matilda-jev-fp4 / PERFORMANCE.json
yue-maincode's picture
Upload validated MATILDA JEV FP4 model and Decision Index scores
c69aaec verified
Raw History Blame Contribute Delete
16.7 kB
{
"complete": true,
"device": "AMD Instinct MI355X",
"torch": "2.14.0+rocm7.2",
"hip": "7.2.53211",
"same_16_inputs": true,
"includes_prompt_tokenization_forward_softmax_CPU_probabilities": true,
"input_ids": [
"language-focus/ACOS/laptop/official-dev/39/201",
"domain-retention/When2Call/3379",
"domain-retention/When2Call/8402",
"candidates-v3:57:2480",
"domain-retention/VAST/8376",
"candidates-v3:57:11139",
"domain-retention/ANLI/f9dc5f0c-ca9b-4cdb-8554-85ee9e2f2d2a",
"domain-retention/HoVer/b68c5781-66fc-4596-96ea-0033c56327a9",
"domain-retention/RAGTruth/16293",
"domain-retention/VAST/17487",
"retrieval-arts/ESCI/1585008",
"repair-science/public-bulk-v1/MegaScience/MegaScience/574552",
"domain-retention/ANLI/4f4b1fc9-99af-40ea-beb3-d40ff74ab948",
"reasoning-gym-batch2-0c14ebb1-3c4f-42e8-80dc-454f3074a36b/order1",
"domain-retention/RAGTruth/13079",
"chess-r4:05:003020"
],
"backend": "FP4 W4A16 portable: Triton dequantization followed by BF16 GEMM; not native FP4 GEMM",
"results": {
"bf16": {
"load_seconds": 57.19352705893107,
"resident_allocated_GiB": 48.60280990600586,
"batches": {
"1": {
"batch_latency_p50_ms": 54.19635958969593,
"batch_latency_p95_ms": 71.69750705361366,
"requests_per_second": 17.545324611772127,
"peak_allocated_GiB": 48.818500995635986,
"measurements": [
{
"seconds": 0.09753149002790451,
"requests": 1,
"tokens": 176
},
{
"seconds": 0.05726980906911194,
"requests": 1,
"tokens": 201
},
{
"seconds": 0.057499883929267526,
"requests": 1,
"tokens": 411
},
{
"seconds": 0.05371670797467232,
"requests": 1,
"tokens": 195
},
{
"seconds": 0.05342476908117533,
"requests": 1,
"tokens": 167
},
{
"seconds": 0.05325274309143424,
"requests": 1,
"tokens": 157
},
{
"seconds": 0.053995239082723856,
"requests": 1,
"tokens": 229
},
{
"seconds": 0.05385898286476731,
"requests": 1,
"tokens": 311
},
{
"seconds": 0.062414316926151514,
"requests": 1,
"tokens": 551
},
{
"seconds": 0.05333736911416054,
"requests": 1,
"tokens": 241
},
{
"seconds": 0.05362454196438193,
"requests": 1,
"tokens": 196
},
{
"seconds": 0.05355773097835481,
"requests": 1,
"tokens": 213
},
{
"seconds": 0.05317683983594179,
"requests": 1,
"tokens": 261
},
{
"seconds": 0.05316391191445291,
"requests": 1,
"tokens": 301
},
{
"seconds": 0.07271995604969561,
"requests": 1,
"tokens": 726
},
{
"seconds": 0.057841310976073146,
"requests": 1,
"tokens": 510
},
{
"seconds": 0.0531228450126946,
"requests": 1,
"tokens": 176
},
{
"seconds": 0.05372273619286716,
"requests": 1,
"tokens": 201
},
{
"seconds": 0.05708522698841989,
"requests": 1,
"tokens": 411
},
{
"seconds": 0.053957157069817185,
"requests": 1,
"tokens": 195
},
{
"seconds": 0.05323703191243112,
"requests": 1,
"tokens": 167
},
{
"seconds": 0.05320191406644881,
"requests": 1,
"tokens": 157
},
{
"seconds": 0.05381570407189429,
"requests": 1,
"tokens": 229
},
{
"seconds": 0.053544011898338795,
"requests": 1,
"tokens": 311
},
{
"seconds": 0.062205130932852626,
"requests": 1,
"tokens": 551
},
{
"seconds": 0.053216683911159635,
"requests": 1,
"tokens": 241
},
{
"seconds": 0.05345195601694286,
"requests": 1,
"tokens": 196
},
{
"seconds": 0.053699644980952144,
"requests": 1,
"tokens": 213
},
{
"seconds": 0.056666122982278466,
"requests": 1,
"tokens": 261
},
{
"seconds": 0.05431815097108483,
"requests": 1,
"tokens": 301
},
{
"seconds": 0.07177247991785407,
"requests": 1,
"tokens": 726
},
{
"seconds": 0.05799283902160823,
"requests": 1,
"tokens": 510
},
{
"seconds": 0.05421221605502069,
"requests": 1,
"tokens": 176
},
{
"seconds": 0.054013164015486836,
"requests": 1,
"tokens": 201
},
{
"seconds": 0.05697794910520315,
"requests": 1,
"tokens": 411
},
{
"seconds": 0.05430118809454143,
"requests": 1,
"tokens": 195
},
{
"seconds": 0.05409477511420846,
"requests": 1,
"tokens": 167
},
{
"seconds": 0.05392967886291444,
"requests": 1,
"tokens": 157
},
{
"seconds": 0.05439830990508199,
"requests": 1,
"tokens": 229
},
{
"seconds": 0.05478060385212302,
"requests": 1,
"tokens": 311
},
{
"seconds": 0.06227720994502306,
"requests": 1,
"tokens": 551
},
{
"seconds": 0.054351587081328034,
"requests": 1,
"tokens": 241
},
{
"seconds": 0.05442318506538868,
"requests": 1,
"tokens": 196
},
{
"seconds": 0.054551840061321855,
"requests": 1,
"tokens": 213
},
{
"seconds": 0.05429398803971708,
"requests": 1,
"tokens": 261
},
{
"seconds": 0.05418050312437117,
"requests": 1,
"tokens": 301
},
{
"seconds": 0.07169750705361366,
"requests": 1,
"tokens": 726
},
{
"seconds": 0.05789261101745069,
"requests": 1,
"tokens": 510
}
]
},
"8": {
"batch_latency_p50_ms": 295.54353503044695,
"batch_latency_p95_ms": 374.23561001196504,
"requests_per_second": 27.134567546742975,
"peak_allocated_GiB": 49.85915184020996,
"measurements": [
{
"seconds": 0.21804877696558833,
"requests": 8,
"tokens": 1847
},
{
"seconds": 0.37578478013165295,
"requests": 8,
"tokens": 2999
},
{
"seconds": 0.21404898399487138,
"requests": 8,
"tokens": 1847
},
{
"seconds": 0.37423561001196504,
"requests": 8,
"tokens": 2999
},
{
"seconds": 0.21380486013367772,
"requests": 8,
"tokens": 1847
},
{
"seconds": 0.37303829309530556,
"requests": 8,
"tokens": 2999
}
]
}
}
},
"fp4": {
"load_seconds": 37.296120763989165,
"resident_allocated_GiB": 16.26628541946411,
"batches": {
"1": {
"batch_latency_p50_ms": 66.10575248487294,
"batch_latency_p95_ms": 76.5055047813803,
"requests_per_second": 14.478255012045283,
"peak_allocated_GiB": 16.507136344909668,
"measurements": [
{
"seconds": 0.06496857199817896,
"requests": 1,
"tokens": 176
},
{
"seconds": 0.06544231995940208,
"requests": 1,
"tokens": 201
},
{
"seconds": 0.07402830617502332,
"requests": 1,
"tokens": 411
},
{
"seconds": 0.06589341699145734,
"requests": 1,
"tokens": 195
},
{
"seconds": 0.06557931704446673,
"requests": 1,
"tokens": 167
},
{
"seconds": 0.06496005994267762,
"requests": 1,
"tokens": 157
},
{
"seconds": 0.06581282708793879,
"requests": 1,
"tokens": 229
},
{
"seconds": 0.0685572309885174,
"requests": 1,
"tokens": 311
},
{
"seconds": 0.0765055047813803,
"requests": 1,
"tokens": 551
},
{
"seconds": 0.06611774396151304,
"requests": 1,
"tokens": 241
},
{
"seconds": 0.06552175804972649,
"requests": 1,
"tokens": 196
},
{
"seconds": 0.06568572297692299,
"requests": 1,
"tokens": 213
},
{
"seconds": 0.06540890503674746,
"requests": 1,
"tokens": 261
},
{
"seconds": 0.06838645786046982,
"requests": 1,
"tokens": 301
},
{
"seconds": 0.0863226349465549,
"requests": 1,
"tokens": 726
},
{
"seconds": 0.07383130700327456,
"requests": 1,
"tokens": 510
},
{
"seconds": 0.06545696989633143,
"requests": 1,
"tokens": 176
},
{
"seconds": 0.06609197799116373,
"requests": 1,
"tokens": 201
},
{
"seconds": 0.07416753610596061,
"requests": 1,
"tokens": 411
},
{
"seconds": 0.06622499111108482,
"requests": 1,
"tokens": 195
},
{
"seconds": 0.06530973804183304,
"requests": 1,
"tokens": 167
},
{
"seconds": 0.06533637479878962,
"requests": 1,
"tokens": 157
},
{
"seconds": 0.06579005508683622,
"requests": 1,
"tokens": 229
},
{
"seconds": 0.06860739993862808,
"requests": 1,
"tokens": 311
},
{
"seconds": 0.07632632995955646,
"requests": 1,
"tokens": 551
},
{
"seconds": 0.0657245060428977,
"requests": 1,
"tokens": 241
},
{
"seconds": 0.06579803698696196,
"requests": 1,
"tokens": 196
},
{
"seconds": 0.06622335803695023,
"requests": 1,
"tokens": 213
},
{
"seconds": 0.0655344061087817,
"requests": 1,
"tokens": 261
},
{
"seconds": 0.06830077082850039,
"requests": 1,
"tokens": 301
},
{
"seconds": 0.08639194094575942,
"requests": 1,
"tokens": 726
},
{
"seconds": 0.07383402087725699,
"requests": 1,
"tokens": 510
},
{
"seconds": 0.065410046139732,
"requests": 1,
"tokens": 176
},
{
"seconds": 0.06643647002056241,
"requests": 1,
"tokens": 201
},
{
"seconds": 0.07415988598950207,
"requests": 1,
"tokens": 411
},
{
"seconds": 0.06597798294387758,
"requests": 1,
"tokens": 195
},
{
"seconds": 0.06547786900773644,
"requests": 1,
"tokens": 167
},
{
"seconds": 0.06554997689090669,
"requests": 1,
"tokens": 157
},
{
"seconds": 0.06611520005390048,
"requests": 1,
"tokens": 229
},
{
"seconds": 0.06863704114221036,
"requests": 1,
"tokens": 311
},
{
"seconds": 0.07638701307587326,
"requests": 1,
"tokens": 551
},
{
"seconds": 0.06636205804534256,
"requests": 1,
"tokens": 241
},
{
"seconds": 0.0660417799372226,
"requests": 1,
"tokens": 196
},
{
"seconds": 0.0660963049158454,
"requests": 1,
"tokens": 213
},
{
"seconds": 0.06588856992311776,
"requests": 1,
"tokens": 261
},
{
"seconds": 0.06834727409295738,
"requests": 1,
"tokens": 301
},
{
"seconds": 0.08631210192106664,
"requests": 1,
"tokens": 726
},
{
"seconds": 0.0739765849430114,
"requests": 1,
"tokens": 510
}
]
},
"8": {
"batch_latency_p50_ms": 310.49984705168754,
"batch_latency_p95_ms": 392.74976099841297,
"requests_per_second": 25.762332831510818,
"peak_allocated_GiB": 17.448820114135742,
"measurements": [
{
"seconds": 0.2306268650572747,
"requests": 8,
"tokens": 1847
},
{
"seconds": 0.3929026599507779,
"requests": 8,
"tokens": 2999
},
{
"seconds": 0.23050653981044888,
"requests": 8,
"tokens": 1847
},
{
"seconds": 0.39274976099841297,
"requests": 8,
"tokens": 2999
},
{
"seconds": 0.2260266519151628,
"requests": 8,
"tokens": 1847
},
{
"seconds": 0.3903728290461004,
"requests": 8,
"tokens": 2999
}
]
}
}
}
}
}