typical-medium / bench.json
guychuk's picture
typical-medium (tm1b): weights, evals, inference package, manifest
970f2df verified
Raw History Blame Contribute Delete
13 kB
{
"model": "runs/tm1b",
"energy_model": null,
"backbone": "Qwen/Qwen3-4B-Base",
"quick": false,
"b_chunk": 4,
"native": {
"256": {
"2": {
"state_tokens": 256,
"suffix_tokens": 22,
"nc_chunk": 8,
"native": [
{
"m": 1,
"t_state_ms": 17.69845560193062,
"t_queries_ms": 38.907185196876526,
"total_ms": 56.605640798807144,
"ratio": 1.0,
"marginal_ms": 38.907185196876526
},
{
"m": 32,
"t_state_ms": 17.29227602481842,
"t_queries_ms": 107.14780166745186,
"total_ms": 124.44007769227028,
"ratio": 2.1983688539904773,
"marginal_ms": 3.3483688021078706
},
{
"m": 256,
"t_state_ms": 17.395589500665665,
"t_queries_ms": 720.7270376384258,
"total_ms": 738.1226271390915,
"ratio": 13.039736265200023,
"marginal_ms": 2.815339990775101
}
],
"native_peak_mem_mb": 14565.473792,
"b_batched": [
{
"m": 1,
"t_state_ms": 19.00634914636612,
"t_queries_ms": 65.67803397774696,
"total_ms": 84.68438312411308,
"ratio": 1.0,
"marginal_ms": 65.67803397774696
},
{
"m": 32,
"t_state_ms": 19.073322415351868,
"t_queries_ms": 408.54376927018166,
"total_ms": 427.6170916855335,
"ratio": 5.0495389576000065,
"marginal_ms": 12.766992789693177
},
{
"m": 256,
"t_state_ms": 19.45146918296814,
"t_queries_ms": 3124.6551759541035,
"total_ms": 3144.1066451370716,
"ratio": 37.127348976836515,
"marginal_ms": 12.205684281070717
}
],
"b_batched_peak_mem_mb": 15050.985472
},
"4": {
"state_tokens": 256,
"suffix_tokens": 34,
"nc_chunk": 8,
"native": [
{
"m": 1,
"t_state_ms": 17.520517110824585,
"t_queries_ms": 39.11682963371277,
"total_ms": 56.63734674453735,
"ratio": 1.0,
"marginal_ms": 39.11682963371277
},
{
"m": 32,
"t_state_ms": 17.56681129336357,
"t_queries_ms": 109.09038409590721,
"total_ms": 126.65719538927078,
"ratio": 2.2362840540634403,
"marginal_ms": 3.4090745029971004
},
{
"m": 256,
"t_state_ms": 17.606686800718307,
"t_queries_ms": 820.20478323102,
"total_ms": 837.8114700317383,
"ratio": 14.792562120021005,
"marginal_ms": 3.2039249344961718
}
],
"native_peak_mem_mb": 14579.467776,
"b_batched": [
{
"m": 1,
"t_state_ms": 19.45377141237259,
"t_queries_ms": 66.85380637645721,
"total_ms": 86.3075777888298,
"ratio": 1.0,
"marginal_ms": 66.85380637645721
},
{
"m": 32,
"t_state_ms": 19.49465647339821,
"t_queries_ms": 415.58846831321716,
"total_ms": 435.0831247866154,
"ratio": 5.041076762125575,
"marginal_ms": 12.987139634788036
},
{
"m": 256,
"t_state_ms": 19.42378655076027,
"t_queries_ms": 3201.4211788773537,
"total_ms": 3220.844965428114,
"ratio": 37.31821756495831,
"marginal_ms": 12.505551479989663
}
],
"b_batched_peak_mem_mb": 15418.156544
},
"10": {
"state_tokens": 256,
"suffix_tokens": 70,
"nc_chunk": 8,
"native": [
{
"m": 1,
"t_state_ms": 17.71542802453041,
"t_queries_ms": 39.49485346674919,
"total_ms": 57.2102814912796,
"ratio": 1.0,
"marginal_ms": 39.49485346674919
},
{
"m": 32,
"t_state_ms": 17.667438834905624,
"t_queries_ms": 113.24474588036537,
"total_ms": 130.912184715271,
"ratio": 2.288263250989694,
"marginal_ms": 3.538898308761418
},
{
"m": 256,
"t_state_ms": 17.80020445585251,
"t_queries_ms": 772.7814465761185,
"total_ms": 790.581651031971,
"ratio": 13.818873643411054,
"marginal_ms": 3.0186775256879628
}
],
"native_peak_mem_mb": 14627.14112,
"b_batched": [
{
"m": 1,
"t_state_ms": 19.491098821163177,
"t_queries_ms": 67.33075901865959,
"total_ms": 86.82185783982277,
"ratio": 1.0,
"marginal_ms": 67.33075901865959
},
{
"m": 32,
"t_state_ms": 19.476331770420074,
"t_queries_ms": 518.6272598803043,
"total_ms": 538.1035916507244,
"ratio": 6.19778941661752,
"marginal_ms": 16.20710187125951
},
{
"m": 256,
"t_state_ms": 19.5954367518425,
"t_queries_ms": 4071.4379996061325,
"total_ms": 4091.033436357975,
"ratio": 47.11985596882185,
"marginal_ms": 15.904054685961455
}
],
"b_batched_peak_mem_mb": 16549.823488
},
"32": {
"state_tokens": 256,
"suffix_tokens": 224,
"nc_chunk": 8,
"native": [
{
"m": 1,
"t_state_ms": 17.933987081050873,
"t_queries_ms": 40.07749259471893,
"total_ms": 58.011479675769806,
"ratio": 1.0,
"marginal_ms": 40.07749259471893
},
{
"m": 32,
"t_state_ms": 17.875414341688156,
"t_queries_ms": 199.15443286299706,
"total_ms": 217.0298472046852,
"ratio": 3.74115344786377,
"marginal_ms": 6.223576026968658
},
{
"m": 256,
"t_state_ms": 17.83185824751854,
"t_queries_ms": 1439.1961731016636,
"total_ms": 1457.0280313491821,
"ratio": 25.116201818891934,
"marginal_ms": 5.621860051178373
}
],
"native_peak_mem_mb": 14892.906496,
"b_batched": [
{
"m": 1,
"t_state_ms": 19.53354850411415,
"t_queries_ms": 75.62482729554176,
"total_ms": 95.15837579965591,
"ratio": 1.0,
"marginal_ms": 75.62482729554176
},
{
"m": 32,
"t_state_ms": 19.454356282949448,
"t_queries_ms": 903.2446146011353,
"total_ms": 922.6989708840847,
"ratio": 9.696455652277129,
"marginal_ms": 28.226394206285477
},
{
"m": 256,
"t_state_ms": 21.368220448493958,
"t_queries_ms": 7146.388553082943,
"total_ms": 7167.756773531437,
"ratio": 75.32449680123014,
"marginal_ms": 27.915580285480246
}
],
"b_batched_peak_mem_mb": 20708.114944
},
"64": {
"state_tokens": 256,
"suffix_tokens": 460,
"nc_chunk": 8,
"native": [
{
"m": 1,
"t_state_ms": 17.79353618621826,
"t_queries_ms": 41.496291756629944,
"total_ms": 59.289827942848206,
"ratio": 1.0,
"marginal_ms": 41.496291756629944
},
{
"m": 32,
"t_state_ms": 17.75621995329857,
"t_queries_ms": 338.5215550661087,
"total_ms": 356.2777750194073,
"ratio": 6.009087686387578,
"marginal_ms": 10.578798595815897
},
{
"m": 256,
"t_state_ms": 17.848066985607147,
"t_queries_ms": 2598.344925791025,
"total_ms": 2616.1929927766323,
"ratio": 44.1254947694988,
"marginal_ms": 10.149784866371192
}
],
"native_peak_mem_mb": 15333.215232,
"b_batched": [
{
"m": 1,
"t_state_ms": 19.58613097667694,
"t_queries_ms": 93.24939921498299,
"total_ms": 112.83553019165993,
"ratio": 1.0,
"marginal_ms": 93.24939921498299
},
{
"m": 32,
"t_state_ms": 19.635286182165146,
"t_queries_ms": 1480.6604906916618,
"total_ms": 1500.295776873827,
"ratio": 13.296306352488953,
"marginal_ms": 46.27064033411443
},
{
"m": 256,
"t_state_ms": 19.47050541639328,
"t_queries_ms": 11805.44163286686,
"total_ms": 11824.912138283253,
"ratio": 104.79777174971146,
"marginal_ms": 46.11500637838617
}
],
"b_batched_peak_mem_mb": 26712.366592
},
"128": {
"state_tokens": 256,
"suffix_tokens": 1029,
"nc_chunk": 8,
"native": [
{
"m": 1,
"t_state_ms": 17.73315668106079,
"t_queries_ms": 48.95908385515213,
"total_ms": 66.69224053621292,
"ratio": 1.0,
"marginal_ms": 48.95908385515213
},
{
"m": 32,
"t_state_ms": 17.543818801641464,
"t_queries_ms": 716.6986837983131,
"total_ms": 734.2425025999546,
"ratio": 11.00941423914633,
"marginal_ms": 22.396833868697286
},
{
"m": 256,
"t_state_ms": 17.649024724960327,
"t_queries_ms": 5805.374853312969,
"total_ms": 5823.0238780379295,
"ratio": 87.31186463702792,
"marginal_ms": 22.677245520753786
}
],
"native_peak_mem_mb": 16306.219008,
"b_batched": [
{
"m": 1,
"t_state_ms": 25.08896216750145,
"t_queries_ms": 140.0025337934494,
"total_ms": 165.09149596095085,
"ratio": 1.0,
"marginal_ms": 140.0025337934494
},
{
"m": 32,
"t_state_ms": 19.624758511781693,
"t_queries_ms": 2846.7205613851547,
"total_ms": 2866.3453198969364,
"ratio": 17.362162134474293,
"marginal_ms": 88.96001754328609
},
{
"m": 256,
"t_state_ms": 19.967511296272278,
"t_queries_ms": 22693.809997290373,
"total_ms": 22713.777508586645,
"ratio": 137.58296498784614,
"marginal_ms": 88.64769530191552
}
],
"b_batched_peak_mem_mb": 33338.615296
},
"256": {
"state_tokens": 256,
"suffix_tokens": 2309,
"nc_chunk": 8,
"native": [
{
"m": 1,
"t_state_ms": 17.903797328472137,
"t_queries_ms": 78.47601175308228,
"total_ms": 96.37980908155441,
"ratio": 1.0,
"marginal_ms": 78.47601175308228
},
{
"m": 32,
"t_state_ms": 17.928332090377808,
"t_queries_ms": 1699.1598010063171,
"total_ms": 1717.088133096695,
"ratio": 17.81584908145786,
"marginal_ms": 53.09874378144741
},
{
"m": 256,
"t_state_ms": 17.86353439092636,
"t_queries_ms": 12893.904343247414,
"total_ms": 12911.76787763834,
"ratio": 133.9675602253237,
"marginal_ms": 50.36681384081021
}
],
"native_peak_mem_mb": 18598.0416,
"b_batched": [
{
"m": 1,
"t_state_ms": 20.501967519521713,
"t_queries_ms": 222.51605615019798,
"total_ms": 243.0180236697197,
"ratio": 1.0,
"marginal_ms": 222.51605615019798
},
{
"m": 32,
"t_state_ms": 19.544314593076706,
"t_queries_ms": 5423.821311444044,
"total_ms": 5443.365626037121,
"ratio": 22.399020220142503,
"marginal_ms": 169.49441598262638
},
{
"m": 256,
"t_state_ms": 19.62793990969658,
"t_queries_ms": 43251.52888149023,
"total_ms": 43271.15682139993,
"ratio": 178.0573974225417,
"marginal_ms": 168.9512846933212
}
],
"b_batched_peak_mem_mb": 27778.753536
}
}
},
"crossover": {}
}