Llama-3.1-8B-Instruct-SAA-800 / trainer_state.json
chchen's picture
End of training
d7d433d verified
{
"best_metric": 0.1529119908809662,
"best_model_checkpoint": "saves/Llama-3.1-8B-Instruct/lora/saa-800/checkpoint-450",
"epoch": 10.0,
"eval_steps": 50,
"global_step": 450,
"is_hyper_param_search": false,
"is_local_process_zero": true,
"is_world_process_zero": true,
"log_history": [
{
"epoch": 0.2222222222222222,
"grad_norm": 6.713047504425049,
"learning_rate": 1.111111111111111e-06,
"logits/chosen": -0.4419662356376648,
"logits/rejected": -0.5169282555580139,
"logps/chosen": -1.7502187490463257,
"logps/rejected": -2.110954523086548,
"loss": 1.8058,
"odds_ratio_loss": 15.742452621459961,
"rewards/accuracies": 0.75,
"rewards/chosen": -0.17502185702323914,
"rewards/margins": 0.03607357665896416,
"rewards/rejected": -0.21109545230865479,
"sft_loss": 0.23150739073753357,
"step": 10
},
{
"epoch": 0.4444444444444444,
"grad_norm": 7.350757122039795,
"learning_rate": 2.222222222222222e-06,
"logits/chosen": -0.390705406665802,
"logits/rejected": -0.4706195294857025,
"logps/chosen": -1.6888424158096313,
"logps/rejected": -2.1693763732910156,
"loss": 1.7416,
"odds_ratio_loss": 15.1643705368042,
"rewards/accuracies": 0.7562500238418579,
"rewards/chosen": -0.16888423264026642,
"rewards/margins": 0.048053398728370667,
"rewards/rejected": -0.21693763136863708,
"sft_loss": 0.22513218224048615,
"step": 20
},
{
"epoch": 0.6666666666666666,
"grad_norm": 8.413761138916016,
"learning_rate": 3.3333333333333333e-06,
"logits/chosen": -0.42878788709640503,
"logits/rejected": -0.5077282190322876,
"logps/chosen": -1.698947548866272,
"logps/rejected": -2.1924192905426025,
"loss": 1.7505,
"odds_ratio_loss": 15.268714904785156,
"rewards/accuracies": 0.800000011920929,
"rewards/chosen": -0.1698947548866272,
"rewards/margins": 0.04934718459844589,
"rewards/rejected": -0.2192419320344925,
"sft_loss": 0.22367195785045624,
"step": 30
},
{
"epoch": 0.8888888888888888,
"grad_norm": 6.453053951263428,
"learning_rate": 4.444444444444444e-06,
"logits/chosen": -0.40729817748069763,
"logits/rejected": -0.480067640542984,
"logps/chosen": -1.677170753479004,
"logps/rejected": -1.9914417266845703,
"loss": 1.736,
"odds_ratio_loss": 15.256231307983398,
"rewards/accuracies": 0.706250011920929,
"rewards/chosen": -0.1677170693874359,
"rewards/margins": 0.03142711520195007,
"rewards/rejected": -0.1991441696882248,
"sft_loss": 0.21038159728050232,
"step": 40
},
{
"epoch": 1.1111111111111112,
"grad_norm": 5.584766864776611,
"learning_rate": 4.998119881260576e-06,
"logits/chosen": -0.4123212695121765,
"logits/rejected": -0.49063047766685486,
"logps/chosen": -1.4437320232391357,
"logps/rejected": -1.951922059059143,
"loss": 1.4945,
"odds_ratio_loss": 13.130969047546387,
"rewards/accuracies": 0.7875000238418579,
"rewards/chosen": -0.14437320828437805,
"rewards/margins": 0.050819020718336105,
"rewards/rejected": -0.19519223272800446,
"sft_loss": 0.18137237429618835,
"step": 50
},
{
"epoch": 1.1111111111111112,
"eval_logits/chosen": -0.3793938159942627,
"eval_logits/rejected": -0.45482414960861206,
"eval_logps/chosen": -1.228076696395874,
"eval_logps/rejected": -1.7094627618789673,
"eval_loss": 1.276250958442688,
"eval_odds_ratio_loss": 11.260908126831055,
"eval_rewards/accuracies": 0.7875000238418579,
"eval_rewards/chosen": -0.12280768156051636,
"eval_rewards/margins": 0.04813859239220619,
"eval_rewards/rejected": -0.17094627022743225,
"eval_runtime": 3.3465,
"eval_samples_per_second": 23.905,
"eval_sft_loss": 0.15016020834445953,
"eval_steps_per_second": 11.953,
"step": 50
},
{
"epoch": 1.3333333333333333,
"grad_norm": 3.5215771198272705,
"learning_rate": 4.983095894354858e-06,
"logits/chosen": -0.4166947901248932,
"logits/rejected": -0.49852150678634644,
"logps/chosen": -1.0756200551986694,
"logps/rejected": -1.5570634603500366,
"loss": 1.1242,
"odds_ratio_loss": 9.97706127166748,
"rewards/accuracies": 0.831250011920929,
"rewards/chosen": -0.10756200551986694,
"rewards/margins": 0.04814432933926582,
"rewards/rejected": -0.15570633113384247,
"sft_loss": 0.1265355795621872,
"step": 60
},
{
"epoch": 1.5555555555555556,
"grad_norm": 3.005864381790161,
"learning_rate": 4.953138276568462e-06,
"logits/chosen": -0.45257633924484253,
"logits/rejected": -0.5196300148963928,
"logps/chosen": -0.6806862950325012,
"logps/rejected": -1.1254621744155884,
"loss": 0.7274,
"odds_ratio_loss": 6.491485595703125,
"rewards/accuracies": 0.800000011920929,
"rewards/chosen": -0.06806863099336624,
"rewards/margins": 0.044477589428424835,
"rewards/rejected": -0.11254620552062988,
"sft_loss": 0.07826091349124908,
"step": 70
},
{
"epoch": 1.7777777777777777,
"grad_norm": 3.241006851196289,
"learning_rate": 4.908427196539701e-06,
"logits/chosen": -0.4250833988189697,
"logits/rejected": -0.49341732263565063,
"logps/chosen": -0.586665153503418,
"logps/rejected": -0.9428589940071106,
"loss": 0.6374,
"odds_ratio_loss": 5.749706268310547,
"rewards/accuracies": 0.762499988079071,
"rewards/chosen": -0.05866651609539986,
"rewards/margins": 0.03561938554048538,
"rewards/rejected": -0.09428589791059494,
"sft_loss": 0.06239296868443489,
"step": 80
},
{
"epoch": 2.0,
"grad_norm": 3.3662965297698975,
"learning_rate": 4.849231551964771e-06,
"logits/chosen": -0.39320772886276245,
"logits/rejected": -0.48544201254844666,
"logps/chosen": -0.3357451558113098,
"logps/rejected": -0.9199529886245728,
"loss": 0.3774,
"odds_ratio_loss": 3.414803981781006,
"rewards/accuracies": 0.8187500238418579,
"rewards/chosen": -0.03357451409101486,
"rewards/margins": 0.05842079594731331,
"rewards/rejected": -0.09199531376361847,
"sft_loss": 0.03593815118074417,
"step": 90
},
{
"epoch": 2.2222222222222223,
"grad_norm": 1.9536067247390747,
"learning_rate": 4.775907352415367e-06,
"logits/chosen": -0.3746485710144043,
"logits/rejected": -0.46515822410583496,
"logps/chosen": -0.22609436511993408,
"logps/rejected": -0.7698042392730713,
"loss": 0.2666,
"odds_ratio_loss": 2.4158308506011963,
"rewards/accuracies": 0.8062499761581421,
"rewards/chosen": -0.022609438747167587,
"rewards/margins": 0.05437099188566208,
"rewards/rejected": -0.07698042690753937,
"sft_loss": 0.02498885616660118,
"step": 100
},
{
"epoch": 2.2222222222222223,
"eval_logits/chosen": -0.36658304929733276,
"eval_logits/rejected": -0.4319043755531311,
"eval_logps/chosen": -0.2119404524564743,
"eval_logps/rejected": -0.7287805676460266,
"eval_loss": 0.24907442927360535,
"eval_odds_ratio_loss": 2.2483649253845215,
"eval_rewards/accuracies": 0.824999988079071,
"eval_rewards/chosen": -0.02119404636323452,
"eval_rewards/margins": 0.05168401449918747,
"eval_rewards/rejected": -0.07287804782390594,
"eval_runtime": 3.3336,
"eval_samples_per_second": 23.998,
"eval_sft_loss": 0.02423796057701111,
"eval_steps_per_second": 11.999,
"step": 100
},
{
"epoch": 2.4444444444444446,
"grad_norm": 1.7341715097427368,
"learning_rate": 4.688895578255228e-06,
"logits/chosen": -0.38399261236190796,
"logits/rejected": -0.45104750990867615,
"logps/chosen": -0.1595337688922882,
"logps/rejected": -0.6476808190345764,
"loss": 0.1984,
"odds_ratio_loss": 1.7913240194320679,
"rewards/accuracies": 0.8062499761581421,
"rewards/chosen": -0.01595337688922882,
"rewards/margins": 0.048814695328474045,
"rewards/rejected": -0.06476806849241257,
"sft_loss": 0.0192393995821476,
"step": 110
},
{
"epoch": 2.6666666666666665,
"grad_norm": 1.6577801704406738,
"learning_rate": 4.588719528532342e-06,
"logits/chosen": -0.42996135354042053,
"logits/rejected": -0.5020943880081177,
"logps/chosen": -0.11305193603038788,
"logps/rejected": -0.6292703747749329,
"loss": 0.1528,
"odds_ratio_loss": 1.378528356552124,
"rewards/accuracies": 0.8125,
"rewards/chosen": -0.011305193416774273,
"rewards/margins": 0.05162184312939644,
"rewards/rejected": -0.06292703002691269,
"sft_loss": 0.014913685619831085,
"step": 120
},
{
"epoch": 2.888888888888889,
"grad_norm": 0.804415762424469,
"learning_rate": 4.475981673796899e-06,
"logits/chosen": -0.4010156989097595,
"logits/rejected": -0.4532565474510193,
"logps/chosen": -0.11344721168279648,
"logps/rejected": -0.5929852724075317,
"loss": 0.1557,
"odds_ratio_loss": 1.430558204650879,
"rewards/accuracies": 0.824999988079071,
"rewards/chosen": -0.011344721540808678,
"rewards/margins": 0.047953806817531586,
"rewards/rejected": -0.059298526495695114,
"sft_loss": 0.012685336172580719,
"step": 130
},
{
"epoch": 3.111111111111111,
"grad_norm": 1.188767910003662,
"learning_rate": 4.351360032772512e-06,
"logits/chosen": -0.37222054600715637,
"logits/rejected": -0.4317198395729065,
"logps/chosen": -0.10829365253448486,
"logps/rejected": -0.6532029509544373,
"loss": 0.1518,
"odds_ratio_loss": 1.385451078414917,
"rewards/accuracies": 0.824999988079071,
"rewards/chosen": -0.010829364880919456,
"rewards/margins": 0.05449093505740166,
"rewards/rejected": -0.06532029807567596,
"sft_loss": 0.013293206691741943,
"step": 140
},
{
"epoch": 3.3333333333333335,
"grad_norm": 1.1802352666854858,
"learning_rate": 4.215604094671835e-06,
"logits/chosen": -0.3585512936115265,
"logits/rejected": -0.4357479512691498,
"logps/chosen": -0.07403306663036346,
"logps/rejected": -0.7158663868904114,
"loss": 0.1014,
"odds_ratio_loss": 0.9041417837142944,
"rewards/accuracies": 0.875,
"rewards/chosen": -0.007403307594358921,
"rewards/margins": 0.06418333202600479,
"rewards/rejected": -0.07158664613962173,
"sft_loss": 0.010936584323644638,
"step": 150
},
{
"epoch": 3.3333333333333335,
"eval_logits/chosen": -0.3283754289150238,
"eval_logits/rejected": -0.3820287585258484,
"eval_logps/chosen": -0.1292387694120407,
"eval_logps/rejected": -0.6057685017585754,
"eval_loss": 0.16315264999866486,
"eval_odds_ratio_loss": 1.4634513854980469,
"eval_rewards/accuracies": 0.8125,
"eval_rewards/chosen": -0.012923876754939556,
"eval_rewards/margins": 0.047652970999479294,
"eval_rewards/rejected": -0.060576848685741425,
"eval_runtime": 3.3355,
"eval_samples_per_second": 23.984,
"eval_sft_loss": 0.016807515174150467,
"eval_steps_per_second": 11.992,
"step": 150
},
{
"epoch": 3.5555555555555554,
"grad_norm": 2.5364763736724854,
"learning_rate": 4.069530311680247e-06,
"logits/chosen": -0.37899985909461975,
"logits/rejected": -0.4429406225681305,
"logps/chosen": -0.09646956622600555,
"logps/rejected": -0.5016245245933533,
"loss": 0.1441,
"odds_ratio_loss": 1.3272716999053955,
"rewards/accuracies": 0.793749988079071,
"rewards/chosen": -0.00964695680886507,
"rewards/margins": 0.04051550105214119,
"rewards/rejected": -0.050162456929683685,
"sft_loss": 0.011343193240463734,
"step": 160
},
{
"epoch": 3.7777777777777777,
"grad_norm": 1.1225122213363647,
"learning_rate": 3.914017188716347e-06,
"logits/chosen": -0.32853808999061584,
"logits/rejected": -0.3947625160217285,
"logps/chosen": -0.0880068689584732,
"logps/rejected": -0.5245143175125122,
"loss": 0.1251,
"odds_ratio_loss": 1.1624243259429932,
"rewards/accuracies": 0.824999988079071,
"rewards/chosen": -0.00880068726837635,
"rewards/margins": 0.04365074634552002,
"rewards/rejected": -0.05245143175125122,
"sft_loss": 0.00884676817804575,
"step": 170
},
{
"epoch": 4.0,
"grad_norm": 2.3566975593566895,
"learning_rate": 3.7500000000000005e-06,
"logits/chosen": -0.394605427980423,
"logits/rejected": -0.4463408589363098,
"logps/chosen": -0.09484803676605225,
"logps/rejected": -0.5903218984603882,
"loss": 0.1306,
"odds_ratio_loss": 1.1676770448684692,
"rewards/accuracies": 0.7875000238418579,
"rewards/chosen": -0.00948480237275362,
"rewards/margins": 0.049547381699085236,
"rewards/rejected": -0.05903219059109688,
"sft_loss": 0.013791071251034737,
"step": 180
},
{
"epoch": 4.222222222222222,
"grad_norm": 1.111855387687683,
"learning_rate": 3.578465164203134e-06,
"logits/chosen": -0.3551129698753357,
"logits/rejected": -0.43426722288131714,
"logps/chosen": -0.06811969727277756,
"logps/rejected": -0.6758677959442139,
"loss": 0.0949,
"odds_ratio_loss": 0.8656018376350403,
"rewards/accuracies": 0.887499988079071,
"rewards/chosen": -0.006811970379203558,
"rewards/margins": 0.06077481061220169,
"rewards/rejected": -0.06758677959442139,
"sft_loss": 0.008338207378983498,
"step": 190
},
{
"epoch": 4.444444444444445,
"grad_norm": 1.3708534240722656,
"learning_rate": 3.400444312011776e-06,
"logits/chosen": -0.345875084400177,
"logits/rejected": -0.3952234089374542,
"logps/chosen": -0.10109534114599228,
"logps/rejected": -0.5074071288108826,
"loss": 0.1429,
"odds_ratio_loss": 1.325073003768921,
"rewards/accuracies": 0.8187500238418579,
"rewards/chosen": -0.010109534487128258,
"rewards/margins": 0.04063117876648903,
"rewards/rejected": -0.05074070766568184,
"sft_loss": 0.010358814150094986,
"step": 200
},
{
"epoch": 4.444444444444445,
"eval_logits/chosen": -0.32978400588035583,
"eval_logits/rejected": -0.38182884454727173,
"eval_logps/chosen": -0.121058389544487,
"eval_logps/rejected": -0.5841361284255981,
"eval_loss": 0.1533508002758026,
"eval_odds_ratio_loss": 1.3751581907272339,
"eval_rewards/accuracies": 0.8125,
"eval_rewards/chosen": -0.01210583932697773,
"eval_rewards/margins": 0.04630777984857559,
"eval_rewards/rejected": -0.058413613587617874,
"eval_runtime": 3.3378,
"eval_samples_per_second": 23.968,
"eval_sft_loss": 0.015834979712963104,
"eval_steps_per_second": 11.984,
"step": 200
},
{
"epoch": 4.666666666666667,
"grad_norm": 1.4620966911315918,
"learning_rate": 3.217008081777726e-06,
"logits/chosen": -0.3843366503715515,
"logits/rejected": -0.4375368058681488,
"logps/chosen": -0.08334760367870331,
"logps/rejected": -0.5660797357559204,
"loss": 0.1171,
"odds_ratio_loss": 1.0661273002624512,
"rewards/accuracies": 0.84375,
"rewards/chosen": -0.00833476148545742,
"rewards/margins": 0.04827320575714111,
"rewards/rejected": -0.05660796910524368,
"sft_loss": 0.010473220609128475,
"step": 210
},
{
"epoch": 4.888888888888889,
"grad_norm": 1.2518880367279053,
"learning_rate": 3.0292596805735275e-06,
"logits/chosen": -0.3598805367946625,
"logits/rejected": -0.4199501872062683,
"logps/chosen": -0.07187429815530777,
"logps/rejected": -0.5510466694831848,
"loss": 0.1029,
"odds_ratio_loss": 0.9251865148544312,
"rewards/accuracies": 0.831250011920929,
"rewards/chosen": -0.007187430746853352,
"rewards/margins": 0.047917239367961884,
"rewards/rejected": -0.05510466545820236,
"sft_loss": 0.010374334640800953,
"step": 220
},
{
"epoch": 5.111111111111111,
"grad_norm": 1.3117250204086304,
"learning_rate": 2.8383282493753282e-06,
"logits/chosen": -0.3635903596878052,
"logits/rejected": -0.43442487716674805,
"logps/chosen": -0.06123008579015732,
"logps/rejected": -0.7030965089797974,
"loss": 0.0878,
"odds_ratio_loss": 0.7977678179740906,
"rewards/accuracies": 0.8374999761581421,
"rewards/chosen": -0.006123009137809277,
"rewards/margins": 0.06418663263320923,
"rewards/rejected": -0.07030965387821198,
"sft_loss": 0.007975312881171703,
"step": 230
},
{
"epoch": 5.333333333333333,
"grad_norm": 2.640739679336548,
"learning_rate": 2.6453620722761897e-06,
"logits/chosen": -0.350188285112381,
"logits/rejected": -0.4023573398590088,
"logps/chosen": -0.08658941090106964,
"logps/rejected": -0.6677496433258057,
"loss": 0.1183,
"odds_ratio_loss": 1.068228006362915,
"rewards/accuracies": 0.8687499761581421,
"rewards/chosen": -0.00865894090384245,
"rewards/margins": 0.05811602994799614,
"rewards/rejected": -0.06677497178316116,
"sft_loss": 0.011527000926434994,
"step": 240
},
{
"epoch": 5.555555555555555,
"grad_norm": 1.3426445722579956,
"learning_rate": 2.4515216705704396e-06,
"logits/chosen": -0.3609291911125183,
"logits/rejected": -0.4190797209739685,
"logps/chosen": -0.0702720507979393,
"logps/rejected": -0.596028745174408,
"loss": 0.1007,
"odds_ratio_loss": 0.925524115562439,
"rewards/accuracies": 0.8374999761581421,
"rewards/chosen": -0.00702720507979393,
"rewards/margins": 0.052575670182704926,
"rewards/rejected": -0.059602878987789154,
"sft_loss": 0.008168576285243034,
"step": 250
},
{
"epoch": 5.555555555555555,
"eval_logits/chosen": -0.32348352670669556,
"eval_logits/rejected": -0.37402915954589844,
"eval_logps/chosen": -0.12062982469797134,
"eval_logps/rejected": -0.6407424807548523,
"eval_loss": 0.15295907855033875,
"eval_odds_ratio_loss": 1.3704458475112915,
"eval_rewards/accuracies": 0.8125,
"eval_rewards/chosen": -0.012062981724739075,
"eval_rewards/margins": 0.05201127007603645,
"eval_rewards/rejected": -0.06407425552606583,
"eval_runtime": 3.3317,
"eval_samples_per_second": 24.012,
"eval_sft_loss": 0.01591448113322258,
"eval_steps_per_second": 12.006,
"step": 250
},
{
"epoch": 5.777777777777778,
"grad_norm": 1.4032063484191895,
"learning_rate": 2.2579728232420524e-06,
"logits/chosen": -0.40077710151672363,
"logits/rejected": -0.4527947008609772,
"logps/chosen": -0.0916273444890976,
"logps/rejected": -0.5904611349105835,
"loss": 0.127,
"odds_ratio_loss": 1.1558631658554077,
"rewards/accuracies": 0.8374999761581421,
"rewards/chosen": -0.009162734262645245,
"rewards/margins": 0.049883387982845306,
"rewards/rejected": -0.05904611945152283,
"sft_loss": 0.011430850252509117,
"step": 260
},
{
"epoch": 6.0,
"grad_norm": 1.6447800397872925,
"learning_rate": 2.0658795558326745e-06,
"logits/chosen": -0.36904090642929077,
"logits/rejected": -0.43377822637557983,
"logps/chosen": -0.05662701651453972,
"logps/rejected": -0.6641786098480225,
"loss": 0.0778,
"odds_ratio_loss": 0.7118644714355469,
"rewards/accuracies": 0.893750011920929,
"rewards/chosen": -0.005662702023983002,
"rewards/margins": 0.060755155980587006,
"rewards/rejected": -0.06641785800457001,
"sft_loss": 0.006575153209269047,
"step": 270
},
{
"epoch": 6.222222222222222,
"grad_norm": 2.223752021789551,
"learning_rate": 1.876397139855047e-06,
"logits/chosen": -0.3556548058986664,
"logits/rejected": -0.4135330319404602,
"logps/chosen": -0.0746704488992691,
"logps/rejected": -0.6784394979476929,
"loss": 0.1018,
"odds_ratio_loss": 0.9240021705627441,
"rewards/accuracies": 0.875,
"rewards/chosen": -0.007467044983059168,
"rewards/margins": 0.06037690117955208,
"rewards/rejected": -0.0678439512848854,
"sft_loss": 0.009404648095369339,
"step": 280
},
{
"epoch": 6.444444444444445,
"grad_norm": 1.75334894657135,
"learning_rate": 1.6906651448541977e-06,
"logits/chosen": -0.3690106272697449,
"logits/rejected": -0.4269256591796875,
"logps/chosen": -0.05620593950152397,
"logps/rejected": -0.6589001417160034,
"loss": 0.0805,
"odds_ratio_loss": 0.7411076426506042,
"rewards/accuracies": 0.887499988079071,
"rewards/chosen": -0.005620594136416912,
"rewards/margins": 0.060269422829151154,
"rewards/rejected": -0.06589001417160034,
"sft_loss": 0.006437716074287891,
"step": 290
},
{
"epoch": 6.666666666666667,
"grad_norm": 4.5506911277771,
"learning_rate": 1.509800584902108e-06,
"logits/chosen": -0.3519323468208313,
"logits/rejected": -0.41471752524375916,
"logps/chosen": -0.10020919144153595,
"logps/rejected": -0.6135331392288208,
"loss": 0.1385,
"odds_ratio_loss": 1.263484001159668,
"rewards/accuracies": 0.8062499761581421,
"rewards/chosen": -0.010020919144153595,
"rewards/margins": 0.051332395523786545,
"rewards/rejected": -0.06135330721735954,
"sft_loss": 0.012116280384361744,
"step": 300
},
{
"epoch": 6.666666666666667,
"eval_logits/chosen": -0.32135990262031555,
"eval_logits/rejected": -0.3725373148918152,
"eval_logps/chosen": -0.12169794738292694,
"eval_logps/rejected": -0.6881048083305359,
"eval_loss": 0.15343782305717468,
"eval_odds_ratio_loss": 1.3728852272033691,
"eval_rewards/accuracies": 0.800000011920929,
"eval_rewards/chosen": -0.012169795110821724,
"eval_rewards/margins": 0.05664069205522537,
"eval_rewards/rejected": -0.06881048530340195,
"eval_runtime": 3.3342,
"eval_samples_per_second": 23.994,
"eval_sft_loss": 0.016149301081895828,
"eval_steps_per_second": 11.997,
"step": 300
},
{
"epoch": 6.888888888888889,
"grad_norm": 1.4514155387878418,
"learning_rate": 1.3348912007436538e-06,
"logits/chosen": -0.33361396193504333,
"logits/rejected": -0.3964681923389435,
"logps/chosen": -0.05151065066456795,
"logps/rejected": -0.6935869455337524,
"loss": 0.075,
"odds_ratio_loss": 0.6938644051551819,
"rewards/accuracies": 0.90625,
"rewards/chosen": -0.005151065066456795,
"rewards/margins": 0.06420762091875076,
"rewards/rejected": -0.069358691573143,
"sft_loss": 0.005607123486697674,
"step": 310
},
{
"epoch": 7.111111111111111,
"grad_norm": 1.1120833158493042,
"learning_rate": 1.1669889179957725e-06,
"logits/chosen": -0.34897491335868835,
"logits/rejected": -0.4280971884727478,
"logps/chosen": -0.06502507627010345,
"logps/rejected": -0.724215030670166,
"loss": 0.0902,
"odds_ratio_loss": 0.8162932395935059,
"rewards/accuracies": 0.856249988079071,
"rewards/chosen": -0.006502507720142603,
"rewards/margins": 0.06591899693012238,
"rewards/rejected": -0.07242151349782944,
"sft_loss": 0.008545474149286747,
"step": 320
},
{
"epoch": 7.333333333333333,
"grad_norm": 0.9275766611099243,
"learning_rate": 1.0071035207430352e-06,
"logits/chosen": -0.34552061557769775,
"logits/rejected": -0.4186931550502777,
"logps/chosen": -0.057514738291502,
"logps/rejected": -0.7304258942604065,
"loss": 0.0812,
"odds_ratio_loss": 0.7468503713607788,
"rewards/accuracies": 0.8999999761581421,
"rewards/chosen": -0.005751474294811487,
"rewards/margins": 0.06729111820459366,
"rewards/rejected": -0.07304258644580841,
"sft_loss": 0.006484520621597767,
"step": 330
},
{
"epoch": 7.555555555555555,
"grad_norm": 1.7526174783706665,
"learning_rate": 8.561965785773413e-07,
"logits/chosen": -0.35446950793266296,
"logits/rejected": -0.41398343443870544,
"logps/chosen": -0.04472974315285683,
"logps/rejected": -0.6881515979766846,
"loss": 0.0655,
"odds_ratio_loss": 0.5992408394813538,
"rewards/accuracies": 0.9125000238418579,
"rewards/chosen": -0.004472973756492138,
"rewards/margins": 0.0643421858549118,
"rewards/rejected": -0.06881515681743622,
"sft_loss": 0.0055578285828232765,
"step": 340
},
{
"epoch": 7.777777777777778,
"grad_norm": 1.4603219032287598,
"learning_rate": 7.151756636052529e-07,
"logits/chosen": -0.35244011878967285,
"logits/rejected": -0.4057006239891052,
"logps/chosen": -0.06629346311092377,
"logps/rejected": -0.6709402203559875,
"loss": 0.0918,
"odds_ratio_loss": 0.841974139213562,
"rewards/accuracies": 0.856249988079071,
"rewards/chosen": -0.0066293468698859215,
"rewards/margins": 0.060464680194854736,
"rewards/rejected": -0.06709402799606323,
"sft_loss": 0.0076185790821909904,
"step": 350
},
{
"epoch": 7.777777777777778,
"eval_logits/chosen": -0.31914329528808594,
"eval_logits/rejected": -0.3697816729545593,
"eval_logps/chosen": -0.12174425274133682,
"eval_logps/rejected": -0.6888757944107056,
"eval_loss": 0.1536635458469391,
"eval_odds_ratio_loss": 1.3741592168807983,
"eval_rewards/accuracies": 0.8125,
"eval_rewards/chosen": -0.012174425646662712,
"eval_rewards/margins": 0.05671315640211105,
"eval_rewards/rejected": -0.06888757646083832,
"eval_runtime": 3.3381,
"eval_samples_per_second": 23.966,
"eval_sft_loss": 0.016247618943452835,
"eval_steps_per_second": 11.983,
"step": 350
},
{
"epoch": 8.0,
"grad_norm": 0.9194740056991577,
"learning_rate": 5.848888922025553e-07,
"logits/chosen": -0.3591064214706421,
"logits/rejected": -0.4297390878200531,
"logps/chosen": -0.08391134440898895,
"logps/rejected": -0.7468418478965759,
"loss": 0.118,
"odds_ratio_loss": 1.0900821685791016,
"rewards/accuracies": 0.84375,
"rewards/chosen": -0.00839113537222147,
"rewards/margins": 0.06629304587841034,
"rewards/rejected": -0.07468418776988983,
"sft_loss": 0.008979488164186478,
"step": 360
},
{
"epoch": 8.222222222222221,
"grad_norm": 1.1211438179016113,
"learning_rate": 4.661198243425813e-07,
"logits/chosen": -0.3469446301460266,
"logits/rejected": -0.416546493768692,
"logps/chosen": -0.06295167654752731,
"logps/rejected": -0.6733905076980591,
"loss": 0.0877,
"odds_ratio_loss": 0.7965422868728638,
"rewards/accuracies": 0.887499988079071,
"rewards/chosen": -0.006295167841017246,
"rewards/margins": 0.0610438771545887,
"rewards/rejected": -0.06733904778957367,
"sft_loss": 0.008044925518333912,
"step": 370
},
{
"epoch": 8.444444444444445,
"grad_norm": 2.9023804664611816,
"learning_rate": 3.595827511743341e-07,
"logits/chosen": -0.36501026153564453,
"logits/rejected": -0.4163384437561035,
"logps/chosen": -0.0608031265437603,
"logps/rejected": -0.7410011291503906,
"loss": 0.0857,
"odds_ratio_loss": 0.7958894371986389,
"rewards/accuracies": 0.8999999761581421,
"rewards/chosen": -0.006080311723053455,
"rewards/margins": 0.068019799888134,
"rewards/rejected": -0.07410011440515518,
"sft_loss": 0.006128143519163132,
"step": 380
},
{
"epoch": 8.666666666666666,
"grad_norm": 1.2382421493530273,
"learning_rate": 2.6591839919146963e-07,
"logits/chosen": -0.3580993711948395,
"logits/rejected": -0.4169589579105377,
"logps/chosen": -0.07410109043121338,
"logps/rejected": -0.7789124250411987,
"loss": 0.1054,
"odds_ratio_loss": 0.9693442583084106,
"rewards/accuracies": 0.862500011920929,
"rewards/chosen": -0.0074101099744439125,
"rewards/margins": 0.07048113644123077,
"rewards/rejected": -0.07789124548435211,
"sft_loss": 0.008436056785285473,
"step": 390
},
{
"epoch": 8.88888888888889,
"grad_norm": 1.7495219707489014,
"learning_rate": 1.8569007682777417e-07,
"logits/chosen": -0.37370580434799194,
"logits/rejected": -0.43429917097091675,
"logps/chosen": -0.05244039371609688,
"logps/rejected": -0.7023984789848328,
"loss": 0.0752,
"odds_ratio_loss": 0.6945605874061584,
"rewards/accuracies": 0.887499988079071,
"rewards/chosen": -0.005244039464741945,
"rewards/margins": 0.06499581038951874,
"rewards/rejected": -0.0702398419380188,
"sft_loss": 0.005755658261477947,
"step": 400
},
{
"epoch": 8.88888888888889,
"eval_logits/chosen": -0.3194546103477478,
"eval_logits/rejected": -0.3705773651599884,
"eval_logps/chosen": -0.1213468462228775,
"eval_logps/rejected": -0.689638614654541,
"eval_loss": 0.15339744091033936,
"eval_odds_ratio_loss": 1.3723235130310059,
"eval_rewards/accuracies": 0.800000011920929,
"eval_rewards/chosen": -0.01213468424975872,
"eval_rewards/margins": 0.056829191744327545,
"eval_rewards/rejected": -0.06896387040615082,
"eval_runtime": 3.3351,
"eval_samples_per_second": 23.987,
"eval_sft_loss": 0.016165101900696754,
"eval_steps_per_second": 11.994,
"step": 400
},
{
"epoch": 9.11111111111111,
"grad_norm": 2.9914894104003906,
"learning_rate": 1.1938028665396172e-07,
"logits/chosen": -0.3595274090766907,
"logits/rejected": -0.41625022888183594,
"logps/chosen": -0.07939668744802475,
"logps/rejected": -0.6182006597518921,
"loss": 0.1064,
"odds_ratio_loss": 0.9655207395553589,
"rewards/accuracies": 0.856249988079071,
"rewards/chosen": -0.007939668372273445,
"rewards/margins": 0.05388040468096733,
"rewards/rejected": -0.06182006746530533,
"sft_loss": 0.009880236349999905,
"step": 410
},
{
"epoch": 9.333333333333334,
"grad_norm": 1.7583844661712646,
"learning_rate": 6.738782355044048e-08,
"logits/chosen": -0.35592731833457947,
"logits/rejected": -0.41578736901283264,
"logps/chosen": -0.06436184048652649,
"logps/rejected": -0.6822584867477417,
"loss": 0.0919,
"odds_ratio_loss": 0.8448025584220886,
"rewards/accuracies": 0.8687499761581421,
"rewards/chosen": -0.006436183117330074,
"rewards/margins": 0.06178967282176018,
"rewards/rejected": -0.06822585314512253,
"sft_loss": 0.007466461509466171,
"step": 420
},
{
"epoch": 9.555555555555555,
"grad_norm": 0.7222520112991333,
"learning_rate": 3.0025376307977474e-08,
"logits/chosen": -0.3527710437774658,
"logits/rejected": -0.41858869791030884,
"logps/chosen": -0.042861782014369965,
"logps/rejected": -0.7103925943374634,
"loss": 0.0601,
"odds_ratio_loss": 0.5496730804443359,
"rewards/accuracies": 0.9375,
"rewards/chosen": -0.004286178387701511,
"rewards/margins": 0.0667530745267868,
"rewards/rejected": -0.07103925198316574,
"sft_loss": 0.005130757577717304,
"step": 430
},
{
"epoch": 9.777777777777779,
"grad_norm": 1.2489662170410156,
"learning_rate": 7.517647080519941e-09,
"logits/chosen": -0.34115296602249146,
"logits/rejected": -0.39803385734558105,
"logps/chosen": -0.05217872932553291,
"logps/rejected": -0.8210271000862122,
"loss": 0.0779,
"odds_ratio_loss": 0.7251914739608765,
"rewards/accuracies": 0.8687499761581421,
"rewards/chosen": -0.0052178725600242615,
"rewards/margins": 0.07688483595848083,
"rewards/rejected": -0.0821027159690857,
"sft_loss": 0.0053517939522862434,
"step": 440
},
{
"epoch": 10.0,
"grad_norm": 1.9128329753875732,
"learning_rate": 0.0,
"logits/chosen": -0.3783785402774811,
"logits/rejected": -0.4389713406562805,
"logps/chosen": -0.07910243421792984,
"logps/rejected": -0.6643815040588379,
"loss": 0.1052,
"odds_ratio_loss": 0.9564045667648315,
"rewards/accuracies": 0.875,
"rewards/chosen": -0.007910243235528469,
"rewards/margins": 0.05852789804339409,
"rewards/rejected": -0.06643815338611603,
"sft_loss": 0.009510872885584831,
"step": 450
},
{
"epoch": 10.0,
"eval_logits/chosen": -0.3184443712234497,
"eval_logits/rejected": -0.36903008818626404,
"eval_logps/chosen": -0.1208958849310875,
"eval_logps/rejected": -0.6922038793563843,
"eval_loss": 0.1529119908809662,
"eval_odds_ratio_loss": 1.3675727844238281,
"eval_rewards/accuracies": 0.8125,
"eval_rewards/chosen": -0.012089588679373264,
"eval_rewards/margins": 0.05713079497218132,
"eval_rewards/rejected": -0.06922038644552231,
"eval_runtime": 3.3378,
"eval_samples_per_second": 23.968,
"eval_sft_loss": 0.016154715791344643,
"eval_steps_per_second": 11.984,
"step": 450
},
{
"epoch": 10.0,
"step": 450,
"total_flos": 8.074648418018918e+16,
"train_loss": 0.3428536836306254,
"train_runtime": 994.0751,
"train_samples_per_second": 7.243,
"train_steps_per_second": 0.453
}
],
"logging_steps": 10,
"max_steps": 450,
"num_input_tokens_seen": 0,
"num_train_epochs": 10,
"save_steps": 50,
"stateful_callbacks": {
"TrainerControl": {
"args": {
"should_epoch_stop": false,
"should_evaluate": false,
"should_log": false,
"should_save": true,
"should_training_stop": true
},
"attributes": {}
}
},
"total_flos": 8.074648418018918e+16,
"train_batch_size": 2,
"trial_name": null,
"trial_params": null
}