Llama-3.1-8B-Instruct-SAA-900 / trainer_state.json
chchen's picture
End of training
530809e verified
{
"best_metric": 0.15150243043899536,
"best_model_checkpoint": "saves/Llama-3.1-8B-Instruct/lora/saa-900/checkpoint-200",
"epoch": 9.876543209876543,
"eval_steps": 50,
"global_step": 500,
"is_hyper_param_search": false,
"is_local_process_zero": true,
"is_world_process_zero": true,
"log_history": [
{
"epoch": 0.19753086419753085,
"grad_norm": 5.683405876159668,
"learning_rate": 1.0000000000000002e-06,
"logits/chosen": -0.4294208884239197,
"logits/rejected": -0.5036857724189758,
"logps/chosen": -1.7164901494979858,
"logps/rejected": -2.1387689113616943,
"loss": 1.77,
"odds_ratio_loss": 15.52961540222168,
"rewards/accuracies": 0.793749988079071,
"rewards/chosen": -0.1716490387916565,
"rewards/margins": 0.0422278493642807,
"rewards/rejected": -0.2138768881559372,
"sft_loss": 0.21705082058906555,
"step": 10
},
{
"epoch": 0.3950617283950617,
"grad_norm": 7.8601579666137695,
"learning_rate": 2.0000000000000003e-06,
"logits/chosen": -0.41924962401390076,
"logits/rejected": -0.4839371144771576,
"logps/chosen": -1.7584301233291626,
"logps/rejected": -2.1139349937438965,
"loss": 1.8145,
"odds_ratio_loss": 15.842885971069336,
"rewards/accuracies": 0.7562500238418579,
"rewards/chosen": -0.1758430302143097,
"rewards/margins": 0.03555048629641533,
"rewards/rejected": -0.21139350533485413,
"sft_loss": 0.23018595576286316,
"step": 20
},
{
"epoch": 0.5925925925925926,
"grad_norm": 8.10688591003418,
"learning_rate": 3e-06,
"logits/chosen": -0.4261442720890045,
"logits/rejected": -0.5029150247573853,
"logps/chosen": -1.730046033859253,
"logps/rejected": -2.217639684677124,
"loss": 1.781,
"odds_ratio_loss": 15.661776542663574,
"rewards/accuracies": 0.800000011920929,
"rewards/chosen": -0.173004612326622,
"rewards/margins": 0.048759374767541885,
"rewards/rejected": -0.2217639982700348,
"sft_loss": 0.21483230590820312,
"step": 30
},
{
"epoch": 0.7901234567901234,
"grad_norm": 6.699493885040283,
"learning_rate": 4.000000000000001e-06,
"logits/chosen": -0.41258639097213745,
"logits/rejected": -0.4953138828277588,
"logps/chosen": -1.566969633102417,
"logps/rejected": -2.103066921234131,
"loss": 1.6158,
"odds_ratio_loss": 14.229395866394043,
"rewards/accuracies": 0.7875000238418579,
"rewards/chosen": -0.15669694542884827,
"rewards/margins": 0.053609758615493774,
"rewards/rejected": -0.21030668914318085,
"sft_loss": 0.19286662340164185,
"step": 40
},
{
"epoch": 0.9876543209876543,
"grad_norm": 9.212828636169434,
"learning_rate": 5e-06,
"logits/chosen": -0.4587825834751129,
"logits/rejected": -0.5460661053657532,
"logps/chosen": -1.524173378944397,
"logps/rejected": -1.9827760457992554,
"loss": 1.5773,
"odds_ratio_loss": 13.699198722839355,
"rewards/accuracies": 0.762499988079071,
"rewards/chosen": -0.15241733193397522,
"rewards/margins": 0.04586028307676315,
"rewards/rejected": -0.19827762246131897,
"sft_loss": 0.20735812187194824,
"step": 50
},
{
"epoch": 0.9876543209876543,
"eval_logits/chosen": -0.4034404754638672,
"eval_logits/rejected": -0.46630287170410156,
"eval_logps/chosen": -1.3146759271621704,
"eval_logps/rejected": -1.754432201385498,
"eval_loss": 1.3696272373199463,
"eval_odds_ratio_loss": 11.865678787231445,
"eval_rewards/accuracies": 0.7666666507720947,
"eval_rewards/chosen": -0.13146759569644928,
"eval_rewards/margins": 0.04397563263773918,
"eval_rewards/rejected": -0.17544323205947876,
"eval_runtime": 3.7458,
"eval_samples_per_second": 24.027,
"eval_sft_loss": 0.18305927515029907,
"eval_steps_per_second": 12.013,
"step": 50
},
{
"epoch": 1.1851851851851851,
"grad_norm": 4.2851243019104,
"learning_rate": 4.993910125649561e-06,
"logits/chosen": -0.42754077911376953,
"logits/rejected": -0.49272727966308594,
"logps/chosen": -1.14907968044281,
"logps/rejected": -1.5600894689559937,
"loss": 1.2014,
"odds_ratio_loss": 10.684017181396484,
"rewards/accuracies": 0.762499988079071,
"rewards/chosen": -0.11490797996520996,
"rewards/margins": 0.041100967675447464,
"rewards/rejected": -0.15600894391536713,
"sft_loss": 0.13294857740402222,
"step": 60
},
{
"epoch": 1.382716049382716,
"grad_norm": 2.6753902435302734,
"learning_rate": 4.975670171853926e-06,
"logits/chosen": -0.42295628786087036,
"logits/rejected": -0.5050143003463745,
"logps/chosen": -0.8449110984802246,
"logps/rejected": -1.3062922954559326,
"loss": 0.8939,
"odds_ratio_loss": 7.972494602203369,
"rewards/accuracies": 0.7875000238418579,
"rewards/chosen": -0.08449110388755798,
"rewards/margins": 0.04613811895251274,
"rewards/rejected": -0.13062922656536102,
"sft_loss": 0.09666236490011215,
"step": 70
},
{
"epoch": 1.5802469135802468,
"grad_norm": 2.68009090423584,
"learning_rate": 4.9453690018345144e-06,
"logits/chosen": -0.41366061568260193,
"logits/rejected": -0.5027936100959778,
"logps/chosen": -0.5736119151115417,
"logps/rejected": -1.0685169696807861,
"loss": 0.6186,
"odds_ratio_loss": 5.562970161437988,
"rewards/accuracies": 0.8062499761581421,
"rewards/chosen": -0.057361193001270294,
"rewards/margins": 0.04949050396680832,
"rewards/rejected": -0.10685168206691742,
"sft_loss": 0.06228654831647873,
"step": 80
},
{
"epoch": 1.7777777777777777,
"grad_norm": 2.270496129989624,
"learning_rate": 4.903154239845798e-06,
"logits/chosen": -0.4178565442562103,
"logits/rejected": -0.48920226097106934,
"logps/chosen": -0.3355949819087982,
"logps/rejected": -0.902151882648468,
"loss": 0.3717,
"odds_ratio_loss": 3.370701551437378,
"rewards/accuracies": 0.8500000238418579,
"rewards/chosen": -0.03355949744582176,
"rewards/margins": 0.05665569379925728,
"rewards/rejected": -0.09021519124507904,
"sft_loss": 0.03466098755598068,
"step": 90
},
{
"epoch": 1.9753086419753085,
"grad_norm": 2.1452736854553223,
"learning_rate": 4.849231551964771e-06,
"logits/chosen": -0.39813241362571716,
"logits/rejected": -0.4744625985622406,
"logps/chosen": -0.22052116692066193,
"logps/rejected": -0.8172726631164551,
"loss": 0.2518,
"odds_ratio_loss": 2.2916011810302734,
"rewards/accuracies": 0.856249988079071,
"rewards/chosen": -0.02205212041735649,
"rewards/margins": 0.05967513844370842,
"rewards/rejected": -0.08172726631164551,
"sft_loss": 0.022620398551225662,
"step": 100
},
{
"epoch": 1.9753086419753085,
"eval_logits/chosen": -0.3780798316001892,
"eval_logits/rejected": -0.448334664106369,
"eval_logps/chosen": -0.18976576626300812,
"eval_logps/rejected": -0.7320783734321594,
"eval_loss": 0.23485496640205383,
"eval_odds_ratio_loss": 2.1322858333587646,
"eval_rewards/accuracies": 0.8111110925674438,
"eval_rewards/chosen": -0.018976576626300812,
"eval_rewards/margins": 0.05423126742243767,
"eval_rewards/rejected": -0.07320784032344818,
"eval_runtime": 3.7439,
"eval_samples_per_second": 24.039,
"eval_sft_loss": 0.02162640169262886,
"eval_steps_per_second": 12.02,
"step": 100
},
{
"epoch": 2.1728395061728394,
"grad_norm": 2.2345616817474365,
"learning_rate": 4.783863644106502e-06,
"logits/chosen": -0.43218794465065,
"logits/rejected": -0.4891538619995117,
"logps/chosen": -0.18616971373558044,
"logps/rejected": -0.6744140386581421,
"loss": 0.2306,
"odds_ratio_loss": 2.0546927452087402,
"rewards/accuracies": 0.793749988079071,
"rewards/chosen": -0.018616970628499985,
"rewards/margins": 0.048824433237314224,
"rewards/rejected": -0.06744140386581421,
"sft_loss": 0.025166219100356102,
"step": 110
},
{
"epoch": 2.3703703703703702,
"grad_norm": 1.977693796157837,
"learning_rate": 4.707368982147318e-06,
"logits/chosen": -0.4379637837409973,
"logits/rejected": -0.5003925561904907,
"logps/chosen": -0.11895088851451874,
"logps/rejected": -0.6111856698989868,
"loss": 0.1586,
"odds_ratio_loss": 1.449549674987793,
"rewards/accuracies": 0.831250011920929,
"rewards/chosen": -0.011895090341567993,
"rewards/margins": 0.04922347515821457,
"rewards/rejected": -0.06111856549978256,
"sft_loss": 0.01368994265794754,
"step": 120
},
{
"epoch": 2.567901234567901,
"grad_norm": 0.721815288066864,
"learning_rate": 4.620120240391065e-06,
"logits/chosen": -0.3538777828216553,
"logits/rejected": -0.4279879927635193,
"logps/chosen": -0.09486258774995804,
"logps/rejected": -0.7161315679550171,
"loss": 0.1321,
"odds_ratio_loss": 1.2123935222625732,
"rewards/accuracies": 0.84375,
"rewards/chosen": -0.009486258029937744,
"rewards/margins": 0.062126897275447845,
"rewards/rejected": -0.07161315530538559,
"sft_loss": 0.010894850827753544,
"step": 130
},
{
"epoch": 2.765432098765432,
"grad_norm": 1.1547597646713257,
"learning_rate": 4.522542485937369e-06,
"logits/chosen": -0.38979417085647583,
"logits/rejected": -0.45057815313339233,
"logps/chosen": -0.0999944657087326,
"logps/rejected": -0.7246894836425781,
"loss": 0.1326,
"odds_ratio_loss": 1.1918091773986816,
"rewards/accuracies": 0.84375,
"rewards/chosen": -0.009999445639550686,
"rewards/margins": 0.06246950477361679,
"rewards/rejected": -0.07246895134449005,
"sft_loss": 0.013394972309470177,
"step": 140
},
{
"epoch": 2.962962962962963,
"grad_norm": 1.2962913513183594,
"learning_rate": 4.415111107797445e-06,
"logits/chosen": -0.3816613554954529,
"logits/rejected": -0.43519797921180725,
"logps/chosen": -0.09587013721466064,
"logps/rejected": -0.5622029304504395,
"loss": 0.1304,
"odds_ratio_loss": 1.182822823524475,
"rewards/accuracies": 0.824999988079071,
"rewards/chosen": -0.009587015025317669,
"rewards/margins": 0.0466332733631134,
"rewards/rejected": -0.05622028559446335,
"sft_loss": 0.012101124040782452,
"step": 150
},
{
"epoch": 2.962962962962963,
"eval_logits/chosen": -0.345350980758667,
"eval_logits/rejected": -0.4031663239002228,
"eval_logps/chosen": -0.10941138863563538,
"eval_logps/rejected": -0.6116734147071838,
"eval_loss": 0.15299223363399506,
"eval_odds_ratio_loss": 1.3987985849380493,
"eval_rewards/accuracies": 0.8111110925674438,
"eval_rewards/chosen": -0.010941138491034508,
"eval_rewards/margins": 0.050226207822561264,
"eval_rewards/rejected": -0.061167340725660324,
"eval_runtime": 3.741,
"eval_samples_per_second": 24.058,
"eval_sft_loss": 0.013112363405525684,
"eval_steps_per_second": 12.029,
"step": 150
},
{
"epoch": 3.1604938271604937,
"grad_norm": 1.6140797138214111,
"learning_rate": 4.2983495008466285e-06,
"logits/chosen": -0.3475257158279419,
"logits/rejected": -0.4172907769680023,
"logps/chosen": -0.0872744545340538,
"logps/rejected": -0.547534167766571,
"loss": 0.1283,
"odds_ratio_loss": 1.1653860807418823,
"rewards/accuracies": 0.7875000238418579,
"rewards/chosen": -0.008727445267140865,
"rewards/margins": 0.046025972813367844,
"rewards/rejected": -0.05475341156125069,
"sft_loss": 0.011779388412833214,
"step": 160
},
{
"epoch": 3.3580246913580245,
"grad_norm": 1.6337134838104248,
"learning_rate": 4.172826515897146e-06,
"logits/chosen": -0.3779260814189911,
"logits/rejected": -0.44266828894615173,
"logps/chosen": -0.08930256962776184,
"logps/rejected": -0.6787833571434021,
"loss": 0.1222,
"odds_ratio_loss": 1.113948106765747,
"rewards/accuracies": 0.856249988079071,
"rewards/chosen": -0.008930256590247154,
"rewards/margins": 0.058948077261447906,
"rewards/rejected": -0.06787833571434021,
"sft_loss": 0.01077614352107048,
"step": 170
},
{
"epoch": 3.5555555555555554,
"grad_norm": 1.329077959060669,
"learning_rate": 4.039153688314146e-06,
"logits/chosen": -0.3644777834415436,
"logits/rejected": -0.4096454977989197,
"logps/chosen": -0.07825501263141632,
"logps/rejected": -0.5899345874786377,
"loss": 0.1135,
"odds_ratio_loss": 1.0353591442108154,
"rewards/accuracies": 0.8374999761581421,
"rewards/chosen": -0.007825501263141632,
"rewards/margins": 0.05116795748472214,
"rewards/rejected": -0.05899345874786377,
"sft_loss": 0.009976235218346119,
"step": 180
},
{
"epoch": 3.753086419753086,
"grad_norm": 2.1350457668304443,
"learning_rate": 3.897982258676867e-06,
"logits/chosen": -0.342695415019989,
"logits/rejected": -0.41270384192466736,
"logps/chosen": -0.09099350869655609,
"logps/rejected": -0.6971117854118347,
"loss": 0.1184,
"odds_ratio_loss": 1.0633059740066528,
"rewards/accuracies": 0.856249988079071,
"rewards/chosen": -0.009099351242184639,
"rewards/margins": 0.06061183288693428,
"rewards/rejected": -0.06971118599176407,
"sft_loss": 0.012052880600094795,
"step": 190
},
{
"epoch": 3.950617283950617,
"grad_norm": 1.1367535591125488,
"learning_rate": 3.7500000000000005e-06,
"logits/chosen": -0.3970497250556946,
"logits/rejected": -0.4613776206970215,
"logps/chosen": -0.07991655170917511,
"logps/rejected": -0.6025404334068298,
"loss": 0.1129,
"odds_ratio_loss": 1.0367439985275269,
"rewards/accuracies": 0.824999988079071,
"rewards/chosen": -0.007991654798388481,
"rewards/margins": 0.05226239562034607,
"rewards/rejected": -0.0602540448307991,
"sft_loss": 0.009189181961119175,
"step": 200
},
{
"epoch": 3.950617283950617,
"eval_logits/chosen": -0.34801948070526123,
"eval_logits/rejected": -0.4030517041683197,
"eval_logps/chosen": -0.10840340703725815,
"eval_logps/rejected": -0.5819076299667358,
"eval_loss": 0.15150243043899536,
"eval_odds_ratio_loss": 1.3828411102294922,
"eval_rewards/accuracies": 0.8222222328186035,
"eval_rewards/chosen": -0.010840341448783875,
"eval_rewards/margins": 0.04735042154788971,
"eval_rewards/rejected": -0.058190762996673584,
"eval_runtime": 3.7401,
"eval_samples_per_second": 24.064,
"eval_sft_loss": 0.013218320906162262,
"eval_steps_per_second": 12.032,
"step": 200
},
{
"epoch": 4.148148148148148,
"grad_norm": 1.6960663795471191,
"learning_rate": 3.595927866972694e-06,
"logits/chosen": -0.3732069432735443,
"logits/rejected": -0.4440130293369293,
"logps/chosen": -0.06717614084482193,
"logps/rejected": -0.516498327255249,
"loss": 0.0942,
"odds_ratio_loss": 0.8479182124137878,
"rewards/accuracies": 0.8374999761581421,
"rewards/chosen": -0.006717613898217678,
"rewards/margins": 0.04493222385644913,
"rewards/rejected": -0.05164983123540878,
"sft_loss": 0.009402485564351082,
"step": 210
},
{
"epoch": 4.345679012345679,
"grad_norm": 0.9651530385017395,
"learning_rate": 3.436516483539781e-06,
"logits/chosen": -0.3677094578742981,
"logits/rejected": -0.4239886403083801,
"logps/chosen": -0.09318893402814865,
"logps/rejected": -0.6319376230239868,
"loss": 0.13,
"odds_ratio_loss": 1.189330816268921,
"rewards/accuracies": 0.831250011920929,
"rewards/chosen": -0.009318893775343895,
"rewards/margins": 0.053874872624874115,
"rewards/rejected": -0.06319376826286316,
"sft_loss": 0.011091579683125019,
"step": 220
},
{
"epoch": 4.54320987654321,
"grad_norm": 1.0926239490509033,
"learning_rate": 3.272542485937369e-06,
"logits/chosen": -0.41018205881118774,
"logits/rejected": -0.48241281509399414,
"logps/chosen": -0.06124984472990036,
"logps/rejected": -0.6271657347679138,
"loss": 0.0887,
"odds_ratio_loss": 0.8028218150138855,
"rewards/accuracies": 0.8500000238418579,
"rewards/chosen": -0.006124985404312611,
"rewards/margins": 0.056591589003801346,
"rewards/rejected": -0.06271658092737198,
"sft_loss": 0.008464094251394272,
"step": 230
},
{
"epoch": 4.7407407407407405,
"grad_norm": 1.5143941640853882,
"learning_rate": 3.1048047389991693e-06,
"logits/chosen": -0.36270713806152344,
"logits/rejected": -0.4348696172237396,
"logps/chosen": -0.05753596872091293,
"logps/rejected": -0.6398877501487732,
"loss": 0.0825,
"odds_ratio_loss": 0.7441199421882629,
"rewards/accuracies": 0.887499988079071,
"rewards/chosen": -0.005753597244620323,
"rewards/margins": 0.058235179632902145,
"rewards/rejected": -0.06398877501487732,
"sft_loss": 0.008074344135820866,
"step": 240
},
{
"epoch": 4.938271604938271,
"grad_norm": 1.6572201251983643,
"learning_rate": 2.9341204441673267e-06,
"logits/chosen": -0.3572860658168793,
"logits/rejected": -0.41670694947242737,
"logps/chosen": -0.09173234552145004,
"logps/rejected": -0.7026180028915405,
"loss": 0.1194,
"odds_ratio_loss": 1.0817253589630127,
"rewards/accuracies": 0.84375,
"rewards/chosen": -0.009173234924674034,
"rewards/margins": 0.061088573187589645,
"rewards/rejected": -0.07026181370019913,
"sft_loss": 0.011259505525231361,
"step": 250
},
{
"epoch": 4.938271604938271,
"eval_logits/chosen": -0.3416781425476074,
"eval_logits/rejected": -0.3981509208679199,
"eval_logps/chosen": -0.10881069302558899,
"eval_logps/rejected": -0.641668975353241,
"eval_loss": 0.15224343538284302,
"eval_odds_ratio_loss": 1.3890637159347534,
"eval_rewards/accuracies": 0.8222222328186035,
"eval_rewards/chosen": -0.01088106818497181,
"eval_rewards/margins": 0.053285837173461914,
"eval_rewards/rejected": -0.06416690349578857,
"eval_runtime": 3.7429,
"eval_samples_per_second": 24.045,
"eval_sft_loss": 0.01333706360310316,
"eval_steps_per_second": 12.023,
"step": 250
},
{
"epoch": 5.135802469135802,
"grad_norm": 0.6316437721252441,
"learning_rate": 2.761321158169134e-06,
"logits/chosen": -0.38472387194633484,
"logits/rejected": -0.43845945596694946,
"logps/chosen": -0.08237903565168381,
"logps/rejected": -0.6897364854812622,
"loss": 0.1098,
"odds_ratio_loss": 1.019060730934143,
"rewards/accuracies": 0.8812500238418579,
"rewards/chosen": -0.00823790393769741,
"rewards/margins": 0.06073574349284172,
"rewards/rejected": -0.06897365301847458,
"sft_loss": 0.007918295450508595,
"step": 260
},
{
"epoch": 5.333333333333333,
"grad_norm": 2.247709035873413,
"learning_rate": 2.587248741756253e-06,
"logits/chosen": -0.33236485719680786,
"logits/rejected": -0.41399604082107544,
"logps/chosen": -0.06683196127414703,
"logps/rejected": -0.745326578617096,
"loss": 0.0981,
"odds_ratio_loss": 0.897613525390625,
"rewards/accuracies": 0.856249988079071,
"rewards/chosen": -0.0066831959411501884,
"rewards/margins": 0.06784946471452713,
"rewards/rejected": -0.07453266531229019,
"sft_loss": 0.008373981341719627,
"step": 270
},
{
"epoch": 5.530864197530864,
"grad_norm": 3.0054218769073486,
"learning_rate": 2.4127512582437486e-06,
"logits/chosen": -0.38011887669563293,
"logits/rejected": -0.4398444592952728,
"logps/chosen": -0.07033652812242508,
"logps/rejected": -0.6755497455596924,
"loss": 0.0967,
"odds_ratio_loss": 0.8789058923721313,
"rewards/accuracies": 0.8812500238418579,
"rewards/chosen": -0.007033653557300568,
"rewards/margins": 0.06052132323384285,
"rewards/rejected": -0.06755497306585312,
"sft_loss": 0.008822395466268063,
"step": 280
},
{
"epoch": 5.728395061728395,
"grad_norm": 1.2473162412643433,
"learning_rate": 2.238678841830867e-06,
"logits/chosen": -0.35760438442230225,
"logits/rejected": -0.432979017496109,
"logps/chosen": -0.05806203559041023,
"logps/rejected": -0.6580389142036438,
"loss": 0.0822,
"odds_ratio_loss": 0.7390257716178894,
"rewards/accuracies": 0.8687499761581421,
"rewards/chosen": -0.005806203465908766,
"rewards/margins": 0.05999768525362015,
"rewards/rejected": -0.0658038854598999,
"sft_loss": 0.008302577771246433,
"step": 290
},
{
"epoch": 5.925925925925926,
"grad_norm": 1.7050279378890991,
"learning_rate": 2.0658795558326745e-06,
"logits/chosen": -0.39829307794570923,
"logits/rejected": -0.4691620469093323,
"logps/chosen": -0.06409647315740585,
"logps/rejected": -0.6852135062217712,
"loss": 0.0898,
"odds_ratio_loss": 0.8166918754577637,
"rewards/accuracies": 0.862500011920929,
"rewards/chosen": -0.00640964787453413,
"rewards/margins": 0.062111712992191315,
"rewards/rejected": -0.06852135807275772,
"sft_loss": 0.00808151438832283,
"step": 300
},
{
"epoch": 5.925925925925926,
"eval_logits/chosen": -0.3401709198951721,
"eval_logits/rejected": -0.39598190784454346,
"eval_logps/chosen": -0.11005988717079163,
"eval_logps/rejected": -0.6839298605918884,
"eval_loss": 0.153494730591774,
"eval_odds_ratio_loss": 1.3989084959030151,
"eval_rewards/accuracies": 0.8111110925674438,
"eval_rewards/chosen": -0.011005990207195282,
"eval_rewards/margins": 0.05738700181245804,
"eval_rewards/rejected": -0.06839299947023392,
"eval_runtime": 3.7412,
"eval_samples_per_second": 24.056,
"eval_sft_loss": 0.0136038763448596,
"eval_steps_per_second": 12.028,
"step": 300
},
{
"epoch": 6.1234567901234565,
"grad_norm": 0.6541029214859009,
"learning_rate": 1.895195261000831e-06,
"logits/chosen": -0.3678016662597656,
"logits/rejected": -0.4242876470088959,
"logps/chosen": -0.07081776112318039,
"logps/rejected": -0.746510922908783,
"loss": 0.0894,
"odds_ratio_loss": 0.7998540997505188,
"rewards/accuracies": 0.9125000238418579,
"rewards/chosen": -0.007081775926053524,
"rewards/margins": 0.0675693154335022,
"rewards/rejected": -0.0746510922908783,
"sft_loss": 0.009410968981683254,
"step": 310
},
{
"epoch": 6.320987654320987,
"grad_norm": 1.3137905597686768,
"learning_rate": 1.7274575140626318e-06,
"logits/chosen": -0.38049232959747314,
"logits/rejected": -0.4344969391822815,
"logps/chosen": -0.06233459711074829,
"logps/rejected": -0.6675801277160645,
"loss": 0.0862,
"odds_ratio_loss": 0.7855814695358276,
"rewards/accuracies": 0.887499988079071,
"rewards/chosen": -0.006233459338545799,
"rewards/margins": 0.060524553060531616,
"rewards/rejected": -0.06675801426172256,
"sft_loss": 0.007671413477510214,
"step": 320
},
{
"epoch": 6.518518518518518,
"grad_norm": 1.7016428709030151,
"learning_rate": 1.56348351646022e-06,
"logits/chosen": -0.35415560007095337,
"logits/rejected": -0.4285738468170166,
"logps/chosen": -0.06706033647060394,
"logps/rejected": -0.7278292775154114,
"loss": 0.0952,
"odds_ratio_loss": 0.8702393770217896,
"rewards/accuracies": 0.8687499761581421,
"rewards/chosen": -0.0067060342989861965,
"rewards/margins": 0.06607688963413239,
"rewards/rejected": -0.07278293371200562,
"sft_loss": 0.008213361725211143,
"step": 330
},
{
"epoch": 6.716049382716049,
"grad_norm": 1.7326641082763672,
"learning_rate": 1.4040721330273063e-06,
"logits/chosen": -0.355478435754776,
"logits/rejected": -0.42061376571655273,
"logps/chosen": -0.057823676615953445,
"logps/rejected": -0.6569600105285645,
"loss": 0.0825,
"odds_ratio_loss": 0.7589547634124756,
"rewards/accuracies": 0.8812500238418579,
"rewards/chosen": -0.005782367195934057,
"rewards/margins": 0.05991363525390625,
"rewards/rejected": -0.06569599360227585,
"sft_loss": 0.006640643812716007,
"step": 340
},
{
"epoch": 6.91358024691358,
"grad_norm": 2.150254011154175,
"learning_rate": 1.2500000000000007e-06,
"logits/chosen": -0.3611661493778229,
"logits/rejected": -0.43402963876724243,
"logps/chosen": -0.06616374105215073,
"logps/rejected": -0.7352410554885864,
"loss": 0.0928,
"odds_ratio_loss": 0.8426556587219238,
"rewards/accuracies": 0.856249988079071,
"rewards/chosen": -0.00661637494340539,
"rewards/margins": 0.06690772622823715,
"rewards/rejected": -0.0735241025686264,
"sft_loss": 0.00857949536293745,
"step": 350
},
{
"epoch": 6.91358024691358,
"eval_logits/chosen": -0.33941933512687683,
"eval_logits/rejected": -0.39487287402153015,
"eval_logps/chosen": -0.11251324415206909,
"eval_logps/rejected": -0.679356038570404,
"eval_loss": 0.1572149693965912,
"eval_odds_ratio_loss": 1.431775450706482,
"eval_rewards/accuracies": 0.7888888716697693,
"eval_rewards/chosen": -0.011251323856413364,
"eval_rewards/margins": 0.05668427795171738,
"eval_rewards/rejected": -0.06793560087680817,
"eval_runtime": 3.7421,
"eval_samples_per_second": 24.05,
"eval_sft_loss": 0.014037418179214,
"eval_steps_per_second": 12.025,
"step": 350
},
{
"epoch": 7.111111111111111,
"grad_norm": 2.0509181022644043,
"learning_rate": 1.1020177413231334e-06,
"logits/chosen": -0.3650658130645752,
"logits/rejected": -0.4295966625213623,
"logps/chosen": -0.057928264141082764,
"logps/rejected": -0.6719712615013123,
"loss": 0.084,
"odds_ratio_loss": 0.771135687828064,
"rewards/accuracies": 0.875,
"rewards/chosen": -0.005792826414108276,
"rewards/margins": 0.061404310166835785,
"rewards/rejected": -0.06719712913036346,
"sft_loss": 0.006877593696117401,
"step": 360
},
{
"epoch": 7.308641975308642,
"grad_norm": 3.04136323928833,
"learning_rate": 9.608463116858544e-07,
"logits/chosen": -0.4004891812801361,
"logits/rejected": -0.45741385221481323,
"logps/chosen": -0.06307969242334366,
"logps/rejected": -0.67022705078125,
"loss": 0.0872,
"odds_ratio_loss": 0.7940786480903625,
"rewards/accuracies": 0.8999999761581421,
"rewards/chosen": -0.006307969335466623,
"rewards/margins": 0.060714732855558395,
"rewards/rejected": -0.06702270358800888,
"sft_loss": 0.00781711470335722,
"step": 370
},
{
"epoch": 7.506172839506172,
"grad_norm": 1.3740649223327637,
"learning_rate": 8.271734841028553e-07,
"logits/chosen": -0.3516360819339752,
"logits/rejected": -0.4145506024360657,
"logps/chosen": -0.06131374090909958,
"logps/rejected": -0.8121838569641113,
"loss": 0.084,
"odds_ratio_loss": 0.779072642326355,
"rewards/accuracies": 0.893750011920929,
"rewards/chosen": -0.006131374277174473,
"rewards/margins": 0.07508701831102371,
"rewards/rejected": -0.08121839910745621,
"sft_loss": 0.006131037604063749,
"step": 380
},
{
"epoch": 7.703703703703704,
"grad_norm": 1.9797685146331787,
"learning_rate": 7.016504991533727e-07,
"logits/chosen": -0.37762802839279175,
"logits/rejected": -0.4341215491294861,
"logps/chosen": -0.06214147061109543,
"logps/rejected": -0.7485532164573669,
"loss": 0.0882,
"odds_ratio_loss": 0.8045924305915833,
"rewards/accuracies": 0.8687499761581421,
"rewards/chosen": -0.006214146967977285,
"rewards/margins": 0.06864118576049805,
"rewards/rejected": -0.07485532760620117,
"sft_loss": 0.007714971899986267,
"step": 390
},
{
"epoch": 7.901234567901234,
"grad_norm": 1.4376161098480225,
"learning_rate": 5.848888922025553e-07,
"logits/chosen": -0.3702755868434906,
"logits/rejected": -0.43015074729919434,
"logps/chosen": -0.06369648873806,
"logps/rejected": -0.6736232042312622,
"loss": 0.0855,
"odds_ratio_loss": 0.7752313613891602,
"rewards/accuracies": 0.893750011920929,
"rewards/chosen": -0.0063696494325995445,
"rewards/margins": 0.06099267676472664,
"rewards/rejected": -0.06736232340335846,
"sft_loss": 0.007961621508002281,
"step": 400
},
{
"epoch": 7.901234567901234,
"eval_logits/chosen": -0.3374575674533844,
"eval_logits/rejected": -0.39353418350219727,
"eval_logps/chosen": -0.11246351897716522,
"eval_logps/rejected": -0.7215471267700195,
"eval_loss": 0.1577637493610382,
"eval_odds_ratio_loss": 1.4393588304519653,
"eval_rewards/accuracies": 0.800000011920929,
"eval_rewards/chosen": -0.011246351525187492,
"eval_rewards/margins": 0.060908351093530655,
"eval_rewards/rejected": -0.072154700756073,
"eval_runtime": 3.7403,
"eval_samples_per_second": 24.062,
"eval_sft_loss": 0.013827885501086712,
"eval_steps_per_second": 12.031,
"step": 400
},
{
"epoch": 8.098765432098766,
"grad_norm": 0.5824891328811646,
"learning_rate": 4.774575140626317e-07,
"logits/chosen": -0.3890485167503357,
"logits/rejected": -0.4562603533267975,
"logps/chosen": -0.052860356867313385,
"logps/rejected": -0.8017857670783997,
"loss": 0.0769,
"odds_ratio_loss": 0.7033621072769165,
"rewards/accuracies": 0.8812500238418579,
"rewards/chosen": -0.005286035593599081,
"rewards/margins": 0.07489253580570221,
"rewards/rejected": -0.08017858117818832,
"sft_loss": 0.00657792529091239,
"step": 410
},
{
"epoch": 8.296296296296296,
"grad_norm": 2.3090314865112305,
"learning_rate": 3.798797596089351e-07,
"logits/chosen": -0.36521631479263306,
"logits/rejected": -0.43496161699295044,
"logps/chosen": -0.052145786583423615,
"logps/rejected": -0.6766245365142822,
"loss": 0.0738,
"odds_ratio_loss": 0.6751989722251892,
"rewards/accuracies": 0.8812500238418579,
"rewards/chosen": -0.0052145784720778465,
"rewards/margins": 0.062447868287563324,
"rewards/rejected": -0.06766244769096375,
"sft_loss": 0.0063272216357290745,
"step": 420
},
{
"epoch": 8.493827160493828,
"grad_norm": 3.1166369915008545,
"learning_rate": 2.9263101785268253e-07,
"logits/chosen": -0.3338077664375305,
"logits/rejected": -0.4001466631889343,
"logps/chosen": -0.05268881469964981,
"logps/rejected": -0.8022252321243286,
"loss": 0.0766,
"odds_ratio_loss": 0.6915222406387329,
"rewards/accuracies": 0.8812500238418579,
"rewards/chosen": -0.005268882028758526,
"rewards/margins": 0.07495363801717758,
"rewards/rejected": -0.08022252470254898,
"sft_loss": 0.007409657351672649,
"step": 430
},
{
"epoch": 8.691358024691358,
"grad_norm": 2.2428250312805176,
"learning_rate": 2.1613635589349756e-07,
"logits/chosen": -0.3932591676712036,
"logits/rejected": -0.4431510865688324,
"logps/chosen": -0.05874771624803543,
"logps/rejected": -0.8061612844467163,
"loss": 0.0783,
"odds_ratio_loss": 0.7208496928215027,
"rewards/accuracies": 0.925000011920929,
"rewards/chosen": -0.005874771624803543,
"rewards/margins": 0.07474134862422943,
"rewards/rejected": -0.08061611652374268,
"sft_loss": 0.006212843116372824,
"step": 440
},
{
"epoch": 8.88888888888889,
"grad_norm": 2.0012757778167725,
"learning_rate": 1.507684480352292e-07,
"logits/chosen": -0.3309887647628784,
"logits/rejected": -0.3963066637516022,
"logps/chosen": -0.07408946752548218,
"logps/rejected": -0.7123581171035767,
"loss": 0.0985,
"odds_ratio_loss": 0.9039106369018555,
"rewards/accuracies": 0.862500011920929,
"rewards/chosen": -0.007408946752548218,
"rewards/margins": 0.06382687389850616,
"rewards/rejected": -0.07123581320047379,
"sft_loss": 0.008154405280947685,
"step": 450
},
{
"epoch": 8.88888888888889,
"eval_logits/chosen": -0.33720430731773376,
"eval_logits/rejected": -0.3934166729450226,
"eval_logps/chosen": -0.11222676187753677,
"eval_logps/rejected": -0.7204592227935791,
"eval_loss": 0.1574162095785141,
"eval_odds_ratio_loss": 1.4357547760009766,
"eval_rewards/accuracies": 0.800000011920929,
"eval_rewards/chosen": -0.011222676374018192,
"eval_rewards/margins": 0.06082325428724289,
"eval_rewards/rejected": -0.07204592972993851,
"eval_runtime": 3.7588,
"eval_samples_per_second": 23.944,
"eval_sft_loss": 0.013840759173035622,
"eval_steps_per_second": 11.972,
"step": 450
},
{
"epoch": 9.08641975308642,
"grad_norm": 1.9788808822631836,
"learning_rate": 9.684576015420277e-08,
"logits/chosen": -0.3609127700328827,
"logits/rejected": -0.41957467794418335,
"logps/chosen": -0.05120778828859329,
"logps/rejected": -0.6972073316574097,
"loss": 0.0681,
"odds_ratio_loss": 0.6181837320327759,
"rewards/accuracies": 0.9312499761581421,
"rewards/chosen": -0.005120779387652874,
"rewards/margins": 0.06459995359182358,
"rewards/rejected": -0.06972073018550873,
"sft_loss": 0.006232709623873234,
"step": 460
},
{
"epoch": 9.283950617283951,
"grad_norm": 0.5987529158592224,
"learning_rate": 5.463099816548578e-08,
"logits/chosen": -0.3785928189754486,
"logits/rejected": -0.44030579924583435,
"logps/chosen": -0.06789538264274597,
"logps/rejected": -0.718836784362793,
"loss": 0.0919,
"odds_ratio_loss": 0.8372634649276733,
"rewards/accuracies": 0.8812500238418579,
"rewards/chosen": -0.006789539009332657,
"rewards/margins": 0.06509413570165634,
"rewards/rejected": -0.0718836709856987,
"sft_loss": 0.008134867064654827,
"step": 470
},
{
"epoch": 9.481481481481481,
"grad_norm": 2.4453930854797363,
"learning_rate": 2.4329828146074096e-08,
"logits/chosen": -0.3719639182090759,
"logits/rejected": -0.44119659066200256,
"logps/chosen": -0.06199900060892105,
"logps/rejected": -0.9121496081352234,
"loss": 0.0796,
"odds_ratio_loss": 0.7194961905479431,
"rewards/accuracies": 0.90625,
"rewards/chosen": -0.006199900526553392,
"rewards/margins": 0.08501505851745605,
"rewards/rejected": -0.09121496230363846,
"sft_loss": 0.0076431455090641975,
"step": 480
},
{
"epoch": 9.679012345679013,
"grad_norm": 2.6841046810150146,
"learning_rate": 6.089874350439507e-09,
"logits/chosen": -0.3508923053741455,
"logits/rejected": -0.410422146320343,
"logps/chosen": -0.050847820937633514,
"logps/rejected": -0.717828631401062,
"loss": 0.0729,
"odds_ratio_loss": 0.6642852425575256,
"rewards/accuracies": 0.90625,
"rewards/chosen": -0.005084782373160124,
"rewards/margins": 0.06669807434082031,
"rewards/rejected": -0.07178286463022232,
"sft_loss": 0.006511971354484558,
"step": 490
},
{
"epoch": 9.876543209876543,
"grad_norm": 1.9767255783081055,
"learning_rate": 0.0,
"logits/chosen": -0.3572847545146942,
"logits/rejected": -0.43199190497398376,
"logps/chosen": -0.0603826642036438,
"logps/rejected": -0.6694537401199341,
"loss": 0.0859,
"odds_ratio_loss": 0.7898350954055786,
"rewards/accuracies": 0.887499988079071,
"rewards/chosen": -0.006038266699761152,
"rewards/margins": 0.060907114297151566,
"rewards/rejected": -0.066945381462574,
"sft_loss": 0.006888846401125193,
"step": 500
},
{
"epoch": 9.876543209876543,
"eval_logits/chosen": -0.33729952573776245,
"eval_logits/rejected": -0.39373528957366943,
"eval_logps/chosen": -0.11289726942777634,
"eval_logps/rejected": -0.7238631248474121,
"eval_loss": 0.15817420184612274,
"eval_odds_ratio_loss": 1.4418776035308838,
"eval_rewards/accuracies": 0.7888888716697693,
"eval_rewards/chosen": -0.011289726011455059,
"eval_rewards/margins": 0.061096593737602234,
"eval_rewards/rejected": -0.07238632440567017,
"eval_runtime": 3.7425,
"eval_samples_per_second": 24.048,
"eval_sft_loss": 0.013986458070576191,
"eval_steps_per_second": 12.024,
"step": 500
},
{
"epoch": 9.876543209876543,
"step": 500,
"total_flos": 8.823432944104243e+16,
"train_loss": 0.3188942909240723,
"train_runtime": 1116.7498,
"train_samples_per_second": 7.253,
"train_steps_per_second": 0.448
}
],
"logging_steps": 10,
"max_steps": 500,
"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": 8.823432944104243e+16,
"train_batch_size": 2,
"trial_name": null,
"trial_params": null
}