{ "platform": "macOS-27.0.1-arm64-arm-64bit", "torch": "2.15.0", "threads": 10, "repeats": 5, "warmup": 2, "scope": "Actual H3 block-0 weights, synthetic activations; model loading and dtype conversion excluded.", "trials": [ { "operator": "attn.qkv_proj", "device": "cpu", "dtype": "torch.bfloat16", "shape": [ 4181, 5376, 21504 ], "median_ms": 3701.2594998814166, "samples_ms": [ 3645.8038750570267, 3700.587541097775, 3701.2594998814166, 3725.67658405751, 3751.146041089669 ], "relative_rmse": 0.0016586730489507318 }, { "operator": "attn.qkv_proj", "device": "cpu", "dtype": "torch.float32", "shape": [ 4181, 5376, 21504 ], "median_ms": 617.0030829962343, "samples_ms": [ 616.0587079357356, 637.2107500210404, 627.203208161518, 617.0030829962343, 613.2661669980735 ], "relative_rmse": 0.0 }, { "operator": "attn.qkv_proj", "device": "mps", "dtype": "torch.bfloat16", "shape": [ 4181, 5376, 21504 ], "median_ms": 38.99029106833041, "samples_ms": [ 38.59824989922345, 38.70695806108415, 38.99029106833041, 39.377457927912474, 40.238582994788885 ], "relative_rmse": 0.0016586730489507318 }, { "operator": "attn.qkv_proj", "device": "mps", "dtype": "torch.float16", "shape": [ 4181, 5376, 21504 ], "median_ms": 41.44991608336568, "samples_ms": [ 40.2035410515964, 40.685457875952125, 41.44991608336568, 41.75474983640015, 41.69837478548288 ], "relative_rmse": 0.0002074193034786731 }, { "operator": "attn.qkv_proj", "device": "mps", "dtype": "torch.float32", "shape": [ 4181, 5376, 21504 ], "median_ms": 203.43483285978436, "samples_ms": [ 186.70570803806186, 197.21666700206697, 204.3533339165151, 206.5801250282675, 203.43483285978436 ], "relative_rmse": 0.0 }, { "operator": "attn.out_proj", "device": "cpu", "dtype": "torch.bfloat16", "shape": [ 4181, 7168, 5376 ], "median_ms": 1214.189250022173, "samples_ms": [ 1221.7448328156024, 1211.8159160017967, 1213.1728339008987, 1214.189250022173, 1216.0887920763344 ], "relative_rmse": 0.0016584296245127916 }, { "operator": "attn.out_proj", "device": "cpu", "dtype": "torch.float32", "shape": [ 4181, 7168, 5376 ], "median_ms": 204.44674999453127, "samples_ms": [ 204.44674999453127, 206.50512515567243, 202.72879186086357, 202.91408407501876, 204.5679169241339 ], "relative_rmse": 0.0 }, { "operator": "attn.out_proj", "device": "mps", "dtype": "torch.bfloat16", "shape": [ 4181, 7168, 5376 ], "median_ms": 13.290167087689042, "samples_ms": [ 13.436208013445139, 13.941709185019135, 13.290167087689042, 13.14554107375443, 13.104083016514778 ], "relative_rmse": 0.0016584297409281135 }, { "operator": "attn.out_proj", "device": "mps", "dtype": "torch.float16", "shape": [ 4181, 7168, 5376 ], "median_ms": 13.19179218262434, "samples_ms": [ 13.19179218262434, 13.081457931548357, 13.279916951432824, 13.22045805864036, 13.152749976143241 ], "relative_rmse": 0.00020729658717755228 }, { "operator": "attn.out_proj", "device": "mps", "dtype": "torch.float32", "shape": [ 4181, 7168, 5376 ], "median_ms": 59.14987483993173, "samples_ms": [ 59.288624906912446, 59.14987483993173, 58.7309580296278, 59.26216603256762, 58.7053750641644 ], "relative_rmse": 0.0 }, { "operator": "mlp.fc1", "device": "cpu", "dtype": "torch.bfloat16", "shape": [ 4181, 5376, 28672 ], "median_ms": 5015.978500014171, "samples_ms": [ 4981.27191606909, 5016.96683303453, 5013.545291963965, 5043.636667076498, 5015.978500014171 ], "relative_rmse": 0.0016574508044868708 }, { "operator": "mlp.fc1", "device": "cpu", "dtype": "torch.float32", "shape": [ 4181, 5376, 28672 ], "median_ms": 781.2532079406083, "samples_ms": [ 775.5420829635113, 781.2532079406083, 811.6851670201868, 786.0322080086917, 775.5197500810027 ], "relative_rmse": 0.0 }, { "operator": "mlp.fc1", "device": "mps", "dtype": "torch.bfloat16", "shape": [ 4181, 5376, 28672 ], "median_ms": 51.55754182487726, "samples_ms": [ 51.55754182487726, 51.43470806069672, 51.36879091151059, 51.687208004295826, 51.64370802231133 ], "relative_rmse": 0.0016574509209021926 }, { "operator": "mlp.fc1", "device": "mps", "dtype": "torch.float16", "shape": [ 4181, 5376, 28672 ], "median_ms": 51.62183311767876, "samples_ms": [ 51.6144591383636, 51.398917101323605, 51.62183311767876, 52.15637502260506, 52.686791867017746 ], "relative_rmse": 0.0002071879425784573 }, { "operator": "mlp.fc1", "device": "mps", "dtype": "torch.float32", "shape": [ 4181, 5376, 28672 ], "median_ms": 232.809916138649, "samples_ms": [ 232.8284999821335, 233.14516711980104, 232.58750000968575, 232.809916138649, 232.6427090447396 ], "relative_rmse": 0.0 }, { "operator": "mlp.fc2", "device": "cpu", "dtype": "torch.bfloat16", "shape": [ 4181, 14336, 5376 ], "median_ms": 2572.5704170763493, "samples_ms": [ 2581.110875122249, 2607.537457952276, 2572.5704170763493, 2570.869208080694, 2538.388082990423 ], "relative_rmse": 0.001660577836446464 }, { "operator": "mlp.fc2", "device": "cpu", "dtype": "torch.float32", "shape": [ 4181, 14336, 5376 ], "median_ms": 393.7346248421818, "samples_ms": [ 394.1986670251936, 387.5164999626577, 393.7346248421818, 407.3841660283506, 387.0072499848902 ], "relative_rmse": 0.0 }, { "operator": "mlp.fc2", "device": "mps", "dtype": "torch.bfloat16", "shape": [ 4181, 14336, 5376 ], "median_ms": 31.237290939316154, "samples_ms": [ 29.555541928857565, 30.560749815776944, 31.237290939316154, 31.85308282263577, 31.76683303900063 ], "relative_rmse": 0.0016605781856924295 }, { "operator": "mlp.fc2", "device": "mps", "dtype": "torch.float16", "shape": [ 4181, 14336, 5376 ], "median_ms": 30.82345798611641, "samples_ms": [ 30.82345798611641, 29.868208104744554, 31.982708955183625, 33.232917077839375, 29.665250098332763 ], "relative_rmse": 0.00020759549806825817 }, { "operator": "mlp.fc2", "device": "mps", "dtype": "torch.float32", "shape": [ 4181, 14336, 5376 ], "median_ms": 119.69162500463426, "samples_ms": [ 119.69162500463426, 119.40862494520843, 119.94224996306002, 119.78904204443097, 119.54558291472495 ], "relative_rmse": 1.3209257758717285e-06 }, { "operator": "mlp", "compiled": false, "device": "mps", "dtype": "torch.bfloat16", "median_ms": 85.36641695536673, "samples_ms": [ 85.36641695536673, 85.00783308409154, 87.21108408644795, 84.90916690789163, 86.90800005570054 ], "equal_fraction": 1.0 }, { "operator": "mlp", "compiled": true, "device": "mps", "dtype": "torch.bfloat16", "median_ms": 83.18099984899163, "samples_ms": [ 82.51995779573917, 83.18099984899163, 84.68637499026954, 82.74875022470951, 83.52845907211304 ], "equal_fraction": 1.0 } ] }