devin-lai's picture
Release v0.4.0: selective MLX model downloads
3da88e3 verified
Raw History Blame Contribute Delete
27.9 kB
{ "environment": { "platform": "macOS-27.0.1-arm64-arm-64bit", "machine": "arm64", "metal": { "device_name": "Apple M5 Pro", "max_recommended_working_set_size": 55662788608, "memory_size": 68719476736, "architecture": "applegpu_g17s", "max_buffer_length": 41747087360, "resource_limit": 499000 }, "versions": { "mlx": "0.32.4", "mlx-metal": "0.32.4", "numpy": "2.5.4", "safetensors": "0.8.0", "transformers": "4.57.6", "torch": "2.15.0", "Pillow": "13.0.0", "scikit-image": "0.26.0" }, "precision_environment": { "MLX_ENABLE_TF32": "0", "MLX_CONV_WINOGRAD": "0" } }, "warmup": 2, "repeats": 5, "scope": "Original block-0 weights; synthetic BF16 inputs. Operator latency only, excludes conversion and loading.", "trials": [ { "operator": "attn.qkv_proj", "variant": "bf16_nt", "shape": [ 4181, 5376, 21504 ], "median_ms": 39.01795903220773, "min_ms": 38.90308388508856, "max_ms": 39.20233412645757, "samples_ms": [ 39.01795903220773, 39.000708842650056, 39.20233412645757, 39.08749995753169, 38.90308388508856 ], "max_abs": 0.0, "relative_rmse": 0.0, "equal_fraction": 1.0 }, { "operator": "attn.qkv_proj", "variant": "bf16_nn", "shape": [ 4181, 5376, 21504 ], "median_ms": 40.58533301576972, "min_ms": 40.212417021393776, "max_ms": 40.690625086426735, "samples_ms": [ 40.212417021393776, 40.641458006575704, 40.690625086426735, 40.58533301576972, 40.2926248498261 ], "max_abs": 0.0, "relative_rmse": 0.0, "equal_fraction": 1.0 }, { "operator": "attn.qkv_proj", "variant": "bf16_pad64", "shape": [ 4181, 5376, 21504 ], "padded_tokens": 4224, "median_ms": 40.12054088525474, "min_ms": 39.86324998550117, "max_ms": 40.58220889419317, "samples_ms": [ 40.12054088525474, 40.07162502966821, 40.58220889419317, 39.86324998550117, 40.49258283339441 ], "max_abs": 0.0, "relative_rmse": 0.0, "equal_fraction": 1.0 }, { "operator": "attn.qkv_proj", "variant": "bf16_pad128", "shape": [ 4181, 5376, 21504 ], "padded_tokens": 4224, "median_ms": 40.043167071416974, "min_ms": 39.82716705650091, "max_ms": 40.46391719020903, "samples_ms": [ 39.82716705650091, 40.04783392883837, 39.89016613923013, 40.46391719020903, 40.043167071416974 ], "max_abs": 0.0, "relative_rmse": 0.0, "equal_fraction": 1.0 }, { "operator": "attn.qkv_proj", "variant": "bf16_pad256", "shape": [ 4181, 5376, 21504 ], "padded_tokens": 4352, "median_ms": 40.20208399742842, "min_ms": 40.15479190275073, "max_ms": 40.87774991057813, "samples_ms": [ 40.20208399742842, 40.200083050876856, 40.15479190275073, 40.59520806185901, 40.87774991057813 ], "max_abs": 0.0, "relative_rmse": 0.0, "equal_fraction": 1.0 }, { "operator": "attn.qkv_proj", "variant": "fp16_round_bf16", "shape": [ 4181, 5376, 21504 ], "median_ms": 40.44137499295175, "min_ms": 40.23237503133714, "max_ms": 40.570707991719246, "samples_ms": [ 40.280082961544394, 40.23237503133714, 40.44137499295175, 40.52645806223154, 40.570707991719246 ], "max_abs": 0.25, "relative_rmse": 0.0014367633266374469, "equal_fraction": 0.9375272393226624 }, { "operator": "attn.qkv_proj", "variant": "mxfp88_g32", "shape": [ 4181, 5376, 21504 ], "weight_gib": 0.11103057861328125, "median_ms": 46.38499999418855, "min_ms": 45.91437499038875, "max_ms": 46.56987520866096, "samples_ms": [ 46.10741604119539, 45.91437499038875, 46.38499999418855, 46.56987520866096, 46.53862491250038 ], "max_abs": 1.796875, "relative_rmse": 0.02671019919216633, "equal_fraction": 0.06673875451087952 }, { "operator": "attn.qkv_proj", "variant": "mxfp44_g32", "shape": [ 4181, 5376, 21504 ], "weight_gib": 0.05719757080078125, "median_ms": 42.059540981426835, "min_ms": 41.91033309325576, "max_ms": 42.56245819851756, "samples_ms": [ 42.16808290220797, 41.91033309325576, 42.007958982139826, 42.059540981426835, 42.56245819851756 ], "max_abs": 9.109375, "relative_rmse": 0.12055830657482147, "equal_fraction": 0.014743390493094921 }, { "operator": "attn.qkv_proj", "variant": "nvfp44_g16", "shape": [ 4181, 5376, 21504 ], "weight_gib": 0.0605621337890625, "median_ms": 43.3145419228822, "min_ms": 42.84545802511275, "max_ms": 43.78341697156429, "samples_ms": [ 42.84545802511275, 43.3145419228822, 43.2329170871526, 43.78341697156429, 43.68899995461106 ], "max_abs": 7.0625, "relative_rmse": 0.09437878429889679, "equal_fraction": 0.018927190452814102 }, { "operator": "attn.qkv_proj", "variant": "affine8_g32", "shape": [ 4181, 5376, 21504 ], "weight_gib": 0.121124267578125, "median_ms": 48.382166773080826, "min_ms": 47.48316714540124, "max_ms": 48.58658299781382, "samples_ms": [ 47.48316714540124, 48.273707972839475, 48.382166773080826, 48.58658299781382, 48.445832915604115 ], "max_abs": 0.5, "relative_rmse": 0.007152453996241093, "equal_fraction": 0.23806069791316986 }, { "operator": "attn.qkv_proj", "variant": "affine6_g32", "shape": [ 4181, 5376, 21504 ], "weight_gib": 0.094207763671875, "median_ms": 48.20779198780656, "min_ms": 47.32833313755691, "max_ms": 48.97204088047147, "samples_ms": [ 47.32833313755691, 47.76024981401861, 48.20779198780656, 48.81024989299476, 48.97204088047147 ], "max_abs": 1.40625, "relative_rmse": 0.020706482231616974, "equal_fraction": 0.0854361429810524 }, { "operator": "attn.qkv_proj", "variant": "affine4_g32", "shape": [ 4181, 5376, 21504 ], "weight_gib": 0.067291259765625, "median_ms": 46.27333302050829, "min_ms": 45.270083006471395, "max_ms": 46.489667147397995, "samples_ms": [ 45.270083006471395, 46.27333302050829, 46.13558296114206, 46.434874879196286, 46.489667147397995 ], "max_abs": 5.60546875, "relative_rmse": 0.08441592752933502, "equal_fraction": 0.021116377785801888 }, { "operator": "attn.out_proj", "variant": "bf16_nt", "shape": [ 4181, 7168, 5376 ], "median_ms": 13.733083149418235, "min_ms": 13.50299990735948, "max_ms": 13.899000128731132, "samples_ms": [ 13.50299990735948, 13.647624989971519, 13.899000128731132, 13.870457885786891, 13.733083149418235 ], "max_abs": 0.0, "relative_rmse": 0.0, "equal_fraction": 1.0 }, { "operator": "attn.out_proj", "variant": "bf16_nn", "shape": [ 4181, 7168, 5376 ], "median_ms": 13.947458937764168, "min_ms": 13.841500040143728, "max_ms": 14.289834070950747, "samples_ms": [ 13.904999941587448, 14.170832931995392, 13.841500040143728, 13.947458937764168, 14.289834070950747 ], "max_abs": 0.0, "relative_rmse": 0.0, "equal_fraction": 1.0 }, { "operator": "attn.out_proj", "variant": "bf16_pad64", "shape": [ 4181, 7168, 5376 ], "padded_tokens": 4224, "median_ms": 14.317499939352274, "min_ms": 14.01154207997024, "max_ms": 14.362999936565757, "samples_ms": [ 14.317499939352274, 14.01154207997024, 14.362999936565757, 14.22337512485683, 14.34200000949204 ], "max_abs": 0.0, "relative_rmse": 0.0, "equal_fraction": 1.0 }, { "operator": "attn.out_proj", "variant": "bf16_pad128", "shape": [ 4181, 7168, 5376 ], "padded_tokens": 4224, "median_ms": 14.29895800538361, "min_ms": 14.083375222980976, "max_ms": 14.456833014264703, "samples_ms": [ 14.083375222980976, 14.456833014264703, 14.22062492929399, 14.29895800538361, 14.390083961188793 ], "max_abs": 0.0, "relative_rmse": 0.0, "equal_fraction": 1.0 }, { "operator": "attn.out_proj", "variant": "bf16_pad256", "shape": [ 4181, 7168, 5376 ], "padded_tokens": 4352, "median_ms": 14.49437509290874, "min_ms": 14.469749992713332, "max_ms": 14.765792060643435, "samples_ms": [ 14.469749992713332, 14.49437509290874, 14.531959081068635, 14.490332920104265, 14.765792060643435 ], "max_abs": 0.0, "relative_rmse": 0.0, "equal_fraction": 1.0 }, { "operator": "attn.out_proj", "variant": "fp16_round_bf16", "shape": [ 4181, 7168, 5376 ], "median_ms": 14.573207823559642, "min_ms": 14.342708047479391, "max_ms": 14.808290870860219, "samples_ms": [ 14.342708047479391, 14.638457912951708, 14.573207823559642, 14.566500205546618, 14.808290870860219 ], "max_abs": 0.25, "relative_rmse": 0.0014379547210410237, "equal_fraction": 0.93744957447052 }, { "operator": "attn.out_proj", "variant": "mxfp88_g32", "shape": [ 4181, 7168, 5376 ], "weight_gib": 0.03701019287109375, "median_ms": 16.644750023260713, "min_ms": 16.09429088421166, "max_ms": 16.698166029527783, "samples_ms": [ 16.09429088421166, 16.173999989405274, 16.698166029527783, 16.687000170350075, 16.644750023260713 ], "max_abs": 1.53125, "relative_rmse": 0.026705145835876465, "equal_fraction": 0.06687343865633011 }, { "operator": "attn.out_proj", "variant": "mxfp44_g32", "shape": [ 4181, 7168, 5376 ], "weight_gib": 0.01906585693359375, "median_ms": 15.215334016829729, "min_ms": 15.133667038753629, "max_ms": 15.53854113444686, "samples_ms": [ 15.215334016829729, 15.133667038753629, 15.319749945774674, 15.53854113444686, 15.184166841208935 ], "max_abs": 6.890625, "relative_rmse": 0.11676037311553955, "equal_fraction": 0.01528572104871273 }, { "operator": "attn.out_proj", "variant": "nvfp44_g16", "shape": [ 4181, 7168, 5376 ], "weight_gib": 0.0201873779296875, "median_ms": 15.884707914665341, "min_ms": 15.523541951552033, "max_ms": 15.917500015348196, "samples_ms": [ 15.906207961961627, 15.523541951552033, 15.884707914665341, 15.917500015348196, 15.744959004223347 ], "max_abs": 6.16796875, "relative_rmse": 0.09526105225086212, "equal_fraction": 0.018807044252753258 }, { "operator": "attn.out_proj", "variant": "affine8_g32", "shape": [ 4181, 7168, 5376 ], "weight_gib": 0.040374755859375, "median_ms": 16.813124995678663, "min_ms": 16.703708097338676, "max_ms": 16.886916011571884, "samples_ms": [ 16.7407498229295, 16.813124995678663, 16.703708097338676, 16.844790894538164, 16.886916011571884 ], "max_abs": 0.5, "relative_rmse": 0.006857909727841616, "equal_fraction": 0.24853673577308655 }, { "operator": "attn.out_proj", "variant": "affine6_g32", "shape": [ 4181, 7168, 5376 ], "weight_gib": 0.031402587890625, "median_ms": 17.113542184233665, "min_ms": 16.857665963470936, "max_ms": 17.204958014190197, "samples_ms": [ 17.113542184233665, 17.105791019275784, 17.204958014190197, 16.857665963470936, 17.14545814320445 ], "max_abs": 1.25, "relative_rmse": 0.019887777045369148, "equal_fraction": 0.08924407511949539 }, { "operator": "attn.out_proj", "variant": "affine4_g32", "shape": [ 4181, 7168, 5376 ], "weight_gib": 0.022430419921875, "median_ms": 15.973791014403105, "min_ms": 15.816291095688939, "max_ms": 16.10216684639454, "samples_ms": [ 16.008083010092378, 15.816291095688939, 15.922083985060453, 15.973791014403105, 16.10216684639454 ], "max_abs": 5.78125, "relative_rmse": 0.08107094466686249, "equal_fraction": 0.022073309868574142 }, { "operator": "mlp.fc1", "variant": "bf16_nt", "shape": [ 4181, 5376, 28672 ], "median_ms": 56.88041704706848, "min_ms": 56.056959088891745, "max_ms": 57.06249992363155, "samples_ms": [ 56.056959088891745, 56.88041704706848, 56.92950007505715, 56.434999918565154, 57.06249992363155 ], "max_abs": 0.0, "relative_rmse": 0.0, "equal_fraction": 1.0 }, { "operator": "mlp.fc1", "variant": "bf16_nn", "shape": [ 4181, 5376, 28672 ], "median_ms": 57.2883749846369, "min_ms": 57.20029189251363, "max_ms": 57.50133306719363, "samples_ms": [ 57.20029189251363, 57.2883749846369, 57.50133306719363, 57.251624995842576, 57.43858404457569 ], "max_abs": 0.0, "relative_rmse": 0.0, "equal_fraction": 1.0 }, { "operator": "mlp.fc1", "variant": "bf16_pad64", "shape": [ 4181, 5376, 28672 ], "padded_tokens": 4224, "median_ms": 57.62783298268914, "min_ms": 56.953833904117346, "max_ms": 58.22679097764194, "samples_ms": [ 56.953833904117346, 57.62783298268914, 57.27287498302758, 58.22679097764194, 58.03979095071554 ], "max_abs": 0.0, "relative_rmse": 0.0, "equal_fraction": 1.0 }, { "operator": "mlp.fc1", "variant": "bf16_pad128", "shape": [ 4181, 5376, 28672 ], "padded_tokens": 4224, "median_ms": 58.27929195947945, "min_ms": 57.83387483097613, "max_ms": 58.72062500566244, "samples_ms": [ 57.83387483097613, 58.05337498895824, 58.49958397448063, 58.27929195947945, 58.72062500566244 ], "max_abs": 0.0, "relative_rmse": 0.0, "equal_fraction": 1.0 }, { "operator": "mlp.fc1", "variant": "bf16_pad256", "shape": [ 4181, 5376, 28672 ], "padded_tokens": 4352, "median_ms": 59.44949993863702, "min_ms": 58.33770893514156, "max_ms": 60.163999907672405, "samples_ms": [ 58.33770893514156, 58.92641679383814, 59.69562497921288, 59.44949993863702, 60.163999907672405 ], "max_abs": 0.0, "relative_rmse": 0.0, "equal_fraction": 1.0 }, { "operator": "mlp.fc1", "variant": "fp16_round_bf16", "shape": [ 4181, 5376, 28672 ], "median_ms": 60.3128750808537, "min_ms": 59.13254199549556, "max_ms": 60.70520798675716, "samples_ms": [ 59.13254199549556, 59.86141599714756, 60.36758283153176, 60.3128750808537, 60.70520798675716 ], "max_abs": 0.5, "relative_rmse": 0.0014354648301377892, "equal_fraction": 0.937492311000824 }, { "operator": "mlp.fc1", "variant": "mxfp88_g32", "shape": [ 4181, 5376, 28672 ], "weight_gib": 0.148040771484375, "median_ms": 68.84741690009832, "min_ms": 68.01695888862014, "max_ms": 69.89704188890755, "samples_ms": [ 68.01695888862014, 68.76904191449285, 68.84741690009832, 69.89704188890755, 69.67624998651445 ], "max_abs": 2.1875, "relative_rmse": 0.026700254529714584, "equal_fraction": 0.06675215065479279 }, { "operator": "mlp.fc1", "variant": "mxfp44_g32", "shape": [ 4181, 5376, 28672 ], "weight_gib": 0.076263427734375, "median_ms": 62.32962501235306, "min_ms": 61.62541685625911, "max_ms": 63.24025010690093, "samples_ms": [ 61.62541685625911, 62.32962501235306, 62.328125117346644, 63.24025010690093, 63.214374938979745 ], "max_abs": 10.34375, "relative_rmse": 0.11928655952215195, "equal_fraction": 0.014761211350560188 }, { "operator": "mlp.fc1", "variant": "nvfp44_g16", "shape": [ 4181, 5376, 28672 ], "weight_gib": 0.08074951171875, "median_ms": 64.51166700571775, "min_ms": 63.704333966597915, "max_ms": 65.21387491375208, "samples_ms": [ 63.704333966597915, 64.38133399933577, 64.51166700571775, 64.73379209637642, 65.21387491375208 ], "max_abs": 8.03125, "relative_rmse": 0.0948139876127243, "equal_fraction": 0.01893121376633644 }, { "operator": "mlp.fc1", "variant": "affine8_g32", "shape": [ 4181, 5376, 28672 ], "weight_gib": 0.1614990234375, "median_ms": 69.97695891186595, "min_ms": 69.09304205328226, "max_ms": 70.55416703224182, "samples_ms": [ 69.09304205328226, 69.7993750218302, 69.97695891186595, 70.26958395726979, 70.55416703224182 ], "max_abs": 0.6875, "relative_rmse": 0.00697753531858325, "equal_fraction": 0.24016006290912628 }, { "operator": "mlp.fc1", "variant": "affine6_g32", "shape": [ 4181, 5376, 28672 ], "weight_gib": 0.1256103515625, "median_ms": 71.28174998797476, "min_ms": 69.93604195304215, "max_ms": 71.53204106725752, "samples_ms": [ 69.93604195304215, 71.13158400170505, 71.44416682422161, 71.53204106725752, 71.28174998797476 ], "max_abs": 1.625, "relative_rmse": 0.020274344831705093, "equal_fraction": 0.08617563545703888 }, { "operator": "mlp.fc1", "variant": "affine4_g32", "shape": [ 4181, 5376, 28672 ], "weight_gib": 0.0897216796875, "median_ms": 66.9265000615269, "min_ms": 65.172916976735, "max_ms": 67.17233289964497, "samples_ms": [ 65.172916976735, 66.3087498396635, 66.94695795886219, 67.17233289964497, 66.9265000615269 ], "max_abs": 7.21875, "relative_rmse": 0.08270496875047684, "equal_fraction": 0.021269498392939568 }, { "operator": "mlp.fc2", "variant": "bf16_nt", "shape": [ 4181, 14336, 5376 ], "median_ms": 34.53045804053545, "min_ms": 33.855958143249154, "max_ms": 35.29229178093374, "samples_ms": [ 33.855958143249154, 35.29229178093374, 34.14345812052488, 34.53045804053545, 34.985417034476995 ], "max_abs": 0.0, "relative_rmse": 0.0, "equal_fraction": 1.0 }, { "operator": "mlp.fc2", "variant": "bf16_nn", "shape": [ 4181, 14336, 5376 ], "median_ms": 34.404292004182935, "min_ms": 33.76633278094232, "max_ms": 35.07920913398266, "samples_ms": [ 33.76633278094232, 34.01512489654124, 34.55324983224273, 34.404292004182935, 35.07920913398266 ], "max_abs": 0.0, "relative_rmse": 0.0, "equal_fraction": 1.0 }, { "operator": "mlp.fc2", "variant": "bf16_pad64", "shape": [ 4181, 14336, 5376 ], "padded_tokens": 4224, "median_ms": 35.306917037814856, "min_ms": 35.11383384466171, "max_ms": 36.45095904357731, "samples_ms": [ 35.11383384466171, 36.45095904357731, 35.306917037814856, 35.516874864697456, 35.22562491707504 ], "max_abs": 0.0, "relative_rmse": 0.0, "equal_fraction": 1.0 }, { "operator": "mlp.fc2", "variant": "bf16_pad128", "shape": [ 4181, 14336, 5376 ], "padded_tokens": 4224, "median_ms": 35.68662493489683, "min_ms": 34.68191693536937, "max_ms": 36.35245794430375, "samples_ms": [ 34.68191693536937, 35.05495795980096, 35.8519998844713, 35.68662493489683, 36.35245794430375 ], "max_abs": 0.0, "relative_rmse": 0.0, "equal_fraction": 1.0 }, { "operator": "mlp.fc2", "variant": "bf16_pad256", "shape": [ 4181, 14336, 5376 ], "padded_tokens": 4352, "median_ms": 35.76362505555153, "min_ms": 35.15912499278784, "max_ms": 36.0618750564754, "samples_ms": [ 35.15912499278784, 35.957250045612454, 36.0618750564754, 35.314542008563876, 35.76362505555153 ], "max_abs": 0.0, "relative_rmse": 0.0, "equal_fraction": 1.0 }, { "operator": "mlp.fc2", "variant": "fp16_round_bf16", "shape": [ 4181, 14336, 5376 ], "median_ms": 34.2127091716975, "min_ms": 34.06991716474295, "max_ms": 35.027166828513145, "samples_ms": [ 34.06991716474295, 34.142165910452604, 34.2127091716975, 34.318125108256936, 35.027166828513145 ], "max_abs": 0.5, "relative_rmse": 0.0014385132817551494, "equal_fraction": 0.9375085830688477 }, { "operator": "mlp.fc2", "variant": "mxfp88_g32", "shape": [ 4181, 14336, 5376 ], "weight_gib": 0.0740203857421875, "median_ms": 38.617667043581605, "min_ms": 37.2016669716686, "max_ms": 38.94404089078307, "samples_ms": [ 37.2016669716686, 38.0708749871701, 38.617667043581605, 38.94404089078307, 38.72366691939533 ], "max_abs": 3.5, "relative_rmse": 0.02668989636003971, "equal_fraction": 0.06687913089990616 }, { "operator": "mlp.fc2", "variant": "mxfp44_g32", "shape": [ 4181, 14336, 5376 ], "weight_gib": 0.0381317138671875, "median_ms": 34.176249988377094, "min_ms": 34.03883310966194, "max_ms": 34.531333949416876, "samples_ms": [ 34.16574979200959, 34.176249988377094, 34.03883310966194, 34.2446668073535, 34.531333949416876 ], "max_abs": 17.015625, "relative_rmse": 0.12333980202674866, "equal_fraction": 0.014510574750602245 }, { "operator": "mlp.fc2", "variant": "nvfp44_g16", "shape": [ 4181, 14336, 5376 ], "weight_gib": 0.040374755859375, "median_ms": 34.81729212217033, "min_ms": 34.36283301562071, "max_ms": 34.83804198913276, "samples_ms": [ 34.65237491764128, 34.83804198913276, 34.82170891948044, 34.36283301562071, 34.81729212217033 ], "max_abs": 11.5, "relative_rmse": 0.0943082720041275, "equal_fraction": 0.019040215760469437 }, { "operator": "mlp.fc2", "variant": "affine8_g32", "shape": [ 4181, 14336, 5376 ], "weight_gib": 0.08074951171875, "median_ms": 40.01833382062614, "min_ms": 39.36866600997746, "max_ms": 40.50199990160763, "samples_ms": [ 40.23308400064707, 40.01833382062614, 40.50199990160763, 39.36866600997746, 39.642584044486284 ], "max_abs": 1.125, "relative_rmse": 0.007229790091514587, "equal_fraction": 0.23662720620632172 }, { "operator": "mlp.fc2", "variant": "affine6_g32", "shape": [ 4181, 14336, 5376 ], "weight_gib": 0.06280517578125, "median_ms": 38.771667052060366, "min_ms": 38.38183404877782, "max_ms": 39.438792038708925, "samples_ms": [ 38.38183404877782, 38.771667052060366, 39.07174989581108, 38.75845903530717, 39.438792038708925 ], "max_abs": 3.5, "relative_rmse": 0.021220030263066292, "equal_fraction": 0.08387544006109238 }, { "operator": "mlp.fc2", "variant": "affine4_g32", "shape": [ 4181, 14336, 5376 ], "weight_gib": 0.04486083984375, "median_ms": 36.1207090318203, "min_ms": 35.76437500305474, "max_ms": 36.558917025104165, "samples_ms": [ 35.76437500305474, 36.558917025104165, 36.03099985048175, 36.1207090318203, 36.28229186870158 ], "max_abs": 13.5, "relative_rmse": 0.0866343304514885, "equal_fraction": 0.020664894953370094 }, { "operator": "mlp", "variant": "eager", "median_ms": 106.36541689746082, "min_ms": 105.06766708567739, "max_ms": 106.84604104608297, "samples_ms": [ 105.30966590158641, 105.06766708567739, 106.36541689746082, 106.41158395446837, 106.84604104608297 ], "max_abs": 0.0, "relative_rmse": 0.0, "equal_fraction": 1.0 }, { "operator": "mlp", "variant": "compiled", "median_ms": 99.15808285586536, "min_ms": 96.69841709546745, "max_ms": 99.61933316662908, "samples_ms": [ 96.69841709546745, 97.61712490580976, 99.33945816010237, 99.15808285586536, 99.61933316662908 ], "max_abs": 0.0, "relative_rmse": 0.0, "equal_fraction": 1.0 } ] }
{
"environment": {
"platform": "macOS-27.0.1-arm64-arm-64bit",
"machine": "arm64",
"metal": {
"device_name": "Apple M5 Pro",
"max_recommended_working_set_size": 55662788608,
"memory_size": 68719476736,
"architecture": "applegpu_g17s",
"max_buffer_length": 41747087360,
"resource_limit": 499000
},
"versions": {
"mlx": "0.32.4",
"mlx-metal": "0.32.4",
"numpy": "2.5.4",
"safetensors": "0.8.0",
"transformers": "4.57.6",
"torch": "2.15.0",
"Pillow": "13.0.0",
"scikit-image": "0.26.0"
},
"precision_environment": {
"MLX_ENABLE_TF32": "0",
"MLX_CONV_WINOGRAD": "0"
}
},
"warmup": 2,
"repeats": 5,
"scope": "Original block-0 weights; synthetic BF16 inputs. Operator latency only, excludes conversion and loading.",
"trials": [
{
"operator": "attn.qkv_proj",
"variant": "bf16_nt",
"shape": [
4181,
5376,
21504
],
"median_ms": 39.01795903220773,
"min_ms": 38.90308388508856,
"max_ms": 39.20233412645757,
"samples_ms": [
39.01795903220773,
39.000708842650056,
39.20233412645757,
39.08749995753169,
38.90308388508856
],
"max_abs": 0.0,
"relative_rmse": 0.0,
"equal_fraction": 1.0
},
{
"operator": "attn.qkv_proj",
"variant": "bf16_nn",
"shape": [
4181,
5376,
21504
],
"median_ms": 40.58533301576972,
"min_ms": 40.212417021393776,
"max_ms": 40.690625086426735,
"samples_ms": [
40.212417021393776,
40.641458006575704,
40.690625086426735,
40.58533301576972,
40.2926248498261
],
"max_abs": 0.0,
"relative_rmse": 0.0,
"equal_fraction": 1.0
},
{
"operator": "attn.qkv_proj",
"variant": "bf16_pad64",
"shape": [
4181,
5376,
21504
],
"padded_tokens": 4224,
"median_ms": 40.12054088525474,
"min_ms": 39.86324998550117,
"max_ms": 40.58220889419317,
"samples_ms": [
40.12054088525474,
40.07162502966821,
40.58220889419317,
39.86324998550117,
40.49258283339441
],
"max_abs": 0.0,
"relative_rmse": 0.0,
"equal_fraction": 1.0
},
{
"operator": "attn.qkv_proj",