gemma-7b-orpo-noisy-1e-5-norm / trainer_state.json
silviasapora's picture
Model save
c2b98d9 verified
{
"best_metric": null,
"best_model_checkpoint": null,
"epoch": 2.9765925925925925,
"eval_steps": 500,
"global_step": 315,
"is_hyper_param_search": false,
"is_local_process_zero": true,
"is_world_process_zero": true,
"log_history": [
{
"epoch": 0.047407407407407405,
"grad_norm": 2580.320556640625,
"learning_rate": 1.5625e-06,
"log_odds_chosen": 5.033224582672119,
"log_odds_ratio": -9.833906173706055,
"logits/chosen": 138.36488342285156,
"logits/rejected": 152.9020233154297,
"logps/chosen": -20.877660751342773,
"logps/rejected": -25.9111385345459,
"loss": 459.1401,
"nll_loss": 8.45106315612793,
"rewards/accuracies": 0.53125,
"rewards/chosen": -10.438830375671387,
"rewards/margins": 2.5167386531829834,
"rewards/rejected": -12.95556926727295,
"step": 5
},
{
"epoch": 0.09481481481481481,
"grad_norm": 956.760009765625,
"learning_rate": 3.125e-06,
"log_odds_chosen": 2.400622606277466,
"log_odds_ratio": -7.376997470855713,
"logits/chosen": 131.07968139648438,
"logits/rejected": 155.00018310546875,
"logps/chosen": -16.098163604736328,
"logps/rejected": -18.498001098632812,
"loss": 398.9935,
"nll_loss": 7.736745357513428,
"rewards/accuracies": 0.543749988079071,
"rewards/chosen": -8.049081802368164,
"rewards/margins": 1.1999186277389526,
"rewards/rejected": -9.249000549316406,
"step": 10
},
{
"epoch": 0.14222222222222222,
"grad_norm": 571.1576538085938,
"learning_rate": 4.6875000000000004e-06,
"log_odds_chosen": 4.6121039390563965,
"log_odds_ratio": -8.182807922363281,
"logits/chosen": 105.5162124633789,
"logits/rejected": 132.4467315673828,
"logps/chosen": -21.264659881591797,
"logps/rejected": -25.8754940032959,
"loss": 389.0222,
"nll_loss": 9.0397367477417,
"rewards/accuracies": 0.581250011920929,
"rewards/chosen": -10.632329940795898,
"rewards/margins": 2.30541729927063,
"rewards/rejected": -12.93774700164795,
"step": 15
},
{
"epoch": 0.18962962962962962,
"grad_norm": 905.5711669921875,
"learning_rate": 6.25e-06,
"log_odds_chosen": -0.20019039511680603,
"log_odds_ratio": -7.8728742599487305,
"logits/chosen": 133.976318359375,
"logits/rejected": 139.43692016601562,
"logps/chosen": -18.147478103637695,
"logps/rejected": -17.94868278503418,
"loss": 419.4126,
"nll_loss": 8.399177551269531,
"rewards/accuracies": 0.53125,
"rewards/chosen": -9.073739051818848,
"rewards/margins": -0.09939660876989365,
"rewards/rejected": -8.97434139251709,
"step": 20
},
{
"epoch": 0.23703703703703705,
"grad_norm": 316.11334228515625,
"learning_rate": 7.8125e-06,
"log_odds_chosen": 4.466104030609131,
"log_odds_ratio": -7.779648780822754,
"logits/chosen": 137.90159606933594,
"logits/rejected": 146.57969665527344,
"logps/chosen": -18.408245086669922,
"logps/rejected": -22.874719619750977,
"loss": 347.403,
"nll_loss": 7.184973239898682,
"rewards/accuracies": 0.518750011920929,
"rewards/chosen": -9.204122543334961,
"rewards/margins": 2.2332377433776855,
"rewards/rejected": -11.437359809875488,
"step": 25
},
{
"epoch": 0.28444444444444444,
"grad_norm": 597.9056396484375,
"learning_rate": 9.375000000000001e-06,
"log_odds_chosen": 1.4436252117156982,
"log_odds_ratio": -7.299798011779785,
"logits/chosen": 127.5577621459961,
"logits/rejected": 165.3824005126953,
"logps/chosen": -18.25235366821289,
"logps/rejected": -19.692398071289062,
"loss": 367.166,
"nll_loss": 7.642889976501465,
"rewards/accuracies": 0.5874999761581421,
"rewards/chosen": -9.126176834106445,
"rewards/margins": 0.7200227379798889,
"rewards/rejected": -9.846199035644531,
"step": 30
},
{
"epoch": 0.33185185185185184,
"grad_norm": 1674.3773193359375,
"learning_rate": 9.997227514697568e-06,
"log_odds_chosen": -2.2156496047973633,
"log_odds_ratio": -7.339943885803223,
"logits/chosen": 163.3466033935547,
"logits/rejected": 147.68124389648438,
"logps/chosen": -15.995862007141113,
"logps/rejected": -13.778966903686523,
"loss": 308.5459,
"nll_loss": 6.2163190841674805,
"rewards/accuracies": 0.5,
"rewards/chosen": -7.997931003570557,
"rewards/margins": -1.1084463596343994,
"rewards/rejected": -6.889483451843262,
"step": 35
},
{
"epoch": 0.37925925925925924,
"grad_norm": 1445.621337890625,
"learning_rate": 9.980295682286924e-06,
"log_odds_chosen": 0.39161425828933716,
"log_odds_ratio": -2.787511110305786,
"logits/chosen": 178.052978515625,
"logits/rejected": 191.52259826660156,
"logps/chosen": -8.379082679748535,
"logps/rejected": -8.763626098632812,
"loss": 205.9666,
"nll_loss": 4.583432197570801,
"rewards/accuracies": 0.5062500238418579,
"rewards/chosen": -4.189541339874268,
"rewards/margins": 0.19227202236652374,
"rewards/rejected": -4.381813049316406,
"step": 40
},
{
"epoch": 0.4266666666666667,
"grad_norm": 437.58502197265625,
"learning_rate": 9.94802437440896e-06,
"log_odds_chosen": 0.5884231925010681,
"log_odds_ratio": -1.4498884677886963,
"logits/chosen": 213.7032928466797,
"logits/rejected": 203.2352752685547,
"logps/chosen": -3.589550733566284,
"logps/rejected": -4.178198337554932,
"loss": 102.45,
"nll_loss": 2.611146926879883,
"rewards/accuracies": 0.5062500238418579,
"rewards/chosen": -1.794775366783142,
"rewards/margins": 0.29432395100593567,
"rewards/rejected": -2.089099168777466,
"step": 45
},
{
"epoch": 0.4740740740740741,
"grad_norm": 180.1491241455078,
"learning_rate": 9.90051298775959e-06,
"log_odds_chosen": 0.2308872491121292,
"log_odds_ratio": -0.8247494697570801,
"logits/chosen": 221.44058227539062,
"logits/rejected": 221.6063995361328,
"logps/chosen": -1.8605806827545166,
"logps/rejected": -2.0761098861694336,
"loss": 83.7704,
"nll_loss": 2.2189698219299316,
"rewards/accuracies": 0.53125,
"rewards/chosen": -0.9302903413772583,
"rewards/margins": 0.10776450484991074,
"rewards/rejected": -1.0380549430847168,
"step": 50
},
{
"epoch": 0.5214814814814814,
"grad_norm": 131.46673583984375,
"learning_rate": 9.837907858981536e-06,
"log_odds_chosen": 0.12095841020345688,
"log_odds_ratio": -0.8109160661697388,
"logits/chosen": 231.9627685546875,
"logits/rejected": 238.06460571289062,
"logps/chosen": -1.6934187412261963,
"logps/rejected": -1.7964763641357422,
"loss": 74.7448,
"nll_loss": 2.007000207901001,
"rewards/accuracies": 0.5625,
"rewards/chosen": -0.8467093706130981,
"rewards/margins": 0.051528822630643845,
"rewards/rejected": -0.8982381820678711,
"step": 55
},
{
"epoch": 0.5688888888888889,
"grad_norm": 149.99351501464844,
"learning_rate": 9.760401813942641e-06,
"log_odds_chosen": 0.2649911046028137,
"log_odds_ratio": -0.7256797552108765,
"logits/chosen": 218.29061889648438,
"logits/rejected": 230.5416259765625,
"logps/chosen": -1.5244457721710205,
"logps/rejected": -1.7421019077301025,
"loss": 70.0635,
"nll_loss": 1.8057339191436768,
"rewards/accuracies": 0.5625,
"rewards/chosen": -0.7622228860855103,
"rewards/margins": 0.10882814228534698,
"rewards/rejected": -0.8710509538650513,
"step": 60
},
{
"epoch": 0.6162962962962963,
"grad_norm": 172.64825439453125,
"learning_rate": 9.668233573825794e-06,
"log_odds_chosen": 0.12981751561164856,
"log_odds_ratio": -0.8062798380851746,
"logits/chosen": 232.5802764892578,
"logits/rejected": 233.1844482421875,
"logps/chosen": -1.5933538675308228,
"logps/rejected": -1.6922533512115479,
"loss": 68.5854,
"nll_loss": 1.815799355506897,
"rewards/accuracies": 0.53125,
"rewards/chosen": -0.7966769337654114,
"rewards/margins": 0.049449823796749115,
"rewards/rejected": -0.8461266756057739,
"step": 65
},
{
"epoch": 0.6637037037037037,
"grad_norm": 136.158447265625,
"learning_rate": 9.56168701985981e-06,
"log_odds_chosen": 0.42648229002952576,
"log_odds_ratio": -0.6542950868606567,
"logits/chosen": 224.38955688476562,
"logits/rejected": 242.11837768554688,
"logps/chosen": -1.4570573568344116,
"logps/rejected": -1.8463979959487915,
"loss": 67.3288,
"nll_loss": 1.753805160522461,
"rewards/accuracies": 0.5874999761581421,
"rewards/chosen": -0.7285286784172058,
"rewards/margins": 0.19467028975486755,
"rewards/rejected": -0.9231989979743958,
"step": 70
},
{
"epoch": 0.7111111111111111,
"grad_norm": 115.01741790771484,
"learning_rate": 9.441090318955843e-06,
"log_odds_chosen": 0.24410974979400635,
"log_odds_ratio": -0.7884246706962585,
"logits/chosen": 249.4243621826172,
"logits/rejected": 242.49984741210938,
"logps/chosen": -1.5666649341583252,
"logps/rejected": -1.7553819417953491,
"loss": 66.3816,
"nll_loss": 1.7346446514129639,
"rewards/accuracies": 0.6312500238418579,
"rewards/chosen": -0.7833324670791626,
"rewards/margins": 0.09435858577489853,
"rewards/rejected": -0.8776909708976746,
"step": 75
},
{
"epoch": 0.7585185185185185,
"grad_norm": 117.21932983398438,
"learning_rate": 9.306814912942445e-06,
"log_odds_chosen": 0.33950307965278625,
"log_odds_ratio": -0.7339600920677185,
"logits/chosen": 235.14981079101562,
"logits/rejected": 246.0205078125,
"logps/chosen": -1.488161325454712,
"logps/rejected": -1.7720199823379517,
"loss": 64.8768,
"nll_loss": 1.7132374048233032,
"rewards/accuracies": 0.606249988079071,
"rewards/chosen": -0.744080662727356,
"rewards/margins": 0.1419292837381363,
"rewards/rejected": -0.8860099911689758,
"step": 80
},
{
"epoch": 0.8059259259259259,
"grad_norm": 94.97506713867188,
"learning_rate": 9.159274374512444e-06,
"log_odds_chosen": 0.3309640884399414,
"log_odds_ratio": -0.7209373116493225,
"logits/chosen": 235.80746459960938,
"logits/rejected": 265.1528015136719,
"logps/chosen": -1.376119613647461,
"logps/rejected": -1.6490930318832397,
"loss": 63.5063,
"nll_loss": 1.6010291576385498,
"rewards/accuracies": 0.643750011920929,
"rewards/chosen": -0.6880598068237305,
"rewards/margins": 0.13648679852485657,
"rewards/rejected": -0.8245465159416199,
"step": 85
},
{
"epoch": 0.8533333333333334,
"grad_norm": 124.18742370605469,
"learning_rate": 8.99892313340537e-06,
"log_odds_chosen": 0.28358036279678345,
"log_odds_ratio": -0.7167706489562988,
"logits/chosen": 244.89981079101562,
"logits/rejected": 260.676025390625,
"logps/chosen": -1.414159893989563,
"logps/rejected": -1.6283620595932007,
"loss": 65.2253,
"nll_loss": 1.647252082824707,
"rewards/accuracies": 0.612500011920929,
"rewards/chosen": -0.7070799469947815,
"rewards/margins": 0.10710110515356064,
"rewards/rejected": -0.8141810297966003,
"step": 90
},
{
"epoch": 0.9007407407407407,
"grad_norm": 100.92960357666016,
"learning_rate": 8.826255076748823e-06,
"log_odds_chosen": 0.3998146057128906,
"log_odds_ratio": -0.6268864870071411,
"logits/chosen": 235.48410034179688,
"logits/rejected": 251.33712768554688,
"logps/chosen": -1.2698018550872803,
"logps/rejected": -1.5803816318511963,
"loss": 63.0803,
"nll_loss": 1.5454211235046387,
"rewards/accuracies": 0.6625000238418579,
"rewards/chosen": -0.6349009275436401,
"rewards/margins": 0.15528982877731323,
"rewards/rejected": -0.7901908159255981,
"step": 95
},
{
"epoch": 0.9481481481481482,
"grad_norm": 141.9030303955078,
"learning_rate": 8.641802027869774e-06,
"log_odds_chosen": 0.21526363492012024,
"log_odds_ratio": -0.7336837649345398,
"logits/chosen": 244.89962768554688,
"logits/rejected": 258.1126403808594,
"logps/chosen": -1.3898332118988037,
"logps/rejected": -1.5845530033111572,
"loss": 63.6891,
"nll_loss": 1.6614280939102173,
"rewards/accuracies": 0.5562499761581421,
"rewards/chosen": -0.6949166059494019,
"rewards/margins": 0.09735982120037079,
"rewards/rejected": -0.7922765016555786,
"step": 100
},
{
"epoch": 0.9955555555555555,
"grad_norm": 100.77186584472656,
"learning_rate": 8.446132108261136e-06,
"log_odds_chosen": 0.35006898641586304,
"log_odds_ratio": -0.6226455569267273,
"logits/chosen": 238.6287841796875,
"logits/rejected": 256.33172607421875,
"logps/chosen": -1.2620939016342163,
"logps/rejected": -1.5345309972763062,
"loss": 59.8393,
"nll_loss": 1.5451446771621704,
"rewards/accuracies": 0.6625000238418579,
"rewards/chosen": -0.6310469508171082,
"rewards/margins": 0.13621851801872253,
"rewards/rejected": -0.7672654986381531,
"step": 105
},
{
"epoch": 1.037925925925926,
"grad_norm": 96.23641967773438,
"learning_rate": 8.23984798774876e-06,
"log_odds_chosen": 0.34819138050079346,
"log_odds_ratio": -0.654488205909729,
"logits/chosen": 246.05015563964844,
"logits/rejected": 255.47967529296875,
"logps/chosen": -1.2216328382492065,
"logps/rejected": -1.5190995931625366,
"loss": 52.3665,
"nll_loss": 1.4913266897201538,
"rewards/accuracies": 0.5944055914878845,
"rewards/chosen": -0.6108164191246033,
"rewards/margins": 0.148733451962471,
"rewards/rejected": -0.7595497965812683,
"step": 110
},
{
"epoch": 1.0853333333333333,
"grad_norm": 102.46453094482422,
"learning_rate": 8.023585028248435e-06,
"log_odds_chosen": 0.5755366086959839,
"log_odds_ratio": -0.5598169565200806,
"logits/chosen": 231.81973266601562,
"logits/rejected": 254.8740234375,
"logps/chosen": -1.1871004104614258,
"logps/rejected": -1.6207106113433838,
"loss": 56.5061,
"nll_loss": 1.442861795425415,
"rewards/accuracies": 0.6937500238418579,
"rewards/chosen": -0.5935502052307129,
"rewards/margins": 0.21680507063865662,
"rewards/rejected": -0.8103553056716919,
"step": 115
},
{
"epoch": 1.1327407407407408,
"grad_norm": 95.08788299560547,
"learning_rate": 7.798009326830167e-06,
"log_odds_chosen": 0.5298383235931396,
"log_odds_ratio": -0.5681466460227966,
"logits/chosen": 227.4845428466797,
"logits/rejected": 237.4995880126953,
"logps/chosen": -1.187435507774353,
"logps/rejected": -1.5832734107971191,
"loss": 58.2104,
"nll_loss": 1.5183286666870117,
"rewards/accuracies": 0.706250011920929,
"rewards/chosen": -0.5937177538871765,
"rewards/margins": 0.19791889190673828,
"rewards/rejected": -0.7916367053985596,
"step": 120
},
{
"epoch": 1.1801481481481482,
"grad_norm": 131.9794158935547,
"learning_rate": 7.563815664117173e-06,
"log_odds_chosen": 0.4175822138786316,
"log_odds_ratio": -0.6133934259414673,
"logits/chosen": 232.56655883789062,
"logits/rejected": 236.88711547851562,
"logps/chosen": -1.2231204509735107,
"logps/rejected": -1.5356894731521606,
"loss": 55.8452,
"nll_loss": 1.4756473302841187,
"rewards/accuracies": 0.6875,
"rewards/chosen": -0.6115602254867554,
"rewards/margins": 0.1562844067811966,
"rewards/rejected": -0.7678447365760803,
"step": 125
},
{
"epoch": 1.2275555555555555,
"grad_norm": 128.4602813720703,
"learning_rate": 7.321725364338566e-06,
"log_odds_chosen": 0.3267204165458679,
"log_odds_ratio": -0.6442457437515259,
"logits/chosen": 226.52706909179688,
"logits/rejected": 236.42977905273438,
"logps/chosen": -1.1632269620895386,
"logps/rejected": -1.3935999870300293,
"loss": 55.0227,
"nll_loss": 1.4348909854888916,
"rewards/accuracies": 0.637499988079071,
"rewards/chosen": -0.5816134810447693,
"rewards/margins": 0.11518646776676178,
"rewards/rejected": -0.6967999935150146,
"step": 130
},
{
"epoch": 1.274962962962963,
"grad_norm": 111.24720764160156,
"learning_rate": 7.072484073626872e-06,
"log_odds_chosen": 0.39476487040519714,
"log_odds_ratio": -0.6248899698257446,
"logits/chosen": 231.5218505859375,
"logits/rejected": 237.8566131591797,
"logps/chosen": -1.1376434564590454,
"logps/rejected": -1.4273903369903564,
"loss": 55.291,
"nll_loss": 1.433115839958191,
"rewards/accuracies": 0.6499999761581421,
"rewards/chosen": -0.5688217282295227,
"rewards/margins": 0.14487341046333313,
"rewards/rejected": -0.7136951684951782,
"step": 135
},
{
"epoch": 1.3223703703703704,
"grad_norm": 123.10855102539062,
"learning_rate": 6.816859463403271e-06,
"log_odds_chosen": 0.4121381640434265,
"log_odds_ratio": -0.6339040994644165,
"logits/chosen": 231.813720703125,
"logits/rejected": 230.7804412841797,
"logps/chosen": -1.231404423713684,
"logps/rejected": -1.5583487749099731,
"loss": 57.6659,
"nll_loss": 1.4830906391143799,
"rewards/accuracies": 0.6312500238418579,
"rewards/chosen": -0.615702211856842,
"rewards/margins": 0.16347214579582214,
"rewards/rejected": -0.7791743874549866,
"step": 140
},
{
"epoch": 1.3697777777777778,
"grad_norm": 130.8789825439453,
"learning_rate": 6.555638865924221e-06,
"log_odds_chosen": 0.5062509775161743,
"log_odds_ratio": -0.5808830857276917,
"logits/chosen": 227.046875,
"logits/rejected": 243.47970581054688,
"logps/chosen": -1.1645643711090088,
"logps/rejected": -1.5737435817718506,
"loss": 55.8075,
"nll_loss": 1.456322193145752,
"rewards/accuracies": 0.668749988079071,
"rewards/chosen": -0.5822821855545044,
"rewards/margins": 0.20458965003490448,
"rewards/rejected": -0.7868717908859253,
"step": 145
},
{
"epoch": 1.417185185185185,
"grad_norm": 204.43161010742188,
"learning_rate": 6.289626849272062e-06,
"log_odds_chosen": 0.27410560846328735,
"log_odds_ratio": -0.6676009893417358,
"logits/chosen": 231.1629180908203,
"logits/rejected": 226.8252716064453,
"logps/chosen": -1.1737444400787354,
"logps/rejected": -1.372044324874878,
"loss": 55.1833,
"nll_loss": 1.4221489429473877,
"rewards/accuracies": 0.606249988079071,
"rewards/chosen": -0.5868722200393677,
"rewards/margins": 0.09915001690387726,
"rewards/rejected": -0.686022162437439,
"step": 150
},
{
"epoch": 1.4645925925925927,
"grad_norm": 107.84815979003906,
"learning_rate": 6.0196427392587085e-06,
"log_odds_chosen": 0.4244801104068756,
"log_odds_ratio": -0.5939399600028992,
"logits/chosen": 218.2533721923828,
"logits/rejected": 238.97201538085938,
"logps/chosen": -1.1000165939331055,
"logps/rejected": -1.4028135538101196,
"loss": 55.1904,
"nll_loss": 1.4290273189544678,
"rewards/accuracies": 0.6625000238418579,
"rewards/chosen": -0.5500082969665527,
"rewards/margins": 0.15139850974082947,
"rewards/rejected": -0.7014067769050598,
"step": 155
},
{
"epoch": 1.512,
"grad_norm": 124.19522094726562,
"learning_rate": 5.746518095875033e-06,
"log_odds_chosen": 0.42281055450439453,
"log_odds_ratio": -0.5916022062301636,
"logits/chosen": 226.8450469970703,
"logits/rejected": 230.33285522460938,
"logps/chosen": -1.174420952796936,
"logps/rejected": -1.5028213262557983,
"loss": 56.0624,
"nll_loss": 1.4398643970489502,
"rewards/accuracies": 0.65625,
"rewards/chosen": -0.587210476398468,
"rewards/margins": 0.16420014202594757,
"rewards/rejected": -0.7514106631278992,
"step": 160
},
{
"epoch": 1.5594074074074074,
"grad_norm": 125.1372299194336,
"learning_rate": 5.471094152058592e-06,
"log_odds_chosen": 0.4360503554344177,
"log_odds_ratio": -0.589511513710022,
"logits/chosen": 234.1468505859375,
"logits/rejected": 225.77023315429688,
"logps/chosen": -1.1270663738250732,
"logps/rejected": -1.4443070888519287,
"loss": 55.663,
"nll_loss": 1.4153871536254883,
"rewards/accuracies": 0.6875,
"rewards/chosen": -0.5635331869125366,
"rewards/margins": 0.15862040221691132,
"rewards/rejected": -0.7221535444259644,
"step": 165
},
{
"epoch": 1.6068148148148147,
"grad_norm": 170.6956787109375,
"learning_rate": 5.1942192226683385e-06,
"log_odds_chosen": 0.4941047728061676,
"log_odds_ratio": -0.5777939558029175,
"logits/chosen": 225.4412078857422,
"logits/rejected": 230.6029052734375,
"logps/chosen": -1.1379103660583496,
"logps/rejected": -1.5056231021881104,
"loss": 54.5688,
"nll_loss": 1.4074928760528564,
"rewards/accuracies": 0.6625000238418579,
"rewards/chosen": -0.5689551830291748,
"rewards/margins": 0.18385633826255798,
"rewards/rejected": -0.7528115510940552,
"step": 170
},
{
"epoch": 1.6542222222222223,
"grad_norm": 93.71338653564453,
"learning_rate": 4.916746091646808e-06,
"log_odds_chosen": 0.4696536958217621,
"log_odds_ratio": -0.5783597230911255,
"logits/chosen": 226.40939331054688,
"logits/rejected": 238.772705078125,
"logps/chosen": -1.1081968545913696,
"logps/rejected": -1.4637243747711182,
"loss": 54.5723,
"nll_loss": 1.4280409812927246,
"rewards/accuracies": 0.675000011920929,
"rewards/chosen": -0.5540984272956848,
"rewards/margins": 0.1777636557817459,
"rewards/rejected": -0.7318621873855591,
"step": 175
},
{
"epoch": 1.7016296296296296,
"grad_norm": 103.26416015625,
"learning_rate": 4.6395293854173395e-06,
"log_odds_chosen": 0.2048105001449585,
"log_odds_ratio": -0.6917891502380371,
"logits/chosen": 233.7334442138672,
"logits/rejected": 230.9724578857422,
"logps/chosen": -1.193915843963623,
"logps/rejected": -1.3438174724578857,
"loss": 55.4722,
"nll_loss": 1.475660800933838,
"rewards/accuracies": 0.5625,
"rewards/chosen": -0.5969579219818115,
"rewards/margins": 0.07495082169771194,
"rewards/rejected": -0.6719087362289429,
"step": 180
},
{
"epoch": 1.749037037037037,
"grad_norm": 87.70002746582031,
"learning_rate": 4.363422940606435e-06,
"log_odds_chosen": 0.4551773965358734,
"log_odds_ratio": -0.5870417356491089,
"logits/chosen": 229.46359252929688,
"logits/rejected": 232.06924438476562,
"logps/chosen": -1.1224743127822876,
"logps/rejected": -1.4738192558288574,
"loss": 54.9357,
"nll_loss": 1.4060946702957153,
"rewards/accuracies": 0.6625000238418579,
"rewards/chosen": -0.5612371563911438,
"rewards/margins": 0.17567242681980133,
"rewards/rejected": -0.7369096279144287,
"step": 185
},
{
"epoch": 1.7964444444444445,
"grad_norm": 95.35106658935547,
"learning_rate": 4.089277174198694e-06,
"log_odds_chosen": 0.45670628547668457,
"log_odds_ratio": -0.5882663726806641,
"logits/chosen": 221.1391143798828,
"logits/rejected": 236.39956665039062,
"logps/chosen": -1.1329078674316406,
"logps/rejected": -1.476986289024353,
"loss": 55.4382,
"nll_loss": 1.4155741930007935,
"rewards/accuracies": 0.7250000238418579,
"rewards/chosen": -0.5664539337158203,
"rewards/margins": 0.1720392405986786,
"rewards/rejected": -0.7384931445121765,
"step": 190
},
{
"epoch": 1.8438518518518519,
"grad_norm": 114.8470458984375,
"learning_rate": 3.817936464224367e-06,
"log_odds_chosen": 0.5381921529769897,
"log_odds_ratio": -0.5697619915008545,
"logits/chosen": 229.50509643554688,
"logits/rejected": 242.05123901367188,
"logps/chosen": -1.1255033016204834,
"logps/rejected": -1.5437520742416382,
"loss": 54.134,
"nll_loss": 1.4217894077301025,
"rewards/accuracies": 0.71875,
"rewards/chosen": -0.5627516508102417,
"rewards/margins": 0.209124356508255,
"rewards/rejected": -0.7718760371208191,
"step": 195
},
{
"epoch": 1.8912592592592592,
"grad_norm": 118.22327423095703,
"learning_rate": 3.55023654904709e-06,
"log_odds_chosen": 0.3448692858219147,
"log_odds_ratio": -0.6549310088157654,
"logits/chosen": 225.15869140625,
"logits/rejected": 235.4357147216797,
"logps/chosen": -1.1163337230682373,
"logps/rejected": -1.368422508239746,
"loss": 55.3987,
"nll_loss": 1.3597387075424194,
"rewards/accuracies": 0.6000000238418579,
"rewards/chosen": -0.5581668615341187,
"rewards/margins": 0.12604445219039917,
"rewards/rejected": -0.684211254119873,
"step": 200
},
{
"epoch": 1.9386666666666668,
"grad_norm": 90.22616577148438,
"learning_rate": 3.2870019532620744e-06,
"log_odds_chosen": 0.401312917470932,
"log_odds_ratio": -0.6004291772842407,
"logits/chosen": 224.0797119140625,
"logits/rejected": 241.08871459960938,
"logps/chosen": -1.0669735670089722,
"logps/rejected": -1.3696752786636353,
"loss": 53.038,
"nll_loss": 1.351078987121582,
"rewards/accuracies": 0.6187499761581421,
"rewards/chosen": -0.5334867835044861,
"rewards/margins": 0.15135079622268677,
"rewards/rejected": -0.6848376393318176,
"step": 205
},
{
"epoch": 1.986074074074074,
"grad_norm": 131.29962158203125,
"learning_rate": 3.0290434481330746e-06,
"log_odds_chosen": 0.3439275622367859,
"log_odds_ratio": -0.6156661510467529,
"logits/chosen": 233.643798828125,
"logits/rejected": 226.2475128173828,
"logps/chosen": -1.0474224090576172,
"logps/rejected": -1.2926290035247803,
"loss": 51.5863,
"nll_loss": 1.3270413875579834,
"rewards/accuracies": 0.606249988079071,
"rewards/chosen": -0.5237112045288086,
"rewards/margins": 0.12260331213474274,
"rewards/rejected": -0.6463145017623901,
"step": 210
},
{
"epoch": 2.0284444444444443,
"grad_norm": 96.40355682373047,
"learning_rate": 2.77715555439007e-06,
"log_odds_chosen": 0.5393110513687134,
"log_odds_ratio": -0.546492338180542,
"logits/chosen": 227.2359619140625,
"logits/rejected": 236.05026245117188,
"logps/chosen": -1.061828374862671,
"logps/rejected": -1.4429984092712402,
"loss": 46.8863,
"nll_loss": 1.3469436168670654,
"rewards/accuracies": 0.7342657446861267,
"rewards/chosen": -0.5309141874313354,
"rewards/margins": 0.19058500230312347,
"rewards/rejected": -0.7214992046356201,
"step": 215
},
{
"epoch": 2.075851851851852,
"grad_norm": 105.6058120727539,
"learning_rate": 2.532114095079137e-06,
"log_odds_chosen": 0.6614600419998169,
"log_odds_ratio": -0.5195887684822083,
"logits/chosen": 222.1413116455078,
"logits/rejected": 226.68173217773438,
"logps/chosen": -0.9896618723869324,
"logps/rejected": -1.4654362201690674,
"loss": 50.1408,
"nll_loss": 1.314392328262329,
"rewards/accuracies": 0.71875,
"rewards/chosen": -0.4948309361934662,
"rewards/margins": 0.23788709938526154,
"rewards/rejected": -0.7327181100845337,
"step": 220
},
{
"epoch": 2.1232592592592594,
"grad_norm": 86.1953353881836,
"learning_rate": 2.2946738060017947e-06,
"log_odds_chosen": 0.6218869686126709,
"log_odds_ratio": -0.5385203957557678,
"logits/chosen": 225.3985137939453,
"logits/rejected": 221.6757049560547,
"logps/chosen": -0.9887048602104187,
"logps/rejected": -1.440615177154541,
"loss": 50.805,
"nll_loss": 1.342362880706787,
"rewards/accuracies": 0.75,
"rewards/chosen": -0.49435243010520935,
"rewards/margins": 0.22595517337322235,
"rewards/rejected": -0.7203075885772705,
"step": 225
},
{
"epoch": 2.1706666666666665,
"grad_norm": 123.00335693359375,
"learning_rate": 2.0655660111037685e-06,
"log_odds_chosen": 0.7268826365470886,
"log_odds_ratio": -0.4982364773750305,
"logits/chosen": 222.36849975585938,
"logits/rejected": 233.1688995361328,
"logps/chosen": -1.0594282150268555,
"logps/rejected": -1.601413369178772,
"loss": 50.4757,
"nll_loss": 1.3385509252548218,
"rewards/accuracies": 0.7562500238418579,
"rewards/chosen": -0.5297141075134277,
"rewards/margins": 0.2709924578666687,
"rewards/rejected": -0.800706684589386,
"step": 230
},
{
"epoch": 2.218074074074074,
"grad_norm": 119.10016632080078,
"learning_rate": 1.8454963699730471e-06,
"log_odds_chosen": 0.6008008718490601,
"log_odds_ratio": -0.5274027585983276,
"logits/chosen": 214.6153106689453,
"logits/rejected": 219.9340057373047,
"logps/chosen": -1.0045088529586792,
"logps/rejected": -1.422736406326294,
"loss": 50.8648,
"nll_loss": 1.3140861988067627,
"rewards/accuracies": 0.75,
"rewards/chosen": -0.5022544264793396,
"rewards/margins": 0.20911379158496857,
"rewards/rejected": -0.711368203163147,
"step": 235
},
{
"epoch": 2.2654814814814817,
"grad_norm": 95.31659698486328,
"learning_rate": 1.6351427043849955e-06,
"log_odds_chosen": 0.5878480076789856,
"log_odds_ratio": -0.5382899045944214,
"logits/chosen": 223.2783203125,
"logits/rejected": 223.6912384033203,
"logps/chosen": -1.025895595550537,
"logps/rejected": -1.4464019536972046,
"loss": 50.6412,
"nll_loss": 1.3646161556243896,
"rewards/accuracies": 0.706250011920929,
"rewards/chosen": -0.5129477977752686,
"rewards/margins": 0.21025320887565613,
"rewards/rejected": -0.7232009768486023,
"step": 240
},
{
"epoch": 2.3128888888888888,
"grad_norm": 114.522705078125,
"learning_rate": 1.4351529105888735e-06,
"log_odds_chosen": 0.7168388366699219,
"log_odds_ratio": -0.4856956899166107,
"logits/chosen": 215.4445343017578,
"logits/rejected": 233.8488311767578,
"logps/chosen": -0.9942634701728821,
"logps/rejected": -1.5078818798065186,
"loss": 49.269,
"nll_loss": 1.3349363803863525,
"rewards/accuracies": 0.7875000238418579,
"rewards/chosen": -0.49713173508644104,
"rewards/margins": 0.25680914521217346,
"rewards/rejected": -0.7539409399032593,
"step": 245
},
{
"epoch": 2.3602962962962963,
"grad_norm": 162.42523193359375,
"learning_rate": 1.2461429637659466e-06,
"log_odds_chosen": 0.7001165151596069,
"log_odds_ratio": -0.48200806975364685,
"logits/chosen": 218.6329803466797,
"logits/rejected": 225.58749389648438,
"logps/chosen": -1.0117690563201904,
"logps/rejected": -1.5104820728302002,
"loss": 49.1429,
"nll_loss": 1.3088269233703613,
"rewards/accuracies": 0.7875000238418579,
"rewards/chosen": -0.5058845281600952,
"rewards/margins": 0.24935641884803772,
"rewards/rejected": -0.7552410364151001,
"step": 250
},
{
"epoch": 2.407703703703704,
"grad_norm": 174.11715698242188,
"learning_rate": 1.0686950208055486e-06,
"log_odds_chosen": 0.8261364698410034,
"log_odds_ratio": -0.47225421667099,
"logits/chosen": 215.776611328125,
"logits/rejected": 220.90621948242188,
"logps/chosen": -0.9497050046920776,
"logps/rejected": -1.5556175708770752,
"loss": 48.4801,
"nll_loss": 1.2853540182113647,
"rewards/accuracies": 0.78125,
"rewards/chosen": -0.4748525023460388,
"rewards/margins": 0.302956223487854,
"rewards/rejected": -0.7778087854385376,
"step": 255
},
{
"epoch": 2.455111111111111,
"grad_norm": 110.5379638671875,
"learning_rate": 9.033556272426075e-07,
"log_odds_chosen": 0.8216450810432434,
"log_odds_ratio": -0.4542720913887024,
"logits/chosen": 208.1492156982422,
"logits/rejected": 226.76187133789062,
"logps/chosen": -0.9726914167404175,
"logps/rejected": -1.557952642440796,
"loss": 49.1974,
"nll_loss": 1.2973281145095825,
"rewards/accuracies": 0.7875000238418579,
"rewards/chosen": -0.48634570837020874,
"rewards/margins": 0.292630672454834,
"rewards/rejected": -0.778976321220398,
"step": 260
},
{
"epoch": 2.5025185185185186,
"grad_norm": 110.2140884399414,
"learning_rate": 7.506340338793111e-07,
"log_odds_chosen": 0.660784125328064,
"log_odds_ratio": -0.5323556065559387,
"logits/chosen": 219.0674591064453,
"logits/rejected": 217.68307495117188,
"logps/chosen": -0.9799796938896179,
"logps/rejected": -1.453094720840454,
"loss": 48.5513,
"nll_loss": 1.2660871744155884,
"rewards/accuracies": 0.7437499761581421,
"rewards/chosen": -0.48998984694480896,
"rewards/margins": 0.23655752837657928,
"rewards/rejected": -0.726547360420227,
"step": 265
},
{
"epoch": 2.549925925925926,
"grad_norm": 125.52155303955078,
"learning_rate": 6.110006282757897e-07,
"log_odds_chosen": 0.838211178779602,
"log_odds_ratio": -0.4636700749397278,
"logits/chosen": 210.531982421875,
"logits/rejected": 217.48464965820312,
"logps/chosen": -0.9940751791000366,
"logps/rejected": -1.5986590385437012,
"loss": 48.4161,
"nll_loss": 1.2988741397857666,
"rewards/accuracies": 0.800000011920929,
"rewards/chosen": -0.4970375895500183,
"rewards/margins": 0.3022919297218323,
"rewards/rejected": -0.7993295192718506,
"step": 270
},
{
"epoch": 2.5973333333333333,
"grad_norm": 99.75615692138672,
"learning_rate": 4.848854859408731e-07,
"log_odds_chosen": 0.8387314677238464,
"log_odds_ratio": -0.45085686445236206,
"logits/chosen": 213.62991333007812,
"logits/rejected": 234.63095092773438,
"logps/chosen": -0.9795564413070679,
"logps/rejected": -1.5703213214874268,
"loss": 50.0056,
"nll_loss": 1.3050158023834229,
"rewards/accuracies": 0.8125,
"rewards/chosen": -0.48977822065353394,
"rewards/margins": 0.29538241028785706,
"rewards/rejected": -0.7851606607437134,
"step": 275
},
{
"epoch": 2.644740740740741,
"grad_norm": 120.89396667480469,
"learning_rate": 3.7267704568529015e-07,
"log_odds_chosen": 0.7018845081329346,
"log_odds_ratio": -0.5149911046028137,
"logits/chosen": 216.6708526611328,
"logits/rejected": 217.61160278320312,
"logps/chosen": -0.9656041860580444,
"logps/rejected": -1.4782164096832275,
"loss": 48.7583,
"nll_loss": 1.2335693836212158,
"rewards/accuracies": 0.706250011920929,
"rewards/chosen": -0.4828020930290222,
"rewards/margins": 0.25630611181259155,
"rewards/rejected": -0.7391082048416138,
"step": 280
},
{
"epoch": 2.6921481481481484,
"grad_norm": 121.43464660644531,
"learning_rate": 2.7472091321728067e-07,
"log_odds_chosen": 0.6797521710395813,
"log_odds_ratio": -0.5134819746017456,
"logits/chosen": 214.6907501220703,
"logits/rejected": 224.23226928710938,
"logps/chosen": -0.9947472810745239,
"logps/rejected": -1.4958022832870483,
"loss": 50.1254,
"nll_loss": 1.3161695003509521,
"rewards/accuracies": 0.75,
"rewards/chosen": -0.49737364053726196,
"rewards/margins": 0.25052759051322937,
"rewards/rejected": -0.7479011416435242,
"step": 285
},
{
"epoch": 2.7395555555555555,
"grad_norm": 128.531005859375,
"learning_rate": 1.9131879666558385e-07,
"log_odds_chosen": 0.7722429037094116,
"log_odds_ratio": -0.4660676121711731,
"logits/chosen": 220.88125610351562,
"logits/rejected": 225.862548828125,
"logps/chosen": -0.9653012156486511,
"logps/rejected": -1.5093042850494385,
"loss": 49.7205,
"nll_loss": 1.3300297260284424,
"rewards/accuracies": 0.768750011920929,
"rewards/chosen": -0.48265060782432556,
"rewards/margins": 0.27200156450271606,
"rewards/rejected": -0.7546521425247192,
"step": 290
},
{
"epoch": 2.786962962962963,
"grad_norm": 99.14717102050781,
"learning_rate": 1.2272757730841744e-07,
"log_odds_chosen": 0.7158989906311035,
"log_odds_ratio": -0.5291553735733032,
"logits/chosen": 215.935302734375,
"logits/rejected": 213.4801788330078,
"logps/chosen": -1.0447461605072021,
"logps/rejected": -1.5658835172653198,
"loss": 49.2845,
"nll_loss": 1.304596185684204,
"rewards/accuracies": 0.768750011920929,
"rewards/chosen": -0.5223730802536011,
"rewards/margins": 0.260568767786026,
"rewards/rejected": -0.7829417586326599,
"step": 295
},
{
"epoch": 2.83437037037037,
"grad_norm": 123.91796875,
"learning_rate": 6.91585183706428e-08,
"log_odds_chosen": 0.8705388903617859,
"log_odds_ratio": -0.4770691394805908,
"logits/chosen": 206.53121948242188,
"logits/rejected": 225.6171417236328,
"logps/chosen": -0.9915706515312195,
"logps/rejected": -1.6319612264633179,
"loss": 49.6516,
"nll_loss": 1.3192683458328247,
"rewards/accuracies": 0.78125,
"rewards/chosen": -0.49578532576560974,
"rewards/margins": 0.32019534707069397,
"rewards/rejected": -0.8159806132316589,
"step": 300
},
{
"epoch": 2.8817777777777778,
"grad_norm": 100.9173812866211,
"learning_rate": 3.077661432604184e-08,
"log_odds_chosen": 0.7253429293632507,
"log_odds_ratio": -0.49313846230506897,
"logits/chosen": 211.61221313476562,
"logits/rejected": 224.7270050048828,
"logps/chosen": -1.0019609928131104,
"logps/rejected": -1.5113885402679443,
"loss": 48.0438,
"nll_loss": 1.2643382549285889,
"rewards/accuracies": 0.7749999761581421,
"rewards/chosen": -0.5009804964065552,
"rewards/margins": 0.254713773727417,
"rewards/rejected": -0.7556942701339722,
"step": 305
},
{
"epoch": 2.9291851851851853,
"grad_norm": 134.68287658691406,
"learning_rate": 7.700082708883006e-09,
"log_odds_chosen": 0.7188720703125,
"log_odds_ratio": -0.4834628999233246,
"logits/chosen": 216.98306274414062,
"logits/rejected": 232.9964599609375,
"logps/chosen": -0.9981603622436523,
"logps/rejected": -1.4897390604019165,
"loss": 49.1365,
"nll_loss": 1.3062077760696411,
"rewards/accuracies": 0.7875000238418579,
"rewards/chosen": -0.49908018112182617,
"rewards/margins": 0.24578937888145447,
"rewards/rejected": -0.7448695302009583,
"step": 310
},
{
"epoch": 2.9765925925925925,
"grad_norm": 172.74346923828125,
"learning_rate": 0.0,
"log_odds_chosen": 0.808211624622345,
"log_odds_ratio": -0.4583892822265625,
"logits/chosen": 211.4867706298828,
"logits/rejected": 224.31991577148438,
"logps/chosen": -0.9515358209609985,
"logps/rejected": -1.5278295278549194,
"loss": 47.6357,
"nll_loss": 1.26894211769104,
"rewards/accuracies": 0.800000011920929,
"rewards/chosen": -0.47576791048049927,
"rewards/margins": 0.2881469428539276,
"rewards/rejected": -0.7639147639274597,
"step": 315
},
{
"epoch": 2.9765925925925925,
"step": 315,
"total_flos": 0.0,
"train_loss": 95.27591378348214,
"train_runtime": 8112.4006,
"train_samples_per_second": 2.496,
"train_steps_per_second": 0.039
}
],
"logging_steps": 5,
"max_steps": 315,
"num_input_tokens_seen": 0,
"num_train_epochs": 3,
"save_steps": 100000,
"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": 1,
"trial_name": null,
"trial_params": null
}