silviasapora's picture
Model save
fd3ce7b 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": 1149.4111328125,
"learning_rate": 7.8125e-06,
"log_odds_chosen": -1.058286428451538,
"log_odds_ratio": -11.003866195678711,
"logps/chosen": -22.701828002929688,
"logps/rejected": -21.64377212524414,
"loss": 487.7823,
"nll_loss": 9.837206840515137,
"po_loss": 5.405989170074463,
"rewards/accuracies": 0.48124998807907104,
"rewards/chosen": -11.350914001464844,
"rewards/margins": -0.5290289521217346,
"rewards/rejected": -10.82188606262207,
"step": 5
},
{
"epoch": 0.09481481481481481,
"grad_norm": 597.1365966796875,
"learning_rate": 1.5625e-05,
"log_odds_chosen": -3.132328510284424,
"log_odds_ratio": -11.042044639587402,
"logps/chosen": -21.983291625976562,
"logps/rejected": -18.8513126373291,
"loss": 453.2121,
"nll_loss": 8.745321273803711,
"po_loss": 5.417555809020996,
"rewards/accuracies": 0.4437499940395355,
"rewards/chosen": -10.991645812988281,
"rewards/margins": -1.5659898519515991,
"rewards/rejected": -9.42565631866455,
"step": 10
},
{
"epoch": 0.14222222222222222,
"grad_norm": 527.598388671875,
"learning_rate": 2.34375e-05,
"log_odds_chosen": -1.3104137182235718,
"log_odds_ratio": -11.265727996826172,
"logps/chosen": -21.87612533569336,
"logps/rejected": -20.564207077026367,
"loss": 432.0127,
"nll_loss": 7.955958366394043,
"po_loss": 5.544436931610107,
"rewards/accuracies": 0.515625,
"rewards/chosen": -10.93806266784668,
"rewards/margins": -0.6559588313102722,
"rewards/rejected": -10.282103538513184,
"step": 15
},
{
"epoch": 0.18962962962962962,
"grad_norm": 2963.93408203125,
"learning_rate": 3.125e-05,
"log_odds_chosen": -3.6984353065490723,
"log_odds_ratio": -8.447766304016113,
"logps/chosen": -15.567995071411133,
"logps/rejected": -11.872796058654785,
"loss": 317.4233,
"nll_loss": 5.814067840576172,
"po_loss": 4.105409622192383,
"rewards/accuracies": 0.4312500059604645,
"rewards/chosen": -7.783997535705566,
"rewards/margins": -1.8475996255874634,
"rewards/rejected": -5.936398029327393,
"step": 20
},
{
"epoch": 0.23703703703703705,
"grad_norm": 273.48370361328125,
"learning_rate": 3.90625e-05,
"log_odds_chosen": -0.019201982766389847,
"log_odds_ratio": -1.406181812286377,
"logps/chosen": -3.055501937866211,
"logps/rejected": -3.0181398391723633,
"loss": 95.9863,
"nll_loss": 2.533766269683838,
"po_loss": 0.46580663323402405,
"rewards/accuracies": 0.5625,
"rewards/chosen": -1.5277509689331055,
"rewards/margins": -0.018681075423955917,
"rewards/rejected": -1.5090699195861816,
"step": 25
},
{
"epoch": 0.28444444444444444,
"grad_norm": 216.39308166503906,
"learning_rate": 4.6875e-05,
"log_odds_chosen": 0.09772934764623642,
"log_odds_ratio": -0.8396835327148438,
"logps/chosen": -1.784551978111267,
"logps/rejected": -1.8702020645141602,
"loss": 68.8806,
"nll_loss": 2.0184409618377686,
"po_loss": 0.13407745957374573,
"rewards/accuracies": 0.543749988079071,
"rewards/chosen": -0.8922759890556335,
"rewards/margins": 0.0428251177072525,
"rewards/rejected": -0.9351010322570801,
"step": 30
},
{
"epoch": 0.33185185185185184,
"grad_norm": 280.289794921875,
"learning_rate": 4.998613757348784e-05,
"log_odds_chosen": 0.25464916229248047,
"log_odds_ratio": -0.7872657775878906,
"logps/chosen": -1.6759769916534424,
"logps/rejected": -1.8980681896209717,
"loss": 62.3358,
"nll_loss": 1.86691415309906,
"po_loss": 0.0810796245932579,
"rewards/accuracies": 0.5375000238418579,
"rewards/chosen": -0.8379884958267212,
"rewards/margins": 0.11104562133550644,
"rewards/rejected": -0.9490340948104858,
"step": 35
},
{
"epoch": 0.37925925925925924,
"grad_norm": 708.8033447265625,
"learning_rate": 4.990147841143462e-05,
"log_odds_chosen": 0.31906959414482117,
"log_odds_ratio": -0.6981687545776367,
"logps/chosen": -1.4914863109588623,
"logps/rejected": -1.7542707920074463,
"loss": 57.8745,
"nll_loss": 1.7927372455596924,
"po_loss": 0.015842173248529434,
"rewards/accuracies": 0.6187499761581421,
"rewards/chosen": -0.7457431554794312,
"rewards/margins": 0.13139230012893677,
"rewards/rejected": -0.8771353960037231,
"step": 40
},
{
"epoch": 0.4266666666666667,
"grad_norm": 216.06666564941406,
"learning_rate": 4.97401218720448e-05,
"log_odds_chosen": 0.29898008704185486,
"log_odds_ratio": -0.6774553060531616,
"logps/chosen": -1.4114806652069092,
"logps/rejected": -1.6652767658233643,
"loss": 55.5358,
"nll_loss": 1.7187302112579346,
"po_loss": 0.01676475629210472,
"rewards/accuracies": 0.6156250238418579,
"rewards/chosen": -0.7057403326034546,
"rewards/margins": 0.12689796090126038,
"rewards/rejected": -0.8326383829116821,
"step": 45
},
{
"epoch": 0.4740740740740741,
"grad_norm": 134.64588928222656,
"learning_rate": 4.9502564938797946e-05,
"log_odds_chosen": 0.3125649392604828,
"log_odds_ratio": -0.6582598686218262,
"logps/chosen": -1.3025346994400024,
"logps/rejected": -1.5579352378845215,
"loss": 50.6171,
"nll_loss": 1.5993175506591797,
"po_loss": -0.01753159798681736,
"rewards/accuracies": 0.6156250238418579,
"rewards/chosen": -0.6512673497200012,
"rewards/margins": 0.1277003139257431,
"rewards/rejected": -0.7789676189422607,
"step": 50
},
{
"epoch": 0.5214814814814814,
"grad_norm": 125.86614227294922,
"learning_rate": 4.918953929490768e-05,
"log_odds_chosen": 0.32853975892066956,
"log_odds_ratio": -0.6720215082168579,
"logps/chosen": -1.3117344379425049,
"logps/rejected": -1.5763285160064697,
"loss": 51.7501,
"nll_loss": 1.6202409267425537,
"po_loss": -0.0030506025068461895,
"rewards/accuracies": 0.6031249761581421,
"rewards/chosen": -0.6558672189712524,
"rewards/margins": 0.13229700922966003,
"rewards/rejected": -0.7881642580032349,
"step": 55
},
{
"epoch": 0.5688888888888889,
"grad_norm": 223.18544006347656,
"learning_rate": 4.88020090697132e-05,
"log_odds_chosen": 0.35989266633987427,
"log_odds_ratio": -0.6441806554794312,
"logps/chosen": -1.3117573261260986,
"logps/rejected": -1.609619379043579,
"loss": 51.8589,
"nll_loss": 1.6353687047958374,
"po_loss": -0.014778362587094307,
"rewards/accuracies": 0.6031249761581421,
"rewards/chosen": -0.6558786630630493,
"rewards/margins": 0.1489311009645462,
"rewards/rejected": -0.8048096895217896,
"step": 60
},
{
"epoch": 0.6162962962962963,
"grad_norm": 195.27818298339844,
"learning_rate": 4.834116786912897e-05,
"log_odds_chosen": 0.36024975776672363,
"log_odds_ratio": -0.6309642195701599,
"logps/chosen": -1.305149793624878,
"logps/rejected": -1.584517478942871,
"loss": 50.395,
"nll_loss": 1.5889090299606323,
"po_loss": -0.014065178111195564,
"rewards/accuracies": 0.643750011920929,
"rewards/chosen": -0.652574896812439,
"rewards/margins": 0.13968387246131897,
"rewards/rejected": -0.7922587394714355,
"step": 65
},
{
"epoch": 0.6637037037037037,
"grad_norm": 90.99329376220703,
"learning_rate": 4.7808435099299045e-05,
"log_odds_chosen": 0.49140042066574097,
"log_odds_ratio": -0.5946860313415527,
"logps/chosen": -1.2680964469909668,
"logps/rejected": -1.666022539138794,
"loss": 48.3341,
"nll_loss": 1.567561388015747,
"po_loss": -0.05712021514773369,
"rewards/accuracies": 0.668749988079071,
"rewards/chosen": -0.6340482234954834,
"rewards/margins": 0.19896303117275238,
"rewards/rejected": -0.833011269569397,
"step": 70
},
{
"epoch": 0.7111111111111111,
"grad_norm": 311.4484558105469,
"learning_rate": 4.720545159477922e-05,
"log_odds_chosen": 0.4681640565395355,
"log_odds_ratio": -0.6271704435348511,
"logps/chosen": -1.2391738891601562,
"logps/rejected": -1.6162137985229492,
"loss": 48.3516,
"nll_loss": 1.547349214553833,
"po_loss": -0.03636247664690018,
"rewards/accuracies": 0.6781250238418579,
"rewards/chosen": -0.6195869445800781,
"rewards/margins": 0.1885199397802353,
"rewards/rejected": -0.8081068992614746,
"step": 75
},
{
"epoch": 0.7585185185185185,
"grad_norm": 290.2730407714844,
"learning_rate": 4.653407456471222e-05,
"log_odds_chosen": 0.37080585956573486,
"log_odds_ratio": -0.645098090171814,
"logps/chosen": -1.2255287170410156,
"logps/rejected": -1.5139461755752563,
"loss": 46.785,
"nll_loss": 1.5034220218658447,
"po_loss": -0.04139198362827301,
"rewards/accuracies": 0.6031249761581421,
"rewards/chosen": -0.6127643585205078,
"rewards/margins": 0.14420875906944275,
"rewards/rejected": -0.7569730877876282,
"step": 80
},
{
"epoch": 0.8059259259259259,
"grad_norm": 123.97457885742188,
"learning_rate": 4.579637187256222e-05,
"log_odds_chosen": 0.3362444341182709,
"log_odds_ratio": -0.6288684010505676,
"logps/chosen": -1.1935008764266968,
"logps/rejected": -1.456559419631958,
"loss": 47.5166,
"nll_loss": 1.5174460411071777,
"po_loss": -0.03255213052034378,
"rewards/accuracies": 0.628125011920929,
"rewards/chosen": -0.5967504382133484,
"rewards/margins": 0.1315293163061142,
"rewards/rejected": -0.728279709815979,
"step": 85
},
{
"epoch": 0.8533333333333334,
"grad_norm": 215.05145263671875,
"learning_rate": 4.499461566702685e-05,
"log_odds_chosen": 0.5429065823554993,
"log_odds_ratio": -0.5807454586029053,
"logps/chosen": -1.1128449440002441,
"logps/rejected": -1.5253103971481323,
"loss": 44.2251,
"nll_loss": 1.449202537536621,
"po_loss": -0.06716945022344589,
"rewards/accuracies": 0.6781250238418579,
"rewards/chosen": -0.5564224720001221,
"rewards/margins": 0.20623275637626648,
"rewards/rejected": -0.7626551985740662,
"step": 90
},
{
"epoch": 0.9007407407407407,
"grad_norm": 107.29192352294922,
"learning_rate": 4.413127538374411e-05,
"log_odds_chosen": 0.4139605462551117,
"log_odds_ratio": -0.6369166374206543,
"logps/chosen": -1.1876246929168701,
"logps/rejected": -1.5150048732757568,
"loss": 46.2734,
"nll_loss": 1.4956274032592773,
"po_loss": -0.049582015722990036,
"rewards/accuracies": 0.606249988079071,
"rewards/chosen": -0.5938123464584351,
"rewards/margins": 0.16369004547595978,
"rewards/rejected": -0.7575024366378784,
"step": 95
},
{
"epoch": 0.9481481481481482,
"grad_norm": 88.76741027832031,
"learning_rate": 4.320901013934887e-05,
"log_odds_chosen": 0.49218830466270447,
"log_odds_ratio": -0.5807846188545227,
"logps/chosen": -1.1516714096069336,
"logps/rejected": -1.5277314186096191,
"loss": 44.1828,
"nll_loss": 1.4526407718658447,
"po_loss": -0.07192973792552948,
"rewards/accuracies": 0.6781250238418579,
"rewards/chosen": -0.5758357048034668,
"rewards/margins": 0.18803004920482635,
"rewards/rejected": -0.7638657093048096,
"step": 100
},
{
"epoch": 0.9955555555555555,
"grad_norm": 205.42556762695312,
"learning_rate": 4.223066054130568e-05,
"log_odds_chosen": 0.6228845119476318,
"log_odds_ratio": -0.5975527763366699,
"logps/chosen": -1.139916181564331,
"logps/rejected": -1.6632674932479858,
"loss": 44.3594,
"nll_loss": 1.4790844917297363,
"po_loss": -0.09285450726747513,
"rewards/accuracies": 0.6499999761581421,
"rewards/chosen": -0.5699580907821655,
"rewards/margins": 0.26167553663253784,
"rewards/rejected": -0.8316337466239929,
"step": 105
},
{
"epoch": 1.037925925925926,
"grad_norm": 264.0072326660156,
"learning_rate": 4.1199239938743797e-05,
"log_odds_chosen": 0.9932449460029602,
"log_odds_ratio": -0.44668498635292053,
"logps/chosen": -1.02596116065979,
"logps/rejected": -1.7672877311706543,
"loss": 30.7868,
"nll_loss": 1.3024835586547852,
"po_loss": -0.2260211557149887,
"rewards/accuracies": 0.7902097702026367,
"rewards/chosen": -0.512980580329895,
"rewards/margins": 0.3706631660461426,
"rewards/rejected": -0.8836438655853271,
"step": 110
},
{
"epoch": 1.0853333333333333,
"grad_norm": 156.9835968017578,
"learning_rate": 4.0117925141242174e-05,
"log_odds_chosen": 1.619765043258667,
"log_odds_ratio": -0.3932790756225586,
"logps/chosen": -0.9797142148017883,
"logps/rejected": -2.256946563720703,
"loss": 29.6065,
"nll_loss": 1.299898386001587,
"po_loss": -0.3746953308582306,
"rewards/accuracies": 0.831250011920929,
"rewards/chosen": -0.48985710740089417,
"rewards/margins": 0.6386162042617798,
"rewards/rejected": -1.1284732818603516,
"step": 115
},
{
"epoch": 1.1327407407407408,
"grad_norm": 388.80926513671875,
"learning_rate": 3.899004663415084e-05,
"log_odds_chosen": 1.9765598773956299,
"log_odds_ratio": -0.3769186735153198,
"logps/chosen": -1.0536551475524902,
"logps/rejected": -2.6980745792388916,
"loss": 28.527,
"nll_loss": 1.3548568487167358,
"po_loss": -0.4633867144584656,
"rewards/accuracies": 0.824999988079071,
"rewards/chosen": -0.5268275737762451,
"rewards/margins": 0.8222096562385559,
"rewards/rejected": -1.3490372896194458,
"step": 120
},
{
"epoch": 1.1801481481481482,
"grad_norm": 701.78466796875,
"learning_rate": 3.781907832058587e-05,
"log_odds_chosen": 2.3174455165863037,
"log_odds_ratio": -0.4550691545009613,
"logps/chosen": -1.3317890167236328,
"logps/rejected": -3.3481040000915527,
"loss": 29.2023,
"nll_loss": 1.430572748184204,
"po_loss": -0.5180009007453918,
"rewards/accuracies": 0.84375,
"rewards/chosen": -0.6658945083618164,
"rewards/margins": 1.00815749168396,
"rewards/rejected": -1.6740520000457764,
"step": 125
},
{
"epoch": 1.2275555555555555,
"grad_norm": 284.3743896484375,
"learning_rate": 3.660862682169282e-05,
"log_odds_chosen": 1.8650093078613281,
"log_odds_ratio": -0.3967917561531067,
"logps/chosen": -1.0224735736846924,
"logps/rejected": -2.5611350536346436,
"loss": 31.3418,
"nll_loss": 1.3258826732635498,
"po_loss": -0.34645071625709534,
"rewards/accuracies": 0.828125,
"rewards/chosen": -0.5112367868423462,
"rewards/margins": 0.7693307995796204,
"rewards/rejected": -1.2805675268173218,
"step": 130
},
{
"epoch": 1.274962962962963,
"grad_norm": 514.7438354492188,
"learning_rate": 3.5362420368134356e-05,
"log_odds_chosen": 3.2559561729431152,
"log_odds_ratio": -0.4518875479698181,
"logps/chosen": -1.2612740993499756,
"logps/rejected": -4.1930646896362305,
"loss": 29.2929,
"nll_loss": 1.4387086629867554,
"po_loss": -0.5233055949211121,
"rewards/accuracies": 0.8343750238418579,
"rewards/chosen": -0.6306370496749878,
"rewards/margins": 1.4658952951431274,
"rewards/rejected": -2.0965323448181152,
"step": 135
},
{
"epoch": 1.3223703703703704,
"grad_norm": 581.5865478515625,
"learning_rate": 3.408429731701635e-05,
"log_odds_chosen": 2.8635735511779785,
"log_odds_ratio": -0.4325293004512787,
"logps/chosen": -1.1112406253814697,
"logps/rejected": -3.690396785736084,
"loss": 27.7038,
"nll_loss": 1.3551785945892334,
"po_loss": -0.48943454027175903,
"rewards/accuracies": 0.828125,
"rewards/chosen": -0.5556203126907349,
"rewards/margins": 1.2895781993865967,
"rewards/rejected": -1.845198392868042,
"step": 140
},
{
"epoch": 1.3697777777777778,
"grad_norm": 112.59197998046875,
"learning_rate": 3.2778194329621104e-05,
"log_odds_chosen": 2.2916512489318848,
"log_odds_ratio": -0.41769519448280334,
"logps/chosen": -1.020237684249878,
"logps/rejected": -3.0198616981506348,
"loss": 27.0053,
"nll_loss": 1.3402351140975952,
"po_loss": -0.4963192939758301,
"rewards/accuracies": 0.8218749761581421,
"rewards/chosen": -0.510118842124939,
"rewards/margins": 0.9998119473457336,
"rewards/rejected": -1.5099308490753174,
"step": 145
},
{
"epoch": 1.417185185185185,
"grad_norm": 805.1986694335938,
"learning_rate": 3.144813424636031e-05,
"log_odds_chosen": 6.398291110992432,
"log_odds_ratio": -0.40958553552627563,
"logps/chosen": -1.284738302230835,
"logps/rejected": -7.353111267089844,
"loss": 18.0904,
"nll_loss": 1.4131615161895752,
"po_loss": -0.8478377461433411,
"rewards/accuracies": 0.859375,
"rewards/chosen": -0.6423691511154175,
"rewards/margins": 3.034186601638794,
"rewards/rejected": -3.676555633544922,
"step": 150
},
{
"epoch": 1.4645925925925927,
"grad_norm": 219.5271759033203,
"learning_rate": 3.0098213696293542e-05,
"log_odds_chosen": 6.040364742279053,
"log_odds_ratio": -0.49418535828590393,
"logps/chosen": -1.2788912057876587,
"logps/rejected": -7.029244899749756,
"loss": 29.1762,
"nll_loss": 1.5258408784866333,
"po_loss": -0.6140841245651245,
"rewards/accuracies": 0.809374988079071,
"rewards/chosen": -0.6394456028938293,
"rewards/margins": 2.8751773834228516,
"rewards/rejected": -3.514622449874878,
"step": 155
},
{
"epoch": 1.512,
"grad_norm": 485.47576904296875,
"learning_rate": 2.8732590479375165e-05,
"log_odds_chosen": 3.2111659049987793,
"log_odds_ratio": -0.4072226583957672,
"logps/chosen": -0.9738284945487976,
"logps/rejected": -3.8901615142822266,
"loss": 27.3374,
"nll_loss": 1.3175849914550781,
"po_loss": -0.4632897973060608,
"rewards/accuracies": 0.8218749761581421,
"rewards/chosen": -0.4869142472743988,
"rewards/margins": 1.458166480064392,
"rewards/rejected": -1.9450807571411133,
"step": 160
},
{
"epoch": 1.5594074074074074,
"grad_norm": 835.3743286132812,
"learning_rate": 2.7355470760292956e-05,
"log_odds_chosen": 3.777056932449341,
"log_odds_ratio": -0.535234808921814,
"logps/chosen": -1.136812448501587,
"logps/rejected": -4.577279090881348,
"loss": 28.2855,
"nll_loss": 1.4577716588974,
"po_loss": -0.5738505125045776,
"rewards/accuracies": 0.840624988079071,
"rewards/chosen": -0.5684062242507935,
"rewards/margins": 1.7202333211898804,
"rewards/rejected": -2.288639545440674,
"step": 165
},
{
"epoch": 1.6068148148148147,
"grad_norm": 275.0011291503906,
"learning_rate": 2.597109611334169e-05,
"log_odds_chosen": 3.221364974975586,
"log_odds_ratio": -0.6203993558883667,
"logps/chosen": -1.2395914793014526,
"logps/rejected": -4.143717288970947,
"loss": 36.586,
"nll_loss": 1.4404484033584595,
"po_loss": -0.2971368730068207,
"rewards/accuracies": 0.8343750238418579,
"rewards/chosen": -0.6197957396507263,
"rewards/margins": 1.4520628452301025,
"rewards/rejected": -2.0718586444854736,
"step": 170
},
{
"epoch": 1.6542222222222223,
"grad_norm": 454.6676940917969,
"learning_rate": 2.458373045823404e-05,
"log_odds_chosen": 3.51635479927063,
"log_odds_ratio": -0.4349392354488373,
"logps/chosen": -1.2439143657684326,
"logps/rejected": -4.460102081298828,
"loss": 20.5614,
"nll_loss": 1.4349825382232666,
"po_loss": -0.7924402356147766,
"rewards/accuracies": 0.8218749761581421,
"rewards/chosen": -0.6219571828842163,
"rewards/margins": 1.6080938577651978,
"rewards/rejected": -2.230051040649414,
"step": 175
},
{
"epoch": 1.7016296296296296,
"grad_norm": 1382.2955322265625,
"learning_rate": 2.3197646927086697e-05,
"log_odds_chosen": 2.992593288421631,
"log_odds_ratio": -0.4546445906162262,
"logps/chosen": -1.1713831424713135,
"logps/rejected": -3.8302390575408936,
"loss": 25.5293,
"nll_loss": 1.433226227760315,
"po_loss": -0.6354348659515381,
"rewards/accuracies": 0.8218749761581421,
"rewards/chosen": -0.5856915712356567,
"rewards/margins": 1.32942795753479,
"rewards/rejected": -1.9151195287704468,
"step": 180
},
{
"epoch": 1.749037037037037,
"grad_norm": 828.3840942382812,
"learning_rate": 2.1817114703032176e-05,
"log_odds_chosen": 2.797692060470581,
"log_odds_ratio": -0.43407559394836426,
"logps/chosen": -1.1162699460983276,
"logps/rejected": -3.64747953414917,
"loss": 25.3097,
"nll_loss": 1.3284680843353271,
"po_loss": -0.5375403165817261,
"rewards/accuracies": 0.793749988079071,
"rewards/chosen": -0.5581349730491638,
"rewards/margins": 1.265605092048645,
"rewards/rejected": -1.823739767074585,
"step": 185
},
{
"epoch": 1.7964444444444445,
"grad_norm": 1088.3760986328125,
"learning_rate": 2.0446385870993467e-05,
"log_odds_chosen": 3.665976047515869,
"log_odds_ratio": -0.43087244033813477,
"logps/chosen": -0.9914714097976685,
"logps/rejected": -4.345323085784912,
"loss": 20.7913,
"nll_loss": 1.3063815832138062,
"po_loss": -0.656653106212616,
"rewards/accuracies": 0.8187500238418579,
"rewards/chosen": -0.49573570489883423,
"rewards/margins": 1.6769258975982666,
"rewards/rejected": -2.172661542892456,
"step": 190
},
{
"epoch": 1.8438518518518519,
"grad_norm": 1150.76123046875,
"learning_rate": 1.9089682321121834e-05,
"log_odds_chosen": 6.179993629455566,
"log_odds_ratio": -1.0467485189437866,
"logps/chosen": -1.6919925212860107,
"logps/rejected": -7.563102722167969,
"loss": 44.1491,
"nll_loss": 1.8505232334136963,
"po_loss": -0.4708642065525055,
"rewards/accuracies": 0.84375,
"rewards/chosen": -0.8459962606430054,
"rewards/margins": 2.9355549812316895,
"rewards/rejected": -3.7815513610839844,
"step": 195
},
{
"epoch": 1.8912592592592592,
"grad_norm": 454.659912109375,
"learning_rate": 1.775118274523545e-05,
"log_odds_chosen": 4.30472469329834,
"log_odds_ratio": -0.5107940435409546,
"logps/chosen": -1.1776766777038574,
"logps/rejected": -5.171801567077637,
"loss": 24.2982,
"nll_loss": 1.444457769393921,
"po_loss": -0.685138463973999,
"rewards/accuracies": 0.846875011920929,
"rewards/chosen": -0.5888383388519287,
"rewards/margins": 1.9970623254776,
"rewards/rejected": -2.5859007835388184,
"step": 200
},
{
"epoch": 1.9386666666666668,
"grad_norm": 800.92578125,
"learning_rate": 1.643500976631037e-05,
"log_odds_chosen": 2.9392075538635254,
"log_odds_ratio": -0.5851477384567261,
"logps/chosen": -1.1809542179107666,
"logps/rejected": -3.785405397415161,
"loss": 31.8492,
"nll_loss": 1.4079935550689697,
"po_loss": -0.412706196308136,
"rewards/accuracies": 0.840624988079071,
"rewards/chosen": -0.5904771089553833,
"rewards/margins": 1.3022253513336182,
"rewards/rejected": -1.8927026987075806,
"step": 205
},
{
"epoch": 1.986074074074074,
"grad_norm": 717.8953247070312,
"learning_rate": 1.514521724066537e-05,
"log_odds_chosen": 3.531224012374878,
"log_odds_ratio": -0.637366771697998,
"logps/chosen": -1.275508165359497,
"logps/rejected": -4.51492977142334,
"loss": 26.7864,
"nll_loss": 1.5013716220855713,
"po_loss": -0.6642956733703613,
"rewards/accuracies": 0.78125,
"rewards/chosen": -0.6377540826797485,
"rewards/margins": 1.6197106838226318,
"rewards/rejected": -2.25746488571167,
"step": 210
},
{
"epoch": 2.0284444444444443,
"grad_norm": 1063.2196044921875,
"learning_rate": 1.3885777771950348e-05,
"log_odds_chosen": 7.453225612640381,
"log_odds_ratio": -0.5666939616203308,
"logps/chosen": -1.2951644659042358,
"logps/rejected": -8.28040885925293,
"loss": 0.3911,
"nll_loss": 1.4602717161178589,
"po_loss": -1.4465978145599365,
"rewards/accuracies": 0.8741258978843689,
"rewards/chosen": -0.6475822329521179,
"rewards/margins": 3.492622137069702,
"rewards/rejected": -4.140204429626465,
"step": 215
},
{
"epoch": 2.075851851851852,
"grad_norm": 1035.703369140625,
"learning_rate": 1.2660570475395683e-05,
"log_odds_chosen": 8.555649757385254,
"log_odds_ratio": -0.9756455421447754,
"logps/chosen": -2.252915859222412,
"logps/rejected": -10.40683364868164,
"loss": 26.9416,
"nll_loss": 2.1952455043792725,
"po_loss": -1.353319764137268,
"rewards/accuracies": 0.8687499761581421,
"rewards/chosen": -1.126457929611206,
"rewards/margins": 4.076959133148193,
"rewards/rejected": -5.20341682434082,
"step": 220
},
{
"epoch": 2.1232592592592594,
"grad_norm": 983.8046875,
"learning_rate": 1.1473369030008974e-05,
"log_odds_chosen": 6.684991359710693,
"log_odds_ratio": -0.562995433807373,
"logps/chosen": -1.3902976512908936,
"logps/rejected": -7.629068851470947,
"loss": -8.0422,
"nll_loss": 1.5127547979354858,
"po_loss": -1.7640743255615234,
"rewards/accuracies": 0.8968750238418579,
"rewards/chosen": -0.6951488256454468,
"rewards/margins": 3.119385242462158,
"rewards/rejected": -3.8145344257354736,
"step": 225
},
{
"epoch": 2.1706666666666665,
"grad_norm": 1034.22802734375,
"learning_rate": 1.0327830055518842e-05,
"log_odds_chosen": 6.934300899505615,
"log_odds_ratio": -0.2921414077281952,
"logps/chosen": -0.9763369560241699,
"logps/rejected": -7.44844913482666,
"loss": -18.2469,
"nll_loss": 1.264357328414917,
"po_loss": -1.8345714807510376,
"rewards/accuracies": 0.859375,
"rewards/chosen": -0.48816847801208496,
"rewards/margins": 3.236056089401245,
"rewards/rejected": -3.72422456741333,
"step": 230
},
{
"epoch": 2.218074074074074,
"grad_norm": 3417.23193359375,
"learning_rate": 9.227481849865235e-06,
"log_odds_chosen": 10.725831031799316,
"log_odds_ratio": -0.9306958317756653,
"logps/chosen": -1.6473735570907593,
"logps/rejected": -11.90982437133789,
"loss": -0.3702,
"nll_loss": 1.8238227367401123,
"po_loss": -1.8353925943374634,
"rewards/accuracies": 0.871874988079071,
"rewards/chosen": -0.8236867785453796,
"rewards/margins": 5.131225109100342,
"rewards/rejected": -5.954912185668945,
"step": 235
},
{
"epoch": 2.2654814814814817,
"grad_norm": 6067.5625,
"learning_rate": 8.175713521924978e-06,
"log_odds_chosen": 13.598230361938477,
"log_odds_ratio": -0.6408636569976807,
"logps/chosen": -1.4218194484710693,
"logps/rejected": -14.529757499694824,
"loss": -35.3649,
"nll_loss": 1.5829206705093384,
"po_loss": -2.688075304031372,
"rewards/accuracies": 0.909375011920929,
"rewards/chosen": -0.7109097242355347,
"rewards/margins": 6.553968906402588,
"rewards/rejected": -7.264878749847412,
"step": 240
},
{
"epoch": 2.3128888888888888,
"grad_norm": 2778.763671875,
"learning_rate": 7.1757645529443665e-06,
"log_odds_chosen": 12.886404037475586,
"log_odds_ratio": -0.262265145778656,
"logps/chosen": -1.1175204515457153,
"logps/rejected": -13.538418769836426,
"loss": -32.4414,
"nll_loss": 1.350950002670288,
"po_loss": -2.3647446632385254,
"rewards/accuracies": 0.90625,
"rewards/chosen": -0.5587602257728577,
"rewards/margins": 6.210448265075684,
"rewards/rejected": -6.769209384918213,
"step": 245
},
{
"epoch": 2.3602962962962963,
"grad_norm": 2757.852783203125,
"learning_rate": 6.230714818829733e-06,
"log_odds_chosen": 12.95228385925293,
"log_odds_ratio": -0.26469850540161133,
"logps/chosen": -0.9851976633071899,
"logps/rejected": -13.434306144714355,
"loss": -44.9231,
"nll_loss": 1.3150312900543213,
"po_loss": -2.7188773155212402,
"rewards/accuracies": 0.903124988079071,
"rewards/chosen": -0.49259883165359497,
"rewards/margins": 6.224554538726807,
"rewards/rejected": -6.717153072357178,
"step": 250
},
{
"epoch": 2.407703703703704,
"grad_norm": 2125.9111328125,
"learning_rate": 5.343475104027743e-06,
"log_odds_chosen": 12.030647277832031,
"log_odds_ratio": -0.5396193265914917,
"logps/chosen": -1.6967741250991821,
"logps/rejected": -13.254587173461914,
"loss": -11.2409,
"nll_loss": 1.7866249084472656,
"po_loss": -2.1379032135009766,
"rewards/accuracies": 0.8999999761581421,
"rewards/chosen": -0.8483870625495911,
"rewards/margins": 5.77890682220459,
"rewards/rejected": -6.627293586730957,
"step": 255
},
{
"epoch": 2.455111111111111,
"grad_norm": 2060.32958984375,
"learning_rate": 4.516778136213037e-06,
"log_odds_chosen": 13.848899841308594,
"log_odds_ratio": -0.3086540997028351,
"logps/chosen": -1.331648349761963,
"logps/rejected": -14.704687118530273,
"loss": -44.157,
"nll_loss": 1.3919355869293213,
"po_loss": -2.7718422412872314,
"rewards/accuracies": 0.890625,
"rewards/chosen": -0.6658241748809814,
"rewards/margins": 6.686519622802734,
"rewards/rejected": -7.352343559265137,
"step": 260
},
{
"epoch": 2.5025185185185186,
"grad_norm": 5455.78369140625,
"learning_rate": 3.7531701693965554e-06,
"log_odds_chosen": 16.588672637939453,
"log_odds_ratio": -0.3408772945404053,
"logps/chosen": -1.6753642559051514,
"logps/rejected": -17.813495635986328,
"loss": -37.1963,
"nll_loss": 1.6350977420806885,
"po_loss": -2.7974817752838135,
"rewards/accuracies": 0.8999999761581421,
"rewards/chosen": -0.8376821279525757,
"rewards/margins": 8.069067001342773,
"rewards/rejected": -8.906747817993164,
"step": 265
},
{
"epoch": 2.549925925925926,
"grad_norm": 4511.32470703125,
"learning_rate": 3.055003141378948e-06,
"log_odds_chosen": 14.822712898254395,
"log_odds_ratio": -0.9747039079666138,
"logps/chosen": -2.418854236602783,
"logps/rejected": -16.78946304321289,
"loss": -6.2935,
"nll_loss": 2.2066473960876465,
"po_loss": -2.403317451477051,
"rewards/accuracies": 0.878125011920929,
"rewards/chosen": -1.2094271183013916,
"rewards/margins": 7.185305118560791,
"rewards/rejected": -8.394731521606445,
"step": 270
},
{
"epoch": 2.5973333333333333,
"grad_norm": 2031.630859375,
"learning_rate": 2.424427429704365e-06,
"log_odds_chosen": 12.37581729888916,
"log_odds_ratio": -0.4414401650428772,
"logps/chosen": -1.395058035850525,
"logps/rejected": -13.333681106567383,
"loss": -32.9732,
"nll_loss": 1.4321500062942505,
"po_loss": -2.4625613689422607,
"rewards/accuracies": 0.909375011920929,
"rewards/chosen": -0.6975290179252625,
"rewards/margins": 5.969311714172363,
"rewards/rejected": -6.666840553283691,
"step": 275
},
{
"epoch": 2.644740740740741,
"grad_norm": 2148.434326171875,
"learning_rate": 1.8633852284264508e-06,
"log_odds_chosen": 12.9013671875,
"log_odds_ratio": -0.5055217742919922,
"logps/chosen": -1.2132359743118286,
"logps/rejected": -13.65849494934082,
"loss": -46.2362,
"nll_loss": 1.4333760738372803,
"po_loss": -2.8782572746276855,
"rewards/accuracies": 0.871874988079071,
"rewards/chosen": -0.6066179871559143,
"rewards/margins": 6.222629547119141,
"rewards/rejected": -6.82924747467041,
"step": 280
},
{
"epoch": 2.6921481481481484,
"grad_norm": 1996.1431884765625,
"learning_rate": 1.3736045660864034e-06,
"log_odds_chosen": 9.410332679748535,
"log_odds_ratio": -0.5190314054489136,
"logps/chosen": -1.217825174331665,
"logps/rejected": -10.21954345703125,
"loss": -21.9301,
"nll_loss": 1.4607911109924316,
"po_loss": -2.1461057662963867,
"rewards/accuracies": 0.875,
"rewards/chosen": -0.6089125871658325,
"rewards/margins": 4.500859260559082,
"rewards/rejected": -5.109771728515625,
"step": 285
},
{
"epoch": 2.7395555555555555,
"grad_norm": 2126.190673828125,
"learning_rate": 9.565939833279192e-07,
"log_odds_chosen": 9.630215644836426,
"log_odds_ratio": -0.6117982268333435,
"logps/chosen": -1.3845949172973633,
"logps/rejected": -10.577371597290039,
"loss": -7.386,
"nll_loss": 1.5414915084838867,
"po_loss": -1.7723032236099243,
"rewards/accuracies": 0.8843749761581421,
"rewards/chosen": -0.6922974586486816,
"rewards/margins": 4.596388816833496,
"rewards/rejected": -5.2886857986450195,
"step": 290
},
{
"epoch": 2.786962962962963,
"grad_norm": 1371.4443359375,
"learning_rate": 6.136378865420872e-07,
"log_odds_chosen": 9.735053062438965,
"log_odds_ratio": -0.34574881196022034,
"logps/chosen": -1.0634145736694336,
"logps/rejected": -10.3881254196167,
"loss": -27.8217,
"nll_loss": 1.3093498945236206,
"po_loss": -2.1787781715393066,
"rewards/accuracies": 0.856249988079071,
"rewards/chosen": -0.5317072868347168,
"rewards/margins": 4.662355899810791,
"rewards/rejected": -5.19406270980835,
"step": 295
},
{
"epoch": 2.83437037037037,
"grad_norm": 1635.2408447265625,
"learning_rate": 3.45792591853214e-07,
"log_odds_chosen": 12.451967239379883,
"log_odds_ratio": -0.2706998884677887,
"logps/chosen": -0.9222389459609985,
"logps/rejected": -12.897394180297852,
"loss": -45.3531,
"nll_loss": 1.2098591327667236,
"po_loss": -2.6271443367004395,
"rewards/accuracies": 0.887499988079071,
"rewards/chosen": -0.46111947298049927,
"rewards/margins": 5.987576961517334,
"rewards/rejected": -6.448697090148926,
"step": 300
},
{
"epoch": 2.8817777777777778,
"grad_norm": 1768.5748291015625,
"learning_rate": 1.538830716302092e-07,
"log_odds_chosen": 12.523870468139648,
"log_odds_ratio": -0.49474939703941345,
"logps/chosen": -1.1987851858139038,
"logps/rejected": -13.224016189575195,
"loss": -39.2375,
"nll_loss": 1.4008312225341797,
"po_loss": -2.6270015239715576,
"rewards/accuracies": 0.8999999761581421,
"rewards/chosen": -0.5993925929069519,
"rewards/margins": 6.012616157531738,
"rewards/rejected": -6.612008094787598,
"step": 305
},
{
"epoch": 2.9291851851851853,
"grad_norm": 2095.619140625,
"learning_rate": 3.8500413544415025e-08,
"log_odds_chosen": 13.987775802612305,
"log_odds_ratio": -0.43595457077026367,
"logps/chosen": -1.334578514099121,
"logps/rejected": -14.836097717285156,
"loss": -48.9085,
"nll_loss": 1.5135501623153687,
"po_loss": -3.041940689086914,
"rewards/accuracies": 0.90625,
"rewards/chosen": -0.6672892570495605,
"rewards/margins": 6.750760078430176,
"rewards/rejected": -7.418048858642578,
"step": 310
},
{
"epoch": 2.9765925925925925,
"grad_norm": 4473.79345703125,
"learning_rate": 0.0,
"log_odds_chosen": 12.231574058532715,
"log_odds_ratio": -0.3107563853263855,
"logps/chosen": -1.1671451330184937,
"logps/rejected": -12.893664360046387,
"loss": -49.9097,
"nll_loss": 1.2714515924453735,
"po_loss": -2.8311288356781006,
"rewards/accuracies": 0.893750011920929,
"rewards/chosen": -0.5835725665092468,
"rewards/margins": 5.863259315490723,
"rewards/rejected": -6.446832180023193,
"step": 315
},
{
"epoch": 2.9765925925925925,
"step": 315,
"total_flos": 0.0,
"train_loss": 42.33665879264711,
"train_runtime": 9540.6023,
"train_samples_per_second": 2.123,
"train_steps_per_second": 0.033
}
],
"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
}