{
  "model": "WaveCut/ERNIE-Image-Turbo-SDNQ-uint4-static",
  "device": "NVIDIA RTX 6000 Ada Generation",
  "torch": "2.8.0+cu128",
  "cases": [
    {
      "run": "warmup",
      "prompt_id": "00-cyrillic-poster",
      "width": 1024,
      "height": 1024,
      "seconds": 42.33458385616541,
      "stage_seconds_sum": 41.638948045670986,
      "unattributed_seconds": 0.6956358104944229,
      "gpu_start_mib": 10172,
      "gpu_end_mib": 15368,
      "torch_peak_allocated_mib": 12175,
      "torch_peak_reserved_mib": 14848,
      "stages": {
        "pe.forward": {
          "seconds": 24.21831013262272,
          "calls": 415
        },
        "text_encoder.forward": {
          "seconds": 0.04355049878358841,
          "calls": 1
        },
        "transformer.forward": {
          "seconds": 17.202288791537285,
          "calls": 8
        },
        "vae.decode": {
          "seconds": 0.1747986227273941,
          "calls": 1
        }
      },
      "use_quantized_matmul": false,
      "image": "/root/ernie-qmm-debug/00-cyrillic-poster_noqmm_warmup.png"
    },
    {
      "run": "hot_repeat",
      "prompt_id": "00-cyrillic-poster",
      "width": 1024,
      "height": 1024,
      "seconds": 24.803174294531345,
      "stage_seconds_sum": 24.33856651186943,
      "unattributed_seconds": 0.4646077826619148,
      "gpu_start_mib": 10278,
      "gpu_end_mib": 15318,
      "torch_peak_allocated_mib": 12175,
      "torch_peak_reserved_mib": 14798,
      "stages": {
        "pe.forward": {
          "seconds": 16.506547823548317,
          "calls": 421
        },
        "text_encoder.forward": {
          "seconds": 0.040486179292201996,
          "calls": 1
        },
        "transformer.forward": {
          "seconds": 7.632116124033928,
          "calls": 8
        },
        "vae.decode": {
          "seconds": 0.15941638499498367,
          "calls": 1
        }
      },
      "use_quantized_matmul": false,
      "image": "/root/ernie-qmm-debug/00-cyrillic-poster_noqmm_hot_repeat.png"
    },
    {
      "run": "warmup",
      "prompt_id": "02-technical-diagram",
      "width": 1200,
      "height": 896,
      "seconds": 94.24524058401585,
      "stage_seconds_sum": 92.17488968372345,
      "unattributed_seconds": 2.0703509002923965,
      "gpu_start_mib": 10278,
      "gpu_end_mib": 15912,
      "torch_peak_allocated_mib": 12259,
      "torch_peak_reserved_mib": 15392,
      "stages": {
        "pe.forward": {
          "seconds": 80.31031914055347,
          "calls": 2048
        },
        "text_encoder.forward": {
          "seconds": 0.11907234787940979,
          "calls": 1
        },
        "transformer.forward": {
          "seconds": 11.57931386679411,
          "calls": 8
        },
        "vae.decode": {
          "seconds": 0.16618432849645615,
          "calls": 1
        }
      },
      "use_quantized_matmul": false,
      "image": "/root/ernie-qmm-debug/02-technical-diagram_noqmm_warmup.png"
    },
    {
      "run": "hot_repeat",
      "prompt_id": "02-technical-diagram",
      "width": 1200,
      "height": 896,
      "seconds": 94.40739246457815,
      "stage_seconds_sum": 92.33396648615599,
      "unattributed_seconds": 2.073425978422165,
      "gpu_start_mib": 10278,
      "gpu_end_mib": 15910,
      "torch_peak_allocated_mib": 12259,
      "torch_peak_reserved_mib": 15390,
      "stages": {
        "pe.forward": {
          "seconds": 80.37759981304407,
          "calls": 2048
        },
        "text_encoder.forward": {
          "seconds": 0.1175723597407341,
          "calls": 1
        },
        "transformer.forward": {
          "seconds": 11.67346066236496,
          "calls": 8
        },
        "vae.decode": {
          "seconds": 0.16533365100622177,
          "calls": 1
        }
      },
      "use_quantized_matmul": false,
      "image": "/root/ernie-qmm-debug/02-technical-diagram_noqmm_hot_repeat.png"
    },
    {
      "run": "warmup",
      "prompt_id": "00-cyrillic-poster",
      "width": 1024,
      "height": 1024,
      "seconds": 47.86481238901615,
      "stage_seconds_sum": 47.40215189009905,
      "unattributed_seconds": 0.4626604989171028,
      "gpu_start_mib": 10280,
      "gpu_end_mib": 15354,
      "torch_peak_allocated_mib": 12175,
      "torch_peak_reserved_mib": 14832,
      "stages": {
        "pe.forward": {
          "seconds": 28.630388408899307,
          "calls": 408
        },
        "text_encoder.forward": {
          "seconds": 3.4749386832118034,
          "calls": 1
        },
        "transformer.forward": {
          "seconds": 15.150532938539982,
          "calls": 8
        },
        "vae.decode": {
          "seconds": 0.14629185944795609,
          "calls": 1
        }
      },
      "use_quantized_matmul": true,
      "image": "/root/ernie-qmm-debug/00-cyrillic-poster_qmm_warmup.png"
    },
    {
      "run": "hot_repeat",
      "prompt_id": "00-cyrillic-poster",
      "width": 1024,
      "height": 1024,
      "seconds": 29.076967030763626,
      "stage_seconds_sum": 28.608746394515038,
      "unattributed_seconds": 0.46822063624858856,
      "gpu_start_mib": 10282,
      "gpu_end_mib": 15334,
      "torch_peak_allocated_mib": 12175,
      "torch_peak_reserved_mib": 14812,
      "stages": {
        "pe.forward": {
          "seconds": 16.936102136969566,
          "calls": 438
        },
        "text_encoder.forward": {
          "seconds": 3.2992821782827377,
          "calls": 1
        },
        "transformer.forward": {
          "seconds": 8.225335031747818,
          "calls": 8
        },
        "vae.decode": {
          "seconds": 0.14802704751491547,
          "calls": 1
        }
      },
      "use_quantized_matmul": true,
      "image": "/root/ernie-qmm-debug/00-cyrillic-poster_qmm_hot_repeat.png"
    },
    {
      "run": "warmup",
      "prompt_id": "02-technical-diagram",
      "width": 1200,
      "height": 896,
      "seconds": 100.24962972849607,
      "stage_seconds_sum": 98.18934270739555,
      "unattributed_seconds": 2.060287021100521,
      "gpu_start_mib": 10282,
      "gpu_end_mib": 15988,
      "torch_peak_allocated_mib": 12259,
      "torch_peak_reserved_mib": 15466,
      "stages": {
        "pe.forward": {
          "seconds": 83.79404803365469,
          "calls": 2048
        },
        "text_encoder.forward": {
          "seconds": 3.445981025695801,
          "calls": 1
        },
        "transformer.forward": {
          "seconds": 10.796071700751781,
          "calls": 8
        },
        "vae.decode": {
          "seconds": 0.15324194729328156,
          "calls": 1
        }
      },
      "use_quantized_matmul": true,
      "image": "/root/ernie-qmm-debug/02-technical-diagram_qmm_warmup.png"
    },
    {
      "run": "hot_repeat",
      "prompt_id": "02-technical-diagram",
      "width": 1200,
      "height": 896,
      "seconds": 38.25821412354708,
      "stage_seconds_sum": 37.577924594283104,
      "unattributed_seconds": 0.6802895292639732,
      "gpu_start_mib": 10282,
      "gpu_end_mib": 15474,
      "torch_peak_allocated_mib": 12244,
      "torch_peak_reserved_mib": 14952,
      "stages": {
        "pe.forward": {
          "seconds": 25.347995460033417,
          "calls": 638
        },
        "text_encoder.forward": {
          "seconds": 3.2810458168387413,
          "calls": 1
        },
        "transformer.forward": {
          "seconds": 8.796003498136997,
          "calls": 8
        },
        "vae.decode": {
          "seconds": 0.15287981927394867,
          "calls": 1
        }
      },
      "use_quantized_matmul": true,
      "image": "/root/ernie-qmm-debug/02-technical-diagram_qmm_hot_repeat.png"
    }
  ],
  "sdnq": "0.1.9"
}