Files
aituner/runs/frontier-tp2-prefill-serving-v0/results/replay-verdict.json

632 lines
19 KiB
JSON

{
"cells": {
"tp2_rho0p0025": {
"structured_attention": {
"legacy_pooled_distribution_bias": {
"e2e": {
"mean": 0.12696146154030977,
"p50": 0.14796251184179712,
"p90": 0.12767235986604622,
"p99": -0.03965040123988098
},
"tpot": {
"mean": 0.15073709978689778,
"p50": 0.18527285925405948,
"p90": 0.07863996768588354,
"p99": 0.1661995397125918
},
"ttft": {
"mean": -0.045433491316312524,
"p50": -0.0875990583320057,
"p90": -0.1142200855839675,
"p99": -0.07674087497700505
}
},
"trialwise_distribution_bias": [
{
"e2e": {
"mean": 0.12073678040141796,
"p50": 0.14420262806746403,
"p90": 0.11915070375608593,
"p99": -0.022788557992646938
},
"tpot": {
"mean": 0.14386158526258666,
"p50": 0.1787588067010671,
"p90": 0.071419448886029,
"p99": 0.17844287277146695
},
"ttft": {
"mean": -0.044844475267623835,
"p50": -0.08906525404741798,
"p90": -0.11412907883561804,
"p99": -0.07227786537885415
}
},
{
"e2e": {
"mean": 0.13325567383154438,
"p50": 0.15920445684060733,
"p90": 0.12953600816854335,
"p99": -0.009483068681846096
},
"tpot": {
"mean": 0.15769576871598226,
"p50": 0.196444372264794,
"p90": 0.08604019319402298,
"p99": 0.19311481061458743
},
"ttft": {
"mean": -0.046021781355297144,
"p50": -0.08612813517467069,
"p90": -0.11510233828040589,
"p99": -0.07592149320892366
}
}
],
"trialwise_distribution_bias_summary": {
"e2e": {
"mean": {
"max": 0.13325567383154438,
"mean": 0.12699622711648118,
"min": 0.12073678040141796
},
"p50": {
"max": 0.15920445684060733,
"mean": 0.15170354245403567,
"min": 0.14420262806746403
},
"p90": {
"max": 0.12953600816854335,
"mean": 0.12434335596231463,
"min": 0.11915070375608593
},
"p99": {
"max": -0.009483068681846096,
"mean": -0.016135813337246518,
"min": -0.022788557992646938
}
},
"tpot": {
"mean": {
"max": 0.15769576871598226,
"mean": 0.15077867698928446,
"min": 0.14386158526258666
},
"p50": {
"max": 0.196444372264794,
"mean": 0.18760158948293054,
"min": 0.1787588067010671
},
"p90": {
"max": 0.08604019319402298,
"mean": 0.07872982104002599,
"min": 0.071419448886029
},
"p99": {
"max": 0.19311481061458743,
"mean": 0.1857788416930272,
"min": 0.17844287277146695
}
},
"ttft": {
"mean": {
"max": -0.044844475267623835,
"mean": -0.045433128311460486,
"min": -0.046021781355297144
},
"p50": {
"max": -0.08612813517467069,
"mean": -0.08759669461104433,
"min": -0.08906525404741798
},
"p90": {
"max": -0.11412907883561804,
"mean": -0.11461570855801197,
"min": -0.11510233828040589
},
"p99": {
"max": -0.07227786537885415,
"mean": -0.07409967929388891,
"min": -0.07592149320892366
}
}
},
"waiting_p99_ms": 777.5911879793948
},
"structured_attention_plus_prefill_moe": {
"legacy_pooled_distribution_bias": {
"e2e": {
"mean": 0.15546889966298288,
"p50": 0.160521791081281,
"p90": 0.14949973986858572,
"p99": 0.063501869119007
},
"tpot": {
"mean": 0.1725158165141388,
"p50": 0.19940581724371484,
"p90": 0.1197438720681557,
"p99": 0.16591937069699617
},
"ttft": {
"mean": 0.052253838015785536,
"p50": 0.011133339751522397,
"p90": -0.05603209782959642,
"p99": 0.008949154409837737
}
},
"paired_relative_error": [
{
"e2e": {
"mean": 0.19703071945786896,
"p50": 0.16651188156434055,
"p90": 0.3164763187383559,
"p99": 0.828609074267838
},
"tpot": {
"mean": 0.19036780497451272,
"p50": 0.17682334345090506,
"p90": 0.25886898169288464,
"p99": 0.7758893841183759
},
"ttft": {
"mean": 0.45065657187067015,
"p50": 0.10957981115881457,
"p90": 1.0421699185581517,
"p99": 5.507647228917042
}
},
{
"e2e": {
"mean": 0.20969149313321356,
"p50": 0.17863746376866904,
"p90": 0.32770775080455516,
"p99": 0.8245897427826396
},
"tpot": {
"mean": 0.20429023428176118,
"p50": 0.19259101506782267,
"p90": 0.2812533323043662,
"p99": 0.7385621812554619
},
"ttft": {
"mean": 0.4457497323795849,
"p50": 0.10597514413253763,
"p90": 1.0162200916250081,
"p99": 5.496662131059387
}
}
],
"trialwise_distribution_bias": [
{
"e2e": {
"mean": 0.14908675997874046,
"p50": 0.1567207723136509,
"p90": 0.14081313742067922,
"p99": 0.08217486261364416
},
"tpot": {
"mean": 0.16551017680032373,
"p50": 0.1928140924228096,
"p90": 0.11224819981281003,
"p99": 0.1781597624045383
},
"ttft": {
"mean": 0.052903132111322415,
"p50": 0.00950848459985634,
"p90": -0.05593511273439991,
"p99": 0.013826387288597744
}
},
{
"e2e": {
"mean": 0.16192232934855333,
"p50": 0.1718867285340943,
"p90": 0.15139946120181186,
"p99": 0.09690951004812934
},
"tpot": {
"mean": 0.17960618440333612,
"p50": 0.21071053715514645,
"p90": 0.1274260991438306,
"p99": 0.19282817544573386
},
"ttft": {
"mean": 0.051605344227498705,
"p50": 0.012763433909333126,
"p90": -0.05697230680643937,
"p99": 0.009844585085406704
}
}
],
"trialwise_distribution_bias_summary": {
"e2e": {
"mean": {
"max": 0.16192232934855333,
"mean": 0.15550454466364688,
"min": 0.14908675997874046
},
"p50": {
"max": 0.1718867285340943,
"mean": 0.1643037504238726,
"min": 0.1567207723136509
},
"p90": {
"max": 0.15139946120181186,
"mean": 0.14610629931124552,
"min": 0.14081313742067922
},
"p99": {
"max": 0.09690951004812934,
"mean": 0.08954218633088676,
"min": 0.08217486261364416
}
},
"tpot": {
"mean": {
"max": 0.17960618440333612,
"mean": 0.17255818060182992,
"min": 0.16551017680032373
},
"p50": {
"max": 0.21071053715514645,
"mean": 0.20176231478897805,
"min": 0.1928140924228096
},
"p90": {
"max": 0.1274260991438306,
"mean": 0.11983714947832032,
"min": 0.11224819981281003
},
"p99": {
"max": 0.19282817544573386,
"mean": 0.1854939689251361,
"min": 0.1781597624045383
}
},
"ttft": {
"mean": {
"max": 0.052903132111322415,
"mean": 0.05225423816941056,
"min": 0.051605344227498705
},
"p50": {
"max": 0.012763433909333126,
"mean": 0.011135959254594732,
"min": 0.00950848459985634
},
"p90": {
"max": -0.05593511273439991,
"mean": -0.056453709770419636,
"min": -0.05697230680643937
},
"p99": {
"max": 0.013826387288597744,
"mean": 0.011835486187002224,
"min": 0.009844585085406704
}
}
},
"waiting_p99_ms": 872.1814900223759
}
},
"tp2_rho0p005": {
"structured_attention": {
"legacy_pooled_distribution_bias": {
"e2e": {
"mean": 0.050164236759908075,
"p50": 0.0912493994395978,
"p90": -0.0004979191794830456,
"p99": -0.0014369075488634014
},
"tpot": {
"mean": 0.05457336599027876,
"p50": 0.0872512146277895,
"p90": 0.01132352325010614,
"p99": -0.05948329733667248
},
"ttft": {
"mean": -0.07082500585057615,
"p50": -0.1284088888361358,
"p90": -0.111136573344055,
"p99": -0.09012556161973535
}
},
"trialwise_distribution_bias": [
{
"e2e": {
"mean": 0.04462683161260099,
"p50": 0.0878889042422101,
"p90": -0.006843953727060337,
"p99": -0.0036641600308731643
},
"tpot": {
"mean": 0.048585629713612626,
"p50": 0.08234652581183928,
"p90": 0.003101343693636994,
"p99": -0.05952673172238401
},
"ttft": {
"mean": -0.07113138559444453,
"p50": -0.1294796632798084,
"p90": -0.11063584772248519,
"p99": -0.08982438463776078
}
},
{
"e2e": {
"mean": 0.05576066061144686,
"p50": 0.09176507061640418,
"p90": 0.003187406757202862,
"p99": 0.004813545234813467
},
"tpot": {
"mean": 0.060629878515092876,
"p50": 0.09375745746657616,
"p90": 0.019190079238111446,
"p99": -0.05840233314016777
},
"ttft": {
"mean": -0.07051842392629604,
"p50": -0.12771160295843229,
"p90": -0.11011498510890981,
"p99": -0.08861845103384182
}
}
],
"trialwise_distribution_bias_summary": {
"e2e": {
"mean": {
"max": 0.05576066061144686,
"mean": 0.05019374611202393,
"min": 0.04462683161260099
},
"p50": {
"max": 0.09176507061640418,
"mean": 0.08982698742930714,
"min": 0.0878889042422101
},
"p90": {
"max": 0.003187406757202862,
"mean": -0.0018282734849287376,
"min": -0.006843953727060337
},
"p99": {
"max": 0.004813545234813467,
"mean": 0.0005746926019701514,
"min": -0.0036641600308731643
}
},
"tpot": {
"mean": {
"max": 0.060629878515092876,
"mean": 0.05460775411435275,
"min": 0.048585629713612626
},
"p50": {
"max": 0.09375745746657616,
"mean": 0.08805199163920771,
"min": 0.08234652581183928
},
"p90": {
"max": 0.019190079238111446,
"mean": 0.01114571146587422,
"min": 0.003101343693636994
},
"p99": {
"max": -0.05840233314016777,
"mean": -0.05896453243127589,
"min": -0.05952673172238401
}
},
"ttft": {
"mean": {
"max": -0.07051842392629604,
"mean": -0.07082490476037029,
"min": -0.07113138559444453
},
"p50": {
"max": -0.12771160295843229,
"mean": -0.12859563311912034,
"min": -0.1294796632798084
},
"p90": {
"max": -0.11011498510890981,
"mean": -0.11037541641569751,
"min": -0.11063584772248519
},
"p99": {
"max": -0.08861845103384182,
"mean": -0.0892214178358013,
"min": -0.08982438463776078
}
}
},
"waiting_p99_ms": 1166.1196883368564
},
"structured_attention_plus_prefill_moe": {
"legacy_pooled_distribution_bias": {
"e2e": {
"mean": 0.09135599701015996,
"p50": 0.12135156828958114,
"p90": 0.046904152670232756,
"p99": 0.05267198541612694
},
"tpot": {
"mean": 0.09686033389444083,
"p50": 0.10920022438478016,
"p90": 0.055959658792935615,
"p99": 0.03647201411457465
},
"ttft": {
"mean": 0.03133826519177879,
"p50": -0.002404965655955949,
"p90": -0.03671038522787783,
"p99": -0.003047423826180572
}
},
"paired_relative_error": [
{
"e2e": {
"mean": 0.1373684769346999,
"p50": 0.10813242136980938,
"p90": 0.27973507165610256,
"p99": 0.6892985502092539
},
"tpot": {
"mean": 0.12750409894388354,
"p50": 0.10486952080742089,
"p90": 0.2347358304319833,
"p99": 0.7326960496792728
},
"ttft": {
"mean": 0.48959967245129044,
"p50": 0.11864693915356177,
"p90": 1.3203954670852478,
"p99": 5.486884563706639
}
},
{
"e2e": {
"mean": 0.15015620590606898,
"p50": 0.11828746639583118,
"p90": 0.29344565738935435,
"p99": 0.7209062198737742
},
"tpot": {
"mean": 0.14093466935703095,
"p50": 0.11608679958107468,
"p90": 0.24292321061978972,
"p99": 0.7622579889720426
},
"ttft": {
"mean": 0.4989603034128753,
"p50": 0.11800815758794723,
"p90": 1.394773550999063,
"p99": 5.683406566684342
}
}
],
"trialwise_distribution_bias": [
{
"e2e": {
"mean": 0.08560139205995335,
"p50": 0.11789837366536833,
"p90": 0.040257153080803736,
"p99": 0.05032404535106296
},
"tpot": {
"mean": 0.09063249747878419,
"p50": 0.10419652159558444,
"p90": 0.04737458218848663,
"p99": 0.03642414837746145
},
"ttft": {
"mean": 0.03099819883665655,
"p50": -0.0036305394992732502,
"p90": -0.03616773291859882,
"p99": -0.0027174231630140012
}
},
{
"e2e": {
"mean": 0.09717193562089739,
"p50": 0.12188146428139615,
"p90": 0.05076425771763047,
"p99": 0.05926112994918994
},
"tpot": {
"mean": 0.10315970439292232,
"p50": 0.11583781275404642,
"p90": 0.06417341590032893,
"p99": 0.037663262642925396
},
"ttft": {
"mean": 0.031678555957153555,
"p50": -0.001606874761951341,
"p90": -0.03560325750951616,
"p99": -0.0013960775328830752
}
}
],
"trialwise_distribution_bias_summary": {
"e2e": {
"mean": {
"max": 0.09717193562089739,
"mean": 0.09138666384042538,
"min": 0.08560139205995335
},
"p50": {
"max": 0.12188146428139615,
"mean": 0.11988991897338225,
"min": 0.11789837366536833
},
"p90": {
"max": 0.05076425771763047,
"mean": 0.0455107053992171,
"min": 0.040257153080803736
},
"p99": {
"max": 0.05926112994918994,
"mean": 0.05479258765012645,
"min": 0.05032404535106296
}
},
"tpot": {
"mean": {
"max": 0.10315970439292232,
"mean": 0.09689610093585325,
"min": 0.09063249747878419
},
"p50": {
"max": 0.11583781275404642,
"mean": 0.11001716717481544,
"min": 0.10419652159558444
},
"p90": {
"max": 0.06417341590032893,
"mean": 0.05577399904440778,
"min": 0.04737458218848663
},
"p99": {
"max": 0.037663262642925396,
"mean": 0.037043705510193425,
"min": 0.03642414837746145
}
},
"ttft": {
"mean": {
"max": 0.031678555957153555,
"mean": 0.03133837739690505,
"min": 0.03099819883665655
},
"p50": {
"max": -0.001606874761951341,
"mean": -0.002618707130612296,
"min": -0.0036305394992732502
},
"p90": {
"max": -0.03560325750951616,
"mean": -0.03588549521405749,
"min": -0.03616773291859882
},
"p99": {
"max": -0.0013960775328830752,
"mean": -0.0020567503479485385,
"min": -0.0027174231630140012
}
}
},
"waiting_p99_ms": 1296.3971075780032
}
}
},
"decision": "reject_global_scale_and_fit_shape_conditioned_curve",
"gates": {
"subcritical_mean_ttft_abs_bias_improves_3pp": false,
"subcritical_ttft_e2e_abs_bias_not_worse": false,
"subcritical_waiting_below_1s": true
},
"schema": "frontier-tp2-prefill-serving-replay-verdict.v1"
}