Text-to-Image
MLX
MiniMax H3
Safetensors
English
apple-silicon
macos
image-to-image
image-editing
image-generation
Instructions to use devin-lai/MiniMax-H3-Image-mlx with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- MLX
How to use devin-lai/MiniMax-H3-Image-mlx with MLX:
# Download the model from the Hub pip install huggingface_hub[hf_xet] hf download devin-lai/MiniMax-H3-Image-mlx --local-dir MiniMax-H3-Image-mlx
- Notebooks
- Google Colab
- Kaggle
- Local Apps Settings
- LM Studio
- Atomic Chat
Download reports/optimization-r3/screen.json from devin-lai/MiniMax-H3-Image-mlx: direct link, hf CLI and curl.
- Browser
- Download file 112 kB
-
https://huggingface.co/devin-lai/MiniMax-H3-Image-mlx/resolve/main/reports/optimization-r3/screen.json
- Command line
-
hf download hf://devin-lai/MiniMax-H3-Image-mlx/reports/optimization-r3/screen.json
-
curl -L -o screen.json https://huggingface.co/devin-lai/MiniMax-H3-Image-mlx/resolve/main/reports/optimization-r3/screen.json
112 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"
}
},
"scope": "Actual saved first-block activations, one GPU job, counterbalanced warmed calls including layout/decode costs. Operator evidence only.",
"trials": [
{
"group": "attention",
"variant": "q",
"order": [
"control",
"candidate",
"candidate",
"control",
"control",
"candidate",
"candidate",
"control",
"control",
"candidate",
"candidate",
"control"
],
"control": {
"median_ms": 25.229458115063608,
"min_ms": 25.031917029991746,
"max_ms": 25.60575003735721,
"samples_ms": [
25.268583092838526,
25.075667072087526,
25.19033313728869,
25.031917029991746,
25.60575003735721,
25.284667033702135
]
},
"candidate": {
"median_ms": 25.469625601544976,
"min_ms": 25.04358277656138,
"max_ms": 27.149833040311933,
"samples_ms": [
25.305458111688495,
25.851749815046787,
25.04358277656138,
25.56595904752612,
25.37329215556383,
27.149833040311933
]
},
"reduction_pct": -0.9519327977083103,
"max_abs": 0.0,
"relative_rmse": 0.0,
"equal_fraction": 1.0
},
{
"group": "attention",
"variant": "k",
"order": [
"control",
"candidate",
"candidate",
"control",
"control",
"candidate",
"candidate",
"control",
"control",
"candidate",
"candidate",
"control"
],
"control": {
"median_ms": 25.859437533654273,
"min_ms": 25.340416003018618,
"max_ms": 26.08179091475904,
"samples_ms": [
25.340416003018618,
25.849125115200877,
25.9084589779377,
25.79154190607369,
25.869749952107668,
26.08179091475904
]
},
"candidate": {
"median_ms": 26.141000096686184,
"min_ms": 25.78525012359023,
"max_ms": 26.738915825262666,
"samples_ms": [
25.89174988679588,
25.78525012359023,
26.06958313845098,
26.21241705492139,
26.308625005185604,
26.738915825262666
]
},
"reduction_pct": -1.0888193630100407,
"max_abs": 0.0,
"relative_rmse": 0.0,
"equal_fraction": 1.0
},
{
"group": "attention",
"variant": "v",
"order": [
"control",
"candidate",
"candidate",
"control",
"control",
"candidate",
"candidate",
"control",
"control",
"candidate",
"candidate",
"control"
],
"control": {
"median_ms": 26.5235414262861,
"min_ms": 26.06820804066956,
"max_ms": 26.9916660618037,
"samples_ms": [
26.572000002488494,
26.06820804066956,
26.71158406883478,
26.47508285008371,
26.14770899526775,
26.9916660618037
]
},
"candidate": {
"median_ms": 26.45568735897541,
"min_ms": 25.472500128671527,
"max_ms": 26.980499969795346,
"samples_ms": [
25.472500128671527,
26.262166909873486,
26.980499969795346,
26.169166900217533,
26.649207808077335,
26.651000138372183
]
},
"reduction_pct": 0.25582581986371977,
"max_abs": 0.0,
"relative_rmse": 0.0,
"equal_fraction": 1.0
},
{
"group": "attention",
"variant": "kv",
"order": [
"control",
"candidate",
"candidate",
"control",
"control",
"candidate",
"candidate",
"control",
"control",
"candidate",
"candidate",
"control"
],
"control": {
"median_ms": 26.858228840865195,
"min_ms": 26.246291119605303,
"max_ms": 27.52129198051989,
"samples_ms": [
27.52129198051989,
26.246291119605303,
26.920874835923314,
26.890749810263515,
26.825707871466875,
26.35725005529821
]
},
"candidate": {
"median_ms": 27.170770918019116,
"min_ms": 26.904917089268565,
"max_ms": 27.513458160683513,
"samples_ms": [
27.103166794404387,
27.238375041633844,
27.05774991773069,
26.904917089268565,
27.513458160683513,
27.28683315217495
]
},
"reduction_pct": -1.163673446248925,
"max_abs": 0.0,
"relative_rmse": 0.0,
"equal_fraction": 1.0
},
{
"group": "attention",
"variant": "qk",
"order": [
"control",
"candidate",
"candidate",
"control",
"control",
"candidate",
"candidate",
"control",
"control",
"candidate",
"candidate",
"control"
],
"control": {
"median_ms": 27.14068756904453,
"min_ms": 26.804166147485375,
"max_ms": 27.53179194405675,
"samples_ms": [
27.08041714504361,
27.53179194405675,
27.262166142463684,
26.89216611906886,
27.20095799304545,
26.804166147485375
]
},
"candidate": {
"median_ms": 27.800749870948493,
"min_ms": 27.702665887773037,
"max_ms": 28.138082940131426,
"samples_ms": [
27.702665887773037,
27.784874895587564,
27.734083821997046,
27.816624846309423,
28.138082940131426,
28.064000187441707
]
},
"reduction_pct": -2.4320028747422118,
"max_abs": 0.0,
"relative_rmse": 0.0,
"equal_fraction": 1.0
},
{
"group": "attention",
"variant": "qkv",
"order": [
"control",
"candidate",
"candidate",
"control",
"control",
"candidate",
"candidate",
"control",
"control",
"candidate",
"candidate",
"control"
],
"control": {
"median_ms": 27.442208491265774,
"min_ms": 26.71104110777378,
"max_ms": 27.863250114023685,
"samples_ms": [
27.492749970406294,
27.391667012125254,
27.72591612301767,
27.863250114023685,
27.305457973852754,
26.71104110777378
]
},
"candidate": {
"median_ms": 28.753166552633047,
"min_ms": 28.096041874960065,
"max_ms": 29.45591602474451,
"samples_ms": [
28.803500114008784,
28.70283299125731,
28.096041874960065,
28.122124960646033,
29.45591602474451,
28.994416119530797
]
},
"reduction_pct": -4.7771594687232355,
"max_abs": 0.0,
"relative_rmse": 0.0,
"equal_fraction": 1.0
},
{
"group": "mpp",
"operator": "fc2",
"tile": [
64,
128,
128
],
"simdgroups": 4,
"walk": 0,
"order": [
"control",
"candidate",
"candidate",
"control",
"control",
"candidate",
"candidate",
"control",
"control",
"candidate",
"candidate",
"control"
],
"control": {
"median_ms": 32.69812499638647,
"min_ms": 32.49645885080099,
"max_ms": 33.13204203732312,
"samples_ms": [
32.848249888047576,
32.50425006262958,
32.49645885080099,
32.93229197151959,
33.13204203732312,
32.54800010472536
]
},
"candidate": {
"median_ms": 46.483063022606075,
"min_ms": 44.436374912038445,
"max_ms": 46.96791688911617,
"samples_ms": [
46.82158399373293,
44.436374912038445,
46.14454205147922,
45.50975002348423,
46.96791688911617,
46.862542163580656
]
},
"reduction_pct": -42.15819111261887,
"max_abs": 0.0,
"relative_rmse": 0.0,
"equal_fraction": 1.0
},
{
"group": "mpp",
"operator": "fc2",
"tile": [
64,
128,
128
],
"simdgroups": 4,
"walk": 8,
"order": [
"control",
"candidate",
"candidate",
"control",
"control",
"candidate",
"candidate",
"control",
"control",
"candidate",
"candidate",
"control"
],
"control": {
"median_ms": 33.702207962051034,
"min_ms": 33.23649987578392,
"max_ms": 34.3878751154989,
"samples_ms": [
33.23649987578392,
33.68545789271593,
33.96387514658272,
34.3878751154989,
33.64429180510342,
33.71895803138614
]
},
"candidate": {
"median_ms": 30.430083512328565,
"min_ms": 29.547041980549693,
"max_ms": 32.16029214672744,
"samples_ms": [
29.918708838522434,
30.43100005015731,
32.16029214672744,
29.547041980549693,
30.42916697449982,
30.4317909758538
]
},
"reduction_pct": 9.708931988690207,
"max_abs": 0.0,
"relative_rmse": 0.0,
"equal_fraction": 1.0
},
{
"group": "mpp",
"operator": "fc2",
"tile": [
64,
128,
256
],
"simdgroups": 4,
"walk": 0,
"order": [
"control",
"candidate",
"candidate",
"control",
"control",
"candidate",
"candidate",
"control",
"control",
"candidate",
"candidate",
"control"
],
"control": {
"median_ms": 33.28241698909551,
"min_ms": 33.0401249229908,
"max_ms": 33.49166689440608,
"samples_ms": [
33.4627500269562,
33.33674999885261,
33.49166689440608,
33.14974997192621,
33.22808397933841,
33.0401249229908
]
},
"candidate": {
"median_ms": 45.94068753067404,
"min_ms": 44.2713750526309,
"max_ms": 46.81620793417096,
"samples_ms": [
45.48799991607666,
46.39337514527142,
46.77283298224211,
46.81620793417096,
45.39870913140476,
44.2713750526309
]
},
"reduction_pct": -38.03290652156011,
"max_abs": 0.0,
"relative_rmse": 0.0,
"equal_fraction": 1.0
},
{
"group": "mpp",
"operator": "fc2",
"tile": [
64,
128,
256
],
"simdgroups": 4,
"walk": 8,
"order": [
"control",
"candidate",
"candidate",
"control",
"control",
"candidate",
"candidate",
"control",
"control",
"candidate",
"candidate",
"control"
],
"control": {
"median_ms": 33.980124979279935,
"min_ms": 33.76808296889067,
"max_ms": 34.726041834801435,
"samples_ms": [
33.94587500952184,
33.76808296889067,
34.01437494903803,
33.8490828871727,
34.5814588945359,
34.726041834801435
]
},
"candidate": {
"median_ms": 30.394604080356658,
"min_ms": 29.724291991442442,
"max_ms": 31.81616705842316,
"samples_ms": [
30.098708113655448,
29.724291991442442,
31.81616705842316,
31.27020807005465,
30.690500047057867,
29.959790874272585
]
},
"reduction_pct": 10.551817867384605,
"max_abs": 0.0,
"relative_rmse": 0.0,
"equal_fraction": 1.0
},
{
"group": "mpp",
"operator": "fc2",
"tile": [
64,
128,
512
],
"simdgroups": 4,
"walk": 0,
"order": [
"control",
"candidate",
"candidate",
"control",
"control",
"candidate",
"candidate",
"control",
"control",
"candidate",
"candidate",
"control"
],
"control": {
"median_ms": 32.97620837111026,
"min_ms": 32.364374957978725,
"max_ms": 33.496208023279905,
"samples_ms": [
32.364374957978725,
33.023957861587405,
32.54604106768966,
32.928458880633116,
33.078499836847186,
33.496208023279905
]
},
"candidate": {
"median_ms": 44.57024997100234,
"min_ms": 43.49741595797241,
"max_ms": 45.911625027656555,
"samples_ms": [
44.575999956578016,
45.71600002236664,
44.51462486758828,
44.564499985426664,
45.911625027656555,
43.49741595797241
]
},
"reduction_pct": -35.158807432965425,
"max_abs": 0.0,
"relative_rmse": 0.0,
"equal_fraction": 1.0
},
{
"group": "mpp",
"operator": "fc2",
"tile": [
64,
128,
512
],
"simdgroups": 4,
"walk": 8,
"order": [
"control",
"candidate",
"candidate",
"control",
"control",
"candidate",
"candidate",
"control",
"control",
"candidate",
"candidate",
"control"
],
"control": {
"median_ms": 33.583750016987324,
"min_ms": 32.85624994896352,
"max_ms": 34.44895800203085,
"samples_ms": [
32.85624994896352,
33.038708148524165,
33.43666600994766,
34.1495000757277,
34.44895800203085,
33.73083402402699
]
},
"candidate": {
"median_ms": 29.923375113867223,
"min_ms": 28.22491596452892,
"max_ms": 30.517749954015017,
"samples_ms": [
30.457957880571485,
28.22491596452892,
28.807041933760047,
29.71458318643272,
30.132167041301727,
30.517749954015017
]
},
"reduction_pct": 10.89924413226221,
"max_abs": 0.0,
"relative_rmse": 0.0,
"equal_fraction": 1.0
},
{
"group": "mpp",
"operator": "fc2",
"tile": [
64,
128,
128
],
"simdgroups": 8,
"walk": 0,
"order": [
"control",
"candidate",
"candidate",
"control",
"control",
"candidate",
"candidate",
"control",
"control",
"candidate",
"candidate",
"control"
],
"control": {
"median_ms": 33.13168743625283,
"min_ms": 32.98945794813335,
"max_ms": 33.945250092074275,
"samples_ms": [
33.16450002603233,
33.098874846473336,
33.945250092074275,
32.98945794813335,
33.26154197566211,
33.072917023673654
]
},
"candidate": {
"median_ms": 53.310583578422666,
"min_ms": 53.09424991719425,
"max_ms": 53.87866706587374,
"samples_ms": [
53.130083018913865,
53.09424991719425,
53.173334104940295,
53.684249985963106,
53.447833051905036,
53.87866706587374
]
},
"reduction_pct": -60.90512649256132,
"max_abs": 0.0,
"relative_rmse": 0.0,
"equal_fraction": 1.0
},
{
"group": "mpp",
"operator": "fc2",
"tile": [
64,
128,
128
],
"simdgroups": 8,
"walk": 8,
"order": [
"control",
"candidate",
"candidate",
"control",
"control",
"candidate",
"candidate",
"control",
"control",
"candidate",
"candidate",
"control"
],
"control": {
"median_ms": 34.71137548331171,
"min_ms": 34.47116701863706,
"max_ms": 35.334791988134384,
"samples_ms": [
34.511875128373504,
34.80858402326703,
34.614166943356395,
35.0277500692755,
35.334791988134384,
34.47116701863706
]
},
"candidate": {
"median_ms": 33.90649997163564,
"min_ms": 32.41004212759435,
"max_ms": 35.40570801123977,
"samples_ms": [
32.41004212759435,
33.353874925523996,
34.77708389982581,
33.19504205137491,
35.40570801123977,
34.45912501774728
]
},
"reduction_pct": 2.3187658237946662,
"max_abs": 0.0,
"relative_rmse": 0.0,
"equal_fraction": 1.0
},
{
"group": "mpp",
"operator": "fc2",
"tile": [
64,
128,
256
],
"simdgroups": 8,
"walk": 0,
"order": [
"control",
"candidate",
"candidate",
"control",
"control",
"candidate",
"candidate",
"control",
"control",
"candidate",
"candidate",
"control"
],
"control": {
"median_ms": 33.76337548252195,
"min_ms": 33.24000001884997,
"max_ms": 34.47937499731779,
"samples_ms": [
33.30691694281995,
33.24000001884997,
33.71295891702175,
33.81379204802215,
34.47937499731779,
33.92070787958801
]
},
"candidate": {
"median_ms": 53.14691655803472,
"min_ms": 52.84675001166761,
"max_ms": 53.39562497101724,
"samples_ms": [
53.18125011399388,
53.11258300207555,
53.39562497101724,
53.26724983751774,
52.84675001166761,
53.013459080830216
]
},
"reduction_pct": -57.40996212166911,
"max_abs": 0.0,
"relative_rmse": 0.0,
"equal_fraction": 1.0
},
{
"group": "mpp",
"operator": "fc2",
"tile": [
64,
128,
256
],
"simdgroups": 8,
"walk": 8,
"order": [
"control",
"candidate",
"candidate",
"control",
"control",
"candidate",
"candidate",
"control",
"control",
"candidate",
"candidate",
"control"
],
"control": {
"median_ms": 35.07014596834779,
"min_ms": 34.801292000338435,
"max_ms": 35.45666695572436,
"samples_ms": [
34.801292000338435,
34.80708389542997,
34.918874967843294,
35.22141696885228,
35.45666695572436,
35.4250418022275
]
},
"candidate": {
"median_ms": 33.909062505699694,
"min_ms": 33.21541682817042,
"max_ms": 35.0338751450181,
"samples_ms": [
33.57775020413101,
33.88270805589855,
33.93541695550084,
33.21541682817042,
35.0338751450181,
34.80620798654854
]
},
"reduction_pct": 3.310746022260691,
"max_abs": 0.0,
"relative_rmse": 0.0,
"equal_fraction": 1.0
},
{
"group": "mpp",
"operator": "fc2",
"tile": [
64,
128,
512
],
"simdgroups": 8,
"walk": 0,
"order": [
"control",
"candidate",
"candidate",
"control",
"control",
"candidate",
"candidate",
"control",
"control",
"candidate",
"candidate",
"control"
],
"control": {
"median_ms": 33.45052036456764,
"min_ms": 33.10804208740592,
"max_ms": 34.30433291941881,
"samples_ms": [
33.10804208740592,
33.283333061262965,
33.33395789377391,
33.56708283536136,
34.30433291941881,
33.7523331400007
]
},
"candidate": {
"median_ms": 53.67537494748831,
"min_ms": 53.04420809261501,
"max_ms": 53.82925015874207,
"samples_ms": [
53.38458297774196,
53.04420809261501,
53.75545797869563,
53.82925015874207,
53.595291916280985,
53.80429094657302
]
},
"reduction_pct": -60.462002870196855,
"max_abs": 0.0,
"relative_rmse": 0.0,
"equal_fraction": 1.0
},
{
"group": "mpp",
"operator": "fc2",
"tile": [
64,
128,
512
],
"simdgroups": 8,
"walk": 8,
"order": [
"control",
"candidate",
"candidate",
"control",
"control",
"candidate",
"candidate",
"control",
"control",
"candidate",
"candidate",
"control"
],
"control": {
"median_ms": 34.256957937031984,
"min_ms": 33.05050008930266,
"max_ms": 34.71174999140203,
"samples_ms": [
33.05050008930266,
34.18091591447592,
34.33299995958805,
34.11058313213289,
34.71174999140203,
34.46020791307092
]
},
"candidate": {
"median_ms": 35.579250077717006,
"min_ms": 30.861708102747798,
"max_ms": 37.281750002875924,
"samples_ms": [
30.861708102747798,
36.76758403889835,
33.990791998803616,
34.494500141590834,
37.281750002875924,
36.66400001384318
]
},
"reduction_pct": -3.859922831196916,
"max_abs": 0.0,
"relative_rmse": 0.0,
"equal_fraction": 1.0
},
{
"group": "mpp",
"operator": "fc2",
"tile": [
128,
128,
128
],
"simdgroups": 4,
"walk": 0,
"order": [
"control",
"candidate",
"candidate",
"control",
"control",
"candidate",
"candidate",
"control",
"control",
"candidate",
"candidate",
"control"
],
"control": {
"median_ms": 35.25187540799379,
"min_ms": 35.1625420153141,
"max_ms": 35.65299999900162,
"samples_ms": [
35.25291685946286,
35.63412488438189,
35.20904202014208,
35.65299999900162,
35.1625420153141,
35.25083395652473
]
},
"candidate": {
"median_ms": 41.61843750625849,
"min_ms": 40.902707958593965,
"max_ms": 42.40504209883511,
"samples_ms": [
40.902707958593965,
41.42266698181629,
41.77608294412494,
41.49625007994473,
41.740624932572246,
42.40504209883511
]
},
"reduction_pct": -18.06020821468408,
"max_abs": 0.0,
"relative_rmse": 0.0,
"equal_fraction": 1.0
},
{
"group": "mpp",
"operator": "fc2",
"tile": [
128,
128,
128
],
"simdgroups": 4,
"walk": 8,
"order": [
"control",
"candidate",
"candidate",
"control",
"control",
"candidate",
"candidate",
"control",
"control",
"candidate",
"candidate",
"control"
],
"control": {
"median_ms": 35.97972891293466,
"min_ms": 35.42249999009073,
"max_ms": 36.18470882065594,
"samples_ms": [
35.42249999009073,
35.77129193581641,
36.18470882065594,
35.905207972973585,
36.17508406750858,
36.05424985289574
]
},
"candidate": {
"median_ms": 41.706104413606226,
"min_ms": 41.30979208275676,
"max_ms": 42.23008407279849,
"samples_ms": [
41.54358385130763,
41.39179200865328,
41.30979208275676,
41.86862497590482,
42.23008407279849,
42.11949999444187
]
},
"reduction_pct": -15.915560438291521,
"max_abs": 0.0,
"relative_rmse": 0.0,
"equal_fraction": 1.0
},
{
"group": "mpp",
"operator": "fc2",
"tile": [
128,
128,
256
],
"simdgroups": 4,
"walk": 0,
"order": [
"control",
"candidate",
"candidate",
"control",
"control",
"candidate",
"candidate",
"control",
"control",
"candidate",
"candidate",
"control"
],
"control": {
"median_ms": 35.98637506365776,
"min_ms": 35.6951670255512,
"max_ms": 36.42770810984075,
"samples_ms": [
36.040708189830184,
35.6951670255512,
36.04087489657104,
36.42770810984075,
35.93204193748534,
35.81187501549721
]
},
"candidate": {
"median_ms": 38.65062491968274,
"min_ms": 38.38008316233754,
"max_ms": 39.08629110082984,
"samples_ms": [
38.97504205815494,
39.08629110082984,
38.65654184482992,
38.402500096708536,
38.644707994535565,
38.38008316233754
]
},
"reduction_pct": -7.403496049024327,
"max_abs": 0.0,
"relative_rmse": 0.0,
"equal_fraction": 1.0
},
{
"group": "mpp",
"operator": "fc2",
"tile": [
128,
128,
256
],
"simdgroups": 4,
"walk": 8,
"order": [
"control",
"candidate",
"candidate",
"control",
"control",
"candidate",
"candidate",
"control",
"control",
"candidate",
"candidate",
"control"
],
"control": {
"median_ms": 36.38777043670416,
"min_ms": 35.99750017747283,
"max_ms": 36.600833060219884,
"samples_ms": [
36.341457860544324,
36.253374768421054,
36.434083012863994,
35.99750017747283,
36.600833060219884,
36.55995801091194
]
},
"candidate": {
"median_ms": 38.49864553194493,
"min_ms": 38.053416879847646,
"max_ms": 39.50450010597706,
"samples_ms": [
38.49220811389387,
38.053416879847646,
39.16970896534622,
38.505082949995995,
39.50450010597706,
38.24616689234972
]
},
"reduction_pct": -5.80105642612152,
"max_abs": 0.0,
"relative_rmse": 0.0,
"equal_fraction": 1.0
},
{
"group": "mpp",
"operator": "fc2",
"tile": [
128,
128,
512
],
"simdgroups": 4,
"walk": 0,
"order": [
"control",
"candidate",
"candidate",
"control",
"control",
"candidate",
"candidate",
"control",
"control",
"candidate",
"candidate",
"control"
],
"control": {
"median_ms": 35.68329208064824,
"min_ms": 34.19204195961356,
"max_ms": 35.95762513577938,
"samples_ms": [
34.19204195961356,
35.526125226169825,
35.32066592015326,
35.90862499549985,
35.84045893512666,
35.95762513577938
]
},
"candidate": {
"median_ms": 41.27647902350873,
"min_ms": 39.94083311408758,
"max_ms": 42.64591680839658,
"samples_ms": [
39.94083311408758,
40.246833115816116,
41.21075011789799,
41.34220792911947,
42.0346250757575,
42.64591680839658
]
},
"reduction_pct": -15.674526134582134,
"max_abs": 0.0,
"relative_rmse": 0.0,
"equal_fraction": 1.0
},
{
"group": "mpp",
"operator": "fc2",
"tile": [
128,
128,
512
],
"simdgroups": 4,
"walk": 8,
"order": [
"control",
"candidate",
"candidate",
"control",
"control",
"candidate",
"candidate",
"control",
"control",
"candidate",
"candidate",
"control"
],
"control": {
"median_ms": 35.55281204171479,
"min_ms": 33.824458019807935,
"max_ms": 36.18891700170934,
"samples_ms": [
33.824458019807935,
34.69358291476965,
35.19083303399384,
35.914791049435735,
36.18891700170934,
36.066832952201366
]
},
"candidate": {
"median_ms": 41.01883352268487,
"min_ms": 39.21595914289355,
"max_ms": 42.39054210484028,
"samples_ms": [
39.21595914289355,
40.27725011110306,
40.73312506079674,
41.77387501113117,
41.30454198457301,
42.39054210484028
]
},
"reduction_pct": -15.374371722148727,
"max_abs": 0.0,
"relative_rmse": 0.0,
"equal_fraction": 1.0
},
{
"group": "mpp",
"operator": "fc2",
"tile": [
128,
128,
128
],
"simdgroups": 8,
"walk": 0,
"order": [
"control",
"candidate",
"candidate",
"control",
"control",
"candidate",
"candidate",
"control",
"control",
"candidate",
"candidate",
"control"
],
"control": {
"median_ms": 35.35712510347366,
"min_ms": 35.100291948765516,
"max_ms": 36.129374988377094,
"samples_ms": [
35.100291948765516,
35.18845792859793,
35.36375006660819,
36.129374988377094,
35.804207902401686,
35.350500140339136
]
},
"candidate": {
"median_ms": 37.16656262986362,
"min_ms": 36.86491586267948,
"max_ms": 37.45754202827811,
"samples_ms": [
37.261624820530415,
37.09358419291675,
36.86491586267948,
37.13512513786554,
37.198000121861696,
37.45754202827811
]
},
"reduction_pct": -5.117603654410763,
"max_abs": 0.0,
"relative_rmse": 0.0,
"equal_fraction": 1.0
},
{
"group": "mpp",
"operator": "fc2",
"tile": [
128,
128,
128
],
"simdgroups": 8,
"walk": 8,
"order": [
"control",
"candidate",
"candidate",
"control",
"control",
"candidate",
"candidate",
"control",
"control",
"candidate",
"candidate",
"control"
],
"control": {
"median_ms": 35.8905210159719,
"min_ms": 35.524999955669045,
"max_ms": 37.069333950057626,
"samples_ms": [
37.069333950057626,
35.88324994780123,
35.55295802652836,
35.897792084142566,
35.91241594403982,
35.524999955669045
]
},
"candidate": {
"median_ms": 34.55870854668319,
"min_ms": 34.21016689389944,
"max_ms": 34.83779216185212,
"samples_ms": [
34.40679213963449,
34.21016689389944,
34.45874992758036,
34.65866716578603,
34.728209022432566,
34.83779216185212
]
},
"reduction_pct": 3.710763821723362,
"max_abs": 0.0,
"relative_rmse": 0.0,
"equal_fraction": 1.0
},
{
"group": "mpp",
"operator": "fc2",
"tile": [
128,
128,
256
],
"simdgroups": 8,
"walk": 0,
"order": [
"control",
"candidate",
"candidate",
"control",
"control",
"candidate",
"candidate",
"control",
"control",
"candidate",
"candidate",
"control"
],
"control": {
"median_ms": 35.92833341099322,
"min_ms": 35.62420909292996,
"max_ms": 36.63462493568659,
"samples_ms": [
35.71220790036023,
35.62420909292996,
36.52383293956518,
36.14445892162621,
35.690875025466084,
36.63462493568659
]
},
"candidate": {
"median_ms": 36.30470857024193,
"min_ms": 36.04741697199643,
"max_ms": 36.62204183638096,
"samples_ms": [
36.62204183638096,
36.06554097495973,
36.58712515607476,
36.35266702622175,
36.04741697199643,
36.256750114262104
]
},
"reduction_pct": -1.047571995459018,
"max_abs": 0.0,
"relative_rmse": 0.0,
"equal_fraction": 1.0
},
{
"group": "mpp",
"operator": "fc2",
"tile": [
128,
128,
256
],
"simdgroups": 8,
"walk": 8,
"order": [
"control",
"candidate",
"candidate",
"control",
"control",
"candidate",
"candidate",
"control",
"control",
"candidate",
"candidate",
"control"
],
"control": {
"median_ms": 36.06770804617554,
"min_ms": 35.60099983587861,
"max_ms": 36.45800007507205,
"samples_ms": [
36.01370798423886,
35.94916709698737,
35.60099983587861,
36.45800007507205,
36.121708108112216,
36.437375005334616
]
},
"candidate": {
"median_ms": 33.5881250211969,
"min_ms": 32.341625075787306,
"max_ms": 35.77370895072818,
"samples_ms": [
33.67020818404853,
33.50604185834527,
33.68429187685251,
35.77370895072818,
33.483332954347134,
32.341625075787306
]
},
"reduction_pct": 6.874800643845069,
"max_abs": 0.0,
"relative_rmse": 0.0,
"equal_fraction": 1.0
},
{
"group": "mpp",
"operator": "fc2",
"tile": [
128,
128,
512
],
"simdgroups": 8,
"walk": 0,
"order": [
"control",
"candidate",
"candidate",
"control",
"control",
"candidate",
"candidate",
"control",
"control",
"candidate",
"candidate",
"control"
],
"control": {
"median_ms": 35.10264598298818,
"min_ms": 33.59350003302097,
"max_ms": 35.836707800626755,
"samples_ms": [
33.59350003302097,
34.17958412319422,
34.87966698594391,
35.80595809035003,
35.836707800626755,
35.325624980032444
]
},
"candidate": {
"median_ms": 34.59241648670286,
"min_ms": 34.03295809403062,
"max_ms": 35.82058288156986,
"samples_ms": [
34.51408306136727,
34.03295809403062,
34.51770800165832,
34.6671249717474,
35.82058288156986,
35.38491600193083
]
},
"reduction_pct": 1.4535357150358141,
"max_abs": 0.0,
"relative_rmse": 0.0,
"equal_fraction": 1.0
},
{
"group": "mpp",
"operator": "fc2",
"tile": [
128,
128,
512
],
"simdgroups": 8,
"walk": 8,
"order": [
"control",
"candidate",
"candidate",
"control",
"control",
"candidate",
"candidate",
"control",
"control",
"candidate",
"candidate",
"control"
],
"control": {
"median_ms": 35.038665984757245,
"min_ms": 34.09208310768008,
"max_ms": 35.7118749525398,
"samples_ms": [
34.09208310768008,
34.88429100252688,
34.823874942958355,
35.7118749525398,
35.65883287228644,
35.19304096698761
]
},
"candidate": {
"median_ms": 31.510208500549197,
"min_ms": 30.011584050953388,
"max_ms": 33.16824999637902,
"samples_ms": [
30.211166944354773,
30.011584050953388,
31.29887511022389,
31.721541890874505,
33.16824999637902,
32.41729107685387
]
},
"reduction_pct": 10.07018214033325,
"max_abs": 0.0,
"relative_rmse": 0.0,
"equal_fraction": 1.0
},
{
"group": "mpp",
"operator": "fc2",
"tile": [
128,
256,
128
],
"simdgroups": 8,
"walk": 0,
"order": [
"control",
"candidate",
"candidate",
"control",
"control",
"candidate",
"candidate",
"control",
"control",
"candidate",
"candidate",
"control"
],
"control": {
"median_ms": 35.65337450709194,
"min_ms": 33.48512505181134,
"max_ms": 35.89512500911951,
"samples_ms": [
33.48512505181134,
35.57479102164507,
35.15933291055262,
35.799749894067645,
35.73195799253881,
35.89512500911951
]
},
"candidate": {
"median_ms": 44.1848955815658,
"min_ms": 42.33650001697242,
"max_ms": 44.950291980057955,
"samples_ms": [
42.33650001697242,
43.28704113140702,
44.05637504532933,
44.31341611780226,
44.46279187686741,
44.950291980057955
]
},
"reduction_pct": -23.929070368295214,
"max_abs": 0.0,
"relative_rmse": 0.0,
"equal_fraction": 1.0
},
{
"group": "mpp",
"operator": "fc2",
"tile": [
128,
256,
128
],
"simdgroups": 8,
"walk": 8,
"order": [
"control",
"candidate",
"candidate",
"control",
"control",
"candidate",
"candidate",
"control",
"control",
"candidate",
"candidate",
"control"
],
"control": {
"median_ms": 35.47704196535051,
"min_ms": 33.64433301612735,
"max_ms": 35.964249866083264,
"samples_ms": [
33.64433301612735,
34.54320807941258,
35.272917011752725,
35.964249866083264,
35.91850004158914,
35.68116691894829
]
},
"candidate": {
"median_ms": 43.20979141630232,
"min_ms": 41.743291076272726,
"max_ms": 44.87358289770782,
"samples_ms": [
41.743291076272726,
43.125208001583815,
43.29437483102083,
42.989624897018075,
44.87358289770782,
44.278624933212996
]
},
"reduction_pct": -21.79648872221136,
"max_abs": 0.0,
"relative_rmse": 0.0,
"equal_fraction": 1.0
},
{
"group": "mpp",
"operator": "fc2",
"tile": [
128,
256,
256
],
"simdgroups": 8,
"walk": 0,
"order": [
"control",
"candidate",
"candidate",
"control",
"control",
"candidate",
"candidate",
"control",
"control",
"candidate",
"candidate",
"control"
],
"control": {
"median_ms": 35.428583389148116,
"min_ms": 33.60733320005238,
"max_ms": 36.47991712205112,
"samples_ms": [
33.60733320005238,
34.95291620492935,
35.43745796196163,
35.419708816334605,
35.5549999512732,
36.47991712205112
]
},
"candidate": {
"median_ms": 39.55368744209409,
"min_ms": 38.16095786169171,
"max_ms": 40.5611670576036,
"samples_ms": [
38.26766600832343,
38.16095786169171,
39.1714998986572,
40.25241709314287,
39.93587498553097,
40.5611670576036
]
},
"reduction_pct": -11.643434928333885,
"max_abs": 0.0,
"relative_rmse": 0.0,
"equal_fraction": 1.0
},
{
"group": "mpp",
"operator": "fc2",
"tile": [
128,
256,
256
],
"simdgroups": 8,
"walk": 8,
"order": [
"control",
"candidate",
"candidate",
"control",
"control",
"candidate",
"candidate",
"control",
"control",
"candidate",
"candidate",
"control"
],
"control": {
"median_ms": 35.1655624108389,
"min_ms": 33.86304201558232,
"max_ms": 35.9415418934077,
"samples_ms": [
33.86304201558232,
34.836542094126344,
34.958167001605034,
35.37295782007277,
35.50325008109212,
35.9415418934077
]
},
"candidate": {
"median_ms": 38.53777039330453,
"min_ms": 37.19449997879565,
"max_ms": 38.989333901554346,
"samples_ms": [
37.392625119537115,
37.19449997879565,
38.19366591051221,
38.989333901554346,
38.881874876096845,
38.95575017668307
]
},
"reduction_pct": -9.589518128753793,
"max_abs": 0.0,
"relative_rmse": 0.0,
"equal_fraction": 1.0
},
{
"group": "mpp",
"operator": "fc2",
"tile": [
128,
256,
512
],
"simdgroups": 8,
"walk": 0,
"order": [
"control",
"candidate",
"candidate",
"control",
"control",
"candidate",
"candidate",
"control",
"control",
"candidate",
"candidate",
"control"
],
"control": {
"median_ms": 35.462542087771,
"min_ms": 34.29579199291766,
"max_ms": 48.445792170241475,
"samples_ms": [
34.29579199291766,
35.15083296224475,
35.159084014594555,
36.64316702634096,
48.445792170241475,
35.76600016094744
]
},
"candidate": {
"median_ms": 41.34410398546606,
"min_ms": 40.06324987858534,
"max_ms": 42.46208304539323,
"samples_ms": [
40.06324987858534,
40.88900005444884,
41.79920791648328,
41.8063341639936,
42.46208304539323,
40.835542138665915
]
},
"reduction_pct": -16.58527999244386,
"max_abs": 0.0,
"relative_rmse": 0.0,
"equal_fraction": 1.0
},
{
"group": "mpp",
"operator": "fc2",
"tile": [
128,
256,
512
],
"simdgroups": 8,
"walk": 8,
"order": [
"control",
"candidate",
"candidate",
"control",
"control",
"candidate",
"candidate",
"control",
"control",
"candidate",
"candidate",
"control"
],
"control": {
"median_ms": 35.62572959344834,
"min_ms": 34.24241696484387,
"max_ms": 36.06008389033377,
"samples_ms": [
34.24241696484387,
35.36587511189282,
34.86437490209937,
36.06008389033377,
35.88558407500386,
35.954375052824616
]
},
"candidate": {
"median_ms": 41.14629200194031,
"min_ms": 38.749750005081296,
"max_ms": 42.71774995140731,
"samples_ms": [
38.749750005081296,
40.57412501424551,
41.63570795208216,
40.889292024075985,
41.403291979804635,
42.71774995140731
]
},
"reduction_pct": -15.495998177416181,
"max_abs": 0.0,
"relative_rmse": 0.0,
"equal_fraction": 1.0
},
{
"group": "mpp",
"operator": "fc2",
"tile": [
64,
256,
128
],
"simdgroups": 8,
"walk": 0,
"order": [
"control",
"candidate",
"candidate",
"control",
"control",
"candidate",
"candidate",
"control",
"control",
"candidate",
"candidate",
"control"
],
"control": {
"median_ms": 34.40391703043133,
"min_ms": 33.469249960035086,
"max_ms": 35.05558311007917,
"samples_ms": [
33.469249960035086,
34.44033395498991,
35.05558311007917,
34.5252079423517,
34.36750010587275,
34.30908406153321
]
},
"candidate": {
"median_ms": 46.19431297760457,
"min_ms": 43.69416693225503,
"max_ms": 47.964375000447035,
"samples_ms": [
43.69416693225503,
47.964375000447035,
46.300583984702826,
46.07574990950525,
46.08804197050631,
46.9970831181854
]
},
"reduction_pct": -34.270504538027645,
"max_abs": 0.0,
"relative_rmse": 0.0,
"equal_fraction": 1.0
},
{
"group": "mpp",
"operator": "fc2",
"tile": [
64,
256,
128
],
"simdgroups": 8,
"walk": 8,
"order": [
"control",
"candidate",
"candidate",
"control",
"control",
"candidate",
"candidate",
"control",
"control",
"candidate",
"candidate",
"control"
],
"control": {
"median_ms": 34.80716701596975,
"min_ms": 33.99754106067121,
"max_ms": 35.62520816922188,
"samples_ms": [
33.99754106067121,
34.39158294349909,
34.533374942839146,
35.456582903862,
35.08095908910036,
35.62520816922188
]
},
"candidate": {
"median_ms": 32.93895802926272,
"min_ms": 31.162999803200364,
"max_ms": 34.333915915340185,
"samples_ms": [
32.54345804452896,
31.162999803200364,
33.01837504841387,
32.85954101011157,
33.51020789705217,
34.333915915340185
]
},
"reduction_pct": 5.367311237510042,
"max_abs": 0.0,
"relative_rmse": 0.0,
"equal_fraction": 1.0
},
{
"group": "mpp",
"operator": "fc2",
"tile": [
64,
256,
256
],
"simdgroups": 8,
"walk": 0,
"order": [
"control",
"candidate",
"candidate",
"control",
"control",
"candidate",
"candidate",
"control",
"control",
"candidate",
"candidate",
"control"
],
"control": {
"median_ms": 34.14493741001934,
"min_ms": 33.469666028395295,
"max_ms": 34.90562504157424,
"samples_ms": [
33.469666028395295,
33.501625061035156,
34.46529200300574,
34.13399984128773,
34.155874978750944,
34.90562504157424
]
},
"candidate": {
"median_ms": 46.14712449256331,
"min_ms": 44.28262519650161,
"max_ms": 48.80308313295245,
"samples_ms": [
44.28262519650161,
45.455832965672016,
48.80308313295245,
46.8384160194546,
47.00362519361079,
44.46966596879065
]
},
"reduction_pct": -35.1507075219359,
"max_abs": 0.0,
"relative_rmse": 0.0,
"equal_fraction": 1.0
},
{
"group": "mpp",
"operator": "fc2",
"tile": [
64,
256,
256
],
"simdgroups": 8,
"walk": 8,
"order": [
"control",
"candidate",
"candidate",
"control",
"control",
"candidate",
"candidate",
"control",
"control",
"candidate",
"candidate",
"control"
],
"control": {
"median_ms": 35.21287499461323,
"min_ms": 34.00954091921449,
"max_ms": 36.09429206699133,
"samples_ms": [
34.00954091921449,
34.875291865319014,
34.96979200281203,
35.45595798641443,
35.598000045865774,
36.09429206699133
]
},
"candidate": {
"median_ms": 32.78872894588858,
"min_ms": 30.727375065907836,
"max_ms": 34.760459093376994,
"samples_ms": [
30.727375065907836,
31.798291951417923,
33.35325000807643,
32.22420788370073,
33.35470799356699,
34.760459093376994
]
},
"reduction_pct": 6.884260512938778,
"max_abs": 0.0,
"relative_rmse": 0.0,
"equal_fraction": 1.0
},
{
"group": "mpp",
"operator": "fc2",
"tile": [
64,
256,
512
],
"simdgroups": 8,
"walk": 0,
"order": [
"control",
"candidate",
"candidate",
"control",
"control",
"candidate",
"candidate",
"control",
"control",
"candidate",
"candidate",
"control"
],
"control": {
"median_ms": 34.38879153691232,
"min_ms": 33.1217919010669,
"max_ms": 35.00820882618427,
"samples_ms": [
33.1217919010669,
34.01741595007479,
34.29966699331999,
34.477916080504656,
34.516500076279044,
35.00820882618427
]
},
"candidate": {
"median_ms": 45.057187555357814,
"min_ms": 43.962083058431745,
"max_ms": 45.707833021879196,
"samples_ms": [
43.962083058431745,
45.59512506239116,
45.06691708229482,
45.707833021879196,
45.047458028420806,
44.293375220149755
]
},
"reduction_pct": -31.022887230550754,
"max_abs": 0.0,
"relative_rmse": 0.0,
"equal_fraction": 1.0
},
{
"group": "mpp",
"operator": "fc2",
"tile": [
64,
256,
512
],
"simdgroups": 8,
"walk": 8,
"order": [
"control",
"candidate",
"candidate",
"control",
"control",
"candidate",
"candidate",
"control",
"control",
"candidate",
"candidate",
"control"
],
"control": {
"median_ms": 34.84214539639652,
"min_ms": 33.75404211692512,
"max_ms": 35.77066585421562,
"samples_ms": [
33.75404211692512,
34.571625059470534,
34.90949980914593,
35.77066585421562,
34.77479098364711,
35.375708946958184
]
},
"candidate": {
"median_ms": 31.807999941520393,
"min_ms": 29.7283751424402,
"max_ms": 33.02587498910725,
"samples_ms": [
29.7283751424402,
30.72129189968109,
32.953582936897874,
31.430125003680587,
33.02587498910725,
32.1858748793602
]
},
"reduction_pct": 8.708262422869995,
"max_abs": 0.0,
"relative_rmse": 0.0,
"equal_fraction": 1.0
},
{
"group": "mpp",
"operator": "fc1",
"tile": [
64,
128,
128
],
"simdgroups": 4,
"walk": 0,
"order": [
"control",
"candidate",
"candidate",
"control",
"control",
"candidate",
"candidate",
"control",
"control",
"candidate",
"candidate",
"control"
],
"control": {
"median_ms": 59.79906243737787,
"min_ms": 58.727791998535395,
"max_ms": 61.80366687476635,
"samples_ms": [
59.981249971315265,
58.727791998535395,
61.2513751257211,
59.616874903440475,
61.80366687476635,
59.577709063887596
]
},
"candidate": {
"median_ms": 98.25556189753115,
"min_ms": 97.74987515993416,
"max_ms": 98.62716705538332,
"samples_ms": [
97.74987515993416,
97.89075003936887,
98.30145793966949,
98.62716705538332,
98.20966585539281,
98.58637489378452
]
},
"reduction_pct": -64.3095357898383,
"max_abs": 0.0,
"relative_rmse": 0.0,
"equal_fraction": 1.0
},
{
"group": "mpp",
"operator": "fc1",
"tile": [
64,
128,
128
],
"simdgroups": 4,
"walk": 8,
"order": [
"control",
"candidate",
"candidate",
"control",
"control",
"candidate",
"candidate",
"control",
"control",
"candidate",
"candidate",
"control"
],
"control": {
"median_ms": 64.00683359242976,
"min_ms": 62.827124958857894,
"max_ms": 64.38720808364451,
"samples_ms": [
62.827124958857894,
63.56199993751943,
64.38720808364451,
64.08250005915761,
64.16583294048905,
63.931167125701904
]
},
"candidate": {
"median_ms": 66.18287554010749,
"min_ms": 65.16166683286428,
"max_ms": 66.43333402462304,
"samples_ms": [
66.12395914271474,
65.16166683286428,
66.24179193750024,
66.38383283279836,
66.43333402462304,
66.04829197749496
]
},
"reduction_pct": -3.3997025404098435,
"max_abs": 0.0,
"relative_rmse": 0.0,
"equal_fraction": 1.0
},
{
"group": "mpp",
"operator": "fc1",
"tile": [
64,
128,
256
],
"simdgroups": 4,
"walk": 0,
"order": [
"control",
"candidate",
"candidate",
"control",
"control",
"candidate",
"candidate",
"control",
"control",
"candidate",
"candidate",
"control"
],
"control": {
"median_ms": 60.38008292671293,
"min_ms": 59.03858388774097,
"max_ms": 62.838332960382104,
"samples_ms": [
60.70383288897574,
59.15812496095896,
62.128250021487474,
59.03858388774097,
62.838332960382104,
60.05633296445012
]
},
"candidate": {
"median_ms": 97.94074995443225,
"min_ms": 97.43583318777382,
"max_ms": 98.3311659656465,
"samples_ms": [
98.17529213614762,
97.43583318777382,
97.72966708987951,
98.05012494325638,
98.3311659656465,
97.83137496560812
]
},
"reduction_pct": -62.207047766577325,
"max_abs": 0.0,
"relative_rmse": 0.0,
"equal_fraction": 1.0
},
{
"group": "mpp",
"operator": "fc1",
"tile": [
64,
128,
256
],
"simdgroups": 4,
"walk": 8,
"order": [
"control",
"candidate",
"candidate",
"control",
"control",
"candidate",
"candidate",
"control",
"control",
"candidate",
"candidate",
"control"
],
"control": {
"median_ms": 63.786458456888795,
"min_ms": 63.55137494392693,
"max_ms": 64.93179081007838,
"samples_ms": [
63.69879189878702,
63.55137494392693,
64.93179081007838,
63.87412501499057,
64.10745903849602,
63.5924160014838
]
},
"candidate": {
"median_ms": 64.37989533878863,
"min_ms": 63.57079092413187,
"max_ms": 65.03366702236235,
"samples_ms": [
65.03366702236235,
63.57079092413187,
64.5558328833431,
64.20395779423416,
64.81754197739065,
63.78645799122751
]
},
"reduction_pct": -0.9303493190500944,
"max_abs": 0.0,
"relative_rmse": 0.0,
"equal_fraction": 1.0
},
{
"group": "mpp",
"operator": "fc1",
"tile": [
64,
128,
128
],
"simdgroups": 8,
"walk": 0,
"order": [
"control",
"candidate",
"candidate",
"control",
"control",
"candidate",
"candidate",
"control",
"control",
"candidate",
"candidate",
"control"
],
"control": {
"median_ms": 63.008271041326225,
"min_ms": 60.26012497022748,
"max_ms": 64.03883313760161,
"samples_ms": [
60.63812505453825,
60.26012497022748,
62.77483399026096,
63.25179198756814,
64.03883313760161,
63.24170809239149
]
},
"candidate": {
"median_ms": 103.73756301123649,
"min_ms": 102.31587500311434,
"max_ms": 104.81399996206164,
"samples_ms": [
102.35991701483727,
102.31587500311434,
103.75658399425447,
103.71854202821851,
104.47499994188547,
104.81399996206164
]
},
"reduction_pct": -64.64118328718543,
"max_abs": 0.0,
"relative_rmse": 0.0,
"equal_fraction": 1.0
},
{
"group": "mpp",
"operator": "fc1",
"tile": [
64,
128,
128
],
"simdgroups": 8,
"walk": 8,
"order": [
"control",
"candidate",
"candidate",
"control",
"control",
"candidate",
"candidate",
"control",
"control",
"candidate",
"candidate",
"control"
],
"control": {
"median_ms": 66.17508351337165,
"min_ms": 63.94620821811259,
"max_ms": 66.85608392581344,
"samples_ms": [
63.94620821811259,
66.34554197080433,
66.85608392581344,
66.3592079654336,
66.00462505593896,
65.00495807267725
]
},
"candidate": {
"median_ms": 69.76683298125863,
"min_ms": 68.16041609272361,
"max_ms": 71.64516602642834,
"samples_ms": [
68.16041609272361,
69.85083315521479,
71.11266697756946,
71.64516602642834,
69.31612500920892,
69.68283280730247
]
},
"reduction_pct": -5.427646294033339,
"max_abs": 0.0,
"relative_rmse": 0.0,
"equal_fraction": 1.0
},
{
"group": "mpp",
"operator": "fc1",
"tile": [
64,
128,
256
],
"simdgroups": 8,
"walk": 0,
"order": [
"control",
"candidate",
"candidate",
"control",
"control",
"candidate",
"candidate",
"control",
"control",
"candidate",
"candidate",
"control"
],
"control": {
"median_ms": 58.05997899733484,
"min_ms": 56.33254093118012,
"max_ms": 59.44283283315599,
"samples_ms": [
59.17141702957451,
56.6554160322994,
58.45966702327132,
56.33254093118012,
59.44283283315599,
57.660290971398354
]
},
"candidate": {
"median_ms": 101.99858341366053,
"min_ms": 101.42887500114739,
"max_ms": 102.98116598278284,
"samples_ms": [
101.42887500114739,
102.98116598278284,
102.88925003260374,
101.66925005614758,
101.92508297041059,
102.07208385691047
]
},
"reduction_pct": -75.67795437601283,
"max_abs": 0.0,
"relative_rmse": 0.0,
"equal_fraction": 1.0
},
{
"group": "mpp",
"operator": "fc1",
"tile": [
64,
128,
256
],
"simdgroups": 8,
"walk": 8,
"order": [
"control",
"candidate",
"candidate",
"control",
"control",
"candidate",
"candidate",
"control",
"control",
"candidate",
"candidate",
"control"
],
"control": {
"median_ms": 63.49252094514668,
"min_ms": 62.75216583162546,
"max_ms": 63.604332972317934,
"samples_ms": [
62.75216583162546,
63.483166974037886,
63.529792008921504,
63.501874916255474,
63.604332972317934,
63.405083026736975
]
},
"candidate": {
"median_ms": 64.78197895921767,
"min_ms": 64.29212517105043,
"max_ms": 65.36220805719495,
"samples_ms": [
64.29212517105043,
64.32679202407598,
65.36220805719495,
64.67574997805059,
65.16262516379356,
64.88820794038475
]
},
"reduction_pct": -2.030881739890278,
"max_abs": 0.0,
"relative_rmse": 0.0,
"equal_fraction": 1.0
},
{
"group": "mpp",
"operator": "fc1",
"tile": [
128,
128,
128
],
"simdgroups": 4,
"walk": 0,
"order": [
"control",
"candidate",
"candidate",
"control",
"control",
"candidate",
"candidate",
"control",
"control",
"candidate",
"candidate",
"control"
],
"control": {
"median_ms": 63.39512497652322,
"min_ms": 62.838624930009246,
"max_ms": 63.97233298048377,
"samples_ms": [
63.315457897260785,
63.42300004325807,
63.51620890200138,
62.838624930009246,
63.97233298048377,
63.36724990978837
]
},
"candidate": {
"median_ms": 84.42670793738216,
"min_ms": 83.68350006639957,
"max_ms": 84.95029201731086,
"samples_ms": [
84.58908298052847,
84.26433289423585,
84.95029201731086,
83.68350006639957,
84.84283415600657,
83.99487496353686
]
},
"reduction_pct": -33.17539474627971,
"max_abs": 0.0,
"relative_rmse": 0.0,
"equal_fraction": 1.0
},
{
"group": "mpp",
"operator": "fc1",
"tile": [
128,
128,
128
],
"simdgroups": 4,
"walk": 8,
"order": [
"control",
"candidate",
"candidate",
"control",
"control",
"candidate",
"candidate",
"control",
"control",
"candidate",
"candidate",
"control"
],
"control": {
"median_ms": 63.20883345324546,
"min_ms": 62.85237497650087,
"max_ms": 64.06000000424683,
"samples_ms": [
62.92629218660295,
63.38979094289243,
64.06000000424683,
63.17124981433153,
63.24641709215939,
62.85237497650087
]
},
"candidate": {
"median_ms": 84.45014606695622,
"min_ms": 84.08629195764661,
"max_ms": 85.61154105700552,
"samples_ms": [
84.45587498135865,
84.08629195764661,
84.73183307796717,
84.4444171525538,
84.31354188360274,
85.61154105700552
]
},
"reduction_pct": -33.604974895514594,
"max_abs": 0.0,
"relative_rmse": 0.0,
"equal_fraction": 1.0
},
{
"group": "mpp",
"operator": "fc1",
"tile": [
128,
128,
256
],
"simdgroups": 4,
"walk": 0,
"order": [
"control",
"candidate",
"candidate",
"control",
"control",
"candidate",
"candidate",
"control",
"control",
"candidate",
"candidate",
"control"
],
"control": {
"median_ms": 63.586417003534734,
"min_ms": 63.08216694742441,
"max_ms": 63.97229200229049,
"samples_ms": [
63.40629211626947,
63.08216694742441,
63.97229200229049,
63.7665418908,
63.80291702225804,
63.30899987369776
]
},
"candidate": {
"median_ms": 78.20197858382016,
"min_ms": 77.38799997605383,
"max_ms": 78.6936660297215,
"samples_ms": [
78.35479103960097,
78.39841698296368,
78.6936660297215,
77.94012501835823,
78.04916612803936,
77.38799997605383
]
},
"reduction_pct": -22.98535169778941,
"max_abs": 0.0,
"relative_rmse": 0.0,
"equal_fraction": 1.0
},
{
"group": "mpp",
"operator": "fc1",
"tile": [
128,
128,
256
],
"simdgroups": 4,
"walk": 8,
"order": [
"control",
"candidate",
"candidate",
"control",
"control",
"candidate",
"candidate",
"control",
"control",
"candidate",
"candidate",
"control"
],
"control": {
"median_ms": 63.69983300101012,
"min_ms": 63.102665822952986,
"max_ms": 64.23237500712276,
"samples_ms": [
63.532917061820626,
63.84362489916384,
64.23237500712276,
63.72249987907708,
63.67716612294316,
63.102665822952986
]
},
"candidate": {
"median_ms": 77.84372905734926,
"min_ms": 76.57879102043808,
"max_ms": 78.39733292348683,
"samples_ms": [
78.39733292348683,
76.57879102043808,
78.0142080038786,
77.67325011081994,
78.15749989822507,
77.33812485821545
]
},
"reduction_pct": -22.203976666806735,
"max_abs": 0.0,
"relative_rmse": 0.0,
"equal_fraction": 1.0
},
{
"group": "mpp",
"operator": "fc1",
"tile": [
128,
128,
128
],
"simdgroups": 8,
"walk": 0,
"order": [
"control",
"candidate",
"candidate",
"control",
"control",
"candidate",
"candidate",
"control",
"control",
"candidate",
"candidate",
"control"
],
"control": {
"median_ms": 61.911812983453274,
"min_ms": 61.290707904845476,
"max_ms": 63.329082913696766,
"samples_ms": [
61.81354192085564,
62.010084046050906,
63.329082913696766,
61.48108304478228,
62.85750004462898,
61.290707904845476
]
},
"candidate": {
"median_ms": 71.19552057702094,
"min_ms": 70.41745795868337,
"max_ms": 71.96087506599724,
"samples_ms": [
71.56070787459612,
70.41745795868337,
71.10524992458522,
71.28579122945666,
71.96087506599724,
70.92629186809063
]
},
"reduction_pct": -14.995050453535995,
"max_abs": 0.0,
"relative_rmse": 0.0,
"equal_fraction": 1.0
},
{
"group": "mpp",
"operator": "fc1",
"tile": [
128,
128,
128
],
"simdgroups": 8,
"walk": 8,
"order": [
"control",
"candidate",
"candidate",
"control",
"control",
"candidate",
"candidate",
"control",
"control",
"candidate",
"candidate",
"control"
],
"control": {
"median_ms": 61.41318753361702,
"min_ms": 60.9392081387341,
"max_ms": 62.59725010022521,
"samples_ms": [
61.35716591961682,
61.39575014822185,
62.59725010022521,
61.43062491901219,
61.99399987235665,
60.9392081387341
]
},
"candidate": {
"median_ms": 66.05352088809013,
"min_ms": 65.7157909590751,
"max_ms": 67.56674987263978,
"samples_ms": [
66.79275003261864,
65.96341682597995,
67.56674987263978,
65.7157909590751,
65.87954191491008,
66.14362495020032
]
},
"reduction_pct": -7.555923313592938,
"max_abs": 0.0,
"relative_rmse": 0.0,
"equal_fraction": 1.0
},
{
"group": "mpp",
"operator": "fc1",
"tile": [
128,
128,
256
],
"simdgroups": 8,
"walk": 0,
"order": [
"control",
"candidate",
"candidate",
"control",
"control",
"candidate",
"candidate",
"control",
"control",
"candidate",
"candidate",
"control"
],
"control": {
"median_ms": 61.43754185177386,
"min_ms": 60.57624984532595,
"max_ms": 61.83066591620445,
"samples_ms": [
60.57624984532595,
61.51666701771319,
61.465916922315955,
61.40916678123176,
61.83066591620445,
61.09408289194107
]
},
"candidate": {
"median_ms": 68.41056258417666,
"min_ms": 67.89737497456372,
"max_ms": 68.78887512721121,
"samples_ms": [
67.97975022345781,
67.89737497456372,
68.78887512721121,
68.7539589125663,
68.60275007784367,
68.21837509050965
]
},
"reduction_pct": -11.349771690452926,
"max_abs": 0.0,
"relative_rmse": 0.0,
"equal_fraction": 1.0
},
{
"group": "mpp",
"operator": "fc1",
"tile": [
128,
128,
256
],
"simdgroups": 8,
"walk": 8,
"order": [
"control",
"candidate",
"candidate",
"control",
"control",
"candidate",
"candidate",
"control",
"control",
"candidate",
"candidate",
"control"
],
"control": {
"median_ms": 61.12581246998161,
"min_ms": 60.77320780605078,
"max_ms": 62.19662516377866,
"samples_ms": [
60.97987503744662,
61.35254097171128,
62.19662516377866,
61.095708049833775,
61.15591689012945,
60.77320780605078
]
},
"candidate": {
"median_ms": 63.03077097982168,
"min_ms": 62.66850000247359,
"max_ms": 64.16391697712243,
"samples_ms": [
64.16391697712243,
62.66850000247359,
63.280042028054595,
62.78149993158877,
63.48766596056521,
62.6695838291198
]
},
"reduction_pct": -3.1164551158736398,
"max_abs": 0.0,
"relative_rmse": 0.0,
"equal_fraction": 1.0
},
{
"group": "mpp",
"operator": "fc1",
"tile": [
128,
256,
128
],
"simdgroups": 8,
"walk": 0,
"order": [
"control",
"candidate",
"candidate",
"control",
"control",
"candidate",
"candidate",
"control",
"control",
"candidate",
"candidate",
"control"
],
"control": {
"median_ms": 60.211145551875234,
"min_ms": 59.43237501196563,
"max_ms": 61.478584073483944,
"samples_ms": [
60.14295807108283,
59.43237501196563,
59.89066604524851,
60.27933303266764,
60.47937506809831,
61.478584073483944
]
},
"candidate": {
"median_ms": 89.8966669337824,
"min_ms": 87.6690000295639,
"max_ms": 90.39637516252697,
"samples_ms": [
89.41675000824034,
87.6690000295639,
89.42070882767439,
90.39220795966685,
90.37262503989041,
90.39637516252697
]
},
"reduction_pct": -49.302369370022106,
"max_abs": 0.0,
"relative_rmse": 0.0,
"equal_fraction": 1.0
},
{
"group": "mpp",
"operator": "fc1",
"tile": [
128,
256,
128
],
"simdgroups": 8,
"walk": 8,
"order": [
"control",
"candidate",
"candidate",
"control",
"control",
"candidate",
"candidate",
"control",
"control",
"candidate",
"candidate",
"control"
],
"control": {
"median_ms": 59.9094380158931,
"min_ms": 59.31204091757536,
"max_ms": 61.26745790243149,
"samples_ms": [
59.31204091757536,
59.49704209342599,
60.211583971977234,
60.44662487693131,
61.26745790243149,
59.60729205980897
]
},
"candidate": {
"median_ms": 89.09420855343342,
"min_ms": 88.59087503515184,
"max_ms": 90.01100016757846,
"samples_ms": [
88.59087503515184,
88.79987499676645,
89.5525838714093,
89.09108396619558,
90.01100016757846,
89.09733314067125
]
},
"reduction_pct": -48.714812730838865,
"max_abs": 0.0,
"relative_rmse": 0.0,
"equal_fraction": 1.0
},
{
"group": "mpp",
"operator": "fc1",
"tile": [
128,
256,
256
],
"simdgroups": 8,
"walk": 0,
"order": [
"control",
"candidate",
"candidate",
"control",
"control",
"candidate",
"candidate",
"control",
"control",
"candidate",
"candidate",
"control"
],
"control": {
"median_ms": 61.477103969082236,
"min_ms": 60.560583136975765,
"max_ms": 61.69387511909008,
"samples_ms": [
60.560583136975765,
61.50449998676777,
61.63812498562038,
61.449707951396704,
61.69387511909008,
60.781167121604085
]
},
"candidate": {
"median_ms": 79.18108406011015,
"min_ms": 78.36299994960427,
"max_ms": 79.72500007599592,
"samples_ms": [
79.4039589818567,
78.9582091383636,
79.65074991807342,
78.36299994960427,
79.72500007599592,
78.7826671730727
]
},
"reduction_pct": -28.797680677885396,
"max_abs": 0.0,
"relative_rmse": 0.0,
"equal_fraction": 1.0
},
{
"group": "mpp",
"operator": "fc1",
"tile": [
128,
256,
256
],
"simdgroups": 8,
"walk": 8,
"order": [
"control",
"candidate",
"candidate",
"control",
"control",
"candidate",
"candidate",
"control",
"control",
"candidate",
"candidate",
"control"
],
"control": {
"median_ms": 60.90152100659907,
"min_ms": 59.61754103191197,
"max_ms": 61.46008404903114,
"samples_ms": [
60.843209037557244,
59.61754103191197,
60.77250000089407,
60.95983297564089,
61.46008404903114,
61.00125005468726
]
},
"candidate": {
"median_ms": 79.39420791808516,
"min_ms": 77.88962498307228,
"max_ms": 80.29516600072384,
"samples_ms": [
77.92449998669326,
77.88962498307228,
79.05454095453024,
79.73387488164008,
80.29516600072384,
80.0469999667257
]
},
"reduction_pct": -30.364901575253423,
"max_abs": 0.0,
"relative_rmse": 0.0,
"equal_fraction": 1.0
},
{
"group": "mpp",
"operator": "fc1",
"tile": [
64,
256,
128
],
"simdgroups": 8,
"walk": 0,
"order": [
"control",
"candidate",
"candidate",
"control",
"control",
"candidate",
"candidate",
"control",
"control",
"candidate",
"candidate",
"control"
],
"control": {
"median_ms": 57.857624953612685,
"min_ms": 55.96287501975894,
"max_ms": 60.450416058301926,
"samples_ms": [
57.10191698744893,
59.08837518654764,
60.450416058301926,
56.51841592043638,
58.61333291977644,
55.96287501975894
]
},
"candidate": {
"median_ms": 100.25337513070554,
"min_ms": 99.88987492397428,
"max_ms": 100.98091699182987,
"samples_ms": [
100.03845812752843,
100.3146250732243,
99.88987492397428,
100.97149992361665,
100.19212518818676,
100.98091699182987
]
},
"reduction_pct": -73.27599466981857,
"max_abs": 0.0,
"relative_rmse": 0.0,
"equal_fraction": 1.0
},
{
"group": "mpp",
"operator": "fc1",
"tile": [
64,
256,
128
],
"simdgroups": 8,
"walk": 8,
"order": [
"control",
"candidate",
"candidate",
"control",
"control",
"candidate",
"candidate",
"control",
"control",
"candidate",
"candidate",
"control"
],
"control": {
"median_ms": 61.68931245338172,
"min_ms": 60.960167087614536,
"max_ms": 65.81370788626373,
"samples_ms": [
61.02383299730718,
61.06674997135997,
60.960167087614536,
62.311874935403466,
63.711458118632436,
65.81370788626373
]
},
"candidate": {
"median_ms": 66.21014606207609,
"min_ms": 65.9826670307666,
"max_ms": 71.96287508122623,
"samples_ms": [
66.2466671783477,
65.9826670307666,
66.04887498542666,
66.17362494580448,
70.00458403490484,
71.96287508122623
]
},
"reduction_pct": -7.328390330352197,
"max_abs": 0.0,
"relative_rmse": 0.0,
"equal_fraction": 1.0
},
{
"group": "mpp",
"operator": "fc1",
"tile": [
64,
256,
256
],
"simdgroups": 8,
"walk": 0,
"order": [
"control",
"candidate",
"candidate",
"control",
"control",
"candidate",
"candidate",
"control",
"control",
"candidate",
"candidate",
"control"
],
"control": {
"median_ms": 59.39589545596391,
"min_ms": 58.55387495830655,
"max_ms": 60.08745916187763,
"samples_ms": [
60.047625098377466,
59.607790783047676,
60.08745916187763,
58.55387495830655,
59.18400012888014,
59.135750168934464
]
},
"candidate": {
"median_ms": 99.15185393765569,
"min_ms": 98.92800007946789,
"max_ms": 99.65758305042982,
"samples_ms": [
99.01033388450742,
98.92800007946789,
99.1397078614682,
99.56570877693594,
99.16400001384318,
99.65758305042982
]
},
"reduction_pct": -66.93384816660746,
"max_abs": 0.0,
"relative_rmse": 0.0,
"equal_fraction": 1.0
},
{
"group": "mpp",
"operator": "fc1",
"tile": [
64,
256,
256
],
"simdgroups": 8,
"walk": 8,
"order": [
"control",
"candidate",
"candidate",
"control",
"control",
"candidate",
"candidate",
"control",
"control",
"candidate",
"candidate",
"control"
],
"control": {
"median_ms": 64.98860463034362,
"min_ms": 60.69420883432031,
"max_ms": 65.99575001746416,
"samples_ms": [
60.69420883432031,
63.48704104311764,
64.50816709548235,
65.77125005424023,
65.99575001746416,
65.46904216520488
]
},
"candidate": {
"median_ms": 68.07666691020131,
"min_ms": 65.6382089946419,
"max_ms": 68.59708297997713,
"samples_ms": [
65.6382089946419,
66.77141692489386,
68.59708297997713,
68.15712503157556,
68.14566696994007,
68.00766685046256
]
},
"reduction_pct": -4.751698082183253,
"max_abs": 0.0,
"relative_rmse": 0.0,
"equal_fraction": 1.0
},
{
"group": "unpack",
"operator": "qkv",
"lanes": 2,
"threads": 128,
"bit_equal": true,
"decode": {
"order": [
"control",
"candidate",
"candidate",
"control",
"control",
"candidate",
"candidate",
"control",
"control",
"candidate",
"candidate",
"control"
],
"control": {
"median_ms": 3.012958448380232,
"min_ms": 2.721833996474743,
"max_ms": 3.4094578586518764,
"samples_ms": [
2.7550829108804464,
3.2708339858800173,
3.4094578586518764,
2.721833996474743,
3.295874921604991,
2.7246668469160795
]
},
"candidate": {
"median_ms": 2.8923124773427844,
"min_ms": 2.3615839891135693,
"max_ms": 3.126000054180622,
"samples_ms": [
3.126000054180622,
2.879916923120618,
2.3615839891135693,
2.362750004976988,
2.919208025559783,
2.904708031564951
]
},
"reduction_pct": 4.004236138812556
},
"with_gemm": {
"order": [
"control",
"candidate",
"candidate",
"control",
"control",
"candidate",
"candidate",
"control",
"control",
"candidate",
"candidate",
"control"
],
"control": {
"median_ms": 49.968103994615376,
"min_ms": 46.252084197476506,
"max_ms": 52.3542920127511,
"samples_ms": [
46.252084197476506,
47.39858419634402,
49.9996249563992,
51.00825009867549,
52.3542920127511,
49.93658303283155
]
},
"candidate": {
"median_ms": 48.895916901528835,
"min_ms": 47.413749853149056,
"max_ms": 52.04641702584922,
"samples_ms": [
47.413749853149056,
48.56554185971618,
47.66920814290643,
51.641334081068635,
52.04641702584922,
49.226291943341494
]
},
"reduction_pct": 2.145742998777944
}
},
{
"group": "unpack",
"operator": "qkv",
"lanes": 2,
"threads": 256,
"bit_equal": true,
"decode": {
"order": [
"control",
"candidate",
"candidate",
"control",
"control",
"candidate",
"candidate",
"control",
"control",
"candidate",
"candidate",
"control"
],
"control": {
"median_ms": 2.797103952616453,
"min_ms": 2.7537078130990267,
"max_ms": 2.9348330572247505,
"samples_ms": [
2.778958063572645,
2.8350839857012033,
2.8144579846411943,
2.779749920591712,
2.9348330572247505,
2.7537078130990267
]
},
"candidate": {
"median_ms": 2.4032086366787553,
"min_ms": 2.3526670411229134,
"max_ms": 2.5988752022385597,
"samples_ms": [
2.42033414542675,
2.5988752022385597,
2.3698341101408005,
2.3526670411229134,
2.4251670110970736,
2.3860831279307604
]
},
"reduction_pct": 14.082255168573276
},
"with_gemm": {
"order": [
"control",
"candidate",
"candidate",
"control",
"control",
"candidate",
"candidate",
"control",
"control",
"candidate",
"candidate",
"control"
],
"control": {
"median_ms": 49.87804149277508,
"min_ms": 48.51629096083343,
"max_ms": 52.49258317053318,
"samples_ms": [
52.49258317053318,
51.22725013643503,
50.627707969397306,
49.12837501615286,
49.09895802848041,
48.51629096083343
]
},
"candidate": {
"median_ms": 49.28572895005345,
"min_ms": 47.99925000406802,
"max_ms": 51.74404103308916,
"samples_ms": [
51.74404103308916,
51.171874860301614,
49.18862506747246,
49.38283283263445,
47.99925000406802,
48.066750168800354
]
},
"reduction_pct": 1.1875216528047239
}
},
{
"group": "unpack",
"operator": "qkv",
"lanes": 4,
"threads": 128,
"bit_equal": true,
"decode": {
"order": [
"control",
"candidate",
"candidate",
"control",
"control",
"candidate",
"candidate",
"control",
"control",
"candidate",
"candidate",
"control"
],
"control": {
"median_ms": 3.3383540576323867,
"min_ms": 2.943041967228055,
"max_ms": 3.570500062778592,
"samples_ms": [
3.4179158974438906,
2.943041967228055,
3.0063330195844173,
3.3247501123696566,
3.351958002895117,
3.570500062778592
]
},
"candidate": {
"median_ms": 2.668646047823131,
"min_ms": 2.437292132526636,
"max_ms": 3.00904200412333,
"samples_ms": [
3.00904200412333,
2.879292005673051,
2.437292132526636,
2.949667163193226,
2.4580000899732113,
2.452624961733818
]
},
"reduction_pct": 20.061024033029707
},
"with_gemm": {
"order": [
"control",
"candidate",
"candidate",
"control",
"control",
"candidate",
"candidate",
"control",
"control",
"candidate",
"candidate",
"control"
],
"control": {
"median_ms": 48.19112503901124,
"min_ms": 46.45187500864267,
"max_ms": 49.081167206168175,
"samples_ms": [
46.45187500864267,
48.059541964903474,
47.92812489904463,
48.322708113119006,
48.849083948880434,
49.081167206168175
]
},
"candidate": {
"median_ms": 47.92097909376025,
"min_ms": 46.94141703657806,
"max_ms": 48.69379196316004,
"samples_ms": [
46.94141703657806,
47.722374787554145,
47.966625075787306,
47.8753331117332,
48.69379196316004,
48.30574989318848
]
},
"reduction_pct": 0.5605719829788192
}
},
{
"group": "unpack",
"operator": "qkv",
"lanes": 4,
"threads": 256,
"bit_equal": true,
"decode": {
"order": [
"control",
"candidate",
"candidate",
"control",
"control",
"candidate",
"candidate",
"control",
"control",
"candidate",
"candidate",
"control"
],
"control": {
"median_ms": 2.8086245292797685,
"min_ms": 2.747541991993785,
"max_ms": 2.934583928436041,
"samples_ms": [
2.934583928436041,
2.7692499570548534,
2.8094579465687275,
2.8877498116344213,
2.747541991993785,
2.8077911119908094
]
},
"candidate": {
"median_ms": 2.3740000324323773,
"min_ms": 2.3532919585704803,
"max_ms": 2.4597500450909138,
"samples_ms": [
2.4221669882535934,
2.3592079523950815,
2.388792112469673,
2.4597500450909138,
2.35754088498652,
2.3532919585704803
]
},
"reduction_pct": 15.474638646656135
},
"with_gemm": {
"order": [
"control",
"candidate",
"candidate",
"control",
"control",
"candidate",
"candidate",
"control",
"control",
"candidate",
"candidate",
"control"
],
"control": {
"median_ms": 48.76722896005958,
"min_ms": 48.59474999830127,
"max_ms": 49.04170800000429,
"samples_ms": [
48.59474999830127,
48.837542068213224,
48.6438749358058,
48.71054203249514,
49.04170800000429,
48.823915887624025
]
},
"candidate": {
"median_ms": 48.40804194100201,
"min_ms": 48.10195881873369,
"max_ms": 48.87629207223654,
"samples_ms": [
48.302749870345,
48.513334011659026,
48.8370000384748,
48.87629207223654,
48.10195881873369,
48.17000008188188
]
},
"reduction_pct": 0.7365335835500253
}
},
{
"group": "unpack",
"operator": "qkv",
"lanes": 8,
"threads": 128,
"bit_equal": true,
"decode": {
"order": [
"control",
"candidate",
"candidate",
"control",
"control",
"candidate",
"candidate",
"control",
"control",
"candidate",
"candidate",
"control"
],
"control": {
"median_ms": 2.8508331160992384,
"min_ms": 2.696041017770767,
"max_ms": 3.293875139206648,
"samples_ms": [
3.293875139206648,
2.806000178679824,
2.696041017770767,
2.895666053518653,
2.9558748938143253,
2.7021251153200865
]
},
"candidate": {
"median_ms": 2.573416568338871,
"min_ms": 2.3308340460062027,
"max_ms": 3.1028748489916325,
"samples_ms": [
2.5110410060733557,
2.6357921306043863,
2.8954169247299433,
3.1028748489916325,
2.376917051151395,
2.3308340460062027
]
},
"reduction_pct": 9.731069356313405
},
"with_gemm": {
"order": [
"control",
"candidate",
"candidate",
"control",
"control",
"candidate",
"candidate",
"control",
"control",
"candidate",
"candidate",
"control"
],
"control": {
"median_ms": 49.64439547620714,
"min_ms": 47.31045803055167,
"max_ms": 49.88791584037244,
"samples_ms": [
47.31045803055167,
49.58124994300306,
49.707541009411216,
49.88791584037244,
49.46058290079236,
49.8622499871999
]
},
"candidate": {
"median_ms": 49.150792066939175,
"min_ms": 48.532291082665324,
"max_ms": 49.52250001952052,
"samples_ms": [
48.77670807763934,
48.532291082665324,
49.33358309790492,
49.21541712246835,
49.52250001952052,
49.08616701141
]
},
"reduction_pct": 0.9942782151603158
}
},
{
"group": "unpack",
"operator": "qkv",
"lanes": 8,
"threads": 256,
"bit_equal": true,
"decode": {
"order": [
"control",
"candidate",
"candidate",
"control",
"control",
"candidate",
"candidate",
"control",
"control",
"candidate",
"candidate",
"control"
],
"control": {
"median_ms": 3.0089375795796514,
"min_ms": 2.972792135551572,
"max_ms": 3.110417164862156,
"samples_ms": [
2.995207905769348,
3.082999959588051,
3.110417164862156,
2.972792135551572,
3.0096250120550394,
3.0082501471042633
]
},
"candidate": {
"median_ms": 2.676436910405755,
"min_ms": 2.6604998856782913,
"max_ms": 2.7663339860737324,
"samples_ms": [
2.667415887117386,
2.663416089490056,
2.7663339860737324,
2.6604998856782913,
2.685457933694124,
2.685458166524768
]
},
"reduction_pct": 11.050434260598607
},
"with_gemm": {
"order": [
"control",
"candidate",
"candidate",
"control",
"control",
"candidate",
"candidate",
"control",
"control",
"candidate",
"candidate",
"control"
],
"control": {
"median_ms": 48.95018809475005,
"min_ms": 48.11400012113154,
"max_ms": 49.70504203811288,
"samples_ms": [
48.11400012113154,
48.64716716110706,
48.56958310119808,
49.25320902839303,
49.42716704681516,
49.70504203811288
]
},
"candidate": {
"median_ms": 48.75424993224442,
"min_ms": 47.985499957576394,
"max_ms": 49.37275010161102,
"samples_ms": [
48.16216602921486,
47.985499957576394,
48.71866689063609,
48.789832973852754,
49.003709107637405,
49.37275010161102
]
},
"reduction_pct": 0.4002807141953335
}
},
{
"group": "unpack",
"operator": "fc2",
"lanes": 2,
"threads": 128,
"bit_equal": true,
"decode": {
"order": [
"control",
"candidate",
"candidate",
"control",
"control",
"candidate",
"candidate",
"control",
"control",
"candidate",
"candidate",
"control"
],
"control": {
"median_ms": 2.1764999255537987,
"min_ms": 1.6474579460918903,
"max_ms": 2.326082903891802,
"samples_ms": [
2.221875125542283,
1.714083133265376,
1.6474579460918903,
2.1587498486042023,
2.194250002503395,
2.326082903891802
]
},
"candidate": {
"median_ms": 2.028645947575569,
"min_ms": 1.4096670784056187,
"max_ms": 2.136250026524067,
"samples_ms": [
2.0127499010413885,
1.9904589280486107,
1.4096670784056187,
2.136250026524067,
2.04454199410975,
2.045540837571025
]
},
"reduction_pct": 6.79319931245157
},
"with_gemm": {
"order": [
"control",
"candidate",
"candidate",
"control",
"control",
"candidate",
"candidate",
"control",
"control",
"candidate",
"candidate",
"control"
],
"control": {
"median_ms": 36.52379149571061,
"min_ms": 36.12149995751679,
"max_ms": 36.804832983762026,
"samples_ms": [
36.58037493005395,
36.38229100033641,
36.12149995751679,
36.63479210808873,
36.46720806136727,
36.804832983762026
]
},
"candidate": {
"median_ms": 36.16639552637935,
"min_ms": 35.967209143564105,
"max_ms": 36.6698750294745,
"samples_ms": [
36.2744580488652,
35.967209143564105,
36.0406250692904,
36.058333003893495,
36.6698750294745,
36.32033383473754
]
},
"reduction_pct": 0.9785292125907508
}
},
{
"group": "unpack",
"operator": "fc2",
"lanes": 2,
"threads": 256,
"bit_equal": true,
"decode": {
"order": [
"control",
"candidate",
"candidate",
"control",
"control",
"candidate",
"candidate",
"control",
"control",
"candidate",
"candidate",
"control"
],
"control": {
"median_ms": 1.591791631653905,
"min_ms": 1.5801249537616968,
"max_ms": 1.723709050565958,
"samples_ms": [
1.5801249537616968,
1.6954168677330017,
1.723709050565958,
1.6006662044674158,
1.582917058840394,
1.5815829392522573
]
},
"candidate": {
"median_ms": 1.359250396490097,
"min_ms": 1.3114579487591982,
"max_ms": 1.3992919120937586,
"samples_ms": [
1.3992919120937586,
1.3679589610546827,
1.3595419004559517,
1.3589588925242424,
1.335666049271822,
1.3114579487591982
]
},
"reduction_pct": 14.608773569326573
},
"with_gemm": {
"order": [
"control",
"candidate",
"candidate",
"control",
"control",
"candidate",
"candidate",
"control",
"control",
"candidate",
"candidate",
"control"
],
"control": {
"median_ms": 37.30770805850625,
"min_ms": 36.52079193852842,
"max_ms": 37.82283398322761,
"samples_ms": [
36.52079193852842,
37.675000028684735,
37.024415796622634,
37.82283398322761,
37.18841611407697,
37.42700000293553
]
},
"candidate": {
"median_ms": 37.264708429574966,
"min_ms": 36.200499860569835,
"max_ms": 37.65112510882318,
"samples_ms": [
36.29908314906061,
36.200499860569835,
37.45891689322889,
37.65112510882318,
37.1370839420706,
37.39233291707933
]
},
"reduction_pct": 0.11525668868174188
}
},
{
"group": "unpack",
"operator": "fc2",
"lanes": 4,
"threads": 128,
"bit_equal": true,
"decode": {
"order": [
"control",
"candidate",
"candidate",
"control",
"control",
"candidate",
"candidate",
"control",
"control",
"candidate",
"candidate",
"control"
],
"control": {
"median_ms": 1.6842915210872889,
"min_ms": 1.571665983647108,
"max_ms": 1.8738750368356705,
"samples_ms": [
1.6727910842746496,
1.5757500659674406,
1.571665983647108,
1.695791957899928,
1.8738750368356705,
1.6997500788420439
]
},
"candidate": {
"median_ms": 1.3427914818748832,
"min_ms": 1.2893339153379202,
"max_ms": 1.4460829552263021,
"samples_ms": [
1.3589579612016678,
1.3011249247938395,
1.2893339153379202,
1.3266250025480986,
1.4460829552263021,
1.361542148515582
]
},
"reduction_pct": 20.275589762035462
},
"with_gemm": {
"order": [
"control",
"candidate",
"candidate",
"control",
"control",
"candidate",
"candidate",
"control",
"control",
"candidate",
"candidate",
"control"
],
"control": {
"median_ms": 36.98645858094096,
"min_ms": 36.061666905879974,
"max_ms": 38.04287500679493,
"samples_ms": [
36.061666905879974,
36.470916820690036,
36.80479200556874,
37.16812515631318,
37.33900003135204,
38.04287500679493
]
},
"candidate": {
"median_ms": 37.370853940956295,
"min_ms": 36.50595899671316,
"max_ms": 38.162291049957275,
"samples_ms": [
36.50595899671316,
37.39512502215803,
37.01962507329881,
37.34658285975456,
38.162291049957275,
38.043915992602706
]
},
"reduction_pct": -1.0392867410490947
}
},
{
"group": "unpack",
"operator": "fc2",
"lanes": 4,
"threads": 256,
"bit_equal": true,
"decode": {
"order": [
"control",
"candidate",
"candidate",
"control",
"control",
"candidate",
"candidate",
"control",
"control",
"candidate",
"candidate",
"control"
],
"control": {
"median_ms": 1.765687484294176,
"min_ms": 1.7103329300880432,
"max_ms": 1.9108750857412815,
"samples_ms": [
1.9108750857412815,
1.7783751245588064,
1.7529998440295458,
1.7103329300880432,
1.7298748716711998,
1.86187494546175
]
},
"candidate": {
"median_ms": 1.4076875522732735,
"min_ms": 1.3741659931838512,
"max_ms": 1.480249920859933,
"samples_ms": [
1.4768331311643124,
1.480249920859933,
1.3874578289687634,
1.3741659931838512,
1.4230830129235983,
1.3922920916229486
]
},
"reduction_pct": 20.275384812166298
},
"with_gemm": {
"order": [
"control",
"candidate",
"candidate",
"control",
"control",
"candidate",
"candidate",
"control",
"control",
"candidate",
"candidate",
"control"
],
"control": {
"median_ms": 38.445541518740356,
"min_ms": 37.49095811508596,
"max_ms": 39.76120799779892,
"samples_ms": [
37.49095811508596,
37.90116705931723,
38.37700001895428,
38.514083018526435,
38.57575007714331,
39.76120799779892
]
},
"candidate": {
"median_ms": 38.93037501256913,
"min_ms": 37.44658292271197,
"max_ms": 39.427584037184715,
"samples_ms": [
37.44658292271197,
37.45900001376867,
38.869749987497926,
39.31016707792878,
38.99100003764033,
39.427584037184715
]
},
"reduction_pct": -1.2610915978187975
}
},
{
"group": "unpack",
"operator": "fc2",
"lanes": 8,
"threads": 128,
"bit_equal": true,
"decode": {
"order": [
"control",
"candidate",
"candidate",
"control",
"control",
"candidate",
"candidate",
"control",
"control",
"candidate",
"candidate",
"control"
],
"control": {
"median_ms": 1.7134579829871655,
"min_ms": 1.6902079805731773,
"max_ms": 1.8400829285383224,
"samples_ms": [
1.7334579024463892,
1.8400829285383224,
1.69179099611938,
1.6934580635279417,
1.6902079805731773,
1.7727920785546303
]
},
"candidate": {
"median_ms": 1.429312047548592,
"min_ms": 1.4042500406503677,
"max_ms": 1.5134159475564957,
"samples_ms": [
1.5043749008327723,
1.5134159475564957,
1.4151660725474358,
1.412124838680029,
1.4434580225497484,
1.4042500406503677
]
},
"reduction_pct": 16.583186647110317
},
"with_gemm": {
"order": [
"control",
"candidate",
"candidate",
"control",
"control",
"candidate",
"candidate",
"control",
"control",
"candidate",
"candidate",
"control"
],
"control": {
"median_ms": 39.86033354885876,
"min_ms": 39.17104098945856,
"max_ms": 40.33637512475252,
"samples_ms": [
39.51695910654962,
40.33637512475252,
40.28416587971151,
39.17104098945856,
39.618375012651086,
40.10229208506644
]
},
"candidate": {
"median_ms": 39.79412489570677,
"min_ms": 39.281709119677544,
"max_ms": 40.1190840639174,
"samples_ms": [
39.690874982625246,
39.76074978709221,
39.281709119677544,
40.024416986852884,
40.1190840639174,
39.82750000432134
]
},
"reduction_pct": 0.16610160341692382
}
},
{
"group": "unpack",
"operator": "fc2",
"lanes": 8,
"threads": 256,
"bit_equal": true,
"decode": {
"order": [
"control",
"candidate",
"candidate",
"control",
"control",
"candidate",
"candidate",
"control",
"control",
"candidate",
"candidate",
"control"
],
"control": {
"median_ms": 1.8059374997392297,
"min_ms": 1.7295829020440578,
"max_ms": 1.976290950551629,
"samples_ms": [
1.976290950551629,
1.744791865348816,
1.80658302269876,
1.8052919767796993,
1.8619170878082514,
1.7295829020440578
]
},
"candidate": {
"median_ms": 1.5007710317149758,
"min_ms": 1.468417001888156,
"max_ms": 1.6903330106288195,
"samples_ms": [
1.6035421285778284,
1.485415967181325,
1.4942919369786978,
1.507250126451254,
1.6903330106288195,
1.468417001888156
]
},
"reduction_pct": 16.89795289528673
},
"with_gemm": {
"order": [
"control",
"candidate",
"candidate",
"control",
"control",
"candidate",
"candidate",
"control",
"control",
"candidate",
"candidate",
"control"
],
"control": {
"median_ms": 39.135062135756016,
"min_ms": 38.90095790848136,
"max_ms": 39.62595900520682,
"samples_ms": [
39.13941606879234,
39.40516686998308,
38.95041602663696,
39.13070820271969,
38.90095790848136,
39.62595900520682
]
},
"candidate": {
"median_ms": 39.152541547082365,
"min_ms": 38.64799998700619,
"max_ms": 39.39266689121723,
"samples_ms": [
38.955500116571784,
38.64799998700619,
39.15004199370742,
39.23345799557865,
39.39266689121723,
39.15504110045731
]
},
"reduction_pct": -0.044664324961884994
}
}
],
"activation_ranges": {
"qkv": {
"shape": [
4181,
5376
],
"max_abs": 10.0625
},
"out": {
"shape": [
4181,
7168
],
"max_abs": 38.75
},
"fc1": {
"shape": [
4181,
5376
],
"max_abs": 3.21875
},
"fc2": {
"shape": [
4181,
14336
],
"max_abs": 4640.0
}
},
"memory": {
"sampled_peak_physical_bytes": 4787685752,
"lifetime_peak_physical_bytes": 5052712832,
"sample_interval_s": 0.1,
"sample_count": 682,
"scope": "Parent process only; includes Metal; excludes compiler XPC services. Counters overlap; do not sum."
}
}
{
"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"
}
},
"scope": "Actual saved first-block activations, one GPU job, counterbalanced warmed calls including layout/decode costs. Operator evidence only.",
"trials": [
{
"group": "attention",
"variant": "q",
"order": [
"control",
"candidate",
"candidate",
"control",
"control",
"candidate",
"candidate",
"control",
"control",
"candidate",
"candidate",
"control"
],
"control": {
"median_ms": 25.229458115063608,
"min_ms": 25.031917029991746,
"max_ms": 25.60575003735721,
"samples_ms": [
25.268583092838526,
25.075667072087526,
25.19033313728869,
25.031917029991746,
25.60575003735721,
25.284667033702135
]
},
"candidate": {
"median_ms": 25.469625601544976,
"min_ms": 25.04358277656138,
"max_ms": 27.149833040311933,
"samples_ms": [
25.305458111688495,
25.851749815046787,
25.04358277656138,
25.56595904752612,
25.37329215556383,
27.149833040311933
]
},
"reduction_pct": -0.9519327977083103,
"max_abs": 0.0,
"relative_rmse": 0.0,
"equal_fraction": 1.0
},
{
"group": "attention",
"variant": "k",
"order": [
"control",
"candidate",
"candidate",
"control",
"control",
"candidate",
"candidate",
"control",
"control",
"candidate",
"candidate",
"control"
],
"control": {
"median_ms": 25.859437533654273,
"min_ms": 25.340416003018618,
"max_ms": 26.08179091475904,
"samples_ms": [
25.340416003018618,