devin-lai's picture
Release v0.4.0: selective MLX model downloads
3da88e3 verified
Raw History Blame Contribute Delete
23.6 kB
{
"platform": "macOS-27.0.1-arm64-arm-64bit",
"coremltools": "9.0",
"repeats": 5,
"scope": "Real checkpoint projection weights; synthetic inputs. Synchronous predict includes NumPy I/O. Component tests, not full image latency.",
"trials": [
{
"case": "dit-fc2",
"precision": "fp16",
"compute_units": "CPU_ONLY",
"shape": [
4181,
14336
],
"weight_shape": [
5376,
14336
],
"package_bytes": 154153164,
"load_compile_s": 0.1675274579320103,
"placement": [
{
"operator": "const",
"preferred": null,
"supported": [],
"estimated_cost": null
},
{
"operator": "const",
"preferred": null,
"supported": [],
"estimated_cost": null
},
{
"operator": "ios18.linear",
"preferred": "MLCPUComputeDevice",
"supported": [
"MLCPUComputeDevice"
],
"estimated_cost": 1.0
}
],
"first_predict_ms": 239.70012506470084,
"samples_ms": [
194.02279192581773,
193.32108297385275,
193.83937492966652,
192.32758390717208,
193.28666618093848
],
"median_ms": 193.32108297385275,
"relative_rmse": 0.008997025899589062,
"max_abs": 1.7035293579101562
},
{
"case": "dit-fc2",
"precision": "fp16",
"compute_units": "CPU_AND_GPU",
"shape": [
4181,
14336
],
"weight_shape": [
5376,
14336
],
"package_bytes": 154153164,
"load_compile_s": 0.13135279202833772,
"placement": [
{
"operator": "const",
"preferred": null,
"supported": [],
"estimated_cost": null
},
{
"operator": "const",
"preferred": null,
"supported": [],
"estimated_cost": null
},
{
"operator": "ios18.linear",
"preferred": "MLGPUComputeDevice",
"supported": [
"MLCPUComputeDevice",
"MLGPUComputeDevice"
],
"estimated_cost": 1.0
}
],
"first_predict_ms": 274.62662500329316,
"samples_ms": [
55.83895789459348,
50.863665994256735,
54.42062485963106,
52.25870804861188,
54.61216694675386
],
"median_ms": 54.42062485963106,
"relative_rmse": 0.00020757669699378312,
"max_abs": 0.0461883544921875
},
{
"case": "dit-fc2",
"precision": "fp16",
"compute_units": "CPU_AND_NE",
"shape": [
4181,
14336
],
"weight_shape": [
5376,
14336
],
"package_bytes": 154153164,
"load_compile_s": 2.5629583750851452,
"placement": [
{
"operator": "const",
"preferred": null,
"supported": [],
"estimated_cost": null
},
{
"operator": "const",
"preferred": null,
"supported": [],
"estimated_cost": null
},
{
"operator": "ios18.linear",
"preferred": "MLNeuralEngineComputeDevice",
"supported": [
"MLCPUComputeDevice",
"MLNeuralEngineComputeDevice"
],
"estimated_cost": 1.0
}
],
"first_predict_ms": 285.8013748191297,
"samples_ms": [
272.20479189418256,
272.3927500192076,
272.29520911350846,
272.19158411026,
272.3133750259876
],
"median_ms": 272.29520911350846,
"relative_rmse": 0.00020931176550220698,
"max_abs": 0.0461883544921875
},
{
"case": "dit-fc2",
"precision": "fp16",
"compute_units": "ALL",
"shape": [
4181,
14336
],
"weight_shape": [
5376,
14336
],
"package_bytes": 154153164,
"load_compile_s": 0.13452245900407434,
"placement": [
{
"operator": "const",
"preferred": null,
"supported": [],
"estimated_cost": null
},
{
"operator": "const",
"preferred": null,
"supported": [],
"estimated_cost": null
},
{
"operator": "ios18.linear",
"preferred": "MLGPUComputeDevice",
"supported": [
"MLCPUComputeDevice",
"MLGPUComputeDevice",
"MLNeuralEngineComputeDevice"
],
"estimated_cost": 1.0
}
],
"first_predict_ms": 75.36920811980963,
"samples_ms": [
54.255584022030234,
53.10550006106496,
53.56983304955065,
57.02450009994209,
53.382958052679896
],
"median_ms": 53.56983304955065,
"relative_rmse": 0.00020757669699378312,
"max_abs": 0.0461883544921875
},
{
"case": "dit-fc2",
"precision": "fp32",
"compute_units": "CPU_ONLY",
"shape": [
4181,
14336
],
"weight_shape": [
5376,
14336
],
"package_bytes": 308304589,
"load_compile_s": 0.6216194580774754,
"placement": [
{
"operator": "const",
"preferred": null,
"supported": [],
"estimated_cost": null
},
{
"operator": "const",
"preferred": null,
"supported": [],
"estimated_cost": null
},
{
"operator": "ios18.linear",
"preferred": "MLCPUComputeDevice",
"supported": [
"MLCPUComputeDevice"
],
"estimated_cost": 1.0
}
],
"first_predict_ms": 391.0232500638813,
"samples_ms": [
366.8390829116106,
368.16129088401794,
370.6123339943588,
371.06195790693164,
371.9222079962492
],
"median_ms": 370.6123339943588,
"relative_rmse": 1.42448004680773e-06,
"max_abs": 0.00049591064453125
},
{
"case": "dit-fc2",
"precision": "fp32",
"compute_units": "CPU_AND_GPU",
"shape": [
4181,
14336
],
"weight_shape": [
5376,
14336
],
"package_bytes": 308304589,
"load_compile_s": 0.25257754093036056,
"placement": [
{
"operator": "const",
"preferred": null,
"supported": [],
"estimated_cost": null
},
{
"operator": "const",
"preferred": null,
"supported": [],
"estimated_cost": null
},
{
"operator": "ios18.linear",
"preferred": "MLGPUComputeDevice",
"supported": [
"MLCPUComputeDevice",
"MLGPUComputeDevice"
],
"estimated_cost": 1.0
}
],
"first_predict_ms": 259.089709026739,
"samples_ms": [
125.88245910592377,
126.40920793637633,
125.82262489013374,
126.03725004009902,
126.47708388976753
],
"median_ms": 126.03725004009902,
"relative_rmse": 1.3215966419011238e-06,
"max_abs": 0.00048065185546875
},
{
"case": "dit-fc2",
"precision": "fp32",
"compute_units": "CPU_AND_NE",
"shape": [
4181,
14336
],
"weight_shape": [
5376,
14336
],
"package_bytes": 308304589,
"load_compile_s": 0.8115240000188351,
"placement": [
{
"operator": "const",
"preferred": null,
"supported": [],
"estimated_cost": null
},
{
"operator": "const",
"preferred": null,
"supported": [],
"estimated_cost": null
},
{
"operator": "ios18.linear",
"preferred": "MLCPUComputeDevice",
"supported": [
"MLCPUComputeDevice"
],
"estimated_cost": 1.0
}
],
"first_predict_ms": 393.7107918318361,
"samples_ms": [
366.1310840398073,
370.0570832006633,
368.28945809975266,
367.6325420383364,
367.7522500511259
],
"median_ms": 367.7522500511259,
"relative_rmse": 1.42448004680773e-06,
"max_abs": 0.00049591064453125
},
{
"case": "dit-fc2",
"precision": "fp32",
"compute_units": "ALL",
"shape": [
4181,
14336
],
"weight_shape": [
5376,
14336
],
"package_bytes": 308304589,
"load_compile_s": 0.23633408383466303,
"placement": [
{
"operator": "const",
"preferred": null,
"supported": [],
"estimated_cost": null
},
{
"operator": "const",
"preferred": null,
"supported": [],
"estimated_cost": null
},
{
"operator": "ios18.linear",
"preferred": "MLGPUComputeDevice",
"supported": [
"MLCPUComputeDevice",
"MLGPUComputeDevice"
],
"estimated_cost": 1.0
}
],
"first_predict_ms": 155.00358305871487,
"samples_ms": [
126.47862522862852,
126.53029197826982,
126.49054219946265,
126.28233293071389,
125.85870805196464
],
"median_ms": 126.47862522862852,
"relative_rmse": 1.3215966419011238e-06,
"max_abs": 0.00048065185546875
},
{
"case": "dit-fc2-conv256",
"precision": "fp16",
"compute_units": "CPU_ONLY",
"shape": [
1,
14336,
1,
256
],
"weight_shape": [
5376,
14336
],
"package_bytes": 154142986,
"load_compile_s": 0.25310612516477704,
"placement": [
{
"operator": "const",
"preferred": null,
"supported": [],
"estimated_cost": null
},
{
"operator": "const",
"preferred": null,
"supported": [],
"estimated_cost": null
},
{
"operator": "const",
"preferred": null,
"supported": [],
"estimated_cost": null
},
{
"operator": "const",
"preferred": null,
"supported": [],
"estimated_cost": null
},
{
"operator": "const",
"preferred": null,
"supported": [],
"estimated_cost": null
},
{
"operator": "const",
"preferred": null,
"supported": [],
"estimated_cost": null
},
{
"operator": "ios18.conv",
"preferred": "MLCPUComputeDevice",
"supported": [
"MLCPUComputeDevice"
],
"estimated_cost": 1.0
}
],
"first_predict_ms": 24.463333189487457,
"samples_ms": [
13.804416870698333,
13.714332832023501,
15.785250114277005,
14.088458148762584,
14.467166969552636
],
"median_ms": 14.088458148762584,
"relative_rmse": 0.017367275431752205,
"max_abs": 4.4668121337890625
},
{
"case": "dit-fc2-conv256",
"precision": "fp16",
"compute_units": "CPU_AND_GPU",
"shape": [
1,
14336,
1,
256
],
"weight_shape": [
5376,
14336
],
"package_bytes": 154142986,
"load_compile_s": 0.15895758289843798,
"placement": [
{
"operator": "const",
"preferred": null,
"supported": [],
"estimated_cost": null
},
{
"operator": "const",
"preferred": null,
"supported": [],
"estimated_cost": null
},
{
"operator": "const",
"preferred": null,
"supported": [],
"estimated_cost": null
},
{
"operator": "const",
"preferred": null,
"supported": [],
"estimated_cost": null
},
{
"operator": "const",
"preferred": null,
"supported": [],
"estimated_cost": null
},
{
"operator": "const",
"preferred": null,
"supported": [],
"estimated_cost": null
},
{
"operator": "ios18.conv",
"preferred": "MLGPUComputeDevice",
"supported": [
"MLCPUComputeDevice",
"MLGPUComputeDevice"
],
"estimated_cost": 1.0
}
],
"first_predict_ms": 229.41312519833446,
"samples_ms": [
4.823666997253895,
4.999083932489157,
4.665416898205876,
3.7090000696480274,
3.94016713835299
],
"median_ms": 4.665416898205876,
"relative_rmse": 0.0002074961521429941,
"max_abs": 0.0461883544921875
},
{
"case": "dit-fc2-conv256",
"precision": "fp16",
"compute_units": "CPU_AND_NE",
"shape": [
1,
14336,
1,
256
],
"weight_shape": [
5376,
14336
],
"package_bytes": 154142986,
"load_compile_s": 2.3588687090668827,
"placement": [
{
"operator": "const",
"preferred": null,
"supported": [],
"estimated_cost": null
},
{
"operator": "const",
"preferred": null,
"supported": [],
"estimated_cost": null
},
{
"operator": "const",
"preferred": null,
"supported": [],
"estimated_cost": null
},
{
"operator": "const",
"preferred": null,
"supported": [],
"estimated_cost": null
},
{
"operator": "const",
"preferred": null,
"supported": [],
"estimated_cost": null
},
{
"operator": "const",
"preferred": null,
"supported": [],
"estimated_cost": null
},
{
"operator": "ios18.conv",
"preferred": "MLNeuralEngineComputeDevice",
"supported": [
"MLCPUComputeDevice",
"MLNeuralEngineComputeDevice"
],
"estimated_cost": 1.0
}
],
"first_predict_ms": 18.719791900366545,
"samples_ms": [
17.255791928619146,
17.22654211334884,
17.179290996864438,
17.18783308751881,
17.248707823455334
],
"median_ms": 17.22654211334884,
"relative_rmse": 0.0002092272334266454,
"max_abs": 0.0461883544921875
},
{
"case": "dit-fc2-conv256",
"precision": "fp16",
"compute_units": "ALL",
"shape": [
1,
14336,
1,
256
],
"weight_shape": [
5376,
14336
],
"package_bytes": 154142986,
"load_compile_s": 0.15283204219304025,
"placement": [
{
"operator": "const",
"preferred": null,
"supported": [],
"estimated_cost": null
},
{
"operator": "const",
"preferred": null,
"supported": [],
"estimated_cost": null
},
{
"operator": "const",
"preferred": null,
"supported": [],
"estimated_cost": null
},
{
"operator": "const",
"preferred": null,
"supported": [],
"estimated_cost": null
},
{
"operator": "const",
"preferred": null,
"supported": [],
"estimated_cost": null
},
{
"operator": "const",
"preferred": null,
"supported": [],
"estimated_cost": null
},
{
"operator": "ios18.conv",
"preferred": "MLGPUComputeDevice",
"supported": [
"MLCPUComputeDevice",
"MLGPUComputeDevice",
"MLNeuralEngineComputeDevice"
],
"estimated_cost": 1.0
}
],
"first_predict_ms": 13.4526661131531,
"samples_ms": [
4.598583094775677,
4.914749879390001,
4.920917097479105,
3.971958067268133,
4.048625007271767
],
"median_ms": 4.598583094775677,
"relative_rmse": 0.0002074961521429941,
"max_abs": 0.0461883544921875
},
{
"case": "dit-fc2-conv256",
"precision": "fp32",
"compute_units": "CPU_ONLY",
"shape": [
1,
14336,
1,
256
],
"weight_shape": [
5376,
14336
],
"package_bytes": 308283658,
"load_compile_s": 0.6159360830206424,
"placement": [
{
"operator": "const",
"preferred": null,
"supported": [],
"estimated_cost": null
},
{
"operator": "const",
"preferred": null,
"supported": [],
"estimated_cost": null
},
{
"operator": "const",
"preferred": null,
"supported": [],
"estimated_cost": null
},
{
"operator": "const",
"preferred": null,
"supported": [],
"estimated_cost": null
},
{
"operator": "const",
"preferred": null,
"supported": [],
"estimated_cost": null
},
{
"operator": "const",
"preferred": null,
"supported": [],
"estimated_cost": null
},
{
"operator": "ios18.conv",
"preferred": "MLCPUComputeDevice",
"supported": [
"MLCPUComputeDevice"
],
"estimated_cost": 1.0
}
],
"first_predict_ms": 83.27387506142259,
"samples_ms": [
65.84020797163248,
65.9275830257684,
66.90237508155406,
67.3022500704974,
67.99295800738037
],
"median_ms": 66.90237508155406,
"relative_rmse": 0.0,
"max_abs": 0.0
},
{
"case": "dit-fc2-conv256",
"precision": "fp32",
"compute_units": "CPU_AND_GPU",
"shape": [
1,
14336,
1,
256
],
"weight_shape": [
5376,
14336
],
"package_bytes": 308283658,
"load_compile_s": 0.27439770800992846,
"placement": [
{
"operator": "const",
"preferred": null,
"supported": [],
"estimated_cost": null
},
{
"operator": "const",
"preferred": null,
"supported": [],
"estimated_cost": null
},
{
"operator": "const",
"preferred": null,
"supported": [],
"estimated_cost": null
},
{
"operator": "const",
"preferred": null,
"supported": [],
"estimated_cost": null
},
{
"operator": "const",
"preferred": null,
"supported": [],
"estimated_cost": null
},
{
"operator": "const",
"preferred": null,
"supported": [],
"estimated_cost": null
},
{
"operator": "ios18.conv",
"preferred": "MLGPUComputeDevice",
"supported": [
"MLCPUComputeDevice",
"MLGPUComputeDevice"
],
"estimated_cost": 1.0
}
],
"first_predict_ms": 100.21312511526048,
"samples_ms": [
12.018541805446148,
11.276375036686659,
11.430124985054135,
11.256833095103502,
10.951624950394034
],
"median_ms": 11.276375036686659,
"relative_rmse": 0.0,
"max_abs": 0.0
},
{
"case": "dit-fc2-conv256",
"precision": "fp32",
"compute_units": "CPU_AND_NE",
"shape": [
1,
14336,
1,
256
],
"weight_shape": [
5376,
14336
],
"package_bytes": 308283658,
"load_compile_s": 0.6161469579674304,
"placement": [
{
"operator": "const",
"preferred": null,
"supported": [],
"estimated_cost": null
},
{
"operator": "const",
"preferred": null,
"supported": [],
"estimated_cost": null
},
{
"operator": "const",
"preferred": null,
"supported": [],
"estimated_cost": null
},
{
"operator": "const",
"preferred": null,
"supported": [],
"estimated_cost": null
},
{
"operator": "const",
"preferred": null,
"supported": [],
"estimated_cost": null
},
{
"operator": "const",
"preferred": null,
"supported": [],
"estimated_cost": null
},
{
"operator": "ios18.conv",
"preferred": "MLCPUComputeDevice",
"supported": [
"MLCPUComputeDevice"
],
"estimated_cost": 1.0
}
],
"first_predict_ms": 82.68562494777143,
"samples_ms": [
67.91662494651973,
69.08479100093246,
66.73616706393659,
66.93325005471706,
67.00825016014278
],
"median_ms": 67.00825016014278,
"relative_rmse": 0.0,
"max_abs": 0.0
},
{
"case": "dit-fc2-conv256",
"precision": "fp32",
"compute_units": "ALL",
"shape": [
1,
14336,
1,
256
],
"weight_shape": [
5376,
14336
],
"package_bytes": 308283658,
"load_compile_s": 0.27939695795066655,
"placement": [
{
"operator": "const",
"preferred": null,
"supported": [],
"estimated_cost": null
},
{
"operator": "const",
"preferred": null,
"supported": [],
"estimated_cost": null
},
{
"operator": "const",
"preferred": null,
"supported": [],
"estimated_cost": null
},
{
"operator": "const",
"preferred": null,
"supported": [],
"estimated_cost": null
},
{
"operator": "const",
"preferred": null,
"supported": [],
"estimated_cost": null
},
{
"operator": "const",
"preferred": null,
"supported": [],
"estimated_cost": null
},
{
"operator": "ios18.conv",
"preferred": "MLGPUComputeDevice",
"supported": [
"MLCPUComputeDevice",
"MLGPUComputeDevice"
],
"estimated_cost": 1.0
}
],
"first_predict_ms": 22.786707850173116,
"samples_ms": [
11.762959184125066,
11.550666997209191,
10.938999941572547,
11.943457880988717,
11.289791204035282
],
"median_ms": 11.550666997209191,
"relative_rmse": 0.0,
"max_abs": 0.0
}
]
}