 3.002938131491343, "n": 300}, {"repeat": 2, "p50_ms": 2.999184012413025, "p95_ms": 3.126788878440857, "p99_ms": 3.1537602758407592, "mean_ms": 3.015333127975464, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_260748fc9c594def_vllm_fa2.json", "configuration": "704x1280_tp2_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp2_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp2_sp1_v1/captures/qkv_rank0_260748fc9c594def.pt", "context": {"phase": "commit", "step": null, "start_frame": 0, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "vllm_fa3", "status": "measured", "q_shape": [10560, 20, 128], "k_shape": [66880, 20, 128], "q_len": 10560, "k_len": 10560, "reference_error": {"max_abs": 0.0, "rmse": 0.0, "relative_l2": 0.0, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.00800240782572681, "rmse": 0.0007835577224721566, "relative_l2": 0.0016786433168286322, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 1.7198079824447632, "p95_ms": 1.7364207804203033, "p99_ms": 1.7390409767627717, "mean_ms": 1.7120709359645843, "n": 300}, {"repeat": 1, "p50_ms": 1.7027199864387512, "p95_ms": 1.8071999549865723, "p99_ms": 1.8100499045848846, "mean_ms": 1.7114493874708812, "n": 300}, {"repeat": 2, "p50_ms": 1.7044800519943237, "p95_ms": 1.7253519475460053, "p99_ms": 1.7306633877754212, "mean_ms": 1.7056227215131123, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_260748fc9c594def_vllm_fa3.json", "configuration": "704x1280_tp2_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp2_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp2_sp1_v1/captures/qkv_rank0_260748fc9c594def.pt", "context": {"phase": "commit", "step": null, "start_frame": 0, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_math", "status": "measured", "q_shape": [10560, 20, 128], "k_shape": [66880, 20, 128], "q_len": 10560, "k_len": 10560, "reference_error": {"max_abs": 0.0625, "rmse": 0.0006826492693374726, "relative_l2": 0.0009982544335540434, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.007774497258635549, "rmse": 0.0007712550091903823, "relative_l2": 0.0016522867806896585, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 45.00217628479004, "p95_ms": 45.099338340759275, "p99_ms": 45.120188598632815, "mean_ms": 45.002456512451175, "n": 300}, {"repeat": 1, "p50_ms": 44.981184005737305, "p95_ms": 45.09846820831299, "p99_ms": 45.12146640777588, "mean_ms": 44.98971495310465, "n": 300}, {"repeat": 2, "p50_ms": 44.98551940917969, "p95_ms": 45.09031581878662, "p99_ms": 45.11096076965332, "mean_ms": 44.98497755686442, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_260748fc9c594def_sdpa_math.json", "configuration": "704x1280_tp2_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp2_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp2_sp1_v1/captures/qkv_rank0_260748fc9c594def.pt", "context": {"phase": "commit", "step": null, "start_frame": 0, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_flash", "status": "measured", "q_shape": [10560, 20, 128], "k_shape": [66880, 20, 128], "q_len": 10560, "k_len": 10560, "reference_error": {"max_abs": 0.03125, "rmse": 0.00039329846845827487, "relative_l2": 0.0005751298030825218, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.00800240782572681, "rmse": 0.000782955795343756, "relative_l2": 0.0016773537871330796, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 3.228976011276245, "p95_ms": 3.5775423407554627, "p99_ms": 3.616546247005463, "mean_ms": 3.272377920150757, "n": 300}, {"repeat": 1, "p50_ms": 3.249168038368225, "p95_ms": 3.4216959476470947, "p99_ms": 3.422529947757721, "mean_ms": 3.270664530595144, "n": 300}, {"repeat": 2, "p50_ms": 3.25, "p95_ms": 3.3672528743743895, "p99_ms": 3.423680386543274, "mean_ms": 3.2718235715230306, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_260748fc9c594def_sdpa_flash.json", "configuration": "704x1280_tp2_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp2_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp2_sp1_v1/captures/qkv_rank0_260748fc9c594def.pt", "context": {"phase": "commit", "step": null, "start_frame": 0, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_efficient", "status": "measured", "q_shape": [10560, 20, 128], "k_shape": [66880, 20, 128], "q_len": 10560, "k_len": 10560, "reference_error": {"max_abs": 0.0625, "rmse": 0.000580045328912504, "relative_l2": 0.0008482142254560468, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.00800240782572681, "rmse": 0.000776607802202966, "relative_l2": 0.001663754257761534, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 6.369279861450195, "p95_ms": 6.427875018119812, "p99_ms": 6.436165208816528, "mean_ms": 6.373615357081095, "n": 300}, {"repeat": 1, "p50_ms": 6.362367868423462, "p95_ms": 6.398319983482361, "p99_ms": 6.401376094818115, "mean_ms": 6.3690405416488645, "n": 300}, {"repeat": 2, "p50_ms": 6.366559982299805, "p95_ms": 6.39961576461792, "p99_ms": 6.402789311408997, "mean_ms": 6.3710209035873415, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_260748fc9c594def_sdpa_efficient.json", "configuration": "704x1280_tp2_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp2_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp2_sp1_v1/captures/qkv_rank0_260748fc9c594def.pt", "context": {"phase": "commit", "step": null, "start_frame": 0, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_cudnn", "status": "measured", "q_shape": [10560, 20, 128], "k_shape": [66880, 20, 128], "q_len": 10560, "k_len": 10560, "reference_error": {"max_abs": 0.03125, "rmse": 0.0005765570886145158, "relative_l2": 0.0008431132878308617, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.00800240782572681, "rmse": 0.0007766164903600255, "relative_l2": 0.0016637728707065225, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 1.9848639965057373, "p95_ms": 2.0185311794281007, "p99_ms": 2.0189453053474424, "mean_ms": 1.982358082930247, "n": 300}, {"repeat": 1, "p50_ms": 1.9880319833755493, "p95_ms": 2.3758208036422728, "p99_ms": 2.402816376686096, "mean_ms": 2.050200108687083, "n": 300}, {"repeat": 2, "p50_ms": 1.9888319969177246, "p95_ms": 2.0983439803123476, "p99_ms": 2.1034636187553404, "mean_ms": 1.997739201784134, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_260748fc9c594def_sdpa_cudnn.json", "configuration": "704x1280_tp2_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp2_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp2_sp1_v1/captures/qkv_rank0_260748fc9c594def.pt", "context": {"phase": "commit", "step": null, "start_frame": 0, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "native_dense", "status": "measured", "q_shape": [10560, 20, 128], "k_shape": [66880, 20, 128], "q_len": 10560, "k_len": 10560, "reference_error": {"max_abs": 0.03125, "rmse": 0.0003745011379038743, "relative_l2": 0.0005476420148320156, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.00800240782572681, "rmse": 0.0007843154270865668, "relative_l2": 0.001680266574121148, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 1.684976041316986, "p95_ms": 1.703139168024063, "p99_ms": 1.708322901725769, "mean_ms": 1.6756296535332997, "n": 300}, {"repeat": 1, "p50_ms": 1.675328016281128, "p95_ms": 1.85662083029747, "p99_ms": 1.8576963758468628, "mean_ms": 1.6887197895844777, "n": 300}, {"repeat": 2, "p50_ms": 1.6912320256233215, "p95_ms": 1.7387903571128844, "p99_ms": 1.7577702879905701, "mean_ms": 1.6796830908457439, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_260748fc9c594def_native_dense.json", "configuration": "704x1280_tp2_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp2_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp2_sp1_v1/captures/qkv_rank0_61c120430b1af4e0.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 27, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "vllm_fa2", "status": "measured", "q_shape": [10560, 20, 128], "k_shape": [512, 20, 128], "q_len": 10560, "k_len": 512, "reference_error": {"max_abs": 0.00048828125, "rmse": 1.8784910665047568e-06, "relative_l2": 0.0006859328745683148, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 7.489595196656651e-05, "rmse": 5.54717567994717e-06, "relative_l2": 0.001952980789202713, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.1687680035829544, "p95_ms": 0.1707536034286022, "p99_ms": 0.17196895822882652, "mean_ms": 0.16926719988385835, "n": 300}, {"repeat": 1, "p50_ms": 0.16841599345207214, "p95_ms": 0.16937600076198578, "p99_ms": 0.17004831850528718, "mean_ms": 0.16872768009702366, "n": 300}, {"repeat": 2, "p50_ms": 0.1687680035829544, "p95_ms": 0.1694783940911293, "p99_ms": 0.170048638433218, "mean_ms": 0.1691403732697169, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_61c120430b1af4e0_vllm_fa2.json", "configuration": "704x1280_tp2_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp2_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp2_sp1_v1/captures/qkv_rank0_61c120430b1af4e0.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 27, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "vllm_fa3", "status": "measured", "q_shape": [10560, 20, 128], "k_shape": [512, 20, 128], "q_len": 10560, "k_len": 512, "reference_error": {"max_abs": 0.000244140625, "rmse": 7.089748448972384e-07, "relative_l2": 0.0002588828672269634, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 7.489595196656651e-05, "rmse": 5.563208173893569e-06, "relative_l2": 0.0019586253107550977, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.14182400703430176, "p95_ms": 0.14409760162234306, "p99_ms": 0.14486431300640107, "mean_ms": 0.14209685330589614, "n": 300}, {"repeat": 1, "p50_ms": 0.14187199622392654, "p95_ms": 0.1444816045463085, "p99_ms": 0.14553856551647187, "mean_ms": 0.14248928010463716, "n": 300}, {"repeat": 2, "p50_ms": 0.14203199744224548, "p95_ms": 0.14454400539398193, "p99_ms": 0.14502816155552864, "mean_ms": 0.1425726934770743, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_61c120430b1af4e0_vllm_fa3.json", "configuration": "704x1280_tp2_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp2_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp2_sp1_v1/captures/qkv_rank0_61c120430b1af4e0.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 27, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_math", "status": "measured", "q_shape": [10560, 20, 128], "k_shape": [512, 20, 128], "q_len": 10560, "k_len": 512, "reference_error": {"max_abs": 0.0009765625, "rmse": 4.688822112686707e-06, "relative_l2": 0.0017121280411937093, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 6.549472242076831e-05, "rmse": 4.704919328637301e-06, "relative_l2": 0.0016564496229664544, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 2.424639940261841, "p95_ms": 2.4287423729896545, "p99_ms": 2.4310160994529726, "mean_ms": 2.4242146094640096, "n": 300}, {"repeat": 1, "p50_ms": 2.4247679710388184, "p95_ms": 2.4285791873931886, "p99_ms": 2.430884230136871, "mean_ms": 2.4251806855201723, "n": 300}, {"repeat": 2, "p50_ms": 2.424448013305664, "p95_ms": 2.4288352489471436, "p99_ms": 2.430792627334595, "mean_ms": 2.424295155207316, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_61c120430b1af4e0_sdpa_math.json", "configuration": "704x1280_tp2_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp2_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp2_sp1_v1/captures/qkv_rank0_61c120430b1af4e0.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 27, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_flash", "status": "measured", "q_shape": [10560, 20, 128], "k_shape": [512, 20, 128], "q_len": 10560, "k_len": 512, "reference_error": {"max_abs": 0.00048828125, "rmse": 1.8788425668066527e-06, "relative_l2": 0.0006860612252518982, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 7.489595196656651e-05, "rmse": 5.54717567994717e-06, "relative_l2": 0.001952980789202713, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.18060800433158875, "p95_ms": 0.18150559663772584, "p99_ms": 0.18227744206786156, "mean_ms": 0.18079509352644285, "n": 300}, {"repeat": 1, "p50_ms": 0.18057599663734436, "p95_ms": 0.18143999576568604, "p99_ms": 0.18189407914876937, "mean_ms": 0.1810126932958762, "n": 300}, {"repeat": 2, "p50_ms": 0.18060800433158875, "p95_ms": 0.18160000443458557, "p99_ms": 0.1824019230902195, "mean_ms": 0.18091669311126074, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_61c120430b1af4e0_sdpa_flash.json", "configuration": "704x1280_tp2_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp2_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp2_sp1_v1/captures/qkv_rank0_61c120430b1af4e0.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 27, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_efficient", "status": "measured", "q_shape": [10560, 20, 128], "k_shape": [512, 20, 128], "q_len": 10560, "k_len": 512, "reference_error": {"max_abs": 0.0009765625, "rmse": 3.2751520082822894e-06, "relative_l2": 0.0011959250015861447, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 7.489595196656651e-05, "rmse": 5.9867920509910385e-06, "relative_l2": 0.002107755466769057, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.3357760012149811, "p95_ms": 0.3367359936237335, "p99_ms": 0.33702432334423066, "mean_ms": 0.33610858738422394, "n": 300}, {"repeat": 1, "p50_ms": 0.3364480137825012, "p95_ms": 0.33744161278009416, "p99_ms": 0.3380799889564514, "mean_ms": 0.33658538629611334, "n": 300}, {"repeat": 2, "p50_ms": 0.3361920118331909, "p95_ms": 0.3376336008310318, "p99_ms": 0.3387868890166283, "mean_ms": 0.3366286923487981, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_61c120430b1af4e0_sdpa_efficient.json", "configuration": "704x1280_tp2_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp2_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp2_sp1_v1/captures/qkv_rank0_61c120430b1af4e0.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 27, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_cudnn", "status": "measured", "q_shape": [10560, 20, 128], "k_shape": [512, 20, 128], "q_len": 10560, "k_len": 512, "reference_error": {"max_abs": 0.0009765625, "rmse": 3.2739101583039602e-06, "relative_l2": 0.0011954715388358518, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 7.489595196656651e-05, "rmse": 5.9867920509910385e-06, "relative_l2": 0.002107755466769057, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.10255999863147736, "p95_ms": 0.10339199751615524, "p99_ms": 0.10390432141721248, "mean_ms": 0.1027820798009634, "n": 300}, {"repeat": 1, "p50_ms": 0.10247999802231789, "p95_ms": 0.10326559655368328, "p99_ms": 0.10374464020133019, "mean_ms": 0.10299765296280385, "n": 300}, {"repeat": 2, "p50_ms": 0.10201600193977356, "p95_ms": 0.10304160229861736, "p99_ms": 0.1036822397261858, "mean_ms": 0.10245354662338893, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_61c120430b1af4e0_sdpa_cudnn.json", "configuration": "704x1280_tp2_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp2_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp2_sp1_v1/captures/qkv_rank0_61c120430b1af4e0.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 27, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "native_dense", "status": "measured", "q_shape": [10560, 20, 128], "k_shape": [512, 20, 128], "q_len": 10560, "k_len": 512, "reference_error": {"max_abs": 0.0, "rmse": 0.0, "relative_l2": 0.0, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 7.489595196656651e-05, "rmse": 5.57652001207769e-06, "relative_l2": 0.001963311977582064, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.11918399855494499, "p95_ms": 0.12870720326900484, "p99_ms": 0.16069983407855032, "mean_ms": 0.12104309312999248, "n": 300}, {"repeat": 1, "p50_ms": 0.12025599926710129, "p95_ms": 0.13347200229763986, "p99_ms": 0.1907283262908458, "mean_ms": 0.12390197339157263, "n": 300}, {"repeat": 2, "p50_ms": 0.12027199938893318, "p95_ms": 0.12807200625538825, "p99_ms": 0.17440927311778054, "mean_ms": 0.12253002673387528, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_61c120430b1af4e0_native_dense.json", "configuration": "704x1280_tp2_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp2_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp2_sp1_v1/captures/qkv_rank0_af3cea46fecfaf39.pt", "context": {"phase": "commit", "step": null, "start_frame": 9, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "vllm_fa2", "status": "measured", "q_shape": [10560, 20, 128], "k_shape": [66880, 20, 128], "q_len": 10560, "k_len": 42240, "reference_error": {"max_abs": 0.0625, "rmse": 0.0004161057855214385, "relative_l2": 0.0005578574380589803, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.015384331463739187, "rmse": 0.000923460013183313, "relative_l2": 0.001694139676876375, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 12.000415802001953, "p95_ms": 12.654529285430907, "p99_ms": 13.379522895812986, "mean_ms": 12.091136223475138, "n": 300}, {"repeat": 1, "p50_ms": 12.104080200195312, "p95_ms": 12.23118085861206, "p99_ms": 12.317995147705078, "mean_ms": 12.115721486409505, "n": 300}, {"repeat": 2, "p50_ms": 12.103968143463135, "p95_ms": 12.217732477188111, "p99_ms": 12.226309490203857, "mean_ms": 12.11566762606303, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_af3cea46fecfaf39_vllm_fa2.json", "configuration": "704x1280_tp2_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp2_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp2_sp1_v1/captures/qkv_rank0_af3cea46fecfaf39.pt", "context": {"phase": "commit", "step": null, "start_frame": 9, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "vllm_fa3", "status": "measured", "q_shape": [10560, 20, 128], "k_shape": [66880, 20, 128], "q_len": 10560, "k_len": 42240, "reference_error": {"max_abs": 0.0, "rmse": 0.0, "relative_l2": 0.0, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.015384331463739187, "rmse": 0.0009240212371461509, "relative_l2": 0.001695169274010505, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 6.649423837661743, "p95_ms": 7.124215817451477, "p99_ms": 7.705114679336547, "mean_ms": 6.755227398872376, "n": 300}, {"repeat": 1, "p50_ms": 6.704800128936768, "p95_ms": 6.953323388099671, "p99_ms": 6.978400549888611, "mean_ms": 6.745012811024984, "n": 300}, {"repeat": 2, "p50_ms": 6.717423915863037, "p95_ms": 6.833408188819885, "p99_ms": 6.847584095001221, "mean_ms": 6.728244694073995, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_af3cea46fecfaf39_vllm_fa3.json", "configuration": "704x1280_tp2_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp2_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp2_sp1_v1/captures/qkv_rank0_af3cea46fecfaf39.pt", "context": {"phase": "commit", "step": null, "start_frame": 9, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_math", "status": "measured", "q_shape": [10560, 20, 128], "k_shape": [66880, 20, 128], "q_len": 10560, "k_len": 42240, "reference_error": {"max_abs": 0.0625, "rmse": 0.0008193180075552764, "relative_l2": 0.001098428958582279, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.015384331463739187, "rmse": 0.0008987087561677207, "relative_l2": 0.0016487320945619732, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 186.26570892333984, "p95_ms": 187.01155548095704, "p99_ms": 187.17695220947266, "mean_ms": 186.33124649047852, "n": 300}, {"repeat": 1, "p50_ms": 186.45015716552734, "p95_ms": 187.102059173584, "p99_ms": 187.3086375427246, "mean_ms": 186.47564626057942, "n": 300}, {"repeat": 2, "p50_ms": 186.32030487060547, "p95_ms": 187.048779296875, "p99_ms": 187.13489868164064, "mean_ms": 186.38927408854167, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_af3cea46fecfaf39_sdpa_math.json", "configuration": "704x1280_tp2_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp2_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp2_sp1_v1/captures/qkv_rank0_af3cea46fecfaf39.pt", "context": {"phase": "commit", "step": null, "start_frame": 9, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_flash", "status": "measured", "q_shape": [10560, 20, 128], "k_shape": [66880, 20, 128], "q_len": 10560, "k_len": 42240, "reference_error": {"max_abs": 0.0625, "rmse": 0.0004168583244416407, "relative_l2": 0.0005588663387968962, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.015865668536260813, "rmse": 0.0009243712902404814, "relative_l2": 0.0016958114662306905, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 13.020768165588379, "p95_ms": 13.365641641616822, "p99_ms": 13.974754343032837, "mean_ms": 13.079486433664957, "n": 300}, {"repeat": 1, "p50_ms": 13.022416114807129, "p95_ms": 13.14044771194458, "p99_ms": 13.259021844863891, "mean_ms": 13.044986883799234, "n": 300}, {"repeat": 2, "p50_ms": 13.021520137786865, "p95_ms": 13.127166604995727, "p99_ms": 13.151354751586913, "mean_ms": 13.03079039891561, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_af3cea46fecfaf39_sdpa_flash.json", "configuration": "704x1280_tp2_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp2_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp2_sp1_v1/captures/qkv_rank0_af3cea46fecfaf39.pt", "context": {"phase": "commit", "step": null, "start_frame": 9, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_efficient", "status": "measured", "q_shape": [10560, 20, 128], "k_shape": [66880, 20, 128], "q_len": 10560, "k_len": 42240, "reference_error": {"max_abs": 0.0625, "rmse": 0.0007224400030458222, "relative_l2": 0.00096854824728146, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.015384331463739187, "rmse": 0.0009043838547521623, "relative_l2": 0.0016591433842170012, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 24.966879844665527, "p95_ms": 25.04010429382324, "p99_ms": 25.4429647064209, "mean_ms": 25.0002796681722, "n": 300}, {"repeat": 1, "p50_ms": 24.987823486328125, "p95_ms": 25.034957313537596, "p99_ms": 25.040547065734863, "mean_ms": 24.991987813313802, "n": 300}, {"repeat": 2, "p50_ms": 24.994112014770508, "p95_ms": 25.033224487304686, "p99_ms": 25.042911853790283, "mean_ms": 24.99017608642578, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_af3cea46fecfaf39_sdpa_efficient.json", "configuration": "704x1280_tp2_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp2_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp2_sp1_v1/captures/qkv_rank0_af3cea46fecfaf39.pt", "context": {"phase": "commit", "step": null, "start_frame": 9, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_cudnn", "status": "measured", "q_shape": [10560, 20, 128], "k_shape": [66880, 20, 128], "q_len": 10560, "k_len": 42240, "reference_error": {"max_abs": 0.0625, "rmse": 0.0007196185666317981, "relative_l2": 0.0009647656531807761, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.015384331463739187, "rmse": 0.0009044680810150898, "relative_l2": 0.0016592979020649031, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 7.352848052978516, "p95_ms": 7.889472007751465, "p99_ms": 8.510259752273559, "mean_ms": 7.4299085903167725, "n": 300}, {"repeat": 1, "p50_ms": 7.423472166061401, "p95_ms": 7.653891015052795, "p99_ms": 7.687718911170959, "mean_ms": 7.416917225519816, "n": 300}, {"repeat": 2, "p50_ms": 7.4235999584198, "p95_ms": 7.560588765144348, "p99_ms": 7.5654136657714846, "mean_ms": 7.407920835812886, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_af3cea46fecfaf39_sdpa_cudnn.json", "configuration": "704x1280_tp2_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp2_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp2_sp1_v1/captures/qkv_rank0_af3cea46fecfaf39.pt", "context": {"phase": "commit", "step": null, "start_frame": 9, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "native_dense", "status": "measured", "q_shape": [10560, 20, 128], "k_shape": [66880, 20, 128], "q_len": 10560, "k_len": 42240, "reference_error": {"max_abs": 0.0625, "rmse": 0.0003967170968027179, "relative_l2": 0.0005318637494531023, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.015384331463739187, "rmse": 0.0009260597688801338, "relative_l2": 0.001698909075890185, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 6.746095895767212, "p95_ms": 6.8879153490066525, "p99_ms": 7.070176744461059, "mean_ms": 6.7632237927118934, "n": 300}, {"repeat": 1, "p50_ms": 6.745919942855835, "p95_ms": 6.812032222747803, "p99_ms": 6.857101845741272, "mean_ms": 6.74043893178304, "n": 300}, {"repeat": 2, "p50_ms": 6.744768142700195, "p95_ms": 6.8115280866622925, "p99_ms": 6.830311822891235, "mean_ms": 6.73720874786377, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_af3cea46fecfaf39_native_dense.json", "configuration": "704x1280_tp2_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp2_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp2_sp1_v1/captures/qkv_rank0_b5faac666beaef04.pt", "context": {"phase": "commit", "step": null, "start_frame": 27, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "vllm_fa2", "status": "measured", "q_shape": [10560, 20, 128], "k_shape": [66880, 20, 128], "q_len": 10560, "k_len": 63360, "reference_error": {"max_abs": 0.0625, "rmse": 0.00041126209582860977, "relative_l2": 0.0005571960847909151, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.00897828162034564, "rmse": 0.0009021776803715486, "relative_l2": 0.0016666561231022962, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 18.050960540771484, "p95_ms": 18.789546775817872, "p99_ms": 19.515076351165767, "mean_ms": 18.154871501922607, "n": 300}, {"repeat": 1, "p50_ms": 18.15070343017578, "p95_ms": 18.470598888397216, "p99_ms": 18.528610877990722, "mean_ms": 18.178576011657714, "n": 300}, {"repeat": 2, "p50_ms": 18.15113639831543, "p95_ms": 18.467505931854248, "p99_ms": 18.47184301376343, "mean_ms": 18.178045743306477, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_b5faac666beaef04_vllm_fa2.json", "configuration": "704x1280_tp2_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp2_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp2_sp1_v1/captures/qkv_rank0_b5faac666beaef04.pt", "context": {"phase": "commit", "step": null, "start_frame": 27, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "vllm_fa3", "status": "measured", "q_shape": [10560, 20, 128], "k_shape": [66880, 20, 128], "q_len": 10560, "k_len": 63360, "reference_error": {"max_abs": 0.0, "rmse": 0.0, "relative_l2": 0.0, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.009036062330048722, "rmse": 0.000902883153287687, "relative_l2": 0.0016679593927141987, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 10.063584327697754, "p95_ms": 10.565547466278076, "p99_ms": 11.03136799812317, "mean_ms": 10.116092913945517, "n": 300}, {"repeat": 1, "p50_ms": 10.119616031646729, "p95_ms": 10.33188600540161, "p99_ms": 10.338859634399414, "mean_ms": 10.101725021998087, "n": 300}, {"repeat": 2, "p50_ms": 10.120895862579346, "p95_ms": 10.2173490524292, "p99_ms": 10.239403476715088, "mean_ms": 10.097782157262166, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_b5faac666beaef04_vllm_fa3.json", "configuration": "704x1280_tp2_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp2_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp2_sp1_v1/captures/qkv_rank0_b5faac666beaef04.pt", "context": {"phase": "commit", "step": null, "start_frame": 27, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_math", "status": "measured", "q_shape": [10560, 20, 128], "k_shape": [66880, 20, 128], "q_len": 10560, "k_len": 63360, "reference_error": {"max_abs": 0.0625, "rmse": 0.0008310221959496787, "relative_l2": 0.0011259056418135818, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.0077161019941787146, "rmse": 0.0008886268973423762, "relative_l2": 0.0016416228109291332, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 279.7746124267578, "p95_ms": 280.5802474975586, "p99_ms": 280.8737753295898, "mean_ms": 279.69611114501953, "n": 300}, {"repeat": 1, "p50_ms": 279.73492431640625, "p95_ms": 280.5655487060547, "p99_ms": 280.7505099487305, "mean_ms": 279.70046864827475, "n": 300}, {"repeat": 2, "p50_ms": 279.97740173339844, "p95_ms": 280.6724136352539, "p99_ms": 280.88391357421875, "mean_ms": 279.9650796508789, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_b5faac666beaef04_sdpa_math.json", "configuration": "704x1280_tp2_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp2_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp2_sp1_v1/captures/qkv_rank0_b5faac666beaef04.pt", "context": {"phase": "commit", "step": null, "start_frame": 27, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_flash", "status": "measured", "q_shape": [10560, 20, 128], "k_shape": [66880, 20, 128], "q_len": 10560, "k_len": 63360, "reference_error": {"max_abs": 0.0625, "rmse": 0.000411873699641212, "relative_l2": 0.0005580247126982319, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.00897828162034564, "rmse": 0.0009021776920801008, "relative_l2": 0.0016666561447323263, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 19.582672119140625, "p95_ms": 20.21386957168579, "p99_ms": 21.31591384887695, "mean_ms": 19.632160771687825, "n": 300}, {"repeat": 1, "p50_ms": 19.52787208557129, "p95_ms": 19.704632091522218, "p99_ms": 19.860179595947265, "mean_ms": 19.555430990854898, "n": 300}, {"repeat": 2, "p50_ms": 19.525328636169434, "p95_ms": 19.854588413238524, "p99_ms": 19.91986801147461, "mean_ms": 19.533009548187255, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_b5faac666beaef04_sdpa_flash.json", "configuration": "704x1280_tp2_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp2_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp2_sp1_v1/captures/qkv_rank0_b5faac666beaef04.pt", "context": {"phase": "commit", "step": null, "start_frame": 27, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_efficient", "status": "measured", "q_shape": [10560, 20, 128], "k_shape": [66880, 20, 128], "q_len": 10560, "k_len": 63360, "reference_error": {"max_abs": 0.0625, "rmse": 0.000723254990839047, "relative_l2": 0.000979897863889093, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.008240741969119725, "rmse": 0.0008973692801195905, "relative_l2": 0.0016577732279735317, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 37.859825134277344, "p95_ms": 37.94212646484375, "p99_ms": 37.94927806854248, "mean_ms": 37.86097127278646, "n": 300}, {"repeat": 1, "p50_ms": 37.84987258911133, "p95_ms": 37.94873600006103, "p99_ms": 37.959396057128906, "mean_ms": 37.86244222005208, "n": 300}, {"repeat": 2, "p50_ms": 37.86003112792969, "p95_ms": 37.9522310256958, "p99_ms": 37.95961982727051, "mean_ms": 37.86667352040609, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_b5faac666beaef04_sdpa_efficient.json", "configuration": "704x1280_tp2_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp2_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp2_sp1_v1/captures/qkv_rank0_b5faac666beaef04.pt", "context": {"phase": "commit", "step": null, "start_frame": 27, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_cudnn", "status": "measured", "q_shape": [10560, 20, 128], "k_shape": [66880, 20, 128], "q_len": 10560, "k_len": 63360, "reference_error": {"max_abs": 0.0625, "rmse": 0.0007222216398857924, "relative_l2": 0.0009784978342943199, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.008240741969119725, "rmse": 0.0008969490156068632, "relative_l2": 0.0016569968438546394, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 11.01257610321045, "p95_ms": 11.6917067527771, "p99_ms": 12.739332208633423, "mean_ms": 11.124882564544677, "n": 300}, {"repeat": 1, "p50_ms": 11.116767883300781, "p95_ms": 11.326591730117798, "p99_ms": 11.507919311523438, "mean_ms": 11.101652917861939, "n": 300}, {"repeat": 2, "p50_ms": 11.119776248931885, "p95_ms": 11.224992322921754, "p99_ms": 11.350433349609375, "mean_ms": 11.094721193313598, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_b5faac666beaef04_sdpa_cudnn.json", "configuration": "704x1280_tp2_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp2_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp2_sp1_v1/captures/qkv_rank0_b5faac666beaef04.pt", "context": {"phase": "commit", "step": null, "start_frame": 27, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "native_dense", "status": "measured", "q_shape": [10560, 20, 128], "k_shape": [66880, 20, 128], "q_len": 10560, "k_len": 63360, "reference_error": {"max_abs": 0.0625, "rmse": 0.0003934063534883984, "relative_l2": 0.0005330043349945823, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.00897828162034564, "rmse": 0.0009014308854093113, "relative_l2": 0.0016652765163756014, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 10.05619192123413, "p95_ms": 10.620579195022582, "p99_ms": 10.97721604347229, "mean_ms": 10.089247048695881, "n": 300}, {"repeat": 1, "p50_ms": 10.14035177230835, "p95_ms": 10.308470869064331, "p99_ms": 10.389544744491577, "mean_ms": 10.090050141016642, "n": 300}, {"repeat": 2, "p50_ms": 10.091023921966553, "p95_ms": 10.303299283981323, "p99_ms": 10.335573015213013, "mean_ms": 10.071286411285401, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_b5faac666beaef04_native_dense.json", "configuration": "704x1280_tp2_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp2_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp2_sp1_v1/captures/qkv_rank0_bb7d90c82a1e72ca.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 9, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "vllm_fa2", "status": "measured", "q_shape": [10560, 20, 128], "k_shape": [66880, 20, 128], "q_len": 10560, "k_len": 42240, "reference_error": {"max_abs": 0.03125, "rmse": 0.0004317953999885311, "relative_l2": 0.0008771030084767145, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.008121707600010186, "rmse": 0.0007496158925283387, "relative_l2": 0.001916130854741869, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 12.002031803131104, "p95_ms": 12.535628747940063, "p99_ms": 13.443294601440428, "mean_ms": 12.117113602956136, "n": 300}, {"repeat": 1, "p50_ms": 12.107424259185791, "p95_ms": 12.425087928771973, "p99_ms": 12.42781229019165, "mean_ms": 12.156342277526855, "n": 300}, {"repeat": 2, "p50_ms": 12.17030382156372, "p95_ms": 12.38516125679016, "p99_ms": 12.42767991065979, "mean_ms": 12.164027293523153, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_bb7d90c82a1e72ca_vllm_fa2.json", "configuration": "704x1280_tp2_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp2_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp2_sp1_v1/captures/qkv_rank0_bb7d90c82a1e72ca.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 9, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "vllm_fa3", "status": "measured", "q_shape": [10560, 20, 128], "k_shape": [66880, 20, 128], "q_len": 10560, "k_len": 42240, "reference_error": {"max_abs": 0.0, "rmse": 0.0, "relative_l2": 0.0, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.008121707600010186, "rmse": 0.0007550902457188012, "relative_l2": 0.0019301241240449252, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 6.701456069946289, "p95_ms": 7.050289583206177, "p99_ms": 7.6967227220535275, "mean_ms": 6.763803731600444, "n": 300}, {"repeat": 1, "p50_ms": 6.725152015686035, "p95_ms": 6.977196621894836, "p99_ms": 6.982438611984253, "mean_ms": 6.7615478340784705, "n": 300}, {"repeat": 2, "p50_ms": 6.763871908187866, "p95_ms": 6.859054183959961, "p99_ms": 6.947770133018493, "mean_ms": 6.7467129643758135, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_bb7d90c82a1e72ca_vllm_fa3.json", "configuration": "704x1280_tp2_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp2_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp2_sp1_v1/captures/qkv_rank0_bb7d90c82a1e72ca.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 9, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_math", "status": "measured", "q_shape": [10560, 20, 128], "k_shape": [66880, 20, 128], "q_len": 10560, "k_len": 42240, "reference_error": {"max_abs": 0.03125, "rmse": 0.0007650394045733917, "relative_l2": 0.0015540192493305376, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.007798464777282188, "rmse": 0.00065519899775415, "relative_l2": 0.0016747870851006727, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 186.3338851928711, "p95_ms": 187.15654373168945, "p99_ms": 187.31292343139648, "mean_ms": 186.43594411214193, "n": 300}, {"repeat": 1, "p50_ms": 186.61975860595703, "p95_ms": 187.19307403564454, "p99_ms": 187.69362869262696, "mean_ms": 186.59350489298504, "n": 300}, {"repeat": 2, "p50_ms": 186.48824310302734, "p95_ms": 187.13734130859376, "p99_ms": 187.29059127807616, "mean_ms": 186.49607650756835, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_bb7d90c82a1e72ca_sdpa_math.json", "configuration": "704x1280_tp2_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp2_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp2_sp1_v1/captures/qkv_rank0_bb7d90c82a1e72ca.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 9, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_flash", "status": "measured", "q_shape": [10560, 20, 128], "k_shape": [66880, 20, 128], "q_len": 10560, "k_len": 42240, "reference_error": {"max_abs": 0.03125, "rmse": 0.0004322114062916214, "relative_l2": 0.0008779480392019039, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.008121707600010186, "rmse": 0.0007496827275028022, "relative_l2": 0.0019163016949788783, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 13.022271633148193, "p95_ms": 13.596496343612671, "p99_ms": 14.250485982894897, "mean_ms": 13.087751976648967, "n": 300}, {"repeat": 1, "p50_ms": 13.04804801940918, "p95_ms": 13.260289812088013, "p99_ms": 13.271254577636718, "mean_ms": 13.086132577260335, "n": 300}, {"repeat": 2, "p50_ms": 13.071424007415771, "p95_ms": 13.18843994140625, "p99_ms": 13.261415853500367, "mean_ms": 13.07902401606242, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_bb7d90c82a1e72ca_sdpa_flash.json", "configuration": "704x1280_tp2_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp2_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp2_sp1_v1/captures/qkv_rank0_bb7d90c82a1e72ca.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 9, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_efficient", "status": "measured", "q_shape": [10560, 20, 128], "k_shape": [66880, 20, 128], "q_len": 10560, "k_len": 42240, "reference_error": {"max_abs": 0.03125, "rmse": 0.0006796747087374645, "relative_l2": 0.0013806185333030388, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.008121707600010186, "rmse": 0.0007536886690942537, "relative_l2": 0.0019265414835988664, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 25.02836799621582, "p95_ms": 25.522205543518066, "p99_ms": 25.534373626708984, "mean_ms": 25.109849071502687, "n": 300}, {"repeat": 1, "p50_ms": 25.011775970458984, "p95_ms": 25.40949773788452, "p99_ms": 25.52259925842285, "mean_ms": 25.036235936482747, "n": 300}, {"repeat": 2, "p50_ms": 24.983023643493652, "p95_ms": 25.046907901763916, "p99_ms": 25.05974708557129, "mean_ms": 25.002111670176188, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_bb7d90c82a1e72ca_sdpa_efficient.json", "configuration": "704x1280_tp2_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp2_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp2_sp1_v1/captures/qkv_rank0_bb7d90c82a1e72ca.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 9, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_cudnn", "status": "measured", "q_shape": [10560, 20, 128], "k_shape": [66880, 20, 128], "q_len": 10560, "k_len": 42240, "reference_error": {"max_abs": 0.03125, "rmse": 0.0006772677581697126, "relative_l2": 0.0013757293112680521, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.008121707600010186, "rmse": 0.0007535540132164486, "relative_l2": 0.0019261972829424949, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 7.412527799606323, "p95_ms": 7.831884837150573, "p99_ms": 8.328992738723755, "mean_ms": 7.465301024119059, "n": 300}, {"repeat": 1, "p50_ms": 7.425920009613037, "p95_ms": 7.641124749183655, "p99_ms": 7.653573231697083, "mean_ms": 7.450386029879252, "n": 300}, {"repeat": 2, "p50_ms": 7.426527976989746, "p95_ms": 7.562248158454895, "p99_ms": 7.565346097946167, "mean_ms": 7.441597437858581, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_bb7d90c82a1e72ca_sdpa_cudnn.json", "configuration": "704x1280_tp2_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp2_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp2_sp1_v1/captures/qkv_rank0_bb7d90c82a1e72ca.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 9, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "native_dense", "status": "measured", "q_shape": [10560, 20, 128], "k_shape": [66880, 20, 128], "q_len": 10560, "k_len": 42240, "reference_error": {"max_abs": 0.03125, "rmse": 0.0004108283631496138, "relative_l2": 0.0008345128116132349, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.008121707600010186, "rmse": 0.000762189080244546, "relative_l2": 0.0019482698117271964, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 6.787263870239258, "p95_ms": 7.095555162429809, "p99_ms": 7.314405312538147, "mean_ms": 6.788903365135193, "n": 300}, {"repeat": 1, "p50_ms": 6.802175998687744, "p95_ms": 6.946760058403015, "p99_ms": 7.094351391792298, "mean_ms": 6.783079791069031, "n": 300}, {"repeat": 2, "p50_ms": 6.807424068450928, "p95_ms": 6.879646253585816, "p99_ms": 6.898843679428101, "mean_ms": 6.783341323534648, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_bb7d90c82a1e72ca_native_dense.json", "configuration": "704x1280_tp2_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp2_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp2_sp1_v1/captures/qkv_rank0_c02ae993eecc25ee.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 0, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "vllm_fa2", "status": "measured", "q_shape": [10560, 20, 128], "k_shape": [66880, 20, 128], "q_len": 10560, "k_len": 10560, "reference_error": {"max_abs": 0.03125, "rmse": 0.0004069444397540273, "relative_l2": 0.0008306485402599196, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.008999835780528098, "rmse": 0.0006803289293460923, "relative_l2": 0.001885584018998539, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 3.020479917526245, "p95_ms": 3.3658897161483763, "p99_ms": 3.5351040363311768, "mean_ms": 3.053091629346212, "n": 300}, {"repeat": 1, "p50_ms": 3.029103994369507, "p95_ms": 3.184169590473175, "p99_ms": 3.271937220096588, "mean_ms": 3.055221759478251, "n": 300}, {"repeat": 2, "p50_ms": 3.0488959550857544, "p95_ms": 3.1834303975105285, "p99_ms": 3.1854138731956483, "mean_ms": 3.0685164856910707, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_c02ae993eecc25ee_vllm_fa2.json", "configuration": "704x1280_tp2_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp2_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp2_sp1_v1/captures/qkv_rank0_c02ae993eecc25ee.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 0, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "vllm_fa3", "status": "measured", "q_shape": [10560, 20, 128], "k_shape": [66880, 20, 128], "q_len": 10560, "k_len": 10560, "reference_error": {"max_abs": 0.0, "rmse": 0.0, "relative_l2": 0.0, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.008999835780528098, "rmse": 0.000694164453688696, "relative_l2": 0.0019239302401712203, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 1.735983967781067, "p95_ms": 1.7423408329486847, "p99_ms": 1.7461517357826233, "mean_ms": 1.7261806905269623, "n": 300}, {"repeat": 1, "p50_ms": 1.7527360320091248, "p95_ms": 2.040366506576538, "p99_ms": 2.0768665647506714, "mean_ms": 1.776481598218282, "n": 300}, {"repeat": 2, "p50_ms": 1.7365760207176208, "p95_ms": 1.7705295622348785, "p99_ms": 1.7716179025173187, "mean_ms": 1.7262891705830892, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_c02ae993eecc25ee_vllm_fa3.json", "configuration": "704x1280_tp2_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp2_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp2_sp1_v1/captures/qkv_rank0_c02ae993eecc25ee.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 0, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_math", "status": "measured", "q_shape": [10560, 20, 128], "k_shape": [66880, 20, 128], "q_len": 10560, "k_len": 10560, "reference_error": {"max_abs": 0.03125, "rmse": 0.0006770698794725477, "relative_l2": 0.0013820243062610017, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.0072374631331095785, "rmse": 0.0005975725893755776, "relative_l2": 0.0016562184498036551, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 45.004207611083984, "p95_ms": 45.09935894012451, "p99_ms": 45.115256996154784, "mean_ms": 45.00156280517578, "n": 300}, {"repeat": 1, "p50_ms": 45.017473220825195, "p95_ms": 45.09772281646728, "p99_ms": 45.12174766540527, "mean_ms": 45.00489649454752, "n": 300}, {"repeat": 2, "p50_ms": 45.001007080078125, "p95_ms": 45.10305614471436, "p99_ms": 45.12100067138672, "mean_ms": 44.99683958689372, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_c02ae993eecc25ee_sdpa_math.json", "configuration": "704x1280_tp2_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp2_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp2_sp1_v1/captures/qkv_rank0_c02ae993eecc25ee.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 0, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_flash", "status": "measured", "q_shape": [10560, 20, 128], "k_shape": [66880, 20, 128], "q_len": 10560, "k_len": 10560, "reference_error": {"max_abs": 0.03125, "rmse": 0.00040723198283547303, "relative_l2": 0.0008312354686401404, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.008999835780528098, "rmse": 0.0006802589472844292, "relative_l2": 0.0018853900583254049, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 3.276911973953247, "p95_ms": 3.6161343693733214, "p99_ms": 3.685415153503418, "mean_ms": 3.3071031483014424, "n": 300}, {"repeat": 1, "p50_ms": 3.275167942047119, "p95_ms": 3.4570319175720217, "p99_ms": 3.4877414202690122, "mean_ms": 3.2875996748606364, "n": 300}, {"repeat": 2, "p50_ms": 3.3058559894561768, "p95_ms": 3.3965168356895448, "p99_ms": 3.3982455277442933, "mean_ms": 3.3070921540260314, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_c02ae993eecc25ee_sdpa_flash.json", "configuration": "704x1280_tp2_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp2_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp2_sp1_v1/captures/qkv_rank0_c02ae993eecc25ee.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 0, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_efficient", "status": "measured", "q_shape": [10560, 20, 128], "k_shape": [66880, 20, 128], "q_len": 10560, "k_len": 10560, "reference_error": {"max_abs": 0.03125, "rmse": 0.0006289115688844325, "relative_l2": 0.0012837243260091988, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.0072374631331095785, "rmse": 0.0006850203216445965, "relative_l2": 0.001898586574032344, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 6.361824035644531, "p95_ms": 6.404295825958252, "p99_ms": 6.4398184537887575, "mean_ms": 6.368898355166118, "n": 300}, {"repeat": 1, "p50_ms": 6.364016056060791, "p95_ms": 6.396105813980102, "p99_ms": 6.406547517776489, "mean_ms": 6.370112218856812, "n": 300}, {"repeat": 2, "p50_ms": 6.3624958992004395, "p95_ms": 6.3924575567245485, "p99_ms": 6.404961433410644, "mean_ms": 6.368379309972127, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_c02ae993eecc25ee_sdpa_efficient.json", "configuration": "704x1280_tp2_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp2_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp2_sp1_v1/captures/qkv_rank0_c02ae993eecc25ee.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 0, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_cudnn", "status": "measured", "q_shape": [10560, 20, 128], "k_shape": [66880, 20, 128], "q_len": 10560, "k_len": 10560, "reference_error": {"max_abs": 0.03125, "rmse": 0.0006274888275802196, "relative_l2": 0.001280820249009827, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.0072374631331095785, "rmse": 0.0006857515197147507, "relative_l2": 0.0019006131457924638, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 2.0223840475082397, "p95_ms": 2.0315552711486817, "p99_ms": 2.0368694829940797, "mean_ms": 2.0227231959501903, "n": 300}, {"repeat": 1, "p50_ms": 2.0392000675201416, "p95_ms": 2.375939226150513, "p99_ms": 2.389026257991791, "mean_ms": 2.073615786631902, "n": 300}, {"repeat": 2, "p50_ms": 2.031567931175232, "p95_ms": 2.123487949371338, "p99_ms": 2.12672315120697, "mean_ms": 2.044800426165263, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_c02ae993eecc25ee_sdpa_cudnn.json", "configuration": "704x1280_tp2_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp2_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp2_sp1_v1/captures/qkv_rank0_c02ae993eecc25ee.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 0, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "native_dense", "status": "measured", "q_shape": [10560, 20, 128], "k_shape": [66880, 20, 128], "q_len": 10560, "k_len": 10560, "reference_error": {"max_abs": 0.03125, "rmse": 0.0003850905597636508, "relative_l2": 0.0007860407468127502, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.008999835780528098, "rmse": 0.0006931785325494421, "relative_l2": 0.001921197683809172, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 1.7095680236816406, "p95_ms": 1.7261264085769654, "p99_ms": 1.7310818696022034, "mean_ms": 1.7068827748298645, "n": 300}, {"repeat": 1, "p50_ms": 1.730127990245819, "p95_ms": 1.927308839559555, "p99_ms": 1.9479689657688142, "mean_ms": 1.7593094424406688, "n": 300}, {"repeat": 2, "p50_ms": 1.7063040137290955, "p95_ms": 1.7266272127628326, "p99_ms": 1.73101349234581, "mean_ms": 1.7036923734347025, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_c02ae993eecc25ee_native_dense.json", "configuration": "704x1280_tp2_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp2_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp2_sp1_v1/captures/qkv_rank0_c0f6ec677589277b.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 0, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "vllm_fa2", "status": "measured", "q_shape": [10560, 20, 128], "k_shape": [512, 20, 128], "q_len": 10560, "k_len": 512, "reference_error": {"max_abs": 0.0009765625, "rmse": 2.09484938485933e-06, "relative_l2": 0.00027373660687887816, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.0004675048160963924, "rmse": 1.8338522143917177e-05, "relative_l2": 0.0017526865015281977, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.1685120016336441, "p95_ms": 0.16937600076198578, "p99_ms": 0.16995296508073807, "mean_ms": 0.16871264000733693, "n": 300}, {"repeat": 1, "p50_ms": 0.16857600212097168, "p95_ms": 0.16937600076198578, "p99_ms": 0.16966527551412583, "mean_ms": 0.16915466671188673, "n": 300}, {"repeat": 2, "p50_ms": 0.16867199540138245, "p95_ms": 0.1695680022239685, "p99_ms": 0.16992031753063203, "mean_ms": 0.169145813335975, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_c0f6ec677589277b_vllm_fa2.json", "configuration": "704x1280_tp2_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp2_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp2_sp1_v1/captures/qkv_rank0_c0f6ec677589277b.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 0, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "vllm_fa3", "status": "measured", "q_shape": [10560, 20, 128], "k_shape": [512, 20, 128], "q_len": 10560, "k_len": 512, "reference_error": {"max_abs": 0.0009765625, "rmse": 8.036688725162759e-07, "relative_l2": 0.00010501642352275925, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.0004675048160963924, "rmse": 1.8353640117122423e-05, "relative_l2": 0.0017541313871825245, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.14161600172519684, "p95_ms": 0.14384159967303276, "p99_ms": 0.1448681530356407, "mean_ms": 0.14193248013655343, "n": 300}, {"repeat": 1, "p50_ms": 0.1414560005068779, "p95_ms": 0.1439696006476879, "p99_ms": 0.14457727819681168, "mean_ms": 0.14204736029108364, "n": 300}, {"repeat": 2, "p50_ms": 0.141744002699852, "p95_ms": 0.1444816045463085, "p99_ms": 0.1453193563222885, "mean_ms": 0.1424775463839372, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_c0f6ec677589277b_vllm_fa3.json", "configuration": "704x1280_tp2_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp2_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp2_sp1_v1/captures/qkv_rank0_c0f6ec677589277b.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 0, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_math", "status": "measured", "q_shape": [10560, 20, 128], "k_shape": [512, 20, 128], "q_len": 10560, "k_len": 512, "reference_error": {"max_abs": 0.001953125, "rmse": 9.690804074942715e-06, "relative_l2": 0.0012663095707861388, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.0004675048160963924, "rmse": 1.6528679360779944e-05, "relative_l2": 0.0015797125295254962, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 2.4239680767059326, "p95_ms": 2.4285759925842285, "p99_ms": 2.430274341106415, "mean_ms": 2.4239335449536643, "n": 300}, {"repeat": 1, "p50_ms": 2.42411208152771, "p95_ms": 2.4283904671669005, "p99_ms": 2.430884230136871, "mean_ms": 2.424922133286794, "n": 300}, {"repeat": 2, "p50_ms": 2.4243040084838867, "p95_ms": 2.4286398887634277, "p99_ms": 2.4307852673530577, "mean_ms": 2.424581224123637, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_c0f6ec677589277b_sdpa_math.json", "configuration": "704x1280_tp2_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp2_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp2_sp1_v1/captures/qkv_rank0_c0f6ec677589277b.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 0, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_flash", "status": "measured", "q_shape": [10560, 20, 128], "k_shape": [512, 20, 128], "q_len": 10560, "k_len": 512, "reference_error": {"max_abs": 0.0009765625, "rmse": 2.0951523180100097e-06, "relative_l2": 0.00027377619153512006, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.0004675048160963924, "rmse": 1.8338522143917177e-05, "relative_l2": 0.0017526865015281977, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.18035200238227844, "p95_ms": 0.1817312054336071, "p99_ms": 0.18211263343691825, "mean_ms": 0.1806131199002266, "n": 300}, {"repeat": 1, "p50_ms": 0.1804479956626892, "p95_ms": 0.1815679967403412, "p99_ms": 0.18192064687609671, "mean_ms": 0.18077685311436653, "n": 300}, {"repeat": 2, "p50_ms": 0.1804800033569336, "p95_ms": 0.18118559420108796, "p99_ms": 0.18156831681728364, "mean_ms": 0.18075146660208702, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_c0f6ec677589277b_sdpa_flash.json", "configuration": "704x1280_tp2_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp2_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp2_sp1_v1/captures/qkv_rank0_c0f6ec677589277b.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 0, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_efficient", "status": "measured", "q_shape": [10560, 20, 128], "k_shape": [512, 20, 128], "q_len": 10560, "k_len": 512, "reference_error": {"max_abs": 0.0009765625, "rmse": 3.6563941705083473e-06, "relative_l2": 0.00047778563026089604, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.0004675048160963924, "rmse": 1.850826879180923e-05, "relative_l2": 0.0017689098730793644, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.3356800079345703, "p95_ms": 0.33664000034332275, "p99_ms": 0.3372175899147987, "mean_ms": 0.33596650610367457, "n": 300}, {"repeat": 1, "p50_ms": 0.3360320031642914, "p95_ms": 0.33709118962287904, "p99_ms": 0.3375680148601532, "mean_ms": 0.3363158404827118, "n": 300}, {"repeat": 2, "p50_ms": 0.33558401465415955, "p95_ms": 0.33667360693216325, "p99_ms": 0.33728032529354096, "mean_ms": 0.33596970707178114, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_c0f6ec677589277b_sdpa_efficient.json", "configuration": "704x1280_tp2_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp2_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp2_sp1_v1/captures/qkv_rank0_c0f6ec677589277b.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 0, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_cudnn", "status": "measured", "q_shape": [10560, 20, 128], "k_shape": [512, 20, 128], "q_len": 10560, "k_len": 512, "reference_error": {"max_abs": 0.0009765625, "rmse": 3.6462431520679517e-06, "relative_l2": 0.00047645918389949073, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.0004675048160963924, "rmse": 1.850827778128504e-05, "relative_l2": 0.0017689107322398127, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.10188800096511841, "p95_ms": 0.1027199998497963, "p99_ms": 0.10360831901431082, "mean_ms": 0.10222719987233479, "n": 300}, {"repeat": 1, "p50_ms": 0.10182400047779083, "p95_ms": 0.10259519889950752, "p99_ms": 0.10297632180154323, "mean_ms": 0.10228682652115823, "n": 300}, {"repeat": 2, "p50_ms": 0.10175999999046326, "p95_ms": 0.10259359888732433, "p99_ms": 0.10310464277863503, "mean_ms": 0.10228522643446922, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_c0f6ec677589277b_sdpa_cudnn.json", "configuration": "704x1280_tp2_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp2_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp2_sp1_v1/captures/qkv_rank0_c0f6ec677589277b.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 0, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "native_dense", "status": "measured", "q_shape": [10560, 20, 128], "k_shape": [512, 20, 128], "q_len": 10560, "k_len": 512, "reference_error": {"max_abs": 0.0, "rmse": 0.0, "relative_l2": 0.0, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.0004675048160963924, "rmse": 1.8355957684623024e-05, "relative_l2": 0.0017543528864529022, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.12124799937009811, "p95_ms": 0.1317936047911644, "p99_ms": 0.166455992758274, "mean_ms": 0.1235368532439073, "n": 300}, {"repeat": 1, "p50_ms": 0.11905599758028984, "p95_ms": 0.12833759412169457, "p99_ms": 0.21107392504811284, "mean_ms": 0.12312277344365914, "n": 300}, {"repeat": 2, "p50_ms": 0.11993599683046341, "p95_ms": 0.12569600343704224, "p99_ms": 0.14695264130830765, "mean_ms": 0.1215545600404342, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_c0f6ec677589277b_native_dense.json", "configuration": "704x1280_tp2_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp2_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp2_sp1_v1/captures/qkv_rank0_c3e120abc9b34d7b.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 27, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "vllm_fa2", "status": "measured", "q_shape": [10560, 20, 128], "k_shape": [66880, 20, 128], "q_len": 10560, "k_len": 63360, "reference_error": {"max_abs": 0.03125, "rmse": 0.00042642263260645743, "relative_l2": 0.0008746236764571733, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.009407552258297702, "rmse": 0.0007486997547117268, "relative_l2": 0.0019289542681111352, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 18.143343925476074, "p95_ms": 18.79463872909546, "p99_ms": 20.044719161987306, "mean_ms": 18.207724278767902, "n": 300}, {"repeat": 1, "p50_ms": 18.156240463256836, "p95_ms": 18.628597068786622, "p99_ms": 18.793514308929442, "mean_ms": 18.204263343811036, "n": 300}, {"repeat": 2, "p50_ms": 18.171584129333496, "p95_ms": 18.460662078857425, "p99_ms": 18.795610485076903, "mean_ms": 18.187425416310628, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_c3e120abc9b34d7b_vllm_fa2.json", "configuration": "704x1280_tp2_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp2_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp2_sp1_v1/captures/qkv_rank0_c3e120abc9b34d7b.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 27, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "vllm_fa3", "status": "measured", "q_shape": [10560, 20, 128], "k_shape": [66880, 20, 128], "q_len": 10560, "k_len": 63360, "reference_error": {"max_abs": 0.0, "rmse": 0.0, "relative_l2": 0.0, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.009407552258297702, "rmse": 0.0007569944080928546, "relative_l2": 0.001950324659835371, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 10.025424003601074, "p95_ms": 10.540249681472778, "p99_ms": 11.23895341873169, "mean_ms": 10.135986467997233, "n": 300}, {"repeat": 1, "p50_ms": 10.118063926696777, "p95_ms": 10.377793550491333, "p99_ms": 10.47996081352234, "mean_ms": 10.123783798217774, "n": 300}, {"repeat": 2, "p50_ms": 10.117263793945312, "p95_ms": 10.357809782028198, "p99_ms": 10.386762371063233, "mean_ms": 10.108561080296834, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_c3e120abc9b34d7b_vllm_fa3.json", "configuration": "704x1280_tp2_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp2_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp2_sp1_v1/captures/qkv_rank0_c3e120abc9b34d7b.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 27, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_math", "status": "measured", "q_shape": [10560, 20, 128], "k_shape": [66880, 20, 128], "q_len": 10560, "k_len": 63360, "reference_error": {"max_abs": 0.03125, "rmse": 0.000770961970083123, "relative_l2": 0.0015812987893282734, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.0070496019520809305, "rmse": 0.0006435143540868139, "relative_l2": 0.0016579540090599918, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 279.77882385253906, "p95_ms": 280.57825164794923, "p99_ms": 280.7829214477539, "mean_ms": 279.71971099853516, "n": 300}, {"repeat": 1, "p50_ms": 279.8446502685547, "p95_ms": 280.72456512451174, "p99_ms": 280.9300961303711, "mean_ms": 279.763579305013, "n": 300}, {"repeat": 2, "p50_ms": 279.9040832519531, "p95_ms": 280.6882888793945, "p99_ms": 281.0607116699219, "mean_ms": 279.85775665283205, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_c3e120abc9b34d7b_sdpa_math.json", "configuration": "704x1280_tp2_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp2_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp2_sp1_v1/captures/qkv_rank0_c3e120abc9b34d7b.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 27, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_flash", "status": "measured", "q_shape": [10560, 20, 128], "k_shape": [66880, 20, 128], "q_len": 10560, "k_len": 63360, "reference_error": {"max_abs": 0.03125, "rmse": 0.0004267985295950855, "relative_l2": 0.000875394668381672, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.009407552258297702, "rmse": 0.0007487217021014184, "relative_l2": 0.0019290108134896436, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 19.627344131469727, "p95_ms": 20.04776449203491, "p99_ms": 20.962484397888183, "mean_ms": 19.640943584442137, "n": 300}, {"repeat": 1, "p50_ms": 19.54425621032715, "p95_ms": 19.912975692749022, "p99_ms": 19.9291335105896, "mean_ms": 19.608970362345378, "n": 300}, {"repeat": 2, "p50_ms": 19.687983512878418, "p95_ms": 19.8765962600708, "p99_ms": 19.91731901168823, "mean_ms": 19.60237767537435, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_c3e120abc9b34d7b_sdpa_flash.json", "configuration": "704x1280_tp2_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp2_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp2_sp1_v1/captures/qkv_rank0_c3e120abc9b34d7b.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 27, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_efficient", "status": "measured", "q_shape": [10560, 20, 128], "k_shape": [66880, 20, 128], "q_len": 10560, "k_len": 63360, "reference_error": {"max_abs": 0.03125, "rmse": 0.0006762072872589348, "relative_l2": 0.0013869500781759987, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.00931566299183828, "rmse": 0.0007445598834358421, "relative_l2": 0.0019182882804214191, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 37.468238830566406, "p95_ms": 37.53666744232178, "p99_ms": 37.54554275512695, "mean_ms": 37.484353408813476, "n": 300}, {"repeat": 1, "p50_ms": 37.48840141296387, "p95_ms": 37.93662719726562, "p99_ms": 37.9536457824707, "mean_ms": 37.57064469655355, "n": 300}, {"repeat": 2, "p50_ms": 37.87161636352539, "p95_ms": 37.960173797607425, "p99_ms": 37.96974769592285, "mean_ms": 37.87692541758219, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_c3e120abc9b34d7b_sdpa_efficient.json", "configuration": "704x1280_tp2_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp2_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp2_sp1_v1/captures/qkv_rank0_c3e120abc9b34d7b.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 27, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_cudnn", "status": "measured", "q_shape": [10560, 20, 128], "k_shape": [66880, 20, 128], "q_len": 10560, "k_len": 63360, "reference_error": {"max_abs": 0.03125, "rmse": 0.0006745384726953353, "relative_l2": 0.0013835272187465658, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.00931566299183828, "rmse": 0.000744271165335079, "relative_l2": 0.0019175444254792436, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 11.102928161621094, "p95_ms": 11.520065927505494, "p99_ms": 12.115680379867554, "mean_ms": 11.143308804829916, "n": 300}, {"repeat": 1, "p50_ms": 11.123007774353027, "p95_ms": 11.327779245376586, "p99_ms": 11.33190824508667, "mean_ms": 11.127692330678304, "n": 300}, {"repeat": 2, "p50_ms": 11.120160102844238, "p95_ms": 11.244409418106079, "p99_ms": 11.324430875778198, "mean_ms": 11.120142504374186, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_c3e120abc9b34d7b_sdpa_cudnn.json", "configuration": "704x1280_tp2_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp2_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp2_sp1_v1/captures/qkv_rank0_c3e120abc9b34d7b.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 27, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "native_dense", "status": "measured", "q_shape": [10560, 20, 128], "k_shape": [66880, 20, 128], "q_len": 10560, "k_len": 63360, "reference_error": {"max_abs": 0.03125, "rmse": 0.0004061682794040982, "relative_l2": 0.0008330805323847559, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.009407552258297702, "rmse": 0.0007567407392910509, "relative_l2": 0.0019496711061045909, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 10.08896017074585, "p95_ms": 10.747852802276611, "p99_ms": 11.013736963272095, "mean_ms": 10.143521823883056, "n": 300}, {"repeat": 1, "p50_ms": 10.162384033203125, "p95_ms": 10.3345787525177, "p99_ms": 10.605819568634033, "mean_ms": 10.115208717981973, "n": 300}, {"repeat": 2, "p50_ms": 10.154576301574707, "p95_ms": 10.330958557128906, "p99_ms": 10.390259923934934, "mean_ms": 10.1260444800059, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_c3e120abc9b34d7b_native_dense.json", "configuration": "704x1280_tp2_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp2_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp2_sp1_v1/captures/qkv_rank0_dfcab6f0a5779c76.pt", "context": {"phase": "commit", "step": null, "start_frame": 0, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "vllm_fa2", "status": "measured", "q_shape": [10560, 20, 128], "k_shape": [512, 20, 128], "q_len": 10560, "k_len": 512, "reference_error": {"max_abs": 0.0009765625, "rmse": 1.9202693341604827e-06, "relative_l2": 0.00024350410183320028, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.0002505700747839068, "rmse": 1.5871712117193417e-05, "relative_l2": 0.0015982757633049652, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.16857600212097168, "p95_ms": 0.169766403734684, "p99_ms": 0.17030496031045914, "mean_ms": 0.1688794669508934, "n": 300}, {"repeat": 1, "p50_ms": 0.16873599588871002, "p95_ms": 0.1699215978384018, "p99_ms": 0.17072063609957694, "mean_ms": 0.16925952022274335, "n": 300}, {"repeat": 2, "p50_ms": 0.16886399686336517, "p95_ms": 0.17024319991469383, "p99_ms": 0.17200832575559616, "mean_ms": 0.16936810647447903, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_dfcab6f0a5779c76_vllm_fa2.json", "configuration": "704x1280_tp2_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp2_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp2_sp1_v1/captures/qkv_rank0_dfcab6f0a5779c76.pt", "context": {"phase": "commit", "step": null, "start_frame": 0, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "vllm_fa3", "status": "measured", "q_shape": [10560, 20, 128], "k_shape": [512, 20, 128], "q_len": 10560, "k_len": 512, "reference_error": {"max_abs": 0.0009765625, "rmse": 7.234391051191076e-07, "relative_l2": 9.173733412769286e-05, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.0002505700747839068, "rmse": 1.5881886872020503e-05, "relative_l2": 0.001599300357496041, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.14195199310779572, "p95_ms": 0.14397920072078704, "p99_ms": 0.14515392258763313, "mean_ms": 0.1422983466585477, "n": 300}, {"repeat": 1, "p50_ms": 0.14195199310779572, "p95_ms": 0.1438400000333786, "p99_ms": 0.14528160348534583, "mean_ms": 0.1424833065768083, "n": 300}, {"repeat": 2, "p50_ms": 0.14207999408245087, "p95_ms": 0.14426239579916, "p99_ms": 0.14522080302238463, "mean_ms": 0.14272981330752374, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_dfcab6f0a5779c76_vllm_fa3.json", "configuration": "704x1280_tp2_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp2_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp2_sp1_v1/captures/qkv_rank0_dfcab6f0a5779c76.pt", "context": {"phase": "commit", "step": null, "start_frame": 0, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_math", "status": "measured", "q_shape": [10560, 20, 128], "k_shape": [512, 20, 128], "q_len": 10560, "k_len": 512, "reference_error": {"max_abs": 0.001953125, "rmse": 5.0415188828009065e-06, "relative_l2": 0.0006393012196740926, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.0002505700747839068, "rmse": 1.5760708729029838e-05, "relative_l2": 0.0015870977616100886, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 2.4238240718841553, "p95_ms": 2.4284815192222595, "p99_ms": 2.4315740132331847, "mean_ms": 2.423786985874176, "n": 300}, {"repeat": 1, "p50_ms": 2.4239519834518433, "p95_ms": 2.427843189239502, "p99_ms": 2.4289278984069824, "mean_ms": 2.424211620489756, "n": 300}, {"repeat": 2, "p50_ms": 2.424288034439087, "p95_ms": 2.428447961807251, "p99_ms": 2.4311371731758116, "mean_ms": 2.4241174419720966, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_dfcab6f0a5779c76_sdpa_math.json", "configuration": "704x1280_tp2_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp2_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp2_sp1_v1/captures/qkv_rank0_dfcab6f0a5779c76.pt", "context": {"phase": "commit", "step": null, "start_frame": 0, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_flash", "status": "measured", "q_shape": [10560, 20, 128], "k_shape": [512, 20, 128], "q_len": 10560, "k_len": 512, "reference_error": {"max_abs": 0.0009765625, "rmse": 1.9191084120058577e-06, "relative_l2": 0.00024335688847019373, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.0002505700747839068, "rmse": 1.5871712376959926e-05, "relative_l2": 0.0015982757894633603, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.1801759973168373, "p95_ms": 0.18147360309958457, "p99_ms": 0.1819526396691799, "mean_ms": 0.180402240306139, "n": 300}, {"repeat": 1, "p50_ms": 0.18041600286960602, "p95_ms": 0.181633597612381, "p99_ms": 0.1820479929447174, "mean_ms": 0.1807367467880249, "n": 300}, {"repeat": 2, "p50_ms": 0.1804479956626892, "p95_ms": 0.18169759809970856, "p99_ms": 0.18240416303277016, "mean_ms": 0.18081066643198332, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_dfcab6f0a5779c76_sdpa_flash.json", "configuration": "704x1280_tp2_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp2_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp2_sp1_v1/captures/qkv_rank0_dfcab6f0a5779c76.pt", "context": {"phase": "commit", "step": null, "start_frame": 0, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_efficient", "status": "measured", "q_shape": [10560, 20, 128], "k_shape": [512, 20, 128], "q_len": 10560, "k_len": 512, "reference_error": {"max_abs": 0.001953125, "rmse": 3.3573656766406083e-06, "relative_l2": 0.0004257383581941129, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.0002505700747839068, "rmse": 1.6050063555201555e-05, "relative_l2": 0.0016162356896578507, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.33582399785518646, "p95_ms": 0.3367375940084457, "p99_ms": 0.33734783083200454, "mean_ms": 0.33601599911848706, "n": 300}, {"repeat": 1, "p50_ms": 0.33587199449539185, "p95_ms": 0.3367375940084457, "p99_ms": 0.3376652881503105, "mean_ms": 0.33614858667055764, "n": 300}, {"repeat": 2, "p50_ms": 0.33529600501060486, "p95_ms": 0.33619361072778703, "p99_ms": 0.33673631370067597, "mean_ms": 0.33556885361671446, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_dfcab6f0a5779c76_sdpa_efficient.json", "configuration": "704x1280_tp2_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp2_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp2_sp1_v1/captures/qkv_rank0_dfcab6f0a5779c76.pt", "context": {"phase": "commit", "step": null, "start_frame": 0, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_cudnn", "status": "measured", "q_shape": [10560, 20, 128], "k_shape": [512, 20, 128], "q_len": 10560, "k_len": 512, "reference_error": {"max_abs": 0.001953125, "rmse": 3.360135438211077e-06, "relative_l2": 0.000426089584082852, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.0002505700747839068, "rmse": 1.605517673548522e-05, "relative_l2": 0.0016167505850932296, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.10201600193977356, "p95_ms": 0.10291360132396221, "p99_ms": 0.10339231751859188, "mean_ms": 0.10234826716283957, "n": 300}, {"repeat": 1, "p50_ms": 0.10233599692583084, "p95_ms": 0.10326719656586647, "p99_ms": 0.10442047789692878, "mean_ms": 0.10309610689679781, "n": 300}, {"repeat": 2, "p50_ms": 0.10255999863147736, "p95_ms": 0.1032319962978363, "p99_ms": 0.1037760004401207, "mean_ms": 0.10293024003505707, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_dfcab6f0a5779c76_sdpa_cudnn.json", "configuration": "704x1280_tp2_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp2_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp2_sp1_v1/captures/qkv_rank0_dfcab6f0a5779c76.pt", "context": {"phase": "commit", "step": null, "start_frame": 0, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "native_dense", "status": "measured", "q_shape": [10560, 20, 128], "k_shape": [512, 20, 128], "q_len": 10560, "k_len": 512, "reference_error": {"max_abs": 0.0, "rmse": 0.0, "relative_l2": 0.0, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.0002505700747839068, "rmse": 1.5885022568664478e-05, "relative_l2": 0.0015996161210324593, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.12044800072908401, "p95_ms": 0.12938399463891984, "p99_ms": 0.13946144238114339, "mean_ms": 0.12191392004489898, "n": 300}, {"repeat": 1, "p50_ms": 0.12009599804878235, "p95_ms": 0.1302431933581829, "p99_ms": 0.143736956119537, "mean_ms": 0.12226549332340558, "n": 300}, {"repeat": 2, "p50_ms": 0.12009599804878235, "p95_ms": 0.13013440668582918, "p99_ms": 0.14397728100419044, "mean_ms": 0.12221130671600501, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_dfcab6f0a5779c76_native_dense.json", "configuration": "704x1280_tp2_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp2_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp2_sp1_v1/captures/qkv_rank0_f5a94ff018466c49.pt", "context": {"phase": "commit", "step": null, "start_frame": 27, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "vllm_fa2", "status": "measured", "q_shape": [10560, 20, 128], "k_shape": [512, 20, 128], "q_len": 10560, "k_len": 512, "reference_error": {"max_abs": 0.00048828125, "rmse": 1.7195402821793385e-06, "relative_l2": 0.0005857406219879689, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 9.157510160012067e-05, "rmse": 4.641503394274589e-06, "relative_l2": 0.0017293633695280313, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.1685439944267273, "p95_ms": 0.16947359442710877, "p99_ms": 0.1697283160686493, "mean_ms": 0.1688038392861684, "n": 300}, {"repeat": 1, "p50_ms": 0.16859199851751328, "p95_ms": 0.1693439930677414, "p99_ms": 0.1696963231265545, "mean_ms": 0.16914069319764774, "n": 300}, {"repeat": 2, "p50_ms": 0.16867199540138245, "p95_ms": 0.16950560137629508, "p99_ms": 0.1700604784488678, "mean_ms": 0.16905194664994874, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_f5a94ff018466c49_vllm_fa2.json", "configuration": "704x1280_tp2_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp2_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp2_sp1_v1/captures/qkv_rank0_f5a94ff018466c49.pt", "context": {"phase": "commit", "step": null, "start_frame": 27, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "vllm_fa3", "status": "measured", "q_shape": [10560, 20, 128], "k_shape": [512, 20, 128], "q_len": 10560, "k_len": 512, "reference_error": {"max_abs": 0.00048828125, "rmse": 6.543344022808417e-07, "relative_l2": 0.0002228911086016263, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 9.157510160012067e-05, "rmse": 4.68971548304834e-06, "relative_l2": 0.0017473265623152226, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.14158400148153305, "p95_ms": 0.14323200285434723, "p99_ms": 0.14403296142816543, "mean_ms": 0.14187893355886141, "n": 300}, {"repeat": 1, "p50_ms": 0.14176000654697418, "p95_ms": 0.14320319518446922, "p99_ms": 0.14364831864833832, "mean_ms": 0.14222015981872876, "n": 300}, {"repeat": 2, "p50_ms": 0.14215999841690063, "p95_ms": 0.1435839980840683, "p99_ms": 0.1443206362426281, "mean_ms": 0.1426148265103499, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_f5a94ff018466c49_vllm_fa3.json", "configuration": "704x1280_tp2_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp2_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp2_sp1_v1/captures/qkv_rank0_f5a94ff018466c49.pt", "context": {"phase": "commit", "step": null, "start_frame": 27, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_math", "status": "measured", "q_shape": [10560, 20, 128], "k_shape": [512, 20, 128], "q_len": 10560, "k_len": 512, "reference_error": {"max_abs": 0.001953125, "rmse": 5.086644814396238e-06, "relative_l2": 0.0017327041002146108, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 9.157510160012067e-05, "rmse": 4.424818279924814e-06, "relative_l2": 0.00164862933409875, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 2.424896001815796, "p95_ms": 2.429070329666138, "p99_ms": 2.4308514046669005, "mean_ms": 2.4246534379323323, "n": 300}, {"repeat": 1, "p50_ms": 2.4249919652938843, "p95_ms": 2.42902557849884, "p99_ms": 2.4313290166854857, "mean_ms": 2.4251806926727295, "n": 300}, {"repeat": 2, "p50_ms": 2.425071954727173, "p95_ms": 2.429476773738861, "p99_ms": 2.4316537761688233, "mean_ms": 2.425149974822998, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_f5a94ff018466c49_sdpa_math.json", "configuration": "704x1280_tp2_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp2_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp2_sp1_v1/captures/qkv_rank0_f5a94ff018466c49.pt", "context": {"phase": "commit", "step": null, "start_frame": 27, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_flash", "status": "measured", "q_shape": [10560, 20, 128], "k_shape": [512, 20, 128], "q_len": 10560, "k_len": 512, "reference_error": {"max_abs": 0.00048828125, "rmse": 1.719604113168052e-06, "relative_l2": 0.000585762365243081, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 9.157510160012067e-05, "rmse": 4.641503394274589e-06, "relative_l2": 0.0017293633695280313, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.18168000131845474, "p95_ms": 0.18281759917736054, "p99_ms": 0.1832972750067711, "mean_ms": 0.18192693347732225, "n": 300}, {"repeat": 1, "p50_ms": 0.18171200156211853, "p95_ms": 0.18284960612654685, "p99_ms": 0.18342463597655295, "mean_ms": 0.18209205339352288, "n": 300}, {"repeat": 2, "p50_ms": 0.18172800540924072, "p95_ms": 0.18275199830532074, "p99_ms": 0.18320064172148703, "mean_ms": 0.18207125345865885, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_f5a94ff018466c49_sdpa_flash.json", "configuration": "704x1280_tp2_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp2_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp2_sp1_v1/captures/qkv_rank0_f5a94ff018466c49.pt", "context": {"phase": "commit", "step": null, "start_frame": 27, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_efficient", "status": "measured", "q_shape": [10560, 20, 128], "k_shape": [512, 20, 128], "q_len": 10560, "k_len": 512, "reference_error": {"max_abs": 0.00048828125, "rmse": 2.9603584645023902e-06, "relative_l2": 0.0010084103444830679, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 9.157510160012067e-05, "rmse": 5.349796685022096e-06, "relative_l2": 0.001993264172317952, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.3384320139884949, "p95_ms": 0.33932959139347074, "p99_ms": 0.33996831595897675, "mean_ms": 0.33869130581617357, "n": 300}, {"repeat": 1, "p50_ms": 0.33801600337028503, "p95_ms": 0.33894718885421754, "p99_ms": 0.3397155138850212, "mean_ms": 0.33837162613868715, "n": 300}, {"repeat": 2, "p50_ms": 0.3384000062942505, "p95_ms": 0.33923359811306, "p99_ms": 0.3402883329987526, "mean_ms": 0.33881919970115026, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_f5a94ff018466c49_sdpa_efficient.json", "configuration": "704x1280_tp2_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp2_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp2_sp1_v1/captures/qkv_rank0_f5a94ff018466c49.pt", "context": {"phase": "commit", "step": null, "start_frame": 27, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_cudnn", "status": "measured", "q_shape": [10560, 20, 128], "k_shape": [512, 20, 128], "q_len": 10560, "k_len": 512, "reference_error": {"max_abs": 0.00048828125, "rmse": 2.9615148266307775e-06, "relative_l2": 0.0010088042452711706, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 9.157510160012067e-05, "rmse": 5.350092648723356e-06, "relative_l2": 0.00199337444451646, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.10198400169610977, "p95_ms": 0.10275360010564327, "p99_ms": 0.10339263752102852, "mean_ms": 0.10224992031852405, "n": 300}, {"repeat": 1, "p50_ms": 0.10208000242710114, "p95_ms": 0.10291360132396221, "p99_ms": 0.10332959704101086, "mean_ms": 0.1024960000316302, "n": 300}, {"repeat": 2, "p50_ms": 0.1024319976568222, "p95_ms": 0.10345599800348282, "p99_ms": 0.1040700826048851, "mean_ms": 0.10274783993760744, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_f5a94ff018466c49_sdpa_cudnn.json", "configuration": "704x1280_tp2_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp2_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp2_sp1_v1/captures/qkv_rank0_f5a94ff018466c49.pt", "context": {"phase": "commit", "step": null, "start_frame": 27, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "native_dense", "status": "measured", "q_shape": [10560, 20, 128], "k_shape": [512, 20, 128], "q_len": 10560, "k_len": 512, "reference_error": {"max_abs": 0.0, "rmse": 0.0, "relative_l2": 0.0, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 9.157510160012067e-05, "rmse": 4.7052763804774716e-06, "relative_l2": 0.0017531243488781104, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.12150400131940842, "p95_ms": 0.13584639877080917, "p99_ms": 0.1679647973179817, "mean_ms": 0.12442752023537954, "n": 300}, {"repeat": 1, "p50_ms": 0.12118399888277054, "p95_ms": 0.14109439402818683, "p99_ms": 0.17517888605594617, "mean_ms": 0.12552405327558516, "n": 300}, {"repeat": 2, "p50_ms": 0.11900799721479416, "p95_ms": 0.13141439482569695, "p99_ms": 0.1663699185848234, "mean_ms": 0.12184384000798067, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_f5a94ff018466c49_native_dense.json", "configuration": "704x1280_tp2_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp2_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp2_sp1_v1/captures/qkv_rank0_ff91739f1c7bab2a.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 9, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "vllm_fa2", "status": "measured", "q_shape": [10560, 20, 128], "k_shape": [512, 20, 128], "q_len": 10560, "k_len": 512, "reference_error": {"max_abs": 0.00048828125, "rmse": 1.891368918804577e-06, "relative_l2": 0.0006979993282992519, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 5.8207717269661424e-05, "rmse": 5.079302519802582e-06, "relative_l2": 0.00184295455142228, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.16860799491405487, "p95_ms": 0.16921600699424744, "p99_ms": 0.16963424265384675, "mean_ms": 0.16877866645654044, "n": 300}, {"repeat": 1, "p50_ms": 0.16867199540138245, "p95_ms": 0.16937760040163993, "p99_ms": 0.17043328136205674, "mean_ms": 0.16904714663823445, "n": 300}, {"repeat": 2, "p50_ms": 0.16865599900484085, "p95_ms": 0.16937600076198578, "p99_ms": 0.16998495802283287, "mean_ms": 0.16920736024777094, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_ff91739f1c7bab2a_vllm_fa2.json", "configuration": "704x1280_tp2_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp2_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp2_sp1_v1/captures/qkv_rank0_ff91739f1c7bab2a.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 9, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "vllm_fa3", "status": "measured", "q_shape": [10560, 20, 128], "k_shape": [512, 20, 128], "q_len": 10560, "k_len": 512, "reference_error": {"max_abs": 0.000244140625, "rmse": 7.205673741641341e-07, "relative_l2": 0.000265921438255848, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 5.8207717269661424e-05, "rmse": 5.1112193995495965e-06, "relative_l2": 0.0018545351490668646, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.14204800128936768, "p95_ms": 0.1435231976211071, "p99_ms": 0.14467456623911856, "mean_ms": 0.1422968533138434, "n": 300}, {"repeat": 1, "p50_ms": 0.14209599792957306, "p95_ms": 0.14361760541796684, "p99_ms": 0.14425823584198952, "mean_ms": 0.1427020796140035, "n": 300}, {"repeat": 2, "p50_ms": 0.14220799505710602, "p95_ms": 0.14352159798145295, "p99_ms": 0.14494271352887153, "mean_ms": 0.1426755196849505, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_ff91739f1c7bab2a_vllm_fa3.json", "configuration": "704x1280_tp2_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp2_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp2_sp1_v1/captures/qkv_rank0_ff91739f1c7bab2a.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 9, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_math", "status": "measured", "q_shape": [10560, 20, 128], "k_shape": [512, 20, 128], "q_len": 10560, "k_len": 512, "reference_error": {"max_abs": 0.00048828125, "rmse": 4.647045013759424e-06, "relative_l2": 0.0017149664805904581, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 5.8207717269661424e-05, "rmse": 4.561900937039327e-06, "relative_l2": 0.0016552225551198243, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 2.4245280027389526, "p95_ms": 2.429027187824249, "p99_ms": 2.4306917452812193, "mean_ms": 2.4243699153264364, "n": 300}, {"repeat": 1, "p50_ms": 2.42467200756073, "p95_ms": 2.4284224152565, "p99_ms": 2.430273060798645, "mean_ms": 2.424575256506602, "n": 300}, {"repeat": 2, "p50_ms": 2.4248000383377075, "p95_ms": 2.4289968371391297, "p99_ms": 2.430284581184387, "mean_ms": 2.425093115965525, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_ff91739f1c7bab2a_sdpa_math.json", "configuration": "704x1280_tp2_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp2_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp2_sp1_v1/captures/qkv_rank0_ff91739f1c7bab2a.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 9, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_flash", "status": "measured", "q_shape": [10560, 20, 128], "k_shape": [512, 20, 128], "q_len": 10560, "k_len": 512, "reference_error": {"max_abs": 0.00048828125, "rmse": 1.8914483900758217e-06, "relative_l2": 0.0006980286567361297, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 5.8207717269661424e-05, "rmse": 5.079302519802582e-06, "relative_l2": 0.00184295455142228, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.1804479956626892, "p95_ms": 0.1813135951757431, "p99_ms": 0.1819212867319584, "mean_ms": 0.18060597280661264, "n": 300}, {"repeat": 1, "p50_ms": 0.18028800189495087, "p95_ms": 0.18140800297260284, "p99_ms": 0.18176159828901292, "mean_ms": 0.18065418660640717, "n": 300}, {"repeat": 2, "p50_ms": 0.18051199615001678, "p95_ms": 0.18150559663772584, "p99_ms": 0.18195295959711075, "mean_ms": 0.18076373343666394, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_ff91739f1c7bab2a_sdpa_flash.json", "configuration": "704x1280_tp2_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp2_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp2_sp1_v1/captures/qkv_rank0_ff91739f1c7bab2a.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 9, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_efficient", "status": "measured", "q_shape": [10560, 20, 128], "k_shape": [512, 20, 128], "q_len": 10560, "k_len": 512, "reference_error": {"max_abs": 0.00048828125, "rmse": 3.3046877027649495e-06, "relative_l2": 0.001219576875687818, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 7.192861749966833e-05, "rmse": 5.690608035435331e-06, "relative_l2": 0.002064758288835563, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.33606401085853577, "p95_ms": 0.33686399459838867, "p99_ms": 0.3373168128728867, "mean_ms": 0.3362346664071083, "n": 300}, {"repeat": 1, "p50_ms": 0.3359519988298416, "p95_ms": 0.33827520459890364, "p99_ms": 0.33881950736045835, "mean_ms": 0.3365658676624298, "n": 300}, {"repeat": 2, "p50_ms": 0.3357760012149811, "p95_ms": 0.3368000090122223, "p99_ms": 0.3371219167113304, "mean_ms": 0.336192427277565, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_ff91739f1c7bab2a_sdpa_efficient.json", "configuration": "704x1280_tp2_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp2_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp2_sp1_v1/captures/qkv_rank0_ff91739f1c7bab2a.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 9, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_cudnn", "status": "measured", "q_shape": [10560, 20, 128], "k_shape": [512, 20, 128], "q_len": 10560, "k_len": 512, "reference_error": {"max_abs": 0.00048828125, "rmse": 3.3046308460644172e-06, "relative_l2": 0.0012195558930342563, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 7.192861749966833e-05, "rmse": 5.69061612904296e-06, "relative_l2": 0.002064761225488964, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.10172799974679947, "p95_ms": 0.10262559913098812, "p99_ms": 0.10310528270900249, "mean_ms": 0.1020446931074063, "n": 300}, {"repeat": 1, "p50_ms": 0.1016319990158081, "p95_ms": 0.1024015974253416, "p99_ms": 0.10275456011295318, "mean_ms": 0.10206709270675977, "n": 300}, {"repeat": 2, "p50_ms": 0.10159999877214432, "p95_ms": 0.10239999741315842, "p99_ms": 0.1029152012616396, "mean_ms": 0.10207039939860503, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_ff91739f1c7bab2a_sdpa_cudnn.json", "configuration": "704x1280_tp2_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp2_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp2_sp1_v1/captures/qkv_rank0_ff91739f1c7bab2a.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 9, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "native_dense", "status": "measured", "q_shape": [10560, 20, 128], "k_shape": [512, 20, 128], "q_len": 10560, "k_len": 512, "reference_error": {"max_abs": 0.0, "rmse": 0.0, "relative_l2": 0.0, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 5.8207717269661424e-05, "rmse": 5.114539942324399e-06, "relative_l2": 0.0018557399620104048, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.12144000083208084, "p95_ms": 0.13338720202445983, "p99_ms": 0.16150527283549307, "mean_ms": 0.12378442679842314, "n": 300}, {"repeat": 1, "p50_ms": 0.12068800255656242, "p95_ms": 0.13240960240364075, "p99_ms": 0.17574496328830705, "mean_ms": 0.12423168001075585, "n": 300}, {"repeat": 2, "p50_ms": 0.1196800023317337, "p95_ms": 0.127144006639719, "p99_ms": 0.14300159841775864, "mean_ms": 0.12155189342796803, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_ff91739f1c7bab2a_native_dense.json", "configuration": "704x1280_tp2_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp2_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp2_sp2_v1/captures/qkv_rank0_1ef9f10991fa086e.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 27, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "vllm_fa2", "status": "measured", "q_shape": [10560, 10, 128], "k_shape": [512, 10, 128], "q_len": 10560, "k_len": 512, "reference_error": {"max_abs": 0.0001220703125, "rmse": 1.2544089262995998e-06, "relative_l2": 0.0007204665967617542, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 7.489595196656651e-05, "rmse": 5.54717567994717e-06, "relative_l2": 0.001952980789202713, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.09699200093746185, "p95_ms": 0.09756799787282944, "p99_ms": 0.09775999933481216, "mean_ms": 0.09720181355873744, "n": 300}, {"repeat": 1, "p50_ms": 0.09721600264310837, "p95_ms": 0.09785600006580353, "p99_ms": 0.09833664365112782, "mean_ms": 0.09756234670678775, "n": 300}, {"repeat": 2, "p50_ms": 0.09721600264310837, "p95_ms": 0.0979200005531311, "p99_ms": 0.09811296202242374, "mean_ms": 0.09750687996546427, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_1ef9f10991fa086e_vllm_fa2.json", "configuration": "704x1280_tp2_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp2_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp2_sp2_v1/captures/qkv_rank0_1ef9f10991fa086e.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 27, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "vllm_fa3", "status": "measured", "q_shape": [10560, 10, 128], "k_shape": [512, 10, 128], "q_len": 10560, "k_len": 512, "reference_error": {"max_abs": 0.0001220703125, "rmse": 4.6788641644058033e-07, "relative_l2": 0.00026872938087135913, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 7.489595196656651e-05, "rmse": 5.563208173893569e-06, "relative_l2": 0.0019586253107550977, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.08073599636554718, "p95_ms": 0.08176319673657417, "p99_ms": 0.08237056136131286, "mean_ms": 0.08096074678003788, "n": 300}, {"repeat": 1, "p50_ms": 0.08105599880218506, "p95_ms": 0.08204959891736507, "p99_ms": 0.08256448276340961, "mean_ms": 0.08147605329751968, "n": 300}, {"repeat": 2, "p50_ms": 0.08092799782752991, "p95_ms": 0.08198399841785431, "p99_ms": 0.08259552299976348, "mean_ms": 0.08137845339874426, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_1ef9f10991fa086e_vllm_fa3.json", "configuration": "704x1280_tp2_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp2_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp2_sp2_v1/captures/qkv_rank0_1ef9f10991fa086e.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 27, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_math", "status": "measured", "q_shape": [10560, 10, 128], "k_shape": [512, 10, 128], "q_len": 10560, "k_len": 512, "reference_error": {"max_abs": 0.000244140625, "rmse": 3.361599094112954e-06, "relative_l2": 0.001930725944495161, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 6.549472242076831e-05, "rmse": 4.704919328637301e-06, "relative_l2": 0.0016564496229664544, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 1.2686880230903625, "p95_ms": 1.2737055480480195, "p99_ms": 1.27542560338974, "mean_ms": 1.2689427189032236, "n": 300}, {"repeat": 1, "p50_ms": 1.2687040567398071, "p95_ms": 1.2734815537929536, "p99_ms": 1.2753958952426911, "mean_ms": 1.2693491208553314, "n": 300}, {"repeat": 2, "p50_ms": 1.2687679529190063, "p95_ms": 1.273535966873169, "p99_ms": 1.27510404586792, "mean_ms": 1.2692007486025492, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_1ef9f10991fa086e_sdpa_math.json", "configuration": "704x1280_tp2_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp2_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp2_sp2_v1/captures/qkv_rank0_1ef9f10991fa086e.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 27, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_flash", "status": "measured", "q_shape": [10560, 10, 128], "k_shape": [512, 10, 128], "q_len": 10560, "k_len": 512, "reference_error": {"max_abs": 0.0001220703125, "rmse": 1.254506867238866e-06, "relative_l2": 0.000720522848892712, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 7.489595196656651e-05, "rmse": 5.54717567994717e-06, "relative_l2": 0.001952980789202713, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.1032479964196682, "p95_ms": 0.10419200360774994, "p99_ms": 0.10448031835258006, "mean_ms": 0.10336362654964129, "n": 300}, {"repeat": 1, "p50_ms": 0.10326399654150009, "p95_ms": 0.10400480218231678, "p99_ms": 0.10470688007771968, "mean_ms": 0.10358186667164167, "n": 300}, {"repeat": 2, "p50_ms": 0.10326399654150009, "p95_ms": 0.10432159714400768, "p99_ms": 0.10480256080627441, "mean_ms": 0.10358528008063635, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_1ef9f10991fa086e_sdpa_flash.json", "configuration": "704x1280_tp2_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp2_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp2_sp2_v1/captures/qkv_rank0_1ef9f10991fa086e.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 27, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_efficient", "status": "measured", "q_shape": [10560, 10, 128], "k_shape": [512, 10, 128], "q_len": 10560, "k_len": 512, "reference_error": {"max_abs": 0.000244140625, "rmse": 2.1976266869528485e-06, "relative_l2": 0.0012622013339560474, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 7.489595196656651e-05, "rmse": 5.9867920509910385e-06, "relative_l2": 0.002107755466769057, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.17921599745750427, "p95_ms": 0.17997279912233352, "p99_ms": 0.18051679626107214, "mean_ms": 0.17935146669546764, "n": 300}, {"repeat": 1, "p50_ms": 0.17974400520324707, "p95_ms": 0.18041600286960602, "p99_ms": 0.18108864054083823, "mean_ms": 0.1800267740090688, "n": 300}, {"repeat": 2, "p50_ms": 0.17961600422859192, "p95_ms": 0.18028960153460502, "p99_ms": 0.18057663664221763, "mean_ms": 0.17982506702343623, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_1ef9f10991fa086e_sdpa_efficient.json", "configuration": "704x1280_tp2_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp2_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp2_sp2_v1/captures/qkv_rank0_1ef9f10991fa086e.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 27, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_cudnn", "status": "measured", "q_shape": [10560, 10, 128], "k_shape": [512, 10, 128], "q_len": 10560, "k_len": 512, "reference_error": {"max_abs": 0.000244140625, "rmse": 2.1965236984882847e-06, "relative_l2": 0.0012615678353188236, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 7.489595196656651e-05, "rmse": 5.9867920509910385e-06, "relative_l2": 0.002107755466769057, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.05961599946022034, "p95_ms": 0.06028800085186958, "p99_ms": 0.06089664172381162, "mean_ms": 0.05986229320367177, "n": 300}, {"repeat": 1, "p50_ms": 0.0594559982419014, "p95_ms": 0.06016319990158081, "p99_ms": 0.06051295883953571, "mean_ms": 0.059942719948788484, "n": 300}, {"repeat": 2, "p50_ms": 0.059487998485565186, "p95_ms": 0.06035200133919716, "p99_ms": 0.0607680007815361, "mean_ms": 0.05995103991279999, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_1ef9f10991fa086e_sdpa_cudnn.json", "configuration": "704x1280_tp2_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp2_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp2_sp2_v1/captures/qkv_rank0_1ef9f10991fa086e.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 27, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "native_dense", "status": "measured", "q_shape": [10560, 10, 128], "k_shape": [512, 10, 128], "q_len": 10560, "k_len": 512, "reference_error": {"max_abs": 0.0, "rmse": 0.0, "relative_l2": 0.0, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 7.489595196656651e-05, "rmse": 5.57652001207769e-06, "relative_l2": 0.001963311977582064, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.12065600231289864, "p95_ms": 0.12559039518237114, "p99_ms": 0.143051839917898, "mean_ms": 0.12229823991656304, "n": 300}, {"repeat": 1, "p50_ms": 0.11988799646496773, "p95_ms": 0.1277040034532547, "p99_ms": 0.16098527610301966, "mean_ms": 0.12193845324218273, "n": 300}, {"repeat": 2, "p50_ms": 0.12028799951076508, "p95_ms": 0.12723999992012977, "p99_ms": 0.1578998348116874, "mean_ms": 0.12257941340406736, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_1ef9f10991fa086e_native_dense.json", "configuration": "704x1280_tp2_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp2_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp2_sp2_v1/captures/qkv_rank0_6be6af8ca00bf828.pt", "context": {"phase": "commit", "step": null, "start_frame": 0, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "vllm_fa2", "status": "measured", "q_shape": [10560, 10, 128], "k_shape": [512, 10, 128], "q_len": 10560, "k_len": 512, "reference_error": {"max_abs": 0.0009765625, "rmse": 1.9141025711039493e-06, "relative_l2": 0.00030498962981155006, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.0002505700747839068, "rmse": 1.5871712117193417e-05, "relative_l2": 0.0015982757633049652, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.09718400239944458, "p95_ms": 0.09782559983432293, "p99_ms": 0.09868831895291805, "mean_ms": 0.09734869343539079, "n": 300}, {"repeat": 1, "p50_ms": 0.09715200215578079, "p95_ms": 0.09769599884748459, "p99_ms": 0.09798432104289531, "mean_ms": 0.09752416012187799, "n": 300}, {"repeat": 2, "p50_ms": 0.09708800166845322, "p95_ms": 0.09779519960284233, "p99_ms": 0.09849919751286507, "mean_ms": 0.09741077348589897, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_6be6af8ca00bf828_vllm_fa2.json", "configuration": "704x1280_tp2_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp2_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp2_sp2_v1/captures/qkv_rank0_6be6af8ca00bf828.pt", "context": {"phase": "commit", "step": null, "start_frame": 0, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "vllm_fa3", "status": "measured", "q_shape": [10560, 10, 128], "k_shape": [512, 10, 128], "q_len": 10560, "k_len": 512, "reference_error": {"max_abs": 0.0009765625, "rmse": 7.442628026974401e-07, "relative_l2": 0.0001185894842334819, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.0002505700747839068, "rmse": 1.5881886872020503e-05, "relative_l2": 0.001599300357496041, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.08081599697470665, "p95_ms": 0.08192159794270992, "p99_ms": 0.08230432085692882, "mean_ms": 0.08098143994808198, "n": 300}, {"repeat": 1, "p50_ms": 0.08097599819302559, "p95_ms": 0.0819855984300375, "p99_ms": 0.0932515233755104, "mean_ms": 0.08547850668430329, "n": 300}, {"repeat": 2, "p50_ms": 0.0811999998986721, "p95_ms": 0.0822416003793478, "p99_ms": 0.08313759975135326, "mean_ms": 0.08161482647061348, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_6be6af8ca00bf828_vllm_fa3.json", "configuration": "704x1280_tp2_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp2_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp2_sp2_v1/captures/qkv_rank0_6be6af8ca00bf828.pt", "context": {"phase": "commit", "step": null, "start_frame": 0, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_math", "status": "measured", "q_shape": [10560, 10, 128], "k_shape": [512, 10, 128], "q_len": 10560, "k_len": 512, "reference_error": {"max_abs": 0.0009765625, "rmse": 4.753911975200445e-06, "relative_l2": 0.0007574797063445034, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.0002505700747839068, "rmse": 1.5760708729029838e-05, "relative_l2": 0.0015870977616100886, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 1.2678719758987427, "p95_ms": 1.2735391676425933, "p99_ms": 1.2772496247291565, "mean_ms": 1.2685435724258423, "n": 300}, {"repeat": 1, "p50_ms": 1.268335998058319, "p95_ms": 1.2719440579414367, "p99_ms": 1.27462975025177, "mean_ms": 1.2689734395345051, "n": 300}, {"repeat": 2, "p50_ms": 1.2689599990844727, "p95_ms": 1.2729040503501892, "p99_ms": 1.2744348239898682, "mean_ms": 1.2697051715850831, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_6be6af8ca00bf828_sdpa_math.json", "configuration": "704x1280_tp2_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp2_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp2_sp2_v1/captures/qkv_rank0_6be6af8ca00bf828.pt", "context": {"phase": "commit", "step": null, "start_frame": 0, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_flash", "status": "measured", "q_shape": [10560, 10, 128], "k_shape": [512, 10, 128], "q_len": 10560, "k_len": 512, "reference_error": {"max_abs": 0.0009765625, "rmse": 1.911747078200072e-06, "relative_l2": 0.00030461430984718527, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.0002505700747839068, "rmse": 1.5871712376959926e-05, "relative_l2": 0.0015982757894633603, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.10345599800348282, "p95_ms": 0.10422399640083313, "p99_ms": 0.10470432005822658, "mean_ms": 0.10352031990885735, "n": 300}, {"repeat": 1, "p50_ms": 0.1035199984908104, "p95_ms": 0.10425759665668011, "p99_ms": 0.10483232103288173, "mean_ms": 0.1037894393503666, "n": 300}, {"repeat": 2, "p50_ms": 0.10355199873447418, "p95_ms": 0.10422399640083313, "p99_ms": 0.10457631908357143, "mean_ms": 0.10376437296470006, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_6be6af8ca00bf828_sdpa_flash.json", "configuration": "704x1280_tp2_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp2_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp2_sp2_v1/captures/qkv_rank0_6be6af8ca00bf828.pt", "context": {"phase": "commit", "step": null, "start_frame": 0, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_efficient", "status": "measured", "q_shape": [10560, 10, 128], "k_shape": [512, 10, 128], "q_len": 10560, "k_len": 512, "reference_error": {"max_abs": 0.0009765625, "rmse": 3.396901395193959e-06, "relative_l2": 0.0005412561033388217, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.0002505700747839068, "rmse": 1.6050063555201555e-05, "relative_l2": 0.0016162356896578507, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.17919999361038208, "p95_ms": 0.17990559935569764, "p99_ms": 0.18048064336180686, "mean_ms": 0.17931679949164392, "n": 300}, {"repeat": 1, "p50_ms": 0.17919999361038208, "p95_ms": 0.1799360066652298, "p99_ms": 0.18063999712467194, "mean_ms": 0.17950122619668643, "n": 300}, {"repeat": 2, "p50_ms": 0.1793919950723648, "p95_ms": 0.1802575945854187, "p99_ms": 0.18070527762174607, "mean_ms": 0.17970005333423614, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_6be6af8ca00bf828_sdpa_efficient.json", "configuration": "704x1280_tp2_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp2_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp2_sp2_v1/captures/qkv_rank0_6be6af8ca00bf828.pt", "context": {"phase": "commit", "step": null, "start_frame": 0, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_cudnn", "status": "measured", "q_shape": [10560, 10, 128], "k_shape": [512, 10, 128], "q_len": 10560, "k_len": 512, "reference_error": {"max_abs": 0.0009765625, "rmse": 3.403164648699807e-06, "relative_l2": 0.0005422540787853846, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.0002505700747839068, "rmse": 1.605517673548522e-05, "relative_l2": 0.0016167505850932296, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.059647999703884125, "p95_ms": 0.0839775986969471, "p99_ms": 0.08976159803569317, "mean_ms": 0.06235967999945084, "n": 300}, {"repeat": 1, "p50_ms": 0.05942400172352791, "p95_ms": 0.060127999633550644, "p99_ms": 0.06054431907832623, "mean_ms": 0.05989845335483551, "n": 300}, {"repeat": 2, "p50_ms": 0.05950399860739708, "p95_ms": 0.06022400036454201, "p99_ms": 0.06109055951237678, "mean_ms": 0.059962986807028454, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_6be6af8ca00bf828_sdpa_cudnn.json", "configuration": "704x1280_tp2_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp2_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp2_sp2_v1/captures/qkv_rank0_6be6af8ca00bf828.pt", "context": {"phase": "commit", "step": null, "start_frame": 0, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "native_dense", "status": "measured", "q_shape": [10560, 10, 128], "k_shape": [512, 10, 128], "q_len": 10560, "k_len": 512, "reference_error": {"max_abs": 0.0, "rmse": 0.0, "relative_l2": 0.0, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.0002505700747839068, "rmse": 1.5885022568664478e-05, "relative_l2": 0.0015996161210324593, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.12166400253772736, "p95_ms": 0.13709920048713684, "p99_ms": 0.16300512343645085, "mean_ms": 0.1251699202011029, "n": 300}, {"repeat": 1, "p50_ms": 0.12211199849843979, "p95_ms": 0.1529567949473858, "p99_ms": 0.41086400777101445, "mean_ms": 0.13287893325090408, "n": 300}, {"repeat": 2, "p50_ms": 0.12223999947309494, "p95_ms": 0.16572640091180807, "p99_ms": 0.4221072006225583, "mean_ms": 0.1356779734045267, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_6be6af8ca00bf828_native_dense.json", "configuration": "704x1280_tp2_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp2_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp2_sp2_v1/captures/qkv_rank0_6e53ae344202af4b.pt", "context": {"phase": "commit", "step": null, "start_frame": 9, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "vllm_fa2", "status": "measured", "q_shape": [10560, 10, 128], "k_shape": [66880, 10, 128], "q_len": 10560, "k_len": 42240, "reference_error": {"max_abs": 0.0625, "rmse": 0.00040144978997597126, "relative_l2": 0.0004991800503927156, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.015384331463739187, "rmse": 0.000923460013183313, "relative_l2": 0.001694139676876375, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 6.398288011550903, "p95_ms": 6.778478312492371, "p99_ms": 7.047342653274536, "mean_ms": 6.468538982073466, "n": 300}, {"repeat": 1, "p50_ms": 6.441488027572632, "p95_ms": 6.715043044090271, "p99_ms": 6.722053565979004, "mean_ms": 6.471869125366211, "n": 300}, {"repeat": 2, "p50_ms": 6.446304082870483, "p95_ms": 6.722019100189209, "p99_ms": 6.77779453754425, "mean_ms": 6.476321070988973, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_6e53ae344202af4b_vllm_fa2.json", "configuration": "704x1280_tp2_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp2_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp2_sp2_v1/captures/qkv_rank0_6e53ae344202af4b.pt", "context": {"phase": "commit", "step": null, "start_frame": 9, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "vllm_fa3", "status": "measured", "q_shape": [10560, 10, 128], "k_shape": [66880, 10, 128], "q_len": 10560, "k_len": 42240, "reference_error": {"max_abs": 0.0, "rmse": 0.0, "relative_l2": 0.0, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.015384331463739187, "rmse": 0.0009240212371461509, "relative_l2": 0.001695169274010505, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 3.519536018371582, "p95_ms": 3.9270143628120424, "p99_ms": 4.071424851417541, "mean_ms": 3.563025493621826, "n": 300}, {"repeat": 1, "p50_ms": 3.555743932723999, "p95_ms": 3.7301520228385927, "p99_ms": 3.7347916197776794, "mean_ms": 3.568769815762838, "n": 300}, {"repeat": 2, "p50_ms": 3.553679943084717, "p95_ms": 3.647465538978577, "p99_ms": 3.649601948261261, "mean_ms": 3.5603150963783263, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_6e53ae344202af4b_vllm_fa3.json", "configuration": "704x1280_tp2_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp2_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp2_sp2_v1/captures/qkv_rank0_6e53ae344202af4b.pt", "context": {"phase": "commit", "step": null, "start_frame": 9, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_math", "status": "measured", "q_shape": [10560, 10, 128], "k_shape": [66880, 10, 128], "q_len": 10560, "k_len": 42240, "reference_error": {"max_abs": 0.0625, "rmse": 0.0007775462376875909, "relative_l2": 0.0009668346572924845, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.015384331463739187, "rmse": 0.0008987087561677207, "relative_l2": 0.0016487320945619732, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 98.2746696472168, "p95_ms": 98.8175235748291, "p99_ms": 98.98556190490723, "mean_ms": 98.2571021270752, "n": 300}, {"repeat": 1, "p50_ms": 98.45473861694336, "p95_ms": 99.02250213623047, "p99_ms": 99.21492874145508, "mean_ms": 98.4370389811198, "n": 300}, {"repeat": 2, "p50_ms": 98.43196868896484, "p95_ms": 98.93200950622558, "p99_ms": 99.14038742065429, "mean_ms": 98.40584635416667, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_6e53ae344202af4b_sdpa_math.json", "configuration": "704x1280_tp2_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp2_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp2_sp2_v1/captures/qkv_rank0_6e53ae344202af4b.pt", "context": {"phase": "commit", "step": null, "start_frame": 9, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_flash", "status": "measured", "q_shape": [10560, 10, 128], "k_shape": [66880, 10, 128], "q_len": 10560, "k_len": 42240, "reference_error": {"max_abs": 0.0625, "rmse": 0.00040246803514717067, "relative_l2": 0.0005004461805254575, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.015865668536260813, "rmse": 0.0009243712902404814, "relative_l2": 0.0016958114662306905, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 6.937999963760376, "p95_ms": 7.29425618648529, "p99_ms": 7.583738384246826, "mean_ms": 7.005809599558512, "n": 300}, {"repeat": 1, "p50_ms": 7.001392126083374, "p95_ms": 7.118435382843018, "p99_ms": 7.1850343465805055, "mean_ms": 7.002111562093099, "n": 300}, {"repeat": 2, "p50_ms": 6.999376058578491, "p95_ms": 7.135071969032287, "p99_ms": 7.204497179985046, "mean_ms": 7.001624851226807, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_6e53ae344202af4b_sdpa_flash.json", "configuration": "704x1280_tp2_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp2_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp2_sp2_v1/captures/qkv_rank0_6e53ae344202af4b.pt", "context": {"phase": "commit", "step": null, "start_frame": 9, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_efficient", "status": "measured", "q_shape": [10560, 10, 128], "k_shape": [66880, 10, 128], "q_len": 10560, "k_len": 42240, "reference_error": {"max_abs": 0.0625, "rmse": 0.0006764314493664915, "relative_l2": 0.0008411041515358461, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.015384331463739187, "rmse": 0.0009043838547521623, "relative_l2": 0.0016591433842170012, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 13.234655857086182, "p95_ms": 13.27618703842163, "p99_ms": 13.281425046920777, "mean_ms": 13.238038171132406, "n": 300}, {"repeat": 1, "p50_ms": 13.233039855957031, "p95_ms": 13.275004577636718, "p99_ms": 13.281182699203491, "mean_ms": 13.238002548217773, "n": 300}, {"repeat": 2, "p50_ms": 13.232751846313477, "p95_ms": 13.275891399383545, "p99_ms": 13.282121725082398, "mean_ms": 13.238072010676065, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_6e53ae344202af4b_sdpa_efficient.json", "configuration": "704x1280_tp2_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp2_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp2_sp2_v1/captures/qkv_rank0_6e53ae344202af4b.pt", "context": {"phase": "commit", "step": null, "start_frame": 9, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_cudnn", "status": "measured", "q_shape": [10560, 10, 128], "k_shape": [66880, 10, 128], "q_len": 10560, "k_len": 42240, "reference_error": {"max_abs": 0.0625, "rmse": 0.0006754088167188533, "relative_l2": 0.000839832565824936, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.015384331463739187, "rmse": 0.0009044680810150898, "relative_l2": 0.0016592979020649031, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 3.8899039030075073, "p95_ms": 4.339363265037536, "p99_ms": 4.526017832756042, "mean_ms": 3.960382076104482, "n": 300}, {"repeat": 1, "p50_ms": 3.9307520389556885, "p95_ms": 4.086672139167786, "p99_ms": 4.121775250434876, "mean_ms": 3.9479667202631634, "n": 300}, {"repeat": 2, "p50_ms": 3.9424959421157837, "p95_ms": 4.087993574142456, "p99_ms": 4.128136692047119, "mean_ms": 3.9482324186960858, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_6e53ae344202af4b_sdpa_cudnn.json", "configuration": "704x1280_tp2_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp2_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp2_sp2_v1/captures/qkv_rank0_6e53ae344202af4b.pt", "context": {"phase": "commit", "step": null, "start_frame": 9, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "native_dense", "status": "measured", "q_shape": [10560, 10, 128], "k_shape": [66880, 10, 128], "q_len": 10560, "k_len": 42240, "reference_error": {"max_abs": 0.0625, "rmse": 0.00038308774886616607, "relative_l2": 0.00047634789345708914, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.015384331463739187, "rmse": 0.0009260597688801338, "relative_l2": 0.001698909075890185, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 3.508288025856018, "p95_ms": 3.9980303883552555, "p99_ms": 4.133527274131775, "mean_ms": 3.5638381838798523, "n": 300}, {"repeat": 1, "p50_ms": 3.5215680599212646, "p95_ms": 3.7763311862945557, "p99_ms": 3.7936199212074277, "mean_ms": 3.56819371064504, "n": 300}, {"repeat": 2, "p50_ms": 3.564527988433838, "p95_ms": 3.699515151977539, "p99_ms": 3.7105942702293397, "mean_ms": 3.5584862875938414, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_6e53ae344202af4b_native_dense.json", "configuration": "704x1280_tp2_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp2_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp2_sp2_v1/captures/qkv_rank0_7a044869db5ec3c0.pt", "context": {"phase": "commit", "step": null, "start_frame": 27, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "vllm_fa2", "status": "measured", "q_shape": [10560, 10, 128], "k_shape": [66880, 10, 128], "q_len": 10560, "k_len": 63360, "reference_error": {"max_abs": 0.0625, "rmse": 0.0003990636627895533, "relative_l2": 0.0005010281921172143, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.00897828162034564, "rmse": 0.0009021776803715486, "relative_l2": 0.0016666561231022962, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 9.668384075164795, "p95_ms": 10.079819011688233, "p99_ms": 10.666861810684203, "mean_ms": 9.728805316289266, "n": 300}, {"repeat": 1, "p50_ms": 9.690592288970947, "p95_ms": 10.0883358001709, "p99_ms": 10.156068515777587, "mean_ms": 9.743685569763183, "n": 300}, {"repeat": 2, "p50_ms": 9.816320419311523, "p95_ms": 10.073279905319215, "p99_ms": 10.088320417404175, "mean_ms": 9.78908823331197, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_7a044869db5ec3c0_vllm_fa2.json", "configuration": "704x1280_tp2_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp2_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp2_sp2_v1/captures/qkv_rank0_7a044869db5ec3c0.pt", "context": {"phase": "commit", "step": null, "start_frame": 27, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "vllm_fa3", "status": "measured", "q_shape": [10560, 10, 128], "k_shape": [66880, 10, 128], "q_len": 10560, "k_len": 63360, "reference_error": {"max_abs": 0.0, "rmse": 0.0, "relative_l2": 0.0, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.009036062330048722, "rmse": 0.000902883153287687, "relative_l2": 0.0016679593927141987, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 5.30078387260437, "p95_ms": 5.809065461158752, "p99_ms": 5.934304556846619, "mean_ms": 5.3451931762695315, "n": 300}, {"repeat": 1, "p50_ms": 5.360256195068359, "p95_ms": 5.522579407691955, "p99_ms": 5.625665192604065, "mean_ms": 5.378524570465088, "n": 300}, {"repeat": 2, "p50_ms": 5.356096029281616, "p95_ms": 5.41424789428711, "p99_ms": 5.45859769821167, "mean_ms": 5.345047674179077, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_7a044869db5ec3c0_vllm_fa3.json", "configuration": "704x1280_tp2_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp2_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp2_sp2_v1/captures/qkv_rank0_7a044869db5ec3c0.pt", "context": {"phase": "commit", "step": null, "start_frame": 27, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_math", "status": "measured", "q_shape": [10560, 10, 128], "k_shape": [66880, 10, 128], "q_len": 10560, "k_len": 63360, "reference_error": {"max_abs": 0.0625, "rmse": 0.0007956178398917287, "relative_l2": 0.0009989057012874973, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.0077161019941787146, "rmse": 0.0008886268973423762, "relative_l2": 0.0016416228109291332, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 147.62710571289062, "p95_ms": 148.4553077697754, "p99_ms": 148.84964630126953, "mean_ms": 147.65776255289714, "n": 300}, {"repeat": 1, "p50_ms": 147.73433685302734, "p95_ms": 148.49377059936523, "p99_ms": 148.75928756713867, "mean_ms": 147.75057876586914, "n": 300}, {"repeat": 2, "p50_ms": 147.7101593017578, "p95_ms": 148.4431465148926, "p99_ms": 148.72955169677735, "mean_ms": 147.7672549947103, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_7a044869db5ec3c0_sdpa_math.json", "configuration": "704x1280_tp2_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp2_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp2_sp2_v1/captures/qkv_rank0_7a044869db5ec3c0.pt", "context": {"phase": "commit", "step": null, "start_frame": 27, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_flash", "status": "measured", "q_shape": [10560, 10, 128], "k_shape": [66880, 10, 128], "q_len": 10560, "k_len": 63360, "reference_error": {"max_abs": 0.0625, "rmse": 0.00039997350014833385, "relative_l2": 0.0005021705015016468, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.00897828162034564, "rmse": 0.0009021776920801008, "relative_l2": 0.0016666561447323263, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 10.485024452209473, "p95_ms": 10.849702167510987, "p99_ms": 11.227838363647459, "mean_ms": 10.5027463722229, "n": 300}, {"repeat": 1, "p50_ms": 10.493167877197266, "p95_ms": 10.677163410186768, "p99_ms": 10.793001127243041, "mean_ms": 10.490747191111247, "n": 300}, {"repeat": 2, "p50_ms": 10.489391803741455, "p95_ms": 10.670843601226807, "p99_ms": 10.78079517364502, "mean_ms": 10.487840518951415, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_7a044869db5ec3c0_sdpa_flash.json", "configuration": "704x1280_tp2_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp2_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp2_sp2_v1/captures/qkv_rank0_7a044869db5ec3c0.pt", "context": {"phase": "commit", "step": null, "start_frame": 27, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_efficient", "status": "measured", "q_shape": [10560, 10, 128], "k_shape": [66880, 10, 128], "q_len": 10560, "k_len": 63360, "reference_error": {"max_abs": 0.0625, "rmse": 0.0006911761459779374, "relative_l2": 0.0008677781696112253, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.008240741969119725, "rmse": 0.0008973692801195905, "relative_l2": 0.0016577732279735317, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 19.840591430664062, "p95_ms": 19.889506340026855, "p99_ms": 19.900302028656007, "mean_ms": 19.851155363718668, "n": 300}, {"repeat": 1, "p50_ms": 19.843119621276855, "p95_ms": 19.891533946990968, "p99_ms": 19.901131076812746, "mean_ms": 19.850606931050617, "n": 300}, {"repeat": 2, "p50_ms": 19.847216606140137, "p95_ms": 19.89368267059326, "p99_ms": 19.906758365631102, "mean_ms": 19.854666989644368, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_7a044869db5ec3c0_sdpa_efficient.json", "configuration": "704x1280_tp2_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp2_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp2_sp2_v1/captures/qkv_rank0_7a044869db5ec3c0.pt", "context": {"phase": "commit", "step": null, "start_frame": 27, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_cudnn", "status": "measured", "q_shape": [10560, 10, 128], "k_shape": [66880, 10, 128], "q_len": 10560, "k_len": 63360, "reference_error": {"max_abs": 0.0625, "rmse": 0.0006919490626298462, "relative_l2": 0.0008687485737569101, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.008240741969119725, "rmse": 0.0008969490156068632, "relative_l2": 0.0016569968438546394, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 5.880336046218872, "p95_ms": 6.281638526916505, "p99_ms": 6.504695124626159, "mean_ms": 5.928320636749268, "n": 300}, {"repeat": 1, "p50_ms": 5.9267518520355225, "p95_ms": 6.043839979171753, "p99_ms": 6.107493963241577, "mean_ms": 5.923622832298279, "n": 300}, {"repeat": 2, "p50_ms": 5.945568084716797, "p95_ms": 6.022708797454834, "p99_ms": 6.050484952926635, "mean_ms": 5.941730351448059, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_7a044869db5ec3c0_sdpa_cudnn.json", "configuration": "704x1280_tp2_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp2_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp2_sp2_v1/captures/qkv_rank0_7a044869db5ec3c0.pt", "context": {"phase": "commit", "step": null, "start_frame": 27, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "native_dense", "status": "measured", "q_shape": [10560, 10, 128], "k_shape": [66880, 10, 128], "q_len": 10560, "k_len": 63360, "reference_error": {"max_abs": 0.0625, "rmse": 0.00038387850098339625, "relative_l2": 0.0004819630782615375, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.00897828162034564, "rmse": 0.0009014308854093113, "relative_l2": 0.0016652765163756014, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 5.315119981765747, "p95_ms": 5.6747297763824465, "p99_ms": 5.853211946487426, "mean_ms": 5.322096420923869, "n": 300}, {"repeat": 1, "p50_ms": 5.333951950073242, "p95_ms": 5.546623873710632, "p99_ms": 5.632070956230163, "mean_ms": 5.351787521044414, "n": 300}, {"repeat": 2, "p50_ms": 5.333375930786133, "p95_ms": 5.438915109634399, "p99_ms": 5.470480551719666, "mean_ms": 5.335509451230367, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_7a044869db5ec3c0_native_dense.json", "configuration": "704x1280_tp2_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp2_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp2_sp2_v1/captures/qkv_rank0_8d9062a68f7c0930.pt", "context": {"phase": "commit", "step": null, "start_frame": 0, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "vllm_fa2", "status": "measured", "q_shape": [10560, 10, 128], "k_shape": [66880, 10, 128], "q_len": 10560, "k_len": 10560, "reference_error": {"max_abs": 0.03125, "rmse": 0.0003748238619279013, "relative_l2": 0.0005077558711958171, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.00800240782572681, "rmse": 0.000782955795343756, "relative_l2": 0.0016773537871330796, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 1.6331839561462402, "p95_ms": 1.6793919801712036, "p99_ms": 1.6835529863834382, "mean_ms": 1.638827729622523, "n": 300}, {"repeat": 1, "p50_ms": 1.6241440176963806, "p95_ms": 1.6920672416687013, "p99_ms": 1.6930246794223784, "mean_ms": 1.6325091199080148, "n": 300}, {"repeat": 2, "p50_ms": 1.6239359974861145, "p95_ms": 1.6816703915596007, "p99_ms": 1.6840338838100433, "mean_ms": 1.6302858662605286, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_8d9062a68f7c0930_vllm_fa2.json", "configuration": "704x1280_tp2_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp2_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp2_sp2_v1/captures/qkv_rank0_8d9062a68f7c0930.pt", "context": {"phase": "commit", "step": null, "start_frame": 0, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "vllm_fa3", "status": "measured", "q_shape": [10560, 10, 128], "k_shape": [66880, 10, 128], "q_len": 10560, "k_len": 10560, "reference_error": {"max_abs": 0.0, "rmse": 0.0, "relative_l2": 0.0, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.00800240782572681, "rmse": 0.0007835577224721566, "relative_l2": 0.0016786433168286322, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.8984320163726807, "p95_ms": 0.9067551910877227, "p99_ms": 0.9073923444747924, "mean_ms": 0.8920828809340795, "n": 300}, {"repeat": 1, "p50_ms": 0.8994240164756775, "p95_ms": 0.908707183599472, "p99_ms": 0.9095388841629028, "mean_ms": 0.8971169084310532, "n": 300}, {"repeat": 2, "p50_ms": 0.9069919884204865, "p95_ms": 0.9285823941230774, "p99_ms": 0.9301513719558716, "mean_ms": 0.9059336541096369, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_8d9062a68f7c0930_vllm_fa3.json", "configuration": "704x1280_tp2_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp2_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp2_sp2_v1/captures/qkv_rank0_8d9062a68f7c0930.pt", "context": {"phase": "commit", "step": null, "start_frame": 0, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_math", "status": "measured", "q_shape": [10560, 10, 128], "k_shape": [66880, 10, 128], "q_len": 10560, "k_len": 10560, "reference_error": {"max_abs": 0.0625, "rmse": 0.0006830392300338293, "relative_l2": 0.0009252804171081794, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.007774497258635549, "rmse": 0.0007712550091903823, "relative_l2": 0.0016522867806896585, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 23.416223526000977, "p95_ms": 23.90531635284424, "p99_ms": 24.314093933105468, "mean_ms": 23.496831073760987, "n": 300}, {"repeat": 1, "p50_ms": 23.450400352478027, "p95_ms": 23.830215454101562, "p99_ms": 23.89939085006714, "mean_ms": 23.501717332204183, "n": 300}, {"repeat": 2, "p50_ms": 23.47468852996826, "p95_ms": 23.853000926971436, "p99_ms": 23.973524551391602, "mean_ms": 23.512590300242106, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_8d9062a68f7c0930_sdpa_math.json", "configuration": "704x1280_tp2_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp2_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp2_sp2_v1/captures/qkv_rank0_8d9062a68f7c0930.pt", "context": {"phase": "commit", "step": null, "start_frame": 0, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_flash", "status": "measured", "q_shape": [10560, 10, 128], "k_shape": [66880, 10, 128], "q_len": 10560, "k_len": 10560, "reference_error": {"max_abs": 0.03125, "rmse": 0.000375507060814067, "relative_l2": 0.0005086813678914139, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.00800240782572681, "rmse": 0.000782955795343756, "relative_l2": 0.0016773537871330796, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 1.7343679666519165, "p95_ms": 1.7412592113018035, "p99_ms": 1.7446438324451448, "mean_ms": 1.7350114130973815, "n": 300}, {"repeat": 1, "p50_ms": 1.7363680005073547, "p95_ms": 1.816388761997223, "p99_ms": 1.8238454616069795, "mean_ms": 1.7597887992858887, "n": 300}, {"repeat": 2, "p50_ms": 1.7340480089187622, "p95_ms": 1.7779584407806397, "p99_ms": 1.7820854938030244, "mean_ms": 1.7428361598650615, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_8d9062a68f7c0930_sdpa_flash.json", "configuration": "704x1280_tp2_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp2_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp2_sp2_v1/captures/qkv_rank0_8d9062a68f7c0930.pt", "context": {"phase": "commit", "step": null, "start_frame": 0, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_efficient", "status": "measured", "q_shape": [10560, 10, 128], "k_shape": [66880, 10, 128], "q_len": 10560, "k_len": 10560, "reference_error": {"max_abs": 0.0625, "rmse": 0.0005714978740305569, "relative_l2": 0.0007741807029637836, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.00800240782572681, "rmse": 0.000776607802202966, "relative_l2": 0.001663754257761534, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 3.340783953666687, "p95_ms": 3.354043257236481, "p99_ms": 3.365794916152954, "mean_ms": 3.3396942901611326, "n": 300}, {"repeat": 1, "p50_ms": 3.3379679918289185, "p95_ms": 3.3498239517211914, "p99_ms": 3.3538676047325136, "mean_ms": 3.337085761229197, "n": 300}, {"repeat": 2, "p50_ms": 3.3376320600509644, "p95_ms": 3.3512815952301027, "p99_ms": 3.3543475437164307, "mean_ms": 3.3378106729189554, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_8d9062a68f7c0930_sdpa_efficient.json", "configuration": "704x1280_tp2_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp2_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp2_sp2_v1/captures/qkv_rank0_8d9062a68f7c0930.pt", "context": {"phase": "commit", "step": null, "start_frame": 0, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_cudnn", "status": "measured", "q_shape": [10560, 10, 128], "k_shape": [66880, 10, 128], "q_len": 10560, "k_len": 10560, "reference_error": {"max_abs": 0.03125, "rmse": 0.0005681196722917395, "relative_l2": 0.0007696044154292979, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.00800240782572681, "rmse": 0.0007766164903600255, "relative_l2": 0.0016637728707065225, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 1.0561280250549316, "p95_ms": 1.0668239712715148, "p99_ms": 1.068064935207367, "mean_ms": 1.0469622433185577, "n": 300}, {"repeat": 1, "p50_ms": 1.0574560165405273, "p95_ms": 1.064545565843582, "p99_ms": 1.0650246584415435, "mean_ms": 1.0528029882907868, "n": 300}, {"repeat": 2, "p50_ms": 1.078383982181549, "p95_ms": 1.1577040076255798, "p99_ms": 1.1588509273529053, "mean_ms": 1.0859459153811137, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_8d9062a68f7c0930_sdpa_cudnn.json", "configuration": "704x1280_tp2_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp2_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp2_sp2_v1/captures/qkv_rank0_8d9062a68f7c0930.pt", "context": {"phase": "commit", "step": null, "start_frame": 0, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "native_dense", "status": "measured", "q_shape": [10560, 10, 128], "k_shape": [66880, 10, 128], "q_len": 10560, "k_len": 10560, "reference_error": {"max_abs": 0.03125, "rmse": 0.00035853552689640736, "relative_l2": 0.00048569084656876604, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.00800240782572681, "rmse": 0.0007843154270865668, "relative_l2": 0.001680266574121148, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.8792480230331421, "p95_ms": 0.8950751841068267, "p99_ms": 0.8955209320783615, "mean_ms": 0.8730627202987671, "n": 300}, {"repeat": 1, "p50_ms": 0.874783992767334, "p95_ms": 0.8813760280609131, "p99_ms": 0.8824649518728256, "mean_ms": 0.8699252269665401, "n": 300}, {"repeat": 2, "p50_ms": 0.8815680146217346, "p95_ms": 0.8988223910331726, "p99_ms": 0.9088092857599258, "mean_ms": 0.880600747068723, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_8d9062a68f7c0930_native_dense.json", "configuration": "704x1280_tp2_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp2_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp2_sp2_v1/captures/qkv_rank0_b0c50d31de479cf8.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 0, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "vllm_fa2", "status": "measured", "q_shape": [10560, 10, 128], "k_shape": [66880, 10, 128], "q_len": 10560, "k_len": 10560, "reference_error": {"max_abs": 0.03125, "rmse": 0.00046034996679920464, "relative_l2": 0.0008378192022859384, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.008999835780528098, "rmse": 0.0006803289293460923, "relative_l2": 0.001885584018998539, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 1.6507519483566284, "p95_ms": 1.6822464108467101, "p99_ms": 1.685703365802765, "mean_ms": 1.6500188791751862, "n": 300}, {"repeat": 1, "p50_ms": 1.6689119935035706, "p95_ms": 1.708675241470337, "p99_ms": 1.7098582732677459, "mean_ms": 1.6609615989526112, "n": 300}, {"repeat": 2, "p50_ms": 1.651199996471405, "p95_ms": 1.7079376459121705, "p99_ms": 1.7091219544410705, "mean_ms": 1.6512055456638337, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_b0c50d31de479cf8_vllm_fa2.json", "configuration": "704x1280_tp2_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp2_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp2_sp2_v1/captures/qkv_rank0_b0c50d31de479cf8.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 0, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "vllm_fa3", "status": "measured", "q_shape": [10560, 10, 128], "k_shape": [66880, 10, 128], "q_len": 10560, "k_len": 10560, "reference_error": {"max_abs": 0.0, "rmse": 0.0, "relative_l2": 0.0, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.008999835780528098, "rmse": 0.000694164453688696, "relative_l2": 0.0019239302401712203, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.9238719940185547, "p95_ms": 0.9310111820697784, "p99_ms": 0.9317129319906234, "mean_ms": 0.9138708265622457, "n": 300}, {"repeat": 1, "p50_ms": 0.9195200204849243, "p95_ms": 0.9341520190238952, "p99_ms": 0.935267191529274, "mean_ms": 0.9164026641845703, "n": 300}, {"repeat": 2, "p50_ms": 0.9180960059165955, "p95_ms": 0.954028770327568, "p99_ms": 0.958368963599205, "mean_ms": 0.9214210120836893, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_b0c50d31de479cf8_vllm_fa3.json", "configuration": "704x1280_tp2_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp2_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp2_sp2_v1/captures/qkv_rank0_b0c50d31de479cf8.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 0, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_math", "status": "measured", "q_shape": [10560, 10, 128], "k_shape": [66880, 10, 128], "q_len": 10560, "k_len": 10560, "reference_error": {"max_abs": 0.03125, "rmse": 0.0007805397578876906, "relative_l2": 0.0014205522851512761, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.0072374631331095785, "rmse": 0.0005975725893755776, "relative_l2": 0.0016562184498036551, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 23.497135162353516, "p95_ms": 24.05724639892578, "p99_ms": 25.293018321990964, "mean_ms": 23.58132595062256, "n": 300}, {"repeat": 1, "p50_ms": 23.51363182067871, "p95_ms": 23.929773616790772, "p99_ms": 23.97041639328003, "mean_ms": 23.568693288167317, "n": 300}, {"repeat": 2, "p50_ms": 23.525839805603027, "p95_ms": 23.943426609039307, "p99_ms": 24.03615951538086, "mean_ms": 23.585102475484213, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_b0c50d31de479cf8_sdpa_math.json", "configuration": "704x1280_tp2_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp2_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp2_sp2_v1/captures/qkv_rank0_b0c50d31de479cf8.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 0, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_flash", "status": "measured", "q_shape": [10560, 10, 128], "k_shape": [66880, 10, 128], "q_len": 10560, "k_len": 10560, "reference_error": {"max_abs": 0.03125, "rmse": 0.00046066797258583256, "relative_l2": 0.0008383979605648351, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.008999835780528098, "rmse": 0.0006802589472844292, "relative_l2": 0.0018853900583254049, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 1.7504479885101318, "p95_ms": 1.768489545583725, "p99_ms": 1.772275856733322, "mean_ms": 1.7528665618101755, "n": 300}, {"repeat": 1, "p50_ms": 1.761247992515564, "p95_ms": 1.8604847550392152, "p99_ms": 1.8684633791446685, "mean_ms": 1.7807796267668405, "n": 300}, {"repeat": 2, "p50_ms": 1.751039981842041, "p95_ms": 1.804227215051651, "p99_ms": 1.8082921385765076, "mean_ms": 1.7573047478993733, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_b0c50d31de479cf8_sdpa_flash.json", "configuration": "704x1280_tp2_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp2_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp2_sp2_v1/captures/qkv_rank0_b0c50d31de479cf8.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 0, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_efficient", "status": "measured", "q_shape": [10560, 10, 128], "k_shape": [66880, 10, 128], "q_len": 10560, "k_len": 10560, "reference_error": {"max_abs": 0.03125, "rmse": 0.0007052555423885271, "relative_l2": 0.0012835379136443405, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.0072374631331095785, "rmse": 0.0006850203216445965, "relative_l2": 0.001898586574032344, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 3.34171199798584, "p95_ms": 3.357039952278137, "p99_ms": 3.3621056962013243, "mean_ms": 3.341416533788045, "n": 300}, {"repeat": 1, "p50_ms": 3.3409759998321533, "p95_ms": 3.353249490261078, "p99_ms": 3.3563865995407105, "mean_ms": 3.3400628209114074, "n": 300}, {"repeat": 2, "p50_ms": 3.3394880294799805, "p95_ms": 3.351871967315674, "p99_ms": 3.3559705591201783, "mean_ms": 3.339358503818512, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_b0c50d31de479cf8_sdpa_efficient.json", "configuration": "704x1280_tp2_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp2_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp2_sp2_v1/captures/qkv_rank0_b0c50d31de479cf8.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 0, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_cudnn", "status": "measured", "q_shape": [10560, 10, 128], "k_shape": [66880, 10, 128], "q_len": 10560, "k_len": 10560, "reference_error": {"max_abs": 0.03125, "rmse": 0.000703363823738897, "relative_l2": 0.0012800950585899512, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.0072374631331095785, "rmse": 0.0006857515197147507, "relative_l2": 0.0019006131457924638, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 1.0772159695625305, "p95_ms": 1.0942095994949341, "p99_ms": 1.096768661737442, "mean_ms": 1.0668951477607092, "n": 300}, {"repeat": 1, "p50_ms": 1.0755360126495361, "p95_ms": 1.084740787744522, "p99_ms": 1.093728266954422, "mean_ms": 1.0693179710706076, "n": 300}, {"repeat": 2, "p50_ms": 1.0940800309181213, "p95_ms": 1.1820832312107086, "p99_ms": 1.1851843214035034, "mean_ms": 1.1082849085330964, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_b0c50d31de479cf8_sdpa_cudnn.json", "configuration": "704x1280_tp2_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp2_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp2_sp2_v1/captures/qkv_rank0_b0c50d31de479cf8.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 0, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "native_dense", "status": "measured", "q_shape": [10560, 10, 128], "k_shape": [66880, 10, 128], "q_len": 10560, "k_len": 10560, "reference_error": {"max_abs": 0.03125, "rmse": 0.0004366424122894928, "relative_l2": 0.0007946723665305643, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.008999835780528098, "rmse": 0.0006931785325494421, "relative_l2": 0.001921197683809172, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.9015839993953705, "p95_ms": 0.9160352140665055, "p99_ms": 0.9167062389850616, "mean_ms": 0.8826253859202067, "n": 300}, {"repeat": 1, "p50_ms": 0.9058400094509125, "p95_ms": 0.9155264109373092, "p99_ms": 0.9169670408964157, "mean_ms": 0.8995394146442414, "n": 300}, {"repeat": 2, "p50_ms": 0.9115200042724609, "p95_ms": 0.9225328266620636, "p99_ms": 0.9237446624040604, "mean_ms": 0.908078505396843, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_b0c50d31de479cf8_native_dense.json", "configuration": "704x1280_tp2_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp2_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp2_sp2_v1/captures/qkv_rank0_b37e08eb3a08f049.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 0, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "vllm_fa2", "status": "measured", "q_shape": [10560, 10, 128], "k_shape": [512, 10, 128], "q_len": 10560, "k_len": 512, "reference_error": {"max_abs": 0.0009765625, "rmse": 1.860271087001333e-06, "relative_l2": 0.000306188413259094, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.0004675048160963924, "rmse": 1.8338522143917177e-05, "relative_l2": 0.0017526865015281977, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.09702400118112564, "p95_ms": 0.09766559861600399, "p99_ms": 0.0979200005531311, "mean_ms": 0.09717130668461323, "n": 300}, {"repeat": 1, "p50_ms": 0.09705600142478943, "p95_ms": 0.09769759885966778, "p99_ms": 0.09811264201998711, "mean_ms": 0.0973459201057752, "n": 300}, {"repeat": 2, "p50_ms": 0.0966079980134964, "p95_ms": 0.09743999689817429, "p99_ms": 0.09779295958578586, "mean_ms": 0.09688757345080376, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_b37e08eb3a08f049_vllm_fa2.json", "configuration": "704x1280_tp2_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp2_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp2_sp2_v1/captures/qkv_rank0_b37e08eb3a08f049.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 0, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "vllm_fa3", "status": "measured", "q_shape": [10560, 10, 128], "k_shape": [512, 10, 128], "q_len": 10560, "k_len": 512, "reference_error": {"max_abs": 0.0009765625, "rmse": 7.443741017377796e-07, "relative_l2": 0.00012251909233812155, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.0004675048160963924, "rmse": 1.8353640117122423e-05, "relative_l2": 0.0017541313871825245, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.08051200211048126, "p95_ms": 0.08156800270080566, "p99_ms": 0.08188863769173622, "mean_ms": 0.08068704009056091, "n": 300}, {"repeat": 1, "p50_ms": 0.08044800162315369, "p95_ms": 0.08163360320031643, "p99_ms": 0.08246496208012104, "mean_ms": 0.08095978659888109, "n": 300}, {"repeat": 2, "p50_ms": 0.08022399991750717, "p95_ms": 0.08144320175051689, "p99_ms": 0.08204959891736507, "mean_ms": 0.08069472009936969, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_b37e08eb3a08f049_vllm_fa3.json", "configuration": "704x1280_tp2_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp2_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp2_sp2_v1/captures/qkv_rank0_b37e08eb3a08f049.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 0, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_math", "status": "measured", "q_shape": [10560, 10, 128], "k_shape": [512, 10, 128], "q_len": 10560, "k_len": 512, "reference_error": {"max_abs": 0.001953125, "rmse": 1.2858232644122518e-05, "relative_l2": 0.002116380713612272, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.0004675048160963924, "rmse": 1.6528679360779944e-05, "relative_l2": 0.0015797125295254962, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 1.269711971282959, "p95_ms": 1.27401762008667, "p99_ms": 1.2758723127841949, "mean_ms": 1.270085759560267, "n": 300}, {"repeat": 1, "p50_ms": 1.269872009754181, "p95_ms": 1.2743696451187134, "p99_ms": 1.2768003559112548, "mean_ms": 1.27056404987971, "n": 300}, {"repeat": 2, "p50_ms": 1.269919991493225, "p95_ms": 1.274531227350235, "p99_ms": 1.2758464431762695, "mean_ms": 1.2703037905693053, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_b37e08eb3a08f049_sdpa_math.json", "configuration": "704x1280_tp2_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp2_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp2_sp2_v1/captures/qkv_rank0_b37e08eb3a08f049.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 0, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_flash", "status": "measured", "q_shape": [10560, 10, 128], "k_shape": [512, 10, 128], "q_len": 10560, "k_len": 512, "reference_error": {"max_abs": 0.0009765625, "rmse": 1.8602129110551138e-06, "relative_l2": 0.00030617883788011446, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.0004675048160963924, "rmse": 1.8338522143917177e-05, "relative_l2": 0.0017526865015281977, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.10313600301742554, "p95_ms": 0.10396960191428661, "p99_ms": 0.10448095835745334, "mean_ms": 0.10330016002058982, "n": 300}, {"repeat": 1, "p50_ms": 0.10313600301742554, "p95_ms": 0.10400000214576721, "p99_ms": 0.10464095957577228, "mean_ms": 0.10346890670557816, "n": 300}, {"repeat": 2, "p50_ms": 0.10284800082445145, "p95_ms": 0.10374400019645691, "p99_ms": 0.10422463640570641, "mean_ms": 0.1031358935435613, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_b37e08eb3a08f049_sdpa_flash.json", "configuration": "704x1280_tp2_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp2_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp2_sp2_v1/captures/qkv_rank0_b37e08eb3a08f049.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 0, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_efficient", "status": "measured", "q_shape": [10560, 10, 128], "k_shape": [512, 10, 128], "q_len": 10560, "k_len": 512, "reference_error": {"max_abs": 0.0009765625, "rmse": 3.2689188609252166e-06, "relative_l2": 0.0005380425928743683, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.0004675048160963924, "rmse": 1.850826879180923e-05, "relative_l2": 0.0017689098730793644, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.17951999604701996, "p95_ms": 0.18032479509711266, "p99_ms": 0.18112255334854127, "mean_ms": 0.1796145066122214, "n": 300}, {"repeat": 1, "p50_ms": 0.17983999848365784, "p95_ms": 0.18057759702205659, "p99_ms": 0.1816355173289776, "mean_ms": 0.18005546659231186, "n": 300}, {"repeat": 2, "p50_ms": 0.17931199818849564, "p95_ms": 0.18016000092029572, "p99_ms": 0.18067232474684716, "mean_ms": 0.17953567961851755, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_b37e08eb3a08f049_sdpa_efficient.json", "configuration": "704x1280_tp2_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp2_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp2_sp2_v1/captures/qkv_rank0_b37e08eb3a08f049.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 0, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_cudnn", "status": "measured", "q_shape": [10560, 10, 128], "k_shape": [512, 10, 128], "q_len": 10560, "k_len": 512, "reference_error": {"max_abs": 0.0009765625, "rmse": 3.2484241686929933e-06, "relative_l2": 0.0005346693010253113, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.0004675048160963924, "rmse": 1.850827778128504e-05, "relative_l2": 0.0017689107322398127, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.05955199897289276, "p95_ms": 0.060257600620388985, "p99_ms": 0.0608771215006709, "mean_ms": 0.05977685336023569, "n": 300}, {"repeat": 1, "p50_ms": 0.05967999994754791, "p95_ms": 0.06051359884440899, "p99_ms": 0.06092192173004148, "mean_ms": 0.06018517337739468, "n": 300}, {"repeat": 2, "p50_ms": 0.05961599946022034, "p95_ms": 0.06039040144532919, "p99_ms": 0.0609327982738614, "mean_ms": 0.06003018662333488, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_b37e08eb3a08f049_sdpa_cudnn.json", "configuration": "704x1280_tp2_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp2_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp2_sp2_v1/captures/qkv_rank0_b37e08eb3a08f049.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 0, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "native_dense", "status": "measured", "q_shape": [10560, 10, 128], "k_shape": [512, 10, 128], "q_len": 10560, "k_len": 512, "reference_error": {"max_abs": 0.0, "rmse": 0.0, "relative_l2": 0.0, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.0004675048160963924, "rmse": 1.8355957684623024e-05, "relative_l2": 0.0017543528864529022, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.12009599804878235, "p95_ms": 0.1270144060254097, "p99_ms": 0.1356761573255062, "mean_ms": 0.1212889600545168, "n": 300}, {"repeat": 1, "p50_ms": 0.12007999792695045, "p95_ms": 0.13126080036163332, "p99_ms": 0.1556495983898637, "mean_ms": 0.12265642662843068, "n": 300}, {"repeat": 2, "p50_ms": 0.12080000340938568, "p95_ms": 0.13047840222716334, "p99_ms": 0.16557151317596425, "mean_ms": 0.12424064007898172, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_b37e08eb3a08f049_native_dense.json", "configuration": "704x1280_tp2_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp2_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp2_sp2_v1/captures/qkv_rank0_be6bbd23e5a64d4b.pt", "context": {"phase": "commit", "step": null, "start_frame": 9, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "vllm_fa2", "status": "measured", "q_shape": [10560, 10, 128], "k_shape": [512, 10, 128], "q_len": 10560, "k_len": 512, "reference_error": {"max_abs": 0.000244140625, "rmse": 1.1910201838863736e-06, "relative_l2": 0.0006735564748317143, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.00012640277610309614, "rmse": 4.8645140986324285e-06, "relative_l2": 0.0017539323332368927, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.097120001912117, "p95_ms": 0.09769759885966778, "p99_ms": 0.09801696129143238, "mean_ms": 0.09722986658414205, "n": 300}, {"repeat": 1, "p50_ms": 0.09689600020647049, "p95_ms": 0.09756959788501263, "p99_ms": 0.09776255935430526, "mean_ms": 0.09768266667922337, "n": 300}, {"repeat": 2, "p50_ms": 0.09663999825716019, "p95_ms": 0.09756959788501263, "p99_ms": 0.09795328080654145, "mean_ms": 0.09719957319398721, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_be6bbd23e5a64d4b_vllm_fa2.json", "configuration": "704x1280_tp2_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp2_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp2_sp2_v1/captures/qkv_rank0_be6bbd23e5a64d4b.pt", "context": {"phase": "commit", "step": null, "start_frame": 9, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "vllm_fa3", "status": "measured", "q_shape": [10560, 10, 128], "k_shape": [512, 10, 128], "q_len": 10560, "k_len": 512, "reference_error": {"max_abs": 0.0001220703125, "rmse": 4.5258658528098746e-07, "relative_l2": 0.0002559508470656344, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.00012640277610309614, "rmse": 4.8949696271402184e-06, "relative_l2": 0.0017649132729756936, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.08126400038599968, "p95_ms": 0.08233760111033916, "p99_ms": 0.08288095779716968, "mean_ms": 0.08137824013829231, "n": 300}, {"repeat": 1, "p50_ms": 0.08118399977684021, "p95_ms": 0.08236800134181976, "p99_ms": 0.08309887900948519, "mean_ms": 0.08226112008094788, "n": 300}, {"repeat": 2, "p50_ms": 0.08116799965500832, "p95_ms": 0.0822720006108284, "p99_ms": 0.08452992267906574, "mean_ms": 0.08255295976996421, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_be6bbd23e5a64d4b_vllm_fa3.json", "configuration": "704x1280_tp2_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp2_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp2_sp2_v1/captures/qkv_rank0_be6bbd23e5a64d4b.pt", "context": {"phase": "commit", "step": null, "start_frame": 9, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_math", "status": "measured", "q_shape": [10560, 10, 128], "k_shape": [512, 10, 128], "q_len": 10560, "k_len": 512, "reference_error": {"max_abs": 0.000244140625, "rmse": 2.5201906286873472e-06, "relative_l2": 0.0014252409310340592, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.00011773784889690386, "rmse": 4.642962646072207e-06, "relative_l2": 0.001674050509843634, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 1.2695840001106262, "p95_ms": 1.2741472482681275, "p99_ms": 1.2751395142078399, "mean_ms": 1.2698830938339234, "n": 300}, {"repeat": 1, "p50_ms": 1.269327998161316, "p95_ms": 1.2741135895252227, "p99_ms": 1.275488978624344, "mean_ms": 1.2699010157585144, "n": 300}, {"repeat": 2, "p50_ms": 1.2693760395050049, "p95_ms": 1.2738879919052124, "p99_ms": 1.2754649126529694, "mean_ms": 1.270048321088155, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_be6bbd23e5a64d4b_sdpa_math.json", "configuration": "704x1280_tp2_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp2_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp2_sp2_v1/captures/qkv_rank0_be6bbd23e5a64d4b.pt", "context": {"phase": "commit", "step": null, "start_frame": 9, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_flash", "status": "measured", "q_shape": [10560, 10, 128], "k_shape": [512, 10, 128], "q_len": 10560, "k_len": 512, "reference_error": {"max_abs": 0.000244140625, "rmse": 1.1921109620054023e-06, "relative_l2": 0.0006741733415100598, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.00012640277610309614, "rmse": 4.8645140986324285e-06, "relative_l2": 0.0017539323332368927, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.10288000106811523, "p95_ms": 0.10390400141477585, "p99_ms": 0.10454399883747101, "mean_ms": 0.10306912017365297, "n": 300}, {"repeat": 1, "p50_ms": 0.1032319962978363, "p95_ms": 0.10416000336408615, "p99_ms": 0.10467551983892917, "mean_ms": 0.10350954656799634, "n": 300}, {"repeat": 2, "p50_ms": 0.1032319962978363, "p95_ms": 0.10422559641301632, "p99_ms": 0.10457791909575462, "mean_ms": 0.10356608010828496, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_be6bbd23e5a64d4b_sdpa_flash.json", "configuration": "704x1280_tp2_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp2_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp2_sp2_v1/captures/qkv_rank0_be6bbd23e5a64d4b.pt", "context": {"phase": "commit", "step": null, "start_frame": 9, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_efficient", "status": "measured", "q_shape": [10560, 10, 128], "k_shape": [512, 10, 128], "q_len": 10560, "k_len": 512, "reference_error": {"max_abs": 0.000244140625, "rmse": 2.0908035499092823e-06, "relative_l2": 0.0011824100780956064, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.00012640277610309614, "rmse": 5.380012688015438e-06, "relative_l2": 0.0019397987168724244, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.17919999361038208, "p95_ms": 0.17987200617790222, "p99_ms": 0.1802259214222431, "mean_ms": 0.17931242600083352, "n": 300}, {"repeat": 1, "p50_ms": 0.17936000227928162, "p95_ms": 0.1800976000726223, "p99_ms": 0.18044831573963166, "mean_ms": 0.17959733307361603, "n": 300}, {"repeat": 2, "p50_ms": 0.1796800047159195, "p95_ms": 0.18032159507274628, "p99_ms": 0.18077023819088936, "mean_ms": 0.17996458624800046, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_be6bbd23e5a64d4b_sdpa_efficient.json", "configuration": "704x1280_tp2_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp2_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp2_sp2_v1/captures/qkv_rank0_be6bbd23e5a64d4b.pt", "context": {"phase": "commit", "step": null, "start_frame": 9, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_cudnn", "status": "measured", "q_shape": [10560, 10, 128], "k_shape": [512, 10, 128], "q_len": 10560, "k_len": 512, "reference_error": {"max_abs": 0.000244140625, "rmse": 2.0908003771529872e-06, "relative_l2": 0.001182408283809855, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.00012640277610309614, "rmse": 5.380203378410327e-06, "relative_l2": 0.0019398674715398903, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.05961599946022034, "p95_ms": 0.06035360135138035, "p99_ms": 0.06093503829091787, "mean_ms": 0.059877226563791436, "n": 300}, {"repeat": 1, "p50_ms": 0.059647999703884125, "p95_ms": 0.060388801433146004, "p99_ms": 0.06093407828360795, "mean_ms": 0.06010527998209, "n": 300}, {"repeat": 2, "p50_ms": 0.05967999994754791, "p95_ms": 0.0604815986007452, "p99_ms": 0.0607731207832694, "mean_ms": 0.060085973367094994, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_be6bbd23e5a64d4b_sdpa_cudnn.json", "configuration": "704x1280_tp2_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp2_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp2_sp2_v1/captures/qkv_rank0_be6bbd23e5a64d4b.pt", "context": {"phase": "commit", "step": null, "start_frame": 9, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "native_dense", "status": "measured", "q_shape": [10560, 10, 128], "k_shape": [512, 10, 128], "q_len": 10560, "k_len": 512, "reference_error": {"max_abs": 0.0, "rmse": 0.0, "relative_l2": 0.0, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.00012640277610309614, "rmse": 4.907322717693605e-06, "relative_l2": 0.0017693672604650258, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.12014399841427803, "p95_ms": 0.12663360312581062, "p99_ms": 0.13978303775191292, "mean_ms": 0.1212834133952856, "n": 300}, {"repeat": 1, "p50_ms": 0.11929599940776825, "p95_ms": 0.12577119618654253, "p99_ms": 0.14334208279848087, "mean_ms": 0.12084629322091739, "n": 300}, {"repeat": 2, "p50_ms": 0.11884800344705582, "p95_ms": 0.12573439627885818, "p99_ms": 0.1320857669413089, "mean_ms": 0.12033493342498938, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_be6bbd23e5a64d4b_native_dense.json", "configuration": "704x1280_tp2_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp2_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp2_sp2_v1/captures/qkv_rank0_c0afb7cc208e55f9.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 27, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "vllm_fa2", "status": "measured", "q_shape": [10560, 10, 128], "k_shape": [66880, 10, 128], "q_len": 10560, "k_len": 63360, "reference_error": {"max_abs": 0.03125, "rmse": 0.0004949989170693232, "relative_l2": 0.0009491279698312842, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.009407552258297702, "rmse": 0.0007486997547117268, "relative_l2": 0.0019289542681111352, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 9.683135986328125, "p95_ms": 10.158057308197021, "p99_ms": 10.80813144683838, "mean_ms": 9.776282126108805, "n": 300}, {"repeat": 1, "p50_ms": 9.815839767456055, "p95_ms": 10.074561357498169, "p99_ms": 10.082510089874267, "mean_ms": 9.810398918787639, "n": 300}, {"repeat": 2, "p50_ms": 9.827999591827393, "p95_ms": 9.952769899368286, "p99_ms": 10.081941108703614, "mean_ms": 9.818196252187093, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_c0afb7cc208e55f9_vllm_fa2.json", "configuration": "704x1280_tp2_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp2_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp2_sp2_v1/captures/qkv_rank0_c0afb7cc208e55f9.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 27, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "vllm_fa3", "status": "measured", "q_shape": [10560, 10, 128], "k_shape": [66880, 10, 128], "q_len": 10560, "k_len": 63360, "reference_error": {"max_abs": 0.0, "rmse": 0.0, "relative_l2": 0.0, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.009407552258297702, "rmse": 0.0007569944080928546, "relative_l2": 0.001950324659835371, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 5.3440001010894775, "p95_ms": 5.682945489883423, "p99_ms": 5.801602540016174, "mean_ms": 5.367571310997009, "n": 300}, {"repeat": 1, "p50_ms": 5.368512153625488, "p95_ms": 5.681926488876343, "p99_ms": 5.754038963317871, "mean_ms": 5.4008695507049564, "n": 300}, {"repeat": 2, "p50_ms": 5.399264097213745, "p95_ms": 5.457475066184998, "p99_ms": 5.464322099685669, "mean_ms": 5.379622505505879, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_c0afb7cc208e55f9_vllm_fa3.json", "configuration": "704x1280_tp2_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp2_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp2_sp2_v1/captures/qkv_rank0_c0afb7cc208e55f9.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 27, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_math", "status": "measured", "q_shape": [10560, 10, 128], "k_shape": [66880, 10, 128], "q_len": 10560, "k_len": 63360, "reference_error": {"max_abs": 0.03125, "rmse": 0.0009080928822004227, "relative_l2": 0.0017412085642612035, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.0070496019520809305, "rmse": 0.0006435143540868139, "relative_l2": 0.0016579540090599918, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 147.61131286621094, "p95_ms": 148.27221145629883, "p99_ms": 148.50404739379883, "mean_ms": 147.60591471354167, "n": 300}, {"repeat": 1, "p50_ms": 147.7382583618164, "p95_ms": 148.43618392944336, "p99_ms": 148.75604782104492, "mean_ms": 147.73871154785155, "n": 300}, {"repeat": 2, "p50_ms": 147.76192474365234, "p95_ms": 148.58675689697264, "p99_ms": 148.84886825561523, "mean_ms": 147.80279108683268, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_c0afb7cc208e55f9_sdpa_math.json", "configuration": "704x1280_tp2_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp2_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp2_sp2_v1/captures/qkv_rank0_c0afb7cc208e55f9.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 27, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_flash", "status": "measured", "q_shape": [10560, 10, 128], "k_shape": [66880, 10, 128], "q_len": 10560, "k_len": 63360, "reference_error": {"max_abs": 0.03125, "rmse": 0.0004954490514520526, "relative_l2": 0.0009499910729979751, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.009407552258297702, "rmse": 0.0007487217021014184, "relative_l2": 0.0019290108134896436, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 10.488048076629639, "p95_ms": 10.856487655639649, "p99_ms": 11.09144317626953, "mean_ms": 10.525528834660848, "n": 300}, {"repeat": 1, "p50_ms": 10.512896060943604, "p95_ms": 10.76735062599182, "p99_ms": 10.784484891891479, "mean_ms": 10.517700153986613, "n": 300}, {"repeat": 2, "p50_ms": 10.509088039398193, "p95_ms": 10.689419031143188, "p99_ms": 10.779234266281128, "mean_ms": 10.503943554560344, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_c0afb7cc208e55f9_sdpa_flash.json", "configuration": "704x1280_tp2_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp2_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp2_sp2_v1/captures/qkv_rank0_c0afb7cc208e55f9.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 27, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_efficient", "status": "measured", "q_shape": [10560, 10, 128], "k_shape": [66880, 10, 128], "q_len": 10560, "k_len": 63360, "reference_error": {"max_abs": 0.03125, "rmse": 0.0007812446179382344, "relative_l2": 0.001497985334100213, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.00931566299183828, "rmse": 0.0007445598834358421, "relative_l2": 0.0019182882804214191, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 19.85979175567627, "p95_ms": 19.909288311004637, "p99_ms": 19.917734394073488, "mean_ms": 19.8633629989624, "n": 300}, {"repeat": 1, "p50_ms": 19.85979175567627, "p95_ms": 19.906097412109375, "p99_ms": 19.919599857330322, "mean_ms": 19.865435059865316, "n": 300}, {"repeat": 2, "p50_ms": 19.86444854736328, "p95_ms": 19.905788612365722, "p99_ms": 19.91442335128784, "mean_ms": 19.865338433583577, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_c0afb7cc208e55f9_sdpa_efficient.json", "configuration": "704x1280_tp2_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp2_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp2_sp2_v1/captures/qkv_rank0_c0afb7cc208e55f9.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 27, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_cudnn", "status": "measured", "q_shape": [10560, 10, 128], "k_shape": [66880, 10, 128], "q_len": 10560, "k_len": 63360, "reference_error": {"max_abs": 0.03125, "rmse": 0.0007805740620818614, "relative_l2": 0.0014966995872093187, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.00931566299183828, "rmse": 0.000744271165335079, "relative_l2": 0.0019175444254792436, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 5.889247894287109, "p95_ms": 6.188627099990844, "p99_ms": 6.495493607521057, "mean_ms": 5.968876587549845, "n": 300}, {"repeat": 1, "p50_ms": 5.89630389213562, "p95_ms": 6.122105455398559, "p99_ms": 6.247523369789123, "mean_ms": 5.925775561332703, "n": 300}, {"repeat": 2, "p50_ms": 5.9443039894104, "p95_ms": 6.050052738189697, "p99_ms": 6.12051552772522, "mean_ms": 5.94433909257253, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_c0afb7cc208e55f9_sdpa_cudnn.json", "configuration": "704x1280_tp2_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp2_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp2_sp2_v1/captures/qkv_rank0_c0afb7cc208e55f9.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 27, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "native_dense", "status": "measured", "q_shape": [10560, 10, 128], "k_shape": [66880, 10, 128], "q_len": 10560, "k_len": 63360, "reference_error": {"max_abs": 0.03125, "rmse": 0.0004744512612742699, "relative_l2": 0.0009097291870116852, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.009407552258297702, "rmse": 0.0007567407392910509, "relative_l2": 0.0019496711061045909, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 5.322687864303589, "p95_ms": 5.69541449546814, "p99_ms": 5.870669875144959, "mean_ms": 5.344372069040935, "n": 300}, {"repeat": 1, "p50_ms": 5.338335990905762, "p95_ms": 5.619673728942871, "p99_ms": 5.682916994094849, "mean_ms": 5.362212165196737, "n": 300}, {"repeat": 2, "p50_ms": 5.345328092575073, "p95_ms": 5.423729395866394, "p99_ms": 5.4377026987075805, "mean_ms": 5.345107639630635, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_c0afb7cc208e55f9_native_dense.json", "configuration": "704x1280_tp2_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp2_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp2_sp2_v1/captures/qkv_rank0_d12872a3fe20c4e4.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 9, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "vllm_fa2", "status": "measured", "q_shape": [10560, 10, 128], "k_shape": [512, 10, 128], "q_len": 10560, "k_len": 512, "reference_error": {"max_abs": 0.0001220703125, "rmse": 1.2620025939467553e-06, "relative_l2": 0.0007169621823165712, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 5.8207717269661424e-05, "rmse": 5.079302519802582e-06, "relative_l2": 0.00184295455142228, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.0968639999628067, "p95_ms": 0.09782559983432293, "p99_ms": 0.09836863651871681, "mean_ms": 0.09704597334067026, "n": 300}, {"repeat": 1, "p50_ms": 0.09670399874448776, "p95_ms": 0.09773119911551476, "p99_ms": 0.0982080027461052, "mean_ms": 0.09708042656381925, "n": 300}, {"repeat": 2, "p50_ms": 0.09679999947547913, "p95_ms": 0.09779199957847595, "p99_ms": 0.0987859196960926, "mean_ms": 0.0970949329684178, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_d12872a3fe20c4e4_vllm_fa2.json", "configuration": "704x1280_tp2_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp2_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp2_sp2_v1/captures/qkv_rank0_d12872a3fe20c4e4.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 9, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "vllm_fa3", "status": "measured", "q_shape": [10560, 10, 128], "k_shape": [512, 10, 128], "q_len": 10560, "k_len": 512, "reference_error": {"max_abs": 0.0001220703125, "rmse": 4.789217308687012e-07, "relative_l2": 0.00027208245923537346, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 5.8207717269661424e-05, "rmse": 5.1112193995495965e-06, "relative_l2": 0.0018545351490668646, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.08076799660921097, "p95_ms": 0.08188799768686295, "p99_ms": 0.08237056136131286, "mean_ms": 0.0809146666030089, "n": 300}, {"repeat": 1, "p50_ms": 0.0806720033288002, "p95_ms": 0.08189119771122932, "p99_ms": 0.08291487805545329, "mean_ms": 0.08116416017214458, "n": 300}, {"repeat": 2, "p50_ms": 0.08070400357246399, "p95_ms": 0.08201759867370129, "p99_ms": 0.08264544323086737, "mean_ms": 0.08140426695346832, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_d12872a3fe20c4e4_vllm_fa3.json", "configuration": "704x1280_tp2_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp2_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp2_sp2_v1/captures/qkv_rank0_d12872a3fe20c4e4.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 9, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_math", "status": "measured", "q_shape": [10560, 10, 128], "k_shape": [512, 10, 128], "q_len": 10560, "k_len": 512, "reference_error": {"max_abs": 0.000244140625, "rmse": 3.3198650472765848e-06, "relative_l2": 0.0018860640229336563, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 5.8207717269661424e-05, "rmse": 4.561900937039327e-06, "relative_l2": 0.0016552225551198243, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 1.2724000215530396, "p95_ms": 1.284168016910553, "p99_ms": 1.2867558252811433, "mean_ms": 1.2744109876950582, "n": 300}, {"repeat": 1, "p50_ms": 1.271727979183197, "p95_ms": 1.2760431706905364, "p99_ms": 1.2769285881519317, "mean_ms": 1.272792743841807, "n": 300}, {"repeat": 2, "p50_ms": 1.2719839811325073, "p95_ms": 1.2762240171432495, "p99_ms": 1.2776393580436707, "mean_ms": 1.2725409070650737, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_d12872a3fe20c4e4_sdpa_math.json", "configuration": "704x1280_tp2_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp2_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp2_sp2_v1/captures/qkv_rank0_d12872a3fe20c4e4.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 9, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_flash", "status": "measured", "q_shape": [10560, 10, 128], "k_shape": [512, 10, 128], "q_len": 10560, "k_len": 512, "reference_error": {"max_abs": 0.0001220703125, "rmse": 1.2621060299333675e-06, "relative_l2": 0.0007170209458175714, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 5.8207717269661424e-05, "rmse": 5.079302519802582e-06, "relative_l2": 0.00184295455142228, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.10278400033712387, "p95_ms": 0.10349119827151299, "p99_ms": 0.10406400263309479, "mean_ms": 0.10295232015351455, "n": 300}, {"repeat": 1, "p50_ms": 0.10283200070261955, "p95_ms": 0.10374400019645691, "p99_ms": 0.10406560257077217, "mean_ms": 0.10313173356155554, "n": 300}, {"repeat": 2, "p50_ms": 0.10320000350475311, "p95_ms": 0.10406560264527798, "p99_ms": 0.10464255958795547, "mean_ms": 0.10352714682618777, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_d12872a3fe20c4e4_sdpa_flash.json", "configuration": "704x1280_tp2_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp2_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp2_sp2_v1/captures/qkv_rank0_d12872a3fe20c4e4.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 9, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_efficient", "status": "measured", "q_shape": [10560, 10, 128], "k_shape": [512, 10, 128], "q_len": 10560, "k_len": 512, "reference_error": {"max_abs": 0.0001220703125, "rmse": 2.228290812209478e-06, "relative_l2": 0.0012659246908212586, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 7.192861749966833e-05, "rmse": 5.690608035435331e-06, "relative_l2": 0.002064758288835563, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.1794240027666092, "p95_ms": 0.18032159507274628, "p99_ms": 0.181536003947258, "mean_ms": 0.1796331730981668, "n": 300}, {"repeat": 1, "p50_ms": 0.1791359931230545, "p95_ms": 0.17974560484290122, "p99_ms": 0.1802249613404274, "mean_ms": 0.1793651190896829, "n": 300}, {"repeat": 2, "p50_ms": 0.17971199750900269, "p95_ms": 0.18041600286960602, "p99_ms": 0.18099232718348504, "mean_ms": 0.17994229301810263, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_d12872a3fe20c4e4_sdpa_efficient.json", "configuration": "704x1280_tp2_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp2_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp2_sp2_v1/captures/qkv_rank0_d12872a3fe20c4e4.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 9, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_cudnn", "status": "measured", "q_shape": [10560, 10, 128], "k_shape": [512, 10, 128], "q_len": 10560, "k_len": 512, "reference_error": {"max_abs": 0.0001220703125, "rmse": 2.227752693478113e-06, "relative_l2": 0.0012656189776778502, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 7.192861749966833e-05, "rmse": 5.69061612904296e-06, "relative_l2": 0.002064761225488964, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.05984000116586685, "p95_ms": 0.060736000537872314, "p99_ms": 0.061133759729564176, "mean_ms": 0.06010869335383177, "n": 300}, {"repeat": 1, "p50_ms": 0.05955199897289276, "p95_ms": 0.060577599331736565, "p99_ms": 0.06109215948730707, "mean_ms": 0.060133333367606005, "n": 300}, {"repeat": 2, "p50_ms": 0.05961599946022034, "p95_ms": 0.060579199343919754, "p99_ms": 0.07760736174881452, "mean_ms": 0.060441706627607346, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_d12872a3fe20c4e4_sdpa_cudnn.json", "configuration": "704x1280_tp2_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp2_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp2_sp2_v1/captures/qkv_rank0_d12872a3fe20c4e4.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 9, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "native_dense", "status": "measured", "q_shape": [10560, 10, 128], "k_shape": [512, 10, 128], "q_len": 10560, "k_len": 512, "reference_error": {"max_abs": 0.0, "rmse": 0.0, "relative_l2": 0.0, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 5.8207717269661424e-05, "rmse": 5.114539942324399e-06, "relative_l2": 0.0018557399620104048, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.11900799721479416, "p95_ms": 0.12683840394020082, "p99_ms": 0.13751136273145662, "mean_ms": 0.12013824028273423, "n": 300}, {"repeat": 1, "p50_ms": 0.12196799740195274, "p95_ms": 0.1691903993487358, "p99_ms": 0.17437087416648864, "mean_ms": 0.13333898638685546, "n": 300}, {"repeat": 2, "p50_ms": 0.11987199634313583, "p95_ms": 0.16521759852766993, "p99_ms": 0.1806358428299425, "mean_ms": 0.12621503981451193, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_d12872a3fe20c4e4_native_dense.json", "configuration": "704x1280_tp2_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp2_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp2_sp2_v1/captures/qkv_rank0_db45c3c5819dc1f4.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 9, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "vllm_fa2", "status": "measured", "q_shape": [10560, 10, 128], "k_shape": [66880, 10, 128], "q_len": 10560, "k_len": 42240, "reference_error": {"max_abs": 0.03125, "rmse": 0.0005011318657125259, "relative_l2": 0.0009549288809502346, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.008121707600010186, "rmse": 0.0007496158925283387, "relative_l2": 0.001916130854741869, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 6.440000057220459, "p95_ms": 6.815888190269471, "p99_ms": 7.0451150178909305, "mean_ms": 6.4922945006688435, "n": 300}, {"repeat": 1, "p50_ms": 6.4501121044158936, "p95_ms": 6.66967511177063, "p99_ms": 6.679089183807373, "mean_ms": 6.500160214106242, "n": 300}, {"repeat": 2, "p50_ms": 6.492048025131226, "p95_ms": 6.667470550537109, "p99_ms": 6.670685319900513, "mean_ms": 6.504316922823588, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_db45c3c5819dc1f4_vllm_fa2.json", "configuration": "704x1280_tp2_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp2_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp2_sp2_v1/captures/qkv_rank0_db45c3c5819dc1f4.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 9, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "vllm_fa3", "status": "measured", "q_shape": [10560, 10, 128], "k_shape": [66880, 10, 128], "q_len": 10560, "k_len": 42240, "reference_error": {"max_abs": 0.0, "rmse": 0.0, "relative_l2": 0.0, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.008121707600010186, "rmse": 0.0007550902457188012, "relative_l2": 0.0019301241240449252, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 3.5519680976867676, "p95_ms": 3.9369135379791262, "p99_ms": 4.083305106163025, "mean_ms": 3.601637864112854, "n": 300}, {"repeat": 1, "p50_ms": 3.5677119493484497, "p95_ms": 3.7281184077262877, "p99_ms": 3.7335507249832154, "mean_ms": 3.5833861351013185, "n": 300}, {"repeat": 2, "p50_ms": 3.580575942993164, "p95_ms": 3.692083239555359, "p99_ms": 3.700616116523743, "mean_ms": 3.5928801012039187, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_db45c3c5819dc1f4_vllm_fa3.json", "configuration": "704x1280_tp2_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp2_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp2_sp2_v1/captures/qkv_rank0_db45c3c5819dc1f4.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 9, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_math", "status": "measured", "q_shape": [10560, 10, 128], "k_shape": [66880, 10, 128], "q_len": 10560, "k_len": 42240, "reference_error": {"max_abs": 0.03125, "rmse": 0.0009080049541833196, "relative_l2": 0.0017302435030003597, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.007798464777282188, "rmse": 0.00065519899775415, "relative_l2": 0.0016747870851006727, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 98.13593673706055, "p95_ms": 98.73351364135742, "p99_ms": 98.98294120788574, "mean_ms": 98.15070147196452, "n": 300}, {"repeat": 1, "p50_ms": 98.38047790527344, "p95_ms": 98.95275993347168, "p99_ms": 99.11241889953614, "mean_ms": 98.3539653523763, "n": 300}, {"repeat": 2, "p50_ms": 98.42852783203125, "p95_ms": 98.96961669921875, "p99_ms": 99.18562019348144, "mean_ms": 98.42006014506022, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_db45c3c5819dc1f4_sdpa_math.json", "configuration": "704x1280_tp2_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp2_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp2_sp2_v1/captures/qkv_rank0_db45c3c5819dc1f4.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 9, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_flash", "status": "measured", "q_shape": [10560, 10, 128], "k_shape": [66880, 10, 128], "q_len": 10560, "k_len": 42240, "reference_error": {"max_abs": 0.03125, "rmse": 0.0005015828158135832, "relative_l2": 0.0009557881862645252, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.008121707600010186, "rmse": 0.0007496827275028022, "relative_l2": 0.0019163016949788783, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 6.998528003692627, "p95_ms": 7.300017666816712, "p99_ms": 7.436006641387939, "mean_ms": 7.033786126772562, "n": 300}, {"repeat": 1, "p50_ms": 7.004656076431274, "p95_ms": 7.18480658531189, "p99_ms": 7.203175597190857, "mean_ms": 7.0193652200698855, "n": 300}, {"repeat": 2, "p50_ms": 7.02457594871521, "p95_ms": 7.11342568397522, "p99_ms": 7.121228303909302, "mean_ms": 7.014650144577026, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_db45c3c5819dc1f4_sdpa_flash.json", "configuration": "704x1280_tp2_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp2_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp2_sp2_v1/captures/qkv_rank0_db45c3c5819dc1f4.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 9, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_efficient", "status": "measured", "q_shape": [10560, 10, 128], "k_shape": [66880, 10, 128], "q_len": 10560, "k_len": 42240, "reference_error": {"max_abs": 0.03125, "rmse": 0.0007808866986142236, "relative_l2": 0.0014880140583284505, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.008121707600010186, "rmse": 0.0007536886690942537, "relative_l2": 0.0019265414835988664, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 13.24516773223877, "p95_ms": 13.28265733718872, "p99_ms": 13.286466302871704, "mean_ms": 13.247462498346964, "n": 300}, {"repeat": 1, "p50_ms": 13.243792057037354, "p95_ms": 13.28067364692688, "p99_ms": 13.287042636871337, "mean_ms": 13.24550038019816, "n": 300}, {"repeat": 2, "p50_ms": 13.245872020721436, "p95_ms": 13.28550124168396, "p99_ms": 13.294513187408448, "mean_ms": 13.24798867225647, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_db45c3c5819dc1f4_sdpa_efficient.json", "configuration": "704x1280_tp2_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp2_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp2_sp2_v1/captures/qkv_rank0_db45c3c5819dc1f4.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 9, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_cudnn", "status": "measured", "q_shape": [10560, 10, 128], "k_shape": [66880, 10, 128], "q_len": 10560, "k_len": 42240, "reference_error": {"max_abs": 0.03125, "rmse": 0.0007800966987692281, "relative_l2": 0.0014865086787676043, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.008121707600010186, "rmse": 0.0007535540132164486, "relative_l2": 0.0019261972829424949, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 3.9342880249023438, "p95_ms": 4.2922782659530645, "p99_ms": 4.493706769943238, "mean_ms": 3.977435725529989, "n": 300}, {"repeat": 1, "p50_ms": 3.9578559398651123, "p95_ms": 4.082694411277771, "p99_ms": 4.104800224304199, "mean_ms": 3.959085008303324, "n": 300}, {"repeat": 2, "p50_ms": 3.9618239402770996, "p95_ms": 4.0884256839752195, "p99_ms": 4.133902487754821, "mean_ms": 3.96596821308136, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_db45c3c5819dc1f4_sdpa_cudnn.json", "configuration": "704x1280_tp2_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp2_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp2_sp2_v1/captures/qkv_rank0_db45c3c5819dc1f4.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 9, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "native_dense", "status": "measured", "q_shape": [10560, 10, 128], "k_shape": [66880, 10, 128], "q_len": 10560, "k_len": 42240, "reference_error": {"max_abs": 0.03125, "rmse": 0.00047973867665905394, "relative_l2": 0.0009141632153030459, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.008121707600010186, "rmse": 0.000762189080244546, "relative_l2": 0.0019482698117271964, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 3.573423981666565, "p95_ms": 3.7475008249282835, "p99_ms": 3.783778283596039, "mean_ms": 3.5840276304880776, "n": 300}, {"repeat": 1, "p50_ms": 3.589583992958069, "p95_ms": 3.7176257014274596, "p99_ms": 3.7435867977142334, "mean_ms": 3.571521384716034, "n": 300}, {"repeat": 2, "p50_ms": 3.571776032447815, "p95_ms": 3.6615200281143188, "p99_ms": 3.6651381278038024, "mean_ms": 3.580957446893056, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_db45c3c5819dc1f4_native_dense.json", "configuration": "704x1280_tp2_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp2_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp2_sp2_v1/captures/qkv_rank0_fcd1494e1f8885be.pt", "context": {"phase": "commit", "step": null, "start_frame": 27, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "vllm_fa2", "status": "measured", "q_shape": [10560, 10, 128], "k_shape": [512, 10, 128], "q_len": 10560, "k_len": 512, "reference_error": {"max_abs": 0.0001220703125, "rmse": 1.1894251643217163e-06, "relative_l2": 0.0007006104958924201, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 9.157510160012067e-05, "rmse": 4.641503394274589e-06, "relative_l2": 0.0017293633695280313, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.09715200215578079, "p95_ms": 0.09830400347709656, "p99_ms": 0.09897664114832878, "mean_ms": 0.09736864017943542, "n": 300}, {"repeat": 1, "p50_ms": 0.09728000313043594, "p95_ms": 0.09843199700117111, "p99_ms": 0.09881663993000984, "mean_ms": 0.09771786679824193, "n": 300}, {"repeat": 2, "p50_ms": 0.09689600020647049, "p95_ms": 0.09801600128412247, "p99_ms": 0.0983702365309, "mean_ms": 0.09723669335246087, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_fcd1494e1f8885be_vllm_fa2.json", "configuration": "704x1280_tp2_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp2_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp2_sp2_v1/captures/qkv_rank0_fcd1494e1f8885be.pt", "context": {"phase": "commit", "step": null, "start_frame": 27, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "vllm_fa3", "status": "measured", "q_shape": [10560, 10, 128], "k_shape": [512, 10, 128], "q_len": 10560, "k_len": 512, "reference_error": {"max_abs": 0.0001220703125, "rmse": 4.4491419069650046e-07, "relative_l2": 0.000262069074308855, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 9.157510160012067e-05, "rmse": 4.68971548304834e-06, "relative_l2": 0.0017473265623152226, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.08060800284147263, "p95_ms": 0.0817295964807272, "p99_ms": 0.0826252832263708, "mean_ms": 0.0807966934889555, "n": 300}, {"repeat": 1, "p50_ms": 0.08073599636554718, "p95_ms": 0.08185599744319916, "p99_ms": 0.08234272107481956, "mean_ms": 0.08124288010100524, "n": 300}, {"repeat": 2, "p50_ms": 0.08073599636554718, "p95_ms": 0.08188799768686295, "p99_ms": 0.08227328062057496, "mean_ms": 0.0811755733191967, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_fcd1494e1f8885be_vllm_fa3.json", "configuration": "704x1280_tp2_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp2_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp2_sp2_v1/captures/qkv_rank0_fcd1494e1f8885be.pt", "context": {"phase": "commit", "step": null, "start_frame": 27, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_math", "status": "measured", "q_shape": [10560, 10, 128], "k_shape": [512, 10, 128], "q_len": 10560, "k_len": 512, "reference_error": {"max_abs": 0.000244140625, "rmse": 2.4444670719230014e-06, "relative_l2": 0.0014398714091687368, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 9.157510160012067e-05, "rmse": 4.424818279924814e-06, "relative_l2": 0.00164862933409875, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 1.2696800231933594, "p95_ms": 1.2740192234516143, "p99_ms": 1.275169222354889, "mean_ms": 1.2702145099639892, "n": 300}, {"repeat": 1, "p50_ms": 1.2698240280151367, "p95_ms": 1.2745296239852906, "p99_ms": 1.2762323367595672, "mean_ms": 1.2707181851069131, "n": 300}, {"repeat": 2, "p50_ms": 1.270143985748291, "p95_ms": 1.2741119861602783, "p99_ms": 1.275969556570053, "mean_ms": 1.2706319975852967, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_fcd1494e1f8885be_sdpa_math.json", "configuration": "704x1280_tp2_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp2_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp2_sp2_v1/captures/qkv_rank0_fcd1494e1f8885be.pt", "context": {"phase": "commit", "step": null, "start_frame": 27, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_flash", "status": "measured", "q_shape": [10560, 10, 128], "k_shape": [512, 10, 128], "q_len": 10560, "k_len": 512, "reference_error": {"max_abs": 0.0001220703125, "rmse": 1.1899970799806743e-06, "relative_l2": 0.0007009473730036922, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 9.157510160012067e-05, "rmse": 4.641503394274589e-06, "relative_l2": 0.0017293633695280313, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.10329599678516388, "p95_ms": 0.10412960313260555, "p99_ms": 0.10473664030432701, "mean_ms": 0.10343871980905534, "n": 300}, {"repeat": 1, "p50_ms": 0.10307200253009796, "p95_ms": 0.10419360324740409, "p99_ms": 0.1048675212264061, "mean_ms": 0.1034435201684634, "n": 300}, {"repeat": 2, "p50_ms": 0.10304000228643417, "p95_ms": 0.10381120070815086, "p99_ms": 0.10467359982430935, "mean_ms": 0.10355050680538018, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_fcd1494e1f8885be_sdpa_flash.json", "configuration": "704x1280_tp2_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp2_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp2_sp2_v1/captures/qkv_rank0_fcd1494e1f8885be.pt", "context": {"phase": "commit", "step": null, "start_frame": 27, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_efficient", "status": "measured", "q_shape": [10560, 10, 128], "k_shape": [512, 10, 128], "q_len": 10560, "k_len": 512, "reference_error": {"max_abs": 0.0001220703125, "rmse": 2.088575177406598e-06, "relative_l2": 0.0012302393918039282, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 9.157510160012067e-05, "rmse": 5.349796685022096e-06, "relative_l2": 0.001993264172317952, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.17977599799633026, "p95_ms": 0.1805167965590954, "p99_ms": 0.1810240000486374, "mean_ms": 0.1799619199335575, "n": 300}, {"repeat": 1, "p50_ms": 0.17948800325393677, "p95_ms": 0.18038559556007386, "p99_ms": 0.18147552341222764, "mean_ms": 0.17974815994501114, "n": 300}, {"repeat": 2, "p50_ms": 0.17919999361038208, "p95_ms": 0.17984159886837006, "p99_ms": 0.18041664287447928, "mean_ms": 0.1794388263920943, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_fcd1494e1f8885be_sdpa_efficient.json", "configuration": "704x1280_tp2_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp2_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp2_sp2_v1/captures/qkv_rank0_fcd1494e1f8885be.pt", "context": {"phase": "commit", "step": null, "start_frame": 27, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_cudnn", "status": "measured", "q_shape": [10560, 10, 128], "k_shape": [512, 10, 128], "q_len": 10560, "k_len": 512, "reference_error": {"max_abs": 0.0001220703125, "rmse": 2.0881607164210427e-06, "relative_l2": 0.0012299952606678736, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 9.157510160012067e-05, "rmse": 5.350092648723356e-06, "relative_l2": 0.00199337444451646, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.060127999633550644, "p95_ms": 0.06090400163084269, "p99_ms": 0.06141695823520421, "mean_ms": 0.06040768002470334, "n": 300}, {"repeat": 1, "p50_ms": 0.06015999987721443, "p95_ms": 0.06108799949288368, "p99_ms": 0.06205023936927318, "mean_ms": 0.06065354666362206, "n": 300}, {"repeat": 2, "p50_ms": 0.060127999633550644, "p95_ms": 0.06090080160647631, "p99_ms": 0.0765075207501649, "mean_ms": 0.060943466735382874, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_fcd1494e1f8885be_sdpa_cudnn.json", "configuration": "704x1280_tp2_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp2_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp2_sp2_v1/captures/qkv_rank0_fcd1494e1f8885be.pt", "context": {"phase": "commit", "step": null, "start_frame": 27, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "native_dense", "status": "measured", "q_shape": [10560, 10, 128], "k_shape": [512, 10, 128], "q_len": 10560, "k_len": 512, "reference_error": {"max_abs": 0.0, "rmse": 0.0, "relative_l2": 0.0, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 9.157510160012067e-05, "rmse": 4.7052763804774716e-06, "relative_l2": 0.0017531243488781104, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.12054400146007538, "p95_ms": 0.15691680237650873, "p99_ms": 0.16855583444237707, "mean_ms": 0.12406965340177219, "n": 300}, {"repeat": 1, "p50_ms": 0.11929599940776825, "p95_ms": 0.12515520676970482, "p99_ms": 0.1310403192043304, "mean_ms": 0.12053557336330414, "n": 300}, {"repeat": 2, "p50_ms": 0.11913599818944931, "p95_ms": 0.12487199753522873, "p99_ms": 0.1341321632266039, "mean_ms": 0.12138101321955522, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_fcd1494e1f8885be_native_dense.json", "configuration": "704x1280_tp2_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp2_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp4_sp1_v1/captures/qkv_rank0_1ef9f10991fa086e.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 27, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "vllm_fa2", "status": "measured", "q_shape": [10560, 10, 128], "k_shape": [512, 10, 128], "q_len": 10560, "k_len": 512, "reference_error": {"max_abs": 0.0001220703125, "rmse": 1.2598059453441608e-06, "relative_l2": 0.0007188998687351333, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.00011838995432788407, "rmse": 5.501095751943986e-06, "relative_l2": 0.0019164315194519484, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.09571199864149094, "p95_ms": 0.09673599898815155, "p99_ms": 0.09743999689817429, "mean_ms": 0.09592522641023, "n": 300}, {"repeat": 1, "p50_ms": 0.09558399766683578, "p95_ms": 0.09645119681954384, "p99_ms": 0.0968639999628067, "mean_ms": 0.09599551975727082, "n": 300}, {"repeat": 2, "p50_ms": 0.09564799815416336, "p95_ms": 0.09644799679517746, "p99_ms": 0.09673631899058818, "mean_ms": 0.09597429322699706, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_1ef9f10991fa086e_vllm_fa2.json", "configuration": "704x1280_tp4_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp4_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp4_sp1_v1/captures/qkv_rank0_1ef9f10991fa086e.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 27, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "vllm_fa3", "status": "measured", "q_shape": [10560, 10, 128], "k_shape": [512, 10, 128], "q_len": 10560, "k_len": 512, "reference_error": {"max_abs": 0.0001220703125, "rmse": 4.781927194246012e-07, "relative_l2": 0.00027287748918388257, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.00011838995432788407, "rmse": 5.538838446509181e-06, "relative_l2": 0.0019295800434470502, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.0801599994301796, "p95_ms": 0.08137760125100613, "p99_ms": 0.08660512290894976, "mean_ms": 0.08054314643144607, "n": 300}, {"repeat": 1, "p50_ms": 0.08006399869918823, "p95_ms": 0.08102559857070446, "p99_ms": 0.08141024149954319, "mean_ms": 0.08057013337810834, "n": 300}, {"repeat": 2, "p50_ms": 0.08019199967384338, "p95_ms": 0.0811551995575428, "p99_ms": 0.10470335960388177, "mean_ms": 0.08101674685875575, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_1ef9f10991fa086e_vllm_fa3.json", "configuration": "704x1280_tp4_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp4_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp4_sp1_v1/captures/qkv_rank0_1ef9f10991fa086e.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 27, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_math", "status": "measured", "q_shape": [10560, 10, 128], "k_shape": [512, 10, 128], "q_len": 10560, "k_len": 512, "reference_error": {"max_abs": 0.000244140625, "rmse": 3.37511428641406e-06, "relative_l2": 0.0019259864794544143, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.00011838995432788407, "rmse": 4.816439484971386e-06, "relative_l2": 0.0016779159746983547, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 1.2644959688186646, "p95_ms": 1.2686480402946472, "p99_ms": 1.2710723495483398, "mean_ms": 1.2648873567581176, "n": 300}, {"repeat": 1, "p50_ms": 1.2649600505828857, "p95_ms": 1.2696016311645508, "p99_ms": 1.2710742700099944, "mean_ms": 1.2658026639620463, "n": 300}, {"repeat": 2, "p50_ms": 1.2657599449157715, "p95_ms": 1.270756757259369, "p99_ms": 1.2721622931957244, "mean_ms": 1.2673813327153525, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_1ef9f10991fa086e_sdpa_math.json", "configuration": "704x1280_tp4_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp4_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp4_sp1_v1/captures/qkv_rank0_1ef9f10991fa086e.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 27, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_flash", "status": "measured", "q_shape": [10560, 10, 128], "k_shape": [512, 10, 128], "q_len": 10560, "k_len": 512, "reference_error": {"max_abs": 0.0001220703125, "rmse": 1.260067318867212e-06, "relative_l2": 0.0007190490197945538, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.00011838995432788407, "rmse": 5.501095751943986e-06, "relative_l2": 0.0019164315194519484, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.10262399911880493, "p95_ms": 0.10342399775981903, "p99_ms": 0.10377632044255733, "mean_ms": 0.1028676263988018, "n": 300}, {"repeat": 1, "p50_ms": 0.10259199887514114, "p95_ms": 0.1034287977963686, "p99_ms": 0.1037151999771595, "mean_ms": 0.102867519830664, "n": 300}, {"repeat": 2, "p50_ms": 0.10263999924063683, "p95_ms": 0.10336159728467464, "p99_ms": 0.1037126399576664, "mean_ms": 0.10289578641454379, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_1ef9f10991fa086e_sdpa_flash.json", "configuration": "704x1280_tp4_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp4_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp4_sp1_v1/captures/qkv_rank0_1ef9f10991fa086e.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 27, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_efficient", "status": "measured", "q_shape": [10560, 10, 128], "k_shape": [512, 10, 128], "q_len": 10560, "k_len": 512, "reference_error": {"max_abs": 0.0001220703125, "rmse": 2.2093836425141034e-06, "relative_l2": 0.0012607700546730092, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.00011838995432788407, "rmse": 6.04963498237457e-06, "relative_l2": 0.0021075276061691445, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.17894400656223297, "p95_ms": 0.1798096053302288, "p99_ms": 0.18032127469778061, "mean_ms": 0.17909418642520905, "n": 300}, {"repeat": 1, "p50_ms": 0.17878399789333344, "p95_ms": 0.1794240027666092, "p99_ms": 0.17996863946318625, "mean_ms": 0.17909898668527602, "n": 300}, {"repeat": 2, "p50_ms": 0.17905599623918533, "p95_ms": 0.17974560484290122, "p99_ms": 0.18022432133555413, "mean_ms": 0.17946719974279404, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_1ef9f10991fa086e_sdpa_efficient.json", "configuration": "704x1280_tp4_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp4_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp4_sp1_v1/captures/qkv_rank0_1ef9f10991fa086e.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 27, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_cudnn", "status": "measured", "q_shape": [10560, 10, 128], "k_shape": [512, 10, 128], "q_len": 10560, "k_len": 512, "reference_error": {"max_abs": 0.0001220703125, "rmse": 2.208640883624106e-06, "relative_l2": 0.0012603462042613685, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.00011838995432788407, "rmse": 6.048666668469126e-06, "relative_l2": 0.002107190272050126, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.0589120015501976, "p95_ms": 0.059620799496769905, "p99_ms": 0.0600364788994193, "mean_ms": 0.05913717329502106, "n": 300}, {"repeat": 1, "p50_ms": 0.05894400179386139, "p95_ms": 0.059779200702905655, "p99_ms": 0.060257280617952345, "mean_ms": 0.05941898657629887, "n": 300}, {"repeat": 2, "p50_ms": 0.05894400179386139, "p95_ms": 0.05968319997191429, "p99_ms": 0.06012863963842392, "mean_ms": 0.059397866601745286, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_1ef9f10991fa086e_sdpa_cudnn.json", "configuration": "704x1280_tp4_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp4_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp4_sp1_v1/captures/qkv_rank0_1ef9f10991fa086e.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 27, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "native_dense", "status": "measured", "q_shape": [10560, 10, 128], "k_shape": [512, 10, 128], "q_len": 10560, "k_len": 512, "reference_error": {"max_abs": 0.0, "rmse": 0.0, "relative_l2": 0.0, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.00011838995432788407, "rmse": 5.5566371510873275e-06, "relative_l2": 0.0019357806260213346, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.12344000115990639, "p95_ms": 0.13258080333471298, "p99_ms": 0.17945856034755675, "mean_ms": 0.12671733332177004, "n": 300}, {"repeat": 1, "p50_ms": 0.12358400225639343, "p95_ms": 0.12934560105204582, "p99_ms": 0.17088639706373213, "mean_ms": 0.12905471990505854, "n": 300}, {"repeat": 2, "p50_ms": 0.12372800335288048, "p95_ms": 0.13158399611711502, "p99_ms": 0.15819327726960178, "mean_ms": 0.1254803200562795, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_1ef9f10991fa086e_native_dense.json", "configuration": "704x1280_tp4_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp4_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp4_sp1_v1/captures/qkv_rank0_5608acc7409c5075.pt", "context": {"phase": "commit", "step": null, "start_frame": 0, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "vllm_fa2", "status": "measured", "q_shape": [10560, 10, 128], "k_shape": [66880, 10, 128], "q_len": 10560, "k_len": 10560, "reference_error": {"max_abs": 0.03125, "rmse": 0.00037299557480778144, "relative_l2": 0.0005059981497425545, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.008153332523755719, "rmse": 0.0007817116204317559, "relative_l2": 0.0016945918305677431, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 1.583232045173645, "p95_ms": 1.5845455467700957, "p99_ms": 1.5850581872463225, "mean_ms": 1.583436906337738, "n": 300}, {"repeat": 1, "p50_ms": 1.5838720202445984, "p95_ms": 1.6920991957187652, "p99_ms": 1.693251873254776, "mean_ms": 1.5989975468317668, "n": 300}, {"repeat": 2, "p50_ms": 1.5958720445632935, "p95_ms": 1.6086768209934235, "p99_ms": 1.6364236390590667, "mean_ms": 1.5976866126060485, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_5608acc7409c5075_vllm_fa2.json", "configuration": "704x1280_tp4_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp4_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp4_sp1_v1/captures/qkv_rank0_5608acc7409c5075.pt", "context": {"phase": "commit", "step": null, "start_frame": 0, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "vllm_fa3", "status": "measured", "q_shape": [10560, 10, 128], "k_shape": [66880, 10, 128], "q_len": 10560, "k_len": 10560, "reference_error": {"max_abs": 0.0, "rmse": 0.0, "relative_l2": 0.0, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.008153332523755719, "rmse": 0.0007828993074923871, "relative_l2": 0.001697166494596794, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.8828320205211639, "p95_ms": 0.9166767925024033, "p99_ms": 0.917376942038536, "mean_ms": 0.8773480542500813, "n": 300}, {"repeat": 1, "p50_ms": 0.8829919993877411, "p95_ms": 0.8919696122407913, "p99_ms": 0.901443223953247, "mean_ms": 0.8806887443860372, "n": 300}, {"repeat": 2, "p50_ms": 0.8894079923629761, "p95_ms": 0.8988239914178848, "p99_ms": 0.9021475344896317, "mean_ms": 0.8843867729107538, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_5608acc7409c5075_vllm_fa3.json", "configuration": "704x1280_tp4_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp4_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp4_sp1_v1/captures/qkv_rank0_5608acc7409c5075.pt", "context": {"phase": "commit", "step": null, "start_frame": 0, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_math", "status": "measured", "q_shape": [10560, 10, 128], "k_shape": [66880, 10, 128], "q_len": 10560, "k_len": 10560, "reference_error": {"max_abs": 0.0625, "rmse": 0.0006799036130907156, "relative_l2": 0.0009223433023420486, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.007745470089854045, "rmse": 0.000768698805078143, "relative_l2": 0.0016663826930615878, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 23.227983474731445, "p95_ms": 23.36819543838501, "p99_ms": 23.390838356018065, "mean_ms": 23.246447010040285, "n": 300}, {"repeat": 1, "p50_ms": 23.32196807861328, "p95_ms": 23.376748466491698, "p99_ms": 23.44213224411011, "mean_ms": 23.308248538970947, "n": 300}, {"repeat": 2, "p50_ms": 23.338319778442383, "p95_ms": 23.47911128997803, "p99_ms": 23.612308654785156, "mean_ms": 23.34317464828491, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_5608acc7409c5075_sdpa_math.json", "configuration": "704x1280_tp4_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp4_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp4_sp1_v1/captures/qkv_rank0_5608acc7409c5075.pt", "context": {"phase": "commit", "step": null, "start_frame": 0, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_flash", "status": "measured", "q_shape": [10560, 10, 128], "k_shape": [66880, 10, 128], "q_len": 10560, "k_len": 10560, "reference_error": {"max_abs": 0.03125, "rmse": 0.00037328728281715655, "relative_l2": 0.0005063938748475639, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.008153332523755719, "rmse": 0.0007816969472063796, "relative_l2": 0.0016945600219989555, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 1.7323839664459229, "p95_ms": 1.7397152483463287, "p99_ms": 1.743267480134964, "mean_ms": 1.7290498089790345, "n": 300}, {"repeat": 1, "p50_ms": 1.724128007888794, "p95_ms": 1.775265622138977, "p99_ms": 1.7770569384098054, "mean_ms": 1.7331361039479574, "n": 300}, {"repeat": 2, "p50_ms": 1.7335360050201416, "p95_ms": 1.7390416145324707, "p99_ms": 1.7463053143024445, "mean_ms": 1.7306018126010896, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_5608acc7409c5075_sdpa_flash.json", "configuration": "704x1280_tp4_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp4_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp4_sp1_v1/captures/qkv_rank0_5608acc7409c5075.pt", "context": {"phase": "commit", "step": null, "start_frame": 0, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_efficient", "status": "measured", "q_shape": [10560, 10, 128], "k_shape": [66880, 10, 128], "q_len": 10560, "k_len": 10560, "reference_error": {"max_abs": 0.0625, "rmse": 0.0005703485011508429, "relative_l2": 0.0007737230835499581, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.008124133004721035, "rmse": 0.0007763011146837528, "relative_l2": 0.0016828629543425926, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 3.33022403717041, "p95_ms": 3.357772719860077, "p99_ms": 3.3650592374801636, "mean_ms": 3.3332120482126872, "n": 300}, {"repeat": 1, "p50_ms": 3.3305920362472534, "p95_ms": 3.3441952228546143, "p99_ms": 3.350346953868866, "mean_ms": 3.332153710524241, "n": 300}, {"repeat": 2, "p50_ms": 3.3278719186782837, "p95_ms": 3.3405519843101503, "p99_ms": 3.3429802513122557, "mean_ms": 3.3293844199180604, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_5608acc7409c5075_sdpa_efficient.json", "configuration": "704x1280_tp4_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp4_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp4_sp1_v1/captures/qkv_rank0_5608acc7409c5075.pt", "context": {"phase": "commit", "step": null, "start_frame": 0, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_cudnn", "status": "measured", "q_shape": [10560, 10, 128], "k_shape": [66880, 10, 128], "q_len": 10560, "k_len": 10560, "reference_error": {"max_abs": 0.0625, "rmse": 0.0005669032343439171, "relative_l2": 0.0007690493052334934, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.008124133004721035, "rmse": 0.0007762663968103241, "relative_l2": 0.0016827876930529443, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 1.0447360277175903, "p95_ms": 1.0485135853290557, "p99_ms": 1.0490243065357208, "mean_ms": 1.034844051996867, "n": 300}, {"repeat": 1, "p50_ms": 1.0449600219726562, "p95_ms": 1.0484816312789917, "p99_ms": 1.0493468236923218, "mean_ms": 1.042398717800776, "n": 300}, {"repeat": 2, "p50_ms": 1.054960012435913, "p95_ms": 1.16672323346138, "p99_ms": 1.1674249243736268, "mean_ms": 1.071065707206726, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_5608acc7409c5075_sdpa_cudnn.json", "configuration": "704x1280_tp4_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp4_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp4_sp1_v1/captures/qkv_rank0_5608acc7409c5075.pt", "context": {"phase": "commit", "step": null, "start_frame": 0, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "native_dense", "status": "measured", "q_shape": [10560, 10, 128], "k_shape": [66880, 10, 128], "q_len": 10560, "k_len": 10560, "reference_error": {"max_abs": 0.0625, "rmse": 0.0003586157201121001, "relative_l2": 0.00048649073367379496, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.008153332523755719, "rmse": 0.0007834640293523508, "relative_l2": 0.001698390696751943, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.8614720106124878, "p95_ms": 0.9417952120304107, "p99_ms": 0.943105628490448, "mean_ms": 0.8584346669912338, "n": 300}, {"repeat": 1, "p50_ms": 0.8641439974308014, "p95_ms": 0.9006912171840668, "p99_ms": 0.9024646127223969, "mean_ms": 0.861714881459872, "n": 300}, {"repeat": 2, "p50_ms": 0.8543040156364441, "p95_ms": 0.868108794093132, "p99_ms": 0.8806083416938781, "mean_ms": 0.8527367466688156, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_5608acc7409c5075_native_dense.json", "configuration": "704x1280_tp4_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp4_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp4_sp1_v1/captures/qkv_rank0_597583b95ac01e65.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 0, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "vllm_fa2", "status": "measured", "q_shape": [10560, 10, 128], "k_shape": [66880, 10, 128], "q_len": 10560, "k_len": 10560, "reference_error": {"max_abs": 0.03125, "rmse": 0.00046040391411085, "relative_l2": 0.0008379173182688868, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.008999890977471114, "rmse": 0.000680420246030594, "relative_l2": 0.001885837090490708, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 1.609503984451294, "p95_ms": 1.6639488160610199, "p99_ms": 1.6673280000686646, "mean_ms": 1.6160801037152608, "n": 300}, {"repeat": 1, "p50_ms": 1.6135680079460144, "p95_ms": 1.7378688275814056, "p99_ms": 1.7392707312107085, "mean_ms": 1.6305924268563587, "n": 300}, {"repeat": 2, "p50_ms": 1.6358400583267212, "p95_ms": 1.6794943451881408, "p99_ms": 1.7081606829166411, "mean_ms": 1.6339054930210113, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_597583b95ac01e65_vllm_fa2.json", "configuration": "704x1280_tp4_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp4_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp4_sp1_v1/captures/qkv_rank0_597583b95ac01e65.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 0, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "vllm_fa3", "status": "measured", "q_shape": [10560, 10, 128], "k_shape": [66880, 10, 128], "q_len": 10560, "k_len": 10560, "reference_error": {"max_abs": 0.0, "rmse": 0.0, "relative_l2": 0.0, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.008999890977471114, "rmse": 0.0006941936267203683, "relative_l2": 0.001924011075344558, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.9006080031394958, "p95_ms": 0.9340479969978333, "p99_ms": 0.9369635021686554, "mean_ms": 0.8949742937088012, "n": 300}, {"repeat": 1, "p50_ms": 0.8918399810791016, "p95_ms": 0.9079088270664215, "p99_ms": 0.9141183722019196, "mean_ms": 0.8917761059602102, "n": 300}, {"repeat": 2, "p50_ms": 0.9010079801082611, "p95_ms": 0.9230224221944809, "p99_ms": 0.9245763045549392, "mean_ms": 0.8933813357353211, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_597583b95ac01e65_vllm_fa3.json", "configuration": "704x1280_tp4_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp4_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp4_sp1_v1/captures/qkv_rank0_597583b95ac01e65.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 0, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_math", "status": "measured", "q_shape": [10560, 10, 128], "k_shape": [66880, 10, 128], "q_len": 10560, "k_len": 10560, "reference_error": {"max_abs": 0.03125, "rmse": 0.0007804743148918358, "relative_l2": 0.0014204330694600887, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.0072374643840786845, "rmse": 0.0005975817622571412, "relative_l2": 0.0016562438558223708, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 23.350895881652832, "p95_ms": 23.568320751190186, "p99_ms": 23.7297359085083, "mean_ms": 23.373950926462808, "n": 300}, {"repeat": 1, "p50_ms": 23.36676788330078, "p95_ms": 23.59921875, "p99_ms": 23.647891063690185, "mean_ms": 23.39398296356201, "n": 300}, {"repeat": 2, "p50_ms": 23.373536109924316, "p95_ms": 23.65311918258667, "p99_ms": 23.748339157104493, "mean_ms": 23.418456528981526, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_597583b95ac01e65_sdpa_math.json", "configuration": "704x1280_tp4_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp4_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp4_sp1_v1/captures/qkv_rank0_597583b95ac01e65.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 0, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_flash", "status": "measured", "q_shape": [10560, 10, 128], "k_shape": [66880, 10, 128], "q_len": 10560, "k_len": 10560, "reference_error": {"max_abs": 0.03125, "rmse": 0.0004607622293019313, "relative_l2": 0.0008385694380593662, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.008999890977471114, "rmse": 0.000680350262901844, "relative_l2": 0.00188564312686209, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 1.7352319955825806, "p95_ms": 1.747555238008499, "p99_ms": 1.7492544269561767, "mean_ms": 1.7368405342102051, "n": 300}, {"repeat": 1, "p50_ms": 1.7511839866638184, "p95_ms": 1.890577632188797, "p99_ms": 1.9205456280708313, "mean_ms": 1.775888106028239, "n": 300}, {"repeat": 2, "p50_ms": 1.7474559545516968, "p95_ms": 1.763201642036438, "p99_ms": 1.7813187086582183, "mean_ms": 1.746914452314377, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_597583b95ac01e65_sdpa_flash.json", "configuration": "704x1280_tp4_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp4_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp4_sp1_v1/captures/qkv_rank0_597583b95ac01e65.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 0, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_efficient", "status": "measured", "q_shape": [10560, 10, 128], "k_shape": [66880, 10, 128], "q_len": 10560, "k_len": 10560, "reference_error": {"max_abs": 0.03125, "rmse": 0.0007051539678343196, "relative_l2": 0.0012833529507139208, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.0072374643840786845, "rmse": 0.000685025846937274, "relative_l2": 0.001898601867941825, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 3.3056960105895996, "p95_ms": 3.3709472179412843, "p99_ms": 3.3758074116706847, "mean_ms": 3.315220369497935, "n": 300}, {"repeat": 1, "p50_ms": 3.3035199642181396, "p95_ms": 3.3177967429161073, "p99_ms": 3.3206232619285583, "mean_ms": 3.3050553568204246, "n": 300}, {"repeat": 2, "p50_ms": 3.304479956626892, "p95_ms": 3.3170926928520204, "p99_ms": 3.3214189124107363, "mean_ms": 3.30567125082016, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_597583b95ac01e65_sdpa_efficient.json", "configuration": "704x1280_tp4_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp4_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp4_sp1_v1/captures/qkv_rank0_597583b95ac01e65.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 0, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_cudnn", "status": "measured", "q_shape": [10560, 10, 128], "k_shape": [66880, 10, 128], "q_len": 10560, "k_len": 10560, "reference_error": {"max_abs": 0.03125, "rmse": 0.000703256718600545, "relative_l2": 0.0012799000304816457, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.0072374643840786845, "rmse": 0.0006857608568728629, "relative_l2": 0.0019006390045592301, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 1.0607680082321167, "p95_ms": 1.0725120306015015, "p99_ms": 1.0730246746540069, "mean_ms": 1.0566915208101273, "n": 300}, {"repeat": 1, "p50_ms": 1.0595519542694092, "p95_ms": 1.0728720545768737, "p99_ms": 1.0739216113090515, "mean_ms": 1.0618801065286, "n": 300}, {"repeat": 2, "p50_ms": 1.0840480327606201, "p95_ms": 1.1370783805847169, "p99_ms": 1.143168612718582, "mean_ms": 1.0806998411814372, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_597583b95ac01e65_sdpa_cudnn.json", "configuration": "704x1280_tp4_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp4_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp4_sp1_v1/captures/qkv_rank0_597583b95ac01e65.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 0, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "native_dense", "status": "measured", "q_shape": [10560, 10, 128], "k_shape": [66880, 10, 128], "q_len": 10560, "k_len": 10560, "reference_error": {"max_abs": 0.03125, "rmse": 0.000436519356467897, "relative_l2": 0.000794448347057223, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.008999890977471114, "rmse": 0.000693217561360296, "relative_l2": 0.001921305835061798, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.8879519999027252, "p95_ms": 0.9159887999296188, "p99_ms": 0.918024656176567, "mean_ms": 0.8837016546726226, "n": 300}, {"repeat": 1, "p50_ms": 0.8905759751796722, "p95_ms": 0.9001616150140762, "p99_ms": 0.901730854511261, "mean_ms": 0.8805973315238953, "n": 300}, {"repeat": 2, "p50_ms": 0.8916800022125244, "p95_ms": 0.903718376159668, "p99_ms": 0.9085212761163711, "mean_ms": 0.8885136012236278, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_597583b95ac01e65_native_dense.json", "configuration": "704x1280_tp4_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp4_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp4_sp1_v1/captures/qkv_rank0_5ff40347734fb848.pt", "context": {"phase": "commit", "step": null, "start_frame": 27, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "vllm_fa2", "status": "measured", "q_shape": [10560, 10, 128], "k_shape": [66880, 10, 128], "q_len": 10560, "k_len": 63360, "reference_error": {"max_abs": 0.0625, "rmse": 0.0003947158564490791, "relative_l2": 0.0004979970884523445, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.008822682812078764, "rmse": 0.00091823471189617, "relative_l2": 0.0016767313668988057, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 9.579391956329346, "p95_ms": 9.904393577575684, "p99_ms": 10.364512834548949, "mean_ms": 9.672374305725098, "n": 300}, {"repeat": 1, "p50_ms": 9.654928207397461, "p95_ms": 9.997792434692382, "p99_ms": 10.159459447860717, "mean_ms": 9.691199982961018, "n": 300}, {"repeat": 2, "p50_ms": 9.651600360870361, "p95_ms": 9.897343730926513, "p99_ms": 10.113007707595823, "mean_ms": 9.68795979499817, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_5ff40347734fb848_vllm_fa2.json", "configuration": "704x1280_tp4_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp4_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp4_sp1_v1/captures/qkv_rank0_5ff40347734fb848.pt", "context": {"phase": "commit", "step": null, "start_frame": 27, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "vllm_fa3", "status": "measured", "q_shape": [10560, 10, 128], "k_shape": [66880, 10, 128], "q_len": 10560, "k_len": 63360, "reference_error": {"max_abs": 0.0, "rmse": 0.0, "relative_l2": 0.0, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.008822682812078764, "rmse": 0.0009188787426784346, "relative_l2": 0.0016779073915032794, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 5.207391977310181, "p95_ms": 5.5925119638442995, "p99_ms": 5.963122086524963, "mean_ms": 5.269198721249898, "n": 300}, {"repeat": 1, "p50_ms": 5.291695833206177, "p95_ms": 5.402822232246399, "p99_ms": 5.454722900390625, "mean_ms": 5.2907355705897015, "n": 300}, {"repeat": 2, "p50_ms": 5.28987193107605, "p95_ms": 5.35635826587677, "p99_ms": 5.404429202079773, "mean_ms": 5.271096954345703, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_5ff40347734fb848_vllm_fa3.json", "configuration": "704x1280_tp4_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp4_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp4_sp1_v1/captures/qkv_rank0_5ff40347734fb848.pt", "context": {"phase": "commit", "step": null, "start_frame": 27, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_math", "status": "measured", "q_shape": [10560, 10, 128], "k_shape": [66880, 10, 128], "q_len": 10560, "k_len": 63360, "reference_error": {"max_abs": 0.0625, "rmse": 0.0007986779682610063, "relative_l2": 0.0010076598046583065, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.007808065173258161, "rmse": 0.0009045391105559228, "relative_l2": 0.0016517226800584737, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 146.19686889648438, "p95_ms": 146.75783309936523, "p99_ms": 147.11515365600587, "mean_ms": 146.18305257161458, "n": 300}, {"repeat": 1, "p50_ms": 146.41136169433594, "p95_ms": 147.18462066650392, "p99_ms": 147.59679473876952, "mean_ms": 146.39368245442708, "n": 300}, {"repeat": 2, "p50_ms": 146.29388427734375, "p95_ms": 146.89365615844727, "p99_ms": 147.12603958129884, "mean_ms": 146.29585439046224, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_5ff40347734fb848_sdpa_math.json", "configuration": "704x1280_tp4_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp4_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp4_sp1_v1/captures/qkv_rank0_5ff40347734fb848.pt", "context": {"phase": "commit", "step": null, "start_frame": 27, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_flash", "status": "measured", "q_shape": [10560, 10, 128], "k_shape": [66880, 10, 128], "q_len": 10560, "k_len": 63360, "reference_error": {"max_abs": 0.0625, "rmse": 0.0003953177074146978, "relative_l2": 0.0004987564195601867, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.008822682812078764, "rmse": 0.0009182347146929427, "relative_l2": 0.0016767313720058186, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 10.32148790359497, "p95_ms": 10.66026725769043, "p99_ms": 10.937569952011108, "mean_ms": 10.375892496109008, "n": 300}, {"repeat": 1, "p50_ms": 10.325504302978516, "p95_ms": 10.578714895248414, "p99_ms": 10.65731879234314, "mean_ms": 10.375975348154704, "n": 300}, {"repeat": 2, "p50_ms": 10.331504344940186, "p95_ms": 10.573653030395509, "p99_ms": 10.659247913360595, "mean_ms": 10.380554695129394, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_5ff40347734fb848_sdpa_flash.json", "configuration": "704x1280_tp4_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp4_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp4_sp1_v1/captures/qkv_rank0_5ff40347734fb848.pt", "context": {"phase": "commit", "step": null, "start_frame": 27, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_efficient", "status": "measured", "q_shape": [10560, 10, 128], "k_shape": [66880, 10, 128], "q_len": 10560, "k_len": 63360, "reference_error": {"max_abs": 0.0625, "rmse": 0.0006893695570318764, "relative_l2": 0.0008697497875002275, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.00873556951780552, "rmse": 0.0009170178640830516, "relative_l2": 0.0016745093566975311, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 19.846208572387695, "p95_ms": 19.879525184631348, "p99_ms": 19.89780912399292, "mean_ms": 19.848111705780028, "n": 300}, {"repeat": 1, "p50_ms": 19.84615993499756, "p95_ms": 19.873674392700195, "p99_ms": 19.890600566864013, "mean_ms": 19.849046885172527, "n": 300}, {"repeat": 2, "p50_ms": 19.84652805328369, "p95_ms": 19.87461585998535, "p99_ms": 19.89737434387207, "mean_ms": 19.847425810496013, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_5ff40347734fb848_sdpa_efficient.json", "configuration": "704x1280_tp4_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp4_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp4_sp1_v1/captures/qkv_rank0_5ff40347734fb848.pt", "context": {"phase": "commit", "step": null, "start_frame": 27, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_cudnn", "status": "measured", "q_shape": [10560, 10, 128], "k_shape": [66880, 10, 128], "q_len": 10560, "k_len": 63360, "reference_error": {"max_abs": 0.0625, "rmse": 0.000690334036151034, "relative_l2": 0.0008709666319349421, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.00873556951780552, "rmse": 0.0009164405810648339, "relative_l2": 0.0016734552160386305, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 5.717072010040283, "p95_ms": 6.11475203037262, "p99_ms": 6.709224758148193, "mean_ms": 5.793439796765646, "n": 300}, {"repeat": 1, "p50_ms": 5.798624038696289, "p95_ms": 5.933223938941955, "p99_ms": 6.045880317687988, "mean_ms": 5.808432737986247, "n": 300}, {"repeat": 2, "p50_ms": 5.816352128982544, "p95_ms": 5.928246474266052, "p99_ms": 5.936228194236755, "mean_ms": 5.817357444763184, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_5ff40347734fb848_sdpa_cudnn.json", "configuration": "704x1280_tp4_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp4_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp4_sp1_v1/captures/qkv_rank0_5ff40347734fb848.pt", "context": {"phase": "commit", "step": null, "start_frame": 27, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "native_dense", "status": "measured", "q_shape": [10560, 10, 128], "k_shape": [66880, 10, 128], "q_len": 10560, "k_len": 63360, "reference_error": {"max_abs": 0.0625, "rmse": 0.0003818083032985858, "relative_l2": 0.0004817121488357402, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.008822682812078764, "rmse": 0.0009182431507726822, "relative_l2": 0.001676746776606984, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 5.119679927825928, "p95_ms": 5.776660823822022, "p99_ms": 6.099176249504089, "mean_ms": 5.198700690269471, "n": 300}, {"repeat": 1, "p50_ms": 5.218640089035034, "p95_ms": 5.4857566356658936, "p99_ms": 5.65899751663208, "mean_ms": 5.25616544564565, "n": 300}, {"repeat": 2, "p50_ms": 5.216592073440552, "p95_ms": 5.343958401679993, "p99_ms": 5.431545686721802, "mean_ms": 5.208139723141988, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_5ff40347734fb848_native_dense.json", "configuration": "704x1280_tp4_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp4_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp4_sp1_v1/captures/qkv_rank0_6be6af8ca00bf828.pt", "context": {"phase": "commit", "step": null, "start_frame": 0, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "vllm_fa2", "status": "measured", "q_shape": [10560, 10, 128], "k_shape": [512, 10, 128], "q_len": 10560, "k_len": 512, "reference_error": {"max_abs": 0.00048828125, "rmse": 1.8694706818654643e-06, "relative_l2": 0.0002963672975621551, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.00031312853836204524, "rmse": 1.6216402217677728e-05, "relative_l2": 0.0016148793135987087, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.09696000069379807, "p95_ms": 0.09795360080897808, "p99_ms": 0.09837663657963275, "mean_ms": 0.09721130654215812, "n": 300}, {"repeat": 1, "p50_ms": 0.09696000069379807, "p95_ms": 0.0979200005531311, "p99_ms": 0.09840031675994396, "mean_ms": 0.0974061864366134, "n": 300}, {"repeat": 2, "p50_ms": 0.09697600081562996, "p95_ms": 0.09792160056531429, "p99_ms": 0.09868831895291805, "mean_ms": 0.09733770668506622, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_6be6af8ca00bf828_vllm_fa2.json", "configuration": "704x1280_tp4_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp4_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp4_sp1_v1/captures/qkv_rank0_6be6af8ca00bf828.pt", "context": {"phase": "commit", "step": null, "start_frame": 0, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "vllm_fa3", "status": "measured", "q_shape": [10560, 10, 128], "k_shape": [512, 10, 128], "q_len": 10560, "k_len": 512, "reference_error": {"max_abs": 0.00048828125, "rmse": 7.123916952712385e-07, "relative_l2": 0.00011293549750808723, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.00031312853836204524, "rmse": 1.6223249904602404e-05, "relative_l2": 0.0016155612273680035, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.07993599772453308, "p95_ms": 0.08092799782752991, "p99_ms": 0.08211263939738274, "mean_ms": 0.0801783467332522, "n": 300}, {"repeat": 1, "p50_ms": 0.07996799796819687, "p95_ms": 0.08089759759604931, "p99_ms": 0.08179231695830821, "mean_ms": 0.08050805350144705, "n": 300}, {"repeat": 2, "p50_ms": 0.07995199784636497, "p95_ms": 0.0809295978397131, "p99_ms": 0.08154144220054146, "mean_ms": 0.08048394689957301, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_6be6af8ca00bf828_vllm_fa3.json", "configuration": "704x1280_tp4_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp4_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp4_sp1_v1/captures/qkv_rank0_6be6af8ca00bf828.pt", "context": {"phase": "commit", "step": null, "start_frame": 0, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_math", "status": "measured", "q_shape": [10560, 10, 128], "k_shape": [512, 10, 128], "q_len": 10560, "k_len": 512, "reference_error": {"max_abs": 0.0009765625, "rmse": 4.690054967584468e-06, "relative_l2": 0.0007435146908931302, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.00031312853836204524, "rmse": 1.612106107377855e-05, "relative_l2": 0.0016053849486371771, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 1.265023946762085, "p95_ms": 1.2675199508666992, "p99_ms": 1.2690259337425231, "mean_ms": 1.2650697592894236, "n": 300}, {"repeat": 1, "p50_ms": 1.2651039958000183, "p95_ms": 1.2680704236030578, "p99_ms": 1.2700608217716216, "mean_ms": 1.2661038943131764, "n": 300}, {"repeat": 2, "p50_ms": 1.264672040939331, "p95_ms": 1.267683243751526, "p99_ms": 1.2695049118995667, "mean_ms": 1.265281172990799, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_6be6af8ca00bf828_sdpa_math.json", "configuration": "704x1280_tp4_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp4_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp4_sp1_v1/captures/qkv_rank0_6be6af8ca00bf828.pt", "context": {"phase": "commit", "step": null, "start_frame": 0, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_flash", "status": "measured", "q_shape": [10560, 10, 128], "k_shape": [512, 10, 128], "q_len": 10560, "k_len": 512, "reference_error": {"max_abs": 0.00048828125, "rmse": 1.8718339469978572e-06, "relative_l2": 0.000296741946123111, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.00031312853836204524, "rmse": 1.6216402217677728e-05, "relative_l2": 0.0016148793135987087, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.10262399911880493, "p95_ms": 0.10355199873447418, "p99_ms": 0.10397248186171054, "mean_ms": 0.10277151962121328, "n": 300}, {"repeat": 1, "p50_ms": 0.10249599814414978, "p95_ms": 0.10326399654150009, "p99_ms": 0.10416384331881999, "mean_ms": 0.10280735969543457, "n": 300}, {"repeat": 2, "p50_ms": 0.10239999741315842, "p95_ms": 0.10329599678516388, "p99_ms": 0.10387328118085862, "mean_ms": 0.10269333290557067, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_6be6af8ca00bf828_sdpa_flash.json", "configuration": "704x1280_tp4_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp4_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp4_sp1_v1/captures/qkv_rank0_6be6af8ca00bf828.pt", "context": {"phase": "commit", "step": null, "start_frame": 0, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_efficient", "status": "measured", "q_shape": [10560, 10, 128], "k_shape": [512, 10, 128], "q_len": 10560, "k_len": 512, "reference_error": {"max_abs": 0.0009765625, "rmse": 3.3061339734111717e-06, "relative_l2": 0.0005241216139857132, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.00031312853836204524, "rmse": 1.636530906935895e-05, "relative_l2": 0.0016297079168366684, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.17878399789333344, "p95_ms": 0.17942560240626335, "p99_ms": 0.1797443251311779, "mean_ms": 0.17890357365210852, "n": 300}, {"repeat": 1, "p50_ms": 0.17875200510025024, "p95_ms": 0.17948800325393677, "p99_ms": 0.18043200284242628, "mean_ms": 0.1790704000989596, "n": 300}, {"repeat": 2, "p50_ms": 0.17897599935531616, "p95_ms": 0.17977759838104249, "p99_ms": 0.1804963232576847, "mean_ms": 0.1792716799179713, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_6be6af8ca00bf828_sdpa_efficient.json", "configuration": "704x1280_tp4_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp4_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp4_sp1_v1/captures/qkv_rank0_6be6af8ca00bf828.pt", "context": {"phase": "commit", "step": null, "start_frame": 0, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_cudnn", "status": "measured", "q_shape": [10560, 10, 128], "k_shape": [512, 10, 128], "q_len": 10560, "k_len": 512, "reference_error": {"max_abs": 0.0009765625, "rmse": 3.3091312137228107e-06, "relative_l2": 0.0005245967666692624, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.00031312853836204524, "rmse": 1.636530906935895e-05, "relative_l2": 0.0016297079168366684, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.05913599953055382, "p95_ms": 0.05971680022776127, "p99_ms": 0.060416001826524734, "mean_ms": 0.059432533370951814, "n": 300}, {"repeat": 1, "p50_ms": 0.05936000123620033, "p95_ms": 0.0601952001452446, "p99_ms": 0.06056479908525942, "mean_ms": 0.059881599992513655, "n": 300}, {"repeat": 2, "p50_ms": 0.059328000992536545, "p95_ms": 0.06028800085186958, "p99_ms": 0.061058879271149634, "mean_ms": 0.05981120013942321, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_6be6af8ca00bf828_sdpa_cudnn.json", "configuration": "704x1280_tp4_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp4_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp4_sp1_v1/captures/qkv_rank0_6be6af8ca00bf828.pt", "context": {"phase": "commit", "step": null, "start_frame": 0, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "native_dense", "status": "measured", "q_shape": [10560, 10, 128], "k_shape": [512, 10, 128], "q_len": 10560, "k_len": 512, "reference_error": {"max_abs": 0.0, "rmse": 0.0, "relative_l2": 0.0, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.00031312853836204524, "rmse": 1.6226204585766558e-05, "relative_l2": 0.0016158554636249872, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.12331200018525124, "p95_ms": 0.13177439719438552, "p99_ms": 0.13841567561030382, "mean_ms": 0.12455552026629448, "n": 300}, {"repeat": 1, "p50_ms": 0.12355200201272964, "p95_ms": 0.12931519374251366, "p99_ms": 0.1424281622469423, "mean_ms": 0.12498677372932435, "n": 300}, {"repeat": 2, "p50_ms": 0.1234240010380745, "p95_ms": 0.12916480004787445, "p99_ms": 0.14979904070496558, "mean_ms": 0.1246984534462293, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_6be6af8ca00bf828_native_dense.json", "configuration": "704x1280_tp4_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp4_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp4_sp1_v1/captures/qkv_rank0_909ee6942527a295.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 9, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "vllm_fa2", "status": "measured", "q_shape": [10560, 10, 128], "k_shape": [66880, 10, 128], "q_len": 10560, "k_len": 42240, "reference_error": {"max_abs": 0.03125, "rmse": 0.0005042570095821124, "relative_l2": 0.0009594630473967333, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.009544603598139023, "rmse": 0.0007526028144098833, "relative_l2": 0.0019224770129287073, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 6.335808038711548, "p95_ms": 6.692385649681092, "p99_ms": 6.834464392662048, "mean_ms": 6.405864334106445, "n": 300}, {"repeat": 1, "p50_ms": 6.378047943115234, "p95_ms": 6.601345705986023, "p99_ms": 6.791525921821594, "mean_ms": 6.4127972141901655, "n": 300}, {"repeat": 2, "p50_ms": 6.434831857681274, "p95_ms": 6.603780817985535, "p99_ms": 6.7174139404296875, "mean_ms": 6.426144421895345, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_909ee6942527a295_vllm_fa2.json", "configuration": "704x1280_tp4_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp4_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp4_sp1_v1/captures/qkv_rank0_909ee6942527a295.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 9, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "vllm_fa3", "status": "measured", "q_shape": [10560, 10, 128], "k_shape": [66880, 10, 128], "q_len": 10560, "k_len": 42240, "reference_error": {"max_abs": 0.0, "rmse": 0.0, "relative_l2": 0.0, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.009544603598139023, "rmse": 0.0007565133960279007, "relative_l2": 0.0019324663500981608, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 3.483567953109741, "p95_ms": 3.7994287848472594, "p99_ms": 3.92439243555069, "mean_ms": 3.5333756732940675, "n": 300}, {"repeat": 1, "p50_ms": 3.5085920095443726, "p95_ms": 3.620772862434387, "p99_ms": 3.6475948429107667, "mean_ms": 3.520047365029653, "n": 300}, {"repeat": 2, "p50_ms": 3.509119987487793, "p95_ms": 3.618177580833435, "p99_ms": 3.6474591302871704, "mean_ms": 3.5193227744102478, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_909ee6942527a295_vllm_fa3.json", "configuration": "704x1280_tp4_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp4_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp4_sp1_v1/captures/qkv_rank0_909ee6942527a295.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 9, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_math", "status": "measured", "q_shape": [10560, 10, 128], "k_shape": [66880, 10, 128], "q_len": 10560, "k_len": 42240, "reference_error": {"max_abs": 0.03125, "rmse": 0.0009080957003560324, "relative_l2": 0.0017278575239113086, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.007523825360065128, "rmse": 0.0006496496486669119, "relative_l2": 0.0016594895635603508, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 97.14822387695312, "p95_ms": 97.58180923461914, "p99_ms": 97.98253105163575, "mean_ms": 97.1283316040039, "n": 300}, {"repeat": 1, "p50_ms": 97.53193664550781, "p95_ms": 97.93649024963379, "p99_ms": 98.15941665649414, "mean_ms": 97.49055109659831, "n": 300}, {"repeat": 2, "p50_ms": 97.58031845092773, "p95_ms": 98.00452537536621, "p99_ms": 98.19216270446778, "mean_ms": 97.55658020019531, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_909ee6942527a295_sdpa_math.json", "configuration": "704x1280_tp4_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp4_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp4_sp1_v1/captures/qkv_rank0_909ee6942527a295.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 9, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_flash", "status": "measured", "q_shape": [10560, 10, 128], "k_shape": [66880, 10, 128], "q_len": 10560, "k_len": 42240, "reference_error": {"max_abs": 0.03125, "rmse": 0.0005046739119944005, "relative_l2": 0.0009602562985590575, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.009544603598139023, "rmse": 0.0007531249905655517, "relative_l2": 0.001923810879394193, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 6.8846399784088135, "p95_ms": 7.174467062950134, "p99_ms": 7.426773209571838, "mean_ms": 6.947688330014547, "n": 300}, {"repeat": 1, "p50_ms": 6.934831857681274, "p95_ms": 7.107673501968383, "p99_ms": 7.116136608123779, "mean_ms": 6.957056959470113, "n": 300}, {"repeat": 2, "p50_ms": 6.936223983764648, "p95_ms": 7.054758286476135, "p99_ms": 7.060099630355835, "mean_ms": 6.944597977002462, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_909ee6942527a295_sdpa_flash.json", "configuration": "704x1280_tp4_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp4_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp4_sp1_v1/captures/qkv_rank0_909ee6942527a295.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 9, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_efficient", "status": "measured", "q_shape": [10560, 10, 128], "k_shape": [66880, 10, 128], "q_len": 10560, "k_len": 42240, "reference_error": {"max_abs": 0.03125, "rmse": 0.0007799356002997485, "relative_l2": 0.0014840039377081624, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.00892804239358469, "rmse": 0.0007526315616923348, "relative_l2": 0.0019225504460711816, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 13.23302412033081, "p95_ms": 13.268326091766358, "p99_ms": 13.278656969070434, "mean_ms": 13.238311058680216, "n": 300}, {"repeat": 1, "p50_ms": 13.235328197479248, "p95_ms": 13.270263814926148, "p99_ms": 13.281856231689453, "mean_ms": 13.241829217274983, "n": 300}, {"repeat": 2, "p50_ms": 13.23360013961792, "p95_ms": 13.265400075912476, "p99_ms": 13.278954572677613, "mean_ms": 13.2381374168396, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_909ee6942527a295_sdpa_efficient.json", "configuration": "704x1280_tp4_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp4_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp4_sp1_v1/captures/qkv_rank0_909ee6942527a295.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 9, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_cudnn", "status": "measured", "q_shape": [10560, 10, 128], "k_shape": [66880, 10, 128], "q_len": 10560, "k_len": 42240, "reference_error": {"max_abs": 0.03125, "rmse": 0.0007793332139748923, "relative_l2": 0.0014828577614369872, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.00892804239358469, "rmse": 0.0007528017423138951, "relative_l2": 0.0019229851618691185, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 3.8505440950393677, "p95_ms": 4.124633574485779, "p99_ms": 4.583659081459046, "mean_ms": 3.893017597993215, "n": 300}, {"repeat": 1, "p50_ms": 3.852720022201538, "p95_ms": 4.038201403617859, "p99_ms": 4.075970349311828, "mean_ms": 3.8810794695218402, "n": 300}, {"repeat": 2, "p50_ms": 3.884719967842102, "p95_ms": 3.972001540660858, "p99_ms": 3.9968421339988707, "mean_ms": 3.8841590372721355, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_909ee6942527a295_sdpa_cudnn.json", "configuration": "704x1280_tp4_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp4_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp4_sp1_v1/captures/qkv_rank0_909ee6942527a295.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 9, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "native_dense", "status": "measured", "q_shape": [10560, 10, 128], "k_shape": [66880, 10, 128], "q_len": 10560, "k_len": 42240, "reference_error": {"max_abs": 0.03125, "rmse": 0.0004802122176232925, "relative_l2": 0.0009137123906315485, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.007523825360065128, "rmse": 0.0007609683905593386, "relative_l2": 0.0019438463561457437, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 3.5074400901794434, "p95_ms": 3.640057504177094, "p99_ms": 3.822569031715393, "mean_ms": 3.518224430878957, "n": 300}, {"repeat": 1, "p50_ms": 3.5152000188827515, "p95_ms": 3.6047248601913453, "p99_ms": 3.6251304864883425, "mean_ms": 3.5043821914990745, "n": 300}, {"repeat": 2, "p50_ms": 3.5173920392990112, "p95_ms": 3.571793520450592, "p99_ms": 3.5911692070960997, "mean_ms": 3.511685225168864, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_909ee6942527a295_native_dense.json", "configuration": "704x1280_tp4_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp4_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp4_sp1_v1/captures/qkv_rank0_aff3e85d5a3411c1.pt", "context": {"phase": "commit", "step": null, "start_frame": 9, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "vllm_fa2", "status": "measured", "q_shape": [10560, 10, 128], "k_shape": [66880, 10, 128], "q_len": 10560, "k_len": 42240, "reference_error": {"max_abs": 0.0625, "rmse": 0.00040462564833064566, "relative_l2": 0.000509108907392721, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.009308827301267808, "rmse": 0.0009185162838646229, "relative_l2": 0.0016820797707702033, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 6.336431980133057, "p95_ms": 6.664740753173828, "p99_ms": 6.898023543357849, "mean_ms": 6.407793067296346, "n": 300}, {"repeat": 1, "p50_ms": 6.337631940841675, "p95_ms": 6.5947198390960695, "p99_ms": 6.663160376548767, "mean_ms": 6.406529178619385, "n": 300}, {"repeat": 2, "p50_ms": 6.385456085205078, "p95_ms": 6.498528146743775, "p99_ms": 6.65882098197937, "mean_ms": 6.414132372538249, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_aff3e85d5a3411c1_vllm_fa2.json", "configuration": "704x1280_tp4_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp4_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp4_sp1_v1/captures/qkv_rank0_aff3e85d5a3411c1.pt", "context": {"phase": "commit", "step": null, "start_frame": 9, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "vllm_fa3", "status": "measured", "q_shape": [10560, 10, 128], "k_shape": [66880, 10, 128], "q_len": 10560, "k_len": 42240, "reference_error": {"max_abs": 0.0, "rmse": 0.0, "relative_l2": 0.0, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.009308827301267808, "rmse": 0.0009180934434011803, "relative_l2": 0.001681305422615125, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 3.477728009223938, "p95_ms": 3.7299455761909486, "p99_ms": 3.9330399298667906, "mean_ms": 3.514591040611267, "n": 300}, {"repeat": 1, "p50_ms": 3.476991891860962, "p95_ms": 3.620710384845734, "p99_ms": 3.6513760352134703, "mean_ms": 3.4997956283887226, "n": 300}, {"repeat": 2, "p50_ms": 3.4860639572143555, "p95_ms": 3.612100863456726, "p99_ms": 3.6161420488357545, "mean_ms": 3.506738775571187, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_aff3e85d5a3411c1_vllm_fa3.json", "configuration": "704x1280_tp4_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp4_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp4_sp1_v1/captures/qkv_rank0_aff3e85d5a3411c1.pt", "context": {"phase": "commit", "step": null, "start_frame": 9, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_math", "status": "measured", "q_shape": [10560, 10, 128], "k_shape": [66880, 10, 128], "q_len": 10560, "k_len": 42240, "reference_error": {"max_abs": 0.0625, "rmse": 0.0007800226017953204, "relative_l2": 0.000981441626797552, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.007675077368443528, "rmse": 0.0008976327122820362, "relative_l2": 0.0016438356656654992, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 97.38183975219727, "p95_ms": 97.86647987365723, "p99_ms": 98.03152320861817, "mean_ms": 97.375191599528, "n": 300}, {"repeat": 1, "p50_ms": 97.5410270690918, "p95_ms": 97.9499267578125, "p99_ms": 98.16388847351074, "mean_ms": 97.5140108235677, "n": 300}, {"repeat": 2, "p50_ms": 97.48006057739258, "p95_ms": 97.92257423400879, "p99_ms": 98.14679290771484, "mean_ms": 97.48162358601888, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_aff3e85d5a3411c1_sdpa_math.json", "configuration": "704x1280_tp4_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp4_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp4_sp1_v1/captures/qkv_rank0_aff3e85d5a3411c1.pt", "context": {"phase": "commit", "step": null, "start_frame": 9, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_flash", "status": "measured", "q_shape": [10560, 10, 128], "k_shape": [66880, 10, 128], "q_len": 10560, "k_len": 42240, "reference_error": {"max_abs": 0.0625, "rmse": 0.0004051937999150958, "relative_l2": 0.0005098237682365311, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.009308827301267808, "rmse": 0.0009185162838646229, "relative_l2": 0.0016820797707702033, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 6.880640029907227, "p95_ms": 7.110044670104981, "p99_ms": 7.172495784759522, "mean_ms": 6.923136952718099, "n": 300}, {"repeat": 1, "p50_ms": 6.894720077514648, "p95_ms": 7.107115030288696, "p99_ms": 7.110725598335266, "mean_ms": 6.928415360450745, "n": 300}, {"repeat": 2, "p50_ms": 6.931120157241821, "p95_ms": 7.107176184654236, "p99_ms": 7.109410142898559, "mean_ms": 6.931052587827047, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_aff3e85d5a3411c1_sdpa_flash.json", "configuration": "704x1280_tp4_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp4_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp4_sp1_v1/captures/qkv_rank0_aff3e85d5a3411c1.pt", "context": {"phase": "commit", "step": null, "start_frame": 9, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_efficient", "status": "measured", "q_shape": [10560, 10, 128], "k_shape": [66880, 10, 128], "q_len": 10560, "k_len": 42240, "reference_error": {"max_abs": 0.0625, "rmse": 0.000674114197242648, "relative_l2": 0.0008481853383047951, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.008502093919382947, "rmse": 0.0009060178476741044, "relative_l2": 0.0016591913723262711, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 13.220879554748535, "p95_ms": 13.255459260940551, "p99_ms": 13.26785701751709, "mean_ms": 13.225792112350463, "n": 300}, {"repeat": 1, "p50_ms": 13.220863819122314, "p95_ms": 13.257017946243286, "p99_ms": 13.267808599472046, "mean_ms": 13.226541347503662, "n": 300}, {"repeat": 2, "p50_ms": 13.221807956695557, "p95_ms": 13.255652475357056, "p99_ms": 13.270109901428222, "mean_ms": 13.22698442141215, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_aff3e85d5a3411c1_sdpa_efficient.json", "configuration": "704x1280_tp4_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp4_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp4_sp1_v1/captures/qkv_rank0_aff3e85d5a3411c1.pt", "context": {"phase": "commit", "step": null, "start_frame": 9, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_cudnn", "status": "measured", "q_shape": [10560, 10, 128], "k_shape": [66880, 10, 128], "q_len": 10560, "k_len": 42240, "reference_error": {"max_abs": 0.0625, "rmse": 0.0006740104752449736, "relative_l2": 0.0008480548330016182, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.008502093919382947, "rmse": 0.0009058996741385277, "relative_l2": 0.0016589749610147616, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 3.818495988845825, "p95_ms": 4.279744029045106, "p99_ms": 4.522517342567443, "mean_ms": 3.8785170141855874, "n": 300}, {"repeat": 1, "p50_ms": 3.8509600162506104, "p95_ms": 3.9994239807128906, "p99_ms": 4.075680103302002, "mean_ms": 3.8712914141019183, "n": 300}, {"repeat": 2, "p50_ms": 3.859615921974182, "p95_ms": 3.962915277481079, "p99_ms": 3.9718153166770933, "mean_ms": 3.864770766099294, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_aff3e85d5a3411c1_sdpa_cudnn.json", "configuration": "704x1280_tp4_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp4_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp4_sp1_v1/captures/qkv_rank0_aff3e85d5a3411c1.pt", "context": {"phase": "commit", "step": null, "start_frame": 9, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "native_dense", "status": "measured", "q_shape": [10560, 10, 128], "k_shape": [66880, 10, 128], "q_len": 10560, "k_len": 42240, "reference_error": {"max_abs": 0.0625, "rmse": 0.0003865444794383193, "relative_l2": 0.0004863587822409084, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.009308827301267808, "rmse": 0.0009196086006770146, "relative_l2": 0.0016840801316191864, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 3.448720097541809, "p95_ms": 3.9498704433441163, "p99_ms": 4.044993891716003, "mean_ms": 3.5022778662045795, "n": 300}, {"repeat": 1, "p50_ms": 3.4529600143432617, "p95_ms": 3.6683231472969053, "p99_ms": 3.7329730439186095, "mean_ms": 3.4948246391614277, "n": 300}, {"repeat": 2, "p50_ms": 3.4939520359039307, "p95_ms": 3.6354159474372865, "p99_ms": 3.669423577785492, "mean_ms": 3.4977959537506105, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_aff3e85d5a3411c1_native_dense.json", "configuration": "704x1280_tp4_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp4_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp4_sp1_v1/captures/qkv_rank0_b37e08eb3a08f049.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 0, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "vllm_fa2", "status": "measured", "q_shape": [10560, 10, 128], "k_shape": [512, 10, 128], "q_len": 10560, "k_len": 512, "reference_error": {"max_abs": 0.0009765625, "rmse": 1.9031704135855596e-06, "relative_l2": 0.00031324409734920114, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.00048152582391679244, "rmse": 2.005116132159013e-05, "relative_l2": 0.0019177690486011601, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.09567999839782715, "p95_ms": 0.09670559875667095, "p99_ms": 0.09740927666425706, "mean_ms": 0.0959115732461214, "n": 300}, {"repeat": 1, "p50_ms": 0.0958079993724823, "p95_ms": 0.0968639999628067, "p99_ms": 0.09760735787451263, "mean_ms": 0.0964887464294831, "n": 300}, {"repeat": 2, "p50_ms": 0.09596800059080124, "p95_ms": 0.0970896016806364, "p99_ms": 0.09763263836503029, "mean_ms": 0.09639711971084276, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_b37e08eb3a08f049_vllm_fa2.json", "configuration": "704x1280_tp4_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp4_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp4_sp1_v1/captures/qkv_rank0_b37e08eb3a08f049.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 0, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "vllm_fa3", "status": "measured", "q_shape": [10560, 10, 128], "k_shape": [512, 10, 128], "q_len": 10560, "k_len": 512, "reference_error": {"max_abs": 0.00048828125, "rmse": 7.025696737956701e-07, "relative_l2": 0.00011563641475406615, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.00048152582391679244, "rmse": 2.0069218185708738e-05, "relative_l2": 0.0019194960755082836, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.08025600016117096, "p95_ms": 0.08118879981338978, "p99_ms": 0.08169952362775802, "mean_ms": 0.0804590933273236, "n": 300}, {"repeat": 1, "p50_ms": 0.08027200028300285, "p95_ms": 0.08124800026416779, "p99_ms": 0.08179519698023796, "mean_ms": 0.08095946652193864, "n": 300}, {"repeat": 2, "p50_ms": 0.08025600016117096, "p95_ms": 0.08115359954535961, "p99_ms": 0.08156960263848305, "mean_ms": 0.08086815988024076, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_b37e08eb3a08f049_vllm_fa3.json", "configuration": "704x1280_tp4_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp4_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp4_sp1_v1/captures/qkv_rank0_b37e08eb3a08f049.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 0, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_math", "status": "measured", "q_shape": [10560, 10, 128], "k_shape": [512, 10, 128], "q_len": 10560, "k_len": 512, "reference_error": {"max_abs": 0.001953125, "rmse": 1.2904659629547824e-05, "relative_l2": 0.0021239866006747678, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.00042757268329138287, "rmse": 1.6223370232935805e-05, "relative_l2": 0.0015516646042451933, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 1.2640000581741333, "p95_ms": 1.2674239873886108, "p99_ms": 1.2685442793369293, "mean_ms": 1.2645151976744333, "n": 300}, {"repeat": 1, "p50_ms": 1.2645119428634644, "p95_ms": 1.2678783774375915, "p99_ms": 1.2695177114009857, "mean_ms": 1.2650548283259073, "n": 300}, {"repeat": 2, "p50_ms": 1.2644480466842651, "p95_ms": 1.2674560546875, "p99_ms": 1.2688982236385344, "mean_ms": 1.2650229334831238, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_b37e08eb3a08f049_sdpa_math.json", "configuration": "704x1280_tp4_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp4_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp4_sp1_v1/captures/qkv_rank0_b37e08eb3a08f049.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 0, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_flash", "status": "measured", "q_shape": [10560, 10, 128], "k_shape": [512, 10, 128], "q_len": 10560, "k_len": 512, "reference_error": {"max_abs": 0.0009765625, "rmse": 1.9051126339996649e-06, "relative_l2": 0.00031356376871237847, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.00048152582391679244, "rmse": 2.005116132159013e-05, "relative_l2": 0.0019177690486011601, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.10246399790048599, "p95_ms": 0.10326399654150009, "p99_ms": 0.10352223850786686, "mean_ms": 0.10259498638411363, "n": 300}, {"repeat": 1, "p50_ms": 0.10252799838781357, "p95_ms": 0.10335999727249146, "p99_ms": 0.1037785604596138, "mean_ms": 0.10289610634247462, "n": 300}, {"repeat": 2, "p50_ms": 0.10255999863147736, "p95_ms": 0.10329759679734707, "p99_ms": 0.1041292830556631, "mean_ms": 0.10286463958521684, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_b37e08eb3a08f049_sdpa_flash.json", "configuration": "704x1280_tp4_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp4_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp4_sp1_v1/captures/qkv_rank0_b37e08eb3a08f049.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 0, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_efficient", "status": "measured", "q_shape": [10560, 10, 128], "k_shape": [512, 10, 128], "q_len": 10560, "k_len": 512, "reference_error": {"max_abs": 0.0009765625, "rmse": 3.3028501872963014e-06, "relative_l2": 0.0005436183319233584, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.00048152582391679244, "rmse": 1.9910171391925443e-05, "relative_l2": 0.0019042842374753196, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.17876800149679184, "p95_ms": 0.17951999604701996, "p99_ms": 0.18009792044758796, "mean_ms": 0.17891509354114532, "n": 300}, {"repeat": 1, "p50_ms": 0.17871999740600586, "p95_ms": 0.17951999604701996, "p99_ms": 0.17983999848365784, "mean_ms": 0.1790464000403881, "n": 300}, {"repeat": 2, "p50_ms": 0.1791680008172989, "p95_ms": 0.17980800569057465, "p99_ms": 0.18041664287447928, "mean_ms": 0.17947957282265029, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_b37e08eb3a08f049_sdpa_efficient.json", "configuration": "704x1280_tp4_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp4_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp4_sp1_v1/captures/qkv_rank0_b37e08eb3a08f049.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 0, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_cudnn", "status": "measured", "q_shape": [10560, 10, 128], "k_shape": [512, 10, 128], "q_len": 10560, "k_len": 512, "reference_error": {"max_abs": 0.0009765625, "rmse": 3.2847158255872896e-06, "relative_l2": 0.0005406335851429366, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.00048152582391679244, "rmse": 2.015425473606218e-05, "relative_l2": 0.001927629293412827, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.059167999774217606, "p95_ms": 0.059969598427414894, "p99_ms": 0.06045503836125135, "mean_ms": 0.05955744004497925, "n": 300}, {"repeat": 1, "p50_ms": 0.059167999774217606, "p95_ms": 0.0599056014791131, "p99_ms": 0.06032608106732368, "mean_ms": 0.05969141338020563, "n": 300}, {"repeat": 2, "p50_ms": 0.059167999774217606, "p95_ms": 0.059939198195934296, "p99_ms": 0.06102399900555611, "mean_ms": 0.05960085328668356, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_b37e08eb3a08f049_sdpa_cudnn.json", "configuration": "704x1280_tp4_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp4_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp4_sp1_v1/captures/qkv_rank0_b37e08eb3a08f049.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 0, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "native_dense", "status": "measured", "q_shape": [10560, 10, 128], "k_shape": [512, 10, 128], "q_len": 10560, "k_len": 512, "reference_error": {"max_abs": 0.0, "rmse": 0.0, "relative_l2": 0.0, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.00048152582391679244, "rmse": 2.0069586320417118e-05, "relative_l2": 0.0019195312853067647, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.12319999933242798, "p95_ms": 0.1301823928952217, "p99_ms": 0.1492611239850521, "mean_ms": 0.12439552033940951, "n": 300}, {"repeat": 1, "p50_ms": 0.12278400361537933, "p95_ms": 0.13498879969120026, "p99_ms": 0.1671705661714077, "mean_ms": 0.12523402693370977, "n": 300}, {"repeat": 2, "p50_ms": 0.12252800166606903, "p95_ms": 0.12720159962773323, "p99_ms": 0.1297308766841888, "mean_ms": 0.1236015996336937, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_b37e08eb3a08f049_native_dense.json", "configuration": "704x1280_tp4_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp4_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp4_sp1_v1/captures/qkv_rank0_be6bbd23e5a64d4b.pt", "context": {"phase": "commit", "step": null, "start_frame": 9, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "vllm_fa2", "status": "measured", "q_shape": [10560, 10, 128], "k_shape": [512, 10, 128], "q_len": 10560, "k_len": 512, "reference_error": {"max_abs": 0.000244140625, "rmse": 1.2131821409825966e-06, "relative_l2": 0.0006374085004020706, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.00010303920358695207, "rmse": 5.124063667211515e-06, "relative_l2": 0.0017274753842448528, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.09627200290560722, "p95_ms": 0.09728160314261913, "p99_ms": 0.09756895788013935, "mean_ms": 0.09652170690397421, "n": 300}, {"repeat": 1, "p50_ms": 0.09636799991130829, "p95_ms": 0.09728000313043594, "p99_ms": 0.09760127812623978, "mean_ms": 0.0968266670157512, "n": 300}, {"repeat": 2, "p50_ms": 0.09647999703884125, "p95_ms": 0.09731200337409973, "p99_ms": 0.09782399982213974, "mean_ms": 0.09690432014564673, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_be6bbd23e5a64d4b_vllm_fa2.json", "configuration": "704x1280_tp4_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp4_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp4_sp1_v1/captures/qkv_rank0_be6bbd23e5a64d4b.pt", "context": {"phase": "commit", "step": null, "start_frame": 9, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "vllm_fa3", "status": "measured", "q_shape": [10560, 10, 128], "k_shape": [512, 10, 128], "q_len": 10560, "k_len": 512, "reference_error": {"max_abs": 0.0001220703125, "rmse": 4.474371539407549e-07, "relative_l2": 0.0002350844408957041, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.00010303920358695207, "rmse": 5.1476296107596015e-06, "relative_l2": 0.0017354201698739462, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.0809599980711937, "p95_ms": 0.08204959891736507, "p99_ms": 0.0856848033517598, "mean_ms": 0.08128864002724488, "n": 300}, {"repeat": 1, "p50_ms": 0.08083199709653854, "p95_ms": 0.08201759867370129, "p99_ms": 0.08278591707348823, "mean_ms": 0.0812822399288416, "n": 300}, {"repeat": 2, "p50_ms": 0.08083199709653854, "p95_ms": 0.08204959891736507, "p99_ms": 0.08256160274147988, "mean_ms": 0.0811962665617466, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_be6bbd23e5a64d4b_vllm_fa3.json", "configuration": "704x1280_tp4_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp4_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp4_sp1_v1/captures/qkv_rank0_be6bbd23e5a64d4b.pt", "context": {"phase": "commit", "step": null, "start_frame": 9, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_math", "status": "measured", "q_shape": [10560, 10, 128], "k_shape": [512, 10, 128], "q_len": 10560, "k_len": 512, "reference_error": {"max_abs": 0.000244140625, "rmse": 2.6223606643450123e-06, "relative_l2": 0.0013777939207213488, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.00010303920358695207, "rmse": 4.923214436427207e-06, "relative_l2": 0.001659763090905371, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 1.2637919783592224, "p95_ms": 1.2677776098251343, "p99_ms": 1.2716940772533416, "mean_ms": 1.2648343443870544, "n": 300}, {"repeat": 1, "p50_ms": 1.2639039754867554, "p95_ms": 1.2679359912872314, "p99_ms": 1.2689004635810852, "mean_ms": 1.2645436775684358, "n": 300}, {"repeat": 2, "p50_ms": 1.2644000053405762, "p95_ms": 1.2681920528411865, "p99_ms": 1.2689606392383574, "mean_ms": 1.2650718943277994, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_be6bbd23e5a64d4b_sdpa_math.json", "configuration": "704x1280_tp4_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp4_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp4_sp1_v1/captures/qkv_rank0_be6bbd23e5a64d4b.pt", "context": {"phase": "commit", "step": null, "start_frame": 9, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_flash", "status": "measured", "q_shape": [10560, 10, 128], "k_shape": [512, 10, 128], "q_len": 10560, "k_len": 512, "reference_error": {"max_abs": 0.000244140625, "rmse": 1.2133666352136358e-06, "relative_l2": 0.0006375054340669893, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.00010303920358695207, "rmse": 5.124063667211515e-06, "relative_l2": 0.0017274753842448528, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.10246399790048599, "p95_ms": 0.10326399654150009, "p99_ms": 0.10371327996253968, "mean_ms": 0.10264661292235057, "n": 300}, {"repeat": 1, "p50_ms": 0.10246399790048599, "p95_ms": 0.10332799702882767, "p99_ms": 0.10365023948252201, "mean_ms": 0.10276106623311837, "n": 300}, {"repeat": 2, "p50_ms": 0.10246399790048599, "p95_ms": 0.10336159728467464, "p99_ms": 0.10387424118816853, "mean_ms": 0.10282890635232131, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_be6bbd23e5a64d4b_sdpa_flash.json", "configuration": "704x1280_tp4_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp4_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp4_sp1_v1/captures/qkv_rank0_be6bbd23e5a64d4b.pt", "context": {"phase": "commit", "step": null, "start_frame": 9, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_efficient", "status": "measured", "q_shape": [10560, 10, 128], "k_shape": [512, 10, 128], "q_len": 10560, "k_len": 512, "reference_error": {"max_abs": 0.000244140625, "rmse": 2.1282189416635235e-06, "relative_l2": 0.0011181707991796824, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.00010303920358695207, "rmse": 5.512863327133934e-06, "relative_l2": 0.0018585513984279965, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.17900800704956055, "p95_ms": 0.17974560484290122, "p99_ms": 0.1803289547562599, "mean_ms": 0.17917653302351633, "n": 300}, {"repeat": 1, "p50_ms": 0.17903999984264374, "p95_ms": 0.17958559691905976, "p99_ms": 0.17996927946805955, "mean_ms": 0.17923370614647866, "n": 300}, {"repeat": 2, "p50_ms": 0.1785919964313507, "p95_ms": 0.17923520132899284, "p99_ms": 0.17964895710349083, "mean_ms": 0.1788194137811661, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_be6bbd23e5a64d4b_sdpa_efficient.json", "configuration": "704x1280_tp4_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp4_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp4_sp1_v1/captures/qkv_rank0_be6bbd23e5a64d4b.pt", "context": {"phase": "commit", "step": null, "start_frame": 9, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_cudnn", "status": "measured", "q_shape": [10560, 10, 128], "k_shape": [512, 10, 128], "q_len": 10560, "k_len": 512, "reference_error": {"max_abs": 0.000244140625, "rmse": 2.12632307463885e-06, "relative_l2": 0.0011171747065763244, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.00010303920358695207, "rmse": 5.512932142800092e-06, "relative_l2": 0.001858574598250844, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.05910399928689003, "p95_ms": 0.0598768012598157, "p99_ms": 0.06083232127130032, "mean_ms": 0.05943967985610167, "n": 300}, {"repeat": 1, "p50_ms": 0.05913599953055382, "p95_ms": 0.059937598183751106, "p99_ms": 0.06110175948590039, "mean_ms": 0.0596294399847587, "n": 300}, {"repeat": 2, "p50_ms": 0.05910399928689003, "p95_ms": 0.059777600690722466, "p99_ms": 0.06019328013062477, "mean_ms": 0.0596230398863554, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_be6bbd23e5a64d4b_sdpa_cudnn.json", "configuration": "704x1280_tp4_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp4_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp4_sp1_v1/captures/qkv_rank0_be6bbd23e5a64d4b.pt", "context": {"phase": "commit", "step": null, "start_frame": 9, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "native_dense", "status": "measured", "q_shape": [10560, 10, 128], "k_shape": [512, 10, 128], "q_len": 10560, "k_len": 512, "reference_error": {"max_abs": 0.0, "rmse": 0.0, "relative_l2": 0.0, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.00010303920358695207, "rmse": 5.155021065710917e-06, "relative_l2": 0.0017379120507933533, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.12326399981975555, "p95_ms": 0.13174080401659014, "p99_ms": 0.1506460788846015, "mean_ms": 0.12488554688791434, "n": 300}, {"repeat": 1, "p50_ms": 0.12323199957609177, "p95_ms": 0.13144160211086275, "p99_ms": 0.22296608656644734, "mean_ms": 0.12717045324544113, "n": 300}, {"repeat": 2, "p50_ms": 0.12275200337171555, "p95_ms": 0.1283295936882496, "p99_ms": 0.15676415756344747, "mean_ms": 0.1257201066116492, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_be6bbd23e5a64d4b_native_dense.json", "configuration": "704x1280_tp4_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp4_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp4_sp1_v1/captures/qkv_rank0_c1119485a5572ce6.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 27, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "vllm_fa2", "status": "measured", "q_shape": [10560, 10, 128], "k_shape": [66880, 10, 128], "q_len": 10560, "k_len": 63360, "reference_error": {"max_abs": 0.03125, "rmse": 0.0004966972535144519, "relative_l2": 0.0009502337763859682, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.009103983744871424, "rmse": 0.0007474701713503258, "relative_l2": 0.0019252806107063632, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 9.647808074951172, "p95_ms": 9.902174139022828, "p99_ms": 10.254633321762086, "mean_ms": 9.634670248031616, "n": 300}, {"repeat": 1, "p50_ms": 9.650176048278809, "p95_ms": 9.825137567520141, "p99_ms": 9.900735559463499, "mean_ms": 9.649464422861735, "n": 300}, {"repeat": 2, "p50_ms": 9.654592037200928, "p95_ms": 9.817057800292968, "p99_ms": 9.899502372741697, "mean_ms": 9.665374501546223, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_c1119485a5572ce6_vllm_fa2.json", "configuration": "704x1280_tp4_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp4_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp4_sp1_v1/captures/qkv_rank0_c1119485a5572ce6.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 27, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "vllm_fa3", "status": "measured", "q_shape": [10560, 10, 128], "k_shape": [66880, 10, 128], "q_len": 10560, "k_len": 63360, "reference_error": {"max_abs": 0.0, "rmse": 0.0, "relative_l2": 0.0, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.009103983744871424, "rmse": 0.0007535473215642128, "relative_l2": 0.001940933702861211, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 5.258304119110107, "p95_ms": 5.457486248016357, "p99_ms": 5.571266350746154, "mean_ms": 5.2818314695358275, "n": 300}, {"repeat": 1, "p50_ms": 5.292864084243774, "p95_ms": 5.40372941493988, "p99_ms": 5.4091568326950075, "mean_ms": 5.288239239056905, "n": 300}, {"repeat": 2, "p50_ms": 5.294032096862793, "p95_ms": 5.358054375648498, "p99_ms": 5.402953686714173, "mean_ms": 5.285219944318135, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_c1119485a5572ce6_vllm_fa3.json", "configuration": "704x1280_tp4_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp4_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp4_sp1_v1/captures/qkv_rank0_c1119485a5572ce6.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 27, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_math", "status": "measured", "q_shape": [10560, 10, 128], "k_shape": [66880, 10, 128], "q_len": 10560, "k_len": 63360, "reference_error": {"max_abs": 0.03125, "rmse": 0.0009081671830546564, "relative_l2": 0.0017374187713698, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.007101973192271238, "rmse": 0.0006476748990884629, "relative_l2": 0.0016682350320462382, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 146.06645965576172, "p95_ms": 146.73526840209962, "p99_ms": 147.16217330932616, "mean_ms": 146.0789347330729, "n": 300}, {"repeat": 1, "p50_ms": 146.4680938720703, "p95_ms": 147.0257209777832, "p99_ms": 147.41828491210939, "mean_ms": 146.46806325276694, "n": 300}, {"repeat": 2, "p50_ms": 146.36370849609375, "p95_ms": 146.92665939331056, "p99_ms": 147.1806413269043, "mean_ms": 146.36039540608724, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_c1119485a5572ce6_sdpa_math.json", "configuration": "704x1280_tp4_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp4_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp4_sp1_v1/captures/qkv_rank0_c1119485a5572ce6.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 27, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_flash", "status": "measured", "q_shape": [10560, 10, 128], "k_shape": [66880, 10, 128], "q_len": 10560, "k_len": 63360, "reference_error": {"max_abs": 0.03125, "rmse": 0.0004971026307267522, "relative_l2": 0.0009510093053758699, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.009103983744871424, "rmse": 0.000747067582265573, "relative_l2": 0.0019242436503182898, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 10.38152027130127, "p95_ms": 10.661125183105469, "p99_ms": 10.806154098510742, "mean_ms": 10.412020788192748, "n": 300}, {"repeat": 1, "p50_ms": 10.39580774307251, "p95_ms": 10.589684629440308, "p99_ms": 10.65817642211914, "mean_ms": 10.406505896250406, "n": 300}, {"repeat": 2, "p50_ms": 10.393856048583984, "p95_ms": 10.582636833190918, "p99_ms": 10.658849411010742, "mean_ms": 10.40682162920634, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_c1119485a5572ce6_sdpa_flash.json", "configuration": "704x1280_tp4_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp4_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp4_sp1_v1/captures/qkv_rank0_c1119485a5572ce6.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 27, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_efficient", "status": "measured", "q_shape": [10560, 10, 128], "k_shape": [66880, 10, 128], "q_len": 10560, "k_len": 63360, "reference_error": {"max_abs": 0.03125, "rmse": 0.0007833553830292504, "relative_l2": 0.0014986407486678461, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.009309301063288977, "rmse": 0.0007472412349038868, "relative_l2": 0.0019246909324579092, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 19.8536958694458, "p95_ms": 19.88556299209595, "p99_ms": 19.90598934173584, "mean_ms": 19.85594013849894, "n": 300}, {"repeat": 1, "p50_ms": 19.855087280273438, "p95_ms": 19.890561294555663, "p99_ms": 19.90552631378174, "mean_ms": 19.856740385691324, "n": 300}, {"repeat": 2, "p50_ms": 19.855408668518066, "p95_ms": 19.886215591430663, "p99_ms": 19.90835874557495, "mean_ms": 19.856585381825766, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_c1119485a5572ce6_sdpa_efficient.json", "configuration": "704x1280_tp4_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp4_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp4_sp1_v1/captures/qkv_rank0_c1119485a5572ce6.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 27, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_cudnn", "status": "measured", "q_shape": [10560, 10, 128], "k_shape": [66880, 10, 128], "q_len": 10560, "k_len": 63360, "reference_error": {"max_abs": 0.03125, "rmse": 0.0007824147019671921, "relative_l2": 0.0014968411274465698, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.009309301063288977, "rmse": 0.000747603303305806, "relative_l2": 0.0019256235225473648, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 5.761951923370361, "p95_ms": 6.10503351688385, "p99_ms": 6.514282131195069, "mean_ms": 5.820701537132263, "n": 300}, {"repeat": 1, "p50_ms": 5.811536073684692, "p95_ms": 5.982627296447754, "p99_ms": 5.992454252243042, "mean_ms": 5.8201523113250735, "n": 300}, {"repeat": 2, "p50_ms": 5.81715202331543, "p95_ms": 5.930817770957947, "p99_ms": 5.9407776880264285, "mean_ms": 5.832908263206482, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_c1119485a5572ce6_sdpa_cudnn.json", "configuration": "704x1280_tp4_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp4_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp4_sp1_v1/captures/qkv_rank0_c1119485a5572ce6.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 27, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "native_dense", "status": "measured", "q_shape": [10560, 10, 128], "k_shape": [66880, 10, 128], "q_len": 10560, "k_len": 63360, "reference_error": {"max_abs": 0.03125, "rmse": 0.0004752843112156705, "relative_l2": 0.0009092685790144595, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.009103983744871424, "rmse": 0.0007563868454594051, "relative_l2": 0.0019482475469563877, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 5.176367998123169, "p95_ms": 5.717836976051331, "p99_ms": 6.127922549247741, "mean_ms": 5.229626774787903, "n": 300}, {"repeat": 1, "p50_ms": 5.253600120544434, "p95_ms": 5.567286515235901, "p99_ms": 5.724570980072022, "mean_ms": 5.287242442766825, "n": 300}, {"repeat": 2, "p50_ms": 5.23364782333374, "p95_ms": 5.423929595947266, "p99_ms": 5.438211302757264, "mean_ms": 5.238194672266642, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_c1119485a5572ce6_native_dense.json", "configuration": "704x1280_tp4_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp4_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp4_sp1_v1/captures/qkv_rank0_d12872a3fe20c4e4.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 9, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "vllm_fa2", "status": "measured", "q_shape": [10560, 10, 128], "k_shape": [512, 10, 128], "q_len": 10560, "k_len": 512, "reference_error": {"max_abs": 0.0001220703125, "rmse": 1.2637034089433026e-06, "relative_l2": 0.0007130395147746247, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 7.557228920109654e-05, "rmse": 5.397119031052476e-06, "relative_l2": 0.0019442396290101993, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.09689600020647049, "p95_ms": 0.09759999811649323, "p99_ms": 0.09846591725945472, "mean_ms": 0.09704469333092372, "n": 300}, {"repeat": 1, "p50_ms": 0.09571199864149094, "p95_ms": 0.09692800045013428, "p99_ms": 0.09817728251218796, "mean_ms": 0.0962492797523737, "n": 300}, {"repeat": 2, "p50_ms": 0.09571199864149094, "p95_ms": 0.09680159948766232, "p99_ms": 0.09801792129874229, "mean_ms": 0.09615722651282946, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_d12872a3fe20c4e4_vllm_fa2.json", "configuration": "704x1280_tp4_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp4_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp4_sp1_v1/captures/qkv_rank0_d12872a3fe20c4e4.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 9, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "vllm_fa3", "status": "measured", "q_shape": [10560, 10, 128], "k_shape": [512, 10, 128], "q_len": 10560, "k_len": 512, "reference_error": {"max_abs": 0.0001220703125, "rmse": 4.796407111317857e-07, "relative_l2": 0.0002706353227436049, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 7.557228920109654e-05, "rmse": 5.442261561043478e-06, "relative_l2": 0.0019605016190195573, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.07993599772453308, "p95_ms": 0.081216000020504, "p99_ms": 0.0817939169704914, "mean_ms": 0.08022197363277277, "n": 300}, {"repeat": 1, "p50_ms": 0.07996799796819687, "p95_ms": 0.08102399855852127, "p99_ms": 0.08150688216090202, "mean_ms": 0.08056277344624202, "n": 300}, {"repeat": 2, "p50_ms": 0.08022399991750717, "p95_ms": 0.08150560222566128, "p99_ms": 0.0823683213442564, "mean_ms": 0.08077109339336554, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_d12872a3fe20c4e4_vllm_fa3.json", "configuration": "704x1280_tp4_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp4_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp4_sp1_v1/captures/qkv_rank0_d12872a3fe20c4e4.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 9, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_math", "status": "measured", "q_shape": [10560, 10, 128], "k_shape": [512, 10, 128], "q_len": 10560, "k_len": 512, "reference_error": {"max_abs": 0.000244140625, "rmse": 3.363860902135059e-06, "relative_l2": 0.0018980448485403493, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 5.5390732143636295e-05, "rmse": 4.640119119681335e-06, "relative_l2": 0.001671540580058898, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 1.2662240266799927, "p95_ms": 1.2763600468635559, "p99_ms": 1.278823344707489, "mean_ms": 1.2688405366738638, "n": 300}, {"repeat": 1, "p50_ms": 1.2648639678955078, "p95_ms": 1.2679056406021119, "p99_ms": 1.2690575623512268, "mean_ms": 1.2654370141029359, "n": 300}, {"repeat": 2, "p50_ms": 1.264847993850708, "p95_ms": 1.2674624502658844, "p99_ms": 1.2699856066703796, "mean_ms": 1.265306879679362, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_d12872a3fe20c4e4_sdpa_math.json", "configuration": "704x1280_tp4_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp4_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp4_sp1_v1/captures/qkv_rank0_d12872a3fe20c4e4.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 9, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_flash", "status": "measured", "q_shape": [10560, 10, 128], "k_shape": [512, 10, 128], "q_len": 10560, "k_len": 512, "reference_error": {"max_abs": 0.0001220703125, "rmse": 1.2639662685951513e-06, "relative_l2": 0.0007131878322653284, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 7.557228920109654e-05, "rmse": 5.397119031052476e-06, "relative_l2": 0.0019442396290101993, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.10230399668216705, "p95_ms": 0.10313920304179192, "p99_ms": 0.10358463898301125, "mean_ms": 0.1024866134673357, "n": 300}, {"repeat": 1, "p50_ms": 0.10249599814414978, "p95_ms": 0.1034895982593298, "p99_ms": 0.10403648234903812, "mean_ms": 0.10281621309618155, "n": 300}, {"repeat": 2, "p50_ms": 0.10255999863147736, "p95_ms": 0.10355359874665737, "p99_ms": 0.10409632287919521, "mean_ms": 0.10292533290882905, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_d12872a3fe20c4e4_sdpa_flash.json", "configuration": "704x1280_tp4_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp4_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp4_sp1_v1/captures/qkv_rank0_d12872a3fe20c4e4.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 9, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_efficient", "status": "measured", "q_shape": [10560, 10, 128], "k_shape": [512, 10, 128], "q_len": 10560, "k_len": 512, "reference_error": {"max_abs": 0.000244140625, "rmse": 2.227277832828202e-06, "relative_l2": 0.0012567324689865993, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 7.557228920109654e-05, "rmse": 5.886458538697318e-06, "relative_l2": 0.002120517613121644, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.1786240041255951, "p95_ms": 0.1791407935321331, "p99_ms": 0.17958495661616325, "mean_ms": 0.17876906702915826, "n": 300}, {"repeat": 1, "p50_ms": 0.17894400656223297, "p95_ms": 0.17968160435557365, "p99_ms": 0.18028896182775497, "mean_ms": 0.1792780807117621, "n": 300}, {"repeat": 2, "p50_ms": 0.17897599935531616, "p95_ms": 0.17971199750900269, "p99_ms": 0.18003391996026039, "mean_ms": 0.17926730697353682, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_d12872a3fe20c4e4_sdpa_efficient.json", "configuration": "704x1280_tp4_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp4_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp4_sp1_v1/captures/qkv_rank0_d12872a3fe20c4e4.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 9, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_cudnn", "status": "measured", "q_shape": [10560, 10, 128], "k_shape": [512, 10, 128], "q_len": 10560, "k_len": 512, "reference_error": {"max_abs": 0.000244140625, "rmse": 2.2265344384561776e-06, "relative_l2": 0.001256313011732181, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 7.557228920109654e-05, "rmse": 5.886458538697318e-06, "relative_l2": 0.002120517613121644, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.05897599831223488, "p95_ms": 0.05967999994754791, "p99_ms": 0.060101439394056794, "mean_ms": 0.05920266651858886, "n": 300}, {"repeat": 1, "p50_ms": 0.0589120015501976, "p95_ms": 0.059651199728250504, "p99_ms": 0.060067839138209816, "mean_ms": 0.05954463997234901, "n": 300}, {"repeat": 2, "p50_ms": 0.05907199904322624, "p95_ms": 0.05987360142171383, "p99_ms": 0.06096511851996183, "mean_ms": 0.059505919826527434, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_d12872a3fe20c4e4_sdpa_cudnn.json", "configuration": "704x1280_tp4_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp4_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp4_sp1_v1/captures/qkv_rank0_d12872a3fe20c4e4.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 9, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "native_dense", "status": "measured", "q_shape": [10560, 10, 128], "k_shape": [512, 10, 128], "q_len": 10560, "k_len": 512, "reference_error": {"max_abs": 0.0, "rmse": 0.0, "relative_l2": 0.0, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 7.557228920109654e-05, "rmse": 5.452818436748183e-06, "relative_l2": 0.001964304591676919, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.12326399981975555, "p95_ms": 0.12992320582270622, "p99_ms": 0.14522623538970944, "mean_ms": 0.12443103996415933, "n": 300}, {"repeat": 1, "p50_ms": 0.12212799862027168, "p95_ms": 0.1277135968208313, "p99_ms": 0.162835514396428, "mean_ms": 0.12820704005658626, "n": 300}, {"repeat": 2, "p50_ms": 0.12214399874210358, "p95_ms": 0.12714560627937316, "p99_ms": 0.13224223315715783, "mean_ms": 0.12329130661984285, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_d12872a3fe20c4e4_native_dense.json", "configuration": "704x1280_tp4_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp4_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp4_sp1_v1/captures/qkv_rank0_fcd1494e1f8885be.pt", "context": {"phase": "commit", "step": null, "start_frame": 27, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "vllm_fa2", "status": "measured", "q_shape": [10560, 10, 128], "k_shape": [512, 10, 128], "q_len": 10560, "k_len": 512, "reference_error": {"max_abs": 0.000244140625, "rmse": 1.2112066643615484e-06, "relative_l2": 0.0006659406286410848, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.00010103960739320239, "rmse": 5.065093424401204e-06, "relative_l2": 0.0017706614158091568, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.09564799815416336, "p95_ms": 0.09647999703884125, "p99_ms": 0.09696064069867134, "mean_ms": 0.09582858669261138, "n": 300}, {"repeat": 1, "p50_ms": 0.09564799815416336, "p95_ms": 0.09651359729468822, "p99_ms": 0.0969299204647541, "mean_ms": 0.09598559983074666, "n": 300}, {"repeat": 2, "p50_ms": 0.09564799815416336, "p95_ms": 0.09660959802567959, "p99_ms": 0.09731296330690384, "mean_ms": 0.09604138645033042, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_fcd1494e1f8885be_vllm_fa2.json", "configuration": "704x1280_tp4_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp4_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp4_sp1_v1/captures/qkv_rank0_fcd1494e1f8885be.pt", "context": {"phase": "commit", "step": null, "start_frame": 27, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "vllm_fa3", "status": "measured", "q_shape": [10560, 10, 128], "k_shape": [512, 10, 128], "q_len": 10560, "k_len": 512, "reference_error": {"max_abs": 0.000244140625, "rmse": 4.633845710941895e-07, "relative_l2": 0.000254776184491774, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.00010103960739320239, "rmse": 5.088589564138508e-06, "relative_l2": 0.0017788752244336688, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.08025600016117096, "p95_ms": 0.08121760003268719, "p99_ms": 0.08195359818637371, "mean_ms": 0.08044597350060939, "n": 300}, {"repeat": 1, "p50_ms": 0.08022399991750717, "p95_ms": 0.08121760003268719, "p99_ms": 0.08204959891736507, "mean_ms": 0.08087818667292596, "n": 300}, {"repeat": 2, "p50_ms": 0.08022399991750717, "p95_ms": 0.08124800026416779, "p99_ms": 0.08170432366430759, "mean_ms": 0.08072554647922515, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_fcd1494e1f8885be_vllm_fa3.json", "configuration": "704x1280_tp4_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp4_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp4_sp1_v1/captures/qkv_rank0_fcd1494e1f8885be.pt", "context": {"phase": "commit", "step": null, "start_frame": 27, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_math", "status": "measured", "q_shape": [10560, 10, 128], "k_shape": [512, 10, 128], "q_len": 10560, "k_len": 512, "reference_error": {"max_abs": 0.000244140625, "rmse": 2.561058085213004e-06, "relative_l2": 0.0014081103427152058, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.00010103960739320239, "rmse": 4.846869048509495e-06, "relative_l2": 0.001694374277546517, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 1.2649120092391968, "p95_ms": 1.2697391510009766, "p99_ms": 1.2725945341587066, "mean_ms": 1.2658024501800538, "n": 300}, {"repeat": 1, "p50_ms": 1.2643200159072876, "p95_ms": 1.2674608528614044, "p99_ms": 1.2685766661167144, "mean_ms": 1.264805442094803, "n": 300}, {"repeat": 2, "p50_ms": 1.2635040283203125, "p95_ms": 1.266659241914749, "p99_ms": 1.2684170067310334, "mean_ms": 1.2642165335019429, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_fcd1494e1f8885be_sdpa_math.json", "configuration": "704x1280_tp4_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp4_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp4_sp1_v1/captures/qkv_rank0_fcd1494e1f8885be.pt", "context": {"phase": "commit", "step": null, "start_frame": 27, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_flash", "status": "measured", "q_shape": [10560, 10, 128], "k_shape": [512, 10, 128], "q_len": 10560, "k_len": 512, "reference_error": {"max_abs": 0.000244140625, "rmse": 1.2113210897697775e-06, "relative_l2": 0.0006660035415448285, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.00010103960739320239, "rmse": 5.065093424401204e-06, "relative_l2": 0.0017706614158091568, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.10259199887514114, "p95_ms": 0.10345759801566601, "p99_ms": 0.10396832190454006, "mean_ms": 0.1027603197346131, "n": 300}, {"repeat": 1, "p50_ms": 0.10233599692583084, "p95_ms": 0.10310400277376175, "p99_ms": 0.10368319973349571, "mean_ms": 0.10259744008382161, "n": 300}, {"repeat": 2, "p50_ms": 0.10233599692583084, "p95_ms": 0.10300960205495358, "p99_ms": 0.10361631922423839, "mean_ms": 0.10260490668316682, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_fcd1494e1f8885be_sdpa_flash.json", "configuration": "704x1280_tp4_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp4_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp4_sp1_v1/captures/qkv_rank0_fcd1494e1f8885be.pt", "context": {"phase": "commit", "step": null, "start_frame": 27, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_efficient", "status": "measured", "q_shape": [10560, 10, 128], "k_shape": [512, 10, 128], "q_len": 10560, "k_len": 512, "reference_error": {"max_abs": 0.000244140625, "rmse": 2.103862457615787e-06, "relative_l2": 0.0011567369374883315, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.00010103960739320239, "rmse": 5.559465973583726e-06, "relative_l2": 0.0019434847626906989, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.17897599935531616, "p95_ms": 0.17961600422859192, "p99_ms": 0.18022432133555413, "mean_ms": 0.17909600014487903, "n": 300}, {"repeat": 1, "p50_ms": 0.17868800461292267, "p95_ms": 0.17948800325393677, "p99_ms": 0.18019647374749184, "mean_ms": 0.1790307200451692, "n": 300}, {"repeat": 2, "p50_ms": 0.1785919964313507, "p95_ms": 0.17923520132899284, "p99_ms": 0.1797443251311779, "mean_ms": 0.1788058666884899, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_fcd1494e1f8885be_sdpa_efficient.json", "configuration": "704x1280_tp4_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp4_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp4_sp1_v1/captures/qkv_rank0_fcd1494e1f8885be.pt", "context": {"phase": "commit", "step": null, "start_frame": 27, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_cudnn", "status": "measured", "q_shape": [10560, 10, 128], "k_shape": [512, 10, 128], "q_len": 10560, "k_len": 512, "reference_error": {"max_abs": 0.000244140625, "rmse": 2.102459930568543e-06, "relative_l2": 0.0011559658058796557, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.00010103960739320239, "rmse": 5.559464214595815e-06, "relative_l2": 0.0019434841477816015, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.05897599831223488, "p95_ms": 0.05984000116586685, "p99_ms": 0.06032096110284328, "mean_ms": 0.05926463996370634, "n": 300}, {"repeat": 1, "p50_ms": 0.05929600074887276, "p95_ms": 0.06019840016961098, "p99_ms": 0.0611523199826479, "mean_ms": 0.05980064008384943, "n": 300}, {"repeat": 2, "p50_ms": 0.05926400050520897, "p95_ms": 0.06019200012087822, "p99_ms": 0.06067456003278494, "mean_ms": 0.05984778672456741, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_fcd1494e1f8885be_sdpa_cudnn.json", "configuration": "704x1280_tp4_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp4_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_evening_704x1280_tp4_sp1_v1/captures/qkv_rank0_fcd1494e1f8885be.pt", "context": {"phase": "commit", "step": null, "start_frame": 27, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "native_dense", "status": "measured", "q_shape": [10560, 10, 128], "k_shape": [512, 10, 128], "q_len": 10560, "k_len": 512, "reference_error": {"max_abs": 0.0, "rmse": 0.0, "relative_l2": 0.0, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.00010103960739320239, "rmse": 5.097650603668637e-06, "relative_l2": 0.0017820427934672088, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.1231519989669323, "p95_ms": 0.132619196921587, "p99_ms": 0.14065791726112356, "mean_ms": 0.12473994662364324, "n": 300}, {"repeat": 1, "p50_ms": 0.12311999872326851, "p95_ms": 0.13069439679384232, "p99_ms": 0.1428684794902801, "mean_ms": 0.12475338677565256, "n": 300}, {"repeat": 2, "p50_ms": 0.12428800016641617, "p95_ms": 0.1361279934644699, "p99_ms": 0.2228934438526628, "mean_ms": 0.12841546662151815, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_fcd1494e1f8885be_native_dense.json", "configuration": "704x1280_tp4_sp1", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_evening_704x1280_tp4_sp1_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_fixed2_480x832_tp1_sp4/captures/qkv_rank0_13f5e4136c72875f.pt", "context": {"phase": "commit", "step": null, "start_frame": 27, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "vllm_fa2", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [29640, 10, 128], "q_len": 4680, "k_len": 28080, "reference_error": {"max_abs": 0.0625, "rmse": 0.0004278408197932095, "relative_l2": 0.0005383218211689015, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.010101945123846434, "rmse": 0.0009676761183591323, "relative_l2": 0.0016959539687613312, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 1.8893920183181763, "p95_ms": 1.9097647964954376, "p99_ms": 1.915727038383484, "mean_ms": 1.892036370833715, "n": 300}, {"repeat": 1, "p50_ms": 1.889024019241333, "p95_ms": 1.8945663809776305, "p99_ms": 1.904389456510544, "mean_ms": 1.890425173441569, "n": 300}, {"repeat": 2, "p50_ms": 1.904304027557373, "p95_ms": 1.911596816778183, "p99_ms": 1.9137923336029052, "mean_ms": 1.9047224525610607, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_13f5e4136c72875f_vllm_fa2.json", "configuration": "480x832_tp1_sp4", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_fixed2_480x832_tp1_sp4", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_fixed2_480x832_tp1_sp4/captures/qkv_rank0_13f5e4136c72875f.pt", "context": {"phase": "commit", "step": null, "start_frame": 27, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "vllm_fa3", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [29640, 10, 128], "q_len": 4680, "k_len": 28080, "reference_error": {"max_abs": 0.0, "rmse": 0.0, "relative_l2": 0.0, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.010101945123846434, "rmse": 0.0009681409449194487, "relative_l2": 0.0016967686260984276, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 1.005087971687317, "p95_ms": 1.0166479527950287, "p99_ms": 1.0175724720954895, "mean_ms": 0.987574718395869, "n": 300}, {"repeat": 1, "p50_ms": 1.0061440467834473, "p95_ms": 1.0130431711673737, "p99_ms": 1.0173826670646668, "mean_ms": 1.0045351483424505, "n": 300}, {"repeat": 2, "p50_ms": 1.0165919661521912, "p95_ms": 1.209516805410385, "p99_ms": 1.210883880853653, "mean_ms": 1.038167679508527, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_13f5e4136c72875f_vllm_fa3.json", "configuration": "480x832_tp1_sp4", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_fixed2_480x832_tp1_sp4", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_fixed2_480x832_tp1_sp4/captures/qkv_rank0_13f5e4136c72875f.pt", "context": {"phase": "commit", "step": null, "start_frame": 27, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_math", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [29640, 10, 128], "q_len": 4680, "k_len": 28080, "reference_error": {"max_abs": 0.0625, "rmse": 0.0007962888877749233, "relative_l2": 0.0010019139465251164, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.010101945123846434, "rmse": 0.0009455398630386555, "relative_l2": 0.0016571578577981548, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 27.94043254852295, "p95_ms": 28.079129123687743, "p99_ms": 28.108948192596436, "mean_ms": 27.988425540924073, "n": 300}, {"repeat": 1, "p50_ms": 28.061087608337402, "p95_ms": 28.147538471221925, "p99_ms": 28.223288803100587, "mean_ms": 28.052209758758544, "n": 300}, {"repeat": 2, "p50_ms": 28.065776824951172, "p95_ms": 28.262207317352296, "p99_ms": 28.35076105117798, "mean_ms": 28.095546340942384, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_13f5e4136c72875f_sdpa_math.json", "configuration": "480x832_tp1_sp4", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_fixed2_480x832_tp1_sp4", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_fixed2_480x832_tp1_sp4/captures/qkv_rank0_13f5e4136c72875f.pt", "context": {"phase": "commit", "step": null, "start_frame": 27, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_flash", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [29640, 10, 128], "q_len": 4680, "k_len": 28080, "reference_error": {"max_abs": 0.0625, "rmse": 0.00042904017884032254, "relative_l2": 0.0005398308897678007, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.010101945123846434, "rmse": 0.0009676761183591323, "relative_l2": 0.0016959539687613312, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 2.0498719215393066, "p95_ms": 2.070134389400482, "p99_ms": 2.0722172331809996, "mean_ms": 2.0535382397969566, "n": 300}, {"repeat": 1, "p50_ms": 2.049903988838196, "p95_ms": 2.067812716960907, "p99_ms": 2.071168587207794, "mean_ms": 2.0545197892189027, "n": 300}, {"repeat": 2, "p50_ms": 2.049776077270508, "p95_ms": 2.0652480125427246, "p99_ms": 2.069155316352844, "mean_ms": 2.0530315740903218, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_13f5e4136c72875f_sdpa_flash.json", "configuration": "480x832_tp1_sp4", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_fixed2_480x832_tp1_sp4", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_fixed2_480x832_tp1_sp4/captures/qkv_rank0_13f5e4136c72875f.pt", "context": {"phase": "commit", "step": null, "start_frame": 27, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_efficient", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [29640, 10, 128], "q_len": 4680, "k_len": 28080, "reference_error": {"max_abs": 0.0625, "rmse": 0.0006798250225789598, "relative_l2": 0.0008553757082079703, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.010101945123846434, "rmse": 0.0009517787171752579, "relative_l2": 0.001668092104528802, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 4.1100640296936035, "p95_ms": 4.146239876747131, "p99_ms": 4.155557990074158, "mean_ms": 4.11340895652771, "n": 300}, {"repeat": 1, "p50_ms": 4.113023996353149, "p95_ms": 4.146603322029113, "p99_ms": 4.160386414527893, "mean_ms": 4.11531286239624, "n": 300}, {"repeat": 2, "p50_ms": 4.119328022003174, "p95_ms": 4.150382614135742, "p99_ms": 4.161754465103149, "mean_ms": 4.119260792732239, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_13f5e4136c72875f_sdpa_efficient.json", "configuration": "480x832_tp1_sp4", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_fixed2_480x832_tp1_sp4", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_fixed2_480x832_tp1_sp4/captures/qkv_rank0_13f5e4136c72875f.pt", "context": {"phase": "commit", "step": null, "start_frame": 27, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_cudnn", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [29640, 10, 128], "q_len": 4680, "k_len": 28080, "reference_error": {"max_abs": 0.0625, "rmse": 0.0006799944528355498, "relative_l2": 0.0008555888903076417, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.010101945123846434, "rmse": 0.0009516249798776634, "relative_l2": 0.0016678226637778573, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 1.1911680102348328, "p95_ms": 1.218400001525879, "p99_ms": 1.2190723049640655, "mean_ms": 1.1794511957963307, "n": 300}, {"repeat": 1, "p50_ms": 1.1976159811019897, "p95_ms": 1.2027519941329956, "p99_ms": 1.2053833889961243, "mean_ms": 1.1918484266599019, "n": 300}, {"repeat": 2, "p50_ms": 1.2119200229644775, "p95_ms": 1.410471969842911, "p99_ms": 1.4162697112560272, "mean_ms": 1.2543241572380066, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_13f5e4136c72875f_sdpa_cudnn.json", "configuration": "480x832_tp1_sp4", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_fixed2_480x832_tp1_sp4", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_fixed2_480x832_tp1_sp4/captures/qkv_rank0_5d233cbdc47812ff.pt", "context": {"phase": "commit", "step": null, "start_frame": 0, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "vllm_fa2", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [29640, 10, 128], "q_len": 4680, "k_len": 4680, "reference_error": {"max_abs": 0.03125, "rmse": 0.0004155959679669699, "relative_l2": 0.0005516502103045647, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.013453558575000635, "rmse": 0.000855206902521633, "relative_l2": 0.0017184019899498433, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.3303680121898651, "p95_ms": 0.3314880132675171, "p99_ms": 0.3317785531282425, "mean_ms": 0.33115157316128413, "n": 300}, {"repeat": 1, "p50_ms": 0.32979199290275574, "p95_ms": 0.3309119939804077, "p99_ms": 0.331397750377655, "mean_ms": 0.3303185068567594, "n": 300}, {"repeat": 2, "p50_ms": 0.3301759958267212, "p95_ms": 0.3317455857992172, "p99_ms": 0.3321628761291504, "mean_ms": 0.3305845339099566, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_5d233cbdc47812ff_vllm_fa2.json", "configuration": "480x832_tp1_sp4", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_fixed2_480x832_tp1_sp4", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_fixed2_480x832_tp1_sp4/captures/qkv_rank0_5d233cbdc47812ff.pt", "context": {"phase": "commit", "step": null, "start_frame": 0, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "vllm_fa3", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [29640, 10, 128], "q_len": 4680, "k_len": 4680, "reference_error": {"max_abs": 0.0, "rmse": 0.0, "relative_l2": 0.0, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.013453558575000635, "rmse": 0.0008552870221296227, "relative_l2": 0.001718562977534716, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.17228800058364868, "p95_ms": 0.18252800405025482, "p99_ms": 0.18284864649176596, "mean_ms": 0.17354410688082378, "n": 300}, {"repeat": 1, "p50_ms": 0.18539199978113174, "p95_ms": 0.1865919977426529, "p99_ms": 0.18716864213347434, "mean_ms": 0.18593845347563426, "n": 300}, {"repeat": 2, "p50_ms": 0.18332800269126892, "p95_ms": 0.18409599363803864, "p99_ms": 0.18470432564616204, "mean_ms": 0.18279541358351709, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_5d233cbdc47812ff_vllm_fa3.json", "configuration": "480x832_tp1_sp4", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_fixed2_480x832_tp1_sp4", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_fixed2_480x832_tp1_sp4/captures/qkv_rank0_5d233cbdc47812ff.pt", "context": {"phase": "commit", "step": null, "start_frame": 0, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_math", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [29640, 10, 128], "q_len": 4680, "k_len": 4680, "reference_error": {"max_abs": 0.03125, "rmse": 0.0007357542109918866, "relative_l2": 0.0009766191120949496, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.013453558575000635, "rmse": 0.000840414736449667, "relative_l2": 0.0016886794894195205, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 4.804112195968628, "p95_ms": 4.831438612937927, "p99_ms": 4.834692664146424, "mean_ms": 4.807215034166972, "n": 300}, {"repeat": 1, "p50_ms": 4.80350399017334, "p95_ms": 4.807617831230163, "p99_ms": 4.8091387367248535, "mean_ms": 4.80720287322998, "n": 300}, {"repeat": 2, "p50_ms": 4.803647994995117, "p95_ms": 4.808035230636596, "p99_ms": 5.125357646942134, "mean_ms": 4.812995206514994, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_5d233cbdc47812ff_sdpa_math.json", "configuration": "480x832_tp1_sp4", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_fixed2_480x832_tp1_sp4", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_fixed2_480x832_tp1_sp4/captures/qkv_rank0_5d233cbdc47812ff.pt", "context": {"phase": "commit", "step": null, "start_frame": 0, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_flash", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [29640, 10, 128], "q_len": 4680, "k_len": 4680, "reference_error": {"max_abs": 0.03125, "rmse": 0.00041651617644221817, "relative_l2": 0.0005528716687354007, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.013453558575000635, "rmse": 0.000855206902521633, "relative_l2": 0.0017184019899498433, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.3577760010957718, "p95_ms": 0.36013118624687196, "p99_ms": 0.360995854139328, "mean_ms": 0.3579851730664571, "n": 300}, {"repeat": 1, "p50_ms": 0.355679988861084, "p95_ms": 0.35683359801769254, "p99_ms": 0.35769855678081514, "mean_ms": 0.35610645363728205, "n": 300}, {"repeat": 2, "p50_ms": 0.35596799850463867, "p95_ms": 0.3569920063018799, "p99_ms": 0.3572176006436348, "mean_ms": 0.35632650752862294, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_5d233cbdc47812ff_sdpa_flash.json", "configuration": "480x832_tp1_sp4", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_fixed2_480x832_tp1_sp4", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_fixed2_480x832_tp1_sp4/captures/qkv_rank0_5d233cbdc47812ff.pt", "context": {"phase": "commit", "step": null, "start_frame": 0, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_efficient", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [29640, 10, 128], "q_len": 4680, "k_len": 4680, "reference_error": {"max_abs": 0.03125, "rmse": 0.0005937389373871991, "relative_l2": 0.0007881120966546252, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.013453558575000635, "rmse": 0.0008510924395452369, "relative_l2": 0.0017101346322550366, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.6989759802818298, "p95_ms": 0.7029535830020904, "p99_ms": 0.7052902460098267, "mean_ms": 0.6995801625649134, "n": 300}, {"repeat": 1, "p50_ms": 0.6987999975681305, "p95_ms": 0.7039087742567063, "p99_ms": 0.7053481811285018, "mean_ms": 0.6999641601244608, "n": 300}, {"repeat": 2, "p50_ms": 0.6987839937210083, "p95_ms": 0.7031520247459412, "p99_ms": 0.7059891164302826, "mean_ms": 0.6999466653664906, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_5d233cbdc47812ff_sdpa_efficient.json", "configuration": "480x832_tp1_sp4", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_fixed2_480x832_tp1_sp4", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_fixed2_480x832_tp1_sp4/captures/qkv_rank0_5d233cbdc47812ff.pt", "context": {"phase": "commit", "step": null, "start_frame": 0, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_cudnn", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [29640, 10, 128], "q_len": 4680, "k_len": 4680, "reference_error": {"max_abs": 0.03125, "rmse": 0.0005926419217863406, "relative_l2": 0.0007866559494983345, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.013453558575000635, "rmse": 0.000851012888406189, "relative_l2": 0.0017099747869179142, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.19657599925994873, "p95_ms": 0.20291200280189514, "p99_ms": 0.20339199900627136, "mean_ms": 0.1991687465707461, "n": 300}, {"repeat": 1, "p50_ms": 0.20235200226306915, "p95_ms": 0.20326559841632844, "p99_ms": 0.20381151482462884, "mean_ms": 0.20262581318616868, "n": 300}, {"repeat": 2, "p50_ms": 0.2011360004544258, "p95_ms": 0.20726560577750205, "p99_ms": 0.208226238489151, "mean_ms": 0.20216597288846969, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_5d233cbdc47812ff_sdpa_cudnn.json", "configuration": "480x832_tp1_sp4", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_fixed2_480x832_tp1_sp4", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_fixed2_480x832_tp1_sp4/captures/qkv_rank0_6b4ac7dbc60c0c5a.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 0, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "vllm_fa2", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [29640, 10, 128], "q_len": 4680, "k_len": 4680, "reference_error": {"max_abs": 0.03125, "rmse": 0.0005266509856320316, "relative_l2": 0.0009067576358098154, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.006950112781240136, "rmse": 0.0007533224446526795, "relative_l2": 0.0019169673843472507, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.33158400654792786, "p95_ms": 0.3337920010089874, "p99_ms": 0.33423998951911926, "mean_ms": 0.3317829332749049, "n": 300}, {"repeat": 1, "p50_ms": 0.3290559947490692, "p95_ms": 0.3301759958267212, "p99_ms": 0.33053279101848604, "mean_ms": 0.32956159859895706, "n": 300}, {"repeat": 2, "p50_ms": 0.3303360044956207, "p95_ms": 0.3320319950580597, "p99_ms": 0.3328963142633438, "mean_ms": 0.3307899730404218, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_6b4ac7dbc60c0c5a_vllm_fa2.json", "configuration": "480x832_tp1_sp4", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_fixed2_480x832_tp1_sp4", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_fixed2_480x832_tp1_sp4/captures/qkv_rank0_6b4ac7dbc60c0c5a.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 0, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "vllm_fa3", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [29640, 10, 128], "q_len": 4680, "k_len": 4680, "reference_error": {"max_abs": 0.0, "rmse": 0.0, "relative_l2": 0.0, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.006950112781240136, "rmse": 0.0007627068873890609, "relative_l2": 0.0019408478232929027, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.17531199753284454, "p95_ms": 0.1865600049495697, "p99_ms": 0.18716960206627845, "mean_ms": 0.17740330666303636, "n": 300}, {"repeat": 1, "p50_ms": 0.18460799753665924, "p95_ms": 0.1876160055398941, "p99_ms": 0.18822400271892548, "mean_ms": 0.1852278400460879, "n": 300}, {"repeat": 2, "p50_ms": 0.18833599984645844, "p95_ms": 0.18902400135993958, "p99_ms": 0.1894102442264557, "mean_ms": 0.1887447464466095, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_6b4ac7dbc60c0c5a_vllm_fa3.json", "configuration": "480x832_tp1_sp4", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_fixed2_480x832_tp1_sp4", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_fixed2_480x832_tp1_sp4/captures/qkv_rank0_6b4ac7dbc60c0c5a.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 0, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_math", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [29640, 10, 128], "q_len": 4680, "k_len": 4680, "reference_error": {"max_abs": 0.03125, "rmse": 0.0008676879616324845, "relative_l2": 0.0014939356541150204, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.006950112781240136, "rmse": 0.0006602534478209683, "relative_l2": 0.0016801362203659833, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 4.803215980529785, "p95_ms": 4.831083059310913, "p99_ms": 4.833890295028686, "mean_ms": 4.80816073735555, "n": 300}, {"repeat": 1, "p50_ms": 4.802448034286499, "p95_ms": 4.807876634597778, "p99_ms": 4.809965863227844, "mean_ms": 4.803691636721293, "n": 300}, {"repeat": 2, "p50_ms": 4.805056095123291, "p95_ms": 4.830888152122498, "p99_ms": 4.8342099332809445, "mean_ms": 4.811474765141805, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_6b4ac7dbc60c0c5a_sdpa_math.json", "configuration": "480x832_tp1_sp4", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_fixed2_480x832_tp1_sp4", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_fixed2_480x832_tp1_sp4/captures/qkv_rank0_6b4ac7dbc60c0c5a.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 0, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_flash", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [29640, 10, 128], "q_len": 4680, "k_len": 4680, "reference_error": {"max_abs": 0.03125, "rmse": 0.0005271479915826146, "relative_l2": 0.0009076133523147284, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.006950112781240136, "rmse": 0.0007533268077478835, "relative_l2": 0.0019169784870447213, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.35702401399612427, "p95_ms": 0.35843199491500854, "p99_ms": 0.35891263127326967, "mean_ms": 0.35715306709210076, "n": 300}, {"repeat": 1, "p50_ms": 0.3563520014286041, "p95_ms": 0.35734720081090926, "p99_ms": 0.35773088455200197, "mean_ms": 0.3566590934991837, "n": 300}, {"repeat": 2, "p50_ms": 0.35601601004600525, "p95_ms": 0.3570591926574707, "p99_ms": 0.35756863594055177, "mean_ms": 0.35618282626072567, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_6b4ac7dbc60c0c5a_sdpa_flash.json", "configuration": "480x832_tp1_sp4", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_fixed2_480x832_tp1_sp4", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_fixed2_480x832_tp1_sp4/captures/qkv_rank0_6b4ac7dbc60c0c5a.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 0, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_efficient", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [29640, 10, 128], "q_len": 4680, "k_len": 4680, "reference_error": {"max_abs": 0.03125, "rmse": 0.0007438730221660497, "relative_l2": 0.0012807581516485914, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.006950112781240136, "rmse": 0.0007546674581946931, "relative_l2": 0.001920390018452802, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.6982719898223877, "p95_ms": 0.7018256098031997, "p99_ms": 0.7029763084650039, "mean_ms": 0.6988095986843109, "n": 300}, {"repeat": 1, "p50_ms": 0.6998880207538605, "p95_ms": 0.7070848196744919, "p99_ms": 0.7094726407527924, "mean_ms": 0.701045333345731, "n": 300}, {"repeat": 2, "p50_ms": 0.7014240026473999, "p95_ms": 0.708191990852356, "p99_ms": 0.7103369891643524, "mean_ms": 0.7023765327533086, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_6b4ac7dbc60c0c5a_sdpa_efficient.json", "configuration": "480x832_tp1_sp4", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_fixed2_480x832_tp1_sp4", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_fixed2_480x832_tp1_sp4/captures/qkv_rank0_6b4ac7dbc60c0c5a.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 0, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_cudnn", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [29640, 10, 128], "q_len": 4680, "k_len": 4680, "reference_error": {"max_abs": 0.03125, "rmse": 0.0007431919192459065, "relative_l2": 0.0012795854674792623, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.006950112781240136, "rmse": 0.0007544565760142951, "relative_l2": 0.0019198533899949183, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.19763199985027313, "p95_ms": 0.2100175976753235, "p99_ms": 0.2119702446460724, "mean_ms": 0.20030432025591532, "n": 300}, {"repeat": 1, "p50_ms": 0.208576001226902, "p95_ms": 0.21235519275069237, "p99_ms": 0.2132486395537853, "mean_ms": 0.20794186676541965, "n": 300}, {"repeat": 2, "p50_ms": 0.20243200659751892, "p95_ms": 0.204225604981184, "p99_ms": 0.20995327681303025, "mean_ms": 0.20315274700522423, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_6b4ac7dbc60c0c5a_sdpa_cudnn.json", "configuration": "480x832_tp1_sp4", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_fixed2_480x832_tp1_sp4", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_fixed2_480x832_tp1_sp4/captures/qkv_rank0_c6a5fcb1e5f12d0d.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 9, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "vllm_fa2", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [29640, 10, 128], "q_len": 4680, "k_len": 18720, "reference_error": {"max_abs": 0.03125, "rmse": 0.0005655613731987039, "relative_l2": 0.0010091701042483979, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.008430754864764634, "rmse": 0.0008224519263401907, "relative_l2": 0.0019414505560311024, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 1.266752004623413, "p95_ms": 1.2721008360385895, "p99_ms": 1.2740243422985076, "mean_ms": 1.267445121606191, "n": 300}, {"repeat": 1, "p50_ms": 1.2678719758987427, "p95_ms": 1.2744031965732574, "p99_ms": 1.275618289709091, "mean_ms": 1.2700544019540152, "n": 300}, {"repeat": 2, "p50_ms": 1.2793599963188171, "p95_ms": 1.2986944496631623, "p99_ms": 1.3006118786334993, "mean_ms": 1.2829703470071157, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_c6a5fcb1e5f12d0d_vllm_fa2.json", "configuration": "480x832_tp1_sp4", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_fixed2_480x832_tp1_sp4", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_fixed2_480x832_tp1_sp4/captures/qkv_rank0_c6a5fcb1e5f12d0d.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 9, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "vllm_fa3", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [29640, 10, 128], "q_len": 4680, "k_len": 18720, "reference_error": {"max_abs": 0.0, "rmse": 0.0, "relative_l2": 0.0, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.008430754864764634, "rmse": 0.0008316797479339394, "relative_l2": 0.001963233421132848, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.6814880073070526, "p95_ms": 0.6882240176200867, "p99_ms": 0.6887999773025513, "mean_ms": 0.6637857075532277, "n": 300}, {"repeat": 1, "p50_ms": 0.6806559860706329, "p95_ms": 0.6860160231590271, "p99_ms": 0.6871407943964004, "mean_ms": 0.6781361055374145, "n": 300}, {"repeat": 2, "p50_ms": 0.6891840100288391, "p95_ms": 0.6973247975111008, "p99_ms": 0.7001289784908294, "mean_ms": 0.6854442660013834, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_c6a5fcb1e5f12d0d_vllm_fa3.json", "configuration": "480x832_tp1_sp4", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_fixed2_480x832_tp1_sp4", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_fixed2_480x832_tp1_sp4/captures/qkv_rank0_c6a5fcb1e5f12d0d.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 9, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_math", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [29640, 10, 128], "q_len": 4680, "k_len": 18720, "reference_error": {"max_abs": 0.03125, "rmse": 0.0009919727936492876, "relative_l2": 0.001770045365575048, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.007760694409859159, "rmse": 0.0007096702226792513, "relative_l2": 0.0016752221063550086, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 18.71345615386963, "p95_ms": 18.806691455841065, "p99_ms": 18.81445592880249, "mean_ms": 18.733917452494303, "n": 300}, {"repeat": 1, "p50_ms": 18.714608192443848, "p95_ms": 18.80666227340698, "p99_ms": 18.83568021774292, "mean_ms": 18.741550744374592, "n": 300}, {"repeat": 2, "p50_ms": 18.716639518737793, "p95_ms": 18.81005516052246, "p99_ms": 18.89844720840454, "mean_ms": 18.748349049886066, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_c6a5fcb1e5f12d0d_sdpa_math.json", "configuration": "480x832_tp1_sp4", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_fixed2_480x832_tp1_sp4", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_fixed2_480x832_tp1_sp4/captures/qkv_rank0_c6a5fcb1e5f12d0d.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 9, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_flash", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [29640, 10, 128], "q_len": 4680, "k_len": 18720, "reference_error": {"max_abs": 0.03125, "rmse": 0.0005659922767473519, "relative_l2": 0.001009938995123407, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.008430754864764634, "rmse": 0.0008226065287387949, "relative_l2": 0.0019418155049151903, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 1.3740959763526917, "p95_ms": 1.38226078748703, "p99_ms": 1.3859283673763274, "mean_ms": 1.3752785034974415, "n": 300}, {"repeat": 1, "p50_ms": 1.3731839656829834, "p95_ms": 1.3775999546051025, "p99_ms": 1.3791104185581207, "mean_ms": 1.374071677128474, "n": 300}, {"repeat": 2, "p50_ms": 1.3735359907150269, "p95_ms": 1.3783696174621582, "p99_ms": 1.379840656518936, "mean_ms": 1.3745849557717642, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_c6a5fcb1e5f12d0d_sdpa_flash.json", "configuration": "480x832_tp1_sp4", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_fixed2_480x832_tp1_sp4", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_fixed2_480x832_tp1_sp4/captures/qkv_rank0_c6a5fcb1e5f12d0d.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 9, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_efficient", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [29640, 10, 128], "q_len": 4680, "k_len": 18720, "reference_error": {"max_abs": 0.03125, "rmse": 0.0007752434725243829, "relative_l2": 0.0013833203133383903, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.00928316173610133, "rmse": 0.000829930267228196, "relative_l2": 0.001959103659647523, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 2.7462559938430786, "p95_ms": 2.7719167947769163, "p99_ms": 2.782939214706421, "mean_ms": 2.749940270582835, "n": 300}, {"repeat": 1, "p50_ms": 2.7452319860458374, "p95_ms": 2.76771993637085, "p99_ms": 2.77408579826355, "mean_ms": 2.7487993621826172, "n": 300}, {"repeat": 2, "p50_ms": 2.7469760179519653, "p95_ms": 2.7739328503608705, "p99_ms": 2.7863110876083375, "mean_ms": 2.7512875763575235, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_c6a5fcb1e5f12d0d_sdpa_efficient.json", "configuration": "480x832_tp1_sp4", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_fixed2_480x832_tp1_sp4", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_fixed2_480x832_tp1_sp4/captures/qkv_rank0_c6a5fcb1e5f12d0d.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 9, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_cudnn", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [29640, 10, 128], "q_len": 4680, "k_len": 18720, "reference_error": {"max_abs": 0.03125, "rmse": 0.0007742347555068339, "relative_l2": 0.0013815203901010595, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.00928316173610133, "rmse": 0.000830095690624396, "relative_l2": 0.0019594941521909135, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.8075999915599823, "p95_ms": 0.8273615747690201, "p99_ms": 0.8280959725379944, "mean_ms": 0.801125546892484, "n": 300}, {"repeat": 1, "p50_ms": 0.8016639947891235, "p95_ms": 0.8198720216751099, "p99_ms": 0.8288995063304901, "mean_ms": 0.8036073611179988, "n": 300}, {"repeat": 2, "p50_ms": 0.8125599920749664, "p95_ms": 0.8282240033149719, "p99_ms": 0.8286086165904999, "mean_ms": 0.8124308244387308, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_c6a5fcb1e5f12d0d_sdpa_cudnn.json", "configuration": "480x832_tp1_sp4", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_fixed2_480x832_tp1_sp4", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_fixed2_480x832_tp1_sp4/captures/qkv_rank0_cf2e5e4d322f9347.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 27, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "vllm_fa2", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [29640, 10, 128], "q_len": 4680, "k_len": 28080, "reference_error": {"max_abs": 0.03125, "rmse": 0.0005552815252655393, "relative_l2": 0.000991735346034173, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.008820491288137777, "rmse": 0.0008120587250742453, "relative_l2": 0.0019695529708380166, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 1.890112042427063, "p95_ms": 1.8972592294216155, "p99_ms": 1.9010624504089355, "mean_ms": 1.8909125324090321, "n": 300}, {"repeat": 1, "p50_ms": 1.903056025505066, "p95_ms": 2.011948752403259, "p99_ms": 2.0135704851150513, "mean_ms": 1.9235284260908763, "n": 300}, {"repeat": 2, "p50_ms": 1.8991999626159668, "p95_ms": 1.9082880020141602, "p99_ms": 1.9124490082263947, "mean_ms": 1.8994260295232137, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_cf2e5e4d322f9347_vllm_fa2.json", "configuration": "480x832_tp1_sp4", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_fixed2_480x832_tp1_sp4", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_fixed2_480x832_tp1_sp4/captures/qkv_rank0_cf2e5e4d322f9347.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 27, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "vllm_fa3", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [29640, 10, 128], "q_len": 4680, "k_len": 28080, "reference_error": {"max_abs": 0.0, "rmse": 0.0, "relative_l2": 0.0, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.008820491288137777, "rmse": 0.0008167053351597663, "relative_l2": 0.0019808227773380664, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 1.0442399978637695, "p95_ms": 1.0569663882255553, "p99_ms": 1.0584355032444, "mean_ms": 1.0208552529414494, "n": 300}, {"repeat": 1, "p50_ms": 1.027616024017334, "p95_ms": 1.0496144056320191, "p99_ms": 1.0522614288330079, "mean_ms": 1.0313667201995849, "n": 300}, {"repeat": 2, "p50_ms": 1.017471969127655, "p95_ms": 1.0964559495449067, "p99_ms": 1.1027596509456634, "mean_ms": 1.0274323225021362, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_cf2e5e4d322f9347_vllm_fa3.json", "configuration": "480x832_tp1_sp4", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_fixed2_480x832_tp1_sp4", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_fixed2_480x832_tp1_sp4/captures/qkv_rank0_cf2e5e4d322f9347.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 27, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_math", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [29640, 10, 128], "q_len": 4680, "k_len": 28080, "reference_error": {"max_abs": 0.03125, "rmse": 0.0009780852179610475, "relative_l2": 0.0017468646766550496, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.007724527303786921, "rmse": 0.0006858574812207812, "relative_l2": 0.0016634666902770664, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 28.065760612487793, "p95_ms": 28.284465980529784, "p99_ms": 28.89194990158081, "mean_ms": 28.09598274230957, "n": 300}, {"repeat": 1, "p50_ms": 28.067520141601562, "p95_ms": 28.266581439971922, "p99_ms": 28.332067718505858, "mean_ms": 28.089946212768556, "n": 300}, {"repeat": 2, "p50_ms": 28.06833553314209, "p95_ms": 28.278346824645997, "p99_ms": 28.341850357055662, "mean_ms": 28.09684181213379, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_cf2e5e4d322f9347_sdpa_math.json", "configuration": "480x832_tp1_sp4", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_fixed2_480x832_tp1_sp4", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_fixed2_480x832_tp1_sp4/captures/qkv_rank0_cf2e5e4d322f9347.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 27, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_flash", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [29640, 10, 128], "q_len": 4680, "k_len": 28080, "reference_error": {"max_abs": 0.03125, "rmse": 0.0005560456884091784, "relative_l2": 0.000993100144906819, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.008820491288137777, "rmse": 0.0008120515445332757, "relative_l2": 0.00196953555528005, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 2.0354559421539307, "p95_ms": 2.0670767188072205, "p99_ms": 2.0701493787765504, "mean_ms": 2.043961173693339, "n": 300}, {"repeat": 1, "p50_ms": 2.0701119899749756, "p95_ms": 2.148139226436615, "p99_ms": 2.1509452748298643, "mean_ms": 2.076390724182129, "n": 300}, {"repeat": 2, "p50_ms": 2.0354719161987305, "p95_ms": 2.0697680830955507, "p99_ms": 2.0722886753082275, "mean_ms": 2.045490026473999, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_cf2e5e4d322f9347_sdpa_flash.json", "configuration": "480x832_tp1_sp4", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_fixed2_480x832_tp1_sp4", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_fixed2_480x832_tp1_sp4/captures/qkv_rank0_cf2e5e4d322f9347.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 27, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_efficient", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [29640, 10, 128], "q_len": 4680, "k_len": 28080, "reference_error": {"max_abs": 0.03125, "rmse": 0.0007705672939716462, "relative_l2": 0.0013762367144560445, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.008820491288137777, "rmse": 0.000811379428513884, "relative_l2": 0.001967905416889973, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 4.065023899078369, "p95_ms": 4.158907008171082, "p99_ms": 4.179219422340393, "mean_ms": 4.080093245506287, "n": 300}, {"repeat": 1, "p50_ms": 4.065439939498901, "p95_ms": 4.1030402183532715, "p99_ms": 4.106600174903869, "mean_ms": 4.072406608263652, "n": 300}, {"repeat": 2, "p50_ms": 4.064064025878906, "p95_ms": 4.10376808643341, "p99_ms": 4.114093074798584, "mean_ms": 4.072324585914612, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_cf2e5e4d322f9347_sdpa_efficient.json", "configuration": "480x832_tp1_sp4", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_fixed2_480x832_tp1_sp4", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_fixed2_480x832_tp1_sp4/captures/qkv_rank0_cf2e5e4d322f9347.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 27, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_cudnn", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [29640, 10, 128], "q_len": 4680, "k_len": 28080, "reference_error": {"max_abs": 0.03125, "rmse": 0.0007701305226022741, "relative_l2": 0.0013754566387909396, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.008820491288137777, "rmse": 0.0008098227142425212, "relative_l2": 0.001964129789434424, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 1.2022079825401306, "p95_ms": 1.2313936233520508, "p99_ms": 1.2321609330177308, "mean_ms": 1.1977302372455596, "n": 300}, {"repeat": 1, "p50_ms": 1.1984959840774536, "p95_ms": 1.206499171257019, "p99_ms": 1.2122246468067168, "mean_ms": 1.1978636781374614, "n": 300}, {"repeat": 2, "p50_ms": 1.2458720207214355, "p95_ms": 1.404097604751587, "p99_ms": 1.4064979672431945, "mean_ms": 1.2559374944368997, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_cf2e5e4d322f9347_sdpa_cudnn.json", "configuration": "480x832_tp1_sp4", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_fixed2_480x832_tp1_sp4", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_fixed2_480x832_tp1_sp4/captures/qkv_rank0_df3a5d6865cf4937.pt", "context": {"phase": "commit", "step": null, "start_frame": 9, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "vllm_fa2", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [29640, 10, 128], "q_len": 4680, "k_len": 18720, "reference_error": {"max_abs": 0.0625, "rmse": 0.0004193072287393182, "relative_l2": 0.0005270094373548114, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.012678517544874701, "rmse": 0.0009683098051771197, "relative_l2": 0.0016927232205301553, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 1.2678080201148987, "p95_ms": 1.274399995803833, "p99_ms": 1.2759366476535796, "mean_ms": 1.2685309890906016, "n": 300}, {"repeat": 1, "p50_ms": 1.2688319683074951, "p95_ms": 1.2746928036212921, "p99_ms": 1.2757136130332947, "mean_ms": 1.2697394120693206, "n": 300}, {"repeat": 2, "p50_ms": 1.275712013244629, "p95_ms": 1.2857888400554658, "p99_ms": 1.2896982181072234, "mean_ms": 1.2766429853439332, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_df3a5d6865cf4937_vllm_fa2.json", "configuration": "480x832_tp1_sp4", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_fixed2_480x832_tp1_sp4", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_fixed2_480x832_tp1_sp4/captures/qkv_rank0_df3a5d6865cf4937.pt", "context": {"phase": "commit", "step": null, "start_frame": 9, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "vllm_fa3", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [29640, 10, 128], "q_len": 4680, "k_len": 18720, "reference_error": {"max_abs": 0.0, "rmse": 0.0, "relative_l2": 0.0, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.012678517544874701, "rmse": 0.0009692782411752042, "relative_l2": 0.001694416164351224, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.6747680008411407, "p95_ms": 0.6786351770162582, "p99_ms": 0.6812182414531708, "mean_ms": 0.6630681578318278, "n": 300}, {"repeat": 1, "p50_ms": 0.6765120029449463, "p95_ms": 0.6811839938163757, "p99_ms": 0.6827203255891799, "mean_ms": 0.6715181865294775, "n": 300}, {"repeat": 2, "p50_ms": 0.6774399876594543, "p95_ms": 0.6855040192604065, "p99_ms": 0.6869769752025604, "mean_ms": 0.6717861334482829, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_df3a5d6865cf4937_vllm_fa3.json", "configuration": "480x832_tp1_sp4", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_fixed2_480x832_tp1_sp4", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_fixed2_480x832_tp1_sp4/captures/qkv_rank0_df3a5d6865cf4937.pt", "context": {"phase": "commit", "step": null, "start_frame": 9, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_math", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [29640, 10, 128], "q_len": 4680, "k_len": 18720, "reference_error": {"max_abs": 0.0625, "rmse": 0.0007898468096038272, "relative_l2": 0.000992724890475454, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.012678517544874701, "rmse": 0.00094767651387136, "relative_l2": 0.001656653719712869, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 18.714303970336914, "p95_ms": 18.80870714187622, "p99_ms": 18.932186946868896, "mean_ms": 18.739232457478842, "n": 300}, {"repeat": 1, "p50_ms": 18.71457576751709, "p95_ms": 18.81114263534546, "p99_ms": 18.93754913330078, "mean_ms": 18.7433131980896, "n": 300}, {"repeat": 2, "p50_ms": 18.714863777160645, "p95_ms": 18.82592487335205, "p99_ms": 18.927104415893556, "mean_ms": 18.743779188791912, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_df3a5d6865cf4937_sdpa_math.json", "configuration": "480x832_tp1_sp4", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_fixed2_480x832_tp1_sp4", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_fixed2_480x832_tp1_sp4/captures/qkv_rank0_df3a5d6865cf4937.pt", "context": {"phase": "commit", "step": null, "start_frame": 9, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_flash", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [29640, 10, 128], "q_len": 4680, "k_len": 18720, "reference_error": {"max_abs": 0.0625, "rmse": 0.0004200155646248864, "relative_l2": 0.0005278997146286704, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.012678517544874701, "rmse": 0.0009683237781535962, "relative_l2": 0.0016927476469912064, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 1.3725600242614746, "p95_ms": 1.375339150428772, "p99_ms": 1.3786566698551177, "mean_ms": 1.372776850859324, "n": 300}, {"repeat": 1, "p50_ms": 1.3737599849700928, "p95_ms": 1.3781472265720367, "p99_ms": 1.3790412831306458, "mean_ms": 1.374523415962855, "n": 300}, {"repeat": 2, "p50_ms": 1.3736000061035156, "p95_ms": 1.3786287665367127, "p99_ms": 1.379809229373932, "mean_ms": 1.3748777544498443, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_df3a5d6865cf4937_sdpa_flash.json", "configuration": "480x832_tp1_sp4", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_fixed2_480x832_tp1_sp4", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_fixed2_480x832_tp1_sp4/captures/qkv_rank0_df3a5d6865cf4937.pt", "context": {"phase": "commit", "step": null, "start_frame": 9, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_efficient", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [29640, 10, 128], "q_len": 4680, "k_len": 18720, "reference_error": {"max_abs": 0.0625, "rmse": 0.0006686263996730479, "relative_l2": 0.0008403681084910054, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.012678517544874701, "rmse": 0.0009547840566502768, "relative_l2": 0.0016690785682876319, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 2.7432479858398438, "p95_ms": 2.766756808757782, "p99_ms": 2.7710460424423218, "mean_ms": 2.746646188100179, "n": 300}, {"repeat": 1, "p50_ms": 2.7399359941482544, "p95_ms": 2.7644495368003845, "p99_ms": 2.7699910473823546, "mean_ms": 2.745941870212555, "n": 300}, {"repeat": 2, "p50_ms": 2.7418559789657593, "p95_ms": 2.7683616280555725, "p99_ms": 2.7709173727035523, "mean_ms": 2.74701376358668, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_df3a5d6865cf4937_sdpa_efficient.json", "configuration": "480x832_tp1_sp4", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_fixed2_480x832_tp1_sp4", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_fixed2_480x832_tp1_sp4/captures/qkv_rank0_df3a5d6865cf4937.pt", "context": {"phase": "commit", "step": null, "start_frame": 9, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_cudnn", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [29640, 10, 128], "q_len": 4680, "k_len": 18720, "reference_error": {"max_abs": 0.0625, "rmse": 0.000667980162633798, "relative_l2": 0.0008395558806181951, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.012678517544874701, "rmse": 0.0009547032318614938, "relative_l2": 0.0016689372767340032, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.8065919876098633, "p95_ms": 0.8277792125940323, "p99_ms": 0.8281651079654694, "mean_ms": 0.7969219183921814, "n": 300}, {"repeat": 1, "p50_ms": 0.8119519948959351, "p95_ms": 0.8303375750780105, "p99_ms": 0.8350835013389587, "mean_ms": 0.8079220269123714, "n": 300}, {"repeat": 2, "p50_ms": 0.802480012178421, "p95_ms": 0.8197455912828445, "p99_ms": 0.8202889627218246, "mean_ms": 0.7995682136217753, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_df3a5d6865cf4937_sdpa_cudnn.json", "configuration": "480x832_tp1_sp4", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_fixed2_480x832_tp1_sp4", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_retry_480x832_tp1_sp2_v1/captures/qkv_rank0_0c1b0b9bae2b8b48.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 0, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "vllm_fa2", "status": "measured", "q_shape": [4680, 20, 128], "k_shape": [512, 20, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.00048828125, "rmse": 2.1010678770034274e-06, "relative_l2": 0.0002564703519247502, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.0006142374604239942, "rmse": 1.9996831763657214e-05, "relative_l2": 0.0020143029066056256, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.08329600095748901, "p95_ms": 0.08416320011019707, "p99_ms": 0.0846745565533638, "mean_ms": 0.08351925360659758, "n": 300}, {"repeat": 1, "p50_ms": 0.08329600095748901, "p95_ms": 0.08428800106048584, "p99_ms": 0.08458112321794033, "mean_ms": 0.08401162691414356, "n": 300}, {"repeat": 2, "p50_ms": 0.08323200047016144, "p95_ms": 0.08416000008583069, "p99_ms": 0.08492831848561763, "mean_ms": 0.08364917330443859, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_0c1b0b9bae2b8b48_vllm_fa2.json", "configuration": "480x832_tp1_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_retry_480x832_tp1_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_retry_480x832_tp1_sp2_v1/captures/qkv_rank0_0c1b0b9bae2b8b48.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 0, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "vllm_fa3", "status": "measured", "q_shape": [4680, 20, 128], "k_shape": [512, 20, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.00048828125, "rmse": 8.032271870925259e-07, "relative_l2": 9.804726520446975e-05, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.0006142374604239942, "rmse": 2.0004182011344022e-05, "relative_l2": 0.0020150433051574963, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.07276800274848938, "p95_ms": 0.07379360310733318, "p99_ms": 0.07417663857340813, "mean_ms": 0.07297482679287592, "n": 300}, {"repeat": 1, "p50_ms": 0.07283200323581696, "p95_ms": 0.07366400212049484, "p99_ms": 0.07441088028252124, "mean_ms": 0.07338442705571652, "n": 300}, {"repeat": 2, "p50_ms": 0.07273600250482559, "p95_ms": 0.07366560213267803, "p99_ms": 0.07430399954319, "mean_ms": 0.07314112027486165, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_0c1b0b9bae2b8b48_vllm_fa3.json", "configuration": "480x832_tp1_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_retry_480x832_tp1_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_retry_480x832_tp1_sp2_v1/captures/qkv_rank0_0c1b0b9bae2b8b48.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 0, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_math", "status": "measured", "q_shape": [4680, 20, 128], "k_shape": [512, 20, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.001953125, "rmse": 1.1329499402960049e-05, "relative_l2": 0.0013829542257114194, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.0004727375432675984, "rmse": 1.7092660525431627e-05, "relative_l2": 0.001721762536432087, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 1.1251360177993774, "p95_ms": 1.129220759868622, "p99_ms": 1.1308531701564788, "mean_ms": 1.125240426858266, "n": 300}, {"repeat": 1, "p50_ms": 1.1254559755325317, "p95_ms": 1.129248023033142, "p99_ms": 1.131204116344452, "mean_ms": 1.1259274673461914, "n": 300}, {"repeat": 2, "p50_ms": 1.1252480149269104, "p95_ms": 1.1297935783863067, "p99_ms": 1.131266531944275, "mean_ms": 1.1259013338883719, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_0c1b0b9bae2b8b48_sdpa_math.json", "configuration": "480x832_tp1_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_retry_480x832_tp1_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_retry_480x832_tp1_sp2_v1/captures/qkv_rank0_0c1b0b9bae2b8b48.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 0, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_flash", "status": "measured", "q_shape": [4680, 20, 128], "k_shape": [512, 20, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.00048828125, "rmse": 2.10120932880474e-06, "relative_l2": 0.00025648761847461284, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.0006142374604239942, "rmse": 1.9996831763657214e-05, "relative_l2": 0.0020143029066056256, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.08895999938249588, "p95_ms": 0.08985599875450134, "p99_ms": 0.09046496331691742, "mean_ms": 0.08909290653963883, "n": 300}, {"repeat": 1, "p50_ms": 0.0888959988951683, "p95_ms": 0.08988799899816513, "p99_ms": 0.09033728241920472, "mean_ms": 0.08929429305096467, "n": 300}, {"repeat": 2, "p50_ms": 0.08908800035715103, "p95_ms": 0.09005120024085045, "p99_ms": 0.09049728356301784, "mean_ms": 0.08940352015197277, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_0c1b0b9bae2b8b48_sdpa_flash.json", "configuration": "480x832_tp1_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_retry_480x832_tp1_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_retry_480x832_tp1_sp2_v1/captures/qkv_rank0_0c1b0b9bae2b8b48.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 0, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_efficient", "status": "measured", "q_shape": [4680, 20, 128], "k_shape": [512, 20, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.0009765625, "rmse": 3.7657847397106228e-06, "relative_l2": 0.0004596767900920422, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.0006142374604239942, "rmse": 2.0105374323793734e-05, "relative_l2": 0.0020252365183375982, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.16606400161981583, "p95_ms": 0.16681759655475617, "p99_ms": 0.16720287904143333, "mean_ms": 0.16615658645828565, "n": 300}, {"repeat": 1, "p50_ms": 0.16607999801635742, "p95_ms": 0.16687999665737152, "p99_ms": 0.16732896000146866, "mean_ms": 0.16632309302687645, "n": 300}, {"repeat": 2, "p50_ms": 0.16633599996566772, "p95_ms": 0.16707199811935425, "p99_ms": 0.1674879938364029, "mean_ms": 0.1665611735979716, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_0c1b0b9bae2b8b48_sdpa_efficient.json", "configuration": "480x832_tp1_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_retry_480x832_tp1_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_retry_480x832_tp1_sp2_v1/captures/qkv_rank0_0c1b0b9bae2b8b48.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 0, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_cudnn", "status": "measured", "q_shape": [4680, 20, 128], "k_shape": [512, 20, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.0009765625, "rmse": 3.7567815210420183e-06, "relative_l2": 0.00045857779720101494, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.0006142374604239942, "rmse": 2.0105374115138647e-05, "relative_l2": 0.0020252364973195416, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.05209600180387497, "p95_ms": 0.05305760148912668, "p99_ms": 0.05331231988966465, "mean_ms": 0.05242218679438035, "n": 300}, {"repeat": 1, "p50_ms": 0.05209600180387497, "p95_ms": 0.053119998425245285, "p99_ms": 0.05430655974894759, "mean_ms": 0.05272618682434162, "n": 300}, {"repeat": 2, "p50_ms": 0.05212799832224846, "p95_ms": 0.05315199866890907, "p99_ms": 0.05556927833706139, "mean_ms": 0.05264010686427355, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_0c1b0b9bae2b8b48_sdpa_cudnn.json", "configuration": "480x832_tp1_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_retry_480x832_tp1_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_retry_480x832_tp1_sp2_v1/captures/qkv_rank0_0c1b0b9bae2b8b48.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 0, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "native_dense", "status": "measured", "q_shape": [4680, 20, 128], "k_shape": [512, 20, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.0, "rmse": 0.0, "relative_l2": 0.0, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.0006142374604239942, "rmse": 2.001643072823305e-05, "relative_l2": 0.002016277131911822, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.11584000289440155, "p95_ms": 0.16998559832572938, "p99_ms": 0.18946784272789932, "mean_ms": 0.13315210700035096, "n": 300}, {"repeat": 1, "p50_ms": 0.11529599875211716, "p95_ms": 0.16135680601000787, "p99_ms": 0.17347648218274117, "mean_ms": 0.12229866651197274, "n": 300}, {"repeat": 2, "p50_ms": 0.11529599875211716, "p95_ms": 0.1678624041378498, "p99_ms": 0.1762380795180796, "mean_ms": 0.12500234675904115, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_0c1b0b9bae2b8b48_native_dense.json", "configuration": "480x832_tp1_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_retry_480x832_tp1_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_retry_480x832_tp1_sp2_v1/captures/qkv_rank0_0e33ce888d278925.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 0, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "vllm_fa2", "status": "measured", "q_shape": [4680, 20, 128], "k_shape": [29640, 20, 128], "q_len": 4680, "k_len": 4680, "reference_error": {"max_abs": 0.03125, "rmse": 0.00046775552374968775, "relative_l2": 0.0009102978526036899, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.006950112781240136, "rmse": 0.0007533224446526795, "relative_l2": 0.0019169673843472507, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.6226399838924408, "p95_ms": 0.6264320015907288, "p99_ms": 0.6272025674581527, "mean_ms": 0.6219401601950327, "n": 300}, {"repeat": 1, "p50_ms": 0.6234399974346161, "p95_ms": 0.6279696136713028, "p99_ms": 0.6284825772047042, "mean_ms": 0.6250199457009633, "n": 300}, {"repeat": 2, "p50_ms": 0.6235520243644714, "p95_ms": 0.6281616002321243, "p99_ms": 0.6289292854070664, "mean_ms": 0.6250577050447464, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_0e33ce888d278925_vllm_fa2.json", "configuration": "480x832_tp1_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_retry_480x832_tp1_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_retry_480x832_tp1_sp2_v1/captures/qkv_rank0_0e33ce888d278925.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 0, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "vllm_fa3", "status": "measured", "q_shape": [4680, 20, 128], "k_shape": [29640, 20, 128], "q_len": 4680, "k_len": 4680, "reference_error": {"max_abs": 0.0, "rmse": 0.0, "relative_l2": 0.0, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.006950112781240136, "rmse": 0.0007627068873890609, "relative_l2": 0.0019408478232929027, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.36987200379371643, "p95_ms": 0.371265609562397, "p99_ms": 0.3716169658303261, "mean_ms": 0.3595541341106097, "n": 300}, {"repeat": 1, "p50_ms": 0.36905598640441895, "p95_ms": 0.3742400109767914, "p99_ms": 0.37487998604774475, "mean_ms": 0.36670069257418314, "n": 300}, {"repeat": 2, "p50_ms": 0.3712960034608841, "p95_ms": 0.37779520004987716, "p99_ms": 0.378656319975853, "mean_ms": 0.3692518402139346, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_0e33ce888d278925_vllm_fa3.json", "configuration": "480x832_tp1_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_retry_480x832_tp1_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_retry_480x832_tp1_sp2_v1/captures/qkv_rank0_0e33ce888d278925.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 0, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_math", "status": "measured", "q_shape": [4680, 20, 128], "k_shape": [29640, 20, 128], "q_len": 4680, "k_len": 4680, "reference_error": {"max_abs": 0.03125, "rmse": 0.0007545789405605385, "relative_l2": 0.0014684841852980555, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.006950112781240136, "rmse": 0.0006602534478209683, "relative_l2": 0.0016801362203659833, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 9.048160076141357, "p95_ms": 9.055041933059693, "p99_ms": 9.056457262039185, "mean_ms": 9.048880414962769, "n": 300}, {"repeat": 1, "p50_ms": 9.047664165496826, "p95_ms": 9.05779333114624, "p99_ms": 9.062465686798095, "mean_ms": 9.049879461924235, "n": 300}, {"repeat": 2, "p50_ms": 9.049760341644287, "p95_ms": 9.058211660385131, "p99_ms": 9.061633758544922, "mean_ms": 9.050615361531575, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_0e33ce888d278925_sdpa_math.json", "configuration": "480x832_tp1_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_retry_480x832_tp1_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_retry_480x832_tp1_sp2_v1/captures/qkv_rank0_0e33ce888d278925.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 0, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_flash", "status": "measured", "q_shape": [4680, 20, 128], "k_shape": [29640, 20, 128], "q_len": 4680, "k_len": 4680, "reference_error": {"max_abs": 0.03125, "rmse": 0.000468004173442929, "relative_l2": 0.0009107817491486063, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.006950112781240136, "rmse": 0.0007533268077478835, "relative_l2": 0.0019169784870447213, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.6717119812965393, "p95_ms": 0.6793072164058686, "p99_ms": 0.6810307347774506, "mean_ms": 0.6725526416301727, "n": 300}, {"repeat": 1, "p50_ms": 0.675104022026062, "p95_ms": 0.6773120164871216, "p99_ms": 0.6777606451511383, "mean_ms": 0.6749605361620585, "n": 300}, {"repeat": 2, "p50_ms": 0.6764479875564575, "p95_ms": 0.6800959706306458, "p99_ms": 0.6805773061513901, "mean_ms": 0.6759166902303696, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_0e33ce888d278925_sdpa_flash.json", "configuration": "480x832_tp1_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_retry_480x832_tp1_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_retry_480x832_tp1_sp2_v1/captures/qkv_rank0_0e33ce888d278925.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 0, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_efficient", "status": "measured", "q_shape": [4680, 20, 128], "k_shape": [29640, 20, 128], "q_len": 4680, "k_len": 4680, "reference_error": {"max_abs": 0.03125, "rmse": 0.0006655913009421192, "relative_l2": 0.0012953055628340119, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.006950112781240136, "rmse": 0.0007546674581946931, "relative_l2": 0.001920390018452802, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 1.3849120140075684, "p95_ms": 1.3982912361621858, "p99_ms": 1.4004214560985566, "mean_ms": 1.3882351994514466, "n": 300}, {"repeat": 1, "p50_ms": 1.3825920224189758, "p95_ms": 1.3865328133106232, "p99_ms": 1.3886748743057251, "mean_ms": 1.3833213857809703, "n": 300}, {"repeat": 2, "p50_ms": 1.3826240301132202, "p95_ms": 1.386241602897644, "p99_ms": 1.3880729866027832, "mean_ms": 1.3830418133735656, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_0e33ce888d278925_sdpa_efficient.json", "configuration": "480x832_tp1_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_retry_480x832_tp1_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_retry_480x832_tp1_sp2_v1/captures/qkv_rank0_0e33ce888d278925.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 0, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_cudnn", "status": "measured", "q_shape": [4680, 20, 128], "k_shape": [29640, 20, 128], "q_len": 4680, "k_len": 4680, "reference_error": {"max_abs": 0.03125, "rmse": 0.0006651537944295404, "relative_l2": 0.001294454132506246, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.006950112781240136, "rmse": 0.0007544565760142951, "relative_l2": 0.0019198533899949183, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.41808000206947327, "p95_ms": 0.42508798837661743, "p99_ms": 0.42547423124313355, "mean_ms": 0.4092653858661652, "n": 300}, {"repeat": 1, "p50_ms": 0.4184640049934387, "p95_ms": 0.4298287943005562, "p99_ms": 0.43072224140167237, "mean_ms": 0.4179823996623357, "n": 300}, {"repeat": 2, "p50_ms": 0.419280007481575, "p95_ms": 0.42950560748577116, "p99_ms": 0.43024160534143446, "mean_ms": 0.41741760075092316, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_0e33ce888d278925_sdpa_cudnn.json", "configuration": "480x832_tp1_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_retry_480x832_tp1_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_retry_480x832_tp1_sp2_v1/captures/qkv_rank0_0e33ce888d278925.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 0, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "native_dense", "status": "measured", "q_shape": [4680, 20, 128], "k_shape": [29640, 20, 128], "q_len": 4680, "k_len": 4680, "reference_error": {"max_abs": 0.03125, "rmse": 0.0004351454555273363, "relative_l2": 0.0008468354805549274, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.006950112781240136, "rmse": 0.0007607101265716498, "relative_l2": 0.0019357666984857367, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.3426239937543869, "p95_ms": 0.3551055997610092, "p99_ms": 0.35558527559041975, "mean_ms": 0.3384732801715533, "n": 300}, {"repeat": 1, "p50_ms": 0.3477119952440262, "p95_ms": 0.35180801153182983, "p99_ms": 0.3552371203899384, "mean_ms": 0.3441518916686376, "n": 300}, {"repeat": 2, "p50_ms": 0.35022400319576263, "p95_ms": 0.35345118790864943, "p99_ms": 0.354402568936348, "mean_ms": 0.3480805331468582, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_0e33ce888d278925_native_dense.json", "configuration": "480x832_tp1_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_retry_480x832_tp1_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_retry_480x832_tp1_sp2_v1/captures/qkv_rank0_23b4681596dc64ac.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 9, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "vllm_fa2", "status": "measured", "q_shape": [4680, 20, 128], "k_shape": [29640, 20, 128], "q_len": 4680, "k_len": 18720, "reference_error": {"max_abs": 0.03125, "rmse": 0.00048489808748047354, "relative_l2": 0.000930507427809316, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.008430754864764634, "rmse": 0.0008224519263401907, "relative_l2": 0.0019414505560311024, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 2.4257280826568604, "p95_ms": 2.467355263233185, "p99_ms": 2.4699256801605225, "mean_ms": 2.431489922205607, "n": 300}, {"repeat": 1, "p50_ms": 2.451743960380554, "p95_ms": 2.6624208927154545, "p99_ms": 2.7147877478599547, "mean_ms": 2.481170663833618, "n": 300}, {"repeat": 2, "p50_ms": 2.450096011161804, "p95_ms": 2.552014446258545, "p99_ms": 2.5553914380073546, "mean_ms": 2.4691570099194844, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_23b4681596dc64ac_vllm_fa2.json", "configuration": "480x832_tp1_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_retry_480x832_tp1_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_retry_480x832_tp1_sp2_v1/captures/qkv_rank0_23b4681596dc64ac.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 9, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "vllm_fa3", "status": "measured", "q_shape": [4680, 20, 128], "k_shape": [29640, 20, 128], "q_len": 4680, "k_len": 18720, "reference_error": {"max_abs": 0.0, "rmse": 0.0, "relative_l2": 0.0, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.008430754864764634, "rmse": 0.0008316797479339394, "relative_l2": 0.001963233421132848, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 1.3669919967651367, "p95_ms": 1.3708415925502777, "p99_ms": 1.3746259331703186, "mean_ms": 1.3621502955754599, "n": 300}, {"repeat": 1, "p50_ms": 1.3675040006637573, "p95_ms": 1.3752959966659546, "p99_ms": 1.3767999410629272, "mean_ms": 1.363146344423294, "n": 300}, {"repeat": 2, "p50_ms": 1.3861920237541199, "p95_ms": 1.5100159645080566, "p99_ms": 1.5120870208740234, "mean_ms": 1.4154293314615884, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_23b4681596dc64ac_vllm_fa3.json", "configuration": "480x832_tp1_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_retry_480x832_tp1_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_retry_480x832_tp1_sp2_v1/captures/qkv_rank0_23b4681596dc64ac.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 9, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_math", "status": "measured", "q_shape": [4680, 20, 128], "k_shape": [29640, 20, 128], "q_len": 4680, "k_len": 18720, "reference_error": {"max_abs": 0.03125, "rmse": 0.0008278957583846513, "relative_l2": 0.0015887114684892767, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.007760694409859159, "rmse": 0.0007096702226792513, "relative_l2": 0.0016752221063550086, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 35.10646438598633, "p95_ms": 35.14780807495117, "p99_ms": 35.15513961791992, "mean_ms": 35.11369505564372, "n": 300}, {"repeat": 1, "p50_ms": 35.10921669006348, "p95_ms": 35.15113410949707, "p99_ms": 35.17693351745606, "mean_ms": 35.11490955352783, "n": 300}, {"repeat": 2, "p50_ms": 35.10592079162598, "p95_ms": 35.13053550720215, "p99_ms": 35.13578514099121, "mean_ms": 35.10819943745931, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_23b4681596dc64ac_sdpa_math.json", "configuration": "480x832_tp1_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_retry_480x832_tp1_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_retry_480x832_tp1_sp2_v1/captures/qkv_rank0_23b4681596dc64ac.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 9, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_flash", "status": "measured", "q_shape": [4680, 20, 128], "k_shape": [29640, 20, 128], "q_len": 4680, "k_len": 18720, "reference_error": {"max_abs": 0.03125, "rmse": 0.00048512975976239584, "relative_l2": 0.0009309520011840356, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.008430754864764634, "rmse": 0.0008226065287387949, "relative_l2": 0.0019418155049151903, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 2.634320020675659, "p95_ms": 2.686695945262909, "p99_ms": 2.710549337863922, "mean_ms": 2.6437768467267353, "n": 300}, {"repeat": 1, "p50_ms": 2.678431987762451, "p95_ms": 2.8751696825027464, "p99_ms": 2.876001527309418, "mean_ms": 2.6824681639671324, "n": 300}, {"repeat": 2, "p50_ms": 2.6791839599609375, "p95_ms": 2.759545648097992, "p99_ms": 2.7702806568145752, "mean_ms": 2.6734749921162924, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_23b4681596dc64ac_sdpa_flash.json", "configuration": "480x832_tp1_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_retry_480x832_tp1_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_retry_480x832_tp1_sp2_v1/captures/qkv_rank0_23b4681596dc64ac.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 9, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_efficient", "status": "measured", "q_shape": [4680, 20, 128], "k_shape": [29640, 20, 128], "q_len": 4680, "k_len": 18720, "reference_error": {"max_abs": 0.03125, "rmse": 0.000678704984213333, "relative_l2": 0.0013024180655840235, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.00928316173610133, "rmse": 0.000829930267228196, "relative_l2": 0.001959103659647523, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 5.444656133651733, "p95_ms": 5.490851306915284, "p99_ms": 5.497316226959229, "mean_ms": 5.447988805770874, "n": 300}, {"repeat": 1, "p50_ms": 5.443615913391113, "p95_ms": 5.4519437789917, "p99_ms": 5.453891777992249, "mean_ms": 5.443305072784423, "n": 300}, {"repeat": 2, "p50_ms": 5.444352149963379, "p95_ms": 5.454352164268494, "p99_ms": 5.462581934928894, "mean_ms": 5.443842236200968, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_23b4681596dc64ac_sdpa_efficient.json", "configuration": "480x832_tp1_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_retry_480x832_tp1_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_retry_480x832_tp1_sp2_v1/captures/qkv_rank0_23b4681596dc64ac.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 9, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_cudnn", "status": "measured", "q_shape": [4680, 20, 128], "k_shape": [29640, 20, 128], "q_len": 4680, "k_len": 18720, "reference_error": {"max_abs": 0.03125, "rmse": 0.0006770580024648157, "relative_l2": 0.0012992575483742636, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.00928316173610133, "rmse": 0.000830095690624396, "relative_l2": 0.0019594941521909135, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 1.6184479594230652, "p95_ms": 1.6643520593643188, "p99_ms": 1.6762578618526458, "mean_ms": 1.616351680358251, "n": 300}, {"repeat": 1, "p50_ms": 1.6605280041694641, "p95_ms": 1.8865344166755675, "p99_ms": 1.887783056497574, "mean_ms": 1.676459307273229, "n": 300}, {"repeat": 2, "p50_ms": 1.6325119733810425, "p95_ms": 1.6655695915222168, "p99_ms": 1.6715222346782683, "mean_ms": 1.6345897603034973, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_23b4681596dc64ac_sdpa_cudnn.json", "configuration": "480x832_tp1_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_retry_480x832_tp1_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_retry_480x832_tp1_sp2_v1/captures/qkv_rank0_23b4681596dc64ac.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 9, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "native_dense", "status": "measured", "q_shape": [4680, 20, 128], "k_shape": [29640, 20, 128], "q_len": 4680, "k_len": 18720, "reference_error": {"max_abs": 0.03125, "rmse": 0.00043555783098289495, "relative_l2": 0.0008358246968470857, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.008430754864764634, "rmse": 0.0008339159662758888, "relative_l2": 0.0019685121580466317, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 1.3587360382080078, "p95_ms": 1.4206415951251983, "p99_ms": 1.4214738738536834, "mean_ms": 1.3676731729507445, "n": 300}, {"repeat": 1, "p50_ms": 1.3863199949264526, "p95_ms": 1.3998431980609893, "p99_ms": 1.4107936573028563, "mean_ms": 1.3762392544746398, "n": 300}, {"repeat": 2, "p50_ms": 1.368175983428955, "p95_ms": 1.4061327576637268, "p99_ms": 1.5045520198345184, "mean_ms": 1.37776362657547, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_23b4681596dc64ac_native_dense.json", "configuration": "480x832_tp1_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_retry_480x832_tp1_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_retry_480x832_tp1_sp2_v1/captures/qkv_rank0_325560d6739aa74a.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 9, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "vllm_fa2", "status": "measured", "q_shape": [4680, 20, 128], "k_shape": [512, 20, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.00048828125, "rmse": 1.8836826195050004e-06, "relative_l2": 0.0005715062122943907, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 7.629575161298766e-05, "rmse": 4.942429805964212e-06, "relative_l2": 0.0018663346713588256, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.08310399949550629, "p95_ms": 0.08403199911117554, "p99_ms": 0.08432032130658626, "mean_ms": 0.08329407977561155, "n": 300}, {"repeat": 1, "p50_ms": 0.0830719992518425, "p95_ms": 0.0838719978928566, "p99_ms": 0.08406399935483932, "mean_ms": 0.08341333299875259, "n": 300}, {"repeat": 2, "p50_ms": 0.08321600034832954, "p95_ms": 0.0840367991477251, "p99_ms": 0.0844803225249052, "mean_ms": 0.08357333332300186, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_325560d6739aa74a_vllm_fa2.json", "configuration": "480x832_tp1_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_retry_480x832_tp1_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_retry_480x832_tp1_sp2_v1/captures/qkv_rank0_325560d6739aa74a.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 9, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "vllm_fa3", "status": "measured", "q_shape": [4680, 20, 128], "k_shape": [512, 20, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.000244140625, "rmse": 7.083348487188031e-07, "relative_l2": 0.0002149076294677405, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 7.629575161298766e-05, "rmse": 4.97898511900548e-06, "relative_l2": 0.0018801384987938588, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.07235199958086014, "p95_ms": 0.07315199822187424, "p99_ms": 0.07721183910965917, "mean_ms": 0.07256351990004381, "n": 300}, {"repeat": 1, "p50_ms": 0.07212799787521362, "p95_ms": 0.07286400347948074, "p99_ms": 0.0732809587568044, "mean_ms": 0.07272234675784905, "n": 300}, {"repeat": 2, "p50_ms": 0.0721919983625412, "p95_ms": 0.07327999919652939, "p99_ms": 0.07364640161395068, "mean_ms": 0.07277141312758127, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_325560d6739aa74a_vllm_fa3.json", "configuration": "480x832_tp1_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_retry_480x832_tp1_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_retry_480x832_tp1_sp2_v1/captures/qkv_rank0_325560d6739aa74a.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 9, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_math", "status": "measured", "q_shape": [4680, 20, 128], "k_shape": [512, 20, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.0009765625, "rmse": 5.525154027990402e-06, "relative_l2": 0.0016763226555170261, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 5.955835214807581e-05, "rmse": 4.4437602027578725e-06, "relative_l2": 0.0016780296459857492, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 1.125711977481842, "p95_ms": 1.1306031882762908, "p99_ms": 1.1321929752826692, "mean_ms": 1.1260901347796122, "n": 300}, {"repeat": 1, "p50_ms": 1.1252319812774658, "p95_ms": 1.128899198770523, "p99_ms": 1.1302115356922149, "mean_ms": 1.1257913613319397, "n": 300}, {"repeat": 2, "p50_ms": 1.1249759793281555, "p95_ms": 1.1287695705890655, "p99_ms": 1.1304963445663452, "mean_ms": 1.1255871999263762, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_325560d6739aa74a_sdpa_math.json", "configuration": "480x832_tp1_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_retry_480x832_tp1_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_retry_480x832_tp1_sp2_v1/captures/qkv_rank0_325560d6739aa74a.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 9, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_flash", "status": "measured", "q_shape": [4680, 20, 128], "k_shape": [512, 20, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.00048828125, "rmse": 1.883728743872256e-06, "relative_l2": 0.0005715202063516441, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 7.629575161298766e-05, "rmse": 4.942430280889429e-06, "relative_l2": 0.001866334850697619, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.08905600011348724, "p95_ms": 0.0899839997291565, "p99_ms": 0.09046528331935405, "mean_ms": 0.08924362669388453, "n": 300}, {"repeat": 1, "p50_ms": 0.08887999877333641, "p95_ms": 0.08972799777984619, "p99_ms": 0.09001727998256684, "mean_ms": 0.08922229312360287, "n": 300}, {"repeat": 2, "p50_ms": 0.08902399986982346, "p95_ms": 0.0899519994854927, "p99_ms": 0.0904969635605812, "mean_ms": 0.08941578656435013, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_325560d6739aa74a_sdpa_flash.json", "configuration": "480x832_tp1_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_retry_480x832_tp1_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_retry_480x832_tp1_sp2_v1/captures/qkv_rank0_325560d6739aa74a.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 9, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_efficient", "status": "measured", "q_shape": [4680, 20, 128], "k_shape": [512, 20, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.0009765625, "rmse": 3.291571659091038e-06, "relative_l2": 0.0009986574340623356, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 7.629575161298766e-05, "rmse": 5.409689151586598e-06, "relative_l2": 0.002042778718414212, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.16572800278663635, "p95_ms": 0.1664319932460785, "p99_ms": 0.16668831527233124, "mean_ms": 0.16589269345005353, "n": 300}, {"repeat": 1, "p50_ms": 0.16582399606704712, "p95_ms": 0.16646560057997703, "p99_ms": 0.1667843233048916, "mean_ms": 0.16607029284040134, "n": 300}, {"repeat": 2, "p50_ms": 0.16595199704170227, "p95_ms": 0.1667535960674286, "p99_ms": 0.1671052858233452, "mean_ms": 0.16625472014149031, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_325560d6739aa74a_sdpa_efficient.json", "configuration": "480x832_tp1_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_retry_480x832_tp1_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_retry_480x832_tp1_sp2_v1/captures/qkv_rank0_325560d6739aa74a.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 9, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_cudnn", "status": "measured", "q_shape": [4680, 20, 128], "k_shape": [512, 20, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.0009765625, "rmse": 3.292963429283332e-06, "relative_l2": 0.0009990796948523147, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 7.629575161298766e-05, "rmse": 5.409689151586598e-06, "relative_l2": 0.002042778718414212, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.05222399905323982, "p95_ms": 0.053249599412083626, "p99_ms": 0.06696992233395566, "mean_ms": 0.053219093071917696, "n": 300}, {"repeat": 1, "p50_ms": 0.05225599929690361, "p95_ms": 0.07425279878079892, "p99_ms": 0.08038368061184878, "mean_ms": 0.05470922653873762, "n": 300}, {"repeat": 2, "p50_ms": 0.05225599929690361, "p95_ms": 0.06010559927672149, "p99_ms": 0.07713983863592146, "mean_ms": 0.05383466651042303, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_325560d6739aa74a_sdpa_cudnn.json", "configuration": "480x832_tp1_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_retry_480x832_tp1_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_retry_480x832_tp1_sp2_v1/captures/qkv_rank0_325560d6739aa74a.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 9, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "native_dense", "status": "measured", "q_shape": [4680, 20, 128], "k_shape": [512, 20, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.0, "rmse": 0.0, "relative_l2": 0.0, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 7.629575161298766e-05, "rmse": 4.99106817785475e-06, "relative_l2": 0.0018847012407146944, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.11475200206041336, "p95_ms": 0.1500943943858147, "p99_ms": 0.2331635215878486, "mean_ms": 0.12051338675121466, "n": 300}, {"repeat": 1, "p50_ms": 0.1130559965968132, "p95_ms": 0.1177936028689146, "p99_ms": 0.12016639813780779, "mean_ms": 0.11425909337898095, "n": 300}, {"repeat": 2, "p50_ms": 0.11337599903345108, "p95_ms": 0.11850080080330372, "p99_ms": 0.13451072335243225, "mean_ms": 0.1146253868440787, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_325560d6739aa74a_native_dense.json", "configuration": "480x832_tp1_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_retry_480x832_tp1_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_retry_480x832_tp1_sp2_v1/captures/qkv_rank0_37c9b2a3872f7438.pt", "context": {"phase": "commit", "step": null, "start_frame": 27, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "vllm_fa2", "status": "measured", "q_shape": [4680, 20, 128], "k_shape": [29640, 20, 128], "q_len": 4680, "k_len": 28080, "reference_error": {"max_abs": 0.0625, "rmse": 0.000430560705924088, "relative_l2": 0.0005870272162627125, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.010101945123846434, "rmse": 0.0009676761183591323, "relative_l2": 0.0016959539687613312, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 3.6590720415115356, "p95_ms": 3.9518239736557006, "p99_ms": 4.060214662551879, "mean_ms": 3.70234698454539, "n": 300}, {"repeat": 1, "p50_ms": 3.687935948371887, "p95_ms": 3.8159440875053408, "p99_ms": 3.8794298553466797, "mean_ms": 3.7123914631207784, "n": 300}, {"repeat": 2, "p50_ms": 3.7145440578460693, "p95_ms": 3.781166410446167, "p99_ms": 3.7927885842323303, "mean_ms": 3.708504642645518, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_37c9b2a3872f7438_vllm_fa2.json", "configuration": "480x832_tp1_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_retry_480x832_tp1_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_retry_480x832_tp1_sp2_v1/captures/qkv_rank0_37c9b2a3872f7438.pt", "context": {"phase": "commit", "step": null, "start_frame": 27, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "vllm_fa3", "status": "measured", "q_shape": [4680, 20, 128], "k_shape": [29640, 20, 128], "q_len": 4680, "k_len": 28080, "reference_error": {"max_abs": 0.0, "rmse": 0.0, "relative_l2": 0.0, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.010101945123846434, "rmse": 0.0009681409449194487, "relative_l2": 0.0016967686260984276, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 2.0410399436950684, "p95_ms": 2.0850576639175413, "p99_ms": 2.090880558490753, "mean_ms": 2.035259413719177, "n": 300}, {"repeat": 1, "p50_ms": 2.0452960729599, "p95_ms": 2.270201539993286, "p99_ms": 2.3316519498825072, "mean_ms": 2.0856663473447163, "n": 300}, {"repeat": 2, "p50_ms": 2.080175995826721, "p95_ms": 2.1194241046905518, "p99_ms": 2.121955895423889, "mean_ms": 2.066770348151525, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_37c9b2a3872f7438_vllm_fa3.json", "configuration": "480x832_tp1_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_retry_480x832_tp1_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_retry_480x832_tp1_sp2_v1/captures/qkv_rank0_37c9b2a3872f7438.pt", "context": {"phase": "commit", "step": null, "start_frame": 27, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_math", "status": "measured", "q_shape": [4680, 20, 128], "k_shape": [29640, 20, 128], "q_len": 4680, "k_len": 28080, "reference_error": {"max_abs": 0.0625, "rmse": 0.0008267063200807613, "relative_l2": 0.0011271328364771012, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.010101945123846434, "rmse": 0.0009455398630386555, "relative_l2": 0.0016571578577981548, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 52.62985610961914, "p95_ms": 52.6719970703125, "p99_ms": 52.88413047790527, "mean_ms": 52.642149429321286, "n": 300}, {"repeat": 1, "p50_ms": 52.63950538635254, "p95_ms": 52.92564525604248, "p99_ms": 52.9398762512207, "mean_ms": 52.694814376831054, "n": 300}, {"repeat": 2, "p50_ms": 52.64713668823242, "p95_ms": 52.91429748535156, "p99_ms": 52.93752044677734, "mean_ms": 52.699918797810874, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_37c9b2a3872f7438_sdpa_math.json", "configuration": "480x832_tp1_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_retry_480x832_tp1_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_retry_480x832_tp1_sp2_v1/captures/qkv_rank0_37c9b2a3872f7438.pt", "context": {"phase": "commit", "step": null, "start_frame": 27, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_flash", "status": "measured", "q_shape": [4680, 20, 128], "k_shape": [29640, 20, 128], "q_len": 4680, "k_len": 28080, "reference_error": {"max_abs": 0.0625, "rmse": 0.0004316709017471864, "relative_l2": 0.0005885408591812906, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.010101945123846434, "rmse": 0.0009676761183591323, "relative_l2": 0.0016959539687613312, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 3.947119951248169, "p95_ms": 4.113465738296509, "p99_ms": 4.183937077522278, "mean_ms": 3.9732098086675007, "n": 300}, {"repeat": 1, "p50_ms": 3.9478559494018555, "p95_ms": 4.053987169265747, "p99_ms": 4.086349306106567, "mean_ms": 3.9688840508461, "n": 300}, {"repeat": 2, "p50_ms": 3.9776159524917603, "p95_ms": 4.030417394638062, "p99_ms": 4.045302062034607, "mean_ms": 3.972197653452555, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_37c9b2a3872f7438_sdpa_flash.json", "configuration": "480x832_tp1_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_retry_480x832_tp1_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_retry_480x832_tp1_sp2_v1/captures/qkv_rank0_37c9b2a3872f7438.pt", "context": {"phase": "commit", "step": null, "start_frame": 27, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_efficient", "status": "measured", "q_shape": [4680, 20, 128], "k_shape": [29640, 20, 128], "q_len": 4680, "k_len": 28080, "reference_error": {"max_abs": 0.0625, "rmse": 0.0007081494522111591, "relative_l2": 0.0009654921963612119, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.010101945123846434, "rmse": 0.0009517787171752579, "relative_l2": 0.001668092104528802, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 8.1419358253479, "p95_ms": 8.1599506855011, "p99_ms": 8.212165870666503, "mean_ms": 8.141505403518677, "n": 300}, {"repeat": 1, "p50_ms": 8.142064094543457, "p95_ms": 8.150468873977662, "p99_ms": 8.157828683853149, "mean_ms": 8.138815015157064, "n": 300}, {"repeat": 2, "p50_ms": 8.142656326293945, "p95_ms": 8.152132987976074, "p99_ms": 8.158673400878905, "mean_ms": 8.139167585372924, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_37c9b2a3872f7438_sdpa_efficient.json", "configuration": "480x832_tp1_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_retry_480x832_tp1_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_retry_480x832_tp1_sp2_v1/captures/qkv_rank0_37c9b2a3872f7438.pt", "context": {"phase": "commit", "step": null, "start_frame": 27, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_cudnn", "status": "measured", "q_shape": [4680, 20, 128], "k_shape": [29640, 20, 128], "q_len": 4680, "k_len": 28080, "reference_error": {"max_abs": 0.0625, "rmse": 0.0007072754035817625, "relative_l2": 0.0009643005169377685, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.010101945123846434, "rmse": 0.0009516249798776634, "relative_l2": 0.0016678226637778573, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 2.409648060798645, "p95_ms": 2.431393551826477, "p99_ms": 2.4332502818107606, "mean_ms": 2.4092631483078004, "n": 300}, {"repeat": 1, "p50_ms": 2.430880069732666, "p95_ms": 2.791785550117493, "p99_ms": 2.8843522882461547, "mean_ms": 2.468227203687032, "n": 300}, {"repeat": 2, "p50_ms": 2.453263998031616, "p95_ms": 2.572096109390259, "p99_ms": 2.5732527112960817, "mean_ms": 2.462216107050578, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_37c9b2a3872f7438_sdpa_cudnn.json", "configuration": "480x832_tp1_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_retry_480x832_tp1_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_retry_480x832_tp1_sp2_v1/captures/qkv_rank0_37c9b2a3872f7438.pt", "context": {"phase": "commit", "step": null, "start_frame": 27, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "native_dense", "status": "measured", "q_shape": [4680, 20, 128], "k_shape": [29640, 20, 128], "q_len": 4680, "k_len": 28080, "reference_error": {"max_abs": 0.0625, "rmse": 0.00038278201745887286, "relative_l2": 0.00052188566920439, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.010101945123846434, "rmse": 0.0009673749392795276, "relative_l2": 0.0016954261208113068, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 2.018847942352295, "p95_ms": 2.030209708213806, "p99_ms": 2.0314566326141357, "mean_ms": 2.0092484299341837, "n": 300}, {"repeat": 1, "p50_ms": 2.029711961746216, "p95_ms": 2.3410255312919617, "p99_ms": 2.368389005661011, "mean_ms": 2.079103257258733, "n": 300}, {"repeat": 2, "p50_ms": 2.0184000730514526, "p95_ms": 2.1748270988464355, "p99_ms": 2.1865305280685425, "mean_ms": 2.038186988433202, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_37c9b2a3872f7438_native_dense.json", "configuration": "480x832_tp1_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_retry_480x832_tp1_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_retry_480x832_tp1_sp2_v1/captures/qkv_rank0_5f58cf168d29ac00.pt", "context": {"phase": "commit", "step": null, "start_frame": 0, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "vllm_fa2", "status": "measured", "q_shape": [4680, 20, 128], "k_shape": [29640, 20, 128], "q_len": 4680, "k_len": 4680, "reference_error": {"max_abs": 0.03125, "rmse": 0.0004238123822568921, "relative_l2": 0.0006066598492131995, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.013453558575000635, "rmse": 0.000855206902521633, "relative_l2": 0.0017184019899498433, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.6211519837379456, "p95_ms": 0.6229119896888733, "p99_ms": 0.6232640147209167, "mean_ms": 0.6211402668555578, "n": 300}, {"repeat": 1, "p50_ms": 0.6219840049743652, "p95_ms": 0.623105576634407, "p99_ms": 0.6236195003986359, "mean_ms": 0.6225672539075215, "n": 300}, {"repeat": 2, "p50_ms": 0.622111976146698, "p95_ms": 0.6308159828186035, "p99_ms": 0.6312323427200317, "mean_ms": 0.6233308804035187, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_5f58cf168d29ac00_vllm_fa2.json", "configuration": "480x832_tp1_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_retry_480x832_tp1_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_retry_480x832_tp1_sp2_v1/captures/qkv_rank0_5f58cf168d29ac00.pt", "context": {"phase": "commit", "step": null, "start_frame": 0, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "vllm_fa3", "status": "measured", "q_shape": [4680, 20, 128], "k_shape": [29640, 20, 128], "q_len": 4680, "k_len": 4680, "reference_error": {"max_abs": 0.0, "rmse": 0.0, "relative_l2": 0.0, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.013453558575000635, "rmse": 0.0008552870221296227, "relative_l2": 0.001718562977534716, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.3638080060482025, "p95_ms": 0.36592160612344743, "p99_ms": 0.3666563254594803, "mean_ms": 0.3513010138273239, "n": 300}, {"repeat": 1, "p50_ms": 0.3620480000972748, "p95_ms": 0.368353596329689, "p99_ms": 0.3698886474967003, "mean_ms": 0.3616702934106191, "n": 300}, {"repeat": 2, "p50_ms": 0.36694398522377014, "p95_ms": 0.36924960315227506, "p99_ms": 0.37059359818696974, "mean_ms": 0.36346346626679105, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_5f58cf168d29ac00_vllm_fa3.json", "configuration": "480x832_tp1_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_retry_480x832_tp1_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_retry_480x832_tp1_sp2_v1/captures/qkv_rank0_5f58cf168d29ac00.pt", "context": {"phase": "commit", "step": null, "start_frame": 0, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_math", "status": "measured", "q_shape": [4680, 20, 128], "k_shape": [29640, 20, 128], "q_len": 4680, "k_len": 4680, "reference_error": {"max_abs": 0.03125, "rmse": 0.0007322738741810228, "relative_l2": 0.0010482024043935313, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.013453558575000635, "rmse": 0.000840414736449667, "relative_l2": 0.0016886794894195205, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 9.048223972320557, "p95_ms": 9.058640241622925, "p99_ms": 9.098154458999634, "mean_ms": 9.052069110870361, "n": 300}, {"repeat": 1, "p50_ms": 9.048399925231934, "p95_ms": 9.056883001327515, "p99_ms": 9.060292263031005, "mean_ms": 9.049553616841633, "n": 300}, {"repeat": 2, "p50_ms": 9.048223972320557, "p95_ms": 9.056337785720824, "p99_ms": 9.060937623977662, "mean_ms": 9.049558162689209, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_5f58cf168d29ac00_sdpa_math.json", "configuration": "480x832_tp1_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_retry_480x832_tp1_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_retry_480x832_tp1_sp2_v1/captures/qkv_rank0_5f58cf168d29ac00.pt", "context": {"phase": "commit", "step": null, "start_frame": 0, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_flash", "status": "measured", "q_shape": [4680, 20, 128], "k_shape": [29640, 20, 128], "q_len": 4680, "k_len": 4680, "reference_error": {"max_abs": 0.03125, "rmse": 0.0004244422871256175, "relative_l2": 0.000607561517047076, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.013453558575000635, "rmse": 0.000855206902521633, "relative_l2": 0.0017184019899498433, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.6696480214595795, "p95_ms": 0.6728336125612259, "p99_ms": 0.6737936043739319, "mean_ms": 0.66953109284242, "n": 300}, {"repeat": 1, "p50_ms": 0.6710239946842194, "p95_ms": 0.672993591427803, "p99_ms": 0.6737308692932129, "mean_ms": 0.6710461870829264, "n": 300}, {"repeat": 2, "p50_ms": 0.6697919964790344, "p95_ms": 0.6728960275650024, "p99_ms": 0.6733456158638, "mean_ms": 0.6705441063642502, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_5f58cf168d29ac00_sdpa_flash.json", "configuration": "480x832_tp1_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_retry_480x832_tp1_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_retry_480x832_tp1_sp2_v1/captures/qkv_rank0_5f58cf168d29ac00.pt", "context": {"phase": "commit", "step": null, "start_frame": 0, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_efficient", "status": "measured", "q_shape": [4680, 20, 128], "k_shape": [29640, 20, 128], "q_len": 4680, "k_len": 4680, "reference_error": {"max_abs": 0.03125, "rmse": 0.0005886275345030929, "relative_l2": 0.0008425820157088546, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.013453558575000635, "rmse": 0.0008510924395452369, "relative_l2": 0.0017101346322550366, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 1.3874720335006714, "p95_ms": 1.3983135879039765, "p99_ms": 1.4003871500492096, "mean_ms": 1.3891360008716582, "n": 300}, {"repeat": 1, "p50_ms": 1.38428795337677, "p95_ms": 1.3879535973072052, "p99_ms": 1.3899295961856841, "mean_ms": 1.3849410156408946, "n": 300}, {"repeat": 2, "p50_ms": 1.384768009185791, "p95_ms": 1.388452798128128, "p99_ms": 1.3901165223121643, "mean_ms": 1.3850752023855846, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_5f58cf168d29ac00_sdpa_efficient.json", "configuration": "480x832_tp1_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_retry_480x832_tp1_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_retry_480x832_tp1_sp2_v1/captures/qkv_rank0_5f58cf168d29ac00.pt", "context": {"phase": "commit", "step": null, "start_frame": 0, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_cudnn", "status": "measured", "q_shape": [4680, 20, 128], "k_shape": [29640, 20, 128], "q_len": 4680, "k_len": 4680, "reference_error": {"max_abs": 0.03125, "rmse": 0.0005870589890288059, "relative_l2": 0.0008403367449221738, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.013453558575000635, "rmse": 0.000851012888406189, "relative_l2": 0.0017099747869179142, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.40296000242233276, "p95_ms": 0.4153071895241737, "p99_ms": 0.4169920086860657, "mean_ms": 0.39959018578131994, "n": 300}, {"repeat": 1, "p50_ms": 0.40855999290943146, "p95_ms": 0.4186912000179291, "p99_ms": 0.41962017357349396, "mean_ms": 0.4086972798903783, "n": 300}, {"repeat": 2, "p50_ms": 0.4143199920654297, "p95_ms": 0.41811200976371765, "p99_ms": 0.41862465351819994, "mean_ms": 0.41030890663464864, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_5f58cf168d29ac00_sdpa_cudnn.json", "configuration": "480x832_tp1_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_retry_480x832_tp1_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_retry_480x832_tp1_sp2_v1/captures/qkv_rank0_5f58cf168d29ac00.pt", "context": {"phase": "commit", "step": null, "start_frame": 0, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "native_dense", "status": "measured", "q_shape": [4680, 20, 128], "k_shape": [29640, 20, 128], "q_len": 4680, "k_len": 4680, "reference_error": {"max_abs": 0.03125, "rmse": 0.0003785237897499875, "relative_l2": 0.0005418321758096813, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.013453558575000635, "rmse": 0.0008563707767124402, "relative_l2": 0.0017207406096682262, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.33452799916267395, "p95_ms": 0.3473919928073883, "p99_ms": 0.34787326902151106, "mean_ms": 0.33112341264883677, "n": 300}, {"repeat": 1, "p50_ms": 0.34142400324344635, "p95_ms": 0.34508960545063017, "p99_ms": 0.3475593614578247, "mean_ms": 0.338245120048523, "n": 300}, {"repeat": 2, "p50_ms": 0.34355199337005615, "p95_ms": 0.3449648037552834, "p99_ms": 0.34550590604543685, "mean_ms": 0.3419915739695231, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_5f58cf168d29ac00_native_dense.json", "configuration": "480x832_tp1_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_retry_480x832_tp1_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_retry_480x832_tp1_sp2_v1/captures/qkv_rank0_60db4e364fe9a358.pt", "context": {"phase": "commit", "step": null, "start_frame": 9, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "vllm_fa2", "status": "measured", "q_shape": [4680, 20, 128], "k_shape": [512, 20, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.0009765625, "rmse": 1.6734795937845932e-06, "relative_l2": 0.0005881900962707838, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 8.546650151804169e-05, "rmse": 4.432017925885755e-06, "relative_l2": 0.001751077866033789, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.08339200168848038, "p95_ms": 0.08419360034167767, "p99_ms": 0.08480415754020214, "mean_ms": 0.0835692801574866, "n": 300}, {"repeat": 1, "p50_ms": 0.08316799998283386, "p95_ms": 0.08396959863603115, "p99_ms": 0.08476927727460862, "mean_ms": 0.08350869335234165, "n": 300}, {"repeat": 2, "p50_ms": 0.08316799998283386, "p95_ms": 0.0841279998421669, "p99_ms": 0.0844169620424509, "mean_ms": 0.08354463999470076, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_60db4e364fe9a358_vllm_fa2.json", "configuration": "480x832_tp1_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_retry_480x832_tp1_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_retry_480x832_tp1_sp2_v1/captures/qkv_rank0_60db4e364fe9a358.pt", "context": {"phase": "commit", "step": null, "start_frame": 9, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "vllm_fa3", "status": "measured", "q_shape": [4680, 20, 128], "k_shape": [512, 20, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.000244140625, "rmse": 6.261797114381149e-07, "relative_l2": 0.00022008795692611522, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 8.546650151804169e-05, "rmse": 4.471097426135494e-06, "relative_l2": 0.0017665180670996055, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.0721919983625412, "p95_ms": 0.07295999675989151, "p99_ms": 0.07349664039909834, "mean_ms": 0.0724172801276048, "n": 300}, {"repeat": 1, "p50_ms": 0.07212799787521362, "p95_ms": 0.07308799773454666, "p99_ms": 0.07362848140299315, "mean_ms": 0.07267882694800695, "n": 300}, {"repeat": 2, "p50_ms": 0.07206399738788605, "p95_ms": 0.073217598721385, "p99_ms": 0.07386816352605818, "mean_ms": 0.0726522666712602, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_60db4e364fe9a358_vllm_fa3.json", "configuration": "480x832_tp1_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_retry_480x832_tp1_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_retry_480x832_tp1_sp2_v1/captures/qkv_rank0_60db4e364fe9a358.pt", "context": {"phase": "commit", "step": null, "start_frame": 9, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_math", "status": "measured", "q_shape": [4680, 20, 128], "k_shape": [512, 20, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.001953125, "rmse": 4.545564526228634e-06, "relative_l2": 0.0015976627657831048, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 8.546650151804169e-05, "rmse": 4.251983876886612e-06, "relative_l2": 0.0016799469176471491, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 1.1246720552444458, "p95_ms": 1.128607988357544, "p99_ms": 1.1296639442443848, "mean_ms": 1.1249627768993378, "n": 300}, {"repeat": 1, "p50_ms": 1.1244159936904907, "p95_ms": 1.128385591506958, "p99_ms": 1.1295430719852448, "mean_ms": 1.1254969636599224, "n": 300}, {"repeat": 2, "p50_ms": 1.1245920062065125, "p95_ms": 1.1282623648643493, "p99_ms": 1.1296339166164397, "mean_ms": 1.125098667939504, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_60db4e364fe9a358_sdpa_math.json", "configuration": "480x832_tp1_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_retry_480x832_tp1_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_retry_480x832_tp1_sp2_v1/captures/qkv_rank0_60db4e364fe9a358.pt", "context": {"phase": "commit", "step": null, "start_frame": 9, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_flash", "status": "measured", "q_shape": [4680, 20, 128], "k_shape": [512, 20, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.0009765625, "rmse": 1.6738170958738367e-06, "relative_l2": 0.0005883087205952757, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 8.546650151804169e-05, "rmse": 4.432017925885755e-06, "relative_l2": 0.001751077866033789, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.0888959988951683, "p95_ms": 0.08975999802350998, "p99_ms": 0.09027264192700386, "mean_ms": 0.08920981302857399, "n": 300}, {"repeat": 1, "p50_ms": 0.08902399986982346, "p95_ms": 0.08982399851083755, "p99_ms": 0.09027264192700386, "mean_ms": 0.08930207992593447, "n": 300}, {"repeat": 2, "p50_ms": 0.08902399986982346, "p95_ms": 0.08982399851083755, "p99_ms": 0.09050048358738422, "mean_ms": 0.0893847465266784, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_60db4e364fe9a358_sdpa_flash.json", "configuration": "480x832_tp1_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_retry_480x832_tp1_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_retry_480x832_tp1_sp2_v1/captures/qkv_rank0_60db4e364fe9a358.pt", "context": {"phase": "commit", "step": null, "start_frame": 9, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_efficient", "status": "measured", "q_shape": [4680, 20, 128], "k_shape": [512, 20, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.0009765625, "rmse": 2.903487950870584e-06, "relative_l2": 0.0010205101177728818, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 8.546650151804169e-05, "rmse": 5.084117071383684e-06, "relative_l2": 0.002008720411536978, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.165583997964859, "p95_ms": 0.1662416063249111, "p99_ms": 0.1670723181962967, "mean_ms": 0.16577877352635065, "n": 300}, {"repeat": 1, "p50_ms": 0.16566400229930878, "p95_ms": 0.1664975941181183, "p99_ms": 0.16694431722164155, "mean_ms": 0.1659234131872654, "n": 300}, {"repeat": 2, "p50_ms": 0.16595199704170227, "p95_ms": 0.16662399470806122, "p99_ms": 0.16704224526882172, "mean_ms": 0.1661927463610967, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_60db4e364fe9a358_sdpa_efficient.json", "configuration": "480x832_tp1_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_retry_480x832_tp1_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_retry_480x832_tp1_sp2_v1/captures/qkv_rank0_60db4e364fe9a358.pt", "context": {"phase": "commit", "step": null, "start_frame": 9, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_cudnn", "status": "measured", "q_shape": [4680, 20, 128], "k_shape": [512, 20, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.0009765625, "rmse": 2.9027484760488563e-06, "relative_l2": 0.001020250209156011, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 8.546650151804169e-05, "rmse": 5.0836227950647876e-06, "relative_l2": 0.002008525124348118, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.052032001316547394, "p95_ms": 0.05299200117588043, "p99_ms": 0.05383999980986114, "mean_ms": 0.05246624025205771, "n": 300}, {"repeat": 1, "p50_ms": 0.052032001316547394, "p95_ms": 0.05302400141954422, "p99_ms": 0.05391519956290708, "mean_ms": 0.052770986917118234, "n": 300}, {"repeat": 2, "p50_ms": 0.052000001072883606, "p95_ms": 0.052960000932216644, "p99_ms": 0.05397728104144332, "mean_ms": 0.0526201602568229, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_60db4e364fe9a358_sdpa_cudnn.json", "configuration": "480x832_tp1_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_retry_480x832_tp1_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_retry_480x832_tp1_sp2_v1/captures/qkv_rank0_60db4e364fe9a358.pt", "context": {"phase": "commit", "step": null, "start_frame": 9, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "native_dense", "status": "measured", "q_shape": [4680, 20, 128], "k_shape": [512, 20, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.0, "rmse": 0.0, "relative_l2": 0.0, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 8.546650151804169e-05, "rmse": 4.48689709008337e-06, "relative_l2": 0.0017727604700619917, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.11363200098276138, "p95_ms": 0.11908159777522087, "p99_ms": 0.1302851128578185, "mean_ms": 0.11489749319851399, "n": 300}, {"repeat": 1, "p50_ms": 0.11356800049543381, "p95_ms": 0.11962880156934262, "p99_ms": 0.13712480083107945, "mean_ms": 0.11563520009318988, "n": 300}, {"repeat": 2, "p50_ms": 0.11363200098276138, "p95_ms": 0.11965760216116905, "p99_ms": 0.13202911779284457, "mean_ms": 0.1151480533182621, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_60db4e364fe9a358_native_dense.json", "configuration": "480x832_tp1_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_retry_480x832_tp1_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_retry_480x832_tp1_sp2_v1/captures/qkv_rank0_89e6d37b3d6f79ec.pt", "context": {"phase": "commit", "step": null, "start_frame": 27, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "vllm_fa2", "status": "measured", "q_shape": [4680, 20, 128], "k_shape": [512, 20, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.000244140625, "rmse": 1.6516243760203818e-06, "relative_l2": 0.0005886583225825171, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 6.036895268355172e-05, "rmse": 3.8066916294210143e-06, "relative_l2": 0.0017721563376145706, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.08291199803352356, "p95_ms": 0.08377599716186523, "p99_ms": 0.08422464057803154, "mean_ms": 0.0830936528245608, "n": 300}, {"repeat": 1, "p50_ms": 0.08297599852085114, "p95_ms": 0.08384319767355919, "p99_ms": 0.08438688173890113, "mean_ms": 0.08333365278939406, "n": 300}, {"repeat": 2, "p50_ms": 0.08297599852085114, "p95_ms": 0.0838719978928566, "p99_ms": 0.08416512005031108, "mean_ms": 0.08330943944553534, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_89e6d37b3d6f79ec_vllm_fa2.json", "configuration": "480x832_tp1_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_retry_480x832_tp1_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_retry_480x832_tp1_sp2_v1/captures/qkv_rank0_89e6d37b3d6f79ec.pt", "context": {"phase": "commit", "step": null, "start_frame": 27, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "vllm_fa3", "status": "measured", "q_shape": [4680, 20, 128], "k_shape": [512, 20, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.000244140625, "rmse": 6.323970961290354e-07, "relative_l2": 0.00022539375127797148, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 6.036895268355172e-05, "rmse": 3.834593322207515e-06, "relative_l2": 0.0017851456119018848, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.07235199958086014, "p95_ms": 0.07334559969604015, "p99_ms": 0.07383072331547737, "mean_ms": 0.07256448023021221, "n": 300}, {"repeat": 1, "p50_ms": 0.07222399860620499, "p95_ms": 0.07302559725940228, "p99_ms": 0.07364800162613387, "mean_ms": 0.07280693327387174, "n": 300}, {"repeat": 2, "p50_ms": 0.07231999933719635, "p95_ms": 0.073217598721385, "p99_ms": 0.07379712305963039, "mean_ms": 0.07280533323685329, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_89e6d37b3d6f79ec_vllm_fa3.json", "configuration": "480x832_tp1_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_retry_480x832_tp1_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_retry_480x832_tp1_sp2_v1/captures/qkv_rank0_89e6d37b3d6f79ec.pt", "context": {"phase": "commit", "step": null, "start_frame": 27, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_math", "status": "measured", "q_shape": [4680, 20, 128], "k_shape": [512, 20, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.001953125, "rmse": 4.581208904922202e-06, "relative_l2": 0.0016327966506945726, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 6.036895268355172e-05, "rmse": 3.591073235205049e-06, "relative_l2": 0.0016717779668364214, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 1.1255359649658203, "p95_ms": 1.1296000480651855, "p99_ms": 1.1313600540161133, "mean_ms": 1.1257561608155568, "n": 300}, {"repeat": 1, "p50_ms": 1.1250560283660889, "p95_ms": 1.1284480094909668, "p99_ms": 1.1302742707729339, "mean_ms": 1.1257985071341197, "n": 300}, {"repeat": 2, "p50_ms": 1.1251680254936218, "p95_ms": 1.1289023995399474, "p99_ms": 1.1300480365753174, "mean_ms": 1.1256846956411997, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_89e6d37b3d6f79ec_sdpa_math.json", "configuration": "480x832_tp1_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_retry_480x832_tp1_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_retry_480x832_tp1_sp2_v1/captures/qkv_rank0_89e6d37b3d6f79ec.pt", "context": {"phase": "commit", "step": null, "start_frame": 27, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_flash", "status": "measured", "q_shape": [4680, 20, 128], "k_shape": [512, 20, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.000244140625, "rmse": 1.651654615647204e-06, "relative_l2": 0.0005886691003406193, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 6.036895268355172e-05, "rmse": 3.8066900674542536e-06, "relative_l2": 0.0017721556104610306, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.08908800035715103, "p95_ms": 0.0899519994854927, "p99_ms": 0.09046592332422733, "mean_ms": 0.08924597320457299, "n": 300}, {"repeat": 1, "p50_ms": 0.08912000060081482, "p95_ms": 0.08992159925401211, "p99_ms": 0.09046976335346699, "mean_ms": 0.08938314681251844, "n": 300}, {"repeat": 2, "p50_ms": 0.08883199840784073, "p95_ms": 0.08972799777984619, "p99_ms": 0.09053087644279002, "mean_ms": 0.08931679969032605, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_89e6d37b3d6f79ec_sdpa_flash.json", "configuration": "480x832_tp1_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_retry_480x832_tp1_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_retry_480x832_tp1_sp2_v1/captures/qkv_rank0_89e6d37b3d6f79ec.pt", "context": {"phase": "commit", "step": null, "start_frame": 27, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_efficient", "status": "measured", "q_shape": [4680, 20, 128], "k_shape": [512, 20, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.00048828125, "rmse": 2.8886375881373482e-06, "relative_l2": 0.0010295443575850658, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 6.036895268355172e-05, "rmse": 4.3905271507836556e-06, "relative_l2": 0.0020439534570110473, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.16582399606704712, "p95_ms": 0.16662559509277344, "p99_ms": 0.16688031673431397, "mean_ms": 0.16598858684301376, "n": 300}, {"repeat": 1, "p50_ms": 0.16598400473594666, "p95_ms": 0.16668959558010102, "p99_ms": 0.16700799763202667, "mean_ms": 0.16622314696510632, "n": 300}, {"repeat": 2, "p50_ms": 0.16564799845218658, "p95_ms": 0.16649599373340607, "p99_ms": 0.16688031673431397, "mean_ms": 0.16585397337873778, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_89e6d37b3d6f79ec_sdpa_efficient.json", "configuration": "480x832_tp1_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_retry_480x832_tp1_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_retry_480x832_tp1_sp2_v1/captures/qkv_rank0_89e6d37b3d6f79ec.pt", "context": {"phase": "commit", "step": null, "start_frame": 27, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_cudnn", "status": "measured", "q_shape": [4680, 20, 128], "k_shape": [512, 20, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.00048828125, "rmse": 2.8872231276408505e-06, "relative_l2": 0.0010290402272541525, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 6.036895268355172e-05, "rmse": 4.3905005043652074e-06, "relative_l2": 0.0020439410521135865, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.052032001316547394, "p95_ms": 0.053121598437428474, "p99_ms": 0.05392864078283309, "mean_ms": 0.05240309342741966, "n": 300}, {"repeat": 1, "p50_ms": 0.05225599929690361, "p95_ms": 0.0533135998994112, "p99_ms": 0.06499680288136002, "mean_ms": 0.053188479878008364, "n": 300}, {"repeat": 2, "p50_ms": 0.05222399905323982, "p95_ms": 0.05321599915623665, "p99_ms": 0.054127998538315285, "mean_ms": 0.05276447985321284, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_89e6d37b3d6f79ec_sdpa_cudnn.json", "configuration": "480x832_tp1_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_retry_480x832_tp1_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_retry_480x832_tp1_sp2_v1/captures/qkv_rank0_89e6d37b3d6f79ec.pt", "context": {"phase": "commit", "step": null, "start_frame": 27, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "native_dense", "status": "measured", "q_shape": [4680, 20, 128], "k_shape": [512, 20, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.0, "rmse": 0.0, "relative_l2": 0.0, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 6.036895268355172e-05, "rmse": 3.843290768525026e-06, "relative_l2": 0.0017891945961940492, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.1138560026884079, "p95_ms": 0.1582528039813042, "p99_ms": 0.20537855774164182, "mean_ms": 0.12411199984451135, "n": 300}, {"repeat": 1, "p50_ms": 0.11473600193858147, "p95_ms": 0.12985600531101227, "p99_ms": 0.13478400528430937, "mean_ms": 0.11838815987110138, "n": 300}, {"repeat": 2, "p50_ms": 0.11340799927711487, "p95_ms": 0.12395839728415015, "p99_ms": 0.14881216004490846, "mean_ms": 0.11584917336702347, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_89e6d37b3d6f79ec_native_dense.json", "configuration": "480x832_tp1_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_retry_480x832_tp1_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_retry_480x832_tp1_sp2_v1/captures/qkv_rank0_994f8af95ec0756f.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 27, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "vllm_fa2", "status": "measured", "q_shape": [4680, 20, 128], "k_shape": [512, 20, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.00048828125, "rmse": 1.901296061381002e-06, "relative_l2": 0.0005748787165766316, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.00018629015193599718, "rmse": 6.425491210484444e-06, "relative_l2": 0.0018716524974291156, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.08380799740552902, "p95_ms": 0.08476799726486206, "p99_ms": 0.08505823947489262, "mean_ms": 0.0840008532255888, "n": 300}, {"repeat": 1, "p50_ms": 0.08377599716186523, "p95_ms": 0.08470559678971767, "p99_ms": 0.08556960329413414, "mean_ms": 0.08417642615735531, "n": 300}, {"repeat": 2, "p50_ms": 0.08364800363779068, "p95_ms": 0.08457760326564312, "p99_ms": 0.08489663824439049, "mean_ms": 0.0840114134301742, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_994f8af95ec0756f_vllm_fa2.json", "configuration": "480x832_tp1_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_retry_480x832_tp1_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_retry_480x832_tp1_sp2_v1/captures/qkv_rank0_994f8af95ec0756f.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 27, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "vllm_fa3", "status": "measured", "q_shape": [4680, 20, 128], "k_shape": [512, 20, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.0001220703125, "rmse": 7.073425660110257e-07, "relative_l2": 0.00021387315462752316, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.00018629015193599718, "rmse": 6.451693484237799e-06, "relative_l2": 0.001879284840156278, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.07276800274848938, "p95_ms": 0.0737616028636694, "p99_ms": 0.07468896247446537, "mean_ms": 0.07300885347028574, "n": 300}, {"repeat": 1, "p50_ms": 0.07289600372314453, "p95_ms": 0.07388959638774395, "p99_ms": 0.07473888270556925, "mean_ms": 0.0734181334078312, "n": 300}, {"repeat": 2, "p50_ms": 0.07280000299215317, "p95_ms": 0.07385760322213172, "p99_ms": 0.07472672246396538, "mean_ms": 0.07341856012741725, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_994f8af95ec0756f_vllm_fa3.json", "configuration": "480x832_tp1_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_retry_480x832_tp1_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_retry_480x832_tp1_sp2_v1/captures/qkv_rank0_994f8af95ec0756f.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 27, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_math", "status": "measured", "q_shape": [4680, 20, 128], "k_shape": [512, 20, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.0009765625, "rmse": 5.153678906367993e-06, "relative_l2": 0.001558274050801384, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.00018629015193599718, "rmse": 5.611731845242477e-06, "relative_l2": 0.0016346161840378117, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 1.1252480149269104, "p95_ms": 1.1289935648441314, "p99_ms": 1.1301567995548247, "mean_ms": 1.1255576543013255, "n": 300}, {"repeat": 1, "p50_ms": 1.1242879629135132, "p95_ms": 1.1280736327171326, "p99_ms": 1.1296425557136536, "mean_ms": 1.1251450673739116, "n": 300}, {"repeat": 2, "p50_ms": 1.1244479417800903, "p95_ms": 1.1285776376724244, "p99_ms": 1.130401222705841, "mean_ms": 1.1252376528580983, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_994f8af95ec0756f_sdpa_math.json", "configuration": "480x832_tp1_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_retry_480x832_tp1_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_retry_480x832_tp1_sp2_v1/captures/qkv_rank0_994f8af95ec0756f.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 27, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_flash", "status": "measured", "q_shape": [4680, 20, 128], "k_shape": [512, 20, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.00048828125, "rmse": 1.9064821762545377e-06, "relative_l2": 0.0005764467980149067, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.00018629015193599718, "rmse": 6.425491210484444e-06, "relative_l2": 0.0018716524974291156, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.08903999999165535, "p95_ms": 0.08995519950985909, "p99_ms": 0.09078783839941025, "mean_ms": 0.08924383995433649, "n": 300}, {"repeat": 1, "p50_ms": 0.08908800035715103, "p95_ms": 0.0899839997291565, "p99_ms": 0.09099679887294755, "mean_ms": 0.08989301323890686, "n": 300}, {"repeat": 2, "p50_ms": 0.08905600011348724, "p95_ms": 0.09005120024085045, "p99_ms": 0.09034112237393856, "mean_ms": 0.08948053350051244, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_994f8af95ec0756f_sdpa_flash.json", "configuration": "480x832_tp1_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_retry_480x832_tp1_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_retry_480x832_tp1_sp2_v1/captures/qkv_rank0_994f8af95ec0756f.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 27, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_efficient", "status": "measured", "q_shape": [4680, 20, 128], "k_shape": [512, 20, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.00048828125, "rmse": 3.3135834531941932e-06, "relative_l2": 0.0010019000415212637, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.00018629015193599718, "rmse": 6.716977998482818e-06, "relative_l2": 0.0019565583757275027, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.16556799411773682, "p95_ms": 0.16617760583758354, "p99_ms": 0.16681599617004395, "mean_ms": 0.16568085337678592, "n": 300}, {"repeat": 1, "p50_ms": 0.16592000424861908, "p95_ms": 0.1666879951953888, "p99_ms": 0.1670403252542019, "mean_ms": 0.1661716268459956, "n": 300}, {"repeat": 2, "p50_ms": 0.16595199704170227, "p95_ms": 0.16678400337696075, "p99_ms": 0.16710432574152947, "mean_ms": 0.16620416005452474, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_994f8af95ec0756f_sdpa_efficient.json", "configuration": "480x832_tp1_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_retry_480x832_tp1_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_retry_480x832_tp1_sp2_v1/captures/qkv_rank0_994f8af95ec0756f.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 27, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_cudnn", "status": "measured", "q_shape": [4680, 20, 128], "k_shape": [512, 20, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.00048828125, "rmse": 3.3129698590422482e-06, "relative_l2": 0.0010017145142771201, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.00018629015193599718, "rmse": 6.716978124601895e-06, "relative_l2": 0.0019565584124641624, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.052352000027894974, "p95_ms": 0.0533440001308918, "p99_ms": 0.05360063835978508, "mean_ms": 0.05270080013821522, "n": 300}, {"repeat": 1, "p50_ms": 0.052352000027894974, "p95_ms": 0.05331199988722801, "p99_ms": 0.05787360094487665, "mean_ms": 0.05298410670210918, "n": 300}, {"repeat": 2, "p50_ms": 0.05244800075888634, "p95_ms": 0.05347360111773014, "p99_ms": 0.08486911632120589, "mean_ms": 0.05724330684791008, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_994f8af95ec0756f_sdpa_cudnn.json", "configuration": "480x832_tp1_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_retry_480x832_tp1_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_retry_480x832_tp1_sp2_v1/captures/qkv_rank0_994f8af95ec0756f.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 27, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "native_dense", "status": "measured", "q_shape": [4680, 20, 128], "k_shape": [512, 20, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.0, "rmse": 0.0, "relative_l2": 0.0, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.00018629015193599718, "rmse": 6.461484411018777e-06, "relative_l2": 0.001882136795277135, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.114656001329422, "p95_ms": 0.12365920245647431, "p99_ms": 0.14167264550924297, "mean_ms": 0.11624981331328552, "n": 300}, {"repeat": 1, "p50_ms": 0.11406399682164192, "p95_ms": 0.12182080298662187, "p99_ms": 0.1357430373132228, "mean_ms": 0.11589034679035345, "n": 300}, {"repeat": 2, "p50_ms": 0.1136000007390976, "p95_ms": 0.12103679776191711, "p99_ms": 0.15111520260572425, "mean_ms": 0.11643199980258942, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_994f8af95ec0756f_native_dense.json", "configuration": "480x832_tp1_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_retry_480x832_tp1_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_retry_480x832_tp1_sp2_v1/captures/qkv_rank0_b29cb74704615250.pt", "context": {"phase": "commit", "step": null, "start_frame": 0, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "vllm_fa2", "status": "measured", "q_shape": [4680, 20, 128], "k_shape": [512, 20, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.00048828125, "rmse": 1.9189446691626973e-06, "relative_l2": 0.000234833882818655, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.00025586326656432123, "rmse": 1.8789914191711385e-05, "relative_l2": 0.0017170595682592, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.08390399813652039, "p95_ms": 0.08470399677753448, "p99_ms": 0.0851247999817133, "mean_ms": 0.08407743968069553, "n": 300}, {"repeat": 1, "p50_ms": 0.08395199850201607, "p95_ms": 0.08476959727704525, "p99_ms": 0.08550400286912918, "mean_ms": 0.08426805317401886, "n": 300}, {"repeat": 2, "p50_ms": 0.08399999886751175, "p95_ms": 0.08489599823951721, "p99_ms": 0.0864668824523687, "mean_ms": 0.08446463959912459, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_b29cb74704615250_vllm_fa2.json", "configuration": "480x832_tp1_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_retry_480x832_tp1_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_retry_480x832_tp1_sp2_v1/captures/qkv_rank0_b29cb74704615250.pt", "context": {"phase": "commit", "step": null, "start_frame": 0, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "vllm_fa3", "status": "measured", "q_shape": [4680, 20, 128], "k_shape": [512, 20, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.00048828125, "rmse": 7.572041374826215e-07, "relative_l2": 9.266405152212257e-05, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.00025586326656432123, "rmse": 1.8794715046210773e-05, "relative_l2": 0.0017174982798504232, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.0727040022611618, "p95_ms": 0.07379680275917053, "p99_ms": 0.12916096702218052, "mean_ms": 0.07421248016258081, "n": 300}, {"repeat": 1, "p50_ms": 0.07260800153017044, "p95_ms": 0.07366560213267803, "p99_ms": 0.074263679087162, "mean_ms": 0.07324970684945584, "n": 300}, {"repeat": 2, "p50_ms": 0.07267200201749802, "p95_ms": 0.07356960140168667, "p99_ms": 0.07465664193034169, "mean_ms": 0.07326389340062936, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_b29cb74704615250_vllm_fa3.json", "configuration": "480x832_tp1_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_retry_480x832_tp1_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_retry_480x832_tp1_sp2_v1/captures/qkv_rank0_b29cb74704615250.pt", "context": {"phase": "commit", "step": null, "start_frame": 0, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_math", "status": "measured", "q_shape": [4680, 20, 128], "k_shape": [512, 20, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.001953125, "rmse": 5.501883747321388e-06, "relative_l2": 0.0006733017079455921, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.00024379653738876295, "rmse": 1.860982023786444e-05, "relative_l2": 0.0017006022261189715, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 1.1250879764556885, "p95_ms": 1.1293776512145997, "p99_ms": 1.131649545431137, "mean_ms": 1.1254003218809763, "n": 300}, {"repeat": 1, "p50_ms": 1.1242879629135132, "p95_ms": 1.1281007826328278, "p99_ms": 1.1302089750766755, "mean_ms": 1.124858235915502, "n": 300}, {"repeat": 2, "p50_ms": 1.1243360042572021, "p95_ms": 1.127942407131195, "p99_ms": 1.1288351833820343, "mean_ms": 1.1250276295344035, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_b29cb74704615250_sdpa_math.json", "configuration": "480x832_tp1_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_retry_480x832_tp1_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_retry_480x832_tp1_sp2_v1/captures/qkv_rank0_b29cb74704615250.pt", "context": {"phase": "commit", "step": null, "start_frame": 0, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_flash", "status": "measured", "q_shape": [4680, 20, 128], "k_shape": [512, 20, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.00048828125, "rmse": 1.9134515308329295e-06, "relative_l2": 0.0002341616513449866, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.00025586326656432123, "rmse": 1.8789914191711385e-05, "relative_l2": 0.0017170595682592, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.08886399865150452, "p95_ms": 0.08985599875450134, "p99_ms": 0.09043616309762001, "mean_ms": 0.08902207973102728, "n": 300}, {"repeat": 1, "p50_ms": 0.0888959988951683, "p95_ms": 0.08988799899816513, "p99_ms": 0.09030464217066765, "mean_ms": 0.08916874644656976, "n": 300}, {"repeat": 2, "p50_ms": 0.08908800035715103, "p95_ms": 0.08998559974133968, "p99_ms": 0.09037024259567261, "mean_ms": 0.08933322675526142, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_b29cb74704615250_sdpa_flash.json", "configuration": "480x832_tp1_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_retry_480x832_tp1_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_retry_480x832_tp1_sp2_v1/captures/qkv_rank0_b29cb74704615250.pt", "context": {"phase": "commit", "step": null, "start_frame": 0, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_efficient", "status": "measured", "q_shape": [4680, 20, 128], "k_shape": [512, 20, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.0009765625, "rmse": 3.385811819117086e-06, "relative_l2": 0.00041434406564911026, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.00025586326656432123, "rmse": 1.8933874262258647e-05, "relative_l2": 0.0017302149245880482, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.16569599509239197, "p95_ms": 0.16665760204195976, "p99_ms": 0.16713599860668182, "mean_ms": 0.16587253272533417, "n": 300}, {"repeat": 1, "p50_ms": 0.1656319946050644, "p95_ms": 0.16627199947834015, "p99_ms": 0.16668863520026206, "mean_ms": 0.16582826644182205, "n": 300}, {"repeat": 2, "p50_ms": 0.16601599752902985, "p95_ms": 0.16668959558010102, "p99_ms": 0.1670732781291008, "mean_ms": 0.16622208004196484, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_b29cb74704615250_sdpa_efficient.json", "configuration": "480x832_tp1_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_retry_480x832_tp1_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_retry_480x832_tp1_sp2_v1/captures/qkv_rank0_b29cb74704615250.pt", "context": {"phase": "commit", "step": null, "start_frame": 0, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_cudnn", "status": "measured", "q_shape": [4680, 20, 128], "k_shape": [512, 20, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.0009765625, "rmse": 3.3792352600343783e-06, "relative_l2": 0.0004135392488506913, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.00025586326656432123, "rmse": 1.893387426216046e-05, "relative_l2": 0.0017302149245790755, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.05251200124621391, "p95_ms": 0.06647040210664287, "p99_ms": 0.10594560138881198, "mean_ms": 0.056417280038197834, "n": 300}, {"repeat": 1, "p50_ms": 0.052319999784231186, "p95_ms": 0.05331199988722801, "p99_ms": 0.05966303970664738, "mean_ms": 0.052992533159752685, "n": 300}, {"repeat": 2, "p50_ms": 0.052319999784231186, "p95_ms": 0.053279999643564224, "p99_ms": 0.053812479563057375, "mean_ms": 0.05284661337733269, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_b29cb74704615250_sdpa_cudnn.json", "configuration": "480x832_tp1_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_retry_480x832_tp1_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_retry_480x832_tp1_sp2_v1/captures/qkv_rank0_b29cb74704615250.pt", "context": {"phase": "commit", "step": null, "start_frame": 0, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "native_dense", "status": "measured", "q_shape": [4680, 20, 128], "k_shape": [512, 20, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.0, "rmse": 0.0, "relative_l2": 0.0, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.00025586326656432123, "rmse": 1.879633017463274e-05, "relative_l2": 0.0017176458734840402, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.11425599828362465, "p95_ms": 0.12185759656131268, "p99_ms": 0.14159712448716155, "mean_ms": 0.11583146661520004, "n": 300}, {"repeat": 1, "p50_ms": 0.11364800110459328, "p95_ms": 0.12041600048542023, "p99_ms": 0.179828477948904, "mean_ms": 0.11640266661842665, "n": 300}, {"repeat": 2, "p50_ms": 0.11353600025177002, "p95_ms": 0.11952320113778114, "p99_ms": 0.1275011222064495, "mean_ms": 0.11473002667228381, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_b29cb74704615250_native_dense.json", "configuration": "480x832_tp1_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_retry_480x832_tp1_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_retry_480x832_tp1_sp2_v1/captures/qkv_rank0_b3c02cfce9305c38.pt", "context": {"phase": "commit", "step": null, "start_frame": 9, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "vllm_fa2", "status": "measured", "q_shape": [4680, 20, 128], "k_shape": [29640, 20, 128], "q_len": 4680, "k_len": 18720, "reference_error": {"max_abs": 0.0625, "rmse": 0.0004315156573927417, "relative_l2": 0.0005866373094328526, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.012678517544874701, "rmse": 0.0009683098051771197, "relative_l2": 0.0016927232205301553, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 2.445904016494751, "p95_ms": 2.4688032627105714, "p99_ms": 2.473408303260803, "mean_ms": 2.439756048520406, "n": 300}, {"repeat": 1, "p50_ms": 2.458032011985779, "p95_ms": 2.739798295497894, "p99_ms": 2.7666568827629088, "mean_ms": 2.4907458066940307, "n": 300}, {"repeat": 2, "p50_ms": 2.4571200609207153, "p95_ms": 2.5714879035949707, "p99_ms": 2.594081251621246, "mean_ms": 2.461342504819234, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_b3c02cfce9305c38_vllm_fa2.json", "configuration": "480x832_tp1_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_retry_480x832_tp1_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_retry_480x832_tp1_sp2_v1/captures/qkv_rank0_b3c02cfce9305c38.pt", "context": {"phase": "commit", "step": null, "start_frame": 9, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "vllm_fa3", "status": "measured", "q_shape": [4680, 20, 128], "k_shape": [29640, 20, 128], "q_len": 4680, "k_len": 18720, "reference_error": {"max_abs": 0.0, "rmse": 0.0, "relative_l2": 0.0, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.012678517544874701, "rmse": 0.0009692782411752042, "relative_l2": 0.001694416164351224, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 1.35862398147583, "p95_ms": 1.379265594482422, "p99_ms": 1.3802246296405791, "mean_ms": 1.352910614013672, "n": 300}, {"repeat": 1, "p50_ms": 1.370959997177124, "p95_ms": 1.403891146183014, "p99_ms": 1.4071241748332977, "mean_ms": 1.3667474122842154, "n": 300}, {"repeat": 2, "p50_ms": 1.3818399906158447, "p95_ms": 1.4082704484462738, "p99_ms": 1.4492169678211213, "mean_ms": 1.3837840004762014, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_b3c02cfce9305c38_vllm_fa3.json", "configuration": "480x832_tp1_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_retry_480x832_tp1_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_retry_480x832_tp1_sp2_v1/captures/qkv_rank0_b3c02cfce9305c38.pt", "context": {"phase": "commit", "step": null, "start_frame": 9, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_math", "status": "measured", "q_shape": [4680, 20, 128], "k_shape": [29640, 20, 128], "q_len": 4680, "k_len": 18720, "reference_error": {"max_abs": 0.0625, "rmse": 0.0008285306313520479, "relative_l2": 0.0011263715974892076, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.012678517544874701, "rmse": 0.00094767651387136, "relative_l2": 0.001656653719712869, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 35.10524940490723, "p95_ms": 35.135972023010254, "p99_ms": 35.1416487121582, "mean_ms": 35.109353612263995, "n": 300}, {"repeat": 1, "p50_ms": 35.1036319732666, "p95_ms": 35.13098335266113, "p99_ms": 35.13758060455322, "mean_ms": 35.10685464223226, "n": 300}, {"repeat": 2, "p50_ms": 35.10419273376465, "p95_ms": 35.130548286437985, "p99_ms": 35.137216606140136, "mean_ms": 35.10774302164714, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_b3c02cfce9305c38_sdpa_math.json", "configuration": "480x832_tp1_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_retry_480x832_tp1_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_retry_480x832_tp1_sp2_v1/captures/qkv_rank0_b3c02cfce9305c38.pt", "context": {"phase": "commit", "step": null, "start_frame": 9, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_flash", "status": "measured", "q_shape": [4680, 20, 128], "k_shape": [29640, 20, 128], "q_len": 4680, "k_len": 18720, "reference_error": {"max_abs": 0.0625, "rmse": 0.00043194246157900244, "relative_l2": 0.0005872175415871053, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.012678517544874701, "rmse": 0.0009683237781535962, "relative_l2": 0.0016927476469912064, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 2.61628794670105, "p95_ms": 2.6669503450393677, "p99_ms": 2.7053915905952453, "mean_ms": 2.620737385749817, "n": 300}, {"repeat": 1, "p50_ms": 2.636207938194275, "p95_ms": 2.8494896411895754, "p99_ms": 2.8759682989120483, "mean_ms": 2.6639212822914122, "n": 300}, {"repeat": 2, "p50_ms": 2.6607840061187744, "p95_ms": 2.7500656247138977, "p99_ms": 2.7984976077079775, "mean_ms": 2.669575893084208, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_b3c02cfce9305c38_sdpa_flash.json", "configuration": "480x832_tp1_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_retry_480x832_tp1_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_retry_480x832_tp1_sp2_v1/captures/qkv_rank0_b3c02cfce9305c38.pt", "context": {"phase": "commit", "step": null, "start_frame": 9, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_efficient", "status": "measured", "q_shape": [4680, 20, 128], "k_shape": [29640, 20, 128], "q_len": 4680, "k_len": 18720, "reference_error": {"max_abs": 0.0625, "rmse": 0.0007101627096293966, "relative_l2": 0.0009654526645770349, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.012678517544874701, "rmse": 0.0009547840566502768, "relative_l2": 0.0016690785682876319, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 5.395375967025757, "p95_ms": 5.486702537536621, "p99_ms": 5.492487483024597, "mean_ms": 5.402759795188904, "n": 300}, {"repeat": 1, "p50_ms": 5.395520210266113, "p95_ms": 5.404443287849427, "p99_ms": 5.408033862113952, "mean_ms": 5.394683834711711, "n": 300}, {"repeat": 2, "p50_ms": 5.395840167999268, "p95_ms": 5.404292631149292, "p99_ms": 5.406817607879638, "mean_ms": 5.395574943224589, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_b3c02cfce9305c38_sdpa_efficient.json", "configuration": "480x832_tp1_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_retry_480x832_tp1_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_retry_480x832_tp1_sp2_v1/captures/qkv_rank0_b3c02cfce9305c38.pt", "context": {"phase": "commit", "step": null, "start_frame": 9, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_cudnn", "status": "measured", "q_shape": [4680, 20, 128], "k_shape": [29640, 20, 128], "q_len": 4680, "k_len": 18720, "reference_error": {"max_abs": 0.0625, "rmse": 0.000709001569842497, "relative_l2": 0.0009638741171737314, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.012678517544874701, "rmse": 0.0009547032318614938, "relative_l2": 0.0016689372767340032, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 1.6053919792175293, "p95_ms": 1.6595840454101562, "p99_ms": 1.6600028467178345, "mean_ms": 1.6038138643900552, "n": 300}, {"repeat": 1, "p50_ms": 1.6289279460906982, "p95_ms": 1.8017119467258453, "p99_ms": 1.8391376042366028, "mean_ms": 1.6516875716050465, "n": 300}, {"repeat": 2, "p50_ms": 1.609600007534027, "p95_ms": 1.649888038635254, "p99_ms": 1.657359983921051, "mean_ms": 1.6143638368447621, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_b3c02cfce9305c38_sdpa_cudnn.json", "configuration": "480x832_tp1_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_retry_480x832_tp1_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_retry_480x832_tp1_sp2_v1/captures/qkv_rank0_b3c02cfce9305c38.pt", "context": {"phase": "commit", "step": null, "start_frame": 9, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "native_dense", "status": "measured", "q_shape": [4680, 20, 128], "k_shape": [29640, 20, 128], "q_len": 4680, "k_len": 18720, "reference_error": {"max_abs": 0.03125, "rmse": 0.0003809453068340918, "relative_l2": 0.0005178878819658402, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.012678517544874701, "rmse": 0.0009680282200371425, "relative_l2": 0.0016922309754837381, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 1.352783977985382, "p95_ms": 1.3758752167224884, "p99_ms": 1.3830102431774138, "mean_ms": 1.3402715746561686, "n": 300}, {"repeat": 1, "p50_ms": 1.343775987625122, "p95_ms": 1.3552719533443451, "p99_ms": 1.360556446313858, "mean_ms": 1.3413798391819, "n": 300}, {"repeat": 2, "p50_ms": 1.3822879791259766, "p95_ms": 1.5539583802223205, "p99_ms": 1.5713939785957336, "mean_ms": 1.413663891951243, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_b3c02cfce9305c38_native_dense.json", "configuration": "480x832_tp1_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_retry_480x832_tp1_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_retry_480x832_tp1_sp2_v1/captures/qkv_rank0_c345c054e4ffb254.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 27, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "vllm_fa2", "status": "measured", "q_shape": [4680, 20, 128], "k_shape": [29640, 20, 128], "q_len": 4680, "k_len": 28080, "reference_error": {"max_abs": 0.03125, "rmse": 0.00047600064624848655, "relative_l2": 0.0009182524244601488, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.008820491288137777, "rmse": 0.0008120587250742453, "relative_l2": 0.0019695529708380166, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 3.6839520931243896, "p95_ms": 3.9479328989982605, "p99_ms": 4.058370122909546, "mean_ms": 3.71243146498998, "n": 300}, {"repeat": 1, "p50_ms": 3.684623956680298, "p95_ms": 3.845604693889618, "p99_ms": 3.8476245975494385, "mean_ms": 3.7187837807337445, "n": 300}, {"repeat": 2, "p50_ms": 3.688271999359131, "p95_ms": 3.793206441402435, "p99_ms": 3.847650146484375, "mean_ms": 3.717250560124715, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_c345c054e4ffb254_vllm_fa2.json", "configuration": "480x832_tp1_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_retry_480x832_tp1_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_retry_480x832_tp1_sp2_v1/captures/qkv_rank0_c345c054e4ffb254.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 27, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "vllm_fa3", "status": "measured", "q_shape": [4680, 20, 128], "k_shape": [29640, 20, 128], "q_len": 4680, "k_len": 28080, "reference_error": {"max_abs": 0.0, "rmse": 0.0, "relative_l2": 0.0, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.008820491288137777, "rmse": 0.0008167053351597663, "relative_l2": 0.0019808227773380664, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 2.082271933555603, "p95_ms": 2.1177615404129027, "p99_ms": 2.120449070930481, "mean_ms": 2.071820477247238, "n": 300}, {"repeat": 1, "p50_ms": 2.0465439558029175, "p95_ms": 2.1406895756721496, "p99_ms": 2.147629671096802, "mean_ms": 2.0600808588663737, "n": 300}, {"repeat": 2, "p50_ms": 2.0826560258865356, "p95_ms": 2.119281494617462, "p99_ms": 2.127937166690826, "mean_ms": 2.0764374442895255, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_c345c054e4ffb254_vllm_fa3.json", "configuration": "480x832_tp1_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_retry_480x832_tp1_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_retry_480x832_tp1_sp2_v1/captures/qkv_rank0_c345c054e4ffb254.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 27, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_math", "status": "measured", "q_shape": [4680, 20, 128], "k_shape": [29640, 20, 128], "q_len": 4680, "k_len": 28080, "reference_error": {"max_abs": 0.03125, "rmse": 0.0008201883403515623, "relative_l2": 0.0015822245998561235, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.007724527303786921, "rmse": 0.0006858574812207812, "relative_l2": 0.0016634666902770664, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 52.62784004211426, "p95_ms": 52.67095546722412, "p99_ms": 52.67605125427246, "mean_ms": 52.636314417521156, "n": 300}, {"repeat": 1, "p50_ms": 52.63064002990723, "p95_ms": 52.672672271728516, "p99_ms": 52.88285549163818, "mean_ms": 52.642219950358076, "n": 300}, {"repeat": 2, "p50_ms": 52.642560958862305, "p95_ms": 52.9040283203125, "p99_ms": 52.933569793701174, "mean_ms": 52.688408088684085, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_c345c054e4ffb254_sdpa_math.json", "configuration": "480x832_tp1_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_retry_480x832_tp1_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_retry_480x832_tp1_sp2_v1/captures/qkv_rank0_c345c054e4ffb254.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 27, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_flash", "status": "measured", "q_shape": [4680, 20, 128], "k_shape": [29640, 20, 128], "q_len": 4680, "k_len": 28080, "reference_error": {"max_abs": 0.03125, "rmse": 0.00047663553481608236, "relative_l2": 0.0009194771874327448, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.008820491288137777, "rmse": 0.0008120515445332757, "relative_l2": 0.00196953555528005, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 3.9474079608917236, "p95_ms": 4.298471832275391, "p99_ms": 4.379907636642456, "mean_ms": 3.986519996325175, "n": 300}, {"repeat": 1, "p50_ms": 3.948207974433899, "p95_ms": 4.187079811096192, "p99_ms": 4.19320246219635, "mean_ms": 3.9890878907839458, "n": 300}, {"repeat": 2, "p50_ms": 3.9838720560073853, "p95_ms": 4.048215866088867, "p99_ms": 4.087874269485473, "mean_ms": 3.9792317954699197, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_c345c054e4ffb254_sdpa_flash.json", "configuration": "480x832_tp1_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_retry_480x832_tp1_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_retry_480x832_tp1_sp2_v1/captures/qkv_rank0_c345c054e4ffb254.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 27, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_efficient", "status": "measured", "q_shape": [4680, 20, 128], "k_shape": [29640, 20, 128], "q_len": 4680, "k_len": 28080, "reference_error": {"max_abs": 0.03125, "rmse": 0.00067228581527902, "relative_l2": 0.0012969059699298481, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.008820491288137777, "rmse": 0.000811379428513884, "relative_l2": 0.001967905416889973, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 8.07696008682251, "p95_ms": 8.09344973564148, "p99_ms": 8.21803316116333, "mean_ms": 8.081405013402303, "n": 300}, {"repeat": 1, "p50_ms": 8.146095752716064, "p95_ms": 8.159782123565673, "p99_ms": 8.169348106384277, "mean_ms": 8.143524796168009, "n": 300}, {"repeat": 2, "p50_ms": 8.146927833557129, "p95_ms": 8.159142208099365, "p99_ms": 8.167232847213745, "mean_ms": 8.14452254295349, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_c345c054e4ffb254_sdpa_efficient.json", "configuration": "480x832_tp1_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_retry_480x832_tp1_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_retry_480x832_tp1_sp2_v1/captures/qkv_rank0_c345c054e4ffb254.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 27, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_cudnn", "status": "measured", "q_shape": [4680, 20, 128], "k_shape": [29640, 20, 128], "q_len": 4680, "k_len": 28080, "reference_error": {"max_abs": 0.03125, "rmse": 0.0006716624988550779, "relative_l2": 0.0012957035307693096, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.008820491288137777, "rmse": 0.0008098227142425212, "relative_l2": 0.001964129789434424, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 2.4111039638519287, "p95_ms": 2.4587279677391054, "p99_ms": 2.4646733665466307, "mean_ms": 2.4187971250216167, "n": 300}, {"repeat": 1, "p50_ms": 2.4277760982513428, "p95_ms": 2.810499107837677, "p99_ms": 2.891939582824707, "mean_ms": 2.4687581888834633, "n": 300}, {"repeat": 2, "p50_ms": 2.4599199295043945, "p95_ms": 2.5638800978660585, "p99_ms": 2.5725447368621825, "mean_ms": 2.468251523176829, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_c345c054e4ffb254_sdpa_cudnn.json", "configuration": "480x832_tp1_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_retry_480x832_tp1_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_retry_480x832_tp1_sp2_v1/captures/qkv_rank0_c345c054e4ffb254.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 27, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "native_dense", "status": "measured", "q_shape": [4680, 20, 128], "k_shape": [29640, 20, 128], "q_len": 4680, "k_len": 28080, "reference_error": {"max_abs": 0.03125, "rmse": 0.0004435518124668058, "relative_l2": 0.0008556554079943809, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.007724527303786921, "rmse": 0.0008217814009871526, "relative_l2": 0.0019931341782525556, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 2.0194880962371826, "p95_ms": 2.0338255643844603, "p99_ms": 2.034597034454346, "mean_ms": 2.0062180312474567, "n": 300}, {"repeat": 1, "p50_ms": 2.037023901939392, "p95_ms": 2.399084734916687, "p99_ms": 2.4355237746238707, "mean_ms": 2.084420471191406, "n": 300}, {"repeat": 2, "p50_ms": 2.0360480546951294, "p95_ms": 2.236961531639099, "p99_ms": 2.2408701181411743, "mean_ms": 2.0687572276592254, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_c345c054e4ffb254_native_dense.json", "configuration": "480x832_tp1_sp2", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_retry_480x832_tp1_sp2_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_cross_480x832_tp1_sp4_v1/captures/qkv_rank0_13f5e4136c72875f.pt", "context": {"phase": "commit", "step": null, "start_frame": 27, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "vllm_fa2", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [29640, 10, 128], "q_len": 4680, "k_len": 28080, "reference_error": {"max_abs": 0.0625, "rmse": 0.0004278408197932095, "relative_l2": 0.0005383218211689015, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.010101945123846434, "rmse": 0.0009676761183591323, "relative_l2": 0.0016959539687613312, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 1.8908480405807495, "p95_ms": 1.9065775454044342, "p99_ms": 1.9127763259410857, "mean_ms": 1.8931733314196268, "n": 300}, {"repeat": 1, "p50_ms": 1.8906559944152832, "p95_ms": 1.89792160987854, "p99_ms": 1.900320372581482, "mean_ms": 1.891904213031133, "n": 300}, {"repeat": 2, "p50_ms": 1.8912960290908813, "p95_ms": 1.898849654197693, "p99_ms": 1.9004511654376983, "mean_ms": 1.8923169088363647, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_13f5e4136c72875f_vllm_fa2.json", "configuration": "480x832_tp1_sp4", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_retry_480x832_tp1_sp4_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_cross_480x832_tp1_sp4_v1/captures/qkv_rank0_13f5e4136c72875f.pt", "context": {"phase": "commit", "step": null, "start_frame": 27, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "vllm_fa3", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [29640, 10, 128], "q_len": 4680, "k_len": 28080, "reference_error": {"max_abs": 0.0, "rmse": 0.0, "relative_l2": 0.0, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.010101945123846434, "rmse": 0.0009681409449194487, "relative_l2": 0.0016967686260984276, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 1.014128029346466, "p95_ms": 1.025761651992798, "p99_ms": 1.0262093222141266, "mean_ms": 1.0024049057563147, "n": 300}, {"repeat": 1, "p50_ms": 1.0067359805107117, "p95_ms": 1.0167424261569977, "p99_ms": 1.017505897283554, "mean_ms": 1.0039885846773784, "n": 300}, {"repeat": 2, "p50_ms": 1.009328007698059, "p95_ms": 1.0685151636600494, "p99_ms": 1.1502479445934295, "mean_ms": 1.0137070961793264, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_13f5e4136c72875f_vllm_fa3.json", "configuration": "480x832_tp1_sp4", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_retry_480x832_tp1_sp4_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_cross_480x832_tp1_sp4_v1/captures/qkv_rank0_13f5e4136c72875f.pt", "context": {"phase": "commit", "step": null, "start_frame": 27, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_math", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [29640, 10, 128], "q_len": 4680, "k_len": 28080, "reference_error": {"max_abs": 0.0625, "rmse": 0.0007962888877749233, "relative_l2": 0.0010019139465251164, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.010101945123846434, "rmse": 0.0009455398630386555, "relative_l2": 0.0016571578577981548, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 27.923664093017578, "p95_ms": 28.05120038986206, "p99_ms": 28.086005668640137, "mean_ms": 27.93373088200887, "n": 300}, {"repeat": 1, "p50_ms": 28.052592277526855, "p95_ms": 28.08032855987549, "p99_ms": 28.23908239364624, "mean_ms": 28.016718101501464, "n": 300}, {"repeat": 2, "p50_ms": 28.06169605255127, "p95_ms": 28.238605880737303, "p99_ms": 28.397269382476807, "mean_ms": 28.07273384730021, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_13f5e4136c72875f_sdpa_math.json", "configuration": "480x832_tp1_sp4", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_retry_480x832_tp1_sp4_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_cross_480x832_tp1_sp4_v1/captures/qkv_rank0_13f5e4136c72875f.pt", "context": {"phase": "commit", "step": null, "start_frame": 27, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_flash", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [29640, 10, 128], "q_len": 4680, "k_len": 28080, "reference_error": {"max_abs": 0.0625, "rmse": 0.00042904017884032254, "relative_l2": 0.0005398308897678007, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.010101945123846434, "rmse": 0.0009676761183591323, "relative_l2": 0.0016959539687613312, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 2.048959970474243, "p95_ms": 2.066596782207489, "p99_ms": 2.0803671073913574, "mean_ms": 2.051882985432943, "n": 300}, {"repeat": 1, "p50_ms": 2.0554879903793335, "p95_ms": 2.0728320717811584, "p99_ms": 2.0808045363426206, "mean_ms": 2.0583241605758666, "n": 300}, {"repeat": 2, "p50_ms": 2.0497119426727295, "p95_ms": 2.065217661857605, "p99_ms": 2.068898935317993, "mean_ms": 2.052366939385732, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_13f5e4136c72875f_sdpa_flash.json", "configuration": "480x832_tp1_sp4", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_retry_480x832_tp1_sp4_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_cross_480x832_tp1_sp4_v1/captures/qkv_rank0_13f5e4136c72875f.pt", "context": {"phase": "commit", "step": null, "start_frame": 27, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_efficient", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [29640, 10, 128], "q_len": 4680, "k_len": 28080, "reference_error": {"max_abs": 0.0625, "rmse": 0.0006798250225789598, "relative_l2": 0.0008553757082079703, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.010101945123846434, "rmse": 0.0009517787171752579, "relative_l2": 0.001668092104528802, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 4.103808164596558, "p95_ms": 4.139294505119324, "p99_ms": 4.153506188392639, "mean_ms": 4.108807902336121, "n": 300}, {"repeat": 1, "p50_ms": 4.107519865036011, "p95_ms": 4.146244621276855, "p99_ms": 4.156269636154175, "mean_ms": 4.1118670018514, "n": 300}, {"repeat": 2, "p50_ms": 4.113216161727905, "p95_ms": 4.148769688606262, "p99_ms": 4.1560007762908935, "mean_ms": 4.114489491780599, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_13f5e4136c72875f_sdpa_efficient.json", "configuration": "480x832_tp1_sp4", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_retry_480x832_tp1_sp4_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_cross_480x832_tp1_sp4_v1/captures/qkv_rank0_13f5e4136c72875f.pt", "context": {"phase": "commit", "step": null, "start_frame": 27, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_cudnn", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [29640, 10, 128], "q_len": 4680, "k_len": 28080, "reference_error": {"max_abs": 0.0625, "rmse": 0.0006799944528355498, "relative_l2": 0.0008555888903076417, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.010101945123846434, "rmse": 0.0009516249798776634, "relative_l2": 0.0016678226637778573, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 1.1967999935150146, "p95_ms": 1.2112031996250152, "p99_ms": 1.2152021872997283, "mean_ms": 1.1895007995764415, "n": 300}, {"repeat": 1, "p50_ms": 1.1916799545288086, "p95_ms": 1.2180208265781403, "p99_ms": 1.2184643363952636, "mean_ms": 1.1863747203350068, "n": 300}, {"repeat": 2, "p50_ms": 1.242576003074646, "p95_ms": 1.4237264335155488, "p99_ms": 1.4480973160266877, "mean_ms": 1.2692056540648142, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_13f5e4136c72875f_sdpa_cudnn.json", "configuration": "480x832_tp1_sp4", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_retry_480x832_tp1_sp4_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_cross_480x832_tp1_sp4_v1/captures/qkv_rank0_13f5e4136c72875f.pt", "context": {"phase": "commit", "step": null, "start_frame": 27, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "native_dense", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [29640, 10, 128], "q_len": 4680, "k_len": 28080, "reference_error": {"max_abs": 0.0625, "rmse": 0.00037934621429436765, "relative_l2": 0.0004773044914956324, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.010101945123846434, "rmse": 0.0009673749392795276, "relative_l2": 0.0016954261208113068, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.992576003074646, "p95_ms": 1.0056111752986907, "p99_ms": 1.0069226050376892, "mean_ms": 0.9798791460196177, "n": 300}, {"repeat": 1, "p50_ms": 0.9927039742469788, "p95_ms": 1.0023727774620057, "p99_ms": 1.0030733036994934, "mean_ms": 0.9871550919612249, "n": 300}, {"repeat": 2, "p50_ms": 0.9932799935340881, "p95_ms": 1.1519807994365692, "p99_ms": 1.15283203125, "mean_ms": 1.0054287999868392, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_13f5e4136c72875f_native_dense.json", "configuration": "480x832_tp1_sp4", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_retry_480x832_tp1_sp4_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_cross_480x832_tp1_sp4_v1/captures/qkv_rank0_2419ddf76b22ab88.pt", "context": {"phase": "commit", "step": null, "start_frame": 9, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "vllm_fa2", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [512, 10, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.000244140625, "rmse": 1.1651005849709376e-06, "relative_l2": 0.000735961118544523, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 8.546650151804169e-05, "rmse": 4.432017925885755e-06, "relative_l2": 0.001751077866033789, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.05238400027155876, "p95_ms": 0.06019680015742779, "p99_ms": 0.06634399801492684, "mean_ms": 0.05330847995976607, "n": 300}, {"repeat": 1, "p50_ms": 0.0525440014898777, "p95_ms": 0.05776799991726876, "p99_ms": 0.06655487991869447, "mean_ms": 0.05383936005334059, "n": 300}, {"repeat": 2, "p50_ms": 0.05222399905323982, "p95_ms": 0.05732000060379505, "p99_ms": 0.06173440016806119, "mean_ms": 0.053099413327872755, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_2419ddf76b22ab88_vllm_fa2.json", "configuration": "480x832_tp1_sp4", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_retry_480x832_tp1_sp4_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_cross_480x832_tp1_sp4_v1/captures/qkv_rank0_2419ddf76b22ab88.pt", "context": {"phase": "commit", "step": null, "start_frame": 9, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "vllm_fa3", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [512, 10, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.0001220703125, "rmse": 4.452483364756388e-07, "relative_l2": 0.00028125079325307616, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 8.546650151804169e-05, "rmse": 4.471097426135494e-06, "relative_l2": 0.0017665180670996055, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.06496000289916992, "p95_ms": 0.073369599878788, "p99_ms": 0.09241856329143047, "mean_ms": 0.0664642133563757, "n": 300}, {"repeat": 1, "p50_ms": 0.06431999802589417, "p95_ms": 0.07187520302832126, "p99_ms": 0.07891455739736557, "mean_ms": 0.06558272015303374, "n": 300}, {"repeat": 2, "p50_ms": 0.06473600119352341, "p95_ms": 0.07418399825692178, "p99_ms": 0.10152000106871074, "mean_ms": 0.06720149339487155, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_2419ddf76b22ab88_vllm_fa3.json", "configuration": "480x832_tp1_sp4", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_retry_480x832_tp1_sp4_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_cross_480x832_tp1_sp4_v1/captures/qkv_rank0_2419ddf76b22ab88.pt", "context": {"phase": "commit", "step": null, "start_frame": 9, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_math", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [512, 10, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.00048828125, "rmse": 2.337693456847005e-06, "relative_l2": 0.001476654903025604, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 8.546650151804169e-05, "rmse": 4.251983876886612e-06, "relative_l2": 0.0016799469176471491, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.6164319813251495, "p95_ms": 0.6191695839166641, "p99_ms": 0.6208694142103195, "mean_ms": 0.6171064533789953, "n": 300}, {"repeat": 1, "p50_ms": 0.617247998714447, "p95_ms": 0.6199039816856384, "p99_ms": 0.6217289632558822, "mean_ms": 0.6178068282206853, "n": 300}, {"repeat": 2, "p50_ms": 0.6171519756317139, "p95_ms": 0.6198416233062745, "p99_ms": 0.6207369631528854, "mean_ms": 0.6180003205935161, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_2419ddf76b22ab88_sdpa_math.json", "configuration": "480x832_tp1_sp4", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_retry_480x832_tp1_sp4_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_cross_480x832_tp1_sp4_v1/captures/qkv_rank0_2419ddf76b22ab88.pt", "context": {"phase": "commit", "step": null, "start_frame": 9, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_flash", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [512, 10, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.000244140625, "rmse": 1.1646446035996768e-06, "relative_l2": 0.0007356730879964678, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 8.546650151804169e-05, "rmse": 4.432017925885755e-06, "relative_l2": 0.001751077866033789, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.0490880012512207, "p95_ms": 0.04940799996256828, "p99_ms": 0.04973983820527785, "mean_ms": 0.04923285350203514, "n": 300}, {"repeat": 1, "p50_ms": 0.0490880012512207, "p95_ms": 0.04964640159159899, "p99_ms": 0.05114880118519058, "mean_ms": 0.04944960019240777, "n": 300}, {"repeat": 2, "p50_ms": 0.0490880012512207, "p95_ms": 0.049375999718904495, "p99_ms": 0.05105600118637083, "mean_ms": 0.04937930688261986, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_2419ddf76b22ab88_sdpa_flash.json", "configuration": "480x832_tp1_sp4", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_retry_480x832_tp1_sp4_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_cross_480x832_tp1_sp4_v1/captures/qkv_rank0_2419ddf76b22ab88.pt", "context": {"phase": "commit", "step": null, "start_frame": 9, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_efficient", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [512, 10, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.000244140625, "rmse": 2.0393986634254526e-06, "relative_l2": 0.001288230510613159, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 8.546650151804169e-05, "rmse": 5.084117071383684e-06, "relative_l2": 0.002008720411536978, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.0865280032157898, "p95_ms": 0.08694559894502163, "p99_ms": 0.0873923223465681, "mean_ms": 0.08666762761771679, "n": 300}, {"repeat": 1, "p50_ms": 0.0865280032157898, "p95_ms": 0.08684799820184708, "p99_ms": 0.08716800063848495, "mean_ms": 0.08675530766447385, "n": 300}, {"repeat": 2, "p50_ms": 0.08649600297212601, "p95_ms": 0.08681759797036648, "p99_ms": 0.08700831942260265, "mean_ms": 0.08673354762295882, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_2419ddf76b22ab88_sdpa_efficient.json", "configuration": "480x832_tp1_sp4", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_retry_480x832_tp1_sp4_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_cross_480x832_tp1_sp4_v1/captures/qkv_rank0_2419ddf76b22ab88.pt", "context": {"phase": "commit", "step": null, "start_frame": 9, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_cudnn", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [512, 10, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.000244140625, "rmse": 2.0372705397818046e-06, "relative_l2": 0.0012868862350395446, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 8.546650151804169e-05, "rmse": 5.0836227950647876e-06, "relative_l2": 0.002008525124348118, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.04745600000023842, "p95_ms": 0.05158400163054466, "p99_ms": 0.05408799830824136, "mean_ms": 0.04789919993529717, "n": 300}, {"repeat": 1, "p50_ms": 0.047648001462221146, "p95_ms": 0.05299680102616549, "p99_ms": 0.0665683201700449, "mean_ms": 0.048841813368101915, "n": 300}, {"repeat": 2, "p50_ms": 0.04742399975657463, "p95_ms": 0.05063040163367987, "p99_ms": 0.054714879207313036, "mean_ms": 0.04809055992712577, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_2419ddf76b22ab88_sdpa_cudnn.json", "configuration": "480x832_tp1_sp4", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_retry_480x832_tp1_sp4_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_cross_480x832_tp1_sp4_v1/captures/qkv_rank0_2419ddf76b22ab88.pt", "context": {"phase": "commit", "step": null, "start_frame": 9, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "native_dense", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [512, 10, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.0, "rmse": 0.0, "relative_l2": 0.0, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 8.546650151804169e-05, "rmse": 4.48689709008337e-06, "relative_l2": 0.0017727604700619917, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.10726399719715118, "p95_ms": 0.11252319999039173, "p99_ms": 0.1256556752324104, "mean_ms": 0.10800704004863898, "n": 300}, {"repeat": 1, "p50_ms": 0.10737599804997444, "p95_ms": 0.11927679926156998, "p99_ms": 0.15268255278468132, "mean_ms": 0.10963466644287109, "n": 300}, {"repeat": 2, "p50_ms": 0.10697600245475769, "p95_ms": 0.11405599676072598, "p99_ms": 0.13188191741704927, "mean_ms": 0.10883381322026253, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_2419ddf76b22ab88_native_dense.json", "configuration": "480x832_tp1_sp4", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_retry_480x832_tp1_sp4_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_cross_480x832_tp1_sp4_v1/captures/qkv_rank0_5d233cbdc47812ff.pt", "context": {"phase": "commit", "step": null, "start_frame": 0, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "vllm_fa2", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [29640, 10, 128], "q_len": 4680, "k_len": 4680, "reference_error": {"max_abs": 0.03125, "rmse": 0.0004155959679669699, "relative_l2": 0.0005516502103045647, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.013453558575000635, "rmse": 0.000855206902521633, "relative_l2": 0.0017184019899498433, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.33059200644493103, "p95_ms": 0.33164799213409424, "p99_ms": 0.3321603161096573, "mean_ms": 0.33078218668699266, "n": 300}, {"repeat": 1, "p50_ms": 0.33049601316452026, "p95_ms": 0.33164959251880644, "p99_ms": 0.33203231513500214, "mean_ms": 0.3308971729874611, "n": 300}, {"repeat": 2, "p50_ms": 0.33055999875068665, "p95_ms": 0.3317455857992172, "p99_ms": 0.33228831708431245, "mean_ms": 0.3312244266271591, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_5d233cbdc47812ff_vllm_fa2.json", "configuration": "480x832_tp1_sp4", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_retry_480x832_tp1_sp4_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_cross_480x832_tp1_sp4_v1/captures/qkv_rank0_5d233cbdc47812ff.pt", "context": {"phase": "commit", "step": null, "start_frame": 0, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "vllm_fa3", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [29640, 10, 128], "q_len": 4680, "k_len": 4680, "reference_error": {"max_abs": 0.0, "rmse": 0.0, "relative_l2": 0.0, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.013453558575000635, "rmse": 0.0008552870221296227, "relative_l2": 0.001718562977534716, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.17244799435138702, "p95_ms": 0.1731856070458889, "p99_ms": 0.17369855642318727, "mean_ms": 0.17265909309188524, "n": 300}, {"repeat": 1, "p50_ms": 0.1751679927110672, "p95_ms": 0.1815039962530136, "p99_ms": 0.18176159828901292, "mean_ms": 0.17683818687995276, "n": 300}, {"repeat": 2, "p50_ms": 0.18412800133228302, "p95_ms": 0.18508799374103546, "p99_ms": 0.18534399569034576, "mean_ms": 0.18396736005942027, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_5d233cbdc47812ff_vllm_fa3.json", "configuration": "480x832_tp1_sp4", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_retry_480x832_tp1_sp4_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_cross_480x832_tp1_sp4_v1/captures/qkv_rank0_5d233cbdc47812ff.pt", "context": {"phase": "commit", "step": null, "start_frame": 0, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_math", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [29640, 10, 128], "q_len": 4680, "k_len": 4680, "reference_error": {"max_abs": 0.03125, "rmse": 0.0007357542109918866, "relative_l2": 0.0009766191120949496, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.013453558575000635, "rmse": 0.000840414736449667, "relative_l2": 0.0016886794894195205, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 4.801136016845703, "p95_ms": 4.829057407379151, "p99_ms": 4.83213604927063, "mean_ms": 4.805223461786906, "n": 300}, {"repeat": 1, "p50_ms": 4.800384044647217, "p95_ms": 4.8056321144104, "p99_ms": 4.807554740905761, "mean_ms": 4.801752525965373, "n": 300}, {"repeat": 2, "p50_ms": 4.800415992736816, "p95_ms": 4.805475091934204, "p99_ms": 4.807427906990052, "mean_ms": 4.801097270647685, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_5d233cbdc47812ff_sdpa_math.json", "configuration": "480x832_tp1_sp4", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_retry_480x832_tp1_sp4_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_cross_480x832_tp1_sp4_v1/captures/qkv_rank0_5d233cbdc47812ff.pt", "context": {"phase": "commit", "step": null, "start_frame": 0, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_flash", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [29640, 10, 128], "q_len": 4680, "k_len": 4680, "reference_error": {"max_abs": 0.03125, "rmse": 0.00041651617644221817, "relative_l2": 0.0005528716687354007, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.013453558575000635, "rmse": 0.000855206902521633, "relative_l2": 0.0017184019899498433, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.35705599188804626, "p95_ms": 0.3585279881954193, "p99_ms": 0.35916959315538405, "mean_ms": 0.3571753602226575, "n": 300}, {"repeat": 1, "p50_ms": 0.3560959994792938, "p95_ms": 0.3572160005569458, "p99_ms": 0.35757279574871065, "mean_ms": 0.3563962670167287, "n": 300}, {"repeat": 2, "p50_ms": 0.35625600814819336, "p95_ms": 0.3575040102005005, "p99_ms": 0.35782559782266615, "mean_ms": 0.3566573864221573, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_5d233cbdc47812ff_sdpa_flash.json", "configuration": "480x832_tp1_sp4", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_retry_480x832_tp1_sp4_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_cross_480x832_tp1_sp4_v1/captures/qkv_rank0_5d233cbdc47812ff.pt", "context": {"phase": "commit", "step": null, "start_frame": 0, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_efficient", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [29640, 10, 128], "q_len": 4680, "k_len": 4680, "reference_error": {"max_abs": 0.03125, "rmse": 0.0005937389373871991, "relative_l2": 0.0007881120966546252, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.013453558575000635, "rmse": 0.0008510924395452369, "relative_l2": 0.0017101346322550366, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.6987519860267639, "p95_ms": 0.701761594414711, "p99_ms": 0.7032319903373718, "mean_ms": 0.6992904520034791, "n": 300}, {"repeat": 1, "p50_ms": 0.6986240148544312, "p95_ms": 0.7014399766921997, "p99_ms": 0.7040044760704041, "mean_ms": 0.699074558019638, "n": 300}, {"repeat": 2, "p50_ms": 0.6983360052108765, "p95_ms": 0.7011520266532898, "p99_ms": 0.7036825388669967, "mean_ms": 0.6989249082406362, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_5d233cbdc47812ff_sdpa_efficient.json", "configuration": "480x832_tp1_sp4", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_retry_480x832_tp1_sp4_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_cross_480x832_tp1_sp4_v1/captures/qkv_rank0_5d233cbdc47812ff.pt", "context": {"phase": "commit", "step": null, "start_frame": 0, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_cudnn", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [29640, 10, 128], "q_len": 4680, "k_len": 4680, "reference_error": {"max_abs": 0.03125, "rmse": 0.0005926419217863406, "relative_l2": 0.0007866559494983345, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.013453558575000635, "rmse": 0.000851012888406189, "relative_l2": 0.0017099747869179142, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.19705599546432495, "p95_ms": 0.20524800494313242, "p99_ms": 0.207136645168066, "mean_ms": 0.19778698638081552, "n": 300}, {"repeat": 1, "p50_ms": 0.2035840004682541, "p95_ms": 0.2096640020608902, "p99_ms": 0.2103696073591709, "mean_ms": 0.2051459198196729, "n": 300}, {"repeat": 2, "p50_ms": 0.20291200280189514, "p95_ms": 0.20742399990558624, "p99_ms": 0.20784127563238144, "mean_ms": 0.2033286400139332, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_5d233cbdc47812ff_sdpa_cudnn.json", "configuration": "480x832_tp1_sp4", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_retry_480x832_tp1_sp4_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_cross_480x832_tp1_sp4_v1/captures/qkv_rank0_5d233cbdc47812ff.pt", "context": {"phase": "commit", "step": null, "start_frame": 0, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "native_dense", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [29640, 10, 128], "q_len": 4680, "k_len": 4680, "reference_error": {"max_abs": 0.03125, "rmse": 0.0003681619947681451, "relative_l2": 0.0004886877099253684, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.013453558575000635, "rmse": 0.0008563707767124402, "relative_l2": 0.0017207406096682262, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.16206400096416473, "p95_ms": 0.17274880409240723, "p99_ms": 0.17389023795723915, "mean_ms": 0.16285994693636893, "n": 300}, {"repeat": 1, "p50_ms": 0.1732959970831871, "p95_ms": 0.1740799993276596, "p99_ms": 0.1748227174580097, "mean_ms": 0.17413631985584896, "n": 300}, {"repeat": 2, "p50_ms": 0.1674879938364029, "p95_ms": 0.17276959717273713, "p99_ms": 0.1742124803364277, "mean_ms": 0.16793258721629778, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_5d233cbdc47812ff_native_dense.json", "configuration": "480x832_tp1_sp4", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_retry_480x832_tp1_sp4_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_cross_480x832_tp1_sp4_v1/captures/qkv_rank0_6b4ac7dbc60c0c5a.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 0, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "vllm_fa2", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [29640, 10, 128], "q_len": 4680, "k_len": 4680, "reference_error": {"max_abs": 0.03125, "rmse": 0.0005266509856320316, "relative_l2": 0.0009067576358098154, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.006950112781240136, "rmse": 0.0007533224446526795, "relative_l2": 0.0019169673843472507, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.33430400490760803, "p95_ms": 0.33612799644470215, "p99_ms": 0.33683582693338393, "mean_ms": 0.334333972632885, "n": 300}, {"repeat": 1, "p50_ms": 0.3301120102405548, "p95_ms": 0.3336319923400879, "p99_ms": 0.3345283192396164, "mean_ms": 0.3307521065076192, "n": 300}, {"repeat": 2, "p50_ms": 0.3309440016746521, "p95_ms": 0.3320319950580597, "p99_ms": 0.3326419124007225, "mean_ms": 0.33119584023952486, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_6b4ac7dbc60c0c5a_vllm_fa2.json", "configuration": "480x832_tp1_sp4", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_retry_480x832_tp1_sp4_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_cross_480x832_tp1_sp4_v1/captures/qkv_rank0_6b4ac7dbc60c0c5a.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 0, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "vllm_fa3", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [29640, 10, 128], "q_len": 4680, "k_len": 4680, "reference_error": {"max_abs": 0.0, "rmse": 0.0, "relative_l2": 0.0, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.006950112781240136, "rmse": 0.0007627068873890609, "relative_l2": 0.0019408478232929027, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.17323199659585953, "p95_ms": 0.1740799993276596, "p99_ms": 0.18694399297237396, "mean_ms": 0.17389653384685516, "n": 300}, {"repeat": 1, "p50_ms": 0.18675200641155243, "p95_ms": 0.18806559443473816, "p99_ms": 0.18848032459616662, "mean_ms": 0.18716480056444804, "n": 300}, {"repeat": 2, "p50_ms": 0.18561600148677826, "p95_ms": 0.18860960528254508, "p99_ms": 0.1890579141676426, "mean_ms": 0.1858622933924198, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_6b4ac7dbc60c0c5a_vllm_fa3.json", "configuration": "480x832_tp1_sp4", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_retry_480x832_tp1_sp4_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_cross_480x832_tp1_sp4_v1/captures/qkv_rank0_6b4ac7dbc60c0c5a.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 0, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_math", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [29640, 10, 128], "q_len": 4680, "k_len": 4680, "reference_error": {"max_abs": 0.03125, "rmse": 0.0008676879616324845, "relative_l2": 0.0014939356541150204, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.006950112781240136, "rmse": 0.0006602534478209683, "relative_l2": 0.0016801362203659833, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 4.802863836288452, "p95_ms": 4.807492756843567, "p99_ms": 4.818234453201294, "mean_ms": 4.804253118832906, "n": 300}, {"repeat": 1, "p50_ms": 4.802783966064453, "p95_ms": 4.807295799255371, "p99_ms": 4.808747687339783, "mean_ms": 4.8040245262781776, "n": 300}, {"repeat": 2, "p50_ms": 4.802608013153076, "p95_ms": 4.808000087738037, "p99_ms": 4.810469026565552, "mean_ms": 4.803711349169413, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_6b4ac7dbc60c0c5a_sdpa_math.json", "configuration": "480x832_tp1_sp4", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_retry_480x832_tp1_sp4_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_cross_480x832_tp1_sp4_v1/captures/qkv_rank0_6b4ac7dbc60c0c5a.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 0, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_flash", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [29640, 10, 128], "q_len": 4680, "k_len": 4680, "reference_error": {"max_abs": 0.03125, "rmse": 0.0005271479915826146, "relative_l2": 0.0009076133523147284, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.006950112781240136, "rmse": 0.0007533268077478835, "relative_l2": 0.0019169784870447213, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.35761600732803345, "p95_ms": 0.36022720783948897, "p99_ms": 0.36096416652202606, "mean_ms": 0.357839359442393, "n": 300}, {"repeat": 1, "p50_ms": 0.35686400532722473, "p95_ms": 0.3579519987106323, "p99_ms": 0.3587216046452522, "mean_ms": 0.3571911460161209, "n": 300}, {"repeat": 2, "p50_ms": 0.356224000453949, "p95_ms": 0.35760320276021956, "p99_ms": 0.35856575578451155, "mean_ms": 0.35667135973771413, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_6b4ac7dbc60c0c5a_sdpa_flash.json", "configuration": "480x832_tp1_sp4", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_retry_480x832_tp1_sp4_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_cross_480x832_tp1_sp4_v1/captures/qkv_rank0_6b4ac7dbc60c0c5a.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 0, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_efficient", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [29640, 10, 128], "q_len": 4680, "k_len": 4680, "reference_error": {"max_abs": 0.03125, "rmse": 0.0007438730221660497, "relative_l2": 0.0012807581516485914, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.006950112781240136, "rmse": 0.0007546674581946931, "relative_l2": 0.001920390018452802, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.6986240148544312, "p95_ms": 0.7011600255966186, "p99_ms": 0.70292285323143, "mean_ms": 0.6989981863896052, "n": 300}, {"repeat": 1, "p50_ms": 0.6988800168037415, "p95_ms": 0.7029871881008148, "p99_ms": 0.7057952100038528, "mean_ms": 0.699609813094139, "n": 300}, {"repeat": 2, "p50_ms": 0.6990880072116852, "p95_ms": 0.7040384024381637, "p99_ms": 0.7052518379688263, "mean_ms": 0.6998602638641993, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_6b4ac7dbc60c0c5a_sdpa_efficient.json", "configuration": "480x832_tp1_sp4", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_retry_480x832_tp1_sp4_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_cross_480x832_tp1_sp4_v1/captures/qkv_rank0_6b4ac7dbc60c0c5a.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 0, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_cudnn", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [29640, 10, 128], "q_len": 4680, "k_len": 4680, "reference_error": {"max_abs": 0.03125, "rmse": 0.0007431919192459065, "relative_l2": 0.0012795854674792623, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.006950112781240136, "rmse": 0.0007544565760142951, "relative_l2": 0.0019198533899949183, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.20204800367355347, "p95_ms": 0.20358560010790824, "p99_ms": 0.20406527668237687, "mean_ms": 0.20170005321502685, "n": 300}, {"repeat": 1, "p50_ms": 0.20416000485420227, "p95_ms": 0.21040159985423088, "p99_ms": 0.21075231552124024, "mean_ms": 0.20590784053007763, "n": 300}, {"repeat": 2, "p50_ms": 0.20908799767494202, "p95_ms": 0.211776003241539, "p99_ms": 0.21222400665283203, "mean_ms": 0.20832330668965976, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_6b4ac7dbc60c0c5a_sdpa_cudnn.json", "configuration": "480x832_tp1_sp4", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_retry_480x832_tp1_sp4_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_cross_480x832_tp1_sp4_v1/captures/qkv_rank0_6b4ac7dbc60c0c5a.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 0, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "native_dense", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [29640, 10, 128], "q_len": 4680, "k_len": 4680, "reference_error": {"max_abs": 0.03125, "rmse": 0.0004880581341662283, "relative_l2": 0.0008403106648385286, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.006950112781240136, "rmse": 0.0007607101265716498, "relative_l2": 0.0019357666984857367, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.166703999042511, "p95_ms": 0.17443199455738068, "p99_ms": 0.1748841579258442, "mean_ms": 0.1671404793858528, "n": 300}, {"repeat": 1, "p50_ms": 0.1719360053539276, "p95_ms": 0.17596960589289665, "p99_ms": 0.17657664343714713, "mean_ms": 0.17270069335897764, "n": 300}, {"repeat": 2, "p50_ms": 0.1714559942483902, "p95_ms": 0.176129599660635, "p99_ms": 0.17744096249341965, "mean_ms": 0.17163765360911687, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_6b4ac7dbc60c0c5a_native_dense.json", "configuration": "480x832_tp1_sp4", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_retry_480x832_tp1_sp4_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_cross_480x832_tp1_sp4_v1/captures/qkv_rank0_6bef1f209a2b6a6e.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 0, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "vllm_fa2", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [512, 10, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.00048828125, "rmse": 1.8211860529862304e-06, "relative_l2": 0.0002644321561886288, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.0006142374604239942, "rmse": 1.9996831763657214e-05, "relative_l2": 0.0020143029066056256, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.04979199916124344, "p95_ms": 0.0541439987719059, "p99_ms": 0.05903104178607456, "mean_ms": 0.05135904004176458, "n": 300}, {"repeat": 1, "p50_ms": 0.04955200105905533, "p95_ms": 0.054529601521790025, "p99_ms": 0.061539198979735345, "mean_ms": 0.05043989329288403, "n": 300}, {"repeat": 2, "p50_ms": 0.04960000142455101, "p95_ms": 0.05392800066620112, "p99_ms": 0.05950976174324743, "mean_ms": 0.05031861333797375, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_6bef1f209a2b6a6e_vllm_fa2.json", "configuration": "480x832_tp1_sp4", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_retry_480x832_tp1_sp4_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_cross_480x832_tp1_sp4_v1/captures/qkv_rank0_6bef1f209a2b6a6e.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 0, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "vllm_fa3", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [512, 10, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.00048828125, "rmse": 7.027818468172491e-07, "relative_l2": 0.0001020423579344844, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.0006142374604239942, "rmse": 2.0004182011344022e-05, "relative_l2": 0.0020150433051574963, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.06088000163435936, "p95_ms": 0.0696976013481617, "p99_ms": 0.08492831803858274, "mean_ms": 0.06243509324888388, "n": 300}, {"repeat": 1, "p50_ms": 0.06019200012087822, "p95_ms": 0.06591200232505799, "p99_ms": 0.07145824022591114, "mean_ms": 0.061230080090463164, "n": 300}, {"repeat": 2, "p50_ms": 0.05990400165319443, "p95_ms": 0.06493440270423889, "p99_ms": 0.07763680227100846, "mean_ms": 0.06130080003291369, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_6bef1f209a2b6a6e_vllm_fa3.json", "configuration": "480x832_tp1_sp4", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_retry_480x832_tp1_sp4_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_cross_480x832_tp1_sp4_v1/captures/qkv_rank0_6bef1f209a2b6a6e.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 0, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_math", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [512, 10, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.001953125, "rmse": 1.475276971591166e-05, "relative_l2": 0.002142069284648931, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.0004727375432675984, "rmse": 1.7092660525431627e-05, "relative_l2": 0.001721762536432087, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.6181439757347107, "p95_ms": 0.6213759779930115, "p99_ms": 0.6225954926013947, "mean_ms": 0.6185405868291854, "n": 300}, {"repeat": 1, "p50_ms": 0.6177439987659454, "p95_ms": 0.6203855782747268, "p99_ms": 0.6213766181468964, "mean_ms": 0.6184658141930898, "n": 300}, {"repeat": 2, "p50_ms": 0.6178400218486786, "p95_ms": 0.6201024264097214, "p99_ms": 0.621218239068985, "mean_ms": 0.618397228717804, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_6bef1f209a2b6a6e_sdpa_math.json", "configuration": "480x832_tp1_sp4", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_retry_480x832_tp1_sp4_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_cross_480x832_tp1_sp4_v1/captures/qkv_rank0_6bef1f209a2b6a6e.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 0, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_flash", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [512, 10, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.00048828125, "rmse": 1.8211270710366762e-06, "relative_l2": 0.000264423592141001, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.0006142374604239942, "rmse": 1.9996831763657214e-05, "relative_l2": 0.0020143029066056256, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.04972799867391586, "p95_ms": 0.05440800059586764, "p99_ms": 0.06541055887937545, "mean_ms": 0.050421226633091766, "n": 300}, {"repeat": 1, "p50_ms": 0.04972799867391586, "p95_ms": 0.05507839843630791, "p99_ms": 0.0669116825610399, "mean_ms": 0.05070229332894087, "n": 300}, {"repeat": 2, "p50_ms": 0.04902400076389313, "p95_ms": 0.05011200159788132, "p99_ms": 0.0634393577277659, "mean_ms": 0.050168426781892775, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_6bef1f209a2b6a6e_sdpa_flash.json", "configuration": "480x832_tp1_sp4", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_retry_480x832_tp1_sp4_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_cross_480x832_tp1_sp4_v1/captures/qkv_rank0_6bef1f209a2b6a6e.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 0, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_efficient", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [512, 10, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.0009765625, "rmse": 3.4240173283262414e-06, "relative_l2": 0.0004971596852896511, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.0006142374604239942, "rmse": 2.0105374323793734e-05, "relative_l2": 0.0020252365183375982, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.08646400272846222, "p95_ms": 0.08681599795818329, "p99_ms": 0.08761599659919739, "mean_ms": 0.08690208114683629, "n": 300}, {"repeat": 1, "p50_ms": 0.08646400272846222, "p95_ms": 0.08685119822621346, "p99_ms": 0.08758463636040688, "mean_ms": 0.08696181431412697, "n": 300}, {"repeat": 2, "p50_ms": 0.08671999722719193, "p95_ms": 0.08707359991967678, "p99_ms": 0.08765407688915729, "mean_ms": 0.08697994617124398, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_6bef1f209a2b6a6e_sdpa_efficient.json", "configuration": "480x832_tp1_sp4", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_retry_480x832_tp1_sp4_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_cross_480x832_tp1_sp4_v1/captures/qkv_rank0_6bef1f209a2b6a6e.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 0, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_cudnn", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [512, 10, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.0009765625, "rmse": 3.402406925503428e-06, "relative_l2": 0.0004940219029608378, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.0006142374604239942, "rmse": 2.0105374115138647e-05, "relative_l2": 0.0020252364973195416, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.04742399975657463, "p95_ms": 0.06441919878125191, "p99_ms": 0.06738335885107515, "mean_ms": 0.05051957346498966, "n": 300}, {"repeat": 1, "p50_ms": 0.046959999948740005, "p95_ms": 0.052139198407530785, "p99_ms": 0.06161951966583727, "mean_ms": 0.04789354672034581, "n": 300}, {"repeat": 2, "p50_ms": 0.04684799909591675, "p95_ms": 0.05101440083235503, "p99_ms": 0.0536076783761382, "mean_ms": 0.04747594663252433, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_6bef1f209a2b6a6e_sdpa_cudnn.json", "configuration": "480x832_tp1_sp4", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_retry_480x832_tp1_sp4_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_cross_480x832_tp1_sp4_v1/captures/qkv_rank0_6bef1f209a2b6a6e.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 0, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "native_dense", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [512, 10, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.0, "rmse": 0.0, "relative_l2": 0.0, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.0006142374604239942, "rmse": 2.001643072823305e-05, "relative_l2": 0.002016277131911822, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.10719999670982361, "p95_ms": 0.12288160361349584, "p99_ms": 0.16607935518026318, "mean_ms": 0.11138848004241785, "n": 300}, {"repeat": 1, "p50_ms": 0.1088160015642643, "p95_ms": 0.13043680116534234, "p99_ms": 0.17380063861608436, "mean_ms": 0.1144806394726038, "n": 300}, {"repeat": 2, "p50_ms": 0.10899200290441513, "p95_ms": 0.17300640568137168, "p99_ms": 0.17955552324652665, "mean_ms": 0.12674890674650668, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_6bef1f209a2b6a6e_native_dense.json", "configuration": "480x832_tp1_sp4", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_retry_480x832_tp1_sp4_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_cross_480x832_tp1_sp4_v1/captures/qkv_rank0_6c601440672149ef.pt", "context": {"phase": "commit", "step": null, "start_frame": 0, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "vllm_fa2", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [512, 10, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.00048828125, "rmse": 1.9783499656596346e-06, "relative_l2": 0.00028311028721459693, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.00025586326656432123, "rmse": 1.8789914191711385e-05, "relative_l2": 0.0017170595682592, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.04980799928307533, "p95_ms": 0.05475359968841076, "p99_ms": 0.07741152033209794, "mean_ms": 0.050796906625231106, "n": 300}, {"repeat": 1, "p50_ms": 0.049855999648571014, "p95_ms": 0.05456479825079441, "p99_ms": 0.057122881785035017, "mean_ms": 0.050670719953874746, "n": 300}, {"repeat": 2, "p50_ms": 0.04983999952673912, "p95_ms": 0.06092000082135204, "p99_ms": 0.07498143725097178, "mean_ms": 0.05172234671811263, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_6c601440672149ef_vllm_fa2.json", "configuration": "480x832_tp1_sp4", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_retry_480x832_tp1_sp4_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_cross_480x832_tp1_sp4_v1/captures/qkv_rank0_6c601440672149ef.pt", "context": {"phase": "commit", "step": null, "start_frame": 0, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "vllm_fa3", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [512, 10, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.00048828125, "rmse": 8.137099459926773e-07, "relative_l2": 0.00011644535118565316, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.00025586326656432123, "rmse": 1.8794715046210773e-05, "relative_l2": 0.0017174982798504232, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.06111999973654747, "p95_ms": 0.07021279819309711, "p99_ms": 0.08403039865195745, "mean_ms": 0.062499733480314415, "n": 300}, {"repeat": 1, "p50_ms": 0.06059199944138527, "p95_ms": 0.06765120103955269, "p99_ms": 0.0730399969965219, "mean_ms": 0.061967573203146456, "n": 300}, {"repeat": 2, "p50_ms": 0.060816001147031784, "p95_ms": 0.06900959648191929, "p99_ms": 0.07395775668323036, "mean_ms": 0.06215882666409016, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_6c601440672149ef_vllm_fa3.json", "configuration": "480x832_tp1_sp4", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_retry_480x832_tp1_sp4_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_cross_480x832_tp1_sp4_v1/captures/qkv_rank0_6c601440672149ef.pt", "context": {"phase": "commit", "step": null, "start_frame": 0, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_math", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [512, 10, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.0009765625, "rmse": 5.799299816110414e-06, "relative_l2": 0.0008299044481925851, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.00024379653738876295, "rmse": 1.860982023786444e-05, "relative_l2": 0.0017006022261189715, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.6168000102043152, "p95_ms": 0.6188480257987976, "p99_ms": 0.6195561760663986, "mean_ms": 0.6172839486598969, "n": 300}, {"repeat": 1, "p50_ms": 0.6171519756317139, "p95_ms": 0.6192351996898651, "p99_ms": 0.6200966674089432, "mean_ms": 0.6176817075411478, "n": 300}, {"repeat": 2, "p50_ms": 0.6170400083065033, "p95_ms": 0.6191072255373001, "p99_ms": 0.6210563403367996, "mean_ms": 0.6175805888573329, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_6c601440672149ef_sdpa_math.json", "configuration": "480x832_tp1_sp4", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_retry_480x832_tp1_sp4_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_cross_480x832_tp1_sp4_v1/captures/qkv_rank0_6c601440672149ef.pt", "context": {"phase": "commit", "step": null, "start_frame": 0, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_flash", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [512, 10, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.00048828125, "rmse": 1.9676565329363196e-06, "relative_l2": 0.00028158001154944263, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.00025586326656432123, "rmse": 1.8789914191711385e-05, "relative_l2": 0.0017170595682592, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.04915200173854828, "p95_ms": 0.05008320137858391, "p99_ms": 0.05047712031751867, "mean_ms": 0.04934282677869002, "n": 300}, {"repeat": 1, "p50_ms": 0.04915200173854828, "p95_ms": 0.04950400069355965, "p99_ms": 0.05032095935195683, "mean_ms": 0.04949546677370866, "n": 300}, {"repeat": 2, "p50_ms": 0.04912000149488449, "p95_ms": 0.04940799996256828, "p99_ms": 0.0514640004187822, "mean_ms": 0.04941055999447902, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_6c601440672149ef_sdpa_flash.json", "configuration": "480x832_tp1_sp4", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_retry_480x832_tp1_sp4_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_cross_480x832_tp1_sp4_v1/captures/qkv_rank0_6c601440672149ef.pt", "context": {"phase": "commit", "step": null, "start_frame": 0, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_efficient", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [512, 10, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.0009765625, "rmse": 3.547975701944155e-06, "relative_l2": 0.0005077304003050362, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.00025586326656432123, "rmse": 1.8933874262258647e-05, "relative_l2": 0.0017302149245880482, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.08671999722719193, "p95_ms": 0.0870399996638298, "p99_ms": 0.08713792040944099, "mean_ms": 0.08683733239769936, "n": 300}, {"repeat": 1, "p50_ms": 0.08671999722719193, "p95_ms": 0.08700799942016602, "p99_ms": 0.0871363203972578, "mean_ms": 0.086972265864412, "n": 300}, {"repeat": 2, "p50_ms": 0.08649600297212601, "p95_ms": 0.08681599795818329, "p99_ms": 0.0870086394250393, "mean_ms": 0.08673653456072013, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_6c601440672149ef_sdpa_efficient.json", "configuration": "480x832_tp1_sp4", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_retry_480x832_tp1_sp4_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_cross_480x832_tp1_sp4_v1/captures/qkv_rank0_6c601440672149ef.pt", "context": {"phase": "commit", "step": null, "start_frame": 0, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_cudnn", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [512, 10, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.0009765625, "rmse": 3.53635267141311e-06, "relative_l2": 0.0005060670952432085, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.00025586326656432123, "rmse": 1.893387426216046e-05, "relative_l2": 0.0017302149245790755, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.0469760000705719, "p95_ms": 0.05236799996346236, "p99_ms": 0.06629279777407641, "mean_ms": 0.04908245335022608, "n": 300}, {"repeat": 1, "p50_ms": 0.046911999583244324, "p95_ms": 0.05203840117901564, "p99_ms": 0.056252479702234244, "mean_ms": 0.047724053499599295, "n": 300}, {"repeat": 2, "p50_ms": 0.047200001776218414, "p95_ms": 0.05196800082921982, "p99_ms": 0.05764127925038337, "mean_ms": 0.048018026798963546, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_6c601440672149ef_sdpa_cudnn.json", "configuration": "480x832_tp1_sp4", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_retry_480x832_tp1_sp4_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_cross_480x832_tp1_sp4_v1/captures/qkv_rank0_6c601440672149ef.pt", "context": {"phase": "commit", "step": null, "start_frame": 0, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "native_dense", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [512, 10, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.0, "rmse": 0.0, "relative_l2": 0.0, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.00025586326656432123, "rmse": 1.879633017463274e-05, "relative_l2": 0.0017176458734840402, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.10808000341057777, "p95_ms": 0.11673920229077339, "p99_ms": 0.12263616219162937, "mean_ms": 0.10954805351793766, "n": 300}, {"repeat": 1, "p50_ms": 0.10779200121760368, "p95_ms": 0.11942560002207757, "p99_ms": 0.15744992479681943, "mean_ms": 0.113271146590511, "n": 300}, {"repeat": 2, "p50_ms": 0.1082879975438118, "p95_ms": 0.11831199936568737, "p99_ms": 0.13233472168445587, "mean_ms": 0.11027957305312157, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_6c601440672149ef_native_dense.json", "configuration": "480x832_tp1_sp4", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_retry_480x832_tp1_sp4_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_cross_480x832_tp1_sp4_v1/captures/qkv_rank0_7baca46400faa6d9.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 27, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "vllm_fa2", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [512, 10, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.0001220703125, "rmse": 1.2177752203517502e-06, "relative_l2": 0.0007455754449189619, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.00018629015193599718, "rmse": 6.425491210484444e-06, "relative_l2": 0.0018716524974291156, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.04934399947524071, "p95_ms": 0.055747199803590775, "p99_ms": 0.061960321813821724, "mean_ms": 0.050178346695999305, "n": 300}, {"repeat": 1, "p50_ms": 0.04948800057172775, "p95_ms": 0.05739200096577406, "p99_ms": 0.06488384187221523, "mean_ms": 0.05085920003553231, "n": 300}, {"repeat": 2, "p50_ms": 0.049215998500585556, "p95_ms": 0.05597120132297278, "p99_ms": 0.05914079908281558, "mean_ms": 0.050213013365864756, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_7baca46400faa6d9_vllm_fa2.json", "configuration": "480x832_tp1_sp4", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_retry_480x832_tp1_sp4_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_cross_480x832_tp1_sp4_v1/captures/qkv_rank0_7baca46400faa6d9.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 27, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "vllm_fa3", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [512, 10, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 6.103515625e-05, "rmse": 4.5961343682608566e-07, "relative_l2": 0.0002813955210497394, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.00018629015193599718, "rmse": 6.451693484237799e-06, "relative_l2": 0.001879284840156278, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.060736000537872314, "p95_ms": 0.06699200309813023, "p99_ms": 0.07778944350779055, "mean_ms": 0.061795093193650244, "n": 300}, {"repeat": 1, "p50_ms": 0.06038400158286095, "p95_ms": 0.06552479974925518, "p99_ms": 0.08571711689233778, "mean_ms": 0.0614716799557209, "n": 300}, {"repeat": 2, "p50_ms": 0.060447998344898224, "p95_ms": 0.07167200148105622, "p99_ms": 0.08981439828872678, "mean_ms": 0.06252277343223492, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_7baca46400faa6d9_vllm_fa3.json", "configuration": "480x832_tp1_sp4", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_retry_480x832_tp1_sp4_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_cross_480x832_tp1_sp4_v1/captures/qkv_rank0_7baca46400faa6d9.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 27, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_math", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [512, 10, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.000244140625, "rmse": 3.1722574616132708e-06, "relative_l2": 0.0019421952662632134, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.00018629015193599718, "rmse": 5.611731845242477e-06, "relative_l2": 0.0016346161840378117, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.6171360015869141, "p95_ms": 0.6197840094566345, "p99_ms": 0.6214739209413529, "mean_ms": 0.6174533355236054, "n": 300}, {"repeat": 1, "p50_ms": 0.617328017950058, "p95_ms": 0.6197792112827301, "p99_ms": 0.6204201459884644, "mean_ms": 0.6179863472779592, "n": 300}, {"repeat": 2, "p50_ms": 0.6171839833259583, "p95_ms": 0.6196831881999969, "p99_ms": 0.6208639740943909, "mean_ms": 0.617784747282664, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_7baca46400faa6d9_sdpa_math.json", "configuration": "480x832_tp1_sp4", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_retry_480x832_tp1_sp4_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_cross_480x832_tp1_sp4_v1/captures/qkv_rank0_7baca46400faa6d9.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 27, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_flash", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [512, 10, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.0001220703125, "rmse": 1.217766212863948e-06, "relative_l2": 0.0007455699301395399, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.00018629015193599718, "rmse": 6.425491210484444e-06, "relative_l2": 0.0018716524974291156, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.049215998500585556, "p95_ms": 0.05020799860358238, "p99_ms": 0.05050368078052997, "mean_ms": 0.04944266660759846, "n": 300}, {"repeat": 1, "p50_ms": 0.04915200173854828, "p95_ms": 0.050177598372101784, "p99_ms": 0.05070752102881654, "mean_ms": 0.049575466675062976, "n": 300}, {"repeat": 2, "p50_ms": 0.049247998744249344, "p95_ms": 0.05027199909090996, "p99_ms": 0.05070816151797761, "mean_ms": 0.04961791989703973, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_7baca46400faa6d9_sdpa_flash.json", "configuration": "480x832_tp1_sp4", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_retry_480x832_tp1_sp4_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_cross_480x832_tp1_sp4_v1/captures/qkv_rank0_7baca46400faa6d9.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 27, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_efficient", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [512, 10, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.0001220703125, "rmse": 2.16051630927381e-06, "relative_l2": 0.0013227629217781355, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.00018629015193599718, "rmse": 6.716977998482818e-06, "relative_l2": 0.0019565583757275027, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.08684799820184708, "p95_ms": 0.08713600039482117, "p99_ms": 0.08736000210046768, "mean_ms": 0.08696735881268979, "n": 300}, {"repeat": 1, "p50_ms": 0.08686399832367897, "p95_ms": 0.08713600039482117, "p99_ms": 0.08739360235631466, "mean_ms": 0.08717973209917546, "n": 300}, {"repeat": 2, "p50_ms": 0.08691199868917465, "p95_ms": 0.08726400136947632, "p99_ms": 0.0875839963555336, "mean_ms": 0.08711903909842174, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_7baca46400faa6d9_sdpa_efficient.json", "configuration": "480x832_tp1_sp4", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_retry_480x832_tp1_sp4_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_cross_480x832_tp1_sp4_v1/captures/qkv_rank0_7baca46400faa6d9.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 27, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_cudnn", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [512, 10, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.0001220703125, "rmse": 2.1591474379475984e-06, "relative_l2": 0.0013219248386647496, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.00018629015193599718, "rmse": 6.716978124601895e-06, "relative_l2": 0.0019565584124641624, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.04726399853825569, "p95_ms": 0.0526799988001585, "p99_ms": 0.06472383946180324, "mean_ms": 0.048377706706523894, "n": 300}, {"repeat": 1, "p50_ms": 0.04838399961590767, "p95_ms": 0.054019201360642914, "p99_ms": 0.06804000221192816, "mean_ms": 0.051983146630227565, "n": 300}, {"repeat": 2, "p50_ms": 0.04700800031423569, "p95_ms": 0.05226079933345318, "p99_ms": 0.05644416116178033, "mean_ms": 0.047799253389239314, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_7baca46400faa6d9_sdpa_cudnn.json", "configuration": "480x832_tp1_sp4", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_retry_480x832_tp1_sp4_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_cross_480x832_tp1_sp4_v1/captures/qkv_rank0_7baca46400faa6d9.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 27, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "native_dense", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [512, 10, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.0, "rmse": 0.0, "relative_l2": 0.0, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.00018629015193599718, "rmse": 6.461484411018777e-06, "relative_l2": 0.001882136795277135, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.10927999764680862, "p95_ms": 0.11778400242328645, "p99_ms": 0.1331593532860279, "mean_ms": 0.11105397333701451, "n": 300}, {"repeat": 1, "p50_ms": 0.11059200018644333, "p95_ms": 0.17758240327239036, "p99_ms": 0.24591680392622875, "mean_ms": 0.13490858676532905, "n": 300}, {"repeat": 2, "p50_ms": 0.10768000036478043, "p95_ms": 0.11606559716165066, "p99_ms": 0.14532000258564934, "mean_ms": 0.10990464021762213, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_7baca46400faa6d9_native_dense.json", "configuration": "480x832_tp1_sp4", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_retry_480x832_tp1_sp4_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_cross_480x832_tp1_sp4_v1/captures/qkv_rank0_8ec3a76708f5e241.pt", "context": {"phase": "commit", "step": null, "start_frame": 27, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "vllm_fa2", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [512, 10, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.000244140625, "rmse": 1.1765090791105027e-06, "relative_l2": 0.0008027447067766402, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 6.036895268355172e-05, "rmse": 3.8066916294210143e-06, "relative_l2": 0.0017721563376145706, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.04966399818658829, "p95_ms": 0.054207999259233475, "p99_ms": 0.05945631824433804, "mean_ms": 0.05025877336661021, "n": 300}, {"repeat": 1, "p50_ms": 0.04972799867391586, "p95_ms": 0.0542191993445158, "p99_ms": 0.060541758388280784, "mean_ms": 0.05053418674816688, "n": 300}, {"repeat": 2, "p50_ms": 0.04992000013589859, "p95_ms": 0.0559088010340929, "p99_ms": 0.07110783755779267, "mean_ms": 0.05127488009631634, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_8ec3a76708f5e241_vllm_fa2.json", "configuration": "480x832_tp1_sp4", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_retry_480x832_tp1_sp4_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_cross_480x832_tp1_sp4_v1/captures/qkv_rank0_8ec3a76708f5e241.pt", "context": {"phase": "commit", "step": null, "start_frame": 27, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "vllm_fa3", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [512, 10, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.0001220703125, "rmse": 4.4691778534592407e-07, "relative_l2": 0.00030493677687725085, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 6.036895268355172e-05, "rmse": 3.834593322207515e-06, "relative_l2": 0.0017851456119018848, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.061039999127388, "p95_ms": 0.06704159639775753, "p99_ms": 0.07584512352943416, "mean_ms": 0.06194549333304167, "n": 300}, {"repeat": 1, "p50_ms": 0.060655999928712845, "p95_ms": 0.0690016031265259, "p99_ms": 0.0912732820212841, "mean_ms": 0.06250271997104088, "n": 300}, {"repeat": 2, "p50_ms": 0.0607680007815361, "p95_ms": 0.06871200166642666, "p99_ms": 0.09341760098934143, "mean_ms": 0.06405856003363927, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_8ec3a76708f5e241_vllm_fa3.json", "configuration": "480x832_tp1_sp4", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_retry_480x832_tp1_sp4_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_cross_480x832_tp1_sp4_v1/captures/qkv_rank0_8ec3a76708f5e241.pt", "context": {"phase": "commit", "step": null, "start_frame": 27, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_math", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [512, 10, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.000244140625, "rmse": 2.2249764824750405e-06, "relative_l2": 0.001518125211035103, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 6.036895268355172e-05, "rmse": 3.591073235205049e-06, "relative_l2": 0.0016717779668364214, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.6178080141544342, "p95_ms": 0.6199727982282639, "p99_ms": 0.6212806540727616, "mean_ms": 0.6183509339888891, "n": 300}, {"repeat": 1, "p50_ms": 0.6177600026130676, "p95_ms": 0.6201935917139053, "p99_ms": 0.6214083057641983, "mean_ms": 0.6184918415546418, "n": 300}, {"repeat": 2, "p50_ms": 0.6179039776325226, "p95_ms": 0.6200976252555848, "p99_ms": 0.6210579401254654, "mean_ms": 0.6186085345347723, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_8ec3a76708f5e241_sdpa_math.json", "configuration": "480x832_tp1_sp4", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_retry_480x832_tp1_sp4_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_cross_480x832_tp1_sp4_v1/captures/qkv_rank0_8ec3a76708f5e241.pt", "context": {"phase": "commit", "step": null, "start_frame": 27, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_flash", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [512, 10, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.000244140625, "rmse": 1.1767946218868126e-06, "relative_l2": 0.0008029395356617809, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 6.036895268355172e-05, "rmse": 3.8066900674542536e-06, "relative_l2": 0.0017721556104610306, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.049056001007556915, "p95_ms": 0.05008000135421753, "p99_ms": 0.05069504175335162, "mean_ms": 0.049292053431272506, "n": 300}, {"repeat": 1, "p50_ms": 0.0490880012512207, "p95_ms": 0.05008000135421753, "p99_ms": 0.06055039837956419, "mean_ms": 0.04958261345823606, "n": 300}, {"repeat": 2, "p50_ms": 0.04918399825692177, "p95_ms": 0.050144001841545105, "p99_ms": 0.050614720545708994, "mean_ms": 0.04959999997168779, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_8ec3a76708f5e241_sdpa_flash.json", "configuration": "480x832_tp1_sp4", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_retry_480x832_tp1_sp4_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_cross_480x832_tp1_sp4_v1/captures/qkv_rank0_8ec3a76708f5e241.pt", "context": {"phase": "commit", "step": null, "start_frame": 27, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_efficient", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [512, 10, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.0001220703125, "rmse": 2.0326950438507145e-06, "relative_l2": 0.0013869295323891096, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 6.036895268355172e-05, "rmse": 4.3905271507836556e-06, "relative_l2": 0.0020439534570110473, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.0870399996638298, "p95_ms": 0.08742560259997845, "p99_ms": 0.08777727782726288, "mean_ms": 0.08718656008442242, "n": 300}, {"repeat": 1, "p50_ms": 0.08691199868917465, "p95_ms": 0.0872960016131401, "p99_ms": 0.08771871738135814, "mean_ms": 0.08720245271921158, "n": 300}, {"repeat": 2, "p50_ms": 0.08687999844551086, "p95_ms": 0.08726400136947632, "p99_ms": 0.08752352327108383, "mean_ms": 0.08711423926055431, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_8ec3a76708f5e241_sdpa_efficient.json", "configuration": "480x832_tp1_sp4", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_retry_480x832_tp1_sp4_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_cross_480x832_tp1_sp4_v1/captures/qkv_rank0_8ec3a76708f5e241.pt", "context": {"phase": "commit", "step": null, "start_frame": 27, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_cudnn", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [512, 10, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.0001220703125, "rmse": 2.0331514618785623e-06, "relative_l2": 0.0013872409512828844, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 6.036895268355172e-05, "rmse": 4.3905005043652074e-06, "relative_l2": 0.0020439410521135865, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.047648001462221146, "p95_ms": 0.05307520162314177, "p99_ms": 0.0589769595861432, "mean_ms": 0.04884245326121648, "n": 300}, {"repeat": 1, "p50_ms": 0.04710400104522705, "p95_ms": 0.052292799577116966, "p99_ms": 0.05539296034723514, "mean_ms": 0.047982613382240136, "n": 300}, {"repeat": 2, "p50_ms": 0.04694399982690811, "p95_ms": 0.052075201459229, "p99_ms": 0.054633278697729104, "mean_ms": 0.047696746637423834, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_8ec3a76708f5e241_sdpa_cudnn.json", "configuration": "480x832_tp1_sp4", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_retry_480x832_tp1_sp4_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_cross_480x832_tp1_sp4_v1/captures/qkv_rank0_8ec3a76708f5e241.pt", "context": {"phase": "commit", "step": null, "start_frame": 27, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "native_dense", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [512, 10, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.0, "rmse": 0.0, "relative_l2": 0.0, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 6.036895268355172e-05, "rmse": 3.843290768525026e-06, "relative_l2": 0.0017891945961940492, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.10804800316691399, "p95_ms": 0.11791999638080597, "p99_ms": 0.12391167730093001, "mean_ms": 0.11006015993654727, "n": 300}, {"repeat": 1, "p50_ms": 0.10763199999928474, "p95_ms": 0.1184704001992941, "p99_ms": 0.15269568711519232, "mean_ms": 0.11593770653009415, "n": 300}, {"repeat": 2, "p50_ms": 0.10740799829363823, "p95_ms": 0.11885440312325955, "p99_ms": 0.14434368297457684, "mean_ms": 0.11055807987848917, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_8ec3a76708f5e241_native_dense.json", "configuration": "480x832_tp1_sp4", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_retry_480x832_tp1_sp4_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_cross_480x832_tp1_sp4_v1/captures/qkv_rank0_a7ce4999e4f2ba20.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 9, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "vllm_fa2", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [512, 10, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.0001220703125, "rmse": 1.2212824683851846e-06, "relative_l2": 0.0007223260174700013, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 7.629575161298766e-05, "rmse": 4.942429805964212e-06, "relative_l2": 0.0018663346713588256, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.04947200044989586, "p95_ms": 0.061447998508811, "p99_ms": 0.07753728181123734, "mean_ms": 0.051176960120598476, "n": 300}, {"repeat": 1, "p50_ms": 0.04918399825692177, "p95_ms": 0.0574464013800025, "p99_ms": 0.08000031799077985, "mean_ms": 0.050777919987837476, "n": 300}, {"repeat": 2, "p50_ms": 0.04887999966740608, "p95_ms": 0.05770239979028702, "p99_ms": 0.08109791904687881, "mean_ms": 0.05071296013891697, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_a7ce4999e4f2ba20_vllm_fa2.json", "configuration": "480x832_tp1_sp4", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_retry_480x832_tp1_sp4_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_cross_480x832_tp1_sp4_v1/captures/qkv_rank0_a7ce4999e4f2ba20.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 9, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "vllm_fa3", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [512, 10, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 6.103515625e-05, "rmse": 4.6673282897397813e-07, "relative_l2": 0.00027604855903732764, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 7.629575161298766e-05, "rmse": 4.97898511900548e-06, "relative_l2": 0.0018801384987938588, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.060816001147031784, "p95_ms": 0.07164320126175881, "p99_ms": 0.09021792091429227, "mean_ms": 0.06251637314756711, "n": 300}, {"repeat": 1, "p50_ms": 0.06019200012087822, "p95_ms": 0.06717599742114544, "p99_ms": 0.07607487805187702, "mean_ms": 0.06144746658702691, "n": 300}, {"repeat": 2, "p50_ms": 0.06083200126886368, "p95_ms": 0.07509759813547134, "p99_ms": 0.10830752298235868, "mean_ms": 0.06390431992709637, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_a7ce4999e4f2ba20_vllm_fa3.json", "configuration": "480x832_tp1_sp4", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_retry_480x832_tp1_sp4_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_cross_480x832_tp1_sp4_v1/captures/qkv_rank0_a7ce4999e4f2ba20.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 9, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_math", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [512, 10, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.000244140625, "rmse": 3.1736303075170077e-06, "relative_l2": 0.0018770397514851157, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 5.955835214807581e-05, "rmse": 4.4437602027578725e-06, "relative_l2": 0.0016780296459857492, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.6165120005607605, "p95_ms": 0.6189135819673538, "p99_ms": 0.6196262121200562, "mean_ms": 0.6169827181100845, "n": 300}, {"repeat": 1, "p50_ms": 0.6166080236434937, "p95_ms": 0.6191727846860886, "p99_ms": 0.6205763435363769, "mean_ms": 0.6172877873977025, "n": 300}, {"repeat": 2, "p50_ms": 0.616703987121582, "p95_ms": 0.618785610795021, "p99_ms": 0.6198419457674027, "mean_ms": 0.6172960009177526, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_a7ce4999e4f2ba20_sdpa_math.json", "configuration": "480x832_tp1_sp4", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_retry_480x832_tp1_sp4_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_cross_480x832_tp1_sp4_v1/captures/qkv_rank0_a7ce4999e4f2ba20.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 9, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_flash", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [512, 10, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.0001220703125, "rmse": 1.2210431593591066e-06, "relative_l2": 0.0007221844784401483, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 7.629575161298766e-05, "rmse": 4.942430280889429e-06, "relative_l2": 0.001866334850697619, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.04919999837875366, "p95_ms": 0.050303999334573746, "p99_ms": 0.050592961497604845, "mean_ms": 0.049403520027796424, "n": 300}, {"repeat": 1, "p50_ms": 0.04912000149488449, "p95_ms": 0.05011200159788132, "p99_ms": 0.05068160053342561, "mean_ms": 0.04961493340631326, "n": 300}, {"repeat": 2, "p50_ms": 0.04912000149488449, "p95_ms": 0.05014560166746378, "p99_ms": 0.05066336173564192, "mean_ms": 0.04953514683991671, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_a7ce4999e4f2ba20_sdpa_flash.json", "configuration": "480x832_tp1_sp4", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_retry_480x832_tp1_sp4_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_cross_480x832_tp1_sp4_v1/captures/qkv_rank0_a7ce4999e4f2ba20.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 9, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_efficient", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [512, 10, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.000244140625, "rmse": 2.168452619391683e-06, "relative_l2": 0.0012825286411493597, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 7.629575161298766e-05, "rmse": 5.409689151586598e-06, "relative_l2": 0.002042778718414212, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.08697599917650223, "p95_ms": 0.08736160211265087, "p99_ms": 0.0875206432491541, "mean_ms": 0.08710154610375563, "n": 300}, {"repeat": 1, "p50_ms": 0.08707199990749359, "p95_ms": 0.08745600283145905, "p99_ms": 0.08761727660894394, "mean_ms": 0.08732298657298088, "n": 300}, {"repeat": 2, "p50_ms": 0.08707199990749359, "p95_ms": 0.08739200234413147, "p99_ms": 0.08775295756757258, "mean_ms": 0.08727274636427561, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_a7ce4999e4f2ba20_sdpa_efficient.json", "configuration": "480x832_tp1_sp4", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_retry_480x832_tp1_sp4_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_cross_480x832_tp1_sp4_v1/captures/qkv_rank0_a7ce4999e4f2ba20.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 9, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_cudnn", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [512, 10, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.000244140625, "rmse": 2.1674583666215477e-06, "relative_l2": 0.001281940591568365, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 7.629575161298766e-05, "rmse": 5.409689151586598e-06, "relative_l2": 0.002042778718414212, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.04681599885225296, "p95_ms": 0.05173279885202647, "p99_ms": 0.06290272191166876, "mean_ms": 0.047846826712290444, "n": 300}, {"repeat": 1, "p50_ms": 0.046911999583244324, "p95_ms": 0.05206720139831305, "p99_ms": 0.06276512101292608, "mean_ms": 0.047813546744485694, "n": 300}, {"repeat": 2, "p50_ms": 0.04694399982690811, "p95_ms": 0.05050080083310604, "p99_ms": 0.056147518754005386, "mean_ms": 0.047565973388652004, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_a7ce4999e4f2ba20_sdpa_cudnn.json", "configuration": "480x832_tp1_sp4", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_retry_480x832_tp1_sp4_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_cross_480x832_tp1_sp4_v1/captures/qkv_rank0_a7ce4999e4f2ba20.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 9, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "native_dense", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [512, 10, 128], "q_len": 4680, "k_len": 512, "reference_error": {"max_abs": 0.0, "rmse": 0.0, "relative_l2": 0.0, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 7.629575161298766e-05, "rmse": 4.99106817785475e-06, "relative_l2": 0.0018847012407146944, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.10700800269842148, "p95_ms": 0.11374080181121826, "p99_ms": 0.11869792178273195, "mean_ms": 0.10988170688350996, "n": 300}, {"repeat": 1, "p50_ms": 0.10719999670982361, "p95_ms": 0.11335679888725281, "p99_ms": 0.12240640066564082, "mean_ms": 0.10886144027113914, "n": 300}, {"repeat": 2, "p50_ms": 0.10710400342941284, "p95_ms": 0.11328479833900928, "p99_ms": 0.12362400248646735, "mean_ms": 0.10868704003592332, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_a7ce4999e4f2ba20_native_dense.json", "configuration": "480x832_tp1_sp4", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_retry_480x832_tp1_sp4_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_cross_480x832_tp1_sp4_v1/captures/qkv_rank0_c6a5fcb1e5f12d0d.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 9, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "vllm_fa2", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [29640, 10, 128], "q_len": 4680, "k_len": 18720, "reference_error": {"max_abs": 0.03125, "rmse": 0.0005655613731987039, "relative_l2": 0.0010091701042483979, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.008430754864764634, "rmse": 0.0008224519263401907, "relative_l2": 0.0019414505560311024, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 1.2668160200119019, "p95_ms": 1.2744111955165862, "p99_ms": 1.2768659710884094, "mean_ms": 1.2676653850078583, "n": 300}, {"repeat": 1, "p50_ms": 1.2666879892349243, "p95_ms": 1.2703039646148682, "p99_ms": 1.2717446529865264, "mean_ms": 1.2679266174634298, "n": 300}, {"repeat": 2, "p50_ms": 1.2847040295600891, "p95_ms": 1.3277119934558868, "p99_ms": 1.3300530850887298, "mean_ms": 1.2927267158031464, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_c6a5fcb1e5f12d0d_vllm_fa2.json", "configuration": "480x832_tp1_sp4", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_retry_480x832_tp1_sp4_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_cross_480x832_tp1_sp4_v1/captures/qkv_rank0_c6a5fcb1e5f12d0d.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 9, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "vllm_fa3", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [29640, 10, 128], "q_len": 4680, "k_len": 18720, "reference_error": {"max_abs": 0.0, "rmse": 0.0, "relative_l2": 0.0, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.008430754864764634, "rmse": 0.0008316797479339394, "relative_l2": 0.001963233421132848, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.6816799938678741, "p95_ms": 0.6849279999732971, "p99_ms": 0.685344300866127, "mean_ms": 0.6659024020036062, "n": 300}, {"repeat": 1, "p50_ms": 0.68476802110672, "p95_ms": 0.69171362221241, "p99_ms": 0.6935404634475708, "mean_ms": 0.6802428793907166, "n": 300}, {"repeat": 2, "p50_ms": 0.6857440173625946, "p95_ms": 0.6922895818948746, "p99_ms": 0.6931532901525498, "mean_ms": 0.6801240543524424, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_c6a5fcb1e5f12d0d_vllm_fa3.json", "configuration": "480x832_tp1_sp4", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_retry_480x832_tp1_sp4_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_cross_480x832_tp1_sp4_v1/captures/qkv_rank0_c6a5fcb1e5f12d0d.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 9, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_math", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [29640, 10, 128], "q_len": 4680, "k_len": 18720, "reference_error": {"max_abs": 0.03125, "rmse": 0.0009919727936492876, "relative_l2": 0.001770045365575048, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.007760694409859159, "rmse": 0.0007096702226792513, "relative_l2": 0.0016752221063550086, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 18.77552032470703, "p95_ms": 18.808274936676025, "p99_ms": 18.875533828735353, "mean_ms": 18.757995777130127, "n": 300}, {"repeat": 1, "p50_ms": 18.792415618896484, "p95_ms": 18.823266220092773, "p99_ms": 18.89453456878662, "mean_ms": 18.767097365061442, "n": 300}, {"repeat": 2, "p50_ms": 18.797216415405273, "p95_ms": 18.85063591003418, "p99_ms": 18.89410936355591, "mean_ms": 18.780764706929524, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_c6a5fcb1e5f12d0d_sdpa_math.json", "configuration": "480x832_tp1_sp4", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_retry_480x832_tp1_sp4_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_cross_480x832_tp1_sp4_v1/captures/qkv_rank0_c6a5fcb1e5f12d0d.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 9, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_flash", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [29640, 10, 128], "q_len": 4680, "k_len": 18720, "reference_error": {"max_abs": 0.03125, "rmse": 0.0005659922767473519, "relative_l2": 0.001009938995123407, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.008430754864764634, "rmse": 0.0008226065287387949, "relative_l2": 0.0019418155049151903, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 1.3731679916381836, "p95_ms": 1.3862096548080445, "p99_ms": 1.3872694528102876, "mean_ms": 1.3763477341334025, "n": 300}, {"repeat": 1, "p50_ms": 1.3734880089759827, "p95_ms": 1.3784976482391358, "p99_ms": 1.3798732447624207, "mean_ms": 1.3744200547536214, "n": 300}, {"repeat": 2, "p50_ms": 1.373695969581604, "p95_ms": 1.378947240114212, "p99_ms": 1.3806114828586578, "mean_ms": 1.3746113042036692, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_c6a5fcb1e5f12d0d_sdpa_flash.json", "configuration": "480x832_tp1_sp4", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_retry_480x832_tp1_sp4_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_cross_480x832_tp1_sp4_v1/captures/qkv_rank0_c6a5fcb1e5f12d0d.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 9, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_efficient", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [29640, 10, 128], "q_len": 4680, "k_len": 18720, "reference_error": {"max_abs": 0.03125, "rmse": 0.0007752434725243829, "relative_l2": 0.0013833203133383903, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.00928316173610133, "rmse": 0.000829930267228196, "relative_l2": 0.001959103659647523, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 2.74126398563385, "p95_ms": 2.763881516456604, "p99_ms": 2.7704109263420107, "mean_ms": 2.7446281655629474, "n": 300}, {"repeat": 1, "p50_ms": 2.7480798959732056, "p95_ms": 2.7768640995025633, "p99_ms": 2.7883296632766723, "mean_ms": 2.752202562491099, "n": 300}, {"repeat": 2, "p50_ms": 2.7467360496520996, "p95_ms": 2.768291103839874, "p99_ms": 2.7757570815086363, "mean_ms": 2.7501537005106607, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_c6a5fcb1e5f12d0d_sdpa_efficient.json", "configuration": "480x832_tp1_sp4", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_retry_480x832_tp1_sp4_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_cross_480x832_tp1_sp4_v1/captures/qkv_rank0_c6a5fcb1e5f12d0d.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 9, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_cudnn", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [29640, 10, 128], "q_len": 4680, "k_len": 18720, "reference_error": {"max_abs": 0.03125, "rmse": 0.0007742347555068339, "relative_l2": 0.0013815203901010595, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.00928316173610133, "rmse": 0.000830095690624396, "relative_l2": 0.0019594941521909135, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.8007200062274933, "p95_ms": 0.8199040293693542, "p99_ms": 0.8207763189077377, "mean_ms": 0.7953172274430593, "n": 300}, {"repeat": 1, "p50_ms": 0.8055519759654999, "p95_ms": 0.8301151812076568, "p99_ms": 0.8346732801198959, "mean_ms": 0.8082679470380147, "n": 300}, {"repeat": 2, "p50_ms": 0.8064799904823303, "p95_ms": 0.8283887833356858, "p99_ms": 0.8301443082094192, "mean_ms": 0.8077065598964691, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_c6a5fcb1e5f12d0d_sdpa_cudnn.json", "configuration": "480x832_tp1_sp4", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_retry_480x832_tp1_sp4_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_cross_480x832_tp1_sp4_v1/captures/qkv_rank0_c6a5fcb1e5f12d0d.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 9, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "native_dense", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [29640, 10, 128], "q_len": 4680, "k_len": 18720, "reference_error": {"max_abs": 0.03125, "rmse": 0.0005012577952112607, "relative_l2": 0.0008944288019311806, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.008430754864764634, "rmse": 0.0008339159662758888, "relative_l2": 0.0019685121580466317, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.7048319876194, "p95_ms": 0.7561280131340027, "p99_ms": 0.7565449541807174, "mean_ms": 0.6824106661478678, "n": 300}, {"repeat": 1, "p50_ms": 0.6845919787883759, "p95_ms": 0.6954720258712769, "p99_ms": 0.6975679993629456, "mean_ms": 0.6779844270149866, "n": 300}, {"repeat": 2, "p50_ms": 0.6720159947872162, "p95_ms": 0.6973776131868362, "p99_ms": 0.6989133042097092, "mean_ms": 0.671793385942777, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_c6a5fcb1e5f12d0d_native_dense.json", "configuration": "480x832_tp1_sp4", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_retry_480x832_tp1_sp4_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_cross_480x832_tp1_sp4_v1/captures/qkv_rank0_cf2e5e4d322f9347.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 27, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "vllm_fa2", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [29640, 10, 128], "q_len": 4680, "k_len": 28080, "reference_error": {"max_abs": 0.03125, "rmse": 0.0005552815252655393, "relative_l2": 0.000991735346034173, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.008820491288137777, "rmse": 0.0008120587250742453, "relative_l2": 0.0019695529708380166, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 1.8922079801559448, "p95_ms": 1.9075312554836272, "p99_ms": 1.91401664853096, "mean_ms": 1.8950632540384929, "n": 300}, {"repeat": 1, "p50_ms": 1.9048640131950378, "p95_ms": 2.0000399470329286, "p99_ms": 2.03046585559845, "mean_ms": 1.923711786667506, "n": 300}, {"repeat": 2, "p50_ms": 1.8989759683609009, "p95_ms": 1.913107168674469, "p99_ms": 1.9165750527381897, "mean_ms": 1.8998897051811219, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_cf2e5e4d322f9347_vllm_fa2.json", "configuration": "480x832_tp1_sp4", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_retry_480x832_tp1_sp4_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_cross_480x832_tp1_sp4_v1/captures/qkv_rank0_cf2e5e4d322f9347.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 27, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "vllm_fa3", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [29640, 10, 128], "q_len": 4680, "k_len": 28080, "reference_error": {"max_abs": 0.0, "rmse": 0.0, "relative_l2": 0.0, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.008820491288137777, "rmse": 0.0008167053351597663, "relative_l2": 0.0019808227773380664, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 1.023743987083435, "p95_ms": 1.0507231771945953, "p99_ms": 1.0532717275619508, "mean_ms": 1.013439466158549, "n": 300}, {"repeat": 1, "p50_ms": 1.028480052947998, "p95_ms": 1.0478144466876984, "p99_ms": 1.0517120361328125, "mean_ms": 1.0307827178637188, "n": 300}, {"repeat": 2, "p50_ms": 1.0174559950828552, "p95_ms": 1.1343407928943634, "p99_ms": 1.1377923119068145, "mean_ms": 1.0302844790617625, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_cf2e5e4d322f9347_vllm_fa3.json", "configuration": "480x832_tp1_sp4", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_retry_480x832_tp1_sp4_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_cross_480x832_tp1_sp4_v1/captures/qkv_rank0_cf2e5e4d322f9347.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 27, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_math", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [29640, 10, 128], "q_len": 4680, "k_len": 28080, "reference_error": {"max_abs": 0.03125, "rmse": 0.0009780852179610475, "relative_l2": 0.0017468646766550496, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.007724527303786921, "rmse": 0.0006858574812207812, "relative_l2": 0.0016634666902770664, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 28.06777572631836, "p95_ms": 28.27443504333496, "p99_ms": 28.324563388824462, "mean_ms": 28.09189599990845, "n": 300}, {"repeat": 1, "p50_ms": 28.070015907287598, "p95_ms": 28.307666110992432, "p99_ms": 28.399074020385743, "mean_ms": 28.113922780354816, "n": 300}, {"repeat": 2, "p50_ms": 28.074912071228027, "p95_ms": 28.344671726226807, "p99_ms": 28.49186201095581, "mean_ms": 28.137030010223388, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_cf2e5e4d322f9347_sdpa_math.json", "configuration": "480x832_tp1_sp4", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_retry_480x832_tp1_sp4_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_cross_480x832_tp1_sp4_v1/captures/qkv_rank0_cf2e5e4d322f9347.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 27, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_flash", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [29640, 10, 128], "q_len": 4680, "k_len": 28080, "reference_error": {"max_abs": 0.03125, "rmse": 0.0005560456884091784, "relative_l2": 0.000993100144906819, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.008820491288137777, "rmse": 0.0008120515445332757, "relative_l2": 0.00196953555528005, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 2.03438401222229, "p95_ms": 2.0703696489334105, "p99_ms": 2.072230622768402, "mean_ms": 2.042091405391693, "n": 300}, {"repeat": 1, "p50_ms": 2.065743923187256, "p95_ms": 2.183743953704834, "p99_ms": 2.184705023765564, "mean_ms": 2.083918396631877, "n": 300}, {"repeat": 2, "p50_ms": 2.0392318964004517, "p95_ms": 2.079878294467926, "p99_ms": 2.0820159912109375, "mean_ms": 2.050890029271444, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_cf2e5e4d322f9347_sdpa_flash.json", "configuration": "480x832_tp1_sp4", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_retry_480x832_tp1_sp4_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_cross_480x832_tp1_sp4_v1/captures/qkv_rank0_cf2e5e4d322f9347.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 27, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_efficient", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [29640, 10, 128], "q_len": 4680, "k_len": 28080, "reference_error": {"max_abs": 0.03125, "rmse": 0.0007705672939716462, "relative_l2": 0.0013762367144560445, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.008820491288137777, "rmse": 0.000811379428513884, "relative_l2": 0.001967905416889973, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 4.123999834060669, "p95_ms": 4.154438233375549, "p99_ms": 4.162926611900329, "mean_ms": 4.123142927487692, "n": 300}, {"repeat": 1, "p50_ms": 4.123471975326538, "p95_ms": 4.15263352394104, "p99_ms": 4.158242721557617, "mean_ms": 4.122332366307576, "n": 300}, {"repeat": 2, "p50_ms": 4.125919818878174, "p95_ms": 4.158255958557129, "p99_ms": 4.163428173065186, "mean_ms": 4.124047247568766, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_cf2e5e4d322f9347_sdpa_efficient.json", "configuration": "480x832_tp1_sp4", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_retry_480x832_tp1_sp4_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_cross_480x832_tp1_sp4_v1/captures/qkv_rank0_cf2e5e4d322f9347.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 27, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_cudnn", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [29640, 10, 128], "q_len": 4680, "k_len": 28080, "reference_error": {"max_abs": 0.03125, "rmse": 0.0007701305226022741, "relative_l2": 0.0013754566387909396, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.008820491288137777, "rmse": 0.0008098227142425212, "relative_l2": 0.001964129789434424, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 1.2088960409164429, "p95_ms": 1.219391942024231, "p99_ms": 1.2197769939899445, "mean_ms": 1.2022045902411143, "n": 300}, {"repeat": 1, "p50_ms": 1.2010400295257568, "p95_ms": 1.211079967021942, "p99_ms": 1.2206508243083953, "mean_ms": 1.2012622940540314, "n": 300}, {"repeat": 2, "p50_ms": 1.242576003074646, "p95_ms": 1.3585920333862305, "p99_ms": 1.372515503168106, "mean_ms": 1.2559867743651072, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_cf2e5e4d322f9347_sdpa_cudnn.json", "configuration": "480x832_tp1_sp4", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_retry_480x832_tp1_sp4_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_cross_480x832_tp1_sp4_v1/captures/qkv_rank0_cf2e5e4d322f9347.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 27, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "native_dense", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [29640, 10, 128], "q_len": 4680, "k_len": 28080, "reference_error": {"max_abs": 0.03125, "rmse": 0.0005156585946133466, "relative_l2": 0.0009209686105076327, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.007724527303786921, "rmse": 0.0008217814009871526, "relative_l2": 0.0019931341782525556, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 1.009872019290924, "p95_ms": 1.024713581800461, "p99_ms": 1.027073233127594, "mean_ms": 0.9953939201434453, "n": 300}, {"repeat": 1, "p50_ms": 1.0024960041046143, "p95_ms": 1.0100239634513855, "p99_ms": 1.013697235584259, "mean_ms": 0.9954527986049652, "n": 300}, {"repeat": 2, "p50_ms": 1.0616320371627808, "p95_ms": 1.150307160615921, "p99_ms": 1.1523183608055114, "mean_ms": 1.0715886920690536, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_cf2e5e4d322f9347_native_dense.json", "configuration": "480x832_tp1_sp4", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_retry_480x832_tp1_sp4_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_cross_480x832_tp1_sp4_v1/captures/qkv_rank0_df3a5d6865cf4937.pt", "context": {"phase": "commit", "step": null, "start_frame": 9, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "vllm_fa2", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [29640, 10, 128], "q_len": 4680, "k_len": 18720, "reference_error": {"max_abs": 0.0625, "rmse": 0.0004193072287393182, "relative_l2": 0.0005270094373548114, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.012678517544874701, "rmse": 0.0009683098051771197, "relative_l2": 0.0016927232205301553, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 1.2683199644088745, "p95_ms": 1.2740896344184875, "p99_ms": 1.2760995066165923, "mean_ms": 1.2692010641098022, "n": 300}, {"repeat": 1, "p50_ms": 1.2588000297546387, "p95_ms": 1.2782272160053254, "p99_ms": 1.2833648037910461, "mean_ms": 1.2636581325531007, "n": 300}, {"repeat": 2, "p50_ms": 1.287663996219635, "p95_ms": 1.3268207669258119, "p99_ms": 1.3278723573684692, "mean_ms": 1.2984683748086294, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_df3a5d6865cf4937_vllm_fa2.json", "configuration": "480x832_tp1_sp4", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_retry_480x832_tp1_sp4_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_cross_480x832_tp1_sp4_v1/captures/qkv_rank0_df3a5d6865cf4937.pt", "context": {"phase": "commit", "step": null, "start_frame": 9, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "vllm_fa3", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [29640, 10, 128], "q_len": 4680, "k_len": 18720, "reference_error": {"max_abs": 0.0, "rmse": 0.0, "relative_l2": 0.0, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.012678517544874701, "rmse": 0.0009692782411752042, "relative_l2": 0.001694416164351224, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.6799359917640686, "p95_ms": 0.685451203584671, "p99_ms": 0.6869478476047516, "mean_ms": 0.670423465569814, "n": 300}, {"repeat": 1, "p50_ms": 0.680512011051178, "p95_ms": 0.6884816199541092, "p99_ms": 0.6894732993841172, "mean_ms": 0.6762714648246765, "n": 300}, {"repeat": 2, "p50_ms": 0.6781439781188965, "p95_ms": 0.6828208267688751, "p99_ms": 0.6850610911846161, "mean_ms": 0.6754716779788336, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_df3a5d6865cf4937_vllm_fa3.json", "configuration": "480x832_tp1_sp4", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_retry_480x832_tp1_sp4_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_cross_480x832_tp1_sp4_v1/captures/qkv_rank0_df3a5d6865cf4937.pt", "context": {"phase": "commit", "step": null, "start_frame": 9, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_math", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [29640, 10, 128], "q_len": 4680, "k_len": 18720, "reference_error": {"max_abs": 0.0625, "rmse": 0.0007898468096038272, "relative_l2": 0.000992724890475454, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.012678517544874701, "rmse": 0.00094767651387136, "relative_l2": 0.001656653719712869, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 18.79646396636963, "p95_ms": 18.811085796356203, "p99_ms": 18.966321086883546, "mean_ms": 18.779477475484214, "n": 300}, {"repeat": 1, "p50_ms": 18.7969913482666, "p95_ms": 18.846524715423584, "p99_ms": 19.037684783935546, "mean_ms": 18.783243548075358, "n": 300}, {"repeat": 2, "p50_ms": 18.798815727233887, "p95_ms": 18.883044910430907, "p99_ms": 18.967518997192382, "mean_ms": 18.789022509257, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_df3a5d6865cf4937_sdpa_math.json", "configuration": "480x832_tp1_sp4", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_retry_480x832_tp1_sp4_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_cross_480x832_tp1_sp4_v1/captures/qkv_rank0_df3a5d6865cf4937.pt", "context": {"phase": "commit", "step": null, "start_frame": 9, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_flash", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [29640, 10, 128], "q_len": 4680, "k_len": 18720, "reference_error": {"max_abs": 0.0625, "rmse": 0.0004200155646248864, "relative_l2": 0.0005278997146286704, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.012678517544874701, "rmse": 0.0009683237781535962, "relative_l2": 0.0016927476469912064, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 1.3721439838409424, "p95_ms": 1.3870239555835724, "p99_ms": 1.3896662938594817, "mean_ms": 1.3728587742646534, "n": 300}, {"repeat": 1, "p50_ms": 1.3618559837341309, "p95_ms": 1.3640976428985596, "p99_ms": 1.3651203727722168, "mean_ms": 1.362102507352829, "n": 300}, {"repeat": 2, "p50_ms": 1.3924959897994995, "p95_ms": 1.4051248133182526, "p99_ms": 1.4067200422286987, "mean_ms": 1.3891843235492707, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_df3a5d6865cf4937_sdpa_flash.json", "configuration": "480x832_tp1_sp4", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_retry_480x832_tp1_sp4_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_cross_480x832_tp1_sp4_v1/captures/qkv_rank0_df3a5d6865cf4937.pt", "context": {"phase": "commit", "step": null, "start_frame": 9, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_efficient", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [29640, 10, 128], "q_len": 4680, "k_len": 18720, "reference_error": {"max_abs": 0.0625, "rmse": 0.0006686263996730479, "relative_l2": 0.0008403681084910054, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.012678517544874701, "rmse": 0.0009547840566502768, "relative_l2": 0.0016690785682876319, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 2.720960021018982, "p95_ms": 2.765859270095825, "p99_ms": 2.792233211994171, "mean_ms": 2.728207573890686, "n": 300}, {"repeat": 1, "p50_ms": 2.7179999351501465, "p95_ms": 2.7441184401512144, "p99_ms": 2.7539251971244814, "mean_ms": 2.723162991205851, "n": 300}, {"repeat": 2, "p50_ms": 2.718239903450012, "p95_ms": 2.745859241485596, "p99_ms": 2.751873342990875, "mean_ms": 2.7238567543029784, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_df3a5d6865cf4937_sdpa_efficient.json", "configuration": "480x832_tp1_sp4", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_retry_480x832_tp1_sp4_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_cross_480x832_tp1_sp4_v1/captures/qkv_rank0_df3a5d6865cf4937.pt", "context": {"phase": "commit", "step": null, "start_frame": 9, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_cudnn", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [29640, 10, 128], "q_len": 4680, "k_len": 18720, "reference_error": {"max_abs": 0.0625, "rmse": 0.000667980162633798, "relative_l2": 0.0008395558806181951, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.012678517544874701, "rmse": 0.0009547032318614938, "relative_l2": 0.0016689372767340032, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.8193280100822449, "p95_ms": 0.8298271715641021, "p99_ms": 0.8344646537303925, "mean_ms": 0.8025899736086527, "n": 300}, {"repeat": 1, "p50_ms": 0.8128800094127655, "p95_ms": 0.8307552129030228, "p99_ms": 0.8335641884803772, "mean_ms": 0.8170810679594676, "n": 300}, {"repeat": 2, "p50_ms": 0.805791974067688, "p95_ms": 0.8206448197364807, "p99_ms": 0.8273599743843079, "mean_ms": 0.8055554125706355, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_df3a5d6865cf4937_sdpa_cudnn.json", "configuration": "480x832_tp1_sp4", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_retry_480x832_tp1_sp4_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_cross_480x832_tp1_sp4_v1/captures/qkv_rank0_df3a5d6865cf4937.pt", "context": {"phase": "commit", "step": null, "start_frame": 9, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "native_dense", "status": "measured", "q_shape": [4680, 10, 128], "k_shape": [29640, 10, 128], "q_len": 4680, "k_len": 18720, "reference_error": {"max_abs": 0.03125, "rmse": 0.00037121631091112655, "relative_l2": 0.0004665660063586139, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.012678517544874701, "rmse": 0.0009680282200371425, "relative_l2": 0.0016922309754837381, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.6773599982261658, "p95_ms": 0.6888335853815079, "p99_ms": 0.6895046669244767, "mean_ms": 0.6672954668601354, "n": 300}, {"repeat": 1, "p50_ms": 0.6708640158176422, "p95_ms": 0.6793056160211564, "p99_ms": 0.6820525026321411, "mean_ms": 0.6684082126617432, "n": 300}, {"repeat": 2, "p50_ms": 0.6708160042762756, "p95_ms": 0.6793008148670197, "p99_ms": 0.6814108681678772, "mean_ms": 0.666720319588979, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_df3a5d6865cf4937_native_dense.json", "configuration": "480x832_tp1_sp4", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_retry_480x832_tp1_sp4_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_retry_704x1280_tp1_sp4_v1/captures/qkv_rank0_1ef9f10991fa086e.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 27, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "vllm_fa2", "status": "measured", "q_shape": [10560, 10, 128], "k_shape": [512, 10, 128], "q_len": 10560, "k_len": 512, "reference_error": {"max_abs": 0.0001220703125, "rmse": 1.2464031024512139e-06, "relative_l2": 0.0007104281737377895, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 7.818887076814496e-05, "rmse": 5.353783328355767e-06, "relative_l2": 0.0018571216816885075, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.09694400057196617, "p95_ms": 0.09759999811649323, "p99_ms": 0.0979859210550785, "mean_ms": 0.09712117334206899, "n": 300}, {"repeat": 1, "p50_ms": 0.0968639999628067, "p95_ms": 0.09753599762916565, "p99_ms": 0.09769791886210441, "mean_ms": 0.09725503996014595, "n": 300}, {"repeat": 2, "p50_ms": 0.09694400057196617, "p95_ms": 0.09756799787282944, "p99_ms": 0.09782431982457637, "mean_ms": 0.09727370674411456, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_1ef9f10991fa086e_vllm_fa2.json", "configuration": "704x1280_tp1_sp4", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_retry_704x1280_tp1_sp4_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_retry_704x1280_tp1_sp4_v1/captures/qkv_rank0_1ef9f10991fa086e.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 27, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "vllm_fa3", "status": "measured", "q_shape": [10560, 10, 128], "k_shape": [512, 10, 128], "q_len": 10560, "k_len": 512, "reference_error": {"max_abs": 0.0001220703125, "rmse": 4.7179196798998337e-07, "relative_l2": 0.00026891324768376913, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 7.818887076814496e-05, "rmse": 5.395104933635903e-06, "relative_l2": 0.0018714553303218902, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.08048000186681747, "p95_ms": 0.08153600245714188, "p99_ms": 0.08211231939494609, "mean_ms": 0.08070752009749413, "n": 300}, {"repeat": 1, "p50_ms": 0.08048000186681747, "p95_ms": 0.08156800270080566, "p99_ms": 0.08240000158548355, "mean_ms": 0.08092586648960909, "n": 300}, {"repeat": 2, "p50_ms": 0.08038400113582611, "p95_ms": 0.08144000172615051, "p99_ms": 0.08195327818393708, "mean_ms": 0.08081055983901024, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_1ef9f10991fa086e_vllm_fa3.json", "configuration": "704x1280_tp1_sp4", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_retry_704x1280_tp1_sp4_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_retry_704x1280_tp1_sp4_v1/captures/qkv_rank0_1ef9f10991fa086e.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 27, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_math", "status": "measured", "q_shape": [10560, 10, 128], "k_shape": [512, 10, 128], "q_len": 10560, "k_len": 512, "reference_error": {"max_abs": 0.000244140625, "rmse": 3.3214523568283587e-06, "relative_l2": 0.001893170297296342, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 7.818887076814496e-05, "rmse": 4.7402353330505205e-06, "relative_l2": 0.0016442940017181583, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 1.2656960487365723, "p95_ms": 1.2696351826190948, "p99_ms": 1.2724262356758118, "mean_ms": 1.2659027210871379, "n": 300}, {"repeat": 1, "p50_ms": 1.2662720084190369, "p95_ms": 1.270883184671402, "p99_ms": 1.2730563497543335, "mean_ms": 1.2670677344004313, "n": 300}, {"repeat": 2, "p50_ms": 1.2662559747695923, "p95_ms": 1.2704976081848145, "p99_ms": 1.2722946846485137, "mean_ms": 1.2669009057680767, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_1ef9f10991fa086e_sdpa_math.json", "configuration": "704x1280_tp1_sp4", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_retry_704x1280_tp1_sp4_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_retry_704x1280_tp1_sp4_v1/captures/qkv_rank0_1ef9f10991fa086e.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 27, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_flash", "status": "measured", "q_shape": [10560, 10, 128], "k_shape": [512, 10, 128], "q_len": 10560, "k_len": 512, "reference_error": {"max_abs": 0.0001220703125, "rmse": 1.246703601658886e-06, "relative_l2": 0.000710599453079839, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 7.818887076814496e-05, "rmse": 5.353783328355767e-06, "relative_l2": 0.0018571216816885075, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.1027199998497963, "p95_ms": 0.10361599922180176, "p99_ms": 0.10422495640814304, "mean_ms": 0.10286719982822735, "n": 300}, {"repeat": 1, "p50_ms": 0.10268799960613251, "p95_ms": 0.10368159972131252, "p99_ms": 0.10393856167793274, "mean_ms": 0.10301290648678939, "n": 300}, {"repeat": 2, "p50_ms": 0.10257599875330925, "p95_ms": 0.10336159728467464, "p99_ms": 0.10390592142939567, "mean_ms": 0.10286197312176228, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_1ef9f10991fa086e_sdpa_flash.json", "configuration": "704x1280_tp1_sp4", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_retry_704x1280_tp1_sp4_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_retry_704x1280_tp1_sp4_v1/captures/qkv_rank0_1ef9f10991fa086e.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 27, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_efficient", "status": "measured", "q_shape": [10560, 10, 128], "k_shape": [512, 10, 128], "q_len": 10560, "k_len": 512, "reference_error": {"max_abs": 0.0001220703125, "rmse": 2.184166658424875e-06, "relative_l2": 0.0012449371533432055, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 7.818887076814496e-05, "rmse": 5.891822370152756e-06, "relative_l2": 0.002043756797238275, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.17903999984264374, "p95_ms": 0.17983999848365784, "p99_ms": 0.18029056191444398, "mean_ms": 0.17917471980055175, "n": 300}, {"repeat": 1, "p50_ms": 0.1791680008172989, "p95_ms": 0.17990559935569764, "p99_ms": 0.18060800433158875, "mean_ms": 0.17947861313819885, "n": 300}, {"repeat": 2, "p50_ms": 0.17923200130462646, "p95_ms": 0.17984159886837006, "p99_ms": 0.18029056191444398, "mean_ms": 0.17948874627550443, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_1ef9f10991fa086e_sdpa_efficient.json", "configuration": "704x1280_tp1_sp4", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_retry_704x1280_tp1_sp4_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_retry_704x1280_tp1_sp4_v1/captures/qkv_rank0_1ef9f10991fa086e.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 27, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_cudnn", "status": "measured", "q_shape": [10560, 10, 128], "k_shape": [512, 10, 128], "q_len": 10560, "k_len": 512, "reference_error": {"max_abs": 0.0001220703125, "rmse": 2.1830788117468677e-06, "relative_l2": 0.0012443170995843189, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 7.818887076814496e-05, "rmse": 5.891822370152756e-06, "relative_l2": 0.002043756797238275, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.05939200147986412, "p95_ms": 0.06016159988939762, "p99_ms": 0.060579519309103484, "mean_ms": 0.059704960212111476, "n": 300}, {"repeat": 1, "p50_ms": 0.05939200147986412, "p95_ms": 0.06013279967010021, "p99_ms": 0.060864641480147835, "mean_ms": 0.05999605340262254, "n": 300}, {"repeat": 2, "p50_ms": 0.05929600074887276, "p95_ms": 0.06015999987721443, "p99_ms": 0.060496958345174746, "mean_ms": 0.060423893531163536, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_1ef9f10991fa086e_sdpa_cudnn.json", "configuration": "704x1280_tp1_sp4", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_retry_704x1280_tp1_sp4_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_retry_704x1280_tp1_sp4_v1/captures/qkv_rank0_1ef9f10991fa086e.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 27, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "native_dense", "status": "measured", "q_shape": [10560, 10, 128], "k_shape": [512, 10, 128], "q_len": 10560, "k_len": 512, "reference_error": {"max_abs": 0.0, "rmse": 0.0, "relative_l2": 0.0, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 7.818887076814496e-05, "rmse": 5.410283113664583e-06, "relative_l2": 0.001876720341154611, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.1242079995572567, "p95_ms": 0.13889919966459274, "p99_ms": 0.2272976019978523, "mean_ms": 0.12820618701477846, "n": 300}, {"repeat": 1, "p50_ms": 0.12377600371837616, "p95_ms": 0.13005280643701553, "p99_ms": 0.15650495946407317, "mean_ms": 0.125393493399024, "n": 300}, {"repeat": 2, "p50_ms": 0.12364800274372101, "p95_ms": 0.13002879917621613, "p99_ms": 0.16797920495271682, "mean_ms": 0.12919530627628167, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_1ef9f10991fa086e_native_dense.json", "configuration": "704x1280_tp1_sp4", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_retry_704x1280_tp1_sp4_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_retry_704x1280_tp1_sp4_v1/captures/qkv_rank0_6be6af8ca00bf828.pt", "context": {"phase": "commit", "step": null, "start_frame": 0, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "vllm_fa2", "status": "measured", "q_shape": [10560, 10, 128], "k_shape": [512, 10, 128], "q_len": 10560, "k_len": 512, "reference_error": {"max_abs": 0.0009765625, "rmse": 1.933911734347754e-06, "relative_l2": 0.00030891720964543515, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.00043489744124647345, "rmse": 1.7589970025427215e-05, "relative_l2": 0.001761251841652134, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.09588799998164177, "p95_ms": 0.09683199971914291, "p99_ms": 0.09712032191455364, "mean_ms": 0.09607637318472068, "n": 300}, {"repeat": 1, "p50_ms": 0.09590400010347366, "p95_ms": 0.09689600020647049, "p99_ms": 0.097344643548131, "mean_ms": 0.09635541332264741, "n": 300}, {"repeat": 2, "p50_ms": 0.09577599912881851, "p95_ms": 0.09673759900033474, "p99_ms": 0.09724896289408207, "mean_ms": 0.09616767970224221, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_6be6af8ca00bf828_vllm_fa2.json", "configuration": "704x1280_tp1_sp4", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_retry_704x1280_tp1_sp4_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_retry_704x1280_tp1_sp4_v1/captures/qkv_rank0_6be6af8ca00bf828.pt", "context": {"phase": "commit", "step": null, "start_frame": 0, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "vllm_fa3", "status": "measured", "q_shape": [10560, 10, 128], "k_shape": [512, 10, 128], "q_len": 10560, "k_len": 512, "reference_error": {"max_abs": 0.00048828125, "rmse": 7.084926050389087e-07, "relative_l2": 0.00011317246527637482, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.00043489744124647345, "rmse": 1.7598296726068673e-05, "relative_l2": 0.0017620855791069708, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.08057600259780884, "p95_ms": 0.08153760246932507, "p99_ms": 0.08269087634980678, "mean_ms": 0.08072266665597756, "n": 300}, {"repeat": 1, "p50_ms": 0.08070400357246399, "p95_ms": 0.0817295964807272, "p99_ms": 0.11122111693024628, "mean_ms": 0.082223786637187, "n": 300}, {"repeat": 2, "p50_ms": 0.08076799660921097, "p95_ms": 0.08160320296883583, "p99_ms": 0.08195615820586681, "mean_ms": 0.08117194664975007, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_6be6af8ca00bf828_vllm_fa3.json", "configuration": "704x1280_tp1_sp4", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_retry_704x1280_tp1_sp4_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_retry_704x1280_tp1_sp4_v1/captures/qkv_rank0_6be6af8ca00bf828.pt", "context": {"phase": "commit", "step": null, "start_frame": 0, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_math", "status": "measured", "q_shape": [10560, 10, 128], "k_shape": [512, 10, 128], "q_len": 10560, "k_len": 512, "reference_error": {"max_abs": 0.0009765625, "rmse": 4.742710175803078e-06, "relative_l2": 0.0007575861750279055, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.00043489744124647345, "rmse": 1.747372544361234e-05, "relative_l2": 0.0017496124822042455, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 1.265663981437683, "p95_ms": 1.2697664141654967, "p99_ms": 1.272647670507431, "mean_ms": 1.2662060793240866, "n": 300}, {"repeat": 1, "p50_ms": 1.2662080526351929, "p95_ms": 1.270603173971176, "p99_ms": 1.2731535935401916, "mean_ms": 1.2669532827536265, "n": 300}, {"repeat": 2, "p50_ms": 1.2654880285263062, "p95_ms": 1.2693519711494445, "p99_ms": 1.2704348754882813, "mean_ms": 1.266306453148524, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_6be6af8ca00bf828_sdpa_math.json", "configuration": "704x1280_tp1_sp4", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_retry_704x1280_tp1_sp4_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_retry_704x1280_tp1_sp4_v1/captures/qkv_rank0_6be6af8ca00bf828.pt", "context": {"phase": "commit", "step": null, "start_frame": 0, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_flash", "status": "measured", "q_shape": [10560, 10, 128], "k_shape": [512, 10, 128], "q_len": 10560, "k_len": 512, "reference_error": {"max_abs": 0.0009765625, "rmse": 1.9285640860462976e-06, "relative_l2": 0.00030806299248437724, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.00043489744124647345, "rmse": 1.7589970025427215e-05, "relative_l2": 0.001761251841652134, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.10275200009346008, "p95_ms": 0.10358399897813797, "p99_ms": 0.10447999835014343, "mean_ms": 0.10289738647639751, "n": 300}, {"repeat": 1, "p50_ms": 0.10252799838781357, "p95_ms": 0.10342399775981903, "p99_ms": 0.10425823666155339, "mean_ms": 0.102897706280152, "n": 300}, {"repeat": 2, "p50_ms": 0.10252799838781357, "p95_ms": 0.10342399775981903, "p99_ms": 0.10425791665911674, "mean_ms": 0.10283669297893842, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_6be6af8ca00bf828_sdpa_flash.json", "configuration": "704x1280_tp1_sp4", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_retry_704x1280_tp1_sp4_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_retry_704x1280_tp1_sp4_v1/captures/qkv_rank0_6be6af8ca00bf828.pt", "context": {"phase": "commit", "step": null, "start_frame": 0, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_efficient", "status": "measured", "q_shape": [10560, 10, 128], "k_shape": [512, 10, 128], "q_len": 10560, "k_len": 512, "reference_error": {"max_abs": 0.0009765625, "rmse": 3.3375096580802105e-06, "relative_l2": 0.0005331236955788764, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.00043489744124647345, "rmse": 1.7758429651179162e-05, "relative_l2": 0.0017781193988833715, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.1788800060749054, "p95_ms": 0.1795536033809185, "p99_ms": 0.1799071989953518, "mean_ms": 0.17900693361957867, "n": 300}, {"repeat": 1, "p50_ms": 0.17907199263572693, "p95_ms": 0.17977759838104249, "p99_ms": 0.18019295379519462, "mean_ms": 0.17932959988713265, "n": 300}, {"repeat": 2, "p50_ms": 0.17919999361038208, "p95_ms": 0.17983999848365784, "p99_ms": 0.18028928190469742, "mean_ms": 0.17945493305722873, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_6be6af8ca00bf828_sdpa_efficient.json", "configuration": "704x1280_tp1_sp4", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_retry_704x1280_tp1_sp4_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_retry_704x1280_tp1_sp4_v1/captures/qkv_rank0_6be6af8ca00bf828.pt", "context": {"phase": "commit", "step": null, "start_frame": 0, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_cudnn", "status": "measured", "q_shape": [10560, 10, 128], "k_shape": [512, 10, 128], "q_len": 10560, "k_len": 512, "reference_error": {"max_abs": 0.0009765625, "rmse": 3.3230769097053417e-06, "relative_l2": 0.0005308182520178843, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.00043489744124647345, "rmse": 1.7758429651179162e-05, "relative_l2": 0.0017781193988833715, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.05923200026154518, "p95_ms": 0.06006719917058945, "p99_ms": 0.060579839311540123, "mean_ms": 0.059508480007449786, "n": 300}, {"repeat": 1, "p50_ms": 0.05920000001788139, "p95_ms": 0.0598768012598157, "p99_ms": 0.06044959835708141, "mean_ms": 0.05968864005059004, "n": 300}, {"repeat": 2, "p50_ms": 0.05923200026154518, "p95_ms": 0.05987200140953064, "p99_ms": 0.060256320610642435, "mean_ms": 0.05964992011586825, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_6be6af8ca00bf828_sdpa_cudnn.json", "configuration": "704x1280_tp1_sp4", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_retry_704x1280_tp1_sp4_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_retry_704x1280_tp1_sp4_v1/captures/qkv_rank0_6be6af8ca00bf828.pt", "context": {"phase": "commit", "step": null, "start_frame": 0, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "native_dense", "status": "measured", "q_shape": [10560, 10, 128], "k_shape": [512, 10, 128], "q_len": 10560, "k_len": 512, "reference_error": {"max_abs": 0.0, "rmse": 0.0, "relative_l2": 0.0, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.00043489744124647345, "rmse": 1.7601063432965562e-05, "relative_l2": 0.001762362604457805, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.1244800016283989, "p95_ms": 0.13306239992380142, "p99_ms": 0.20070111304521554, "mean_ms": 0.1320198403050502, "n": 300}, {"repeat": 1, "p50_ms": 0.12425599992275238, "p95_ms": 0.13077919706702232, "p99_ms": 0.16046431988477705, "mean_ms": 0.12592618681490422, "n": 300}, {"repeat": 2, "p50_ms": 0.12387199699878693, "p95_ms": 0.1300271987915039, "p99_ms": 0.1577628743648529, "mean_ms": 0.12524767982463042, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_6be6af8ca00bf828_native_dense.json", "configuration": "704x1280_tp1_sp4", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_retry_704x1280_tp1_sp4_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_retry_704x1280_tp1_sp4_v1/captures/qkv_rank0_6e53ae344202af4b.pt", "context": {"phase": "commit", "step": null, "start_frame": 9, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "vllm_fa2", "status": "measured", "q_shape": [10560, 10, 128], "k_shape": [66880, 10, 128], "q_len": 10560, "k_len": 42240, "reference_error": {"max_abs": 0.0625, "rmse": 0.0003986688479897109, "relative_l2": 0.0004935892557198819, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.014540188720360447, "rmse": 0.0009034149203516021, "relative_l2": 0.001673807418418208, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 6.334144115447998, "p95_ms": 6.490131044387818, "p99_ms": 6.49504319190979, "mean_ms": 6.355364802678426, "n": 300}, {"repeat": 1, "p50_ms": 6.381792068481445, "p95_ms": 6.488352227210998, "p99_ms": 6.544999008178711, "mean_ms": 6.374140796661377, "n": 300}, {"repeat": 2, "p50_ms": 6.384704113006592, "p95_ms": 6.494795083999634, "p99_ms": 6.4988502025604244, "mean_ms": 6.384365542729696, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_6e53ae344202af4b_vllm_fa2.json", "configuration": "704x1280_tp1_sp4", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_retry_704x1280_tp1_sp4_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_retry_704x1280_tp1_sp4_v1/captures/qkv_rank0_6e53ae344202af4b.pt", "context": {"phase": "commit", "step": null, "start_frame": 9, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "vllm_fa3", "status": "measured", "q_shape": [10560, 10, 128], "k_shape": [66880, 10, 128], "q_len": 10560, "k_len": 42240, "reference_error": {"max_abs": 0.0, "rmse": 0.0, "relative_l2": 0.0, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.014540188720360447, "rmse": 0.0009039246564090026, "relative_l2": 0.0016747518349593698, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 3.478111982345581, "p95_ms": 3.7926767706871036, "p99_ms": 3.9742780923843384, "mean_ms": 3.5189878439903257, "n": 300}, {"repeat": 1, "p50_ms": 3.5086079835891724, "p95_ms": 3.616425573825836, "p99_ms": 3.652393321990967, "mean_ms": 3.516112316449483, "n": 300}, {"repeat": 2, "p50_ms": 3.5122560262680054, "p95_ms": 3.5798095941543577, "p99_ms": 3.6036061811447144, "mean_ms": 3.510686717828115, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_6e53ae344202af4b_vllm_fa3.json", "configuration": "704x1280_tp1_sp4", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_retry_704x1280_tp1_sp4_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_retry_704x1280_tp1_sp4_v1/captures/qkv_rank0_6e53ae344202af4b.pt", "context": {"phase": "commit", "step": null, "start_frame": 9, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_math", "status": "measured", "q_shape": [10560, 10, 128], "k_shape": [66880, 10, 128], "q_len": 10560, "k_len": 42240, "reference_error": {"max_abs": 0.0625, "rmse": 0.0007836689783857957, "relative_l2": 0.0009702553628724663, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.014540188720360447, "rmse": 0.0008877789123615756, "relative_l2": 0.0016448377107251286, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 97.23382568359375, "p95_ms": 97.71466445922852, "p99_ms": 97.8499568939209, "mean_ms": 97.21794741312662, "n": 300}, {"repeat": 1, "p50_ms": 97.45851135253906, "p95_ms": 98.0177448272705, "p99_ms": 98.25253700256347, "mean_ms": 97.45134671529134, "n": 300}, {"repeat": 2, "p50_ms": 97.55630493164062, "p95_ms": 97.96712875366211, "p99_ms": 98.23548248291016, "mean_ms": 97.55094441731771, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_6e53ae344202af4b_sdpa_math.json", "configuration": "704x1280_tp1_sp4", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_retry_704x1280_tp1_sp4_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_retry_704x1280_tp1_sp4_v1/captures/qkv_rank0_6e53ae344202af4b.pt", "context": {"phase": "commit", "step": null, "start_frame": 9, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_flash", "status": "measured", "q_shape": [10560, 10, 128], "k_shape": [66880, 10, 128], "q_len": 10560, "k_len": 42240, "reference_error": {"max_abs": 0.0625, "rmse": 0.0003990278719017121, "relative_l2": 0.0004940337608433795, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.014540188720360447, "rmse": 0.0009034055105428151, "relative_l2": 0.001673789984338475, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 6.8855040073394775, "p95_ms": 7.109081602096557, "p99_ms": 7.131869344711304, "mean_ms": 6.932968222300212, "n": 300}, {"repeat": 1, "p50_ms": 6.884976148605347, "p95_ms": 7.062619400024414, "p99_ms": 7.110849871635437, "mean_ms": 6.923896862665813, "n": 300}, {"repeat": 2, "p50_ms": 6.882143974304199, "p95_ms": 7.051775932312012, "p99_ms": 7.0544095659255985, "mean_ms": 6.921153917312622, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_6e53ae344202af4b_sdpa_flash.json", "configuration": "704x1280_tp1_sp4", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_retry_704x1280_tp1_sp4_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_retry_704x1280_tp1_sp4_v1/captures/qkv_rank0_6e53ae344202af4b.pt", "context": {"phase": "commit", "step": null, "start_frame": 9, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_efficient", "status": "measured", "q_shape": [10560, 10, 128], "k_shape": [66880, 10, 128], "q_len": 10560, "k_len": 42240, "reference_error": {"max_abs": 0.0625, "rmse": 0.0006728782206242722, "relative_l2": 0.0008330860607313483, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.014540188720360447, "rmse": 0.0008957329515698156, "relative_l2": 0.0016595746046298252, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 13.21940803527832, "p95_ms": 13.259039878845215, "p99_ms": 13.277187376022338, "mean_ms": 13.225409809748331, "n": 300}, {"repeat": 1, "p50_ms": 13.219215869903564, "p95_ms": 13.254913949966431, "p99_ms": 13.263502540588378, "mean_ms": 13.225385046005249, "n": 300}, {"repeat": 2, "p50_ms": 13.218688011169434, "p95_ms": 13.257700967788697, "p99_ms": 13.268348884582519, "mean_ms": 13.22531242052714, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_6e53ae344202af4b_sdpa_efficient.json", "configuration": "704x1280_tp1_sp4", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_retry_704x1280_tp1_sp4_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_retry_704x1280_tp1_sp4_v1/captures/qkv_rank0_6e53ae344202af4b.pt", "context": {"phase": "commit", "step": null, "start_frame": 9, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_cudnn", "status": "measured", "q_shape": [10560, 10, 128], "k_shape": [66880, 10, 128], "q_len": 10560, "k_len": 42240, "reference_error": {"max_abs": 0.0625, "rmse": 0.0006721463512854045, "relative_l2": 0.0008321799381584975, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.014540188720360447, "rmse": 0.0008957258433842463, "relative_l2": 0.0016595614348963292, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 3.8149759769439697, "p95_ms": 4.265809583663943, "p99_ms": 4.576552872657776, "mean_ms": 3.8833080530166626, "n": 300}, {"repeat": 1, "p50_ms": 3.82366406917572, "p95_ms": 4.036012578010559, "p99_ms": 4.060933041572571, "mean_ms": 3.866941661834717, "n": 300}, {"repeat": 2, "p50_ms": 3.8589439392089844, "p95_ms": 3.964644706249237, "p99_ms": 3.972452909946442, "mean_ms": 3.86821941057841, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_6e53ae344202af4b_sdpa_cudnn.json", "configuration": "704x1280_tp1_sp4", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_retry_704x1280_tp1_sp4_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_retry_704x1280_tp1_sp4_v1/captures/qkv_rank0_6e53ae344202af4b.pt", "context": {"phase": "commit", "step": null, "start_frame": 9, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "native_dense", "status": "measured", "q_shape": [10560, 10, 128], "k_shape": [66880, 10, 128], "q_len": 10560, "k_len": 42240, "reference_error": {"max_abs": 0.0625, "rmse": 0.00038285899479922116, "relative_l2": 0.00047401518137551307, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.014540188720360447, "rmse": 0.0009027966776015421, "relative_l2": 0.0016726619654506716, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 3.459328055381775, "p95_ms": 3.81016800403595, "p99_ms": 3.9507545804977418, "mean_ms": 3.4953364260991413, "n": 300}, {"repeat": 1, "p50_ms": 3.4442399740219116, "p95_ms": 3.9919344186782837, "p99_ms": 4.095877366065979, "mean_ms": 3.4790328566233315, "n": 300}, {"repeat": 2, "p50_ms": 3.453487992286682, "p95_ms": 3.7833073019981383, "p99_ms": 3.809434189796448, "mean_ms": 3.4943771688143412, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_6e53ae344202af4b_native_dense.json", "configuration": "704x1280_tp1_sp4", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_retry_704x1280_tp1_sp4_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_retry_704x1280_tp1_sp4_v1/captures/qkv_rank0_7a044869db5ec3c0.pt", "context": {"phase": "commit", "step": null, "start_frame": 27, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "vllm_fa2", "status": "measured", "q_shape": [10560, 10, 128], "k_shape": [66880, 10, 128], "q_len": 10560, "k_len": 63360, "reference_error": {"max_abs": 0.0625, "rmse": 0.00039043216308780375, "relative_l2": 0.0004884921109147573, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.00992615478884451, "rmse": 0.0009261728612270919, "relative_l2": 0.0017342987872258817, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 9.568495750427246, "p95_ms": 9.910834980010986, "p99_ms": 10.314844636917114, "mean_ms": 9.629564447402954, "n": 300}, {"repeat": 1, "p50_ms": 9.580895900726318, "p95_ms": 9.81520962715149, "p99_ms": 10.147988767623902, "mean_ms": 9.62740790049235, "n": 300}, {"repeat": 2, "p50_ms": 9.576704025268555, "p95_ms": 9.740544414520263, "p99_ms": 9.94279390335083, "mean_ms": 9.623962885538736, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_7a044869db5ec3c0_vllm_fa2.json", "configuration": "704x1280_tp1_sp4", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_retry_704x1280_tp1_sp4_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_retry_704x1280_tp1_sp4_v1/captures/qkv_rank0_7a044869db5ec3c0.pt", "context": {"phase": "commit", "step": null, "start_frame": 27, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "vllm_fa3", "status": "measured", "q_shape": [10560, 10, 128], "k_shape": [66880, 10, 128], "q_len": 10560, "k_len": 63360, "reference_error": {"max_abs": 0.0, "rmse": 0.0, "relative_l2": 0.0, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.010579403963054013, "rmse": 0.0009308798019245433, "relative_l2": 0.0017431127374989645, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 5.196191787719727, "p95_ms": 5.569089579582214, "p99_ms": 6.064563326835632, "mean_ms": 5.246133125623067, "n": 300}, {"repeat": 1, "p50_ms": 5.245375871658325, "p95_ms": 5.453324723243713, "p99_ms": 5.509185805320739, "mean_ms": 5.274914247194926, "n": 300}, {"repeat": 2, "p50_ms": 5.2450079917907715, "p95_ms": 5.351364707946777, "p99_ms": 5.403841662406921, "mean_ms": 5.237803192138672, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_7a044869db5ec3c0_vllm_fa3.json", "configuration": "704x1280_tp1_sp4", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_retry_704x1280_tp1_sp4_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_retry_704x1280_tp1_sp4_v1/captures/qkv_rank0_7a044869db5ec3c0.pt", "context": {"phase": "commit", "step": null, "start_frame": 27, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_math", "status": "measured", "q_shape": [10560, 10, 128], "k_shape": [66880, 10, 128], "q_len": 10560, "k_len": 63360, "reference_error": {"max_abs": 0.0625, "rmse": 0.0008074246660203004, "relative_l2": 0.001010215388992425, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.007813834310592682, "rmse": 0.0009001338955114059, "relative_l2": 0.0016855396963996849, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 146.3445587158203, "p95_ms": 146.89871978759766, "p99_ms": 147.21963272094726, "mean_ms": 146.35705774943034, "n": 300}, {"repeat": 1, "p50_ms": 146.41619110107422, "p95_ms": 147.13813705444335, "p99_ms": 147.34997344970702, "mean_ms": 146.4600715637207, "n": 300}, {"repeat": 2, "p50_ms": 146.36428833007812, "p95_ms": 147.0126823425293, "p99_ms": 147.41322174072266, "mean_ms": 146.39627604166665, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_7a044869db5ec3c0_sdpa_math.json", "configuration": "704x1280_tp1_sp4", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_retry_704x1280_tp1_sp4_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_retry_704x1280_tp1_sp4_v1/captures/qkv_rank0_7a044869db5ec3c0.pt", "context": {"phase": "commit", "step": null, "start_frame": 27, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_flash", "status": "measured", "q_shape": [10560, 10, 128], "k_shape": [66880, 10, 128], "q_len": 10560, "k_len": 63360, "reference_error": {"max_abs": 0.0625, "rmse": 0.0003910356077344753, "relative_l2": 0.0004892471151821863, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.00992615478884451, "rmse": 0.0009261736710817458, "relative_l2": 0.001734300303713789, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 10.314352035522461, "p95_ms": 10.656587028503418, "p99_ms": 10.937231874465942, "mean_ms": 10.383163951237997, "n": 300}, {"repeat": 1, "p50_ms": 10.374896049499512, "p95_ms": 10.575713777542115, "p99_ms": 10.59233003616333, "mean_ms": 10.383795738220215, "n": 300}, {"repeat": 2, "p50_ms": 10.31712007522583, "p95_ms": 10.592142295837402, "p99_ms": 10.720157594680785, "mean_ms": 10.383571192423503, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_7a044869db5ec3c0_sdpa_flash.json", "configuration": "704x1280_tp1_sp4", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_retry_704x1280_tp1_sp4_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_retry_704x1280_tp1_sp4_v1/captures/qkv_rank0_7a044869db5ec3c0.pt", "context": {"phase": "commit", "step": null, "start_frame": 27, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_efficient", "status": "measured", "q_shape": [10560, 10, 128], "k_shape": [66880, 10, 128], "q_len": 10560, "k_len": 63360, "reference_error": {"max_abs": 0.0625, "rmse": 0.0006881223231312975, "relative_l2": 0.000860949373472543, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.008974668806738695, "rmse": 0.0009140189670257117, "relative_l2": 0.0017115400940532074, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 19.841663360595703, "p95_ms": 19.86812105178833, "p99_ms": 19.881741485595704, "mean_ms": 19.84277759552002, "n": 300}, {"repeat": 1, "p50_ms": 19.842415809631348, "p95_ms": 19.866203784942627, "p99_ms": 19.879205226898193, "mean_ms": 19.842970027923585, "n": 300}, {"repeat": 2, "p50_ms": 19.841456413269043, "p95_ms": 19.869067859649657, "p99_ms": 19.891674728393554, "mean_ms": 19.8427024269104, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_7a044869db5ec3c0_sdpa_efficient.json", "configuration": "704x1280_tp1_sp4", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_retry_704x1280_tp1_sp4_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_retry_704x1280_tp1_sp4_v1/captures/qkv_rank0_7a044869db5ec3c0.pt", "context": {"phase": "commit", "step": null, "start_frame": 27, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_cudnn", "status": "measured", "q_shape": [10560, 10, 128], "k_shape": [66880, 10, 128], "q_len": 10560, "k_len": 63360, "reference_error": {"max_abs": 0.0625, "rmse": 0.0006891798601231863, "relative_l2": 0.0008622725187622458, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.008974668806738695, "rmse": 0.0009150847472603187, "relative_l2": 0.001713535813692281, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 5.756848096847534, "p95_ms": 6.112135934829712, "p99_ms": 6.585379099845885, "mean_ms": 5.795299417177836, "n": 300}, {"repeat": 1, "p50_ms": 5.809679985046387, "p95_ms": 5.943579053878785, "p99_ms": 5.989906430244446, "mean_ms": 5.808113598823548, "n": 300}, {"repeat": 2, "p50_ms": 5.817807912826538, "p95_ms": 5.9258030414581295, "p99_ms": 5.942891349792481, "mean_ms": 5.814850026766459, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_7a044869db5ec3c0_sdpa_cudnn.json", "configuration": "704x1280_tp1_sp4", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_retry_704x1280_tp1_sp4_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_retry_704x1280_tp1_sp4_v1/captures/qkv_rank0_7a044869db5ec3c0.pt", "context": {"phase": "commit", "step": null, "start_frame": 27, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "native_dense", "status": "measured", "q_shape": [10560, 10, 128], "k_shape": [66880, 10, 128], "q_len": 10560, "k_len": 63360, "reference_error": {"max_abs": 0.0625, "rmse": 0.0003829076717093357, "relative_l2": 0.00047907778744315057, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.010579403963054013, "rmse": 0.0009298519468354258, "relative_l2": 0.0017411880343370311, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 5.179247856140137, "p95_ms": 5.6964175462722775, "p99_ms": 5.910116386413574, "mean_ms": 5.20969886302948, "n": 300}, {"repeat": 1, "p50_ms": 5.227263927459717, "p95_ms": 5.478964638710022, "p99_ms": 5.535717206001282, "mean_ms": 5.240874977111816, "n": 300}, {"repeat": 2, "p50_ms": 5.226111888885498, "p95_ms": 5.32021267414093, "p99_ms": 5.329506254196167, "mean_ms": 5.221868057250976, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_7a044869db5ec3c0_native_dense.json", "configuration": "704x1280_tp1_sp4", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_retry_704x1280_tp1_sp4_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_retry_704x1280_tp1_sp4_v1/captures/qkv_rank0_8d9062a68f7c0930.pt", "context": {"phase": "commit", "step": null, "start_frame": 0, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "vllm_fa2", "status": "measured", "q_shape": [10560, 10, 128], "k_shape": [66880, 10, 128], "q_len": 10560, "k_len": 10560, "reference_error": {"max_abs": 0.03125, "rmse": 0.00037573398084062725, "relative_l2": 0.0005104879221844336, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.008178553063441285, "rmse": 0.0007568670538157323, "relative_l2": 0.001639986913325226, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 1.596735954284668, "p95_ms": 1.6228895843029023, "p99_ms": 1.623820185661316, "mean_ms": 1.599337172905604, "n": 300}, {"repeat": 1, "p50_ms": 1.5964800119400024, "p95_ms": 1.6634191989898681, "p99_ms": 1.7221765959262847, "mean_ms": 1.60815061489741, "n": 300}, {"repeat": 2, "p50_ms": 1.5970399975776672, "p95_ms": 1.6285216212272644, "p99_ms": 1.6638522148132324, "mean_ms": 1.6079876267910003, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_8d9062a68f7c0930_vllm_fa2.json", "configuration": "704x1280_tp1_sp4", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_retry_704x1280_tp1_sp4_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_retry_704x1280_tp1_sp4_v1/captures/qkv_rank0_8d9062a68f7c0930.pt", "context": {"phase": "commit", "step": null, "start_frame": 0, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "vllm_fa3", "status": "measured", "q_shape": [10560, 10, 128], "k_shape": [66880, 10, 128], "q_len": 10560, "k_len": 10560, "reference_error": {"max_abs": 0.0, "rmse": 0.0, "relative_l2": 0.0, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.008178553063441285, "rmse": 0.0007573742732439552, "relative_l2": 0.0016410859613288037, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.8922080099582672, "p95_ms": 0.9245455771684646, "p99_ms": 0.9252172988653183, "mean_ms": 0.8888708271582921, "n": 300}, {"repeat": 1, "p50_ms": 0.8897280097007751, "p95_ms": 0.8937632232904434, "p99_ms": 0.8955843073129653, "mean_ms": 0.8817827206850052, "n": 300}, {"repeat": 2, "p50_ms": 0.8899360001087189, "p95_ms": 0.8944400072097778, "p99_ms": 0.8953289455175399, "mean_ms": 0.8801239466667176, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_8d9062a68f7c0930_vllm_fa3.json", "configuration": "704x1280_tp1_sp4", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_retry_704x1280_tp1_sp4_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_retry_704x1280_tp1_sp4_v1/captures/qkv_rank0_8d9062a68f7c0930.pt", "context": {"phase": "commit", "step": null, "start_frame": 0, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_math", "status": "measured", "q_shape": [10560, 10, 128], "k_shape": [66880, 10, 128], "q_len": 10560, "k_len": 10560, "reference_error": {"max_abs": 0.03125, "rmse": 0.0006815312572388022, "relative_l2": 0.0009259569087501649, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.007719305713610236, "rmse": 0.0007465987096451862, "relative_l2": 0.0016177373650375622, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 23.34814453125, "p95_ms": 23.51060094833374, "p99_ms": 23.757438831329345, "mean_ms": 23.352075436909992, "n": 300}, {"repeat": 1, "p50_ms": 23.35366439819336, "p95_ms": 23.55852403640747, "p99_ms": 23.73246681213379, "mean_ms": 23.371453768412273, "n": 300}, {"repeat": 2, "p50_ms": 23.341567993164062, "p95_ms": 23.517773342132568, "p99_ms": 23.66481771469116, "mean_ms": 23.348388710021972, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_8d9062a68f7c0930_sdpa_math.json", "configuration": "704x1280_tp1_sp4", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_retry_704x1280_tp1_sp4_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_retry_704x1280_tp1_sp4_v1/captures/qkv_rank0_8d9062a68f7c0930.pt", "context": {"phase": "commit", "step": null, "start_frame": 0, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_flash", "status": "measured", "q_shape": [10560, 10, 128], "k_shape": [66880, 10, 128], "q_len": 10560, "k_len": 10560, "reference_error": {"max_abs": 0.03125, "rmse": 0.000376122799409107, "relative_l2": 0.0005110161873753698, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.008178553063441285, "rmse": 0.0007568670538157323, "relative_l2": 0.001639986913325226, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 1.7336959838867188, "p95_ms": 1.7379903435707091, "p99_ms": 1.7388253021240234, "mean_ms": 1.731640748977661, "n": 300}, {"repeat": 1, "p50_ms": 1.7341600060462952, "p95_ms": 1.7607760548591613, "p99_ms": 1.7652501821517945, "mean_ms": 1.7360644261042277, "n": 300}, {"repeat": 2, "p50_ms": 1.7345119714736938, "p95_ms": 1.7463535845279694, "p99_ms": 1.7514630615711213, "mean_ms": 1.7357829364140829, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_8d9062a68f7c0930_sdpa_flash.json", "configuration": "704x1280_tp1_sp4", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_retry_704x1280_tp1_sp4_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_retry_704x1280_tp1_sp4_v1/captures/qkv_rank0_8d9062a68f7c0930.pt", "context": {"phase": "commit", "step": null, "start_frame": 0, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_efficient", "status": "measured", "q_shape": [10560, 10, 128], "k_shape": [66880, 10, 128], "q_len": 10560, "k_len": 10560, "reference_error": {"max_abs": 0.03125, "rmse": 0.0005676074721655145, "relative_l2": 0.0007711752831986643, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.007912222877458852, "rmse": 0.000753810530580441, "relative_l2": 0.0016333640089711773, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 3.304287910461426, "p95_ms": 3.36054722070694, "p99_ms": 3.366661183834076, "mean_ms": 3.3093390893936157, "n": 300}, {"repeat": 1, "p50_ms": 3.3033279180526733, "p95_ms": 3.315942478179932, "p99_ms": 3.3185949516296387, "mean_ms": 3.3039300258954367, "n": 300}, {"repeat": 2, "p50_ms": 3.3030879497528076, "p95_ms": 3.3152064800262453, "p99_ms": 3.317858953475952, "mean_ms": 3.3037856006622315, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_8d9062a68f7c0930_sdpa_efficient.json", "configuration": "704x1280_tp1_sp4", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_retry_704x1280_tp1_sp4_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_retry_704x1280_tp1_sp4_v1/captures/qkv_rank0_8d9062a68f7c0930.pt", "context": {"phase": "commit", "step": null, "start_frame": 0, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_cudnn", "status": "measured", "q_shape": [10560, 10, 128], "k_shape": [66880, 10, 128], "q_len": 10560, "k_len": 10560, "reference_error": {"max_abs": 0.03125, "rmse": 0.000564871653276463, "relative_l2": 0.0007674582850793561, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.007912222877458852, "rmse": 0.0007538871835807566, "relative_l2": 0.0016335301014397962, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 1.0452640056610107, "p95_ms": 1.048641610145569, "p99_ms": 1.0495366299152373, "mean_ms": 1.0358363723754882, "n": 300}, {"repeat": 1, "p50_ms": 1.0450880527496338, "p95_ms": 1.0561647772789002, "p99_ms": 1.0604815983772278, "mean_ms": 1.0397959446907044, "n": 300}, {"repeat": 2, "p50_ms": 1.045744001865387, "p95_ms": 1.1229791462421417, "p99_ms": 1.1338560581207275, "mean_ms": 1.0572567466894787, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_8d9062a68f7c0930_sdpa_cudnn.json", "configuration": "704x1280_tp1_sp4", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_retry_704x1280_tp1_sp4_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_retry_704x1280_tp1_sp4_v1/captures/qkv_rank0_8d9062a68f7c0930.pt", "context": {"phase": "commit", "step": null, "start_frame": 0, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "native_dense", "status": "measured", "q_shape": [10560, 10, 128], "k_shape": [66880, 10, 128], "q_len": 10560, "k_len": 10560, "reference_error": {"max_abs": 0.03125, "rmse": 0.00036106901725436763, "relative_l2": 0.0004905634884845297, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.008178553063441285, "rmse": 0.0007583933289138638, "relative_l2": 0.0016432940610923942, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.864192008972168, "p95_ms": 0.8833344131708145, "p99_ms": 0.8876527863740921, "mean_ms": 0.8510691195726394, "n": 300}, {"repeat": 1, "p50_ms": 0.8598720133304596, "p95_ms": 0.8802239894866943, "p99_ms": 0.8884495997428894, "mean_ms": 0.8616356253623962, "n": 300}, {"repeat": 2, "p50_ms": 0.8702239990234375, "p95_ms": 0.9064335733652115, "p99_ms": 0.9090928155183792, "mean_ms": 0.8698675193389257, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_8d9062a68f7c0930_native_dense.json", "configuration": "704x1280_tp1_sp4", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_retry_704x1280_tp1_sp4_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_retry_704x1280_tp1_sp4_v1/captures/qkv_rank0_b0c50d31de479cf8.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 0, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "vllm_fa2", "status": "measured", "q_shape": [10560, 10, 128], "k_shape": [66880, 10, 128], "q_len": 10560, "k_len": 10560, "reference_error": {"max_abs": 0.03125, "rmse": 0.0004602674431208671, "relative_l2": 0.0008376689748024865, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.008999900738183353, "rmse": 0.0006804288244503694, "relative_l2": 0.0018858608044861479, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 1.6094080209732056, "p95_ms": 1.6638143897056579, "p99_ms": 1.664832636117935, "mean_ms": 1.61791979153951, "n": 300}, {"repeat": 1, "p50_ms": 1.6121280193328857, "p95_ms": 1.7080063462257384, "p99_ms": 1.7230762112140656, "mean_ms": 1.62897140900294, "n": 300}, {"repeat": 2, "p50_ms": 1.6359360218048096, "p95_ms": 1.6654096126556397, "p99_ms": 1.6665311861038208, "mean_ms": 1.629786559343338, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_b0c50d31de479cf8_vllm_fa2.json", "configuration": "704x1280_tp1_sp4", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_retry_704x1280_tp1_sp4_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_retry_704x1280_tp1_sp4_v1/captures/qkv_rank0_b0c50d31de479cf8.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 0, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "vllm_fa3", "status": "measured", "q_shape": [10560, 10, 128], "k_shape": [66880, 10, 128], "q_len": 10560, "k_len": 10560, "reference_error": {"max_abs": 0.0, "rmse": 0.0, "relative_l2": 0.0, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.008999900738183353, "rmse": 0.0006941992574560287, "relative_l2": 0.0019240266183567644, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.9022079706192017, "p95_ms": 0.9092191874980926, "p99_ms": 0.9101759791374207, "mean_ms": 0.8963633062442143, "n": 300}, {"repeat": 1, "p50_ms": 0.9066720008850098, "p95_ms": 0.916545620560646, "p99_ms": 0.9174457573890686, "mean_ms": 0.9021849606434504, "n": 300}, {"repeat": 2, "p50_ms": 0.9011360108852386, "p95_ms": 0.9173455744981766, "p99_ms": 0.9345443135499955, "mean_ms": 0.9015195735295614, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_b0c50d31de479cf8_vllm_fa3.json", "configuration": "704x1280_tp1_sp4", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_retry_704x1280_tp1_sp4_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_retry_704x1280_tp1_sp4_v1/captures/qkv_rank0_b0c50d31de479cf8.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 0, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_math", "status": "measured", "q_shape": [10560, 10, 128], "k_shape": [66880, 10, 128], "q_len": 10560, "k_len": 10560, "reference_error": {"max_abs": 0.03125, "rmse": 0.0007805294045195659, "relative_l2": 0.0014205333787108735, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.007237463011211975, "rmse": 0.0005975797342955189, "relative_l2": 0.001656238180934626, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 23.363487243652344, "p95_ms": 23.636722755432128, "p99_ms": 23.82139097213745, "mean_ms": 23.393171164194744, "n": 300}, {"repeat": 1, "p50_ms": 23.36684799194336, "p95_ms": 23.59315776824951, "p99_ms": 23.703374214172364, "mean_ms": 23.393641045888266, "n": 300}, {"repeat": 2, "p50_ms": 23.370128631591797, "p95_ms": 23.592729091644287, "p99_ms": 23.656777744293212, "mean_ms": 23.399381364186606, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_b0c50d31de479cf8_sdpa_math.json", "configuration": "704x1280_tp1_sp4", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_retry_704x1280_tp1_sp4_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_retry_704x1280_tp1_sp4_v1/captures/qkv_rank0_b0c50d31de479cf8.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 0, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_flash", "status": "measured", "q_shape": [10560, 10, 128], "k_shape": [66880, 10, 128], "q_len": 10560, "k_len": 10560, "reference_error": {"max_abs": 0.03125, "rmse": 0.00046059234331743477, "relative_l2": 0.0008382602806153126, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.008999900738183353, "rmse": 0.0006803588393833245, "relative_l2": 0.001885666835491748, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 1.7359039783477783, "p95_ms": 1.7510720491409302, "p99_ms": 1.7763542878627776, "mean_ms": 1.7391064568360646, "n": 300}, {"repeat": 1, "p50_ms": 1.7462559938430786, "p95_ms": 1.824513602256775, "p99_ms": 1.8281929552555085, "mean_ms": 1.761087573369344, "n": 300}, {"repeat": 2, "p50_ms": 1.7464479804039001, "p95_ms": 1.7923872292041778, "p99_ms": 1.8077186787128448, "mean_ms": 1.7560096005598704, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_b0c50d31de479cf8_sdpa_flash.json", "configuration": "704x1280_tp1_sp4", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_retry_704x1280_tp1_sp4_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_retry_704x1280_tp1_sp4_v1/captures/qkv_rank0_b0c50d31de479cf8.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 0, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_efficient", "status": "measured", "q_shape": [10560, 10, 128], "k_shape": [66880, 10, 128], "q_len": 10560, "k_len": 10560, "reference_error": {"max_abs": 0.03125, "rmse": 0.0007052316042652887, "relative_l2": 0.001283494289619101, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.007237463011211975, "rmse": 0.0006850272498120121, "relative_l2": 0.0018986056939444714, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 3.3071519136428833, "p95_ms": 3.371491289138794, "p99_ms": 3.3797859215736388, "mean_ms": 3.316605445543925, "n": 300}, {"repeat": 1, "p50_ms": 3.304784059524536, "p95_ms": 3.3171648263931273, "p99_ms": 3.3211542129516602, "mean_ms": 3.305575787226359, "n": 300}, {"repeat": 2, "p50_ms": 3.304640054702759, "p95_ms": 3.3179519176483154, "p99_ms": 3.3216284418106077, "mean_ms": 3.3055989305178324, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_b0c50d31de479cf8_sdpa_efficient.json", "configuration": "704x1280_tp1_sp4", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_retry_704x1280_tp1_sp4_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_retry_704x1280_tp1_sp4_v1/captures/qkv_rank0_b0c50d31de479cf8.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 0, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_cudnn", "status": "measured", "q_shape": [10560, 10, 128], "k_shape": [66880, 10, 128], "q_len": 10560, "k_len": 10560, "reference_error": {"max_abs": 0.03125, "rmse": 0.0007033764957477413, "relative_l2": 0.0012801180637459337, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.007237463011211975, "rmse": 0.0006857538348473918, "relative_l2": 0.0019006194802364555, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 1.058127999305725, "p95_ms": 1.0657919645309448, "p99_ms": 1.0673964715003967, "mean_ms": 1.0478980267047882, "n": 300}, {"repeat": 1, "p50_ms": 1.0560319423675537, "p95_ms": 1.0645776271820069, "p99_ms": 1.065218505859375, "mean_ms": 1.0536855443318685, "n": 300}, {"repeat": 2, "p50_ms": 1.06387197971344, "p95_ms": 1.1693455517292022, "p99_ms": 1.170306181907654, "mean_ms": 1.0847258647282918, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_b0c50d31de479cf8_sdpa_cudnn.json", "configuration": "704x1280_tp1_sp4", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_retry_704x1280_tp1_sp4_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_retry_704x1280_tp1_sp4_v1/captures/qkv_rank0_b0c50d31de479cf8.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 0, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "native_dense", "status": "measured", "q_shape": [10560, 10, 128], "k_shape": [66880, 10, 128], "q_len": 10560, "k_len": 10560, "reference_error": {"max_abs": 0.03125, "rmse": 0.0004365866080838657, "relative_l2": 0.0007945707693908502, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.008999900738183353, "rmse": 0.0006932121785677012, "relative_l2": 0.0019212908533222126, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.8838559985160828, "p95_ms": 0.9001408070325851, "p99_ms": 0.9030774396657943, "mean_ms": 0.8807166916131973, "n": 300}, {"repeat": 1, "p50_ms": 0.8843199908733368, "p95_ms": 0.9299264162778854, "p99_ms": 0.9356220960617065, "mean_ms": 0.8846096009016037, "n": 300}, {"repeat": 2, "p50_ms": 0.8877440094947815, "p95_ms": 0.9093520164489746, "p99_ms": 0.9115856194496155, "mean_ms": 0.8853976531823476, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_b0c50d31de479cf8_native_dense.json", "configuration": "704x1280_tp1_sp4", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_retry_704x1280_tp1_sp4_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_retry_704x1280_tp1_sp4_v1/captures/qkv_rank0_b37e08eb3a08f049.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 0, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "vllm_fa2", "status": "measured", "q_shape": [10560, 10, 128], "k_shape": [512, 10, 128], "q_len": 10560, "k_len": 512, "reference_error": {"max_abs": 0.00048828125, "rmse": 1.839457799312546e-06, "relative_l2": 0.00030271724389130546, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.0004067786029473136, "rmse": 1.930963898854266e-05, "relative_l2": 0.0018472856840449385, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.09548799693584442, "p95_ms": 0.09638399630784988, "p99_ms": 0.09692800045013428, "mean_ms": 0.09568992001314958, "n": 300}, {"repeat": 1, "p50_ms": 0.09548799693584442, "p95_ms": 0.09638559632003307, "p99_ms": 0.09683327972888947, "mean_ms": 0.09587050673862298, "n": 300}, {"repeat": 2, "p50_ms": 0.09551999717950821, "p95_ms": 0.09644799679517746, "p99_ms": 0.09689664021134377, "mean_ms": 0.09590293320516745, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_b37e08eb3a08f049_vllm_fa2.json", "configuration": "704x1280_tp1_sp4", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_retry_704x1280_tp1_sp4_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_retry_704x1280_tp1_sp4_v1/captures/qkv_rank0_b37e08eb3a08f049.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 0, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "vllm_fa3", "status": "measured", "q_shape": [10560, 10, 128], "k_shape": [512, 10, 128], "q_len": 10560, "k_len": 512, "reference_error": {"max_abs": 0.00048828125, "rmse": 6.687721192290474e-07, "relative_l2": 0.00011005898194567184, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.0004067786029473136, "rmse": 1.936247177353708e-05, "relative_l2": 0.0018523400119599413, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.07983999699354172, "p95_ms": 0.08067520298063754, "p99_ms": 0.08131200075149536, "mean_ms": 0.08002912017206351, "n": 300}, {"repeat": 1, "p50_ms": 0.0799039974808693, "p95_ms": 0.08089599758386612, "p99_ms": 0.081312320753932, "mean_ms": 0.08039509368439515, "n": 300}, {"repeat": 2, "p50_ms": 0.07985599711537361, "p95_ms": 0.08083199709653854, "p99_ms": 0.08144096173346042, "mean_ms": 0.08034709346791108, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_b37e08eb3a08f049_vllm_fa3.json", "configuration": "704x1280_tp1_sp4", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_retry_704x1280_tp1_sp4_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_retry_704x1280_tp1_sp4_v1/captures/qkv_rank0_b37e08eb3a08f049.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 0, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_math", "status": "measured", "q_shape": [10560, 10, 128], "k_shape": [512, 10, 128], "q_len": 10560, "k_len": 512, "reference_error": {"max_abs": 0.001953125, "rmse": 1.2862039923273465e-05, "relative_l2": 0.002116689645094553, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.0003842328223661873, "rmse": 1.6787201932200433e-05, "relative_l2": 0.0016059729455804724, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 1.264639973640442, "p95_ms": 1.2677119970321655, "p99_ms": 1.2688118600845337, "mean_ms": 1.2649934907754263, "n": 300}, {"repeat": 1, "p50_ms": 1.2647680044174194, "p95_ms": 1.2685152113437652, "p99_ms": 1.270624041557312, "mean_ms": 1.26543381412824, "n": 300}, {"repeat": 2, "p50_ms": 1.2649919986724854, "p95_ms": 1.2684575974941255, "p99_ms": 1.2697619330883025, "mean_ms": 1.2655314167340597, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_b37e08eb3a08f049_sdpa_math.json", "configuration": "704x1280_tp1_sp4", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_retry_704x1280_tp1_sp4_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_retry_704x1280_tp1_sp4_v1/captures/qkv_rank0_b37e08eb3a08f049.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 0, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_flash", "status": "measured", "q_shape": [10560, 10, 128], "k_shape": [512, 10, 128], "q_len": 10560, "k_len": 512, "reference_error": {"max_abs": 0.00048828125, "rmse": 1.8461338846084558e-06, "relative_l2": 0.0003038159188060093, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.0004067786029473136, "rmse": 1.930963898854266e-05, "relative_l2": 0.0018472856840449385, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.10252799838781357, "p95_ms": 0.10336319729685783, "p99_ms": 0.10400064215064049, "mean_ms": 0.10267018631100655, "n": 300}, {"repeat": 1, "p50_ms": 0.10260799899697304, "p95_ms": 0.10339359752833843, "p99_ms": 0.10380864068865776, "mean_ms": 0.10298879981040955, "n": 300}, {"repeat": 2, "p50_ms": 0.10262399911880493, "p95_ms": 0.1035199984908104, "p99_ms": 0.10393792167305946, "mean_ms": 0.1030287998666366, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_b37e08eb3a08f049_sdpa_flash.json", "configuration": "704x1280_tp1_sp4", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_retry_704x1280_tp1_sp4_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_retry_704x1280_tp1_sp4_v1/captures/qkv_rank0_b37e08eb3a08f049.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 0, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_efficient", "status": "measured", "q_shape": [10560, 10, 128], "k_shape": [512, 10, 128], "q_len": 10560, "k_len": 512, "reference_error": {"max_abs": 0.0009765625, "rmse": 3.253070969936434e-06, "relative_l2": 0.0005353537757539228, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.0004067786029473136, "rmse": 1.9527557065743348e-05, "relative_l2": 0.001868133144970867, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.17875200510025024, "p95_ms": 0.17936000227928162, "p99_ms": 0.1799369665980339, "mean_ms": 0.1789408004283905, "n": 300}, {"repeat": 1, "p50_ms": 0.178847998380661, "p95_ms": 0.17974560484290122, "p99_ms": 0.18045087575912475, "mean_ms": 0.17921984006961186, "n": 300}, {"repeat": 2, "p50_ms": 0.17923200130462646, "p95_ms": 0.1799039989709854, "p99_ms": 0.18028864189982413, "mean_ms": 0.1794506664077441, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_b37e08eb3a08f049_sdpa_efficient.json", "configuration": "704x1280_tp1_sp4", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_retry_704x1280_tp1_sp4_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_retry_704x1280_tp1_sp4_v1/captures/qkv_rank0_b37e08eb3a08f049.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 0, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_cudnn", "status": "measured", "q_shape": [10560, 10, 128], "k_shape": [512, 10, 128], "q_len": 10560, "k_len": 512, "reference_error": {"max_abs": 0.0009765625, "rmse": 3.2429616662711494e-06, "relative_l2": 0.0005336901004337502, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.0004067786029473136, "rmse": 1.952755709178034e-05, "relative_l2": 0.0018681331474617357, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.05920000001788139, "p95_ms": 0.06015999987721443, "p99_ms": 0.06401824109256242, "mean_ms": 0.05974688003460566, "n": 300}, {"repeat": 1, "p50_ms": 0.059167999774217606, "p95_ms": 0.06006399914622307, "p99_ms": 0.06032352108508348, "mean_ms": 0.05967264005293449, "n": 300}, {"repeat": 2, "p50_ms": 0.05907199904322624, "p95_ms": 0.05990400165319443, "p99_ms": 0.060256960615515706, "mean_ms": 0.059569706531862415, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_b37e08eb3a08f049_sdpa_cudnn.json", "configuration": "704x1280_tp1_sp4", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_retry_704x1280_tp1_sp4_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_retry_704x1280_tp1_sp4_v1/captures/qkv_rank0_b37e08eb3a08f049.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 0, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "native_dense", "status": "measured", "q_shape": [10560, 10, 128], "k_shape": [512, 10, 128], "q_len": 10560, "k_len": 512, "reference_error": {"max_abs": 0.0, "rmse": 0.0, "relative_l2": 0.0, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.0004067786029473136, "rmse": 1.9372534423903118e-05, "relative_l2": 0.001853302670556291, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.12412799894809723, "p95_ms": 0.13855520412325859, "p99_ms": 0.20534687876701352, "mean_ms": 0.1273834667354822, "n": 300}, {"repeat": 1, "p50_ms": 0.12376000359654427, "p95_ms": 0.12912480682134628, "p99_ms": 0.14060895681381214, "mean_ms": 0.1249771733333667, "n": 300}, {"repeat": 2, "p50_ms": 0.12380799651145935, "p95_ms": 0.13015199974179267, "p99_ms": 0.15445695877075194, "mean_ms": 0.12548309328655402, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_b37e08eb3a08f049_native_dense.json", "configuration": "704x1280_tp1_sp4", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_retry_704x1280_tp1_sp4_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_retry_704x1280_tp1_sp4_v1/captures/qkv_rank0_be6bbd23e5a64d4b.pt", "context": {"phase": "commit", "step": null, "start_frame": 9, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "vllm_fa2", "status": "measured", "q_shape": [10560, 10, 128], "k_shape": [512, 10, 128], "q_len": 10560, "k_len": 512, "reference_error": {"max_abs": 0.0001220703125, "rmse": 1.1620982132694709e-06, "relative_l2": 0.0008353918200216838, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 6.307664314882458e-05, "rmse": 4.137486204813533e-06, "relative_l2": 0.00178708309216853, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.0958079993724823, "p95_ms": 0.09673599898815155, "p99_ms": 0.09724864289164543, "mean_ms": 0.09601535973449549, "n": 300}, {"repeat": 1, "p50_ms": 0.09574399888515472, "p95_ms": 0.09667359851300716, "p99_ms": 0.09696000069379807, "mean_ms": 0.09612917299071948, "n": 300}, {"repeat": 2, "p50_ms": 0.09575999900698662, "p95_ms": 0.09660959802567959, "p99_ms": 0.09721600264310837, "mean_ms": 0.09612607990701993, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_be6bbd23e5a64d4b_vllm_fa2.json", "configuration": "704x1280_tp1_sp4", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_retry_704x1280_tp1_sp4_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_retry_704x1280_tp1_sp4_v1/captures/qkv_rank0_be6bbd23e5a64d4b.pt", "context": {"phase": "commit", "step": null, "start_frame": 9, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "vllm_fa3", "status": "measured", "q_shape": [10560, 10, 128], "k_shape": [512, 10, 128], "q_len": 10560, "k_len": 512, "reference_error": {"max_abs": 0.0001220703125, "rmse": 4.382912228427194e-07, "relative_l2": 0.0003150722530757441, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 6.307664314882458e-05, "rmse": 4.161822254469216e-06, "relative_l2": 0.0017975944366702345, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.08086399734020233, "p95_ms": 0.08182719722390175, "p99_ms": 0.08252832256257533, "mean_ms": 0.08139509332676728, "n": 300}, {"repeat": 1, "p50_ms": 0.08099199831485748, "p95_ms": 0.08195359818637371, "p99_ms": 0.08233760111033916, "mean_ms": 0.08129727999369303, "n": 300}, {"repeat": 2, "p50_ms": 0.08115199953317642, "p95_ms": 0.08217599987983704, "p99_ms": 0.08268831633031368, "mean_ms": 0.08156469335158666, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_be6bbd23e5a64d4b_vllm_fa3.json", "configuration": "704x1280_tp1_sp4", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_retry_704x1280_tp1_sp4_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_retry_704x1280_tp1_sp4_v1/captures/qkv_rank0_be6bbd23e5a64d4b.pt", "context": {"phase": "commit", "step": null, "start_frame": 9, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_math", "status": "measured", "q_shape": [10560, 10, 128], "k_shape": [512, 10, 128], "q_len": 10560, "k_len": 512, "reference_error": {"max_abs": 0.000244140625, "rmse": 2.1518178359759678e-06, "relative_l2": 0.0015468666914938704, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 6.095633961755026e-05, "rmse": 3.932673083918934e-06, "relative_l2": 0.001698619216451148, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 1.2637120485305786, "p95_ms": 1.2673631727695465, "p99_ms": 1.269417908191681, "mean_ms": 1.2640514143308004, "n": 300}, {"repeat": 1, "p50_ms": 1.2643200159072876, "p95_ms": 1.2676527798175812, "p99_ms": 1.2691558849811555, "mean_ms": 1.2647107191880544, "n": 300}, {"repeat": 2, "p50_ms": 1.264016032218933, "p95_ms": 1.2670480132102966, "p99_ms": 1.2696332621574402, "mean_ms": 1.2642842694123586, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_be6bbd23e5a64d4b_sdpa_math.json", "configuration": "704x1280_tp1_sp4", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_retry_704x1280_tp1_sp4_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_retry_704x1280_tp1_sp4_v1/captures/qkv_rank0_be6bbd23e5a64d4b.pt", "context": {"phase": "commit", "step": null, "start_frame": 9, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_flash", "status": "measured", "q_shape": [10560, 10, 128], "k_shape": [512, 10, 128], "q_len": 10560, "k_len": 512, "reference_error": {"max_abs": 0.0001220703125, "rmse": 1.1621403387428585e-06, "relative_l2": 0.0008354221025532986, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 6.307664314882458e-05, "rmse": 4.137486204813533e-06, "relative_l2": 0.00178708309216853, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.10262399911880493, "p95_ms": 0.10349119827151299, "p99_ms": 0.10397088184952735, "mean_ms": 0.10278133317828178, "n": 300}, {"repeat": 1, "p50_ms": 0.10247999802231789, "p95_ms": 0.1035199984908104, "p99_ms": 0.11020416207611529, "mean_ms": 0.10420202635228634, "n": 300}, {"repeat": 2, "p50_ms": 0.10246399790048599, "p95_ms": 0.1032319962978363, "p99_ms": 0.10403232239186763, "mean_ms": 0.10275605301062266, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_be6bbd23e5a64d4b_sdpa_flash.json", "configuration": "704x1280_tp1_sp4", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_retry_704x1280_tp1_sp4_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_retry_704x1280_tp1_sp4_v1/captures/qkv_rank0_be6bbd23e5a64d4b.pt", "context": {"phase": "commit", "step": null, "start_frame": 9, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_efficient", "status": "measured", "q_shape": [10560, 10, 128], "k_shape": [512, 10, 128], "q_len": 10560, "k_len": 512, "reference_error": {"max_abs": 0.0001220703125, "rmse": 2.0252572287517085e-06, "relative_l2": 0.0014558866910043535, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 6.307664314882458e-05, "rmse": 4.736439880333926e-06, "relative_l2": 0.0020457860662761937, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.1788800060749054, "p95_ms": 0.1795536033809185, "p99_ms": 0.18000032708048821, "mean_ms": 0.1790041603644689, "n": 300}, {"repeat": 1, "p50_ms": 0.1788800060749054, "p95_ms": 0.17952159643173218, "p99_ms": 0.18009632036089898, "mean_ms": 0.17916213328639666, "n": 300}, {"repeat": 2, "p50_ms": 0.1788800060749054, "p95_ms": 0.17958559691905976, "p99_ms": 0.18009888038039207, "mean_ms": 0.17914410655697185, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_be6bbd23e5a64d4b_sdpa_efficient.json", "configuration": "704x1280_tp1_sp4", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_retry_704x1280_tp1_sp4_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_retry_704x1280_tp1_sp4_v1/captures/qkv_rank0_be6bbd23e5a64d4b.pt", "context": {"phase": "commit", "step": null, "start_frame": 9, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_cudnn", "status": "measured", "q_shape": [10560, 10, 128], "k_shape": [512, 10, 128], "q_len": 10560, "k_len": 512, "reference_error": {"max_abs": 0.0001220703125, "rmse": 2.0244064947004947e-06, "relative_l2": 0.00145527512800625, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 6.307664314882458e-05, "rmse": 4.736452756624415e-06, "relative_l2": 0.002045791627865982, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.05929600074887276, "p95_ms": 0.06006399914622307, "p99_ms": 0.060416641794145104, "mean_ms": 0.05956085342913866, "n": 300}, {"repeat": 1, "p50_ms": 0.05926400050520897, "p95_ms": 0.06003359891474247, "p99_ms": 0.0605145588517189, "mean_ms": 0.059769813356300196, "n": 300}, {"repeat": 2, "p50_ms": 0.05910399928689003, "p95_ms": 0.059777600690722466, "p99_ms": 0.06130720097571609, "mean_ms": 0.059518399847050506, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_be6bbd23e5a64d4b_sdpa_cudnn.json", "configuration": "704x1280_tp1_sp4", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_retry_704x1280_tp1_sp4_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_retry_704x1280_tp1_sp4_v1/captures/qkv_rank0_be6bbd23e5a64d4b.pt", "context": {"phase": "commit", "step": null, "start_frame": 9, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "native_dense", "status": "measured", "q_shape": [10560, 10, 128], "k_shape": [512, 10, 128], "q_len": 10560, "k_len": 512, "reference_error": {"max_abs": 0.0, "rmse": 0.0, "relative_l2": 0.0, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 6.307664314882458e-05, "rmse": 4.17386793468327e-06, "relative_l2": 0.0018027972652426081, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.12344000115990639, "p95_ms": 0.13030880093574526, "p99_ms": 0.16266464009881, "mean_ms": 0.12656277348597844, "n": 300}, {"repeat": 1, "p50_ms": 0.12332800030708313, "p95_ms": 0.1289920061826706, "p99_ms": 0.14041024252772316, "mean_ms": 0.12485226680835088, "n": 300}, {"repeat": 2, "p50_ms": 0.12427200004458427, "p95_ms": 0.13561919704079628, "p99_ms": 0.1498364734649658, "mean_ms": 0.12695328009625276, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_be6bbd23e5a64d4b_native_dense.json", "configuration": "704x1280_tp1_sp4", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_retry_704x1280_tp1_sp4_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_retry_704x1280_tp1_sp4_v1/captures/qkv_rank0_c0afb7cc208e55f9.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 27, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "vllm_fa2", "status": "measured", "q_shape": [10560, 10, 128], "k_shape": [66880, 10, 128], "q_len": 10560, "k_len": 63360, "reference_error": {"max_abs": 0.03125, "rmse": 0.0004943599240033328, "relative_l2": 0.0009516841611292657, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.008982203813774792, "rmse": 0.000752808786776018, "relative_l2": 0.001938066427686598, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 9.572495937347412, "p95_ms": 9.987800025939942, "p99_ms": 10.525089263916016, "mean_ms": 9.65882069269816, "n": 300}, {"repeat": 1, "p50_ms": 9.729471683502197, "p95_ms": 9.817766857147216, "p99_ms": 10.150214023590088, "mean_ms": 9.664067452748617, "n": 300}, {"repeat": 2, "p50_ms": 9.730192184448242, "p95_ms": 9.740390491485595, "p99_ms": 9.785109024047852, "mean_ms": 9.673703241348267, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_c0afb7cc208e55f9_vllm_fa2.json", "configuration": "704x1280_tp1_sp4", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_retry_704x1280_tp1_sp4_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_retry_704x1280_tp1_sp4_v1/captures/qkv_rank0_c0afb7cc208e55f9.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 27, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "vllm_fa3", "status": "measured", "q_shape": [10560, 10, 128], "k_shape": [66880, 10, 128], "q_len": 10560, "k_len": 63360, "reference_error": {"max_abs": 0.0, "rmse": 0.0, "relative_l2": 0.0, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.008982203813774792, "rmse": 0.0007603426648000307, "relative_l2": 0.0019574619984146654, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 5.22048020362854, "p95_ms": 5.628963327407837, "p99_ms": 6.123413863182068, "mean_ms": 5.275665820439657, "n": 300}, {"repeat": 1, "p50_ms": 5.29531192779541, "p95_ms": 5.451236748695374, "p99_ms": 5.455430235862732, "mean_ms": 5.298136107126871, "n": 300}, {"repeat": 2, "p50_ms": 5.249232053756714, "p95_ms": 5.359073805809021, "p99_ms": 5.509737553596497, "mean_ms": 5.274968962669373, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_c0afb7cc208e55f9_vllm_fa3.json", "configuration": "704x1280_tp1_sp4", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_retry_704x1280_tp1_sp4_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_retry_704x1280_tp1_sp4_v1/captures/qkv_rank0_c0afb7cc208e55f9.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 27, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_math", "status": "measured", "q_shape": [10560, 10, 128], "k_shape": [66880, 10, 128], "q_len": 10560, "k_len": 63360, "reference_error": {"max_abs": 0.03125, "rmse": 0.0009064916090442196, "relative_l2": 0.0017450720914791433, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.007363246840188076, "rmse": 0.0006496499745258247, "relative_l2": 0.0016724895185775267, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 146.1050033569336, "p95_ms": 146.7683135986328, "p99_ms": 147.11599472045899, "mean_ms": 146.11681208292643, "n": 300}, {"repeat": 1, "p50_ms": 146.55425262451172, "p95_ms": 147.11891021728516, "p99_ms": 147.3283720397949, "mean_ms": 146.51880457560222, "n": 300}, {"repeat": 2, "p50_ms": 146.45425415039062, "p95_ms": 147.0396484375, "p99_ms": 147.38763793945313, "mean_ms": 146.45558461507161, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_c0afb7cc208e55f9_sdpa_math.json", "configuration": "704x1280_tp1_sp4", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_retry_704x1280_tp1_sp4_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_retry_704x1280_tp1_sp4_v1/captures/qkv_rank0_c0afb7cc208e55f9.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 27, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_flash", "status": "measured", "q_shape": [10560, 10, 128], "k_shape": [66880, 10, 128], "q_len": 10560, "k_len": 63360, "reference_error": {"max_abs": 0.03125, "rmse": 0.0004948005912266606, "relative_l2": 0.0009525324823551728, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.008982203813774792, "rmse": 0.0007528591023141345, "relative_l2": 0.001938195962379773, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 10.329760074615479, "p95_ms": 10.661350393295288, "p99_ms": 10.840666484832763, "mean_ms": 10.400847155253093, "n": 300}, {"repeat": 1, "p50_ms": 10.39580774307251, "p95_ms": 10.599376106262207, "p99_ms": 10.663092432022095, "mean_ms": 10.406796261469523, "n": 300}, {"repeat": 2, "p50_ms": 10.378560066223145, "p95_ms": 10.574737977981567, "p99_ms": 10.674197845458984, "mean_ms": 10.39782346089681, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_c0afb7cc208e55f9_sdpa_flash.json", "configuration": "704x1280_tp1_sp4", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_retry_704x1280_tp1_sp4_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_retry_704x1280_tp1_sp4_v1/captures/qkv_rank0_c0afb7cc208e55f9.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 27, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_efficient", "status": "measured", "q_shape": [10560, 10, 128], "k_shape": [66880, 10, 128], "q_len": 10560, "k_len": 63360, "reference_error": {"max_abs": 0.03125, "rmse": 0.000782264437619323, "relative_l2": 0.0015059244063885388, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.00919098333101509, "rmse": 0.0007510854117432321, "relative_l2": 0.001933629689763201, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 19.851776123046875, "p95_ms": 19.880597591400146, "p99_ms": 19.897043991088868, "mean_ms": 19.85355178833008, "n": 300}, {"repeat": 1, "p50_ms": 19.852895736694336, "p95_ms": 19.88291492462158, "p99_ms": 19.907496833801268, "mean_ms": 19.854271443684897, "n": 300}, {"repeat": 2, "p50_ms": 19.850671768188477, "p95_ms": 19.877358531951906, "p99_ms": 19.899044284820558, "mean_ms": 19.852107327779134, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_c0afb7cc208e55f9_sdpa_efficient.json", "configuration": "704x1280_tp1_sp4", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_retry_704x1280_tp1_sp4_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_retry_704x1280_tp1_sp4_v1/captures/qkv_rank0_c0afb7cc208e55f9.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 27, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_cudnn", "status": "measured", "q_shape": [10560, 10, 128], "k_shape": [66880, 10, 128], "q_len": 10560, "k_len": 63360, "reference_error": {"max_abs": 0.03125, "rmse": 0.0007809252206089539, "relative_l2": 0.001503346301742106, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.00919098333101509, "rmse": 0.0007507384628611804, "relative_l2": 0.001932736488206259, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 5.75766396522522, "p95_ms": 6.715207958221436, "p99_ms": 6.917356996536255, "mean_ms": 5.823713916142782, "n": 300}, {"repeat": 1, "p50_ms": 5.811327934265137, "p95_ms": 6.100995397567749, "p99_ms": 6.1058361530303955, "mean_ms": 5.839136754671733, "n": 300}, {"repeat": 2, "p50_ms": 5.831807851791382, "p95_ms": 5.933281588554382, "p99_ms": 5.939339389801026, "mean_ms": 5.841869762738546, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_c0afb7cc208e55f9_sdpa_cudnn.json", "configuration": "704x1280_tp1_sp4", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_retry_704x1280_tp1_sp4_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_retry_704x1280_tp1_sp4_v1/captures/qkv_rank0_c0afb7cc208e55f9.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 27, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "native_dense", "status": "measured", "q_shape": [10560, 10, 128], "k_shape": [66880, 10, 128], "q_len": 10560, "k_len": 63360, "reference_error": {"max_abs": 0.03125, "rmse": 0.00047503087541053186, "relative_l2": 0.0009144741275033551, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.008982203813774792, "rmse": 0.0007623522854336237, "relative_l2": 0.0019626356605062473, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 5.169087886810303, "p95_ms": 5.711974239349367, "p99_ms": 6.057156276702881, "mean_ms": 5.223190925916036, "n": 300}, {"repeat": 1, "p50_ms": 5.249727964401245, "p95_ms": 5.480713438987732, "p99_ms": 5.49656500339508, "mean_ms": 5.2678515227635705, "n": 300}, {"repeat": 2, "p50_ms": 5.229295969009399, "p95_ms": 5.330148792266845, "p99_ms": 5.356945791244507, "mean_ms": 5.240111893018087, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_c0afb7cc208e55f9_native_dense.json", "configuration": "704x1280_tp1_sp4", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_retry_704x1280_tp1_sp4_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_retry_704x1280_tp1_sp4_v1/captures/qkv_rank0_d12872a3fe20c4e4.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 9, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "vllm_fa2", "status": "measured", "q_shape": [10560, 10, 128], "k_shape": [512, 10, 128], "q_len": 10560, "k_len": 512, "reference_error": {"max_abs": 0.0001220703125, "rmse": 1.2647666855784978e-06, "relative_l2": 0.0007143580127864644, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 6.37440089465216e-05, "rmse": 5.101539129336812e-06, "relative_l2": 0.0018193710309880842, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.09683199971914291, "p95_ms": 0.0976335983723402, "p99_ms": 0.09792064055800438, "mean_ms": 0.09696618683636188, "n": 300}, {"repeat": 1, "p50_ms": 0.09683199971914291, "p95_ms": 0.09782719984650612, "p99_ms": 0.09827328316867351, "mean_ms": 0.09712992000083129, "n": 300}, {"repeat": 2, "p50_ms": 0.09679999947547913, "p95_ms": 0.09788800030946732, "p99_ms": 0.09814432226121425, "mean_ms": 0.09715519987046718, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_d12872a3fe20c4e4_vllm_fa2.json", "configuration": "704x1280_tp1_sp4", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_retry_704x1280_tp1_sp4_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_retry_704x1280_tp1_sp4_v1/captures/qkv_rank0_d12872a3fe20c4e4.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 9, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "vllm_fa3", "status": "measured", "q_shape": [10560, 10, 128], "k_shape": [512, 10, 128], "q_len": 10560, "k_len": 512, "reference_error": {"max_abs": 0.0001220703125, "rmse": 4.7089486263600354e-07, "relative_l2": 0.0002659680414891294, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 6.37440089465216e-05, "rmse": 5.160676778595407e-06, "relative_l2": 0.0018404613966942096, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.08054400235414505, "p95_ms": 0.08137600123882294, "p99_ms": 0.08185023725032804, "mean_ms": 0.08099306675295036, "n": 300}, {"repeat": 1, "p50_ms": 0.08048000186681747, "p95_ms": 0.08137760125100613, "p99_ms": 0.08227424062788487, "mean_ms": 0.0809340801090002, "n": 300}, {"repeat": 2, "p50_ms": 0.08032000064849854, "p95_ms": 0.08137760125100613, "p99_ms": 0.0819526381790638, "mean_ms": 0.08091402659813562, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_d12872a3fe20c4e4_vllm_fa3.json", "configuration": "704x1280_tp1_sp4", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_retry_704x1280_tp1_sp4_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_retry_704x1280_tp1_sp4_v1/captures/qkv_rank0_d12872a3fe20c4e4.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 9, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_math", "status": "measured", "q_shape": [10560, 10, 128], "k_shape": [512, 10, 128], "q_len": 10560, "k_len": 512, "reference_error": {"max_abs": 0.000244140625, "rmse": 3.310928750177461e-06, "relative_l2": 0.0018700591258636117, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 6.09328142618451e-05, "rmse": 4.570944062662757e-06, "relative_l2": 0.0016301439626430338, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 1.2662879824638367, "p95_ms": 1.2694432497024537, "p99_ms": 1.270466504096985, "mean_ms": 1.2664493894577027, "n": 300}, {"repeat": 1, "p50_ms": 1.2666720151901245, "p95_ms": 1.2700175583362578, "p99_ms": 1.2716809570789338, "mean_ms": 1.2670418151219687, "n": 300}, {"repeat": 2, "p50_ms": 1.2664639949798584, "p95_ms": 1.2697648108005524, "p99_ms": 1.270499210357666, "mean_ms": 1.267304321527481, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_d12872a3fe20c4e4_sdpa_math.json", "configuration": "704x1280_tp1_sp4", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_retry_704x1280_tp1_sp4_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_retry_704x1280_tp1_sp4_v1/captures/qkv_rank0_d12872a3fe20c4e4.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 9, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_flash", "status": "measured", "q_shape": [10560, 10, 128], "k_shape": [512, 10, 128], "q_len": 10560, "k_len": 512, "reference_error": {"max_abs": 0.0001220703125, "rmse": 1.2650667444823654e-06, "relative_l2": 0.0007145274902756565, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 6.37440089465216e-05, "rmse": 5.101539129336812e-06, "relative_l2": 0.0018193710309880842, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.10255999863147736, "p95_ms": 0.10333119705319405, "p99_ms": 0.10377696044743061, "mean_ms": 0.10270037325719993, "n": 300}, {"repeat": 1, "p50_ms": 0.10265599936246872, "p95_ms": 0.1034591980278492, "p99_ms": 0.10396800190210342, "mean_ms": 0.10295114661256473, "n": 300}, {"repeat": 2, "p50_ms": 0.10268799960613251, "p95_ms": 0.1035199984908104, "p99_ms": 0.10406432263553142, "mean_ms": 0.10292917328576247, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_d12872a3fe20c4e4_sdpa_flash.json", "configuration": "704x1280_tp1_sp4", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_retry_704x1280_tp1_sp4_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_retry_704x1280_tp1_sp4_v1/captures/qkv_rank0_d12872a3fe20c4e4.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 9, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_efficient", "status": "measured", "q_shape": [10560, 10, 128], "k_shape": [512, 10, 128], "q_len": 10560, "k_len": 512, "reference_error": {"max_abs": 0.000244140625, "rmse": 2.2391697807520104e-06, "relative_l2": 0.0012647145857877136, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 6.37440089465216e-05, "rmse": 5.786214176812004e-06, "relative_l2": 0.0020635479186754463, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.17865599691867828, "p95_ms": 0.1793919950723648, "p99_ms": 0.17996927946805955, "mean_ms": 0.1788228264451027, "n": 300}, {"repeat": 1, "p50_ms": 0.17894400656223297, "p95_ms": 0.17971199750900269, "p99_ms": 0.18041600286960602, "mean_ms": 0.17916266654928525, "n": 300}, {"repeat": 2, "p50_ms": 0.1791040003299713, "p95_ms": 0.17974560484290122, "p99_ms": 0.18025599420070648, "mean_ms": 0.179402879824241, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_d12872a3fe20c4e4_sdpa_efficient.json", "configuration": "704x1280_tp1_sp4", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_retry_704x1280_tp1_sp4_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_retry_704x1280_tp1_sp4_v1/captures/qkv_rank0_d12872a3fe20c4e4.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 9, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_cudnn", "status": "measured", "q_shape": [10560, 10, 128], "k_shape": [512, 10, 128], "q_len": 10560, "k_len": 512, "reference_error": {"max_abs": 0.000244140625, "rmse": 2.2386998347344338e-06, "relative_l2": 0.0012644491536672594, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 6.37440089465216e-05, "rmse": 5.786232531806798e-06, "relative_l2": 0.002063554464650093, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.0597120001912117, "p95_ms": 0.060575999319553375, "p99_ms": 0.061633599922060966, "mean_ms": 0.060031893228491146, "n": 300}, {"repeat": 1, "p50_ms": 0.0597120001912117, "p95_ms": 0.060579199343919754, "p99_ms": 0.06155775919556616, "mean_ms": 0.06021567997833093, "n": 300}, {"repeat": 2, "p50_ms": 0.0597120001912117, "p95_ms": 0.060447998344898224, "p99_ms": 0.0628800019621849, "mean_ms": 0.06016832017650207, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_d12872a3fe20c4e4_sdpa_cudnn.json", "configuration": "704x1280_tp1_sp4", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_retry_704x1280_tp1_sp4_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_retry_704x1280_tp1_sp4_v1/captures/qkv_rank0_d12872a3fe20c4e4.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 9, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "native_dense", "status": "measured", "q_shape": [10560, 10, 128], "k_shape": [512, 10, 128], "q_len": 10560, "k_len": 512, "reference_error": {"max_abs": 0.0, "rmse": 0.0, "relative_l2": 0.0, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 6.37440089465216e-05, "rmse": 5.166876854269409e-06, "relative_l2": 0.0018426725407794024, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.12464000284671783, "p95_ms": 0.1436655975878239, "p99_ms": 0.2223964732885359, "mean_ms": 0.12973600002626579, "n": 300}, {"repeat": 1, "p50_ms": 0.12411199882626534, "p95_ms": 0.13577279821038246, "p99_ms": 0.1556783930957317, "mean_ms": 0.12633279994130134, "n": 300}, {"repeat": 2, "p50_ms": 0.12377600371837616, "p95_ms": 0.1333695948123932, "p99_ms": 0.14684255272150035, "mean_ms": 0.12613823985060055, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_d12872a3fe20c4e4_native_dense.json", "configuration": "704x1280_tp1_sp4", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_retry_704x1280_tp1_sp4_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_retry_704x1280_tp1_sp4_v1/captures/qkv_rank0_db45c3c5819dc1f4.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 9, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "vllm_fa2", "status": "measured", "q_shape": [10560, 10, 128], "k_shape": [66880, 10, 128], "q_len": 10560, "k_len": 42240, "reference_error": {"max_abs": 0.03125, "rmse": 0.0005027876536022947, "relative_l2": 0.0009617304029817336, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.008166105858370631, "rmse": 0.0007497399548139017, "relative_l2": 0.001914449886474462, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 6.337984085083008, "p95_ms": 6.610123109817506, "p99_ms": 6.84154628276825, "mean_ms": 6.409853665033976, "n": 300}, {"repeat": 1, "p50_ms": 6.435248136520386, "p95_ms": 6.508671832084656, "p99_ms": 6.550465431213379, "mean_ms": 6.412469765345255, "n": 300}, {"repeat": 2, "p50_ms": 6.438463926315308, "p95_ms": 6.490870499610901, "p99_ms": 6.540395812988281, "mean_ms": 6.415934828122457, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_db45c3c5819dc1f4_vllm_fa2.json", "configuration": "704x1280_tp1_sp4", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_retry_704x1280_tp1_sp4_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_retry_704x1280_tp1_sp4_v1/captures/qkv_rank0_db45c3c5819dc1f4.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 9, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "vllm_fa3", "status": "measured", "q_shape": [10560, 10, 128], "k_shape": [66880, 10, 128], "q_len": 10560, "k_len": 42240, "reference_error": {"max_abs": 0.0, "rmse": 0.0, "relative_l2": 0.0, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.008166105858370631, "rmse": 0.0007599393361606038, "relative_l2": 0.0019404938559013684, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 3.5131839513778687, "p95_ms": 3.7344511151313786, "p99_ms": 3.7719487929344178, "mean_ms": 3.533585387865702, "n": 300}, {"repeat": 1, "p50_ms": 3.5406079292297363, "p95_ms": 3.616007936000824, "p99_ms": 3.6184557247161866, "mean_ms": 3.5255394172668457, "n": 300}, {"repeat": 2, "p50_ms": 3.541248083114624, "p95_ms": 3.586820864677429, "p99_ms": 3.5885122632980346, "mean_ms": 3.535105810960134, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_db45c3c5819dc1f4_vllm_fa3.json", "configuration": "704x1280_tp1_sp4", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_retry_704x1280_tp1_sp4_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_retry_704x1280_tp1_sp4_v1/captures/qkv_rank0_db45c3c5819dc1f4.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 9, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_math", "status": "measured", "q_shape": [10560, 10, 128], "k_shape": [66880, 10, 128], "q_len": 10560, "k_len": 42240, "reference_error": {"max_abs": 0.03125, "rmse": 0.0009083413062591364, "relative_l2": 0.0017374719610846999, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.007458894141629369, "rmse": 0.0006564985344018477, "relative_l2": 0.0016763592984293276, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 97.20513916015625, "p95_ms": 97.70345726013184, "p99_ms": 97.87994224548339, "mean_ms": 97.16420806884766, "n": 300}, {"repeat": 1, "p50_ms": 97.51614379882812, "p95_ms": 97.93409461975098, "p99_ms": 98.12938018798827, "mean_ms": 97.49314768473307, "n": 300}, {"repeat": 2, "p50_ms": 97.5572624206543, "p95_ms": 98.01304397583007, "p99_ms": 98.23575035095215, "mean_ms": 97.53576957702637, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_db45c3c5819dc1f4_sdpa_math.json", "configuration": "704x1280_tp1_sp4", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_retry_704x1280_tp1_sp4_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_retry_704x1280_tp1_sp4_v1/captures/qkv_rank0_db45c3c5819dc1f4.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 9, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_flash", "status": "measured", "q_shape": [10560, 10, 128], "k_shape": [66880, 10, 128], "q_len": 10560, "k_len": 42240, "reference_error": {"max_abs": 0.03125, "rmse": 0.0005032223291816684, "relative_l2": 0.0009625618488557965, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.008166105858370631, "rmse": 0.000749741695940512, "relative_l2": 0.0019144543324155112, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 6.932944059371948, "p95_ms": 7.170681381225586, "p99_ms": 7.249136185646057, "mean_ms": 6.951531093915303, "n": 300}, {"repeat": 1, "p50_ms": 6.885312080383301, "p95_ms": 7.112633490562439, "p99_ms": 7.2327747774124145, "mean_ms": 6.93240566889445, "n": 300}, {"repeat": 2, "p50_ms": 6.933808088302612, "p95_ms": 7.055513715744018, "p99_ms": 7.108962068557739, "mean_ms": 6.94576180934906, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_db45c3c5819dc1f4_sdpa_flash.json", "configuration": "704x1280_tp1_sp4", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_retry_704x1280_tp1_sp4_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_retry_704x1280_tp1_sp4_v1/captures/qkv_rank0_db45c3c5819dc1f4.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 9, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_efficient", "status": "measured", "q_shape": [10560, 10, 128], "k_shape": [66880, 10, 128], "q_len": 10560, "k_len": 42240, "reference_error": {"max_abs": 0.03125, "rmse": 0.0007803494134979441, "relative_l2": 0.0014926495321294663, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.007458894141629369, "rmse": 0.0007553631967359011, "relative_l2": 0.0019288087515583728, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 13.23043155670166, "p95_ms": 13.260950565338135, "p99_ms": 13.268474760055541, "mean_ms": 13.236069984436035, "n": 300}, {"repeat": 1, "p50_ms": 13.230672359466553, "p95_ms": 13.2620849609375, "p99_ms": 13.27021987915039, "mean_ms": 13.237118953069052, "n": 300}, {"repeat": 2, "p50_ms": 13.23036813735962, "p95_ms": 13.258972358703613, "p99_ms": 13.265813655853272, "mean_ms": 13.235873276392619, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_db45c3c5819dc1f4_sdpa_efficient.json", "configuration": "704x1280_tp1_sp4", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_retry_704x1280_tp1_sp4_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_retry_704x1280_tp1_sp4_v1/captures/qkv_rank0_db45c3c5819dc1f4.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 9, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "sdpa_cudnn", "status": "measured", "q_shape": [10560, 10, 128], "k_shape": [66880, 10, 128], "q_len": 10560, "k_len": 42240, "reference_error": {"max_abs": 0.03125, "rmse": 0.0007793452495270251, "relative_l2": 0.001490728770922439, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.007458894141629369, "rmse": 0.0007556984185515758, "relative_l2": 0.001929664735506996, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 3.8521119356155396, "p95_ms": 4.158561420440674, "p99_ms": 4.163817791938782, "mean_ms": 3.8932477855682373, "n": 300}, {"repeat": 1, "p50_ms": 3.8810240030288696, "p95_ms": 3.997942304611206, "p99_ms": 4.036962041854858, "mean_ms": 3.884686726729075, "n": 300}, {"repeat": 2, "p50_ms": 3.884768009185791, "p95_ms": 3.9647088289260863, "p99_ms": 3.971238977909088, "mean_ms": 3.8888174931208295, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_db45c3c5819dc1f4_sdpa_cudnn.json", "configuration": "704x1280_tp1_sp4", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_retry_704x1280_tp1_sp4_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_retry_704x1280_tp1_sp4_v1/captures/qkv_rank0_db45c3c5819dc1f4.pt", "context": {"phase": "denoise", "step": 0, "start_frame": 9, "role": "self", "layer": "transformer.blocks.0.self_attn"}, "backend": "native_dense", "status": "measured", "q_shape": [10560, 10, 128], "k_shape": [66880, 10, 128], "q_len": 10560, "k_len": 42240, "reference_error": {"max_abs": 0.03125, "rmse": 0.00047990733938625967, "relative_l2": 0.0009179650208095976, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 0.008166105858370631, "rmse": 0.0007626402458939546, "relative_l2": 0.001947390588960866, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 3.4941920042037964, "p95_ms": 3.7117342948913574, "p99_ms": 3.8101127433776854, "mean_ms": 3.5063396374384563, "n": 300}, {"repeat": 1, "p50_ms": 3.501039981842041, "p95_ms": 3.608131265640259, "p99_ms": 3.6246741771698, "mean_ms": 3.515599893728892, "n": 300}, {"repeat": 2, "p50_ms": 3.475279927253723, "p95_ms": 3.857465541362763, "p99_ms": 3.9556971740722657, "mean_ms": 3.4987991491953534, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_db45c3c5819dc1f4_native_dense.json", "configuration": "704x1280_tp1_sp4", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_retry_704x1280_tp1_sp4_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_retry_704x1280_tp1_sp4_v1/captures/qkv_rank0_fcd1494e1f8885be.pt", "context": {"phase": "commit", "step": null, "start_frame": 27, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "vllm_fa2", "status": "measured", "q_shape": [10560, 10, 128], "k_shape": [512, 10, 128], "q_len": 10560, "k_len": 512, "reference_error": {"max_abs": 0.0001220703125, "rmse": 1.1589297699408591e-06, "relative_l2": 0.0008614213152255522, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 6.248058869558204e-05, "rmse": 3.99686897246697e-06, "relative_l2": 0.0017776198736656203, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.09696000069379807, "p95_ms": 0.09779199957847595, "p99_ms": 0.09801824130117893, "mean_ms": 0.09709589372078578, "n": 300}, {"repeat": 1, "p50_ms": 0.09699200093746185, "p95_ms": 0.09795360080897808, "p99_ms": 0.0985299177467823, "mean_ms": 0.09734912013014158, "n": 300}, {"repeat": 2, "p50_ms": 0.09699200093746185, "p95_ms": 0.09772959910333157, "p99_ms": 0.09792032055556774, "mean_ms": 0.09726218687991302, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_fcd1494e1f8885be_vllm_fa2.json", "configuration": "704x1280_tp1_sp4", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_retry_704x1280_tp1_sp4_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_retry_704x1280_tp1_sp4_v1/captures/qkv_rank0_fcd1494e1f8885be.pt", "context": {"phase": "commit", "step": null, "start_frame": 27, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "vllm_fa3", "status": "measured", "q_shape": [10560, 10, 128], "k_shape": [512, 10, 128], "q_len": 10560, "k_len": 512, "reference_error": {"max_abs": 6.103515625e-05, "rmse": 4.3836694929191606e-07, "relative_l2": 0.0003258339235083463, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 6.248058869558204e-05, "rmse": 4.060999388811001e-06, "relative_l2": 0.001806142075265148, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.0806720033288002, "p95_ms": 0.08166560344398022, "p99_ms": 0.08230656087398529, "mean_ms": 0.08090720017751057, "n": 300}, {"repeat": 1, "p50_ms": 0.08051200211048126, "p95_ms": 0.08166560344398022, "p99_ms": 0.08240704156458377, "mean_ms": 0.0810640000551939, "n": 300}, {"repeat": 2, "p50_ms": 0.08038400113582611, "p95_ms": 0.08131200075149536, "p99_ms": 0.08220992013812065, "mean_ms": 0.08087135987977187, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_fcd1494e1f8885be_vllm_fa3.json", "configuration": "704x1280_tp1_sp4", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_retry_704x1280_tp1_sp4_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_retry_704x1280_tp1_sp4_v1/captures/qkv_rank0_fcd1494e1f8885be.pt", "context": {"phase": "commit", "step": null, "start_frame": 27, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_math", "status": "measured", "q_shape": [10560, 10, 128], "k_shape": [512, 10, 128], "q_len": 10560, "k_len": 512, "reference_error": {"max_abs": 0.0001220703125, "rmse": 2.115691239720501e-06, "relative_l2": 0.0015725728837082305, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 5.958972380441796e-05, "rmse": 3.7838496823589655e-06, "relative_l2": 0.0016828788836109959, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 1.2646080255508423, "p95_ms": 1.2676495850086211, "p99_ms": 1.269633902311325, "mean_ms": 1.2647731188933056, "n": 300}, {"repeat": 1, "p50_ms": 1.2646880149841309, "p95_ms": 1.2680319547653198, "p99_ms": 1.270081570148468, "mean_ms": 1.2654572788874308, "n": 300}, {"repeat": 2, "p50_ms": 1.2645599842071533, "p95_ms": 1.2680016040802002, "p99_ms": 1.2699590981006623, "mean_ms": 1.2650705071290334, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_fcd1494e1f8885be_sdpa_math.json", "configuration": "704x1280_tp1_sp4", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_retry_704x1280_tp1_sp4_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_retry_704x1280_tp1_sp4_v1/captures/qkv_rank0_fcd1494e1f8885be.pt", "context": {"phase": "commit", "step": null, "start_frame": 27, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_flash", "status": "measured", "q_shape": [10560, 10, 128], "k_shape": [512, 10, 128], "q_len": 10560, "k_len": 512, "reference_error": {"max_abs": 0.0001220703125, "rmse": 1.1589184721386126e-06, "relative_l2": 0.0008614129176781573, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 6.248058869558204e-05, "rmse": 3.99686897246697e-06, "relative_l2": 0.0017776198736656203, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.1027199998497963, "p95_ms": 0.10364799946546555, "p99_ms": 0.1040649626404047, "mean_ms": 0.10292309326430162, "n": 300}, {"repeat": 1, "p50_ms": 0.10265599936246872, "p95_ms": 0.10355199873447418, "p99_ms": 0.10416000336408615, "mean_ms": 0.10297290630638599, "n": 300}, {"repeat": 2, "p50_ms": 0.10262399911880493, "p95_ms": 0.1035199984908104, "p99_ms": 0.1038089606910944, "mean_ms": 0.10291946654518445, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_fcd1494e1f8885be_sdpa_flash.json", "configuration": "704x1280_tp1_sp4", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_retry_704x1280_tp1_sp4_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_retry_704x1280_tp1_sp4_v1/captures/qkv_rank0_fcd1494e1f8885be.pt", "context": {"phase": "commit", "step": null, "start_frame": 27, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_efficient", "status": "measured", "q_shape": [10560, 10, 128], "k_shape": [512, 10, 128], "q_len": 10560, "k_len": 512, "reference_error": {"max_abs": 0.0001220703125, "rmse": 2.019348551903254e-06, "relative_l2": 0.0015009622934857247, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 6.248058869558204e-05, "rmse": 4.753777167533007e-06, "relative_l2": 0.0021142571413264835, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.17881600558757782, "p95_ms": 0.1796800047159195, "p99_ms": 0.18067296475172043, "mean_ms": 0.17901173373063406, "n": 300}, {"repeat": 1, "p50_ms": 0.1791680008172989, "p95_ms": 0.17980800569057465, "p99_ms": 0.18041696280241012, "mean_ms": 0.17939157297213873, "n": 300}, {"repeat": 2, "p50_ms": 0.17878399789333344, "p95_ms": 0.17955200374126434, "p99_ms": 0.18012959331274034, "mean_ms": 0.17907221352060637, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_fcd1494e1f8885be_sdpa_efficient.json", "configuration": "704x1280_tp1_sp4", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_retry_704x1280_tp1_sp4_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_retry_704x1280_tp1_sp4_v1/captures/qkv_rank0_fcd1494e1f8885be.pt", "context": {"phase": "commit", "step": null, "start_frame": 27, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "sdpa_cudnn", "status": "measured", "q_shape": [10560, 10, 128], "k_shape": [512, 10, 128], "q_len": 10560, "k_len": 512, "reference_error": {"max_abs": 0.0001220703125, "rmse": 2.0189844471173747e-06, "relative_l2": 0.0015006916579117086, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 6.248058869558204e-05, "rmse": 4.753777266973011e-06, "relative_l2": 0.0021142571855527334, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.059039998799562454, "p95_ms": 0.05981120094656944, "p99_ms": 0.06012863963842392, "mean_ms": 0.059284373223781585, "n": 300}, {"repeat": 1, "p50_ms": 0.059167999774217606, "p95_ms": 0.05993599817156792, "p99_ms": 0.060385921560227866, "mean_ms": 0.05965642669548591, "n": 300}, {"repeat": 2, "p50_ms": 0.05920000001788139, "p95_ms": 0.05987360142171383, "p99_ms": 0.06051231883466244, "mean_ms": 0.059661973416805264, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_fcd1494e1f8885be_sdpa_cudnn.json", "configuration": "704x1280_tp1_sp4", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_retry_704x1280_tp1_sp4_v1", "audit_status": "verified"}, {"input_file": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/capture_retry_704x1280_tp1_sp4_v1/captures/qkv_rank0_fcd1494e1f8885be.pt", "context": {"phase": "commit", "step": null, "start_frame": 27, "role": "cross", "layer": "transformer.blocks.0.cross_attn"}, "backend": "native_dense", "status": "measured", "q_shape": [10560, 10, 128], "k_shape": [512, 10, 128], "q_len": 10560, "k_len": 512, "reference_error": {"max_abs": 0.0, "rmse": 0.0, "relative_l2": 0.0, "finite": true, "allclose": true}, "oracle_error": {"max_abs": 6.248058869558204e-05, "rmse": 4.076969633579108e-06, "relative_l2": 0.0018132448911649572, "finite": true, "allclose": true}, "repeats": [{"repeat": 0, "p50_ms": 0.12358400225639343, "p95_ms": 0.1298000045120716, "p99_ms": 0.13720735266804676, "mean_ms": 0.12469919987022876, "n": 300}, {"repeat": 1, "p50_ms": 0.1236800029873848, "p95_ms": 0.1288656048476696, "p99_ms": 0.13860735669732083, "mean_ms": 0.1248234668125709, "n": 300}, {"repeat": 2, "p50_ms": 0.1234240010380745, "p95_ms": 0.13134880140423774, "p99_ms": 0.15885536730289426, "mean_ms": 0.1277186135450999, "n": 300}], "error": null, "dispatch_trace": "qkv_rank0_fcd1494e1f8885be_native_dense.json", "configuration": "704x1280_tp1_sp4", "replay_directory": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/attention_replay_retry_704x1280_tp1_sp4_v1", "audit_status": "verified"}], "attention_status": "已保存后端逐点实测；仅代表相同输入的独立算子，不代表端到端收益。", "campaign": {"status": "completed_with_failures", "jobs": [{"label": "evening-validate-480x832_tp1_sp4", "status": "completed", "started": 1788945936.0198112, "ended": 1788945946.0022347}, {"label": "evening-replay-480x832_tp1_sp4", "status": "completed", "started": 1788945946.0047214, "ended": 1788946106.8209887}, {"label": "evening-audit-480x832_tp1_sp4", "status": "completed", "started": 1788946106.8236759, "ended": 1788946114.8181038}, {"label": "evening-capture-704x1280_tp1_sp4", "status": "not_run_queue_timeout", "started": 1788946114.8202505, "ended": 1788946716.2599561}, {"label": "evening-capture-480x832_tp1_sp2", "status": "not_run_queue_timeout", "started": 1788946716.262718, "ended": 1788947318.7554166}, {"label": "evening-capture-704x1280_tp1_sp2", "status": "completed", "started": 1788947318.7623315, "ended": 1788948137.4634793}, {"label": "evening-validate-704x1280_tp1_sp2", "status": "completed", "started": 1788948137.4702916, "ended": 1788948161.7235909}, {"label": "evening-replay-704x1280_tp1_sp2", "status": "completed", "started": 1788948161.72998, "ended": 1788949564.4334788}, {"label": "evening-audit-704x1280_tp1_sp2", "status": "completed", "started": 1788949564.4408958, "ended": 1788949570.689307}, {"label": "evening-capture-480x832_tp2_sp2", "status": "completed", "started": 1788949570.6962814, "ended": 1788949832.9189928}, {"label": "evening-validate-480x832_tp2_sp2", "status": "completed", "started": 1788949832.926899, "ended": 1788949846.714658}, {"label": "evening-replay-480x832_tp2_sp2", "status": "completed", "started": 1788949846.7218082, "ended": 1788950049.0325992}, {"label": "evening-audit-480x832_tp2_sp2", "status": "completed", "started": 1788950049.0396278, "ended": 1788950050.759793}, {"label": "evening-capture-704x1280_tp2_sp2", "status": "completed", "started": 1788950050.7665331, "ended": 1788950334.47379}, {"label": "evening-validate-704x1280_tp2_sp2", "status": "completed", "started": 1788950334.4807324, "ended": 1788950362.1071312}, {"label": "evening-replay-704x1280_tp2_sp2", "status": "completed", "started": 1788950362.1138687, "ended": 1788951125.1744592}, {"label": "evening-audit-704x1280_tp2_sp2", "status": "completed", "started": 1788951125.181335, "ended": 1788951128.596791}, {"label": "evening-capture-480x832_tp2_sp1", "status": "completed", "started": 1788951128.6037672, "ended": 1788951370.115043}, {"label": "evening-validate-480x832_tp2_sp1", "status": "completed", "started": 1788951370.122375, "ended": 1788951382.6390166}, {"label": "evening-replay-480x832_tp2_sp1", "status": "completed", "started": 1788951382.6459148, "ended": 1788951706.526112}, {"label": "evening-audit-480x832_tp2_sp1", "status": "completed", "started": 1788951706.5335088, "ended": 1788951709.5980165}, {"label": "evening-capture-704x1280_tp2_sp1", "status": "completed", "started": 1788951709.6049678, "ended": 1788952008.1143358}, {"label": "evening-validate-704x1280_tp2_sp1", "status": "completed", "started": 1788952008.121107, "ended": 1788952030.8083289}, {"label": "evening-replay-704x1280_tp2_sp1", "status": "completed", "started": 1788952030.8158998, "ended": 1788953424.907544}, {"label": "evening-audit-704x1280_tp2_sp1", "status": "completed", "started": 1788953424.9147801, "ended": 1788953431.2460754}, {"label": "evening-capture-480x832_tp4_sp1", "status": "completed", "started": 1788953431.252491, "ended": 1788953702.274908}, {"label": "evening-validate-480x832_tp4_sp1", "status": "completed", "started": 1788953702.2822938, "ended": 1788953715.5786185}, {"label": "evening-replay-480x832_tp4_sp1", "status": "completed", "started": 1788953715.5858333, "ended": 1788953912.8431268}, {"label": "evening-audit-480x832_tp4_sp1", "status": "completed", "started": 1788953912.8503678, "ended": 1788953914.648871}, {"label": "evening-capture-704x1280_tp4_sp1", "status": "completed", "started": 1788953914.6561165, "ended": 1788954193.3949037}, {"label": "evening-validate-704x1280_tp4_sp1", "status": "completed", "started": 1788954193.4021046, "ended": 1788954219.2590485}, {"label": "evening-replay-704x1280_tp4_sp1", "status": "completed", "started": 1788954219.266854, "ended": 1788954975.8240306}, {"label": "evening-audit-704x1280_tp4_sp1", "status": "completed", "started": 1788954975.831479, "ended": 1788954979.3039432}, {"label": "evening-capture-480x832_tp1_sp1", "status": "completed", "started": 1788954979.311385, "ended": 1788955180.3223088}, {"label": "evening-validate-480x832_tp1_sp1", "status": "completed", "started": 1788955180.330202, "ended": 1788955192.9686265}, {"label": "evening-replay-480x832_tp1_sp1", "status": "completed", "started": 1788955192.9756956, "ended": 1788955784.8771882}, {"label": "evening-audit-480x832_tp1_sp1", "status": "completed", "started": 1788955784.8845253, "ended": 1788955790.5376623}, {"label": "evening-capture-704x1280_tp1_sp1", "status": "failed", "started": 1788955790.5452666, "ended": 1788956223.6382039}]}, "tail_progress": [{"run": "tail_original_480x832_r0_v1", "workflow_status": "completed", "latest_logged_epoch": 250, "latest_logged_chunk": 9, "target_measurement_epochs": 250, "scope": "Latest log observation, not final sample audit or proof of process liveness"}, {"run": "tail_original_480x832_r1_v1", "workflow_status": "completed", "latest_logged_epoch": 250, "latest_logged_chunk": 9, "target_measurement_epochs": 250, "scope": "Latest log observation, not final sample audit or proof of process liveness"}, {"run": "tail_original_480x832_r2_v1", "workflow_status": "completed", "latest_logged_epoch": 250, "latest_logged_chunk": 9, "target_measurement_epochs": 250, "scope": "Latest log observation, not final sample audit or proof of process liveness"}, {"run": "tail_original_704x1280_r0_v1", "workflow_status": "completed", "latest_logged_epoch": 250, "latest_logged_chunk": 9, "target_measurement_epochs": 250, "scope": "Latest log observation, not final sample audit or proof of process liveness"}, {"run": "tail_original_704x1280_r1_v1", "workflow_status": "completed", "latest_logged_epoch": 250, "latest_logged_chunk": 9, "target_measurement_epochs": 250, "scope": "Latest log observation, not final sample audit or proof of process liveness"}, {"run": "tail_original_704x1280_r2_v1", "workflow_status": "completed", "latest_logged_epoch": 250, "latest_logged_chunk": 9, "target_measurement_epochs": 250, "scope": "Latest log observation, not final sample audit or proof of process liveness"}], "recovery_workflows": [{"label": "Attention缺项补测", "status": "completed_pending_report_refresh", "phase_cn": "completed_pending_report_refresh", "coordinator_live": false, "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/matrix_retry_v1.json"}, {"label": "高分辨率纯DiT补采", "status": "failed", "phase_cn": "failed", "coordinator_live": false, "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/dit_recovery_workflow_v1.json"}, {"label": "矩阵乘/卷积/转换采集", "status": "completed_pending_gpu_replay_and_counters", "phase_cn": "completed_pending_gpu_replay_and_counters", "coordinator_live": false, "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/operator_recovery_workflow_v2.json"}, {"label": "真实算子回放", "status": "completed_pending_dispatch_audit", "phase_cn": "completed_pending_dispatch_audit", "coordinator_live": false, "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/operator_replay_workflow_v2.json"}, {"label": "时间线完整性补采", "status": "completed_pending_coverage_audit", "phase_cn": "completed_pending_coverage_audit", "coordinator_live": false, "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/flush_trace_workflow_v1.json"}, {"label": "本机计算校准", "status": "completed_pending_audit", "phase_cn": "completed_pending_audit", "coordinator_live": false, "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/calibration_workflow_v1.json"}, {"label": "高分辨率输入额度修复重采", "status": "completed_pending_report_update", "phase_cn": "completed_pending_report_update", "coordinator_live": false, "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/dit_recovery_workflow_v2.json"}, {"label": "未插桩基线扩样", "status": "completed_pending_timing_audit", "phase_cn": "completed_pending_timing_audit", "coordinator_live": false, "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/tail_baseline_workflow_v1.json"}, {"label": "各worker显式写出补采", "status": "completed_pending_coverage_audit", "phase_cn": "completed_pending_coverage_audit", "coordinator_live": false, "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/worker_flush_trace_workflow_v2.json"}], "ncu_pilot": [{"id": "0", "kernel": "void prepare_varlen_num_blocks_kernel<1, 0>(int, int, int, const int *, const int *, const int *, const int *, const int *, const int *, int, int, int, int, int, FastDivmod, FastDivmod, int *, int *, int *, int *, int *, bool, bool, bool, int)", "metrics": {"dram__bytes.avg.per_second": {"value": "0.048387", "unit": "Gbyte/s"}, "dram__bytes.max.per_second": {"value": "0.258065", "unit": "Gbyte/s"}, "dram__bytes.min.per_second": {"value": "0.000000", "unit": "Gbyte/s"}, "dram__bytes.sum.per_second": {"value": "2.322581", "unit": "Gbyte/s"}, "gpu__compute_memory_access_throughput.avg.pct_of_peak_sustained_elapsed": {"value": "0.366092", "unit": "%"}, "gpu__compute_memory_request_throughput.avg.pct_of_peak_sustained_elapsed": {"value": "0.280683", "unit": "%"}, "gpu__compute_memory_throughput.avg.pct_of_peak_sustained_elapsed": {"value": "0.366092", "unit": "%"}, "gpu__dram_throughput.avg.pct_of_peak_sustained_elapsed": {"value": "0.049156", "unit": "%"}, "gpu__time_duration.avg": {"value": "2.976000", "unit": "us"}, "gpu__time_duration.max": {"value": "2.976000", "unit": "us"}, "gpu__time_duration.min": {"value": "2.976000", "unit": "us"}, "gpu__time_duration.sum": {"value": "2.976000", "unit": "us"}, "l1tex__throughput.avg.pct_of_peak_sustained_active": {"value": "35.270541", "unit": "%"}, "launch__registers_per_thread": {"value": "16", "unit": "register/thread"}, "launch__registers_per_thread_allocated": {"value": "16.000000", "unit": "register/thread"}, "launch__shared_mem_per_block": {"value": "1.028000", "unit": "Kbyte/block"}, "launch__shared_mem_per_block_allocated": {"value": "1.152000", "unit": "Kbyte/block"}, "launch__shared_mem_per_block_driver": {"value": "1.024000", "unit": "Kbyte/block"}, "launch__shared_mem_per_block_dynamic": {"value": "0.000000", "unit": "Kbyte/block"}, "launch__shared_mem_per_block_static": {"value": "4", "unit": "byte/block"}, "lts__throughput.avg.pct_of_peak_sustained_elapsed": {"value": "0.366092", "unit": "%"}, "sm__memory_throughput.avg.pct_of_peak_sustained_elapsed": {"value": "0.005489", "unit": "%"}, "sm__throughput.avg.pct_of_peak_sustained_elapsed": {"value": "0.005489", "unit": "%"}, "sm__warps_active.avg.pct_of_peak_sustained_active": {"value": "1.588594", "unit": "%"}}}, {"id": "1", "kernel": "void device_kernel<enable_sm90_or_later<FlashAttnFwdSm90<CollectiveMainloopFwdSm90<2, cute::tuple<cute::C<1>, cute::C<1>, cute::C<1>>, cute::tuple<cute::C<128>, cute::C<160>, cute::C<128>>, 128, bfloat16_t, float, arch::Sm90, 0, 0, 0, 1, 0, 0, 0, 1, 1, 1, 0, 0, bfloat16_t, 1>, CollectiveEpilogueFwd<cute::tuple<cute::C<128>, cute::C<128>, cute::C<160>>, cute::tuple<cute::C<1>, cute::C<1>, cute::C<1>>, bfloat16_t, arch::Sm90, 256, 1, 1, 0, 0, 1>, VarlenDynamicPersistentTileScheduler<128, 160, 256, 128, 0, 1, 1, 0, 0, 1>>>>(T1::Params)", "metrics": {"dram__bytes.avg.per_second": {"value": "4.451478", "unit": "Gbyte/s"}, "dram__bytes.max.per_second": {"value": "4.572974", "unit": "Gbyte/s"}, "dram__bytes.min.per_second": {"value": "2.275255", "unit": "Gbyte/s"}, "dram__bytes.sum.per_second": {"value": "213.670942", "unit": "Gbyte/s"}, "gpu__compute_memory_access_throughput.avg.pct_of_peak_sustained_elapsed": {"value": "49.119066", "unit": "%"}, "gpu__compute_memory_request_throughput.avg.pct_of_peak_sustained_elapsed": {"value": "48.788437", "unit": "%"}, "gpu__compute_memory_throughput.avg.pct_of_peak_sustained_elapsed": {"value": "49.119066", "unit": "%"}, "gpu__dram_throughput.avg.pct_of_peak_sustained_elapsed": {"value": "4.440563", "unit": "%"}, "gpu__time_duration.avg": {"value": "923.072000", "unit": "us"}, "gpu__time_duration.max": {"value": "923.072000", "unit": "us"}, "gpu__time_duration.min": {"value": "923.072000", "unit": "us"}, "gpu__time_duration.sum": {"value": "923.072000", "unit": "us"}, "l1tex__throughput.avg.pct_of_peak_sustained_active": {"value": "53.185107", "unit": "%"}, "launch__registers_per_thread": {"value": "168", "unit": "register/thread"}, "launch__registers_per_thread_allocated": {"value": "168.000000", "unit": "register/thread"}, "launch__shared_mem_per_block": {"value": "200.704000", "unit": "Kbyte/block"}, "launch__shared_mem_per_block_allocated": {"value": "200.704000", "unit": "Kbyte/block"}, "launch__shared_mem_per_block_driver": {"value": "1.024000", "unit": "Kbyte/block"}, "launch__shared_mem_per_block_dynamic": {"value": "199.680000", "unit": "Kbyte/block"}, "launch__shared_mem_per_block_static": {"value": "0", "unit": "byte/block"}, "lts__throughput.avg.pct_of_peak_sustained_elapsed": {"value": "23.476523", "unit": "%"}, "sm__memory_throughput.avg.pct_of_peak_sustained_elapsed": {"value": "42.004568", "unit": "%"}, "sm__throughput.avg.pct_of_peak_sustained_elapsed": {"value": "79.479644", "unit": "%"}, "sm__warps_active.avg.pct_of_peak_sustained_active": {"value": "18.750314", "unit": "%"}}}], "counter_gate": {"status": "awaiting_user_clarification", "block_gpu_counter_jobs": true, "reason": "Root SSH GPU processes killed as foreign under normal-user gpu run, including --user root. User requested normal-user scheduling; requested exception to test root-owned gpu run.", "failed_pids": [2268586, 2280634], "guard_changed": false}, "timeline_warning": "旧原版/rebase记录有尾部缺失。新worker写出补采已通过逐块存在性检查；所有表仍只代表已记录事件，诊断插桩耗时不等于干净性能。", "report_url": "/analysis-report-cn.html", "capability_probe": {"status": "passed", "phase_cn": "单卡权限验证通过：三项硬件计数器有效，实际UID2005，30秒观察完成且无该PID违规记录", "scope": "root启动后实际UID为2005，仅本次进程保留CAP_SYS_ADMIN；不修改guard或驱动。身份预检成功不等于GPU计数器验证成功。", "evidence": "raw/capability_counter_probe_workflow_v1.json; raw/ncu_capability_probe_v3; analysis/capability_counter_probe_audit.json"}, "capability_results": {"operator_inputs": 18, "operator_kernels": 45, "status": "passed", "scope": "真实输入独立回放；不是整模型MFU或优化验收", "attention_pair_status": "passed"}, "same_gpu_compute": {"state": "audited_passed", "verified_live": false, "completed_launches": 3, "scope": "同一物理GPU6计算校准，独立启动；结束后需审核，不能仅凭文件存在宣称完成。"}, "cross_sdpa_experiment": {"status": "comparison_completed", "probe_status": "passed", "verified_live": {"probe": false, "model": false}, "model_status": "completed_pending_output_audit", "numerical_passed": null, "scope": "先实际SDPAImpl Flash-only验证，再同卡eager参考/候选；未通过原定数值门槛不认领加速。"}}
