deepseek-8b-orpo-lora / trainer_state.json
zfz1's picture
Model save
3cbea15 verified
raw
history blame
21.5 kB
{
"best_metric": null,
"best_model_checkpoint": null,
"epoch": 1.9936102236421727,
"eval_steps": 10000,
"global_step": 312,
"is_hyper_param_search": false,
"is_local_process_zero": true,
"is_world_process_zero": true,
"log_history": [
{
"epoch": 0.06389776357827476,
"grad_norm": 0.341777060295131,
"learning_rate": 9.375e-07,
"log_odds_chosen": 0.05486620217561722,
"log_odds_ratio": -0.6975381374359131,
"logits/chosen": 33.646114349365234,
"logits/rejected": 33.55038833618164,
"logps/chosen": -0.9772504568099976,
"logps/rejected": -1.0204073190689087,
"loss": 1.6735,
"nll_loss": 1.5879063606262207,
"rewards/accuracies": 0.543749988079071,
"rewards/chosen": -0.09772505611181259,
"rewards/margins": 0.004315672442317009,
"rewards/rejected": -0.10204073041677475,
"step": 10
},
{
"epoch": 0.12779552715654952,
"grad_norm": 0.357474094120966,
"learning_rate": 1.875e-06,
"log_odds_chosen": 0.054913729429244995,
"log_odds_ratio": -0.6985839009284973,
"logits/chosen": 34.386383056640625,
"logits/rejected": 33.98283386230469,
"logps/chosen": -0.9762754440307617,
"logps/rejected": -1.0154238939285278,
"loss": 1.668,
"nll_loss": 1.6016145944595337,
"rewards/accuracies": 0.515625,
"rewards/chosen": -0.09762755781412125,
"rewards/margins": 0.003914830274879932,
"rewards/rejected": -0.10154237598180771,
"step": 20
},
{
"epoch": 0.19169329073482427,
"grad_norm": 0.3765211315354108,
"learning_rate": 2.8125e-06,
"log_odds_chosen": 0.06909728795289993,
"log_odds_ratio": -0.6911064386367798,
"logits/chosen": 33.32851028442383,
"logits/rejected": 32.749427795410156,
"logps/chosen": -0.9813788533210754,
"logps/rejected": -1.0349732637405396,
"loss": 1.6757,
"nll_loss": 1.6085166931152344,
"rewards/accuracies": 0.5249999761581421,
"rewards/chosen": -0.09813789278268814,
"rewards/margins": 0.005359448026865721,
"rewards/rejected": -0.10349734127521515,
"step": 30
},
{
"epoch": 0.25559105431309903,
"grad_norm": 0.39833625263223615,
"learning_rate": 2.993961440992859e-06,
"log_odds_chosen": 0.11841567605733871,
"log_odds_ratio": -0.6757725477218628,
"logits/chosen": 33.839942932128906,
"logits/rejected": 33.162986755371094,
"logps/chosen": -0.9702426791191101,
"logps/rejected": -1.0581519603729248,
"loss": 1.664,
"nll_loss": 1.6022984981536865,
"rewards/accuracies": 0.6031249761581421,
"rewards/chosen": -0.09702426195144653,
"rewards/margins": 0.008790932595729828,
"rewards/rejected": -0.10581519454717636,
"step": 40
},
{
"epoch": 0.3194888178913738,
"grad_norm": 0.45952260963075464,
"learning_rate": 2.9695130976348534e-06,
"log_odds_chosen": 0.09267185628414154,
"log_odds_ratio": -0.684535801410675,
"logits/chosen": 34.052207946777344,
"logits/rejected": 32.802547454833984,
"logps/chosen": -0.9801093935966492,
"logps/rejected": -1.0446739196777344,
"loss": 1.6582,
"nll_loss": 1.5952296257019043,
"rewards/accuracies": 0.565625011920929,
"rewards/chosen": -0.09801094233989716,
"rewards/margins": 0.0064564356580376625,
"rewards/rejected": -0.10446737706661224,
"step": 50
},
{
"epoch": 0.38338658146964855,
"grad_norm": 0.49504068932785444,
"learning_rate": 2.9265847744427307e-06,
"log_odds_chosen": 0.06481704860925674,
"log_odds_ratio": -0.698478102684021,
"logits/chosen": 32.32544708251953,
"logits/rejected": 32.51976776123047,
"logps/chosen": -0.9770153164863586,
"logps/rejected": -1.0272716283798218,
"loss": 1.6387,
"nll_loss": 1.572589635848999,
"rewards/accuracies": 0.5687500238418579,
"rewards/chosen": -0.0977015346288681,
"rewards/margins": 0.0050256275571882725,
"rewards/rejected": -0.10272715240716934,
"step": 60
},
{
"epoch": 0.4472843450479233,
"grad_norm": 0.5262433549192981,
"learning_rate": 2.865716319988224e-06,
"log_odds_chosen": 0.030134931206703186,
"log_odds_ratio": -0.7101849317550659,
"logits/chosen": 33.02894973754883,
"logits/rejected": 33.309715270996094,
"logps/chosen": -0.9551456570625305,
"logps/rejected": -0.9776851534843445,
"loss": 1.6289,
"nll_loss": 1.5476250648498535,
"rewards/accuracies": 0.5531250238418579,
"rewards/chosen": -0.09551456570625305,
"rewards/margins": 0.0022539461497217417,
"rewards/rejected": -0.09776850789785385,
"step": 70
},
{
"epoch": 0.5111821086261981,
"grad_norm": 0.5188886669640671,
"learning_rate": 2.7876731904027993e-06,
"log_odds_chosen": 0.08591620624065399,
"log_odds_ratio": -0.6787055730819702,
"logits/chosen": 34.61581039428711,
"logits/rejected": 34.63213348388672,
"logps/chosen": -0.9702480435371399,
"logps/rejected": -1.0235538482666016,
"loss": 1.5921,
"nll_loss": 1.5600519180297852,
"rewards/accuracies": 0.5625,
"rewards/chosen": -0.0970248132944107,
"rewards/margins": 0.005330582614988089,
"rewards/rejected": -0.10235539823770523,
"step": 80
},
{
"epoch": 0.5750798722044729,
"grad_norm": 0.5939864943475555,
"learning_rate": 2.6934368233226715e-06,
"log_odds_chosen": 0.0763852447271347,
"log_odds_ratio": -0.6891772150993347,
"logits/chosen": 34.269248962402344,
"logits/rejected": 34.00069046020508,
"logps/chosen": -0.9529930949211121,
"logps/rejected": -1.0018848180770874,
"loss": 1.586,
"nll_loss": 1.5087509155273438,
"rewards/accuracies": 0.5687500238418579,
"rewards/chosen": -0.09529931843280792,
"rewards/margins": 0.004889167379587889,
"rewards/rejected": -0.1001884788274765,
"step": 90
},
{
"epoch": 0.6389776357827476,
"grad_norm": 0.6337044866684244,
"learning_rate": 2.584192295741087e-06,
"log_odds_chosen": 0.057517312467098236,
"log_odds_ratio": -0.7008931040763855,
"logits/chosen": 32.44559860229492,
"logits/rejected": 32.7852783203125,
"logps/chosen": -0.9386932253837585,
"logps/rejected": -0.9813023805618286,
"loss": 1.5499,
"nll_loss": 1.4884004592895508,
"rewards/accuracies": 0.5562499761581421,
"rewards/chosen": -0.09386932104825974,
"rewards/margins": 0.004260920453816652,
"rewards/rejected": -0.0981302410364151,
"step": 100
},
{
"epoch": 0.7028753993610224,
"grad_norm": 0.658610355052368,
"learning_rate": 2.461313420977536e-06,
"log_odds_chosen": 0.06078052520751953,
"log_odds_ratio": -0.694057822227478,
"logits/chosen": 34.87937927246094,
"logits/rejected": 34.10810470581055,
"logps/chosen": -0.9328826665878296,
"logps/rejected": -0.9768407940864563,
"loss": 1.5083,
"nll_loss": 1.4501432180404663,
"rewards/accuracies": 0.543749988079071,
"rewards/chosen": -0.09328827261924744,
"rewards/margins": 0.004395808558911085,
"rewards/rejected": -0.09768407791852951,
"step": 110
},
{
"epoch": 0.7667731629392971,
"grad_norm": 0.7021451570051785,
"learning_rate": 2.3263454721781537e-06,
"log_odds_chosen": 0.012948527932167053,
"log_odds_ratio": -0.7222181558609009,
"logits/chosen": 32.433719635009766,
"logits/rejected": 32.53590774536133,
"logps/chosen": -0.9092488288879395,
"logps/rejected": -0.9225506782531738,
"loss": 1.4673,
"nll_loss": 1.4067564010620117,
"rewards/accuracies": 0.4937500059604645,
"rewards/chosen": -0.09092487394809723,
"rewards/margins": 0.0013302009319886565,
"rewards/rejected": -0.09225507825613022,
"step": 120
},
{
"epoch": 0.8306709265175719,
"grad_norm": 0.6227085414802311,
"learning_rate": 2.18098574960932e-06,
"log_odds_chosen": 0.08671535551548004,
"log_odds_ratio": -0.6804493069648743,
"logits/chosen": 33.88345718383789,
"logits/rejected": 33.001914978027344,
"logps/chosen": -0.8342811465263367,
"logps/rejected": -0.8885458707809448,
"loss": 1.3999,
"nll_loss": 1.3158425092697144,
"rewards/accuracies": 0.581250011920929,
"rewards/chosen": -0.08342811465263367,
"rewards/margins": 0.005426469258964062,
"rewards/rejected": -0.0888545885682106,
"step": 130
},
{
"epoch": 0.8945686900958466,
"grad_norm": 0.6264864351444569,
"learning_rate": 2.027062236122014e-06,
"log_odds_chosen": 0.04714164510369301,
"log_odds_ratio": -0.6999929547309875,
"logits/chosen": 31.832406997680664,
"logits/rejected": 31.350921630859375,
"logps/chosen": -0.8387459516525269,
"logps/rejected": -0.876442551612854,
"loss": 1.3491,
"nll_loss": 1.2940082550048828,
"rewards/accuracies": 0.534375011920929,
"rewards/chosen": -0.08387459814548492,
"rewards/margins": 0.0037696503568440676,
"rewards/rejected": -0.08764425665140152,
"step": 140
},
{
"epoch": 0.9584664536741214,
"grad_norm": 0.5655314348380543,
"learning_rate": 1.866510609206841e-06,
"log_odds_chosen": 0.044148243963718414,
"log_odds_ratio": -0.7019084692001343,
"logits/chosen": 33.6451416015625,
"logits/rejected": 33.04177474975586,
"logps/chosen": -0.8360700607299805,
"logps/rejected": -0.8655373454093933,
"loss": 1.3242,
"nll_loss": 1.2609388828277588,
"rewards/accuracies": 0.53125,
"rewards/chosen": -0.08360700309276581,
"rewards/margins": 0.0029467367567121983,
"rewards/rejected": -0.08655373752117157,
"step": 150
},
{
"epoch": 1.0223642172523961,
"grad_norm": 0.5571454673635674,
"learning_rate": 1.7013498987264833e-06,
"log_odds_chosen": 0.06723789870738983,
"log_odds_ratio": -0.6888397932052612,
"logits/chosen": 33.27342987060547,
"logits/rejected": 32.30434036254883,
"logps/chosen": -0.7994817495346069,
"logps/rejected": -0.8369095921516418,
"loss": 1.2899,
"nll_loss": 1.211586594581604,
"rewards/accuracies": 0.565625011920929,
"rewards/chosen": -0.07994817197322845,
"rewards/margins": 0.0037427886854857206,
"rewards/rejected": -0.08369095623493195,
"step": 160
},
{
"epoch": 1.0862619808306708,
"grad_norm": 0.5333966997862959,
"learning_rate": 1.5336570964437077e-06,
"log_odds_chosen": 0.038995109498500824,
"log_odds_ratio": -0.6969125866889954,
"logits/chosen": 32.00867462158203,
"logits/rejected": 31.789836883544922,
"logps/chosen": -0.8085753321647644,
"logps/rejected": -0.8331602215766907,
"loss": 1.2863,
"nll_loss": 1.2334020137786865,
"rewards/accuracies": 0.503125011920929,
"rewards/chosen": -0.0808575376868248,
"rewards/margins": 0.0024584876373410225,
"rewards/rejected": -0.08331602811813354,
"step": 170
},
{
"epoch": 1.1501597444089458,
"grad_norm": 0.5317464882215207,
"learning_rate": 1.3655410366448499e-06,
"log_odds_chosen": 0.10975570976734161,
"log_odds_ratio": -0.6656275987625122,
"logits/chosen": 32.76184844970703,
"logits/rejected": 32.081581115722656,
"logps/chosen": -0.7589792013168335,
"logps/rejected": -0.8160526156425476,
"loss": 1.2451,
"nll_loss": 1.1476625204086304,
"rewards/accuracies": 0.6000000238418579,
"rewards/chosen": -0.0758979320526123,
"rewards/margins": 0.0057073310017585754,
"rewards/rejected": -0.08160526305437088,
"step": 180
},
{
"epoch": 1.2140575079872205,
"grad_norm": 0.5186911330716555,
"learning_rate": 1.199115876325091e-06,
"log_odds_chosen": 0.04539128392934799,
"log_odds_ratio": -0.6990563869476318,
"logits/chosen": 33.35085678100586,
"logits/rejected": 33.3908805847168,
"logps/chosen": -0.7675826549530029,
"logps/rejected": -0.7923057675361633,
"loss": 1.2217,
"nll_loss": 1.161102294921875,
"rewards/accuracies": 0.53125,
"rewards/chosen": -0.07675826549530029,
"rewards/margins": 0.0024723131209611893,
"rewards/rejected": -0.07923058420419693,
"step": 190
},
{
"epoch": 1.2779552715654952,
"grad_norm": 0.5411315465201,
"learning_rate": 1.036474508437579e-06,
"log_odds_chosen": 0.07649532705545425,
"log_odds_ratio": -0.6858269572257996,
"logits/chosen": 30.826059341430664,
"logits/rejected": 30.824636459350586,
"logps/chosen": -0.786870002746582,
"logps/rejected": -0.8327391743659973,
"loss": 1.215,
"nll_loss": 1.165637493133545,
"rewards/accuracies": 0.578125,
"rewards/chosen": -0.07868699729442596,
"rewards/margins": 0.004586914554238319,
"rewards/rejected": -0.08327391743659973,
"step": 200
},
{
"epoch": 1.34185303514377,
"grad_norm": 0.5587119516125117,
"learning_rate": 8.796622425502193e-07,
"log_odds_chosen": 0.04922567307949066,
"log_odds_ratio": -0.6928130984306335,
"logits/chosen": 32.06999969482422,
"logits/rejected": 31.100805282592773,
"logps/chosen": -0.760046124458313,
"logps/rejected": -0.7879078388214111,
"loss": 1.2018,
"nll_loss": 1.1223350763320923,
"rewards/accuracies": 0.5218750238418579,
"rewards/chosen": -0.07600460201501846,
"rewards/margins": 0.00278617930598557,
"rewards/rejected": -0.07879078388214111,
"step": 210
},
{
"epoch": 1.4057507987220448,
"grad_norm": 0.5532006875711314,
"learning_rate": 7.30651083891141e-07,
"log_odds_chosen": 0.09409201890230179,
"log_odds_ratio": -0.6719040870666504,
"logits/chosen": 31.775470733642578,
"logits/rejected": 31.210159301757812,
"logps/chosen": -0.7408252954483032,
"logps/rejected": -0.7944933772087097,
"loss": 1.1848,
"nll_loss": 1.1082121133804321,
"rewards/accuracies": 0.6000000238418579,
"rewards/chosen": -0.07408253848552704,
"rewards/margins": 0.005366811528801918,
"rewards/rejected": -0.07944934070110321,
"step": 220
},
{
"epoch": 1.4696485623003195,
"grad_norm": 0.5478964489714558,
"learning_rate": 5.913149342387704e-07,
"log_odds_chosen": 0.022008871659636497,
"log_odds_ratio": -0.7079219818115234,
"logits/chosen": 30.458974838256836,
"logits/rejected": 30.637874603271484,
"logps/chosen": -0.7661860585212708,
"logps/rejected": -0.783053994178772,
"loss": 1.1753,
"nll_loss": 1.1201671361923218,
"rewards/accuracies": 0.5406249761581421,
"rewards/chosen": -0.07661859691143036,
"rewards/margins": 0.00168679840862751,
"rewards/rejected": -0.07830540090799332,
"step": 230
},
{
"epoch": 1.5335463258785942,
"grad_norm": 0.5614636317534148,
"learning_rate": 4.63406026519703e-07,
"log_odds_chosen": 0.12717078626155853,
"log_odds_ratio": -0.6581142544746399,
"logits/chosen": 31.024799346923828,
"logits/rejected": 30.477502822875977,
"logps/chosen": -0.7121531367301941,
"logps/rejected": -0.7798753380775452,
"loss": 1.1701,
"nll_loss": 1.073432207107544,
"rewards/accuracies": 0.596875011920929,
"rewards/chosen": -0.07121531665325165,
"rewards/margins": 0.006772211752831936,
"rewards/rejected": -0.07798753678798676,
"step": 240
},
{
"epoch": 1.5974440894568689,
"grad_norm": 0.5532326330060124,
"learning_rate": 3.4853288946298335e-07,
"log_odds_chosen": 0.031235849484801292,
"log_odds_ratio": -0.7052245140075684,
"logits/chosen": 31.759140014648438,
"logits/rejected": 31.2039737701416,
"logps/chosen": -0.7514272332191467,
"logps/rejected": -0.7712705135345459,
"loss": 1.1628,
"nll_loss": 1.0986436605453491,
"rewards/accuracies": 0.518750011920929,
"rewards/chosen": -0.07514272630214691,
"rewards/margins": 0.0019843343179672956,
"rewards/rejected": -0.07712705433368683,
"step": 250
},
{
"epoch": 1.6613418530351438,
"grad_norm": 0.5579681347407102,
"learning_rate": 2.48140119418046e-07,
"log_odds_chosen": 0.09361070394515991,
"log_odds_ratio": -0.6759673953056335,
"logits/chosen": 30.127365112304688,
"logits/rejected": 29.955150604248047,
"logps/chosen": -0.7377598881721497,
"logps/rejected": -0.789117157459259,
"loss": 1.1595,
"nll_loss": 1.0882426500320435,
"rewards/accuracies": 0.581250011920929,
"rewards/chosen": -0.0737759917974472,
"rewards/margins": 0.0051357327029109,
"rewards/rejected": -0.07891170680522919,
"step": 260
},
{
"epoch": 1.7252396166134185,
"grad_norm": 0.5388500509178578,
"learning_rate": 1.634902137174483e-07,
"log_odds_chosen": 0.06861741840839386,
"log_odds_ratio": -0.6897167563438416,
"logits/chosen": 31.801860809326172,
"logits/rejected": 30.827133178710938,
"logps/chosen": -0.7297223210334778,
"logps/rejected": -0.7675324082374573,
"loss": 1.15,
"nll_loss": 1.091435194015503,
"rewards/accuracies": 0.5562499761581421,
"rewards/chosen": -0.07297223806381226,
"rewards/margins": 0.003781010629609227,
"rewards/rejected": -0.07675323635339737,
"step": 270
},
{
"epoch": 1.7891373801916934,
"grad_norm": 0.5597110218019701,
"learning_rate": 9.564769404039419e-08,
"log_odds_chosen": 0.10912013053894043,
"log_odds_ratio": -0.6706396341323853,
"logits/chosen": 30.64389419555664,
"logits/rejected": 30.014019012451172,
"logps/chosen": -0.7340375185012817,
"logps/rejected": -0.7948423624038696,
"loss": 1.1518,
"nll_loss": 1.0911478996276855,
"rewards/accuracies": 0.590624988079071,
"rewards/chosen": -0.0734037533402443,
"rewards/margins": 0.00608047703281045,
"rewards/rejected": -0.0794842392206192,
"step": 280
},
{
"epoch": 1.8530351437699681,
"grad_norm": 0.5432963139659418,
"learning_rate": 4.546571943496969e-08,
"log_odds_chosen": 0.071634940803051,
"log_odds_ratio": -0.6880494356155396,
"logits/chosen": 31.523670196533203,
"logits/rejected": 31.051345825195312,
"logps/chosen": -0.7210476398468018,
"logps/rejected": -0.7614242434501648,
"loss": 1.1457,
"nll_loss": 1.078147292137146,
"rewards/accuracies": 0.53125,
"rewards/chosen": -0.07210476696491241,
"rewards/margins": 0.0040376619435846806,
"rewards/rejected": -0.07614242285490036,
"step": 290
},
{
"epoch": 1.9169329073482428,
"grad_norm": 0.5550233900085839,
"learning_rate": 1.357535734809795e-08,
"log_odds_chosen": 0.08968226611614227,
"log_odds_ratio": -0.6760362386703491,
"logits/chosen": 30.686954498291016,
"logits/rejected": 29.28194236755371,
"logps/chosen": -0.7248884439468384,
"logps/rejected": -0.7753577828407288,
"loss": 1.1474,
"nll_loss": 1.0696710348129272,
"rewards/accuracies": 0.5531250238418579,
"rewards/chosen": -0.07248884439468384,
"rewards/margins": 0.005046932026743889,
"rewards/rejected": -0.07753578573465347,
"step": 300
},
{
"epoch": 1.9808306709265175,
"grad_norm": 0.49758643306678957,
"learning_rate": 3.77647586240204e-10,
"log_odds_chosen": 0.048125725239515305,
"log_odds_ratio": -0.6949858665466309,
"logits/chosen": 30.8554744720459,
"logits/rejected": 30.65484046936035,
"logps/chosen": -0.7389064431190491,
"logps/rejected": -0.7689411640167236,
"loss": 1.1435,
"nll_loss": 1.0751014947891235,
"rewards/accuracies": 0.534375011920929,
"rewards/chosen": -0.07389064878225327,
"rewards/margins": 0.0030034759547561407,
"rewards/rejected": -0.0768941268324852,
"step": 310
},
{
"epoch": 1.9936102236421727,
"step": 312,
"total_flos": 0.0,
"train_loss": 1.3676127867820935,
"train_runtime": 5524.9246,
"train_samples_per_second": 7.24,
"train_steps_per_second": 0.056
}
],
"logging_steps": 10,
"max_steps": 312,
"num_input_tokens_seen": 0,
"num_train_epochs": 2,
"save_steps": 500,
"stateful_callbacks": {
"TrainerControl": {
"args": {
"should_epoch_stop": false,
"should_evaluate": false,
"should_log": false,
"should_save": true,
"should_training_stop": true
},
"attributes": {}
}
},
"total_flos": 0.0,
"train_batch_size": 16,
"trial_name": null,
"trial_params": null
}