onnx-web-models / webgpu_bench.json
zouinkhim's picture
Model card: both graph layouts + browser WebGPU benchmark (Apple GPU, real EM)
e743c69 verified
Raw History Blame Contribute Delete
15 kB
[
{
"_device": "Measured in Chrome 154 on macOS, Apple GPU (WebGPU `metal-3`), onnxruntime-web 1.27.0-dev.20260506; best of 3 runs after one warm-up run."
},
{
"m": "setup_54",
"n": 148,
"v": "fp32",
"out": 60,
"mb": 42.9,
"first_ms": 2727,
"run_ms": 1203,
"vs_torch_max": 0.001,
"pos": 0,
"torch_pos": 0
},
{
"m": "setup_54",
"n": 148,
"v": "fp16",
"out": 60,
"mb": 21.5,
"first_ms": 3505,
"run_ms": 905,
"vs_torch_max": 0.082,
"pos": 0,
"torch_pos": 0,
"vs_fp32_max": 0.137,
"flip_pct": 0.0
},
{
"m": "setup_54",
"n": 148,
"v": "int8w",
"out": 60,
"mb": 10.9,
"first_ms": 1269,
"run_ms": 1216,
"vs_torch_max": 0.106,
"pos": 0,
"torch_pos": 0,
"vs_fp32_max": 0.183,
"flip_pct": 0.0
},
{
"m": "setup_54",
"n": 148,
"v": "int8w_fp16",
"out": 60,
"mb": 10.9,
"first_ms": 885,
"run_ms": 879,
"vs_torch_max": 0.153,
"pos": 0,
"torch_pos": 0,
"vs_fp32_max": 0.193,
"flip_pct": 0.0
},
{
"m": "setup_54",
"n": 132,
"v": "fp32",
"out": 44,
"mb": 42.9,
"first_ms": 1580,
"run_ms": 796,
"vs_torch_max": 0.0,
"pos": 0,
"torch_pos": 0
},
{
"m": "setup_54",
"n": 132,
"v": "fp16",
"out": 44,
"mb": 21.5,
"first_ms": 2357,
"run_ms": 573,
"vs_torch_max": 0.072,
"pos": 0,
"torch_pos": 0,
"vs_fp32_max": 0.126,
"flip_pct": 0.0
},
{
"m": "setup_54",
"n": 132,
"v": "int8w",
"out": 44,
"mb": 10.9,
"first_ms": 4422,
"run_ms": 1096,
"vs_torch_max": 0.096,
"pos": 0,
"torch_pos": 0,
"vs_fp32_max": 0.252,
"flip_pct": 0.0
},
{
"m": "setup_54",
"n": 132,
"v": "int8w_fp16",
"out": 44,
"mb": 10.9,
"first_ms": 588,
"run_ms": 558,
"vs_torch_max": 0.09,
"pos": 0,
"torch_pos": 0,
"vs_fp32_max": 0.191,
"flip_pct": 0.0
},
{
"m": "setup_54",
"n": 116,
"v": "fp32",
"out": 28,
"mb": 42.9,
"first_ms": 543,
"run_ms": 516,
"vs_torch_max": 0.001,
"pos": 0,
"torch_pos": 0
},
{
"m": "setup_54",
"n": 116,
"v": "fp16",
"out": 28,
"mb": 21.5,
"first_ms": 367,
"run_ms": 323,
"vs_torch_max": 0.048,
"pos": 0,
"torch_pos": 0,
"vs_fp32_max": 0.098,
"flip_pct": 0.0
},
{
"m": "setup_54",
"n": 116,
"v": "int8w",
"out": 28,
"mb": 10.9,
"first_ms": 568,
"run_ms": 531,
"vs_torch_max": 0.103,
"pos": 0,
"torch_pos": 0,
"vs_fp32_max": 0.167,
"flip_pct": 0.0
},
{
"m": "setup_54",
"n": 116,
"v": "int8w_fp16",
"out": 28,
"mb": 10.9,
"first_ms": 367,
"run_ms": 328,
"vs_torch_max": 0.101,
"pos": 0,
"torch_pos": 0,
"vs_fp32_max": 0.185,
"flip_pct": 0.0
},
{
"m": "setup_53",
"n": 172,
"v": "fp32",
"out": 84,
"mb": 22.5,
"first_ms": 1174,
"run_ms": 1137,
"vs_torch_max": 0.001,
"pos": 0,
"torch_pos": 0
},
{
"m": "setup_53",
"n": 172,
"v": "fp16",
"out": 84,
"mb": 11.3,
"first_ms": 780,
"run_ms": 741,
"vs_torch_max": 0.071,
"pos": 0,
"torch_pos": 0,
"vs_fp32_max": 0.188,
"flip_pct": 0.0
},
{
"m": "setup_53",
"n": 172,
"v": "int8w",
"out": 84,
"mb": 5.8,
"first_ms": 1121,
"run_ms": 1013,
"vs_torch_max": 0.208,
"pos": 0,
"torch_pos": 0,
"vs_fp32_max": 0.433,
"flip_pct": 0.0
},
{
"m": "setup_53",
"n": 172,
"v": "int8w_fp16",
"out": 84,
"mb": 5.8,
"first_ms": 729,
"run_ms": 737,
"vs_torch_max": 0.166,
"pos": 0,
"torch_pos": 0,
"vs_fp32_max": 0.384,
"flip_pct": 0.0
},
{
"m": "setup_53",
"n": 140,
"v": "fp32",
"out": 52,
"mb": 22.5,
"first_ms": 528,
"run_ms": 533,
"vs_torch_max": 0.0,
"pos": 0,
"torch_pos": 0
},
{
"m": "setup_53",
"n": 140,
"v": "fp16",
"out": 52,
"mb": 11.3,
"first_ms": 392,
"run_ms": 336,
"vs_torch_max": 0.076,
"pos": 0,
"torch_pos": 0,
"vs_fp32_max": 0.14,
"flip_pct": 0.0
},
{
"m": "setup_53",
"n": 140,
"v": "int8w",
"out": 52,
"mb": 5.8,
"first_ms": 487,
"run_ms": 494,
"vs_torch_max": 0.268,
"pos": 0,
"torch_pos": 0,
"vs_fp32_max": 0.433,
"flip_pct": 0.0
},
{
"m": "setup_53",
"n": 140,
"v": "int8w_fp16",
"out": 52,
"mb": 5.8,
"first_ms": 341,
"run_ms": 328,
"vs_torch_max": 0.225,
"pos": 0,
"torch_pos": 0,
"vs_fp32_max": 0.384,
"flip_pct": 0.0
},
{
"m": "setup_53",
"n": 124,
"v": "fp32",
"out": 36,
"mb": 22.5,
"first_ms": 354,
"run_ms": 348,
"vs_torch_max": 0.0,
"pos": 0,
"torch_pos": 0
},
{
"m": "setup_53",
"n": 124,
"v": "fp16",
"out": 36,
"mb": 11.3,
"first_ms": 209,
"run_ms": 157,
"vs_torch_max": 0.073,
"pos": 0,
"torch_pos": 0,
"vs_fp32_max": 0.118,
"flip_pct": 0.0
},
{
"m": "setup_53",
"n": 124,
"v": "int8w",
"out": 36,
"mb": 5.8,
"first_ms": 328,
"run_ms": 317,
"vs_torch_max": 0.194,
"pos": 0,
"torch_pos": 0,
"vs_fp32_max": 0.285,
"flip_pct": 0.0
},
{
"m": "setup_53",
"n": 124,
"v": "int8w_fp16",
"out": 36,
"mb": 5.8,
"first_ms": 186,
"run_ms": 160,
"vs_torch_max": 0.201,
"pos": 0,
"torch_pos": 0,
"vs_fp32_max": 0.384,
"flip_pct": 0.0
},
{
"m": "setup_56",
"n": 148,
"v": "fp32",
"out": 60,
"mb": 42.9,
"first_ms": 1216,
"run_ms": 1208,
"vs_torch_max": 0.001,
"pos": 0,
"torch_pos": 0
},
{
"m": "setup_56",
"n": 148,
"v": "fp16",
"out": 60,
"mb": 21.5,
"first_ms": 915,
"run_ms": 921,
"vs_torch_max": 0.2,
"pos": 0,
"torch_pos": 0,
"vs_fp32_max": 0.543,
"flip_pct": 0.0
},
{
"m": "setup_56",
"n": 148,
"v": "int8w",
"out": 60,
"mb": 10.9,
"first_ms": 1244,
"run_ms": 1219,
"vs_torch_max": 0.379,
"pos": 0,
"torch_pos": 0,
"vs_fp32_max": 0.945,
"flip_pct": 0.0
},
{
"m": "setup_56",
"n": 148,
"v": "int8w_fp16",
"out": 60,
"mb": 10.9,
"first_ms": 881,
"run_ms": 880,
"vs_torch_max": 0.34,
"pos": 0,
"torch_pos": 0,
"vs_fp32_max": 1.011,
"flip_pct": 0.0
},
{
"m": "setup_56",
"n": 132,
"v": "fp32",
"out": 44,
"mb": 42.9,
"first_ms": 826,
"run_ms": 811,
"vs_torch_max": 0.001,
"pos": 0,
"torch_pos": 0
},
{
"m": "setup_56",
"n": 132,
"v": "fp16",
"out": 44,
"mb": 21.5,
"first_ms": 616,
"run_ms": 574,
"vs_torch_max": 0.16,
"pos": 0,
"torch_pos": 0,
"vs_fp32_max": 0.404,
"flip_pct": 0.0
},
{
"m": "setup_56",
"n": 132,
"v": "int8w",
"out": 44,
"mb": 10.9,
"first_ms": 816,
"run_ms": 822,
"vs_torch_max": 0.447,
"pos": 0,
"torch_pos": 0,
"vs_fp32_max": 0.646,
"flip_pct": 0.0
},
{
"m": "setup_56",
"n": 132,
"v": "int8w_fp16",
"out": 44,
"mb": 10.9,
"first_ms": 586,
"run_ms": 559,
"vs_torch_max": 0.266,
"pos": 0,
"torch_pos": 0,
"vs_fp32_max": 0.627,
"flip_pct": 0.0
},
{
"m": "setup_56",
"n": 116,
"v": "fp32",
"out": 28,
"mb": 42.9,
"first_ms": 523,
"run_ms": 527,
"vs_torch_max": 0.001,
"pos": 0,
"torch_pos": 0
},
{
"m": "setup_56",
"n": 116,
"v": "fp16",
"out": 28,
"mb": 21.5,
"first_ms": 454,
"run_ms": 324,
"vs_torch_max": 0.119,
"pos": 0,
"torch_pos": 0,
"vs_fp32_max": 0.268,
"flip_pct": 0.0
},
{
"m": "setup_56",
"n": 116,
"v": "int8w",
"out": 28,
"mb": 10.9,
"first_ms": 536,
"run_ms": 526,
"vs_torch_max": 0.342,
"pos": 0,
"torch_pos": 0,
"vs_fp32_max": 0.533,
"flip_pct": 0.0
},
{
"m": "setup_56",
"n": 116,
"v": "int8w_fp16",
"out": 28,
"mb": 10.9,
"first_ms": 370,
"run_ms": 326,
"vs_torch_max": 0.332,
"pos": 0,
"torch_pos": 0,
"vs_fp32_max": 0.531,
"flip_pct": 0.0
},
{
"m": "setup_55",
"n": 172,
"v": "fp32",
"out": 84,
"mb": 22.5,
"first_ms": 1135,
"run_ms": 1080,
"vs_torch_max": 0.001,
"pos": 342,
"torch_pos": 342
},
{
"m": "setup_55",
"n": 172,
"v": "fp16",
"out": 84,
"mb": 11.3,
"first_ms": 758,
"run_ms": 758,
"vs_torch_max": 0.117,
"pos": 342,
"torch_pos": 342,
"vs_fp32_max": 0.269,
"flip_pct": 0.0
},
{
"m": "setup_55",
"n": 172,
"v": "int8w",
"out": 84,
"mb": 5.8,
"first_ms": 998,
"run_ms": 989,
"vs_torch_max": 0.558,
"pos": 342,
"torch_pos": 342,
"vs_fp32_max": 1.121,
"flip_pct": 0.0
},
{
"m": "setup_55",
"n": 172,
"v": "int8w_fp16",
"out": 84,
"mb": 5.8,
"first_ms": 729,
"run_ms": 736,
"vs_torch_max": 0.65,
"pos": 342,
"torch_pos": 342,
"vs_fp32_max": 0.971,
"flip_pct": 0.0
},
{
"m": "setup_55",
"n": 140,
"v": "fp32",
"out": 52,
"mb": 22.5,
"first_ms": 537,
"run_ms": 506,
"vs_torch_max": 0.001,
"pos": 0,
"torch_pos": 0
},
{
"m": "setup_55",
"n": 140,
"v": "fp16",
"out": 52,
"mb": 11.3,
"first_ms": 342,
"run_ms": 319,
"vs_torch_max": 0.131,
"pos": 0,
"torch_pos": 0,
"vs_fp32_max": 0.269,
"flip_pct": 0.0
},
{
"m": "setup_55",
"n": 140,
"v": "int8w",
"out": 52,
"mb": 5.8,
"first_ms": 490,
"run_ms": 481,
"vs_torch_max": 0.531,
"pos": 0,
"torch_pos": 0,
"vs_fp32_max": 1.059,
"flip_pct": 0.0
},
{
"m": "setup_55",
"n": 140,
"v": "int8w_fp16",
"out": 52,
"mb": 5.8,
"first_ms": 5997,
"run_ms": 336,
"vs_torch_max": 0.512,
"pos": 0,
"torch_pos": 0,
"vs_fp32_max": 0.971,
"flip_pct": 0.0
},
{
"m": "setup_55",
"n": 124,
"v": "fp32",
"out": 36,
"mb": 22.5,
"first_ms": 415,
"run_ms": 331,
"vs_torch_max": 0.001,
"pos": 0,
"torch_pos": 0
},
{
"m": "setup_55",
"n": 124,
"v": "fp16",
"out": 36,
"mb": 11.3,
"first_ms": 862,
"run_ms": 156,
"vs_torch_max": 0.103,
"pos": 0,
"torch_pos": 0,
"vs_fp32_max": 0.269,
"flip_pct": 0.0
},
{
"m": "setup_55",
"n": 124,
"v": "int8w",
"out": 36,
"mb": 5.8,
"first_ms": 421,
"run_ms": 315,
"vs_torch_max": 0.314,
"pos": 0,
"torch_pos": 0,
"vs_fp32_max": 0.785,
"flip_pct": 0.0
},
{
"m": "setup_55",
"n": 124,
"v": "int8w_fp16",
"out": 36,
"mb": 5.8,
"first_ms": 308,
"run_ms": 160,
"vs_torch_max": 0.402,
"pos": 0,
"torch_pos": 0,
"vs_fp32_max": 0.611,
"flip_pct": 0.0
},
{
"m": "setup_61",
"n": 148,
"v": "fp32",
"out": 60,
"mb": 42.9,
"first_ms": 1237,
"run_ms": 1220,
"vs_torch_max": 0.001,
"pos": 2526,
"torch_pos": 2526
},
{
"m": "setup_61",
"n": 148,
"v": "fp16",
"out": 60,
"mb": 21.5,
"first_ms": 918,
"run_ms": 900,
"vs_torch_max": 0.124,
"pos": 2526,
"torch_pos": 2526,
"vs_fp32_max": 0.249,
"flip_pct": 0.0019
},
{
"m": "setup_61",
"n": 148,
"v": "int8w",
"out": 60,
"mb": 10.9,
"first_ms": 1321,
"run_ms": 1224,
"vs_torch_max": 0.328,
"pos": 2533,
"torch_pos": 2526,
"vs_fp32_max": 0.584,
"flip_pct": 0.0032
},
{
"m": "setup_61",
"n": 148,
"v": "int8w_fp16",
"out": 60,
"mb": 10.9,
"first_ms": 896,
"run_ms": 869,
"vs_torch_max": 0.362,
"pos": 2534,
"torch_pos": 2526,
"vs_fp32_max": 0.791,
"flip_pct": 0.0037
},
{
"m": "setup_61",
"n": 132,
"v": "fp32",
"out": 44,
"mb": 42.9,
"first_ms": 811,
"run_ms": 809,
"vs_torch_max": 0.001,
"pos": 62,
"torch_pos": 62
},
{
"m": "setup_61",
"n": 132,
"v": "fp16",
"out": 44,
"mb": 21.5,
"first_ms": 608,
"run_ms": 574,
"vs_torch_max": 0.164,
"pos": 62,
"torch_pos": 62,
"vs_fp32_max": 0.251,
"flip_pct": 0.0
},
{
"m": "setup_61",
"n": 132,
"v": "int8w",
"out": 44,
"mb": 10.9,
"first_ms": 812,
"run_ms": 808,
"vs_torch_max": 0.407,
"pos": 62,
"torch_pos": 62,
"vs_fp32_max": 0.612,
"flip_pct": 0.0
},
{
"m": "setup_61",
"n": 132,
"v": "int8w_fp16",
"out": 44,
"mb": 10.9,
"first_ms": 588,
"run_ms": 555,
"vs_torch_max": 0.482,
"pos": 62,
"torch_pos": 62,
"vs_fp32_max": 0.803,
"flip_pct": 0.0
},
{
"m": "setup_61",
"n": 116,
"v": "fp32",
"out": 28,
"mb": 42.9,
"first_ms": 530,
"run_ms": 514,
"vs_torch_max": 0.001,
"pos": 0,
"torch_pos": 0
},
{
"m": "setup_61",
"n": 116,
"v": "fp16",
"out": 28,
"mb": 21.5,
"first_ms": 429,
"run_ms": 322,
"vs_torch_max": 0.145,
"pos": 0,
"torch_pos": 0,
"vs_fp32_max": 0.28,
"flip_pct": 0.0
},
{
"m": "setup_61",
"n": 116,
"v": "int8w",
"out": 28,
"mb": 10.9,
"first_ms": 534,
"run_ms": 525,
"vs_torch_max": 0.303,
"pos": 0,
"torch_pos": 0,
"vs_fp32_max": 0.554,
"flip_pct": 0.0
},
{
"m": "setup_61",
"n": 116,
"v": "int8w_fp16",
"out": 28,
"mb": 10.9,
"first_ms": 352,
"run_ms": 327,
"vs_torch_max": 0.434,
"pos": 0,
"torch_pos": 0,
"vs_fp32_max": 0.613,
"flip_pct": 0.0
},
{
"m": "setup_67",
"n": 148,
"v": "fp32",
"out": 60,
"mb": 42.9,
"first_ms": 1252,
"run_ms": 1233,
"vs_torch_max": 0.0,
"pos": 38086,
"torch_pos": 38086
},
{
"m": "setup_67",
"n": 148,
"v": "fp16",
"out": 60,
"mb": 21.5,
"first_ms": 917,
"run_ms": 890,
"vs_torch_max": 0.03,
"pos": 38088,
"torch_pos": 38086,
"vs_fp32_max": 0.061,
"flip_pct": 0.0148
},
{
"m": "setup_67",
"n": 148,
"v": "int8w",
"out": 60,
"mb": 10.9,
"first_ms": 1391,
"run_ms": 1221,
"vs_torch_max": 0.042,
"pos": 38121,
"torch_pos": 38086,
"vs_fp32_max": 0.109,
"flip_pct": 0.0282
},
{
"m": "setup_67",
"n": 148,
"v": "int8w_fp16",
"out": 60,
"mb": 10.9,
"first_ms": 885,
"run_ms": 870,
"vs_torch_max": 0.048,
"pos": 38135,
"torch_pos": 38086,
"vs_fp32_max": 0.099,
"flip_pct": 0.0356
},
{
"m": "setup_67",
"n": 132,
"v": "fp32",
"out": 44,
"mb": 42.9,
"first_ms": 2401,
"run_ms": 809,
"vs_torch_max": 0.0,
"pos": 16728,
"torch_pos": 16728
},
{
"m": "setup_67",
"n": 132,
"v": "fp16",
"out": 44,
"mb": 21.5,
"first_ms": 608,
"run_ms": 569,
"vs_torch_max": 0.035,
"pos": 16729,
"torch_pos": 16728,
"vs_fp32_max": 0.071,
"flip_pct": 0.0129
},
{
"m": "setup_67",
"n": 132,
"v": "int8w",
"out": 44,
"mb": 10.9,
"first_ms": 824,
"run_ms": 796,
"vs_torch_max": 0.037,
"pos": 16746,
"torch_pos": 16728,
"vs_fp32_max": 0.083,
"flip_pct": 0.0399
},
{
"m": "setup_67",
"n": 132,
"v": "int8w_fp16",
"out": 44,
"mb": 10.9,
"first_ms": 582,
"run_ms": 556,
"vs_torch_max": 0.047,
"pos": 16753,
"torch_pos": 16728,
"vs_fp32_max": 0.088,
"flip_pct": 0.0434
},
{
"m": "setup_67",
"n": 116,
"v": "fp32",
"out": 28,
"mb": 42.9,
"first_ms": 529,
"run_ms": 532,
"vs_torch_max": 0.0,
"pos": 6514,
"torch_pos": 6514
},
{
"m": "setup_67",
"n": 116,
"v": "fp16",
"out": 28,
"mb": 21.5,
"first_ms": 375,
"run_ms": 323,
"vs_torch_max": 0.039,
"pos": 6523,
"torch_pos": 6514,
"vs_fp32_max": 0.058,
"flip_pct": 0.0501
},
{
"m": "setup_67",
"n": 116,
"v": "int8w",
"out": 28,
"mb": 10.9,
"first_ms": 551,
"run_ms": 529,
"vs_torch_max": 0.043,
"pos": 6527,
"torch_pos": 6514,
"vs_fp32_max": 0.068,
"flip_pct": 0.0866
},
{
"m": "setup_67",
"n": 116,
"v": "int8w_fp16",
"out": 28,
"mb": 10.9,
"first_ms": 343,
"run_ms": 327,
"vs_torch_max": 0.066,
"pos": 6528,
"torch_pos": 6514,
"vs_fp32_max": 0.128,
"flip_pct": 0.0729
}
]