{ "scope": "Replay of the original ATen operator on saved real input. Evidence checks do not define a new numerical tolerance or accept a replacement implementation. Profiling durations are diagnostic, not clean timing; no NCU counters.", "status": "evidence_checks_passed", "families": [ "dit", "output", "vae" ], "rows": [ { "input": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/operator_capture_480x832_v2/operators/dit_rank0_cd8cbc742f5c3f37.pt", "sha256": "824a77bcaead8af0dacf908827027966a81c2d964d919eb4200bac0771b4a369", "result": { "op": "aten.linear.default", "stage": "dit", "shape": [ 4680, 5120 ], "exact": true, "max_abs": 0.0, "rmse": 0.0, "finite": true, "math_policy": { "matmul_precision": "highest", "matmul_allow_tf32": false, "cudnn_allow_tf32": true, "cudnn_benchmark": false, "cudnn_deterministic": false }, "scope": "Original ATen op on saved real tensors; isolated diagnostic" }, "trace": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/operator_replay_480x832_v2/000_dit.trace.json", "trace_sha256": "fe3973c512ac66a158aafb4266b812669c6f053e30fb0ca53526ae0f2e164fd3", "kernels": [ { "name": "nvjet_sm90_tst_256x144_64x4_2x1_v_bz_coopA_bias_TNT", "duration_us": 24.224, "pid": 0, "tid": 7, "args": { "External id": 5, "queued": 0, "device": 0, "context": 1, "stream": 7, "correlation": 17, "registers per thread": 168, "shared memory": 221380, "blocks per SM": 1.0, "warps per SM": 12.0, "grid": [ 2, 66, 1 ], "block": [ 384, 1, 1 ], "est. achieved occupancy %": 0, "occupancy": { "activeBlocksPerMultiprocessor": 0, "limitingFactors": "SMEM", "blockLimitRegs": 1, "blockLimitSharedMem": 0, "blockLimitWarps": 5, "blockLimitBlocks": 32, "blockLimitBarriers": 2147483647, "allocatedRegistersPerBlock": 64512, "allocatedSharedMemPerBlock": 222464 }, "graph id": 0, "graph node id": 0, "channel": 0, "channel_type": 1 } } ], "checks": { "exit_zero": true, "input_hash_matches": true, "finite_output": true, "actual_gpu_kernel_recorded": true, "operator_recorded": true }, "evidence_checks_passed": true }, { "input": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/operator_capture_480x832_v2/operators/dit_rank0_9296b219534b05e0.pt", "sha256": "787b43c2c5ab957d3bd3c45047d646e6e407e506ea71b3249408d5d20bf00b0c", "result": { "op": "aten.linear.default", "stage": "dit", "shape": [ 1, 4680, 5120 ], "exact": true, "max_abs": 0.0, "rmse": 0.0, "finite": true, "math_policy": { "matmul_precision": "highest", "matmul_allow_tf32": false, "cudnn_allow_tf32": true, "cudnn_benchmark": false, "cudnn_deterministic": false }, "scope": "Original ATen op on saved real tensors; isolated diagnostic" }, "trace": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/operator_replay_480x832_v2/001_dit.trace.json", "trace_sha256": "b64e20d27b003ea1bf1c386df1163113bb4e5d5f18d22898e4da0b245b16d1e3", "kernels": [ { "name": "nvjet_sm90_tst_256x144_64x4_2x1_v_bz_coopA_bias_TNT", "duration_us": 93.889, "pid": 0, "tid": 7, "args": { "External id": 6, "queued": 0, "device": 0, "context": 1, "stream": 7, "correlation": 17, "registers per thread": 168, "shared memory": 221380, "blocks per SM": 1.0, "warps per SM": 12.0, "grid": [ 2, 66, 1 ], "block": [ 384, 1, 1 ], "est. achieved occupancy %": 0, "occupancy": { "activeBlocksPerMultiprocessor": 0, "limitingFactors": "SMEM", "blockLimitRegs": 1, "blockLimitSharedMem": 0, "blockLimitWarps": 5, "blockLimitBlocks": 32, "blockLimitBarriers": 2147483647, "allocatedRegistersPerBlock": 64512, "allocatedSharedMemPerBlock": 222464 }, "graph id": 0, "graph node id": 0, "channel": 0, "channel_type": 1 } } ], "checks": { "exit_zero": true, "input_hash_matches": true, "finite_output": true, "actual_gpu_kernel_recorded": true, "operator_recorded": true }, "evidence_checks_passed": true }, { "input": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/operator_capture_480x832_v2/operators/dit_rank0_f56603026672560c.pt", "sha256": "4a5b492ee606c1a2c9fe4844a0c3125609fa599f4c027f56b4329c89d2879b7e", "result": { "op": "aten.linear.default", "stage": "dit", "shape": [ 1, 4680, 5120 ], "exact": true, "max_abs": 0.0, "rmse": 0.0, "finite": true, "math_policy": { "matmul_precision": "highest", "matmul_allow_tf32": false, "cudnn_allow_tf32": true, "cudnn_benchmark": false, "cudnn_deterministic": false }, "scope": "Original ATen op on saved real tensors; isolated diagnostic" }, "trace": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/operator_replay_480x832_v2/002_dit.trace.json", "trace_sha256": "33c1e7526180abeac91cc75ab706402cf93a70d9aba36250b5ceb127110e3925", "kernels": [ { "name": "nvjet_sm90_tst_256x144_64x4_2x1_v_bz_coopA_bias_TNT", "duration_us": 290.945, "pid": 0, "tid": 7, "args": { "External id": 6, "queued": 0, "device": 0, "context": 1, "stream": 7, "correlation": 17, "registers per thread": 168, "shared memory": 221380, "blocks per SM": 1.0, "warps per SM": 12.0, "grid": [ 2, 66, 1 ], "block": [ 384, 1, 1 ], "est. achieved occupancy %": 0, "occupancy": { "activeBlocksPerMultiprocessor": 0, "limitingFactors": "SMEM", "blockLimitRegs": 1, "blockLimitSharedMem": 0, "blockLimitWarps": 5, "blockLimitBlocks": 32, "blockLimitBarriers": 2147483647, "allocatedRegistersPerBlock": 64512, "allocatedSharedMemPerBlock": 222464 }, "graph id": 0, "graph node id": 0, "channel": 0, "channel_type": 1 } } ], "checks": { "exit_zero": true, "input_hash_matches": true, "finite_output": true, "actual_gpu_kernel_recorded": true, "operator_recorded": true }, "evidence_checks_passed": true }, { "input": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/operator_capture_480x832_v2/operators/dit_rank0_10a11d3772070763.pt", "sha256": "ea1184eaad12555ac80bf5ee33de7aa2251926b289572cadadb4043884848587", "result": { "op": "aten.linear.default", "stage": "dit", "shape": [ 3, 5120 ], "exact": true, "max_abs": 0.0, "rmse": 0.0, "finite": true, "math_policy": { "matmul_precision": "highest", "matmul_allow_tf32": false, "cudnn_allow_tf32": true, "cudnn_benchmark": false, "cudnn_deterministic": false }, "scope": "Original ATen op on saved real tensors; isolated diagnostic" }, "trace": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/operator_replay_480x832_v2/003_dit.trace.json", "trace_sha256": "55367134f6d06de731e86b33c8495bee5200257d2ad15548a9867cca92d6cf5e", "kernels": [ { "name": "nvjet_sm90_tst_64x8_64x16_4x1_v_bz_bias_TNT", "duration_us": 2.88, "pid": 0, "tid": 7, "args": { "External id": 5, "queued": 0, "device": 0, "context": 1, "stream": 7, "correlation": 16, "registers per thread": 168, "shared memory": 164308, "blocks per SM": 0.606061, "warps per SM": 7.272727, "grid": [ 4, 20, 1 ], "block": [ 384, 1, 1 ], "est. achieved occupancy %": 0, "occupancy": { "activeBlocksPerMultiprocessor": 0, "limitingFactors": "SMEM", "blockLimitRegs": 1, "blockLimitSharedMem": 0, "blockLimitWarps": 5, "blockLimitBlocks": 32, "blockLimitBarriers": 2147483647, "allocatedRegistersPerBlock": 64512, "allocatedSharedMemPerBlock": 165376 }, "graph id": 0, "graph node id": 0, "channel": 0, "channel_type": 1 } } ], "checks": { "exit_zero": true, "input_hash_matches": true, "finite_output": true, "actual_gpu_kernel_recorded": true, "operator_recorded": true }, "evidence_checks_passed": true }, { "input": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/operator_capture_480x832_v2/operators/dit_rank0_ca590b3a668a9ecf.pt", "sha256": "a6cb8d5375b991c9c5d1a1e5674fa86eb61d10648b2ccf71bb7ec40c56f2ac1e", "result": { "op": "aten.linear.default", "stage": "dit", "shape": [ 3, 5120 ], "exact": true, "max_abs": 0.0, "rmse": 0.0, "finite": true, "math_policy": { "matmul_precision": "highest", "matmul_allow_tf32": false, "cudnn_allow_tf32": true, "cudnn_benchmark": false, "cudnn_deterministic": false }, "scope": "Original ATen op on saved real tensors; isolated diagnostic" }, "trace": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/operator_replay_480x832_v2/004_dit.trace.json", "trace_sha256": "1fc44b5f196e7bd835b32c4b648b0085fbe0b7df2512d798771bc33ddbb425fb", "kernels": [ { "name": "nvjet_sm90_tst_64x8_64x16_4x1_v_bz_bias_TNT", "duration_us": 17.023, "pid": 0, "tid": 7, "args": { "External id": 5, "queued": 0, "device": 0, "context": 1, "stream": 7, "correlation": 16, "registers per thread": 168, "shared memory": 164308, "blocks per SM": 0.606061, "warps per SM": 7.272727, "grid": [ 4, 20, 1 ], "block": [ 384, 1, 1 ], "est. achieved occupancy %": 0, "occupancy": { "activeBlocksPerMultiprocessor": 0, "limitingFactors": "SMEM", "blockLimitRegs": 1, "blockLimitSharedMem": 0, "blockLimitWarps": 5, "blockLimitBlocks": 32, "blockLimitBarriers": 2147483647, "allocatedRegistersPerBlock": 64512, "allocatedSharedMemPerBlock": 165376 }, "graph id": 0, "graph node id": 0, "channel": 0, "channel_type": 1 } } ], "checks": { "exit_zero": true, "input_hash_matches": true, "finite_output": true, "actual_gpu_kernel_recorded": true, "operator_recorded": true }, "evidence_checks_passed": true }, { "input": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/operator_capture_480x832_v2/operators/dit_rank0_22805ed4eb786e66.pt", "sha256": "d6a2e382ecf0fdad711df2cd1bcb239a004321ae666d26aa04920718cf49bf81", "result": { "op": "aten.linear.default", "stage": "dit", "shape": [ 1, 3, 30720 ], "exact": true, "max_abs": 0.0, "rmse": 0.0, "finite": true, "math_policy": { "matmul_precision": "highest", "matmul_allow_tf32": false, "cudnn_allow_tf32": true, "cudnn_benchmark": false, "cudnn_deterministic": false }, "scope": "Original ATen op on saved real tensors; isolated diagnostic" }, "trace": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/operator_replay_480x832_v2/005_dit.trace.json", "trace_sha256": "1915c84b146a56a0dbb97bbbfae1eee5a8a834c2b091173de810d5b3b2b37abd", "kernels": [ { "name": "nvjet_sm90_tst_64x8_64x16_2x1_v_bz_bias_TNT", "duration_us": 78.528, "pid": 0, "tid": 7, "args": { "External id": 6, "queued": 0, "device": 0, "context": 1, "stream": 7, "correlation": 17, "registers per thread": 168, "shared memory": 164308, "blocks per SM": 1.0, "warps per SM": 12.0, "grid": [ 2, 66, 1 ], "block": [ 384, 1, 1 ], "est. achieved occupancy %": 0, "occupancy": { "activeBlocksPerMultiprocessor": 0, "limitingFactors": "SMEM", "blockLimitRegs": 1, "blockLimitSharedMem": 0, "blockLimitWarps": 5, "blockLimitBlocks": 32, "blockLimitBarriers": 2147483647, "allocatedRegistersPerBlock": 64512, "allocatedSharedMemPerBlock": 165376 }, "graph id": 0, "graph node id": 0, "channel": 0, "channel_type": 1 } } ], "checks": { "exit_zero": true, "input_hash_matches": true, "finite_output": true, "actual_gpu_kernel_recorded": true, "operator_recorded": true }, "evidence_checks_passed": true }, { "input": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/operator_capture_480x832_v2/operators/dit_rank0_fcebe966800b94b1.pt", "sha256": "4f1e9988bf86b3f91f274f9b8c3bf33bf42b39945181ce6f09fbce8e0132cb6f", "result": { "op": "aten.linear.default", "stage": "dit", "shape": [ 1, 1170, 15360 ], "exact": true, "max_abs": 0.0, "rmse": 0.0, "finite": true, "math_policy": { "matmul_precision": "highest", "matmul_allow_tf32": false, "cudnn_allow_tf32": true, "cudnn_benchmark": false, "cudnn_deterministic": false }, "scope": "Original ATen op on saved real tensors; isolated diagnostic" }, "trace": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/operator_replay_480x832_v2/006_dit.trace.json", "trace_sha256": "b88096a8e6eb0e66e6acb7fd33c1aa33e120e33da8b6cbfe5820273520d35f22", "kernels": [ { "name": "nvjet_sm90_tst_256x152_64x4_1x2_h_bz_coopA_bias_TNT", "duration_us": 251.458, "pid": 0, "tid": 7, "args": { "External id": 6, "queued": 0, "device": 0, "context": 1, "stream": 7, "correlation": 17, "registers per thread": 168, "shared memory": 225476, "blocks per SM": 1.0, "warps per SM": 12.0, "grid": [ 2, 66, 1 ], "block": [ 384, 1, 1 ], "est. achieved occupancy %": 0, "occupancy": { "activeBlocksPerMultiprocessor": 0, "limitingFactors": "SMEM", "blockLimitRegs": 1, "blockLimitSharedMem": 0, "blockLimitWarps": 5, "blockLimitBlocks": 32, "blockLimitBarriers": 2147483647, "allocatedRegistersPerBlock": 64512, "allocatedSharedMemPerBlock": 226560 }, "graph id": 0, "graph node id": 0, "channel": 0, "channel_type": 1 } } ], "checks": { "exit_zero": true, "input_hash_matches": true, "finite_output": true, "actual_gpu_kernel_recorded": true, "operator_recorded": true }, "evidence_checks_passed": true }, { "input": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/operator_capture_480x832_v2/operators/dit_rank0_10a9e2fea034c6cf.pt", "sha256": "a64f7da6eb1fcc9779d40258d974eb529c2397aaba4df6e32363d7c5d4e73065", "result": { "op": "aten.linear.default", "stage": "dit", "shape": [ 1, 1170, 5120 ], "exact": true, "max_abs": 0.0, "rmse": 0.0, "finite": true, "math_policy": { "matmul_precision": "highest", "matmul_allow_tf32": false, "cudnn_allow_tf32": true, "cudnn_benchmark": false, "cudnn_deterministic": false }, "scope": "Original ATen op on saved real tensors; isolated diagnostic" }, "trace": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/operator_replay_480x832_v2/007_dit.trace.json", "trace_sha256": "8f6dfc221248878721be9741abffc9ca69947ebbba496c1d7849986187a7c93f", "kernels": [ { "name": "nvjet_sm90_tst_128x200_64x5_2x1_v_bz_coopA_bias_TNT", "duration_us": 85.281, "pid": 0, "tid": 7, "args": { "External id": 6, "queued": 0, "device": 0, "context": 1, "stream": 7, "correlation": 16, "registers per thread": 168, "shared memory": 226492, "blocks per SM": 1.0, "warps per SM": 12.0, "grid": [ 2, 66, 1 ], "block": [ 384, 1, 1 ], "est. achieved occupancy %": 0, "occupancy": { "activeBlocksPerMultiprocessor": 0, "limitingFactors": "SMEM", "blockLimitRegs": 1, "blockLimitSharedMem": 0, "blockLimitWarps": 5, "blockLimitBlocks": 32, "blockLimitBarriers": 2147483647, "allocatedRegistersPerBlock": 64512, "allocatedSharedMemPerBlock": 227584 }, "graph id": 0, "graph node id": 0, "channel": 0, "channel_type": 1 } } ], "checks": { "exit_zero": true, "input_hash_matches": true, "finite_output": true, "actual_gpu_kernel_recorded": true, "operator_recorded": true }, "evidence_checks_passed": true }, { "input": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/operator_capture_480x832_v2/operators/vae_rank0_6638f430ac58a438.pt", "sha256": "32bcd9b4d8be26c876626649d4ff2e623424efe309222bcd7e80afc47310ded4", "result": { "op": "aten.conv3d.default", "stage": "vae", "shape": [ 1, 16, 3, 60, 104 ], "exact": true, "max_abs": 0.0, "rmse": 0.0, "finite": true, "math_policy": { "matmul_precision": "highest", "matmul_allow_tf32": false, "cudnn_allow_tf32": true, "cudnn_benchmark": false, "cudnn_deterministic": false }, "scope": "Original ATen op on saved real tensors; isolated diagnostic" }, "trace": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/operator_replay_480x832_v2/008_vae.trace.json", "trace_sha256": "ffd9abb9e1212d20346f78e9646cc77515899d5a81f7ca53e3d2b3060f5f14f6", "kernels": [ { "name": "void cutlass::Kernel2(cutlass_80_tensorop_s1688gemm_128x64_16x6_nn_align4::Params)", "duration_us": 3.872, "pid": 0, "tid": 7, "args": { "External id": 4, "queued": 0, "device": 0, "context": 1, "stream": 7, "correlation": 17, "registers per thread": 136, "shared memory": 73728, "blocks per SM": 1.151515, "warps per SM": 4.606061, "grid": [ 8, 19, 1 ], "block": [ 128, 1, 1 ], "est. achieved occupancy %": 0, "occupancy": { "activeBlocksPerMultiprocessor": 0, "limitingFactors": "SMEM", "blockLimitRegs": 3, "blockLimitSharedMem": 0, "blockLimitWarps": 16, "blockLimitBlocks": 32, "blockLimitBarriers": 2147483647, "allocatedRegistersPerBlock": 17408, "allocatedSharedMemPerBlock": 74752 }, "graph id": 0, "graph node id": 0, "channel": 0, "channel_type": 1 } }, { "name": "void at::native::elementwise_kernel<128, 2, at::native::gpu_kernel_impl_nocast >(at::TensorIteratorBase&, at::native::CUDAFunctor_add const&)::{lambda(int)#1}>(int, at::native::gpu_kernel_impl_nocast >(at::TensorIteratorBase&, at::native::CUDAFunctor_add const&)::{lambda(int)#1})", "duration_us": 2.336, "pid": 0, "tid": 7, "args": { "External id": 7, "queued": 0, "device": 0, "context": 1, "stream": 7, "correlation": 25, "registers per thread": 18, "shared memory": 0, "blocks per SM": 8.863636, "warps per SM": 35.454544, "grid": [ 1170, 1, 1 ], "block": [ 128, 1, 1 ], "est. achieved occupancy %": 55, "occupancy": { "activeBlocksPerMultiprocessor": 16, "limitingFactors": "WARPS", "blockLimitRegs": 21, "blockLimitSharedMem": 228, "blockLimitWarps": 16, "blockLimitBlocks": 32, "blockLimitBarriers": 2147483647, "allocatedRegistersPerBlock": 3072, "allocatedSharedMemPerBlock": 1024 }, "graph id": 0, "graph node id": 0, "channel": 0, "channel_type": 1 } } ], "checks": { "exit_zero": true, "input_hash_matches": true, "finite_output": true, "actual_gpu_kernel_recorded": true, "operator_recorded": true }, "evidence_checks_passed": true }, { "input": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/operator_capture_480x832_v2/operators/vae_rank0_29ef146661136bf4.pt", "sha256": "2b9e80cee384305d79045df290612b04bcfe041bd67dc69a2bc096e45a6c5872", "result": { "op": "aten.conv3d.default", "stage": "vae", "shape": [ 1, 384, 1, 60, 26 ], "exact": true, "max_abs": 0.0, "rmse": 0.0, "finite": true, "math_policy": { "matmul_precision": "highest", "matmul_allow_tf32": false, "cudnn_allow_tf32": true, "cudnn_benchmark": false, "cudnn_deterministic": false }, "scope": "Original ATen op on saved real tensors; isolated diagnostic" }, "trace": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/operator_replay_480x832_v2/009_vae.trace.json", "trace_sha256": "2ed2af2526ac2a519a16747285f649426b641456ec56a042fb2087147ca09f18", "kernels": [ { "name": "void cudnn::engines_precompiled::nchwToNhwcKernel(cudnn::engines_precompiled::nchw2nhwc_params_t, float const*, float*)", "duration_us": 1.696, "pid": 0, "tid": 7, "args": { "External id": 4, "queued": 0, "device": 0, "context": 1, "stream": 7, "correlation": 12, "registers per thread": 32, "shared memory": 4224, "blocks per SM": 1.234848, "warps per SM": 9.878788, "grid": [ 163, 1, 1 ], "block": [ 256, 1, 1 ], "est. achieved occupancy %": 15, "occupancy": { "activeBlocksPerMultiprocessor": 8, "limitingFactors": "WARPS|REGS", "blockLimitRegs": 8, "blockLimitSharedMem": 44, "blockLimitWarps": 8, "blockLimitBlocks": 32, "blockLimitBarriers": 2147483647, "allocatedRegistersPerBlock": 8192, "allocatedSharedMemPerBlock": 5248 }, "graph id": 0, "graph node id": 0, "channel": 0, "channel_type": 1 } }, { "name": "void cudnn::engines_precompiled::nchwToNhwcKernel(cudnn::engines_precompiled::nchw2nhwc_params_t, float const*, float*)", "duration_us": 1.824, "pid": 0, "tid": 7, "args": { "External id": 4, "queued": 0, "device": 0, "context": 1, "stream": 7, "correlation": 16, "registers per thread": 32, "shared memory": 4224, "blocks per SM": 2.909091, "warps per SM": 23.272728, "grid": [ 1, 1, 384 ], "block": [ 256, 1, 1 ], "est. achieved occupancy %": 36, "occupancy": { "activeBlocksPerMultiprocessor": 8, "limitingFactors": "WARPS|REGS", "blockLimitRegs": 8, "blockLimitSharedMem": 44, "blockLimitWarps": 8, "blockLimitBlocks": 32, "blockLimitBarriers": 2147483647, "allocatedRegistersPerBlock": 8192, "allocatedSharedMemPerBlock": 5248 }, "graph id": 0, "graph node id": 0, "channel": 0, "channel_type": 1 } }, { "name": "sm80_xmma_fprop_implicit_gemm_tf32f32_tf32f32_f32_nhwckrsc_nchw_tilesize128x64x32_stage5_warpsize2x2x1_g1_tensor16x8x8_execute_kernel__5x_cudnn", "duration_us": 15.457, "pid": 0, "tid": 7, "args": { "External id": 4, "queued": 0, "device": 0, "context": 1, "stream": 7, "correlation": 21, "registers per thread": 168, "shared memory": 122880, "blocks per SM": 0.590909, "warps per SM": 2.363636, "grid": [ 6, 13, 1 ], "block": [ 128, 1, 1 ], "est. achieved occupancy %": 0, "occupancy": { "activeBlocksPerMultiprocessor": 0, "limitingFactors": "SMEM", "blockLimitRegs": 3, "blockLimitSharedMem": 0, "blockLimitWarps": 16, "blockLimitBlocks": 32, "blockLimitBarriers": 2147483647, "allocatedRegistersPerBlock": 21504, "allocatedSharedMemPerBlock": 123904 }, "graph id": 0, "graph node id": 0, "channel": 0, "channel_type": 1 } }, { "name": "void at::native::elementwise_kernel<128, 2, at::native::gpu_kernel_impl_nocast >(at::TensorIteratorBase&, at::native::CUDAFunctor_add const&)::{lambda(int)#1}>(int, at::native::gpu_kernel_impl_nocast >(at::TensorIteratorBase&, at::native::CUDAFunctor_add const&)::{lambda(int)#1})", "duration_us": 3.104, "pid": 0, "tid": 7, "args": { "External id": 7, "queued": 0, "device": 0, "context": 1, "stream": 7, "correlation": 31, "registers per thread": 18, "shared memory": 0, "blocks per SM": 17.727272, "warps per SM": 70.909088, "grid": [ 2340, 1, 1 ], "block": [ 128, 1, 1 ], "est. achieved occupancy %": 100, "occupancy": { "activeBlocksPerMultiprocessor": 16, "limitingFactors": "WARPS", "blockLimitRegs": 21, "blockLimitSharedMem": 228, "blockLimitWarps": 16, "blockLimitBlocks": 32, "blockLimitBarriers": 2147483647, "allocatedRegistersPerBlock": 3072, "allocatedSharedMemPerBlock": 1024 }, "graph id": 0, "graph node id": 0, "channel": 0, "channel_type": 1 } } ], "checks": { "exit_zero": true, "input_hash_matches": true, "finite_output": true, "actual_gpu_kernel_recorded": true, "operator_recorded": true }, "evidence_checks_passed": true }, { "input": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/operator_capture_480x832_v2/operators/vae_rank0_551fa1ff3095f40c.pt", "sha256": "6309c12f24013edabd88edced7c2ce22672322d203206b12ba218c88696effb5", "result": { "op": "aten.conv3d.default", "stage": "vae", "shape": [ 1, 384, 1, 60, 26 ], "exact": true, "max_abs": 0.0, "rmse": 0.0, "finite": true, "math_policy": { "matmul_precision": "highest", "matmul_allow_tf32": false, "cudnn_allow_tf32": true, "cudnn_benchmark": false, "cudnn_deterministic": false }, "scope": "Original ATen op on saved real tensors; isolated diagnostic" }, "trace": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/operator_replay_480x832_v2/010_vae.trace.json", "trace_sha256": "19edee2ec48a654f70c5630ce3b4db42f221d3a4de6c29016fa97ec92b487379", "kernels": [ { "name": "void cudnn::engines_precompiled::nchwToNhwcKernel(cudnn::engines_precompiled::nchw2nhwc_params_t, float const*, float*)", "duration_us": 5.824, "pid": 0, "tid": 7, "args": { "External id": 4, "queued": 0, "device": 0, "context": 1, "stream": 7, "correlation": 12, "registers per thread": 32, "shared memory": 4224, "blocks per SM": 14.818182, "warps per SM": 118.545456, "grid": [ 163, 12, 1 ], "block": [ 256, 1, 1 ], "est. achieved occupancy %": 100, "occupancy": { "activeBlocksPerMultiprocessor": 8, "limitingFactors": "WARPS|REGS", "blockLimitRegs": 8, "blockLimitSharedMem": 44, "blockLimitWarps": 8, "blockLimitBlocks": 32, "blockLimitBarriers": 2147483647, "allocatedRegistersPerBlock": 8192, "allocatedSharedMemPerBlock": 5248 }, "graph id": 0, "graph node id": 0, "channel": 0, "channel_type": 1 } }, { "name": "void cudnn::engines_precompiled::nchwToNhwcKernel(cudnn::engines_precompiled::nchw2nhwc_params_t, float const*, float*)", "duration_us": 11.296, "pid": 0, "tid": 7, "args": { "External id": 4, "queued": 0, "device": 0, "context": 1, "stream": 7, "correlation": 16, "registers per thread": 32, "shared memory": 4224, "blocks per SM": 34.909092, "warps per SM": 279.272736, "grid": [ 1, 12, 384 ], "block": [ 256, 1, 1 ], "est. achieved occupancy %": 100, "occupancy": { "activeBlocksPerMultiprocessor": 8, "limitingFactors": "WARPS|REGS", "blockLimitRegs": 8, "blockLimitSharedMem": 44, "blockLimitWarps": 8, "blockLimitBlocks": 32, "blockLimitBarriers": 2147483647, "allocatedRegistersPerBlock": 8192, "allocatedSharedMemPerBlock": 5248 }, "graph id": 0, "graph node id": 0, "channel": 0, "channel_type": 1 } }, { "name": "sm90_xmma_fprop_implicit_gemm_f32f32_tf32f32_f32_nhwckrsc_nhwc_tilesize64x64x32_warpgroupsize1x1x1_g1_execute_segment_k_off_kernel__5x_cudnn", "duration_us": 92.256, "pid": 0, "tid": 7, "args": { "External id": 4, "queued": 0, "device": 0, "context": 1, "stream": 7, "correlation": 22, "registers per thread": 168, "shared memory": 231424, "blocks per SM": 0.909091, "warps per SM": 10.909091, "grid": [ 60, 2, 1 ], "block": [ 384, 1, 1 ], "est. achieved occupancy %": 0, "occupancy": { "activeBlocksPerMultiprocessor": 0, "limitingFactors": "SMEM", "blockLimitRegs": 1, "blockLimitSharedMem": 0, "blockLimitWarps": 5, "blockLimitBlocks": 32, "blockLimitBarriers": 2147483647, "allocatedRegistersPerBlock": 64512, "allocatedSharedMemPerBlock": 232448 }, "graph id": 0, "graph node id": 0, "channel": 0, "channel_type": 1 } }, { "name": "void cudnn::engines_precompiled::nhwcToNchwKernel(cudnn::engines_precompiled::nhwc2nchw_params_t, float const*, float*)", "duration_us": 2.368, "pid": 0, "tid": 7, "args": { "External id": 4, "queued": 0, "device": 0, "context": 1, "stream": 7, "correlation": 26, "registers per thread": 32, "shared memory": 4224, "blocks per SM": 4.454545, "warps per SM": 35.636364, "grid": [ 49, 12, 1 ], "block": [ 256, 1, 1 ], "est. achieved occupancy %": 56, "occupancy": { "activeBlocksPerMultiprocessor": 8, "limitingFactors": "WARPS|REGS", "blockLimitRegs": 8, "blockLimitSharedMem": 44, "blockLimitWarps": 8, "blockLimitBlocks": 32, "blockLimitBarriers": 2147483647, "allocatedRegistersPerBlock": 8192, "allocatedSharedMemPerBlock": 5248 }, "graph id": 0, "graph node id": 0, "channel": 0, "channel_type": 1 } }, { "name": "void at::native::elementwise_kernel<128, 2, at::native::gpu_kernel_impl_nocast >(at::TensorIteratorBase&, at::native::CUDAFunctor_add const&)::{lambda(int)#1}>(int, at::native::gpu_kernel_impl_nocast >(at::TensorIteratorBase&, at::native::CUDAFunctor_add const&)::{lambda(int)#1})", "duration_us": 3.232, "pid": 0, "tid": 7, "args": { "External id": 7, "queued": 0, "device": 0, "context": 1, "stream": 7, "correlation": 36, "registers per thread": 18, "shared memory": 0, "blocks per SM": 17.727272, "warps per SM": 70.909088, "grid": [ 2340, 1, 1 ], "block": [ 128, 1, 1 ], "est. achieved occupancy %": 100, "occupancy": { "activeBlocksPerMultiprocessor": 16, "limitingFactors": "WARPS", "blockLimitRegs": 21, "blockLimitSharedMem": 228, "blockLimitWarps": 16, "blockLimitBlocks": 32, "blockLimitBarriers": 2147483647, "allocatedRegistersPerBlock": 3072, "allocatedSharedMemPerBlock": 1024 }, "graph id": 0, "graph node id": 0, "channel": 0, "channel_type": 1 } } ], "checks": { "exit_zero": true, "input_hash_matches": true, "finite_output": true, "actual_gpu_kernel_recorded": true, "operator_recorded": true }, "evidence_checks_passed": true }, { "input": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/operator_capture_480x832_v2/operators/vae_rank0_c16d11a7624e1a8c.pt", "sha256": "1e43418295ca6422f27afcf0b9940aa818fbb12fcf1013c12dfcf7a60f86e7d7", "result": { "op": "aten.conv3d.default", "stage": "vae", "shape": [ 1, 384, 1, 120, 52 ], "exact": true, "max_abs": 0.0, "rmse": 0.0, "finite": true, "math_policy": { "matmul_precision": "highest", "matmul_allow_tf32": false, "cudnn_allow_tf32": true, "cudnn_benchmark": false, "cudnn_deterministic": false }, "scope": "Original ATen op on saved real tensors; isolated diagnostic" }, "trace": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/operator_replay_480x832_v2/011_vae.trace.json", "trace_sha256": "d219fed997a5569d3377d5a4d27ccc850395b8af1c1d7e5878ba90a1c31dbf49", "kernels": [ { "name": "void cutlass::Kernel2(cutlass_80_tensorop_s1688gemm_64x128_32x3_nn_align4::Params)", "duration_us": 11.008, "pid": 0, "tid": 7, "args": { "External id": 4, "queued": 0, "device": 0, "context": 1, "stream": 7, "correlation": 18, "registers per thread": 168, "shared memory": 73728, "blocks per SM": 2.363636, "warps per SM": 9.454545, "grid": [ 24, 13, 1 ], "block": [ 128, 1, 1 ], "est. achieved occupancy %": 0, "occupancy": { "activeBlocksPerMultiprocessor": 0, "limitingFactors": "SMEM", "blockLimitRegs": 3, "blockLimitSharedMem": 0, "blockLimitWarps": 16, "blockLimitBlocks": 32, "blockLimitBarriers": 2147483647, "allocatedRegistersPerBlock": 21504, "allocatedSharedMemPerBlock": 74752 }, "graph id": 0, "graph node id": 0, "channel": 0, "channel_type": 1 } }, { "name": "void at::native::elementwise_kernel<128, 2, at::native::gpu_kernel_impl_nocast >(at::TensorIteratorBase&, at::native::CUDAFunctor_add const&)::{lambda(int)#1}>(int, at::native::gpu_kernel_impl_nocast >(at::TensorIteratorBase&, at::native::CUDAFunctor_add const&)::{lambda(int)#1})", "duration_us": 7.392, "pid": 0, "tid": 7, "args": { "External id": 7, "queued": 0, "device": 0, "context": 1, "stream": 7, "correlation": 26, "registers per thread": 18, "shared memory": 0, "blocks per SM": 70.909088, "warps per SM": 283.636353, "grid": [ 9360, 1, 1 ], "block": [ 128, 1, 1 ], "est. achieved occupancy %": 100, "occupancy": { "activeBlocksPerMultiprocessor": 16, "limitingFactors": "WARPS", "blockLimitRegs": 21, "blockLimitSharedMem": 228, "blockLimitWarps": 16, "blockLimitBlocks": 32, "blockLimitBarriers": 2147483647, "allocatedRegistersPerBlock": 3072, "allocatedSharedMemPerBlock": 1024 }, "graph id": 0, "graph node id": 0, "channel": 0, "channel_type": 1 } } ], "checks": { "exit_zero": true, "input_hash_matches": true, "finite_output": true, "actual_gpu_kernel_recorded": true, "operator_recorded": true }, "evidence_checks_passed": true }, { "input": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/operator_capture_480x832_v2/operators/vae_rank0_081e405bd1e07dce.pt", "sha256": "3193a32ddc024e9d34eb5446f9909408dde38d963ebc5cb9654998ecf661a820", "result": { "op": "aten.conv3d.default", "stage": "vae", "shape": [ 1, 384, 1, 120, 52 ], "exact": true, "max_abs": 0.0, "rmse": 0.0, "finite": true, "math_policy": { "matmul_precision": "highest", "matmul_allow_tf32": false, "cudnn_allow_tf32": true, "cudnn_benchmark": false, "cudnn_deterministic": false }, "scope": "Original ATen op on saved real tensors; isolated diagnostic" }, "trace": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/operator_replay_480x832_v2/012_vae.trace.json", "trace_sha256": "3bed073d85a198cc0b7060c5b3e8574ac9482bd08fad1467777e7086a9e1979a", "kernels": [ { "name": "void cudnn::engines_precompiled::nchwToNhwcKernel(cudnn::engines_precompiled::nchw2nhwc_params_t, float const*, float*)", "duration_us": 8.832, "pid": 0, "tid": 7, "args": { "External id": 4, "queued": 0, "device": 0, "context": 1, "stream": 7, "correlation": 12, "registers per thread": 32, "shared memory": 4224, "blocks per SM": 28.09091, "warps per SM": 224.72728, "grid": [ 618, 6, 1 ], "block": [ 256, 1, 1 ], "est. achieved occupancy %": 100, "occupancy": { "activeBlocksPerMultiprocessor": 8, "limitingFactors": "WARPS|REGS", "blockLimitRegs": 8, "blockLimitSharedMem": 44, "blockLimitWarps": 8, "blockLimitBlocks": 32, "blockLimitBarriers": 2147483647, "allocatedRegistersPerBlock": 8192, "allocatedSharedMemPerBlock": 5248 }, "graph id": 0, "graph node id": 0, "channel": 0, "channel_type": 1 } }, { "name": "void cudnn::engines_precompiled::nchwToNhwcKernel(cudnn::engines_precompiled::nchw2nhwc_params_t, float const*, float*)", "duration_us": 6.048, "pid": 0, "tid": 7, "args": { "External id": 4, "queued": 0, "device": 0, "context": 1, "stream": 7, "correlation": 16, "registers per thread": 32, "shared memory": 4224, "blocks per SM": 17.454546, "warps per SM": 139.636368, "grid": [ 1, 6, 384 ], "block": [ 256, 1, 1 ], "est. achieved occupancy %": 100, "occupancy": { "activeBlocksPerMultiprocessor": 8, "limitingFactors": "WARPS|REGS", "blockLimitRegs": 8, "blockLimitSharedMem": 44, "blockLimitWarps": 8, "blockLimitBlocks": 32, "blockLimitBarriers": 2147483647, "allocatedRegistersPerBlock": 8192, "allocatedSharedMemPerBlock": 5248 }, "graph id": 0, "graph node id": 0, "channel": 0, "channel_type": 1 } }, { "name": "void cask_plugin__5x_cudnn::xmma__5x_cudnn::init_device_workspace_kernel >(xmma__5x_cudnn::implicit_gemm::fprop::Warp_specialized_params_non_template, bool)", "duration_us": 2.176, "pid": 0, "tid": 7, "args": { "External id": 4, "queued": 0, "device": 0, "context": 1, "stream": 7, "correlation": 20, "registers per thread": 16, "shared memory": 0, "blocks per SM": 0.007576, "warps per SM": 0.000237, "grid": [ 1, 1, 1 ], "block": [ 1, 1, 1 ], "est. achieved occupancy %": 0, "occupancy": { "activeBlocksPerMultiprocessor": 32, "limitingFactors": "BLOCKS", "blockLimitRegs": 128, "blockLimitSharedMem": 228, "blockLimitWarps": 64, "blockLimitBlocks": 32, "blockLimitBarriers": 2147483647, "allocatedRegistersPerBlock": 512, "allocatedSharedMemPerBlock": 1024 }, "graph id": 0, "graph node id": 0, "channel": 0, "channel_type": 1 } }, { "name": "sm90_xmma_fprop_implicit_gemm_f32f32_tf32f32_f32_nhwckrsc_nhwc_tilesize128x128x32_warpgroupsize1x1x1_g1_execute_segment_k_on_kernel__5x_cudnn", "duration_us": 119.968, "pid": 0, "tid": 7, "args": { "External id": 4, "queued": 0, "device": 0, "context": 1, "stream": 7, "correlation": 23, "registers per thread": 168, "shared memory": 231424, "blocks per SM": 1.0, "warps per SM": 12.0, "grid": [ 132, 1, 1 ], "block": [ 384, 1, 1 ], "est. achieved occupancy %": 0, "occupancy": { "activeBlocksPerMultiprocessor": 0, "limitingFactors": "SMEM", "blockLimitRegs": 1, "blockLimitSharedMem": 0, "blockLimitWarps": 5, "blockLimitBlocks": 32, "blockLimitBarriers": 2147483647, "allocatedRegistersPerBlock": 64512, "allocatedSharedMemPerBlock": 232448 }, "graph id": 0, "graph node id": 0, "channel": 0, "channel_type": 1 } }, { "name": "void cudnn::engines_precompiled::nhwcToNchwKernel(cudnn::engines_precompiled::nhwc2nchw_params_t, float const*, float*)", "duration_us": 5.312, "pid": 0, "tid": 7, "args": { "External id": 4, "queued": 0, "device": 0, "context": 1, "stream": 7, "correlation": 27, "registers per thread": 32, "shared memory": 4224, "blocks per SM": 17.727272, "warps per SM": 141.818176, "grid": [ 195, 12, 1 ], "block": [ 256, 1, 1 ], "est. achieved occupancy %": 100, "occupancy": { "activeBlocksPerMultiprocessor": 8, "limitingFactors": "WARPS|REGS", "blockLimitRegs": 8, "blockLimitSharedMem": 44, "blockLimitWarps": 8, "blockLimitBlocks": 32, "blockLimitBarriers": 2147483647, "allocatedRegistersPerBlock": 8192, "allocatedSharedMemPerBlock": 5248 }, "graph id": 0, "graph node id": 0, "channel": 0, "channel_type": 1 } }, { "name": "void at::native::elementwise_kernel<128, 2, at::native::gpu_kernel_impl_nocast >(at::TensorIteratorBase&, at::native::CUDAFunctor_add const&)::{lambda(int)#1}>(int, at::native::gpu_kernel_impl_nocast >(at::TensorIteratorBase&, at::native::CUDAFunctor_add const&)::{lambda(int)#1})", "duration_us": 7.552, "pid": 0, "tid": 7, "args": { "External id": 7, "queued": 0, "device": 0, "context": 1, "stream": 7, "correlation": 37, "registers per thread": 18, "shared memory": 0, "blocks per SM": 70.909088, "warps per SM": 283.636353, "grid": [ 9360, 1, 1 ], "block": [ 128, 1, 1 ], "est. achieved occupancy %": 100, "occupancy": { "activeBlocksPerMultiprocessor": 16, "limitingFactors": "WARPS", "blockLimitRegs": 21, "blockLimitSharedMem": 228, "blockLimitWarps": 16, "blockLimitBlocks": 32, "blockLimitBarriers": 2147483647, "allocatedRegistersPerBlock": 3072, "allocatedSharedMemPerBlock": 1024 }, "graph id": 0, "graph node id": 0, "channel": 0, "channel_type": 1 } } ], "checks": { "exit_zero": true, "input_hash_matches": true, "finite_output": true, "actual_gpu_kernel_recorded": true, "operator_recorded": true }, "evidence_checks_passed": true }, { "input": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/operator_capture_480x832_v2/operators/vae_rank0_8601bc3a88374d99.pt", "sha256": "c7f10fe8b8ca84229e785645c70c003456c3950c79cdd176694e5bc86c9247f4", "result": { "op": "aten.conv3d.default", "stage": "vae", "shape": [ 1, 384, 1, 120, 52 ], "exact": true, "max_abs": 0.0, "rmse": 0.0, "finite": true, "math_policy": { "matmul_precision": "highest", "matmul_allow_tf32": false, "cudnn_allow_tf32": true, "cudnn_benchmark": false, "cudnn_deterministic": false }, "scope": "Original ATen op on saved real tensors; isolated diagnostic" }, "trace": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/operator_replay_480x832_v2/013_vae.trace.json", "trace_sha256": "83ea7aa0395a48a682f41d310f6e1f134334b98540b6660943aa0a0b7181c56e", "kernels": [ { "name": "void cudnn::engines_precompiled::nchwToNhwcKernel(cudnn::engines_precompiled::nchw2nhwc_params_t, float const*, float*)", "duration_us": 18.784, "pid": 0, "tid": 7, "args": { "External id": 4, "queued": 0, "device": 0, "context": 1, "stream": 7, "correlation": 12, "registers per thread": 32, "shared memory": 4224, "blocks per SM": 56.18182, "warps per SM": 449.454559, "grid": [ 618, 12, 1 ], "block": [ 256, 1, 1 ], "est. achieved occupancy %": 100, "occupancy": { "activeBlocksPerMultiprocessor": 8, "limitingFactors": "WARPS|REGS", "blockLimitRegs": 8, "blockLimitSharedMem": 44, "blockLimitWarps": 8, "blockLimitBlocks": 32, "blockLimitBarriers": 2147483647, "allocatedRegistersPerBlock": 8192, "allocatedSharedMemPerBlock": 5248 }, "graph id": 0, "graph node id": 0, "channel": 0, "channel_type": 1 } }, { "name": "void cudnn::engines_precompiled::nchwToNhwcKernel(cudnn::engines_precompiled::nchw2nhwc_params_t, float const*, float*)", "duration_us": 11.52, "pid": 0, "tid": 7, "args": { "External id": 4, "queued": 0, "device": 0, "context": 1, "stream": 7, "correlation": 16, "registers per thread": 32, "shared memory": 4224, "blocks per SM": 34.909092, "warps per SM": 279.272736, "grid": [ 1, 12, 384 ], "block": [ 256, 1, 1 ], "est. achieved occupancy %": 100, "occupancy": { "activeBlocksPerMultiprocessor": 8, "limitingFactors": "WARPS|REGS", "blockLimitRegs": 8, "blockLimitSharedMem": 44, "blockLimitWarps": 8, "blockLimitBlocks": 32, "blockLimitBarriers": 2147483647, "allocatedRegistersPerBlock": 8192, "allocatedSharedMemPerBlock": 5248 }, "graph id": 0, "graph node id": 0, "channel": 0, "channel_type": 1 } }, { "name": "void cask_plugin__5x_cudnn::xmma__5x_cudnn::init_device_workspace_kernel >(xmma__5x_cudnn::implicit_gemm::fprop::Warp_specialized_params_non_template, bool)", "duration_us": 2.24, "pid": 0, "tid": 7, "args": { "External id": 4, "queued": 0, "device": 0, "context": 1, "stream": 7, "correlation": 20, "registers per thread": 16, "shared memory": 0, "blocks per SM": 0.007576, "warps per SM": 0.000237, "grid": [ 1, 1, 1 ], "block": [ 1, 1, 1 ], "est. achieved occupancy %": 0, "occupancy": { "activeBlocksPerMultiprocessor": 32, "limitingFactors": "BLOCKS", "blockLimitRegs": 128, "blockLimitSharedMem": 228, "blockLimitWarps": 64, "blockLimitBlocks": 32, "blockLimitBarriers": 2147483647, "allocatedRegistersPerBlock": 512, "allocatedSharedMemPerBlock": 1024 }, "graph id": 0, "graph node id": 0, "channel": 0, "channel_type": 1 } }, { "name": "sm90_xmma_fprop_implicit_gemm_f32f32_tf32f32_f32_nhwckrsc_nhwc_tilesize128x128x32_warpgroupsize1x1x1_g1_execute_segment_k_on_kernel__5x_cudnn", "duration_us": 234.433, "pid": 0, "tid": 7, "args": { "External id": 4, "queued": 0, "device": 0, "context": 1, "stream": 7, "correlation": 23, "registers per thread": 168, "shared memory": 231424, "blocks per SM": 1.0, "warps per SM": 12.0, "grid": [ 132, 1, 1 ], "block": [ 384, 1, 1 ], "est. achieved occupancy %": 0, "occupancy": { "activeBlocksPerMultiprocessor": 0, "limitingFactors": "SMEM", "blockLimitRegs": 1, "blockLimitSharedMem": 0, "blockLimitWarps": 5, "blockLimitBlocks": 32, "blockLimitBarriers": 2147483647, "allocatedRegistersPerBlock": 64512, "allocatedSharedMemPerBlock": 232448 }, "graph id": 0, "graph node id": 0, "channel": 0, "channel_type": 1 } }, { "name": "void cudnn::engines_precompiled::nhwcToNchwKernel(cudnn::engines_precompiled::nhwc2nchw_params_t, float const*, float*)", "duration_us": 5.28, "pid": 0, "tid": 7, "args": { "External id": 4, "queued": 0, "device": 0, "context": 1, "stream": 7, "correlation": 27, "registers per thread": 32, "shared memory": 4224, "blocks per SM": 17.727272, "warps per SM": 141.818176, "grid": [ 195, 12, 1 ], "block": [ 256, 1, 1 ], "est. achieved occupancy %": 100, "occupancy": { "activeBlocksPerMultiprocessor": 8, "limitingFactors": "WARPS|REGS", "blockLimitRegs": 8, "blockLimitSharedMem": 44, "blockLimitWarps": 8, "blockLimitBlocks": 32, "blockLimitBarriers": 2147483647, "allocatedRegistersPerBlock": 8192, "allocatedSharedMemPerBlock": 5248 }, "graph id": 0, "graph node id": 0, "channel": 0, "channel_type": 1 } }, { "name": "void at::native::elementwise_kernel<128, 2, at::native::gpu_kernel_impl_nocast >(at::TensorIteratorBase&, at::native::CUDAFunctor_add const&)::{lambda(int)#1}>(int, at::native::gpu_kernel_impl_nocast >(at::TensorIteratorBase&, at::native::CUDAFunctor_add const&)::{lambda(int)#1})", "duration_us": 7.68, "pid": 0, "tid": 7, "args": { "External id": 7, "queued": 0, "device": 0, "context": 1, "stream": 7, "correlation": 37, "registers per thread": 18, "shared memory": 0, "blocks per SM": 70.909088, "warps per SM": 283.636353, "grid": [ 9360, 1, 1 ], "block": [ 128, 1, 1 ], "est. achieved occupancy %": 100, "occupancy": { "activeBlocksPerMultiprocessor": 16, "limitingFactors": "WARPS", "blockLimitRegs": 21, "blockLimitSharedMem": 228, "blockLimitWarps": 16, "blockLimitBlocks": 32, "blockLimitBarriers": 2147483647, "allocatedRegistersPerBlock": 3072, "allocatedSharedMemPerBlock": 1024 }, "graph id": 0, "graph node id": 0, "channel": 0, "channel_type": 1 } } ], "checks": { "exit_zero": true, "input_hash_matches": true, "finite_output": true, "actual_gpu_kernel_recorded": true, "operator_recorded": true }, "evidence_checks_passed": true }, { "input": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/operator_capture_480x832_v2/operators/vae_rank0_99475e040d65d084.pt", "sha256": "0cc0a21ce474602c53c3e5621e64598ed7be5a2c16603fc968441df41de71330", "result": { "op": "aten.conv3d.default", "stage": "vae", "shape": [ 1, 192, 1, 240, 104 ], "exact": true, "max_abs": 0.0, "rmse": 0.0, "finite": true, "math_policy": { "matmul_precision": "highest", "matmul_allow_tf32": false, "cudnn_allow_tf32": true, "cudnn_benchmark": false, "cudnn_deterministic": false }, "scope": "Original ATen op on saved real tensors; isolated diagnostic" }, "trace": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/operator_replay_480x832_v2/014_vae.trace.json", "trace_sha256": "97c555c1cea1ae9dfc7712a2ccb7e844fe2fdeb95ea6ab80d554bc14b89ecc97", "kernels": [ { "name": "void cudnn::engines_precompiled::nchwToNhwcKernel(cudnn::engines_precompiled::nchw2nhwc_params_t, float const*, float*)", "duration_us": 37.343, "pid": 0, "tid": 7, "args": { "External id": 4, "queued": 0, "device": 0, "context": 1, "stream": 7, "correlation": 12, "registers per thread": 32, "shared memory": 4224, "blocks per SM": 109.318184, "warps per SM": 874.545471, "grid": [ 2405, 6, 1 ], "block": [ 256, 1, 1 ], "est. achieved occupancy %": 100, "occupancy": { "activeBlocksPerMultiprocessor": 8, "limitingFactors": "WARPS|REGS", "blockLimitRegs": 8, "blockLimitSharedMem": 44, "blockLimitWarps": 8, "blockLimitBlocks": 32, "blockLimitBarriers": 2147483647, "allocatedRegistersPerBlock": 8192, "allocatedSharedMemPerBlock": 5248 }, "graph id": 0, "graph node id": 0, "channel": 0, "channel_type": 1 } }, { "name": "void cudnn::engines_precompiled::nchwToNhwcKernel(cudnn::engines_precompiled::nchw2nhwc_params_t, float const*, float*)", "duration_us": 4.769, "pid": 0, "tid": 7, "args": { "External id": 4, "queued": 0, "device": 0, "context": 1, "stream": 7, "correlation": 16, "registers per thread": 32, "shared memory": 4224, "blocks per SM": 8.727273, "warps per SM": 69.818184, "grid": [ 1, 6, 192 ], "block": [ 256, 1, 1 ], "est. achieved occupancy %": 100, "occupancy": { "activeBlocksPerMultiprocessor": 8, "limitingFactors": "WARPS|REGS", "blockLimitRegs": 8, "blockLimitSharedMem": 44, "blockLimitWarps": 8, "blockLimitBlocks": 32, "blockLimitBarriers": 2147483647, "allocatedRegistersPerBlock": 8192, "allocatedSharedMemPerBlock": 5248 }, "graph id": 0, "graph node id": 0, "channel": 0, "channel_type": 1 } }, { "name": "sm90_xmma_fprop_implicit_gemm_f32f32_tf32f32_f32_nhwckrsc_nhwc_tilesize128x128x32_warpgroupsize1x1x1_g1_execute_segment_k_off_kernel__5x_cudnn", "duration_us": 163.84, "pid": 0, "tid": 7, "args": { "External id": 4, "queued": 0, "device": 0, "context": 1, "stream": 7, "correlation": 23, "registers per thread": 168, "shared memory": 231424, "blocks per SM": 1.0, "warps per SM": 12.0, "grid": [ 132, 1, 1 ], "block": [ 384, 1, 1 ], "est. achieved occupancy %": 0, "occupancy": { "activeBlocksPerMultiprocessor": 0, "limitingFactors": "SMEM", "blockLimitRegs": 1, "blockLimitSharedMem": 0, "blockLimitWarps": 5, "blockLimitBlocks": 32, "blockLimitBarriers": 2147483647, "allocatedRegistersPerBlock": 64512, "allocatedSharedMemPerBlock": 232448 }, "graph id": 0, "graph node id": 0, "channel": 0, "channel_type": 1 } }, { "name": "void cudnn::engines_precompiled::nhwcToNchwKernel(cudnn::engines_precompiled::nhwc2nchw_params_t, float const*, float*)", "duration_us": 9.249, "pid": 0, "tid": 7, "args": { "External id": 4, "queued": 0, "device": 0, "context": 1, "stream": 7, "correlation": 27, "registers per thread": 32, "shared memory": 4224, "blocks per SM": 35.454544, "warps per SM": 283.636353, "grid": [ 780, 6, 1 ], "block": [ 256, 1, 1 ], "est. achieved occupancy %": 100, "occupancy": { "activeBlocksPerMultiprocessor": 8, "limitingFactors": "WARPS|REGS", "blockLimitRegs": 8, "blockLimitSharedMem": 44, "blockLimitWarps": 8, "blockLimitBlocks": 32, "blockLimitBarriers": 2147483647, "allocatedRegistersPerBlock": 8192, "allocatedSharedMemPerBlock": 5248 }, "graph id": 0, "graph node id": 0, "channel": 0, "channel_type": 1 } }, { "name": "void at::native::elementwise_kernel<128, 2, at::native::gpu_kernel_impl_nocast >(at::TensorIteratorBase&, at::native::CUDAFunctor_add const&)::{lambda(int)#1}>(int, at::native::gpu_kernel_impl_nocast >(at::TensorIteratorBase&, at::native::CUDAFunctor_add const&)::{lambda(int)#1})", "duration_us": 13.215, "pid": 0, "tid": 7, "args": { "External id": 7, "queued": 0, "device": 0, "context": 1, "stream": 7, "correlation": 37, "registers per thread": 18, "shared memory": 0, "blocks per SM": 141.818176, "warps per SM": 567.272705, "grid": [ 18720, 1, 1 ], "block": [ 128, 1, 1 ], "est. achieved occupancy %": 100, "occupancy": { "activeBlocksPerMultiprocessor": 16, "limitingFactors": "WARPS", "blockLimitRegs": 21, "blockLimitSharedMem": 228, "blockLimitWarps": 16, "blockLimitBlocks": 32, "blockLimitBarriers": 2147483647, "allocatedRegistersPerBlock": 3072, "allocatedSharedMemPerBlock": 1024 }, "graph id": 0, "graph node id": 0, "channel": 0, "channel_type": 1 } } ], "checks": { "exit_zero": true, "input_hash_matches": true, "finite_output": true, "actual_gpu_kernel_recorded": true, "operator_recorded": true }, "evidence_checks_passed": true }, { "input": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/operator_capture_480x832_v2/operators/vae_rank0_0bd1bd0a5c44ff3b.pt", "sha256": "1e47965836d719f164296b5c8d52cac8c22a10d9ff8c5f4fdfa48498909f8856", "result": { "op": "aten.conv3d.default", "stage": "vae", "shape": [ 1, 96, 1, 480, 208 ], "exact": true, "max_abs": 0.0, "rmse": 0.0, "finite": true, "math_policy": { "matmul_precision": "highest", "matmul_allow_tf32": false, "cudnn_allow_tf32": true, "cudnn_benchmark": false, "cudnn_deterministic": false }, "scope": "Original ATen op on saved real tensors; isolated diagnostic" }, "trace": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/operator_replay_480x832_v2/015_vae.trace.json", "trace_sha256": "6d47b822d1f005d7a17baec0d6c08f53687391c779ad9e06dca571464ead28e8", "kernels": [ { "name": "void cudnn::engines_precompiled::nchwToNhwcKernel(cudnn::engines_precompiled::nchw2nhwc_params_t, float const*, float*)", "duration_us": 73.216, "pid": 0, "tid": 7, "args": { "External id": 4, "queued": 0, "device": 0, "context": 1, "stream": 7, "correlation": 12, "registers per thread": 32, "shared memory": 4224, "blocks per SM": 215.681824, "warps per SM": 1725.45459, "grid": [ 9490, 3, 1 ], "block": [ 256, 1, 1 ], "est. achieved occupancy %": 100, "occupancy": { "activeBlocksPerMultiprocessor": 8, "limitingFactors": "WARPS|REGS", "blockLimitRegs": 8, "blockLimitSharedMem": 44, "blockLimitWarps": 8, "blockLimitBlocks": 32, "blockLimitBarriers": 2147483647, "allocatedRegistersPerBlock": 8192, "allocatedSharedMemPerBlock": 5248 }, "graph id": 0, "graph node id": 0, "channel": 0, "channel_type": 1 } }, { "name": "void cudnn::engines_precompiled::nchwToNhwcKernel(cudnn::engines_precompiled::nchw2nhwc_params_t, float const*, float*)", "duration_us": 2.688, "pid": 0, "tid": 7, "args": { "External id": 4, "queued": 0, "device": 0, "context": 1, "stream": 7, "correlation": 16, "registers per thread": 32, "shared memory": 4224, "blocks per SM": 2.181818, "warps per SM": 17.454546, "grid": [ 1, 3, 96 ], "block": [ 256, 1, 1 ], "est. achieved occupancy %": 27, "occupancy": { "activeBlocksPerMultiprocessor": 8, "limitingFactors": "WARPS|REGS", "blockLimitRegs": 8, "blockLimitSharedMem": 44, "blockLimitWarps": 8, "blockLimitBlocks": 32, "blockLimitBarriers": 2147483647, "allocatedRegistersPerBlock": 8192, "allocatedSharedMemPerBlock": 5248 }, "graph id": 0, "graph node id": 0, "channel": 0, "channel_type": 1 } }, { "name": "sm90_xmma_fprop_implicit_gemm_f32f32_tf32f32_f32_nhwckrsc_nhwc_tilesize256x32x32_warpgroupsize1x1x1_g1_execute_segment_k_off_kernel__5x_cudnn", "duration_us": 344.96, "pid": 0, "tid": 7, "args": { "External id": 4, "queued": 0, "device": 0, "context": 1, "stream": 7, "correlation": 23, "registers per thread": 168, "shared memory": 231424, "blocks per SM": 1.0, "warps per SM": 12.0, "grid": [ 66, 2, 1 ], "block": [ 384, 1, 1 ], "est. achieved occupancy %": 0, "occupancy": { "activeBlocksPerMultiprocessor": 0, "limitingFactors": "SMEM", "blockLimitRegs": 1, "blockLimitSharedMem": 0, "blockLimitWarps": 5, "blockLimitBlocks": 32, "blockLimitBarriers": 2147483647, "allocatedRegistersPerBlock": 64512, "allocatedSharedMemPerBlock": 232448 }, "graph id": 0, "graph node id": 0, "channel": 0, "channel_type": 1 } }, { "name": "void cudnn::engines_precompiled::nhwcToNchwKernel(cudnn::engines_precompiled::nhwc2nchw_params_t, float const*, float*)", "duration_us": 19.393, "pid": 0, "tid": 7, "args": { "External id": 4, "queued": 0, "device": 0, "context": 1, "stream": 7, "correlation": 27, "registers per thread": 32, "shared memory": 4224, "blocks per SM": 70.909088, "warps per SM": 567.272705, "grid": [ 3120, 3, 1 ], "block": [ 256, 1, 1 ], "est. achieved occupancy %": 100, "occupancy": { "activeBlocksPerMultiprocessor": 8, "limitingFactors": "WARPS|REGS", "blockLimitRegs": 8, "blockLimitSharedMem": 44, "blockLimitWarps": 8, "blockLimitBlocks": 32, "blockLimitBarriers": 2147483647, "allocatedRegistersPerBlock": 8192, "allocatedSharedMemPerBlock": 5248 }, "graph id": 0, "graph node id": 0, "channel": 0, "channel_type": 1 } }, { "name": "void at::native::elementwise_kernel<128, 2, at::native::gpu_kernel_impl_nocast >(at::TensorIteratorBase&, at::native::CUDAFunctor_add const&)::{lambda(int)#1}>(int, at::native::gpu_kernel_impl_nocast >(at::TensorIteratorBase&, at::native::CUDAFunctor_add const&)::{lambda(int)#1})", "duration_us": 27.84, "pid": 0, "tid": 7, "args": { "External id": 7, "queued": 0, "device": 0, "context": 1, "stream": 7, "correlation": 37, "registers per thread": 18, "shared memory": 0, "blocks per SM": 283.636353, "warps per SM": 1134.54541, "grid": [ 37440, 1, 1 ], "block": [ 128, 1, 1 ], "est. achieved occupancy %": 100, "occupancy": { "activeBlocksPerMultiprocessor": 16, "limitingFactors": "WARPS", "blockLimitRegs": 21, "blockLimitSharedMem": 228, "blockLimitWarps": 16, "blockLimitBlocks": 32, "blockLimitBarriers": 2147483647, "allocatedRegistersPerBlock": 3072, "allocatedSharedMemPerBlock": 1024 }, "graph id": 0, "graph node id": 0, "channel": 0, "channel_type": 1 } } ], "checks": { "exit_zero": true, "input_hash_matches": true, "finite_output": true, "actual_gpu_kernel_recorded": true, "operator_recorded": true }, "evidence_checks_passed": true }, { "input": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/operator_capture_480x832_v2/operators/output_rank0_f18eb91117dda1c8.pt", "sha256": "897dd2bd3ad5fe7258ee5c3ae9cfd93da4e5e94644902267f0b3d8bd156f63b0", "result": { "op": "aten.to.dtype", "stage": "output", "shape": [ 1, 3, 9, 480, 832 ], "exact": true, "max_abs": 0.0, "rmse": 0.0, "finite": true, "math_policy": { "matmul_precision": "highest", "matmul_allow_tf32": false, "cudnn_allow_tf32": true, "cudnn_benchmark": false, "cudnn_deterministic": false }, "scope": "Original ATen op on saved real tensors; isolated diagnostic" }, "trace": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/operator_replay_480x832_v2/016_output.trace.json", "trace_sha256": "a3559b906e1f6675c7c04cc5a0a249501ee447441da27f63b0dc30d56bd249d4", "kernels": [ { "name": "void at::native::unrolled_elementwise_kernel, 4, TrivialOffsetCalculator<1, unsigned int>, TrivialOffsetCalculator<1, unsigned int>, at::native::memory::LoadWithCast<1>, at::native::memory::StoreWithCast<1> >(int, at::native::direct_copy_kernel_cuda(at::TensorIteratorBase&)::{lambda()#3}::operator()() const::{lambda()#1}::operator()() const::{lambda(unsigned char)#1}, std::array, TrivialOffsetCalculator<1, unsigned int>, TrivialOffsetCalculator<1, unsigned int>, at::native::memory::LoadWithCast<1>, at::native::memory::StoreWithCast<1>)", "duration_us": 31.265, "pid": 0, "tid": 7, "args": { "External id": 4, "queued": 0, "device": 0, "context": 1, "stream": 7, "correlation": 11, "registers per thread": 32, "shared memory": 0, "blocks per SM": 159.545456, "warps per SM": 638.181824, "grid": [ 21060, 1, 1 ], "block": [ 128, 1, 1 ], "est. achieved occupancy %": 100, "occupancy": { "activeBlocksPerMultiprocessor": 16, "limitingFactors": "WARPS|REGS", "blockLimitRegs": 16, "blockLimitSharedMem": 228, "blockLimitWarps": 16, "blockLimitBlocks": 32, "blockLimitBarriers": 2147483647, "allocatedRegistersPerBlock": 4096, "allocatedSharedMemPerBlock": 1024 }, "graph id": 0, "graph node id": 0, "channel": 0, "channel_type": 1 } } ], "checks": { "exit_zero": true, "input_hash_matches": true, "finite_output": true, "actual_gpu_kernel_recorded": true, "operator_recorded": true }, "evidence_checks_passed": true }, { "input": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/operator_capture_480x832_v2/operators/output_rank0_e753c57113db57aa.pt", "sha256": "a9aab61b63e7cab483e81582c9af41d89d525814ab14de753e8a7b16de6c4f1b", "result": { "op": "aten.to.dtype", "stage": "output", "shape": [ 1, 3, 12, 480, 832 ], "exact": true, "max_abs": 0.0, "rmse": 0.0, "finite": true, "math_policy": { "matmul_precision": "highest", "matmul_allow_tf32": false, "cudnn_allow_tf32": true, "cudnn_benchmark": false, "cudnn_deterministic": false }, "scope": "Original ATen op on saved real tensors; isolated diagnostic" }, "trace": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/operator_replay_480x832_v2/017_output.trace.json", "trace_sha256": "46b9179267d5c728836b8e2d658454a756add25b11bc36eea8cffda2b96bd7a7", "kernels": [ { "name": "void at::native::unrolled_elementwise_kernel, 4, TrivialOffsetCalculator<1, unsigned int>, TrivialOffsetCalculator<1, unsigned int>, at::native::memory::LoadWithCast<1>, at::native::memory::StoreWithCast<1> >(int, at::native::direct_copy_kernel_cuda(at::TensorIteratorBase&)::{lambda()#3}::operator()() const::{lambda()#1}::operator()() const::{lambda(unsigned char)#1}, std::array, TrivialOffsetCalculator<1, unsigned int>, TrivialOffsetCalculator<1, unsigned int>, at::native::memory::LoadWithCast<1>, at::native::memory::StoreWithCast<1>)", "duration_us": 40.288, "pid": 0, "tid": 7, "args": { "External id": 4, "queued": 0, "device": 0, "context": 1, "stream": 7, "correlation": 11, "registers per thread": 32, "shared memory": 0, "blocks per SM": 212.72728, "warps per SM": 850.909119, "grid": [ 28080, 1, 1 ], "block": [ 128, 1, 1 ], "est. achieved occupancy %": 100, "occupancy": { "activeBlocksPerMultiprocessor": 16, "limitingFactors": "WARPS|REGS", "blockLimitRegs": 16, "blockLimitSharedMem": 228, "blockLimitWarps": 16, "blockLimitBlocks": 32, "blockLimitBarriers": 2147483647, "allocatedRegistersPerBlock": 4096, "allocatedSharedMemPerBlock": 1024 }, "graph id": 0, "graph node id": 0, "channel": 0, "channel_type": 1 } } ], "checks": { "exit_zero": true, "input_hash_matches": true, "finite_output": true, "actual_gpu_kernel_recorded": true, "operator_recorded": true }, "evidence_checks_passed": true } ], "nonexact_output_count": 0, "hardware_counter_status": "missing_for_these_operator_inputs" }