{ "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" } }, "repeats": 5, "warmup": 2, "scope": "Synthetic BF16 Q/K/V; dense online softmax. Different reduction/rounding may affect generated images.", "trials": [ { "tokens": 65, "bq": 0, "bk": 0, "simdgroups": 0, "median_ms": 0.659167068079114, "min_ms": 0.42537483386695385, "max_ms": 0.7002090569585562, "samples_ms": [ 0.659167068079114, 0.7002090569585562, 0.6239160429686308, 0.697124982252717, 0.42537483386695385 ], "max_abs": 0.0, "relative_rmse": 0.0, "equal_fraction": 1.0 }, { "tokens": 65, "bq": 16, "bk": 32, "simdgroups": 2, "median_ms": 0.9386250749230385, "min_ms": 0.8443330880254507, "max_ms": 1.0369578376412392, "samples_ms": [ 0.8443330880254507, 1.0369578376412392, 0.9523339103907347, 0.9172081481665373, 0.9386250749230385 ], "max_abs": 0.00390625, "relative_rmse": 0.0021990817040205, "equal_fraction": 0.6370128393173218 }, { "tokens": 65, "bq": 16, "bk": 32, "simdgroups": 4, "median_ms": 0.9888748172670603, "min_ms": 0.9206659160554409, "max_ms": 1.0241249110549688, "samples_ms": [ 1.0212077759206295, 0.9888748172670603, 0.9206659160554409, 0.9763329289853573, 1.0241249110549688 ], "max_abs": 0.00390625, "relative_rmse": 0.0021990817040205, "equal_fraction": 0.6370128393173218 }, { "tokens": 65, "bq": 16, "bk": 64, "simdgroups": 2, "median_ms": 1.0865828953683376, "min_ms": 1.0042921639978886, "max_ms": 1.2037497945129871, "samples_ms": [ 1.1348749976605177, 1.0812499094754457, 1.2037497945129871, 1.0042921639978886, 1.0865828953683376 ], "max_abs": 0.0078125, "relative_rmse": 0.0022570982109755278, "equal_fraction": 0.627047598361969 }, { "tokens": 65, "bq": 16, "bk": 64, "simdgroups": 4, "median_ms": 1.1444170959293842, "min_ms": 1.1029590386897326, "max_ms": 1.2210828717797995, "samples_ms": [ 1.193916192278266, 1.1444170959293842, 1.1029590386897326, 1.1432920582592487, 1.2210828717797995 ], "max_abs": 0.0078125, "relative_rmse": 0.0022570982109755278, "equal_fraction": 0.627047598361969 }, { "tokens": 65, "bq": 16, "bk": 128, "simdgroups": 2, "median_ms": 1.110791927203536, "min_ms": 1.0907920077443123, "max_ms": 1.275666058063507, "samples_ms": [ 1.275666058063507, 1.0907920077443123, 1.110791927203536, 1.105833100154996, 1.1968340259045362 ], "max_abs": 0.0078125, "relative_rmse": 0.002258604858070612, "equal_fraction": 0.6266033053398132 }, { "tokens": 65, "bq": 16, "bk": 128, "simdgroups": 4, "median_ms": 1.157082850113511, "min_ms": 0.9266659617424011, "max_ms": 1.2570410035550594, "samples_ms": [ 1.157082850113511, 1.2570410035550594, 1.0289170313626528, 0.9266659617424011, 1.221915939822793 ], "max_abs": 0.0078125, "relative_rmse": 0.002258604858070612, "equal_fraction": 0.6266033053398132 }, { "tokens": 65, "bq": 32, "bk": 32, "simdgroups": 2, "median_ms": 1.2491252273321152, "min_ms": 1.2336669024080038, "max_ms": 1.284166006371379, "samples_ms": [ 1.2491252273321152, 1.246333122253418, 1.284166006371379, 1.2553329579532146, 1.2336669024080038 ], "max_abs": 0.00390625, "relative_rmse": 0.0021990817040205, "equal_fraction": 0.6370128393173218 }, { "tokens": 65, "bq": 32, "bk": 32, "simdgroups": 4, "median_ms": 0.8222081232815981, "min_ms": 0.7596670184284449, "max_ms": 1.1705411598086357, "samples_ms": [ 1.1705411598086357, 1.1244169436395168, 0.8222081232815981, 0.7790829986333847, 0.7596670184284449 ], "max_abs": 0.00390625, "relative_rmse": 0.0021990817040205, "equal_fraction": 0.6370128393173218 }, { "tokens": 65, "bq": 32, "bk": 64, "simdgroups": 2, "median_ms": 0.9984159842133522, "min_ms": 0.9292080067098141, "max_ms": 1.3260000851005316, "samples_ms": [ 1.3260000851005316, 1.0508750565350056, 0.9984159842133522, 0.9478749707341194, 0.9292080067098141 ], "max_abs": 0.0078125, "relative_rmse": 0.0022570982109755278, "equal_fraction": 0.627047598361969 }, { "tokens": 65, "bq": 32, "bk": 64, "simdgroups": 4, "median_ms": 1.1746669188141823, "min_ms": 1.0720000136643648, "max_ms": 1.1764578521251678, "samples_ms": [ 1.0720000136643648, 1.1764578521251678, 1.1746669188141823, 1.176374964416027, 1.089167082682252 ], "max_abs": 0.0078125, "relative_rmse": 0.0022570982109755278, "equal_fraction": 0.627047598361969 }, { "tokens": 65, "bq": 32, "bk": 128, "simdgroups": 2, "median_ms": 1.4922497794032097, "min_ms": 1.3934581074863672, "max_ms": 1.5366659499704838, "samples_ms": [ 1.5366659499704838, 1.4588329941034317, 1.4922497794032097, 1.5216669999063015, 1.3934581074863672 ], "max_abs": 0.0078125, "relative_rmse": 0.002258604858070612, "equal_fraction": 0.6266033053398132 }, { "tokens": 65, "bq": 32, "bk": 128, "simdgroups": 4, "median_ms": 1.4812089502811432, "min_ms": 1.4430421870201826, "max_ms": 1.8411669880151749, "samples_ms": [ 1.4435420744121075, 1.4812089502811432, 1.8411669880151749, 1.5069169458001852, 1.4430421870201826 ], "max_abs": 0.0078125, "relative_rmse": 0.002258604858070612, "equal_fraction": 0.6266033053398132 }, { "tokens": 4181, "bq": 0, "bk": 0, "simdgroups": 0, "median_ms": 25.476875016465783, "min_ms": 24.982792092487216, "max_ms": 25.54537495598197, "samples_ms": [ 25.476875016465783, 24.982792092487216, 25.54537495598197, 25.49154101870954, 25.047583039849997 ], "max_abs": 0.0, "relative_rmse": 0.0, "equal_fraction": 1.0 }, { "tokens": 4181, "bq": 16, "bk": 32, "simdgroups": 2, "median_ms": 66.1243749782443, "min_ms": 65.3223330155015, "max_ms": 66.38374994508922, "samples_ms": [ 66.38374994508922, 66.16762490011752, 65.96424989402294, 65.3223330155015, 66.1243749782443 ], "max_abs": 0.0009765625, "relative_rmse": 0.0024988260120153427, "equal_fraction": 0.5868481993675232 }, { "tokens": 4181, "bq": 16, "bk": 32, "simdgroups": 4, "median_ms": 64.00691601447761, "min_ms": 63.77850007265806, "max_ms": 64.42370801232755, "samples_ms": [ 64.42370801232755, 63.77937481738627, 64.00691601447761, 63.77850007265806, 64.25516703166068 ], "max_abs": 0.0009765625, "relative_rmse": 0.0024988260120153427, "equal_fraction": 0.5868481993675232 }, { "tokens": 4181, "bq": 16, "bk": 64, "simdgroups": 2, "median_ms": 77.37500010989606, "min_ms": 76.40204206109047, "max_ms": 77.76308408938348, "samples_ms": [ 76.40204206109047, 77.57425005547702, 77.09983410313725, 77.76308408938348, 77.37500010989606 ], "max_abs": 0.0009765625, "relative_rmse": 0.002500565955415368, "equal_fraction": 0.5864733457565308 }, { "tokens": 4181, "bq": 16, "bk": 64, "simdgroups": 4, "median_ms": 75.66545787267387, "min_ms": 74.57149983383715, "max_ms": 77.01599993743002, "samples_ms": [ 75.66545787267387, 74.57149983383715, 75.59287501499057, 75.99666714668274, 77.01599993743002 ], "max_abs": 0.0009765625, "relative_rmse": 0.002500565955415368, "equal_fraction": 0.5864733457565308 }, { "tokens": 4181, "bq": 16, "bk": 128, "simdgroups": 2, "median_ms": 101.380875101313, "min_ms": 98.50454214029014, "max_ms": 105.114207835868, "samples_ms": [ 98.50454214029014, 101.380875101313, 100.96195805817842, 101.94479208439589, 105.114207835868 ], "max_abs": 0.0009765625, "relative_rmse": 0.0025032442063093185, "equal_fraction": 0.5861243009567261 }, { "tokens": 4181, "bq": 16, "bk": 128, "simdgroups": 4, "median_ms": 107.6576670166105, "min_ms": 107.06820781342685, "max_ms": 108.40670811012387, "samples_ms": [ 107.06820781342685, 107.6576670166105, 107.09299985319376, 108.40670811012387, 108.04883390665054 ], "max_abs": 0.0009765625, "relative_rmse": 0.0025032442063093185, "equal_fraction": 0.5861243009567261 }, { "tokens": 4181, "bq": 32, "bk": 32, "simdgroups": 2, "median_ms": 156.10979194752872, "min_ms": 154.5753749087453, "max_ms": 161.5379590075463, "samples_ms": [ 161.5379590075463, 158.93783303909004, 156.10979194752872, 155.60425003059208, 154.5753749087453 ], "max_abs": 0.0009765625, "relative_rmse": 0.0024988260120153427, "equal_fraction": 0.5868481993675232 }, { "tokens": 4181, "bq": 32, "bk": 32, "simdgroups": 4, "median_ms": 150.2824581693858, "min_ms": 149.2766251321882, "max_ms": 151.7170409206301, "samples_ms": [ 149.8422920703888, 151.7170409206301, 150.2824581693858, 149.2766251321882, 150.5736659746617 ], "max_abs": 0.0009765625, "relative_rmse": 0.0024988260120153427, "equal_fraction": 0.5868481993675232 }, { "tokens": 4181, "bq": 32, "bk": 64, "simdgroups": 2, "median_ms": 132.72304087877274, "min_ms": 132.2320420295, "max_ms": 133.52966704405844, "samples_ms": [ 132.2320420295, 132.53358309157193, 132.72304087877274, 133.52966704405844, 132.75087508372962 ], "max_abs": 0.0009765625, "relative_rmse": 0.002500565955415368, "equal_fraction": 0.5864733457565308 }, { "tokens": 4181, "bq": 32, "bk": 64, "simdgroups": 4, "median_ms": 134.08570806495845, "min_ms": 133.57874983921647, "max_ms": 134.52237495221198, "samples_ms": [ 134.1719999909401, 133.94024991430342, 133.57874983921647, 134.52237495221198, 134.08570806495845 ], "max_abs": 0.0009765625, "relative_rmse": 0.002500565955415368, "equal_fraction": 0.5864733457565308 }, { "tokens": 4181, "bq": 32, "bk": 128, "simdgroups": 2, "median_ms": 170.27479084208608, "min_ms": 168.38433407247066, "max_ms": 172.6567498408258, "samples_ms": [ 172.6567498408258, 170.5612908117473, 170.27479084208608, 169.14899996481836, 168.38433407247066 ], "max_abs": 0.0009765625, "relative_rmse": 0.0025032442063093185, "equal_fraction": 0.5861243009567261 }, { "tokens": 4181, "bq": 32, "bk": 128, "simdgroups": 4, "median_ms": 167.5910830963403, "min_ms": 167.38187498413026, "max_ms": 169.39575015567243, "samples_ms": [ 168.2442920282483, 169.39575015567243, 167.5910830963403, 167.38187498413026, 167.47320792637765 ], "max_abs": 0.0009765625, "relative_rmse": 0.0025032442063093185, "equal_fraction": 0.5861243009567261 } ] }