{
  "execution_mode": "fresh Metal tensor execution; eager versus MLX compiled expression",
  "environment": {
    "python": "3.12.8",
    "platform": "macOS-27.0.1-arm64-arm-64bit",
    "mlx": "0.32.4"
  },
  "source_sha256": "62653ea247e893a6c1b2a70709fa0e9a363e1b2ce25e6b522c0c8308fd230a05",
  "cases": [
    {
      "rows": 1,
      "width": 896,
      "dtype": "float32",
      "max_abs_error": 0.0,
      "timings": {
        "eager": {
          "median_ms": 0.45025,
          "samples_ms": [
            0.507958,
            0.464666,
            0.450875,
            0.45325,
            0.449625,
            0.503083,
            0.501917,
            0.456292,
            0.472334,
            0.473333,
            0.516667,
            0.401541,
            0.393125,
            0.401959,
            0.222625,
            0.180541,
            0.174334,
            0.180542,
            0.175917,
            0.1745
          ]
        },
        "compiled": {
          "median_ms": 0.290083,
          "samples_ms": [
            0.174417,
            0.1545,
            0.160917,
            0.161667,
            0.161125,
            0.412792,
            0.464209,
            0.464167,
            0.509875,
            0.450125,
            0.459958,
            0.473375,
            0.533375,
            0.470042,
            0.38675,
            0.187459,
            0.175958,
            0.176375,
            0.183375,
            0.193416
          ]
        }
      }
    },
    {
      "rows": 32,
      "width": 896,
      "dtype": "float32",
      "max_abs_error": 0.0,
      "timings": {
        "eager": {
          "median_ms": 0.222667,
          "samples_ms": [
            0.254083,
            0.229667,
            0.22275,
            0.194417,
            0.196792,
            0.204542,
            0.21225,
            0.204917,
            0.198125,
            0.216666,
            0.216125,
            0.470042,
            0.483125,
            0.5025,
            0.475292,
            0.515,
            0.21125,
            0.225208,
            0.222583,
            0.223292
          ]
        },
        "compiled": {
          "median_ms": 0.192312,
          "samples_ms": [
            0.194792,
            0.200791,
            0.196542,
            0.181,
            0.189917,
            0.192041,
            0.200792,
            0.20125,
            0.195417,
            0.192083,
            0.187625,
            0.178333,
            0.178083,
            0.179417,
            0.183041,
            0.196666,
            0.192083,
            0.192541,
            0.212208,
            0.23525
          ]
        }
      }
    },
    {
      "rows": 32,
      "width": 1003,
      "dtype": "float32",
      "max_abs_error": 0.0,
      "timings": {
        "eager": {
          "median_ms": 0.228791,
          "samples_ms": [
            0.269583,
            0.224083,
            0.249292,
            0.234417,
            0.23475,
            0.229583,
            0.2335,
            0.231584,
            0.232292,
            0.226375,
            0.227375,
            0.224875,
            0.222291,
            0.216875,
            0.228,
            0.222334,
            0.235542,
            0.223666,
            0.230167,
            0.223459
          ]
        },
        "compiled": {
          "median_ms": 0.196583,
          "samples_ms": [
            0.196291,
            0.208083,
            0.211208,
            0.214792,
            0.203625,
            0.191208,
            0.196875,
            0.181042,
            0.190583,
            0.187208,
            0.190167,
            0.192334,
            0.194334,
            0.192125,
            0.246042,
            0.191875,
            0.231625,
            0.269167,
            0.23475,
            0.208583
          ]
        }
      }
    }
  ],
  "limits": "Expression compilation may fuse operations. No kernel trace or measured HBM byte count; no whole-model, CUDA or Triton speedup claim."
}