{
  "status": "derived_from_audited_measurements",
  "generated_at": "2026-09-10T04:43:11.430149+00:00",
  "rows": [
    {
      "input": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/missing_operator_480x832_tp1sp4_v1/operators/dit_rank0_80ac9ebe75667de45dc0a836a408223ec527b2f2b8104a118ed8ed9fec89b6ee.pt",
      "input_sha256": "b8c205d5140f4a6d1aabaa7d0946d8eb3aa5bdc6007bff3e8efd26fcf1740c87",
      "stage": "dit",
      "op": "aten.linear.default",
      "input_tensor": {
        "shape": [
          1,
          1170,
          5120
        ],
        "stride": [
          5990400,
          5120,
          1
        ],
        "dtype": "torch.bfloat16",
        "logical_bytes": 11980800
      },
      "output_tensor": {
        "shape": [
          1,
          1170,
          13824
        ],
        "stride": [
          16174080,
          13824,
          1
        ],
        "dtype": "torch.bfloat16",
        "logical_bytes": 32348160
      },
      "shape_parameters": {
        "M": 1170,
        "N": 13824,
        "K": 5120
      },
      "mac_flops": 165622579200,
      "bias_additions": 16174080,
      "formula": "2*M*N*K (multiply and add each count one); bias additions listed separately",
      "profiled_kernel_count": 1,
      "summed_profiled_kernel_us": 215.52,
      "dram_read_bytes": 199619072.0,
      "dram_write_bytes": 29967872.0,
      "dram_total_bytes": 229586944.0,
      "nominal_mac_tflops_per_s": 768.4789309576837,
      "nominal_mac_flops_per_dram_byte": 721.3937182769417,
      "ncu_output": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/ncu_missing_80ac9ebe75667de4_v1",
      "ncu_metrics_sha256": "622ae0f78b6fa48ef97e64ccfee2d66b43a3c7e23e6daa664771165b7eaa1290",
      "tensor_policy": {
        "matmul_precision": "highest",
        "matmul_allow_tf32": false,
        "cudnn_allow_tf32": true,
        "cudnn_benchmark": false,
        "cudnn_deterministic": false
      }
    },
    {
      "input": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/missing_operator_480x832_tp1sp4_v1/operators/dit_rank0_ef1fe591ff44862bb6352212f3f2ae5b15b2c8407175dfb2d8e1bd607c7824b4.pt",
      "input_sha256": "d89811d1f0f1ab9382656380f7ff354f19e521d815cd71b6e39b7d2a72def4ec",
      "stage": "dit",
      "op": "aten.linear.default",
      "input_tensor": {
        "shape": [
          1,
          1170,
          13824
        ],
        "stride": [
          16174080,
          13824,
          1
        ],
        "dtype": "torch.bfloat16",
        "logical_bytes": 32348160
      },
      "output_tensor": {
        "shape": [
          1,
          1170,
          5120
        ],
        "stride": [
          5990400,
          5120,
          1
        ],
        "dtype": "torch.bfloat16",
        "logical_bytes": 11980800
      },
      "shape_parameters": {
        "M": 1170,
        "N": 5120,
        "K": 13824
      },
      "mac_flops": 165622579200,
      "bias_additions": 5990400,
      "formula": "2*M*N*K (multiply and add each count one); bias additions listed separately",
      "profiled_kernel_count": 1,
      "summed_profiled_kernel_us": 220.60799999999998,
      "dram_read_bytes": 206323712.0,
      "dram_write_bytes": 10662912.0,
      "dram_total_bytes": 216986624.0,
      "nominal_mac_tflops_per_s": 750.755091383812,
      "nominal_mac_flops_per_dram_byte": 763.2847414594552,
      "ncu_output": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/ncu_missing_ef1fe591ff44862b_v1",
      "ncu_metrics_sha256": "7288e1b9653aef916cb445b78fc52edb136cb42f7c4fa49aaf8bb7effa0a5b72",
      "tensor_policy": {
        "matmul_precision": "highest",
        "matmul_allow_tf32": false,
        "cudnn_allow_tf32": true,
        "cudnn_benchmark": false,
        "cudnn_deterministic": false
      }
    },
    {
      "input": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/missing_operator_480x832_tp1sp4_v1/operators/dit_rank0_6d01e6f8a09a63ac0dbf9852da9b0037804b2820e8d918766d6045af85973c72.pt",
      "input_sha256": "7de91ad08752fccf4380732c881403df8972bc20e43941ce68d7ee08b8d5d769",
      "stage": "dit",
      "op": "aten.linear.default",
      "input_tensor": {
        "shape": [
          1,
          1170,
          5120
        ],
        "stride": [
          23961600,
          5120,
          1
        ],
        "dtype": "torch.bfloat16",
        "logical_bytes": 11980800
      },
      "output_tensor": {
        "shape": [
          1,
          1170,
          5120
        ],
        "stride": [
          5990400,
          5120,
          1
        ],
        "dtype": "torch.bfloat16",
        "logical_bytes": 11980800
      },
      "shape_parameters": {
        "M": 1170,
        "N": 5120,
        "K": 5120
      },
      "mac_flops": 61341696000,
      "bias_additions": 5990400,
      "formula": "2*M*N*K (multiply and add each count one); bias additions listed separately",
      "profiled_kernel_count": 1,
      "summed_profiled_kernel_us": 85.696,
      "dram_read_bytes": 74979584.0,
      "dram_write_bytes": 9854720.0,
      "dram_total_bytes": 84834304.0,
      "nominal_mac_tflops_per_s": 715.8058252427185,
      "nominal_mac_flops_per_dram_byte": 723.0765516741907,
      "ncu_output": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/ncu_missing_6d01e6f8a09a63ac_v1",
      "ncu_metrics_sha256": "b48f09f385e3618afcf0cd2a3bd14eab33684a19bb27cf1a4514188def9dbadc",
      "tensor_policy": {
        "matmul_precision": "highest",
        "matmul_allow_tf32": false,
        "cudnn_allow_tf32": true,
        "cudnn_benchmark": false,
        "cudnn_deterministic": false
      }
    },
    {
      "input": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/missing_operator_480x832_tp1sp4_v1/operators/vae_rank0_41fea95495c28b56014e1289b62a91b3eac42977f9eb1f98bc80fab2cbac7d12.pt",
      "input_sha256": "7ea1828a641121c7e50c8fec13aa00d78a4f4b0f419741b8b6e74c3fd8f2b955",
      "stage": "vae",
      "op": "aten.conv3d.default",
      "input_tensor": {
        "shape": [
          1,
          192,
          6,
          242,
          106
        ],
        "stride": [
          29551104,
          153912,
          25652,
          106,
          1
        ],
        "dtype": "torch.float32",
        "logical_bytes": 118204416
      },
      "output_tensor": {
        "shape": [
          1,
          192,
          4,
          240,
          104
        ],
        "stride": [
          19169280,
          99840,
          24960,
          104,
          1
        ],
        "dtype": "torch.float32",
        "logical_bytes": 76677120
      },
      "shape_parameters": {
        "input": [
          1,
          192,
          6,
          242,
          106
        ],
        "weight": [
          192,
          192,
          3,
          3,
          3
        ],
        "output": [
          1,
          192,
          4,
          240,
          104
        ],
        "stride": [
          1,
          1,
          1
        ],
        "padding": [
          0,
          0,
          0
        ],
        "dilation": [
          1,
          1,
          1
        ],
        "groups": 1
      },
      "mac_flops": 198747095040,
      "bias_additions": 19169280,
      "formula": "2*output_elements*(Cin/groups)*Kd*Kh*Kw; padding already materialized in saved input, implicit padding zero; bias additions separate",
      "profiled_kernel_count": 5,
      "summed_profiled_kernel_us": 872.4159999999999,
      "dram_read_bytes": 520401920.0,
      "dram_write_bytes": 277813248.0,
      "dram_total_bytes": 798215168.0,
      "nominal_mac_tflops_per_s": 227.81229945347175,
      "nominal_mac_flops_per_dram_byte": 248.9893740530874,
      "ncu_output": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/ncu_missing_41fea95495c28b56_v1",
      "ncu_metrics_sha256": "c94cb12a6416d045456155d33b733aa1da369b185e3a198c1660a54c127f1ddf",
      "tensor_policy": {
        "matmul_precision": "highest",
        "matmul_allow_tf32": false,
        "cudnn_allow_tf32": true,
        "cudnn_benchmark": false,
        "cudnn_deterministic": false
      }
    },
    {
      "input": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/missing_operator_480x832_tp1sp4_v1/operators/vae_rank0_74aff8f080d200c8e9e314b4c2b370fd0aaa6e07060d40720d63f5b26a1710b2.pt",
      "input_sha256": "3da5719db48ceca956fb363d6dbfa423964bbde96b9ce8b8dd9399261d9405da",
      "stage": "vae",
      "op": "aten.conv3d.default",
      "input_tensor": {
        "shape": [
          1,
          96,
          6,
          482,
          210
        ],
        "stride": [
          58302720,
          607320,
          101220,
          210,
          1
        ],
        "dtype": "torch.float32",
        "logical_bytes": 233210880
      },
      "output_tensor": {
        "shape": [
          1,
          96,
          4,
          480,
          208
        ],
        "stride": [
          38338560,
          399360,
          99840,
          208,
          1
        ],
        "dtype": "torch.float32",
        "logical_bytes": 153354240
      },
      "shape_parameters": {
        "input": [
          1,
          96,
          6,
          482,
          210
        ],
        "weight": [
          96,
          96,
          3,
          3,
          3
        ],
        "output": [
          1,
          96,
          4,
          480,
          208
        ],
        "stride": [
          1,
          1,
          1
        ],
        "padding": [
          0,
          0,
          0
        ],
        "dilation": [
          1,
          1,
          1
        ],
        "groups": 1
      },
      "mac_flops": 198747095040,
      "bias_additions": 38338560,
      "formula": "2*output_elements*(Cin/groups)*Kd*Kh*Kw; padding already materialized in saved input, implicit padding zero; bias additions separate",
      "profiled_kernel_count": 5,
      "summed_profiled_kernel_us": 1672.16,
      "dram_read_bytes": 1008533760.0,
      "dram_write_bytes": 622219264.0,
      "dram_total_bytes": 1630753024.0,
      "nominal_mac_tflops_per_s": 118.85650598028897,
      "nominal_mac_flops_per_dram_byte": 121.87442985848482,
      "ncu_output": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/ncu_missing_74aff8f080d200c8_v1",
      "ncu_metrics_sha256": "e88a2c4abf33df6793433b2805a6160ca9ad2a1b98f30016b5ec378239163b93",
      "tensor_policy": {
        "matmul_precision": "highest",
        "matmul_allow_tf32": false,
        "cudnn_allow_tf32": true,
        "cudnn_benchmark": false,
        "cudnn_deterministic": false
      }
    },
    {
      "input": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/missing_operator_480x832_tp1sp4_v1/operators/vae_rank0_7fc47a1ee259ec1d5074d37dbe0236e362a2ed59c1f4b5e25f7aaee2322a4ad9.pt",
      "input_sha256": "538cab189b8721f3989e60d331a5e8f3a0b3ea6e3f542b4f5be45a3f1807f370",
      "stage": "vae",
      "op": "aten.conv3d.default",
      "input_tensor": {
        "shape": [
          1,
          384,
          4,
          122,
          54
        ],
        "stride": [
          10119168,
          26352,
          6588,
          54,
          1
        ],
        "dtype": "torch.float32",
        "logical_bytes": 40476672
      },
      "output_tensor": {
        "shape": [
          1,
          384,
          2,
          120,
          52
        ],
        "stride": [
          4792320,
          12480,
          6240,
          52,
          1
        ],
        "dtype": "torch.float32",
        "logical_bytes": 19169280
      },
      "shape_parameters": {
        "input": [
          1,
          384,
          4,
          122,
          54
        ],
        "weight": [
          384,
          384,
          3,
          3,
          3
        ],
        "output": [
          1,
          384,
          2,
          120,
          52
        ],
        "stride": [
          1,
          1,
          1
        ],
        "padding": [
          0,
          0,
          0
        ],
        "dilation": [
          1,
          1,
          1
        ],
        "groups": 1
      },
      "mac_flops": 99373547520,
      "bias_additions": 4792320,
      "formula": "2*output_elements*(Cin/groups)*Kd*Kh*Kw; padding already materialized in saved input, implicit padding zero; bias additions separate",
      "profiled_kernel_count": 5,
      "summed_profiled_kernel_us": 486.23999999999995,
      "dram_read_bytes": 208113664.0,
      "dram_write_bytes": 41915904.0,
      "dram_total_bytes": 250029568.0,
      "nominal_mac_tflops_per_s": 204.37139585389932,
      "nominal_mac_flops_per_dram_byte": 397.44718320674775,
      "ncu_output": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/ncu_missing_7fc47a1ee259ec1d_v1",
      "ncu_metrics_sha256": "c56332cf4b9d241312b9b44c5073a49d50f7891aedc91607ce1ff55ae26f221b",
      "tensor_policy": {
        "matmul_precision": "highest",
        "matmul_allow_tf32": false,
        "cudnn_allow_tf32": true,
        "cudnn_benchmark": false,
        "cudnn_deterministic": false
      }
    },
    {
      "input": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/missing_operator_480x832_tp1sp4_v1/operators/vae_rank0_e49e1cfe69d26b07424554feee2fa507bb297d6abecb47d5ae727bd3f1d68d78.pt",
      "input_sha256": "6fa08b3e20b8b3a0f9f34750ff5afbc4a19f18369bd70ccbf0ed0d1071ca5f08",
      "stage": "vae",
      "op": "aten.conv3d.default",
      "input_tensor": {
        "shape": [
          1,
          192,
          4,
          122,
          54
        ],
        "stride": [
          5059584,
          26352,
          6588,
          54,
          1
        ],
        "dtype": "torch.float32",
        "logical_bytes": 20238336
      },
      "output_tensor": {
        "shape": [
          1,
          384,
          2,
          120,
          52
        ],
        "stride": [
          4792320,
          12480,
          6240,
          52,
          1
        ],
        "dtype": "torch.float32",
        "logical_bytes": 19169280
      },
      "shape_parameters": {
        "input": [
          1,
          192,
          4,
          122,
          54
        ],
        "weight": [
          384,
          192,
          3,
          3,
          3
        ],
        "output": [
          1,
          384,
          2,
          120,
          52
        ],
        "stride": [
          1,
          1,
          1
        ],
        "padding": [
          0,
          0,
          0
        ],
        "dilation": [
          1,
          1,
          1
        ],
        "groups": 1
      },
      "mac_flops": 49686773760,
      "bias_additions": 4792320,
      "formula": "2*output_elements*(Cin/groups)*Kd*Kh*Kw; padding already materialized in saved input, implicit padding zero; bias additions separate",
      "profiled_kernel_count": 5,
      "summed_profiled_kernel_us": 253.02399999999997,
      "dram_read_bytes": 105263360.0,
      "dram_write_bytes": 18538752.0,
      "dram_total_bytes": 123802112.0,
      "nominal_mac_tflops_per_s": 196.3717819653472,
      "nominal_mac_flops_per_dram_byte": 401.340275681242,
      "ncu_output": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/ncu_missing_e49e1cfe69d26b07_v1",
      "ncu_metrics_sha256": "430a85d9d1d993854321e6bd2ab791581ce2262a73d999e2ead2d5ef446254af",
      "tensor_policy": {
        "matmul_precision": "highest",
        "matmul_allow_tf32": false,
        "cudnn_allow_tf32": true,
        "cudnn_benchmark": false,
        "cudnn_deterministic": false
      }
    },
    {
      "input": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/missing_operator_480x832_tp1sp4_v1/operators/vae_rank0_c87dc674a51efcf2c90504bcf0ff5a5065a9e71f70ec401bf34738e9648472e4.pt",
      "input_sha256": "3547fbf5eb18cf7da7abff35f3a785e4dd1cfc51a243c23698a162ff25086677",
      "stage": "vae",
      "op": "aten.conv3d.default",
      "input_tensor": {
        "shape": [
          1,
          384,
          4,
          120,
          52
        ],
        "stride": [
          9584640,
          24960,
          6240,
          52,
          1
        ],
        "dtype": "torch.float32",
        "logical_bytes": 38338560
      },
      "output_tensor": {
        "shape": [
          1,
          768,
          2,
          120,
          52
        ],
        "stride": [
          9584640,
          12480,
          6240,
          52,
          1
        ],
        "dtype": "torch.float32",
        "logical_bytes": 38338560
      },
      "shape_parameters": {
        "input": [
          1,
          384,
          4,
          120,
          52
        ],
        "weight": [
          768,
          384,
          3,
          1,
          1
        ],
        "output": [
          1,
          768,
          2,
          120,
          52
        ],
        "stride": [
          1,
          1,
          1
        ],
        "padding": [
          0,
          0,
          0
        ],
        "dilation": [
          1,
          1,
          1
        ],
        "groups": 1
      },
      "mac_flops": 22083010560,
      "bias_additions": 9584640,
      "formula": "2*output_elements*(Cin/groups)*Kd*Kh*Kw; padding already materialized in saved input, implicit padding zero; bias additions separate",
      "profiled_kernel_count": 6,
      "summed_profiled_kernel_us": 168.224,
      "dram_read_bytes": 178596096.0,
      "dram_write_bytes": 81244416.0,
      "dram_total_bytes": 259840512.0,
      "nominal_mac_tflops_per_s": 131.27146281148944,
      "nominal_mac_flops_per_dram_byte": 84.986788203373,
      "ncu_output": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/ncu_missing_c87dc674a51efcf2_v1",
      "ncu_metrics_sha256": "8b7ac04408b03cb6445e5e687b2ec244e5bb81c30e766768b69d3f57ae977ef5",
      "tensor_policy": {
        "matmul_precision": "highest",
        "matmul_allow_tf32": false,
        "cudnn_allow_tf32": true,
        "cudnn_benchmark": false,
        "cudnn_deterministic": false
      }
    },
    {
      "input": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/missing_operator_480x832_tp1sp4_v1/operators/vae_rank0_27b8a5265a15159078a3d3c8f3e73351dd15f85f461ec2ccec0cb050c22a50c7.pt",
      "input_sha256": "3558a52a268a76189298b60a739c4bc3a01099db809b69c10633a6c964ffd111",
      "stage": "vae",
      "op": "aten.conv3d.default",
      "input_tensor": {
        "shape": [
          1,
          96,
          6,
          482,
          210
        ],
        "stride": [
          58302720,
          607320,
          101220,
          210,
          1
        ],
        "dtype": "torch.float32",
        "logical_bytes": 233210880
      },
      "output_tensor": {
        "shape": [
          1,
          3,
          4,
          480,
          208
        ],
        "stride": [
          1198080,
          399360,
          99840,
          208,
          1
        ],
        "dtype": "torch.float32",
        "logical_bytes": 4792320
      },
      "shape_parameters": {
        "input": [
          1,
          96,
          6,
          482,
          210
        ],
        "weight": [
          3,
          96,
          3,
          3,
          3
        ],
        "output": [
          1,
          3,
          4,
          480,
          208
        ],
        "stride": [
          1,
          1,
          1
        ],
        "padding": [
          0,
          0,
          0
        ],
        "dilation": [
          1,
          1,
          1
        ],
        "groups": 1
      },
      "mac_flops": 6210846720,
      "bias_additions": 1198080,
      "formula": "2*output_elements*(Cin/groups)*Kd*Kh*Kw; padding already materialized in saved input, implicit padding zero; bias additions separate",
      "profiled_kernel_count": 5,
      "summed_profiled_kernel_us": 704.128,
      "dram_read_bytes": 766645248.0,
      "dram_write_bytes": 222150656.0,
      "dram_total_bytes": 988795904.0,
      "nominal_mac_tflops_per_s": 8.82062170514452,
      "nominal_mac_flops_per_dram_byte": 6.281222135806906,
      "ncu_output": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/ncu_missing_27b8a5265a151590_v1",
      "ncu_metrics_sha256": "aa5929288bfd8257714b295ee3f0e82d7c093f32a9919b1378d00c535256eab9",
      "tensor_policy": {
        "matmul_precision": "highest",
        "matmul_allow_tf32": false,
        "cudnn_allow_tf32": true,
        "cudnn_benchmark": false,
        "cudnn_deterministic": false
      }
    },
    {
      "input": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/missing_operator_480x832_tp1sp4_v1/operators/dit_rank0_301783745952cfe05062a5d817409de46ddabe1c22e93fb925a1eda1ffc10892.pt",
      "input_sha256": "34457f152c46e24a2e317b09c8ddc97428430b2e02f99cc2ddeb74c8c157ed9e",
      "stage": "dit",
      "op": "aten.linear.default",
      "input_tensor": {
        "shape": [
          1,
          4680,
          5120
        ],
        "stride": [
          23961600,
          5120,
          1
        ],
        "dtype": "torch.bfloat16",
        "logical_bytes": 47923200
      },
      "output_tensor": {
        "shape": [
          1,
          4680,
          64
        ],
        "stride": [
          299520,
          64,
          1
        ],
        "dtype": "torch.bfloat16",
        "logical_bytes": 599040
      },
      "shape_parameters": {
        "M": 4680,
        "N": 64,
        "K": 5120
      },
      "mac_flops": 3067084800,
      "bias_additions": 299520,
      "formula": "2*M*N*K (multiply and add each count one); bias additions listed separately",
      "profiled_kernel_count": 1,
      "summed_profiled_kernel_us": 17.6,
      "dram_read_bytes": 48598016.0,
      "dram_write_bytes": 3023104.0,
      "dram_total_bytes": 51621120.0,
      "nominal_mac_tflops_per_s": 174.26618181818182,
      "nominal_mac_flops_per_dram_byte": 59.41530908279402,
      "ncu_output": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/ncu_missing_301783745952cfe0_v1",
      "ncu_metrics_sha256": "08ca6ae18d06e5584d4f754370f2141d95de2694d3c930c640c2ce62701483fd",
      "tensor_policy": {
        "matmul_precision": "highest",
        "matmul_allow_tf32": false,
        "cudnn_allow_tf32": true,
        "cudnn_benchmark": false,
        "cudnn_deterministic": false
      }
    },
    {
      "input": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/missing_operator_480x832_tp1sp4_v1/operators/vae_rank0_d35bb120bedeb7d0451dfb63719ddcc3395ccc69ee6fbd9aae8cb2c4a15c4317.pt",
      "input_sha256": "03221b37d95f7c65a91055eae771f796fbf8e0d7758ce6840701ea0c066358ea",
      "stage": "vae",
      "op": "aten.conv3d.default",
      "input_tensor": {
        "shape": [
          1,
          384,
          3,
          60,
          26
        ],
        "stride": [
          1797120,
          4680,
          1560,
          26,
          1
        ],
        "dtype": "torch.float32",
        "logical_bytes": 7188480
      },
      "output_tensor": {
        "shape": [
          1,
          768,
          1,
          60,
          26
        ],
        "stride": [
          1198080,
          1560,
          1560,
          26,
          1
        ],
        "dtype": "torch.float32",
        "logical_bytes": 4792320
      },
      "shape_parameters": {
        "input": [
          1,
          384,
          3,
          60,
          26
        ],
        "weight": [
          768,
          384,
          3,
          1,
          1
        ],
        "output": [
          1,
          768,
          1,
          60,
          26
        ],
        "stride": [
          1,
          1,
          1
        ],
        "padding": [
          0,
          0,
          0
        ],
        "dilation": [
          1,
          1,
          1
        ],
        "groups": 1
      },
      "mac_flops": 2760376320,
      "bias_additions": 1198080,
      "formula": "2*output_elements*(Cin/groups)*Kd*Kh*Kw; padding already materialized in saved input, implicit padding zero; bias additions separate",
      "profiled_kernel_count": 5,
      "summed_profiled_kernel_us": 51.711999999999996,
      "dram_read_bytes": 31155712.0,
      "dram_write_bytes": 1792.0,
      "dram_total_bytes": 31157504.0,
      "nominal_mac_tflops_per_s": 53.379801980198025,
      "nominal_mac_flops_per_dram_byte": 88.59426993895275,
      "ncu_output": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/ncu_missing_d35bb120bedeb7d0_v1",
      "ncu_metrics_sha256": "f40bc37db6e1b54e537917e6ff1078ae1a57d0bb921ebf469b6fb5130244a517",
      "tensor_policy": {
        "matmul_precision": "highest",
        "matmul_allow_tf32": false,
        "cudnn_allow_tf32": true,
        "cudnn_benchmark": false,
        "cudnn_deterministic": false
      }
    },
    {
      "input": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/missing_operator_480x832_tp1sp4_v1/operators/vae_rank0_6a92f0b3792a465a048d27644b5b28b8b2fc97b523bf50d8ed01619a250e09f1.pt",
      "input_sha256": "5da1ee3cce399169645cbf03383f0c7594f9d9b878d0bc38e2e663bb1e06edd8",
      "stage": "vae",
      "op": "aten.conv3d.default",
      "input_tensor": {
        "shape": [
          1,
          192,
          2,
          120,
          52
        ],
        "stride": [
          2396160,
          6240,
          1198080,
          52,
          1
        ],
        "dtype": "torch.float32",
        "logical_bytes": 9584640
      },
      "output_tensor": {
        "shape": [
          1,
          384,
          2,
          120,
          52
        ],
        "stride": [
          4792320,
          12480,
          6240,
          52,
          1
        ],
        "dtype": "torch.float32",
        "logical_bytes": 19169280
      },
      "shape_parameters": {
        "input": [
          1,
          192,
          2,
          120,
          52
        ],
        "weight": [
          384,
          192,
          1,
          1,
          1
        ],
        "output": [
          1,
          384,
          2,
          120,
          52
        ],
        "stride": [
          1,
          1,
          1
        ],
        "padding": [
          0,
          0,
          0
        ],
        "dilation": [
          1,
          1,
          1
        ],
        "groups": 1
      },
      "mac_flops": 1840250880,
      "bias_additions": 4792320,
      "formula": "2*output_elements*(Cin/groups)*Kd*Kh*Kw; padding already materialized in saved input, implicit padding zero; bias additions separate",
      "profiled_kernel_count": 3,
      "summed_profiled_kernel_us": 43.936,
      "dram_read_bytes": 38689280.0,
      "dram_write_bytes": 2311424.0,
      "dram_total_bytes": 41000704.0,
      "nominal_mac_tflops_per_s": 41.88480699198834,
      "nominal_mac_flops_per_dram_byte": 44.883397124107915,
      "ncu_output": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/ncu_missing_6a92f0b3792a465a_v1",
      "ncu_metrics_sha256": "127bc53381dd110a9af457e5b5bd55ea04be1f7d55642390aa1398d86f4b1b29",
      "tensor_policy": {
        "matmul_precision": "highest",
        "matmul_allow_tf32": false,
        "cudnn_allow_tf32": true,
        "cudnn_benchmark": false,
        "cudnn_deterministic": false
      }
    },
    {
      "input": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/missing_operator_480x832_tp1sp4_v1/operators/vae_rank0_ddb198f9c3730e1d5aedce8162e1380e9af61aebde3fde5888ccc05d604c69ef.pt",
      "input_sha256": "480f421d034402c9b8b81d025dce79d2061ec150ba4509115415f8dd5d3b37b5",
      "stage": "vae",
      "op": "aten.conv3d.default",
      "input_tensor": {
        "shape": [
          1,
          96,
          3,
          482,
          210
        ],
        "stride": [
          29151360,
          303660,
          101220,
          210,
          1
        ],
        "dtype": "torch.float32",
        "logical_bytes": 116605440
      },
      "output_tensor": {
        "shape": [
          1,
          3,
          1,
          480,
          208
        ],
        "stride": [
          299520,
          99840,
          99840,
          208,
          1
        ],
        "dtype": "torch.float32",
        "logical_bytes": 1198080
      },
      "shape_parameters": {
        "input": [
          1,
          96,
          3,
          482,
          210
        ],
        "weight": [
          3,
          96,
          3,
          3,
          3
        ],
        "output": [
          1,
          3,
          1,
          480,
          208
        ],
        "stride": [
          1,
          1,
          1
        ],
        "padding": [
          0,
          0,
          0
        ],
        "dilation": [
          1,
          1,
          1
        ],
        "groups": 1
      },
      "mac_flops": 1552711680,
      "bias_additions": 299520,
      "formula": "2*output_elements*(Cin/groups)*Kd*Kh*Kw; padding already materialized in saved input, implicit padding zero; bias additions separate",
      "profiled_kernel_count": 5,
      "summed_profiled_kernel_us": 284.0,
      "dram_read_bytes": 236945920.0,
      "dram_write_bytes": 103383552.0,
      "dram_total_bytes": 340329472.0,
      "nominal_mac_tflops_per_s": 5.467294647887323,
      "nominal_mac_flops_per_dram_byte": 4.562377953561424,
      "ncu_output": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/ncu_missing_ddb198f9c3730e1d_v1",
      "ncu_metrics_sha256": "01f2062c537480f807ac36f5686f41e50a309a38501119f779274fc8d8958fef",
      "tensor_policy": {
        "matmul_precision": "highest",
        "matmul_allow_tf32": false,
        "cudnn_allow_tf32": true,
        "cudnn_benchmark": false,
        "cudnn_deterministic": false
      }
    }
  ],
  "scope": "13 additional saved operators only. FLOPs count nominal dense multiply-add work including explicitly materialized input padding; not executed instruction FLOPs. Rate divides by summed NCU kernel duration (including layout and bias kernels), not unprofiled operator latency. DRAM traffic sums each profiled kernel, reflects replay cache/writeback behavior, not required logical bytes. Bias additions separate; dtype conversion FLOPs undefined. No local bandwidth ceiling or full-model MFU claimed; no mapping from this isolated op to all integrated calls implied."
}