deepseek-8b-orpo-lora / trainer_state.json
zfz1's picture
Model save
9cb3e8d verified
{
"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": 0.4242139016276009,
"learning_rate": 3.125e-05,
"log_odds_chosen": 0.05670202523469925,
"log_odds_ratio": -0.6963189840316772,
"logits/chosen": 33.65892791748047,
"logits/rejected": 33.56386947631836,
"logps/chosen": -0.9726333618164062,
"logps/rejected": -1.0167349576950073,
"loss": 1.6695,
"nll_loss": 1.5828170776367188,
"rewards/accuracies": 0.5406249761581421,
"rewards/chosen": -0.09726335108280182,
"rewards/margins": 0.004410145338624716,
"rewards/rejected": -0.10167349874973297,
"step": 10
},
{
"epoch": 0.12779552715654952,
"grad_norm": 0.527757575829653,
"learning_rate": 6.25e-05,
"log_odds_chosen": 0.060249220579862595,
"log_odds_ratio": -0.6938394904136658,
"logits/chosen": 34.7052001953125,
"logits/rejected": 34.30677032470703,
"logps/chosen": -0.8907906413078308,
"logps/rejected": -0.9303584098815918,
"loss": 1.4944,
"nll_loss": 1.4315097332000732,
"rewards/accuracies": 0.546875,
"rewards/chosen": -0.08907906711101532,
"rewards/margins": 0.003956770058721304,
"rewards/rejected": -0.09303583949804306,
"step": 20
},
{
"epoch": 0.19169329073482427,
"grad_norm": 0.4014606132018809,
"learning_rate": 9.375e-05,
"log_odds_chosen": 0.04292842000722885,
"log_odds_ratio": -0.698261022567749,
"logits/chosen": 30.987014770507812,
"logits/rejected": 30.439056396484375,
"logps/chosen": -0.6581910252571106,
"logps/rejected": -0.6830389499664307,
"loss": 1.0721,
"nll_loss": 1.006190538406372,
"rewards/accuracies": 0.5375000238418579,
"rewards/chosen": -0.06581910699605942,
"rewards/margins": 0.002484795870259404,
"rewards/rejected": -0.0683038979768753,
"step": 30
},
{
"epoch": 0.25559105431309903,
"grad_norm": 0.09410150706534143,
"learning_rate": 9.979871469976196e-05,
"log_odds_chosen": 0.07667034864425659,
"log_odds_ratio": -0.6961523294448853,
"logits/chosen": 28.151453018188477,
"logits/rejected": 27.650014877319336,
"logps/chosen": -0.43666666746139526,
"logps/rejected": -0.46118512749671936,
"loss": 0.8277,
"nll_loss": 0.7572250366210938,
"rewards/accuracies": 0.5687500238418579,
"rewards/chosen": -0.043666668236255646,
"rewards/margins": 0.0024518456775695086,
"rewards/rejected": -0.046118512749671936,
"step": 40
},
{
"epoch": 0.3194888178913738,
"grad_norm": 0.09768125665941661,
"learning_rate": 9.898376992116179e-05,
"log_odds_chosen": 0.06605090945959091,
"log_odds_ratio": -0.7164387106895447,
"logits/chosen": 29.047494888305664,
"logits/rejected": 27.90814781188965,
"logps/chosen": -0.37932097911834717,
"logps/rejected": -0.39148497581481934,
"loss": 0.7778,
"nll_loss": 0.7044206261634827,
"rewards/accuracies": 0.5375000238418579,
"rewards/chosen": -0.03793209791183472,
"rewards/margins": 0.0012163992505520582,
"rewards/rejected": -0.03914849832653999,
"step": 50
},
{
"epoch": 0.38338658146964855,
"grad_norm": 0.07343907441931204,
"learning_rate": 9.755282581475769e-05,
"log_odds_chosen": 0.08987905830144882,
"log_odds_ratio": -0.7063173055648804,
"logits/chosen": 28.47494125366211,
"logits/rejected": 28.794668197631836,
"logps/chosen": -0.3645591139793396,
"logps/rejected": -0.38652682304382324,
"loss": 0.7576,
"nll_loss": 0.6854395866394043,
"rewards/accuracies": 0.559374988079071,
"rewards/chosen": -0.0364559181034565,
"rewards/margins": 0.0021967687644064426,
"rewards/rejected": -0.038652680814266205,
"step": 60
},
{
"epoch": 0.4472843450479233,
"grad_norm": 0.06951733417216101,
"learning_rate": 9.552387733294081e-05,
"log_odds_chosen": 0.1109728068113327,
"log_odds_ratio": -0.7004402875900269,
"logits/chosen": 29.863407135009766,
"logits/rejected": 30.27783203125,
"logps/chosen": -0.35166820883750916,
"logps/rejected": -0.3710673451423645,
"loss": 0.7621,
"nll_loss": 0.6829724907875061,
"rewards/accuracies": 0.559374988079071,
"rewards/chosen": -0.035166822373867035,
"rewards/margins": 0.0019399106968194246,
"rewards/rejected": -0.03710673004388809,
"step": 70
},
{
"epoch": 0.5111821086261981,
"grad_norm": 0.06675882825106033,
"learning_rate": 9.292243968009331e-05,
"log_odds_chosen": 0.22688157856464386,
"log_odds_ratio": -0.6433924436569214,
"logits/chosen": 31.850915908813477,
"logits/rejected": 32.051448822021484,
"logps/chosen": -0.35137858986854553,
"logps/rejected": -0.4119800925254822,
"loss": 0.7548,
"nll_loss": 0.6953305602073669,
"rewards/accuracies": 0.640625,
"rewards/chosen": -0.03513786196708679,
"rewards/margins": 0.006060150917619467,
"rewards/rejected": -0.0411980114877224,
"step": 80
},
{
"epoch": 0.5750798722044729,
"grad_norm": 0.07132675829804414,
"learning_rate": 8.978122744408906e-05,
"log_odds_chosen": 0.20013344287872314,
"log_odds_ratio": -0.6580983400344849,
"logits/chosen": 32.63883972167969,
"logits/rejected": 32.50263595581055,
"logps/chosen": -0.35542625188827515,
"logps/rejected": -0.41074585914611816,
"loss": 0.7549,
"nll_loss": 0.6860114932060242,
"rewards/accuracies": 0.6312500238418579,
"rewards/chosen": -0.03554262965917587,
"rewards/margins": 0.0055319564417004585,
"rewards/rejected": -0.04107458516955376,
"step": 90
},
{
"epoch": 0.6389776357827476,
"grad_norm": 0.09149445807314589,
"learning_rate": 8.613974319136958e-05,
"log_odds_chosen": 0.2682611346244812,
"log_odds_ratio": -0.6336508989334106,
"logits/chosen": 32.11505889892578,
"logits/rejected": 32.7309455871582,
"logps/chosen": -0.35972970724105835,
"logps/rejected": -0.4426427483558655,
"loss": 0.753,
"nll_loss": 0.6944054365158081,
"rewards/accuracies": 0.653124988079071,
"rewards/chosen": -0.035972971469163895,
"rewards/margins": 0.008291301317512989,
"rewards/rejected": -0.04426427185535431,
"step": 100
},
{
"epoch": 0.7028753993610224,
"grad_norm": 0.08926217777865493,
"learning_rate": 8.20437806992512e-05,
"log_odds_chosen": 0.2868707776069641,
"log_odds_ratio": -0.6233163475990295,
"logits/chosen": 35.566490173339844,
"logits/rejected": 34.921634674072266,
"logps/chosen": -0.3582688868045807,
"logps/rejected": -0.4581901431083679,
"loss": 0.744,
"nll_loss": 0.6832225918769836,
"rewards/accuracies": 0.6343749761581421,
"rewards/chosen": -0.03582689166069031,
"rewards/margins": 0.00999213196337223,
"rewards/rejected": -0.04581902176141739,
"step": 110
},
{
"epoch": 0.7667731629392971,
"grad_norm": 0.14837704688840048,
"learning_rate": 7.754484907260513e-05,
"log_odds_chosen": 0.30330824851989746,
"log_odds_ratio": -0.6247963905334473,
"logits/chosen": 34.01176834106445,
"logits/rejected": 34.39574432373047,
"logps/chosen": -0.35339340567588806,
"logps/rejected": -0.457806259393692,
"loss": 0.7419,
"nll_loss": 0.6851586103439331,
"rewards/accuracies": 0.6187499761581421,
"rewards/chosen": -0.035339340567588806,
"rewards/margins": 0.01044128555804491,
"rewards/rejected": -0.045780621469020844,
"step": 120
},
{
"epoch": 0.8306709265175719,
"grad_norm": 0.10699968939091851,
"learning_rate": 7.269952498697734e-05,
"log_odds_chosen": 0.3560214638710022,
"log_odds_ratio": -0.6051545143127441,
"logits/chosen": 36.380924224853516,
"logits/rejected": 35.524024963378906,
"logps/chosen": -0.355410635471344,
"logps/rejected": -0.47948652505874634,
"loss": 0.7296,
"nll_loss": 0.6667423844337463,
"rewards/accuracies": 0.668749988079071,
"rewards/chosen": -0.03554106876254082,
"rewards/margins": 0.01240758690983057,
"rewards/rejected": -0.04794865846633911,
"step": 130
},
{
"epoch": 0.8945686900958466,
"grad_norm": 0.1720595584929413,
"learning_rate": 6.756874120406714e-05,
"log_odds_chosen": 0.5141419172286987,
"log_odds_ratio": -0.5401668548583984,
"logits/chosen": 34.641624450683594,
"logits/rejected": 34.34906768798828,
"logps/chosen": -0.347829669713974,
"logps/rejected": -0.5514229536056519,
"loss": 0.7143,
"nll_loss": 0.6645579934120178,
"rewards/accuracies": 0.731249988079071,
"rewards/chosen": -0.03478296846151352,
"rewards/margins": 0.020359333604574203,
"rewards/rejected": -0.05514230206608772,
"step": 140
},
{
"epoch": 0.9584664536741214,
"grad_norm": 0.12630063541107736,
"learning_rate": 6.22170203068947e-05,
"log_odds_chosen": 0.5517296195030212,
"log_odds_ratio": -0.543639600276947,
"logits/chosen": 36.98326873779297,
"logits/rejected": 36.52146530151367,
"logps/chosen": -0.3562212884426117,
"logps/rejected": -0.578203022480011,
"loss": 0.7162,
"nll_loss": 0.6639058589935303,
"rewards/accuracies": 0.731249988079071,
"rewards/chosen": -0.03562213480472565,
"rewards/margins": 0.02219817042350769,
"rewards/rejected": -0.05782030150294304,
"step": 150
},
{
"epoch": 1.0223642172523961,
"grad_norm": 0.1403636887974213,
"learning_rate": 5.6711663290882776e-05,
"log_odds_chosen": 0.6598159074783325,
"log_odds_ratio": -0.5024036169052124,
"logits/chosen": 36.992469787597656,
"logits/rejected": 35.901676177978516,
"logps/chosen": -0.33446019887924194,
"logps/rejected": -0.5835962891578674,
"loss": 0.7105,
"nll_loss": 0.657503604888916,
"rewards/accuracies": 0.7562500238418579,
"rewards/chosen": -0.033446017652750015,
"rewards/margins": 0.0249136071652174,
"rewards/rejected": -0.05835963040590286,
"step": 160
},
{
"epoch": 1.0862619808306708,
"grad_norm": 0.14799945317167665,
"learning_rate": 5.112190321479026e-05,
"log_odds_chosen": 0.6551031470298767,
"log_odds_ratio": -0.5045552849769592,
"logits/chosen": 35.59492874145508,
"logits/rejected": 35.479576110839844,
"logps/chosen": -0.36466851830482483,
"logps/rejected": -0.6384136080741882,
"loss": 0.7194,
"nll_loss": 0.6874456405639648,
"rewards/accuracies": 0.765625,
"rewards/chosen": -0.03646685555577278,
"rewards/margins": 0.027374515309929848,
"rewards/rejected": -0.06384135782718658,
"step": 170
},
{
"epoch": 1.1501597444089458,
"grad_norm": 0.4144450852693044,
"learning_rate": 4.551803455482833e-05,
"log_odds_chosen": 0.8569015264511108,
"log_odds_ratio": -0.43813252449035645,
"logits/chosen": 36.478363037109375,
"logits/rejected": 35.625038146972656,
"logps/chosen": -0.3283715844154358,
"logps/rejected": -0.6759421229362488,
"loss": 0.7011,
"nll_loss": 0.6358424425125122,
"rewards/accuracies": 0.809374988079071,
"rewards/chosen": -0.0328371599316597,
"rewards/margins": 0.03475705534219742,
"rewards/rejected": -0.06759421527385712,
"step": 180
},
{
"epoch": 1.2140575079872205,
"grad_norm": 0.1393106135709619,
"learning_rate": 3.9970529210836366e-05,
"log_odds_chosen": 0.7694636583328247,
"log_odds_ratio": -0.48203492164611816,
"logits/chosen": 37.65850067138672,
"logits/rejected": 37.32286071777344,
"logps/chosen": -0.34162312746047974,
"logps/rejected": -0.6649552583694458,
"loss": 0.696,
"nll_loss": 0.6590245962142944,
"rewards/accuracies": 0.78125,
"rewards/chosen": -0.034162312746047974,
"rewards/margins": 0.032333213835954666,
"rewards/rejected": -0.06649552285671234,
"step": 190
},
{
"epoch": 1.2779552715654952,
"grad_norm": 0.18676893862042468,
"learning_rate": 3.4549150281252636e-05,
"log_odds_chosen": 0.8638985753059387,
"log_odds_ratio": -0.4615907073020935,
"logits/chosen": 34.856327056884766,
"logits/rejected": 34.688533782958984,
"logps/chosen": -0.3576621413230896,
"logps/rejected": -0.74553382396698,
"loss": 0.6981,
"nll_loss": 0.6673237085342407,
"rewards/accuracies": 0.78125,
"rewards/chosen": -0.03576621413230896,
"rewards/margins": 0.03878717124462128,
"rewards/rejected": -0.07455337792634964,
"step": 200
},
{
"epoch": 1.34185303514377,
"grad_norm": 0.2564156817901977,
"learning_rate": 2.932207475167398e-05,
"log_odds_chosen": 0.8325332403182983,
"log_odds_ratio": -0.45953792333602905,
"logits/chosen": 36.798072814941406,
"logits/rejected": 35.09117889404297,
"logps/chosen": -0.3448064923286438,
"logps/rejected": -0.7093031406402588,
"loss": 0.6982,
"nll_loss": 0.6432278752326965,
"rewards/accuracies": 0.8062499761581421,
"rewards/chosen": -0.03448065370321274,
"rewards/margins": 0.03644966334104538,
"rewards/rejected": -0.07093031704425812,
"step": 210
},
{
"epoch": 1.4057507987220448,
"grad_norm": 0.1531688356816753,
"learning_rate": 2.43550361297047e-05,
"log_odds_chosen": 1.0281397104263306,
"log_odds_ratio": -0.4170387387275696,
"logits/chosen": 35.599212646484375,
"logits/rejected": 34.3337287902832,
"logps/chosen": -0.3171806335449219,
"logps/rejected": -0.7794741988182068,
"loss": 0.6881,
"nll_loss": 0.6331400275230408,
"rewards/accuracies": 0.84375,
"rewards/chosen": -0.03171805664896965,
"rewards/margins": 0.04622935503721237,
"rewards/rejected": -0.07794742286205292,
"step": 220
},
{
"epoch": 1.4696485623003195,
"grad_norm": 0.13237811516695533,
"learning_rate": 1.971049780795901e-05,
"log_odds_chosen": 0.9918270111083984,
"log_odds_ratio": -0.440161794424057,
"logits/chosen": 34.80529022216797,
"logits/rejected": 34.433349609375,
"logps/chosen": -0.3528688848018646,
"logps/rejected": -0.8333184123039246,
"loss": 0.684,
"nll_loss": 0.653908908367157,
"rewards/accuracies": 0.8125,
"rewards/chosen": -0.03528688848018646,
"rewards/margins": 0.04804495349526405,
"rewards/rejected": -0.08333183825016022,
"step": 230
},
{
"epoch": 1.5335463258785942,
"grad_norm": 0.17618952870677607,
"learning_rate": 1.544686755065677e-05,
"log_odds_chosen": 1.110446810722351,
"log_odds_ratio": -0.39400768280029297,
"logits/chosen": 35.90544891357422,
"logits/rejected": 34.70677185058594,
"logps/chosen": -0.3103191554546356,
"logps/rejected": -0.8037627935409546,
"loss": 0.6899,
"nll_loss": 0.6227988004684448,
"rewards/accuracies": 0.824999988079071,
"rewards/chosen": -0.03103191778063774,
"rewards/margins": 0.04934436455368996,
"rewards/rejected": -0.0803762823343277,
"step": 240
},
{
"epoch": 1.5974440894568689,
"grad_norm": 0.17643350651397863,
"learning_rate": 1.1617762982099446e-05,
"log_odds_chosen": 1.0278904438018799,
"log_odds_ratio": -0.40579158067703247,
"logits/chosen": 36.86786651611328,
"logits/rejected": 35.534324645996094,
"logps/chosen": -0.34132060408592224,
"logps/rejected": -0.8093164563179016,
"loss": 0.6839,
"nll_loss": 0.6438942551612854,
"rewards/accuracies": 0.8218749761581421,
"rewards/chosen": -0.034132059663534164,
"rewards/margins": 0.04679957777261734,
"rewards/rejected": -0.0809316486120224,
"step": 250
},
{
"epoch": 1.6613418530351438,
"grad_norm": 0.20335250536853633,
"learning_rate": 8.271337313934869e-06,
"log_odds_chosen": 1.0964257717132568,
"log_odds_ratio": -0.40502405166625977,
"logits/chosen": 35.01730728149414,
"logits/rejected": 33.87272262573242,
"logps/chosen": -0.35952043533325195,
"logps/rejected": -0.8931809663772583,
"loss": 0.6875,
"nll_loss": 0.6542503833770752,
"rewards/accuracies": 0.8500000238418579,
"rewards/chosen": -0.035952042788267136,
"rewards/margins": 0.0533660463988781,
"rewards/rejected": -0.08931808918714523,
"step": 260
},
{
"epoch": 1.7252396166134185,
"grad_norm": 0.1417159132341433,
"learning_rate": 5.449673790581611e-06,
"log_odds_chosen": 1.1632859706878662,
"log_odds_ratio": -0.39327362179756165,
"logits/chosen": 37.34235763549805,
"logits/rejected": 35.14636993408203,
"logps/chosen": -0.3363361358642578,
"logps/rejected": -0.8959217071533203,
"loss": 0.6837,
"nll_loss": 0.6559703350067139,
"rewards/accuracies": 0.8531249761581421,
"rewards/chosen": -0.03363361582159996,
"rewards/margins": 0.055958546698093414,
"rewards/rejected": -0.08959217369556427,
"step": 270
},
{
"epoch": 1.7891373801916934,
"grad_norm": 0.39828151397493416,
"learning_rate": 3.18825646801314e-06,
"log_odds_chosen": 1.0620207786560059,
"log_odds_ratio": -0.41589412093162537,
"logits/chosen": 35.572288513183594,
"logits/rejected": 34.087669372558594,
"logps/chosen": -0.3491634726524353,
"logps/rejected": -0.8419939875602722,
"loss": 0.6856,
"nll_loss": 0.6521117091178894,
"rewards/accuracies": 0.8500000238418579,
"rewards/chosen": -0.03491634875535965,
"rewards/margins": 0.04928305745124817,
"rewards/rejected": -0.08419940620660782,
"step": 280
},
{
"epoch": 1.8530351437699681,
"grad_norm": 0.17410175622655502,
"learning_rate": 1.5155239811656563e-06,
"log_odds_chosen": 1.1792383193969727,
"log_odds_ratio": -0.3946213126182556,
"logits/chosen": 36.629905700683594,
"logits/rejected": 35.32917404174805,
"logps/chosen": -0.33119866251945496,
"logps/rejected": -0.899772047996521,
"loss": 0.6837,
"nll_loss": 0.6425603628158569,
"rewards/accuracies": 0.815625011920929,
"rewards/chosen": -0.033119868487119675,
"rewards/margins": 0.05685734748840332,
"rewards/rejected": -0.0899772122502327,
"step": 290
},
{
"epoch": 1.9169329073482428,
"grad_norm": 0.13374493386406422,
"learning_rate": 4.52511911603265e-07,
"log_odds_chosen": 1.0675084590911865,
"log_odds_ratio": -0.42029696702957153,
"logits/chosen": 36.18767547607422,
"logits/rejected": 33.20973205566406,
"logps/chosen": -0.34383276104927063,
"logps/rejected": -0.8607551455497742,
"loss": 0.6844,
"nll_loss": 0.6432604193687439,
"rewards/accuracies": 0.809374988079071,
"rewards/chosen": -0.034383274614810944,
"rewards/margins": 0.05169224739074707,
"rewards/rejected": -0.08607552200555801,
"step": 300
},
{
"epoch": 1.9808306709265175,
"grad_norm": 0.1377772193573712,
"learning_rate": 1.2588252874673468e-08,
"log_odds_chosen": 1.0935267210006714,
"log_odds_ratio": -0.41092976927757263,
"logits/chosen": 36.207176208496094,
"logits/rejected": 34.82603073120117,
"logps/chosen": -0.344975084066391,
"logps/rejected": -0.8865535855293274,
"loss": 0.6781,
"nll_loss": 0.6345787644386292,
"rewards/accuracies": 0.8062499761581421,
"rewards/chosen": -0.03449751064181328,
"rewards/margins": 0.05415785312652588,
"rewards/rejected": -0.08865536749362946,
"step": 310
},
{
"epoch": 1.9936102236421727,
"step": 312,
"total_flos": 0.0,
"train_loss": 0.7846963420892373,
"train_runtime": 5577.3844,
"train_samples_per_second": 7.172,
"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
}