{
 "generated_at": "2026-09-10T16:18:15+0800",
 "tag": "704x1280",
 "steady_gpu_ms": 38096.377424,
 "steady_share_with_shape": 0.9582048358488564,
 "steady_share_by_ncu_status": {
  "exact_card": 0.7827697175536048,
  "no_card": 0.2172302824463952
 },
 "by_role": {
  "dit": {
   "exact_card": 0.8287599517790589,
   "no_card": 0.17124004822094116
  },
  "vae": {
   "no_card": 0.5545470202674143,
   "exact_card": 0.4454529797325857
  }
 },
 "uncovered_top": [
  {
   "role": "dit",
   "unit": "_c10d_functional.all_to_all_single",
   "kernel": "ncclDevKernel_SendRecv(ncclDevKernelArgsStorage<(unsigned lo",
   "steady_ms": 2415.3595849999997
  },
  {
   "role": "dit",
   "unit": "vllm_omni.ar_diffusion_paged_write_attn",
   "kernel": "void at::native::vectorized_gather_kernel<(int)16, long>(cha",
   "steady_ms": 551.392408
  },
  {
   "role": "dit",
   "unit": "vllm_omni.ar_diffusion_paged_write_attn",
   "kernel": "void at::native::index_elementwise_kernel<(int)128, (int)4, ",
   "steady_ms": 452.83913099999995
  },
  {
   "role": "vae",
   "unit": "vae_frame/*",
   "kernel": "fmha_cutlassF_f32_aligned_32x128_gmem_sm80(PyTorchMemEffAtte",
   "steady_ms": 430.418209
  },
  {
   "role": "dit",
   "unit": "_c10d_functional.all_to_all_single",
   "kernel": "nvjet_sm90_tst_192x192_64x4_1x2_h_bz_coopB_bias_TNN",
   "steady_ms": 377.45315899999997
  },
  {
   "role": "dit",
   "unit": "triton_red_fused__to_copy_add_mul_native_layer_norm_split_squeeze_view_9",
   "kernel": "triton_red_fused__to_copy_add_mul_native_layer_norm_split_sq",
   "steady_ms": 178.94312400000004
  },
  {
   "role": "dit",
   "unit": "triton_poi_fused_gelu_view_13",
   "kernel": "triton_poi_fused_gelu_view_13",
   "steady_ms": 167.14757899999998
  },
  {
   "role": "dit",
   "unit": "triton_red_fused__to_copy_add_mul_native_layer_norm_split_squeeze_1",
   "kernel": "triton_red_fused__to_copy_add_mul_native_layer_norm_split_sq",
   "steady_ms": 154.37514600000003
  },
  {
   "role": "dit",
   "unit": "rotary_kernel_0",
   "kernel": "rotary_kernel",
   "steady_ms": 147.97998600000003
  },
  {
   "role": "dit",
   "unit": "triton_red_fused__to_copy_add_mul_native_layer_norm_split_squeeze_view_12",
   "kernel": "triton_red_fused__to_copy_add_mul_native_layer_norm_split_sq",
   "steady_ms": 147.877554
  },
  {
   "role": "dit",
   "unit": "triton_poi_fused_all_to_all_single_clone_split_with_sizes_stack_transpose_view_4",
   "kernel": "triton_poi_fused_all_to_all_single_clone_split_with_sizes_st",
   "steady_ms": 130.831945
  },
  {
   "role": "dit",
   "unit": "triton_poi_fused__to_copy_add_mul_split_squeeze_view_14",
   "kernel": "triton_poi_fused__to_copy_add_mul_split_squeeze_view_14",
   "steady_ms": 102.65756100000002
  },
  {
   "role": "dit",
   "unit": "triton_poi_fused_all_to_all_single_clone_transpose_unsqueeze_view_5",
   "kernel": "triton_poi_fused_all_to_all_single_clone_transpose_unsqueeze",
   "steady_ms": 97.14903899999999
  },
  {
   "role": "vae",
   "unit": "ncclGroupEnd",
   "kernel": "ncclDevKernel_SendRecv(ncclDevKernelArgsStorage<(unsigned lo",
   "steady_ms": 93.00771500000005
  },
  {
   "role": "vae",
   "unit": "vae_frame/*",
   "kernel": "void at::native::elementwise_kernel<(int)128, (int)2, void a",
   "steady_ms": 92.06099800000001
  },
  {
   "role": "dit",
   "unit": "triton_poi_fused_clone_transpose_view_8",
   "kernel": "triton_poi_fused_clone_transpose_view_8",
   "steady_ms": 83.73603900000002
  },
  {
   "role": "dit",
   "unit": "triton_poi_fused__to_copy_all_to_all_single_clone_div_mul_sqrt_transpose_view_11",
   "kernel": "triton_poi_fused__to_copy_all_to_all_single_clone_div_mul_sq",
   "steady_ms": 79.903088
  },
  {
   "role": "dit",
   "unit": "triton_poi_fused_silu_view_6",
   "kernel": "triton_poi_fused_silu_view_6",
   "steady_ms": 75.43051399999999
  },
  {
   "role": "vae",
   "unit": "vae/up_blocks.3.resnets.0.conv1/WanDistCausalConv3d",
   "kernel": "void at::native::elementwise_kernel<(int)128, (int)2, void a",
   "steady_ms": 71.233532
  },
  {
   "role": "vae",
   "unit": "vae/up_blocks.2.upsamplers.0.resample.1/WanDistConv2d",
   "kernel": "sm90_xmma_fprop_implicit_gemm_f32f32_tf32f32_f32_nhwckrsc_nh",
   "steady_ms": 70.899704
  },
  {
   "role": "vae",
   "unit": "vae/up_blocks.3.resnets.0.conv2/WanDistCausalConv3d",
   "kernel": "void at::native::elementwise_kernel<(int)128, (int)2, void a",
   "steady_ms": 70.44274700000001
  },
  {
   "role": "vae",
   "unit": "vae/up_blocks.3.resnets.1.conv1/WanDistCausalConv3d",
   "kernel": "void at::native::elementwise_kernel<(int)128, (int)2, void a",
   "steady_ms": 70.423207
  },
  {
   "role": "vae",
   "unit": "vae/up_blocks.3.resnets.1.conv2/WanDistCausalConv3d",
   "kernel": "void at::native::elementwise_kernel<(int)128, (int)2, void a",
   "steady_ms": 70.39531999999998
  },
  {
   "role": "vae",
   "unit": "vae/up_blocks.3.resnets.2.conv1/WanDistCausalConv3d",
   "kernel": "void at::native::elementwise_kernel<(int)128, (int)2, void a",
   "steady_ms": 70.340875
  },
  {
   "role": "vae",
   "unit": "vae/up_blocks.3.resnets.2.conv2/WanDistCausalConv3d",
   "kernel": "void at::native::elementwise_kernel<(int)128, (int)2, void a",
   "steady_ms": 70.33023299999999
  },
  {
   "role": "dit",
   "unit": "triton_red_fused__to_copy_div_mul_pow_split_with_sizes_sqrt_sum_view_3",
   "kernel": "triton_red_fused__to_copy_div_mul_pow_split_with_sizes_sqrt_",
   "steady_ms": 67.446052
  },
  {
   "role": "dit",
   "unit": "triton_poi_fused_add_view_7",
   "kernel": "triton_poi_fused_add_view_7",
   "steady_ms": 66.119455
  },
  {
   "role": "dit",
   "unit": "triton_red_fused__to_copy_div_mul_pow_split_with_sizes_sqrt_sum_view_2",
   "kernel": "triton_red_fused__to_copy_div_mul_pow_split_with_sizes_sqrt_",
   "steady_ms": 59.509965
  },
  {
   "role": "dit",
   "unit": "tfm_sub/sp_prepare/_LingBotSPPrepare",
   "kernel": "void at::native::elementwise_kernel<(int)128, (int)4, void a",
   "steady_ms": 59.25756499999999
  },
  {
   "role": "vae",
   "unit": "vae/conv_out/WanDistCausalConv3d",
   "kernel": "void at::native::elementwise_kernel<(int)128, (int)2, void a",
   "steady_ms": 58.451939
  }
 ],
 "positions_with_shape": 44,
 "vae_modules_with_shape": 39,
 "seconds": 0.2812676429748535,
 "scope": "Shapes come from the eager full-dispatch run of the same frozen source and configuration (not the traced run); compiled fused kernels get inferred activation shapes. NCU cards are isolated replays; exact_card means identical operator shape, representative_only means same operator family with a different KV length."
}