{
  "scenario": "F3",
  "backend": "cudnn",
  "sequence": 41778,
  "batch": 1,
  "heads": 28,
  "head_dim": 128,
  "pid": 2815986,
  "torch": "2.11.0+cu130",
  "cuda": "13.0",
  "cudnn": 91900,
  "gpu": "NVIDIA RTX 6000D",
  "capability": [
    12,
    0
  ],
  "started_at": 1787624712.331527,
  "results": [
    {
      "layout": "model_strided",
      "dtype": "torch.bfloat16",
      "q_shape": [
        1,
        28,
        41778,
        128
      ],
      "q_stride": [
        449197056,
        128,
        10752,
        1
      ],
      "k_stride": [
        449197056,
        128,
        10752,
        1
      ],
      "v_stride": [
        449197056,
        128,
        10752,
        1
      ],
      "q_contiguous": false,
      "warm": {
        "samples_ms": [
          175.15554809570312,
          175.17213439941406,
          175.1822967529297,
          175.19020080566406,
          175.20431518554688,
          175.221923828125,
          175.2449951171875,
          175.28250122070312,
          175.2829132080078,
          175.3275909423828,
          175.37689208984375,
          175.50755310058594
        ],
        "min_ms": 175.15554809570312,
        "median_ms": 175.23345947265625,
        "mean_ms": 175.2624053955078,
        "p90_ms": 175.37689208984375,
        "max_ms": 175.50755310058594,
        "stdev_ms": 0.10215529569329801,
        "cv": 0.0005828705560828527
      },
      "cold": {
        "samples_ms": [
          175.13352966308594,
          175.16204833984375,
          175.17669677734375,
          175.19964599609375,
          175.27442932128906
        ],
        "min_ms": 175.13352966308594,
        "median_ms": 175.17669677734375,
        "mean_ms": 175.18927001953125,
        "p90_ms": 175.27442932128906,
        "max_ms": 175.27442932128906,
        "stdev_ms": 0.05330539645821187,
        "cv": 0.00030427318095605416
      },
      "cold_over_warm": 0.9996760738760547,
      "effective_tflops": 142.79277988761325,
      "io_lower_bound_gbps": 6.835788208512291,
      "cuda_graph": {
        "supported": true,
        "samples_ms": [
          173.94415283203125,
          174.21372985839844,
          174.38047790527344,
          174.42230224609375,
          174.53097534179688,
          174.53546142578125,
          174.54541015625,
          174.6174774169922,
          174.6393280029297,
          174.6550750732422,
          174.6965789794922,
          174.7626953125
        ],
        "min_ms": 173.94415283203125,
        "median_ms": 174.54043579101562,
        "mean_ms": 174.49530537923178,
        "p90_ms": 174.6965789794922,
        "max_ms": 174.7626953125,
        "stdev_ms": 0.2296573277480753,
        "cv": 0.001316123246118052
      },
      "output_checksum": 2.151111857529031e-06,
      "trace": "/data/wxy/profile_results/h3-sdpa-kernel-matrix-20260825-run1/jobs/gpu1-F3-cudnn-model_strided.trace.json",
      "trace_error": null
    },
    {
      "layout": "contiguous",
      "dtype": "torch.bfloat16",
      "q_shape": [
        1,
        28,
        41778,
        128
      ],
      "q_stride": [
        149732352,
        5347584,
        128,
        1
      ],
      "k_stride": [
        149732352,
        5347584,
        128,
        1
      ],
      "v_stride": [
        149732352,
        5347584,
        128,
        1
      ],
      "q_contiguous": true,
      "warm": {
        "samples_ms": [
          174.1165771484375,
          174.2126007080078,
          174.32179260253906,
          174.33328247070312,
          174.3406982421875,
          174.38082885742188,
          174.42166137695312,
          174.46243286132812,
          174.48867797851562,
          174.54888916015625,
          174.76480102539062,
          174.82150268554688
        ],
        "min_ms": 174.1165771484375,
        "median_ms": 174.4012451171875,
        "mean_ms": 174.43447875976562,
        "p90_ms": 174.76480102539062,
        "max_ms": 174.82150268554688,
        "stdev_ms": 0.20466748891529848,
        "cv": 0.0011733201507551172
      },
      "cold": {
        "samples_ms": [
          174.0714569091797,
          174.08380126953125,
          174.09657287597656,
          174.12640380859375,
          174.13961791992188
        ],
        "min_ms": 174.0714569091797,
        "median_ms": 174.09657287597656,
        "mean_ms": 174.10357055664062,
        "p90_ms": 174.13961791992188,
        "max_ms": 174.13961791992188,
        "stdev_ms": 0.028687315990781403,
        "cv": 0.00016477155465027433
      },
      "cold_over_warm": 0.9982530386121601,
      "effective_tflops": 143.4741637917242,
      "io_lower_bound_gbps": 6.86840747722362,
      "cuda_graph": {
        "supported": true,
        "samples_ms": [
          174.41177368164062,
          174.41845703125,
          174.42796325683594,
          174.45094299316406,
          174.45269775390625,
          174.455810546875,
          174.51263427734375,
          174.69337463378906,
          174.7523193359375,
          174.7738494873047,
          174.78799438476562,
          174.8524169921875
        ],
        "min_ms": 174.41177368164062,
        "median_ms": 174.48422241210938,
        "mean_ms": 174.58251953125,
        "p90_ms": 174.78799438476562,
        "max_ms": 174.8524169921875,
        "stdev_ms": 0.1726415150932476,
        "cv": 0.0009888820229928292
      },
      "output_checksum": 2.1056450805190252e-06,
      "trace": "/data/wxy/profile_results/h3-sdpa-kernel-matrix-20260825-run1/jobs/gpu1-F3-cudnn-contiguous.trace.json",
      "trace_error": null
    }
  ],
  "status": "success",
  "finished_at": 1787624726.660268
}