ElEmperador / trainer_state.json
AINovice2005's picture
Add Training Data
db32ac2 verified
{
"best_metric": null,
"best_model_checkpoint": null,
"epoch": 4.8,
"eval_steps": 12,
"global_step": 60,
"is_hyper_param_search": false,
"is_local_process_zero": true,
"is_world_process_zero": true,
"log_history": [
{
"epoch": 0.08,
"grad_norm": 5.910620212554932,
"learning_rate": 8e-07,
"log_odds_chosen": -0.012692844495177269,
"log_odds_ratio": -0.7021465301513672,
"logits/chosen": -2.7243709564208984,
"logits/rejected": -2.7299304008483887,
"logps/chosen": -1.1940056085586548,
"logps/rejected": -1.1935062408447266,
"loss": 1.4594,
"nll_loss": 1.3891654014587402,
"rewards/accuracies": 0.25,
"rewards/chosen": -0.11940056085586548,
"rewards/margins": -4.9931928515434265e-05,
"rewards/rejected": -0.11935063451528549,
"step": 1
},
{
"epoch": 0.16,
"grad_norm": 5.268943786621094,
"learning_rate": 1.6e-06,
"log_odds_chosen": 0.22696708142757416,
"log_odds_ratio": -0.643199622631073,
"logits/chosen": -2.8990988731384277,
"logits/rejected": -2.82047700881958,
"logps/chosen": -1.2680704593658447,
"logps/rejected": -1.4174962043762207,
"loss": 1.6127,
"nll_loss": 1.5484061241149902,
"rewards/accuracies": 0.75,
"rewards/chosen": -0.1268070489168167,
"rewards/margins": 0.014942573383450508,
"rewards/rejected": -0.14174960553646088,
"step": 2
},
{
"epoch": 0.24,
"grad_norm": 4.226258277893066,
"learning_rate": 2.4e-06,
"log_odds_chosen": 0.044683486223220825,
"log_odds_ratio": -0.7104766368865967,
"logits/chosen": -2.8773069381713867,
"logits/rejected": -2.9280641078948975,
"logps/chosen": -1.150111198425293,
"logps/rejected": -1.1700233221054077,
"loss": 1.6466,
"nll_loss": 1.575596809387207,
"rewards/accuracies": 0.25,
"rewards/chosen": -0.11501111835241318,
"rewards/margins": 0.001991217490285635,
"rewards/rejected": -0.11700233817100525,
"step": 3
},
{
"epoch": 0.32,
"grad_norm": 4.236293792724609,
"learning_rate": 3.2e-06,
"log_odds_chosen": 0.11046111583709717,
"log_odds_ratio": -0.6462637782096863,
"logits/chosen": -3.1065869331359863,
"logits/rejected": -2.97692608833313,
"logps/chosen": -1.0575246810913086,
"logps/rejected": -1.1389795541763306,
"loss": 1.3322,
"nll_loss": 1.2675998210906982,
"rewards/accuracies": 0.5,
"rewards/chosen": -0.10575246810913086,
"rewards/margins": 0.008145488798618317,
"rewards/rejected": -0.11389796435832977,
"step": 4
},
{
"epoch": 0.4,
"grad_norm": 4.288541316986084,
"learning_rate": 4e-06,
"log_odds_chosen": 0.2469336837530136,
"log_odds_ratio": -0.6131026148796082,
"logits/chosen": -3.0275461673736572,
"logits/rejected": -3.003474473953247,
"logps/chosen": -0.9929811954498291,
"logps/rejected": -1.1053597927093506,
"loss": 1.3442,
"nll_loss": 1.2829171419143677,
"rewards/accuracies": 0.75,
"rewards/chosen": -0.09929811954498291,
"rewards/margins": 0.011237863451242447,
"rewards/rejected": -0.11053598672151566,
"step": 5
},
{
"epoch": 0.48,
"grad_norm": 4.556058883666992,
"learning_rate": 4.8e-06,
"log_odds_chosen": 0.048641495406627655,
"log_odds_ratio": -0.691319465637207,
"logits/chosen": -2.9148268699645996,
"logits/rejected": -2.8536345958709717,
"logps/chosen": -1.3467779159545898,
"logps/rejected": -1.3747429847717285,
"loss": 1.4166,
"nll_loss": 1.3474537134170532,
"rewards/accuracies": 0.5,
"rewards/chosen": -0.13467779755592346,
"rewards/margins": 0.0027965055778622627,
"rewards/rejected": -0.13747431337833405,
"step": 6
},
{
"epoch": 0.56,
"grad_norm": 4.545846462249756,
"learning_rate": 5.6e-06,
"log_odds_chosen": 0.32562392950057983,
"log_odds_ratio": -0.5898037552833557,
"logits/chosen": -2.815234899520874,
"logits/rejected": -2.7621359825134277,
"logps/chosen": -1.061276912689209,
"logps/rejected": -1.2360599040985107,
"loss": 1.3822,
"nll_loss": 1.3232545852661133,
"rewards/accuracies": 0.625,
"rewards/chosen": -0.10612770169973373,
"rewards/margins": 0.017478302121162415,
"rewards/rejected": -0.12360600382089615,
"step": 7
},
{
"epoch": 0.64,
"grad_norm": 3.971618890762329,
"learning_rate": 6.4e-06,
"log_odds_chosen": 0.1728549599647522,
"log_odds_ratio": -0.6392964720726013,
"logits/chosen": -2.9990975856781006,
"logits/rejected": -2.9307050704956055,
"logps/chosen": -1.190705418586731,
"logps/rejected": -1.3564025163650513,
"loss": 1.5385,
"nll_loss": 1.4745732545852661,
"rewards/accuracies": 0.375,
"rewards/chosen": -0.11907054483890533,
"rewards/margins": 0.016569709405303,
"rewards/rejected": -0.1356402486562729,
"step": 8
},
{
"epoch": 0.72,
"grad_norm": 2.60530686378479,
"learning_rate": 7.2e-06,
"log_odds_chosen": 0.17511004209518433,
"log_odds_ratio": -0.63337641954422,
"logits/chosen": -3.026963233947754,
"logits/rejected": -2.9191253185272217,
"logps/chosen": -1.0979580879211426,
"logps/rejected": -1.2179211378097534,
"loss": 1.2909,
"nll_loss": 1.2275505065917969,
"rewards/accuracies": 0.625,
"rewards/chosen": -0.10979580879211426,
"rewards/margins": 0.011996296234428883,
"rewards/rejected": -0.12179210036993027,
"step": 9
},
{
"epoch": 0.8,
"grad_norm": 1.7484970092773438,
"learning_rate": 8e-06,
"log_odds_chosen": 0.2105366289615631,
"log_odds_ratio": -0.6329271793365479,
"logits/chosen": -2.628000259399414,
"logits/rejected": -2.4390053749084473,
"logps/chosen": -0.9455797076225281,
"logps/rejected": -1.1043422222137451,
"loss": 1.1688,
"nll_loss": 1.1055198907852173,
"rewards/accuracies": 0.375,
"rewards/chosen": -0.09455796331167221,
"rewards/margins": 0.015876244753599167,
"rewards/rejected": -0.11043421179056168,
"step": 10
},
{
"epoch": 0.88,
"grad_norm": 1.6517988443374634,
"learning_rate": 7.84e-06,
"log_odds_chosen": 0.2828138470649719,
"log_odds_ratio": -0.5968219041824341,
"logits/chosen": -3.0771772861480713,
"logits/rejected": -3.008608341217041,
"logps/chosen": -0.6074647903442383,
"logps/rejected": -0.7685144543647766,
"loss": 1.1134,
"nll_loss": 1.053712010383606,
"rewards/accuracies": 0.625,
"rewards/chosen": -0.060746483504772186,
"rewards/margins": 0.016104966402053833,
"rewards/rejected": -0.07685144990682602,
"step": 11
},
{
"epoch": 0.96,
"grad_norm": 2.5685229301452637,
"learning_rate": 7.68e-06,
"log_odds_chosen": 0.12323664128780365,
"log_odds_ratio": -0.6441441774368286,
"logits/chosen": -2.9566874504089355,
"logits/rejected": -2.8413052558898926,
"logps/chosen": -1.161773443222046,
"logps/rejected": -1.2560498714447021,
"loss": 1.3902,
"nll_loss": 1.3257489204406738,
"rewards/accuracies": 0.625,
"rewards/chosen": -0.11617735028266907,
"rewards/margins": 0.009427637793123722,
"rewards/rejected": -0.12560498714447021,
"step": 12
},
{
"epoch": 0.96,
"eval_log_odds_chosen": 0.18088191747665405,
"eval_log_odds_ratio": -0.6067904233932495,
"eval_logits/chosen": -3.32541561126709,
"eval_logits/rejected": -2.9764161109924316,
"eval_logps/chosen": -0.7066247463226318,
"eval_logps/rejected": -0.8023566603660583,
"eval_loss": 0.9977810382843018,
"eval_nll_loss": 0.9371020197868347,
"eval_rewards/accuracies": 1.0,
"eval_rewards/chosen": -0.0706624761223793,
"eval_rewards/margins": 0.009573191404342651,
"eval_rewards/rejected": -0.08023566752672195,
"eval_runtime": 4.0974,
"eval_samples_per_second": 0.244,
"eval_steps_per_second": 0.244,
"step": 12
},
{
"epoch": 1.04,
"grad_norm": 2.8649556636810303,
"learning_rate": 7.519999999999999e-06,
"log_odds_chosen": 0.7436250448226929,
"log_odds_ratio": -0.42550715804100037,
"logits/chosen": -3.073974847793579,
"logits/rejected": -2.865658760070801,
"logps/chosen": -1.1165134906768799,
"logps/rejected": -1.6563222408294678,
"loss": 1.2329,
"nll_loss": 1.190321922302246,
"rewards/accuracies": 0.875,
"rewards/chosen": -0.11165135353803635,
"rewards/margins": 0.05398087203502655,
"rewards/rejected": -0.1656322181224823,
"step": 13
},
{
"epoch": 1.12,
"grad_norm": 1.9388059377670288,
"learning_rate": 7.36e-06,
"log_odds_chosen": 0.4244161546230316,
"log_odds_ratio": -0.5222975015640259,
"logits/chosen": -2.839306354522705,
"logits/rejected": -2.7786505222320557,
"logps/chosen": -0.7469608783721924,
"logps/rejected": -0.990905225276947,
"loss": 1.0928,
"nll_loss": 1.0406169891357422,
"rewards/accuracies": 0.75,
"rewards/chosen": -0.07469607889652252,
"rewards/margins": 0.02439444325864315,
"rewards/rejected": -0.09909053146839142,
"step": 14
},
{
"epoch": 1.2,
"grad_norm": 2.8289878368377686,
"learning_rate": 7.2e-06,
"log_odds_chosen": 0.02547745406627655,
"log_odds_ratio": -0.7126201391220093,
"logits/chosen": -2.8896398544311523,
"logits/rejected": -2.7489066123962402,
"logps/chosen": -1.080715537071228,
"logps/rejected": -1.107945442199707,
"loss": 1.4342,
"nll_loss": 1.3629157543182373,
"rewards/accuracies": 0.5,
"rewards/chosen": -0.10807155072689056,
"rewards/margins": 0.0027229972183704376,
"rewards/rejected": -0.1107945516705513,
"step": 15
},
{
"epoch": 1.28,
"grad_norm": 2.7624049186706543,
"learning_rate": 7.0399999999999995e-06,
"log_odds_chosen": 0.18877235054969788,
"log_odds_ratio": -0.6148291826248169,
"logits/chosen": -2.8152570724487305,
"logits/rejected": -2.7555055618286133,
"logps/chosen": -1.103430986404419,
"logps/rejected": -1.2451709508895874,
"loss": 1.3929,
"nll_loss": 1.3314636945724487,
"rewards/accuracies": 0.5,
"rewards/chosen": -0.1103430837392807,
"rewards/margins": 0.014174011535942554,
"rewards/rejected": -0.12451709806919098,
"step": 16
},
{
"epoch": 1.3599999999999999,
"grad_norm": 2.5873303413391113,
"learning_rate": 6.879999999999999e-06,
"log_odds_chosen": -0.10925386846065521,
"log_odds_ratio": -0.7573699355125427,
"logits/chosen": -3.056697368621826,
"logits/rejected": -3.0701730251312256,
"logps/chosen": -1.0829575061798096,
"logps/rejected": -1.009861946105957,
"loss": 1.3078,
"nll_loss": 1.2320821285247803,
"rewards/accuracies": 0.375,
"rewards/chosen": -0.1082957535982132,
"rewards/margins": -0.007309555076062679,
"rewards/rejected": -0.10098619759082794,
"step": 17
},
{
"epoch": 1.44,
"grad_norm": 1.7549221515655518,
"learning_rate": 6.719999999999999e-06,
"log_odds_chosen": 0.25229498744010925,
"log_odds_ratio": -0.605067253112793,
"logits/chosen": -3.0717437267303467,
"logits/rejected": -3.031426191329956,
"logps/chosen": -0.6300977468490601,
"logps/rejected": -0.7678447961807251,
"loss": 1.1493,
"nll_loss": 1.0887819528579712,
"rewards/accuracies": 0.5,
"rewards/chosen": -0.06300977617502213,
"rewards/margins": 0.013774698600172997,
"rewards/rejected": -0.07678447663784027,
"step": 18
},
{
"epoch": 1.52,
"grad_norm": 2.258375644683838,
"learning_rate": 6.559999999999999e-06,
"log_odds_chosen": 0.09016017615795135,
"log_odds_ratio": -0.6553211212158203,
"logits/chosen": -2.851407766342163,
"logits/rejected": -2.8655269145965576,
"logps/chosen": -1.025189995765686,
"logps/rejected": -1.0774447917938232,
"loss": 1.2653,
"nll_loss": 1.1997864246368408,
"rewards/accuracies": 0.5,
"rewards/chosen": -0.10251899808645248,
"rewards/margins": 0.005225480534136295,
"rewards/rejected": -0.10774447023868561,
"step": 19
},
{
"epoch": 1.6,
"grad_norm": 1.6248959302902222,
"learning_rate": 6.4e-06,
"log_odds_chosen": 0.22705766558647156,
"log_odds_ratio": -0.6247647404670715,
"logits/chosen": -2.6382391452789307,
"logits/rejected": -2.520052671432495,
"logps/chosen": -0.8028525710105896,
"logps/rejected": -0.9037848114967346,
"loss": 0.9198,
"nll_loss": 0.8573005199432373,
"rewards/accuracies": 0.375,
"rewards/chosen": -0.08028525859117508,
"rewards/margins": 0.010093222372233868,
"rewards/rejected": -0.09037847816944122,
"step": 20
},
{
"epoch": 1.6800000000000002,
"grad_norm": 2.189142942428589,
"learning_rate": 6.2399999999999995e-06,
"log_odds_chosen": 0.1090141013264656,
"log_odds_ratio": -0.6825957894325256,
"logits/chosen": -2.9855175018310547,
"logits/rejected": -2.871065855026245,
"logps/chosen": -0.9656553268432617,
"logps/rejected": -1.0403013229370117,
"loss": 1.2366,
"nll_loss": 1.1683895587921143,
"rewards/accuracies": 0.5,
"rewards/chosen": -0.09656553715467453,
"rewards/margins": 0.007464595139026642,
"rewards/rejected": -0.10403012484312057,
"step": 21
},
{
"epoch": 1.76,
"grad_norm": 2.530574083328247,
"learning_rate": 6.079999999999999e-06,
"log_odds_chosen": 0.11634103953838348,
"log_odds_ratio": -0.6487681269645691,
"logits/chosen": -3.062095880508423,
"logits/rejected": -2.9822239875793457,
"logps/chosen": -0.9827238321304321,
"logps/rejected": -1.0462595224380493,
"loss": 1.0995,
"nll_loss": 1.0346102714538574,
"rewards/accuracies": 0.625,
"rewards/chosen": -0.09827239066362381,
"rewards/margins": 0.006353557575494051,
"rewards/rejected": -0.10462595522403717,
"step": 22
},
{
"epoch": 1.8399999999999999,
"grad_norm": 1.5682123899459839,
"learning_rate": 5.92e-06,
"log_odds_chosen": 0.05784512683749199,
"log_odds_ratio": -0.686516284942627,
"logits/chosen": -3.078280210494995,
"logits/rejected": -3.0346872806549072,
"logps/chosen": -0.8539592623710632,
"logps/rejected": -0.9186102151870728,
"loss": 1.2162,
"nll_loss": 1.1475117206573486,
"rewards/accuracies": 0.375,
"rewards/chosen": -0.0853959321975708,
"rewards/margins": 0.006465096492320299,
"rewards/rejected": -0.09186102449893951,
"step": 23
},
{
"epoch": 1.92,
"grad_norm": 2.3258304595947266,
"learning_rate": 5.76e-06,
"log_odds_chosen": 0.18300886452198029,
"log_odds_ratio": -0.6389400959014893,
"logits/chosen": -3.1072592735290527,
"logits/rejected": -2.9962942600250244,
"logps/chosen": -1.1144001483917236,
"logps/rejected": -1.2568085193634033,
"loss": 1.5199,
"nll_loss": 1.4559848308563232,
"rewards/accuracies": 0.75,
"rewards/chosen": -0.111440010368824,
"rewards/margins": 0.014240838587284088,
"rewards/rejected": -0.1256808340549469,
"step": 24
},
{
"epoch": 1.92,
"eval_log_odds_chosen": 0.09352552890777588,
"eval_log_odds_ratio": -0.6474775075912476,
"eval_logits/chosen": -3.3521740436553955,
"eval_logits/rejected": -3.045846700668335,
"eval_logps/chosen": -0.6791949272155762,
"eval_logps/rejected": -0.7263939380645752,
"eval_loss": 0.958454430103302,
"eval_nll_loss": 0.8937066793441772,
"eval_rewards/accuracies": 1.0,
"eval_rewards/chosen": -0.06791949272155762,
"eval_rewards/margins": 0.0047199055552482605,
"eval_rewards/rejected": -0.07263939827680588,
"eval_runtime": 3.9151,
"eval_samples_per_second": 0.255,
"eval_steps_per_second": 0.255,
"step": 24
},
{
"epoch": 2.0,
"grad_norm": 1.6808022260665894,
"learning_rate": 5.6e-06,
"log_odds_chosen": -0.04988832771778107,
"log_odds_ratio": -0.7225002646446228,
"logits/chosen": -3.167830467224121,
"logits/rejected": -3.061304807662964,
"logps/chosen": -0.8571638464927673,
"logps/rejected": -0.827689528465271,
"loss": 1.2537,
"nll_loss": 1.1814517974853516,
"rewards/accuracies": 0.375,
"rewards/chosen": -0.08571638911962509,
"rewards/margins": -0.002947437111288309,
"rewards/rejected": -0.08276894688606262,
"step": 25
},
{
"epoch": 2.08,
"grad_norm": 2.2546420097351074,
"learning_rate": 5.4400000000000004e-06,
"log_odds_chosen": 0.04561331868171692,
"log_odds_ratio": -0.7332349419593811,
"logits/chosen": -2.59240460395813,
"logits/rejected": -2.582451343536377,
"logps/chosen": -0.9324518442153931,
"logps/rejected": -0.953067421913147,
"loss": 1.1664,
"nll_loss": 1.0930918455123901,
"rewards/accuracies": 0.5,
"rewards/chosen": -0.09324519336223602,
"rewards/margins": 0.0020615560933947563,
"rewards/rejected": -0.09530673921108246,
"step": 26
},
{
"epoch": 2.16,
"grad_norm": 2.0570733547210693,
"learning_rate": 5.28e-06,
"log_odds_chosen": 0.24187083542346954,
"log_odds_ratio": -0.6241340637207031,
"logits/chosen": -3.080639600753784,
"logits/rejected": -3.0304229259490967,
"logps/chosen": -0.7420560121536255,
"logps/rejected": -0.8350604176521301,
"loss": 1.0571,
"nll_loss": 0.9946718811988831,
"rewards/accuracies": 0.5,
"rewards/chosen": -0.07420559972524643,
"rewards/margins": 0.009300438687205315,
"rewards/rejected": -0.0835060402750969,
"step": 27
},
{
"epoch": 2.24,
"grad_norm": 2.023972511291504,
"learning_rate": 5.12e-06,
"log_odds_chosen": 0.19169293344020844,
"log_odds_ratio": -0.6201201677322388,
"logits/chosen": -2.8682751655578613,
"logits/rejected": -2.872384548187256,
"logps/chosen": -0.8573394417762756,
"logps/rejected": -1.0006366968154907,
"loss": 1.1392,
"nll_loss": 1.077183723449707,
"rewards/accuracies": 0.375,
"rewards/chosen": -0.08573393523693085,
"rewards/margins": 0.014329724945127964,
"rewards/rejected": -0.10006366670131683,
"step": 28
},
{
"epoch": 2.32,
"grad_norm": 2.2322771549224854,
"learning_rate": 4.96e-06,
"log_odds_chosen": 0.22571854293346405,
"log_odds_ratio": -0.6048489212989807,
"logits/chosen": -2.7974281311035156,
"logits/rejected": -2.671107292175293,
"logps/chosen": -0.9511339068412781,
"logps/rejected": -1.108659267425537,
"loss": 1.2377,
"nll_loss": 1.1771889925003052,
"rewards/accuracies": 0.625,
"rewards/chosen": -0.09511338919401169,
"rewards/margins": 0.015752539038658142,
"rewards/rejected": -0.11086593568325043,
"step": 29
},
{
"epoch": 2.4,
"grad_norm": 1.5203155279159546,
"learning_rate": 4.8e-06,
"log_odds_chosen": 0.08203547447919846,
"log_odds_ratio": -0.6810359358787537,
"logits/chosen": -3.129624366760254,
"logits/rejected": -3.040184497833252,
"logps/chosen": -1.0037201642990112,
"logps/rejected": -1.065619945526123,
"loss": 1.2065,
"nll_loss": 1.1384248733520508,
"rewards/accuracies": 0.5,
"rewards/chosen": -0.10037201642990112,
"rewards/margins": 0.006189983803778887,
"rewards/rejected": -0.10656200349330902,
"step": 30
},
{
"epoch": 2.48,
"grad_norm": 1.4941612482070923,
"learning_rate": 4.64e-06,
"log_odds_chosen": 0.3264733850955963,
"log_odds_ratio": -0.5566755533218384,
"logits/chosen": -2.98291015625,
"logits/rejected": -2.946389675140381,
"logps/chosen": -0.6631190776824951,
"logps/rejected": -0.8497029542922974,
"loss": 0.9711,
"nll_loss": 0.915428876876831,
"rewards/accuracies": 0.625,
"rewards/chosen": -0.06631191074848175,
"rewards/margins": 0.018658379092812538,
"rewards/rejected": -0.08497028797864914,
"step": 31
},
{
"epoch": 2.56,
"grad_norm": 2.090597629547119,
"learning_rate": 4.48e-06,
"log_odds_chosen": 0.07261727005243301,
"log_odds_ratio": -0.684439480304718,
"logits/chosen": -3.051530361175537,
"logits/rejected": -3.0167644023895264,
"logps/chosen": -0.861786425113678,
"logps/rejected": -0.8696616291999817,
"loss": 1.0085,
"nll_loss": 0.9400495290756226,
"rewards/accuracies": 0.5,
"rewards/chosen": -0.08617864549160004,
"rewards/margins": 0.0007875119335949421,
"rewards/rejected": -0.08696617186069489,
"step": 32
},
{
"epoch": 2.64,
"grad_norm": 1.5811887979507446,
"learning_rate": 4.32e-06,
"log_odds_chosen": 0.12410806119441986,
"log_odds_ratio": -0.6687257289886475,
"logits/chosen": -2.9889068603515625,
"logits/rejected": -2.7933883666992188,
"logps/chosen": -1.034374713897705,
"logps/rejected": -1.1076936721801758,
"loss": 1.2432,
"nll_loss": 1.176369071006775,
"rewards/accuracies": 0.5,
"rewards/chosen": -0.10343748331069946,
"rewards/margins": 0.007331886328756809,
"rewards/rejected": -0.1107693687081337,
"step": 33
},
{
"epoch": 2.7199999999999998,
"grad_norm": 2.2703254222869873,
"learning_rate": 4.16e-06,
"log_odds_chosen": 0.12302179634571075,
"log_odds_ratio": -0.6416699290275574,
"logits/chosen": -3.0783886909484863,
"logits/rejected": -3.0010714530944824,
"logps/chosen": -0.9351593255996704,
"logps/rejected": -1.035468339920044,
"loss": 1.1554,
"nll_loss": 1.0912635326385498,
"rewards/accuracies": 0.75,
"rewards/chosen": -0.09351593255996704,
"rewards/margins": 0.010030903853476048,
"rewards/rejected": -0.10354683548212051,
"step": 34
},
{
"epoch": 2.8,
"grad_norm": 2.956712484359741,
"learning_rate": 4e-06,
"log_odds_chosen": 0.0465146005153656,
"log_odds_ratio": -0.6841952800750732,
"logits/chosen": -3.121398448944092,
"logits/rejected": -3.0879604816436768,
"logps/chosen": -0.7368811964988708,
"logps/rejected": -0.7471375465393066,
"loss": 1.2526,
"nll_loss": 1.1842174530029297,
"rewards/accuracies": 0.625,
"rewards/chosen": -0.07368811964988708,
"rewards/margins": 0.0010256306268274784,
"rewards/rejected": -0.07471375167369843,
"step": 35
},
{
"epoch": 2.88,
"grad_norm": 1.5912374258041382,
"learning_rate": 3.84e-06,
"log_odds_chosen": 0.011004272848367691,
"log_odds_ratio": -0.6958502531051636,
"logits/chosen": -3.0698819160461426,
"logits/rejected": -3.0204977989196777,
"logps/chosen": -0.8866332769393921,
"logps/rejected": -0.892663836479187,
"loss": 1.1397,
"nll_loss": 1.0701532363891602,
"rewards/accuracies": 0.375,
"rewards/chosen": -0.08866333961486816,
"rewards/margins": 0.0006030518561601639,
"rewards/rejected": -0.08926638960838318,
"step": 36
},
{
"epoch": 2.88,
"eval_log_odds_chosen": 0.06313765048980713,
"eval_log_odds_ratio": -0.6620765328407288,
"eval_logits/chosen": -3.3640849590301514,
"eval_logits/rejected": -3.0697708129882812,
"eval_logps/chosen": -0.661213755607605,
"eval_logps/rejected": -0.6922562718391418,
"eval_loss": 0.9372102618217468,
"eval_nll_loss": 0.8710026144981384,
"eval_rewards/accuracies": 1.0,
"eval_rewards/chosen": -0.06612137705087662,
"eval_rewards/margins": 0.0031042546033859253,
"eval_rewards/rejected": -0.06922563165426254,
"eval_runtime": 4.1385,
"eval_samples_per_second": 0.242,
"eval_steps_per_second": 0.242,
"step": 36
},
{
"epoch": 2.96,
"grad_norm": 2.163959503173828,
"learning_rate": 3.68e-06,
"log_odds_chosen": 0.1411035656929016,
"log_odds_ratio": -0.6452722549438477,
"logits/chosen": -2.993727684020996,
"logits/rejected": -2.9780263900756836,
"logps/chosen": -0.691759467124939,
"logps/rejected": -0.7671821117401123,
"loss": 1.3539,
"nll_loss": 1.2893797159194946,
"rewards/accuracies": 0.625,
"rewards/chosen": -0.06917594373226166,
"rewards/margins": 0.007542273961007595,
"rewards/rejected": -0.07671821117401123,
"step": 37
},
{
"epoch": 3.04,
"grad_norm": 5.024839878082275,
"learning_rate": 3.5199999999999998e-06,
"log_odds_chosen": 0.24200591444969177,
"log_odds_ratio": -0.6079150438308716,
"logits/chosen": -3.187926769256592,
"logits/rejected": -3.077181816101074,
"logps/chosen": -1.1563198566436768,
"logps/rejected": -1.3247044086456299,
"loss": 1.3491,
"nll_loss": 1.2883188724517822,
"rewards/accuracies": 0.875,
"rewards/chosen": -0.11563197523355484,
"rewards/margins": 0.016838468611240387,
"rewards/rejected": -0.13247044384479523,
"step": 38
},
{
"epoch": 3.12,
"grad_norm": 1.4955006837844849,
"learning_rate": 3.3599999999999996e-06,
"log_odds_chosen": 0.12877097725868225,
"log_odds_ratio": -0.6657677292823792,
"logits/chosen": -2.92303466796875,
"logits/rejected": -2.85076642036438,
"logps/chosen": -0.8132851719856262,
"logps/rejected": -0.8765811920166016,
"loss": 1.188,
"nll_loss": 1.1214345693588257,
"rewards/accuracies": 0.5,
"rewards/chosen": -0.08132852613925934,
"rewards/margins": 0.006329594645649195,
"rewards/rejected": -0.0876581147313118,
"step": 39
},
{
"epoch": 3.2,
"grad_norm": 2.0386250019073486,
"learning_rate": 3.2e-06,
"log_odds_chosen": 0.4739803671836853,
"log_odds_ratio": -0.528855562210083,
"logits/chosen": -2.883831739425659,
"logits/rejected": -2.7613730430603027,
"logps/chosen": -0.5880856513977051,
"logps/rejected": -0.8737905025482178,
"loss": 1.0372,
"nll_loss": 0.9843271374702454,
"rewards/accuracies": 0.75,
"rewards/chosen": -0.05880856513977051,
"rewards/margins": 0.02857048809528351,
"rewards/rejected": -0.08737905323505402,
"step": 40
},
{
"epoch": 3.2800000000000002,
"grad_norm": 1.866861343383789,
"learning_rate": 3.0399999999999997e-06,
"log_odds_chosen": 0.175949364900589,
"log_odds_ratio": -0.6121445894241333,
"logits/chosen": -2.689525842666626,
"logits/rejected": -2.5920565128326416,
"logps/chosen": -0.8869262933731079,
"logps/rejected": -1.0022177696228027,
"loss": 1.0809,
"nll_loss": 1.0196477174758911,
"rewards/accuracies": 0.75,
"rewards/chosen": -0.08869262784719467,
"rewards/margins": 0.011529151350259781,
"rewards/rejected": -0.10022178292274475,
"step": 41
},
{
"epoch": 3.36,
"grad_norm": 1.7331321239471436,
"learning_rate": 2.88e-06,
"log_odds_chosen": 0.21247801184654236,
"log_odds_ratio": -0.6190555691719055,
"logits/chosen": -2.9846858978271484,
"logits/rejected": -3.0182762145996094,
"logps/chosen": -0.7801945209503174,
"logps/rejected": -0.904188871383667,
"loss": 1.1836,
"nll_loss": 1.1216562986373901,
"rewards/accuracies": 0.75,
"rewards/chosen": -0.07801946252584457,
"rewards/margins": 0.012399429455399513,
"rewards/rejected": -0.09041889011859894,
"step": 42
},
{
"epoch": 3.44,
"grad_norm": 1.5410895347595215,
"learning_rate": 2.7200000000000002e-06,
"log_odds_chosen": 0.16515614092350006,
"log_odds_ratio": -0.6415831446647644,
"logits/chosen": -3.103332996368408,
"logits/rejected": -3.0033347606658936,
"logps/chosen": -0.6259253025054932,
"logps/rejected": -0.6716697216033936,
"loss": 0.9397,
"nll_loss": 0.8755762577056885,
"rewards/accuracies": 0.375,
"rewards/chosen": -0.0625925362110138,
"rewards/margins": 0.004574435297399759,
"rewards/rejected": -0.06716696918010712,
"step": 43
},
{
"epoch": 3.52,
"grad_norm": 2.591630697250366,
"learning_rate": 2.56e-06,
"log_odds_chosen": 0.25676825642585754,
"log_odds_ratio": -0.6107382774353027,
"logits/chosen": -3.0657176971435547,
"logits/rejected": -2.972097873687744,
"logps/chosen": -0.8358463048934937,
"logps/rejected": -0.9837089776992798,
"loss": 0.9461,
"nll_loss": 0.8850574493408203,
"rewards/accuracies": 0.625,
"rewards/chosen": -0.08358462154865265,
"rewards/margins": 0.014786271378397942,
"rewards/rejected": -0.09837089478969574,
"step": 44
},
{
"epoch": 3.6,
"grad_norm": 3.252107858657837,
"learning_rate": 2.4e-06,
"log_odds_chosen": 0.2689153254032135,
"log_odds_ratio": -0.59128737449646,
"logits/chosen": -3.036353826522827,
"logits/rejected": -2.9656076431274414,
"logps/chosen": -0.8418894410133362,
"logps/rejected": -0.9865555167198181,
"loss": 1.1173,
"nll_loss": 1.0581891536712646,
"rewards/accuracies": 0.625,
"rewards/chosen": -0.08418895304203033,
"rewards/margins": 0.014466602355241776,
"rewards/rejected": -0.09865555167198181,
"step": 45
},
{
"epoch": 3.68,
"grad_norm": 2.0005674362182617,
"learning_rate": 2.24e-06,
"log_odds_chosen": 0.08924929052591324,
"log_odds_ratio": -0.6686105728149414,
"logits/chosen": -3.0594611167907715,
"logits/rejected": -3.006617307662964,
"logps/chosen": -0.9009665250778198,
"logps/rejected": -0.9710016250610352,
"loss": 1.2238,
"nll_loss": 1.1569262742996216,
"rewards/accuracies": 0.625,
"rewards/chosen": -0.09009665250778198,
"rewards/margins": 0.00700351083651185,
"rewards/rejected": -0.0971001610159874,
"step": 46
},
{
"epoch": 3.76,
"grad_norm": 1.9500302076339722,
"learning_rate": 2.08e-06,
"log_odds_chosen": -0.01365312933921814,
"log_odds_ratio": -0.7387136220932007,
"logits/chosen": -3.102898120880127,
"logits/rejected": -3.0505154132843018,
"logps/chosen": -0.9508839845657349,
"logps/rejected": -0.9444030523300171,
"loss": 1.2465,
"nll_loss": 1.1726138591766357,
"rewards/accuracies": 0.5,
"rewards/chosen": -0.0950884073972702,
"rewards/margins": -0.0006480957381427288,
"rewards/rejected": -0.09444031119346619,
"step": 47
},
{
"epoch": 3.84,
"grad_norm": 1.4190654754638672,
"learning_rate": 1.92e-06,
"log_odds_chosen": 0.19316454231739044,
"log_odds_ratio": -0.6173151731491089,
"logits/chosen": -3.0933785438537598,
"logits/rejected": -3.019509792327881,
"logps/chosen": -0.7028050422668457,
"logps/rejected": -0.811437726020813,
"loss": 1.1042,
"nll_loss": 1.0424754619598389,
"rewards/accuracies": 0.5,
"rewards/chosen": -0.07028050720691681,
"rewards/margins": 0.01086327526718378,
"rewards/rejected": -0.08114377409219742,
"step": 48
},
{
"epoch": 3.84,
"eval_log_odds_chosen": 0.036497294902801514,
"eval_log_odds_ratio": -0.6750649809837341,
"eval_logits/chosen": -3.3623147010803223,
"eval_logits/rejected": -3.080049991607666,
"eval_logps/chosen": -0.6492528319358826,
"eval_logps/rejected": -0.6668488383293152,
"eval_loss": 0.9257785677909851,
"eval_nll_loss": 0.8582720756530762,
"eval_rewards/accuracies": 1.0,
"eval_rewards/chosen": -0.06492528319358826,
"eval_rewards/margins": 0.0017596036195755005,
"eval_rewards/rejected": -0.06668488681316376,
"eval_runtime": 4.1545,
"eval_samples_per_second": 0.241,
"eval_steps_per_second": 0.241,
"step": 48
},
{
"epoch": 3.92,
"grad_norm": 2.1789004802703857,
"learning_rate": 1.7599999999999999e-06,
"log_odds_chosen": -0.04005882143974304,
"log_odds_ratio": -0.7334250211715698,
"logits/chosen": -2.9112021923065186,
"logits/rejected": -2.9076766967773438,
"logps/chosen": -0.765835165977478,
"logps/rejected": -0.7128950357437134,
"loss": 1.2473,
"nll_loss": 1.1739940643310547,
"rewards/accuracies": 0.375,
"rewards/chosen": -0.07658351957798004,
"rewards/margins": -0.005294016562402248,
"rewards/rejected": -0.07128950208425522,
"step": 49
},
{
"epoch": 4.0,
"grad_norm": 2.178314447402954,
"learning_rate": 1.6e-06,
"log_odds_chosen": -0.07708044350147247,
"log_odds_ratio": -0.7379593253135681,
"logits/chosen": -3.1038739681243896,
"logits/rejected": -3.0927176475524902,
"logps/chosen": -0.9102978706359863,
"logps/rejected": -0.8448383808135986,
"loss": 1.2398,
"nll_loss": 1.1660352945327759,
"rewards/accuracies": 0.375,
"rewards/chosen": -0.09102979302406311,
"rewards/margins": -0.006545952521264553,
"rewards/rejected": -0.08448383212089539,
"step": 50
},
{
"epoch": 4.08,
"grad_norm": 1.5347027778625488,
"learning_rate": 1.44e-06,
"log_odds_chosen": 0.10280649363994598,
"log_odds_ratio": -0.6480928063392639,
"logits/chosen": -2.852008581161499,
"logits/rejected": -2.8604021072387695,
"logps/chosen": -0.6744309067726135,
"logps/rejected": -0.7366931438446045,
"loss": 1.0266,
"nll_loss": 0.9618306756019592,
"rewards/accuracies": 0.5,
"rewards/chosen": -0.06744308769702911,
"rewards/margins": 0.00622622249647975,
"rewards/rejected": -0.07366931438446045,
"step": 51
},
{
"epoch": 4.16,
"grad_norm": 1.4483627080917358,
"learning_rate": 1.28e-06,
"log_odds_chosen": 0.07863382995128632,
"log_odds_ratio": -0.6696581840515137,
"logits/chosen": -3.0924391746520996,
"logits/rejected": -3.02586030960083,
"logps/chosen": -0.7803107500076294,
"logps/rejected": -0.8022903203964233,
"loss": 1.0676,
"nll_loss": 1.0006568431854248,
"rewards/accuracies": 0.375,
"rewards/chosen": -0.07803107798099518,
"rewards/margins": 0.002197958528995514,
"rewards/rejected": -0.08022903650999069,
"step": 52
},
{
"epoch": 4.24,
"grad_norm": 2.0634541511535645,
"learning_rate": 1.12e-06,
"log_odds_chosen": 0.4366196393966675,
"log_odds_ratio": -0.5555592775344849,
"logits/chosen": -2.8958981037139893,
"logits/rejected": -2.7654147148132324,
"logps/chosen": -0.9317148923873901,
"logps/rejected": -1.197984218597412,
"loss": 1.0649,
"nll_loss": 1.0093390941619873,
"rewards/accuracies": 0.5,
"rewards/chosen": -0.09317149221897125,
"rewards/margins": 0.026626940816640854,
"rewards/rejected": -0.1197984367609024,
"step": 53
},
{
"epoch": 4.32,
"grad_norm": 1.9966628551483154,
"learning_rate": 9.6e-07,
"log_odds_chosen": 0.22570586204528809,
"log_odds_ratio": -0.598256528377533,
"logits/chosen": -3.159971237182617,
"logits/rejected": -3.048675298690796,
"logps/chosen": -0.7773997187614441,
"logps/rejected": -0.9069120287895203,
"loss": 1.0975,
"nll_loss": 1.0376590490341187,
"rewards/accuracies": 0.625,
"rewards/chosen": -0.07773996889591217,
"rewards/margins": 0.012951226904988289,
"rewards/rejected": -0.09069120138883591,
"step": 54
},
{
"epoch": 4.4,
"grad_norm": 1.502306342124939,
"learning_rate": 8e-07,
"log_odds_chosen": 0.04895332455635071,
"log_odds_ratio": -0.7371360659599304,
"logits/chosen": -3.0577011108398438,
"logits/rejected": -2.9929468631744385,
"logps/chosen": -0.8072652816772461,
"logps/rejected": -0.8173755407333374,
"loss": 1.0562,
"nll_loss": 0.9824643135070801,
"rewards/accuracies": 0.5,
"rewards/chosen": -0.08072653412818909,
"rewards/margins": 0.0010110288858413696,
"rewards/rejected": -0.08173755556344986,
"step": 55
},
{
"epoch": 4.48,
"grad_norm": 1.947930097579956,
"learning_rate": 6.4e-07,
"log_odds_chosen": 0.16152703762054443,
"log_odds_ratio": -0.633980929851532,
"logits/chosen": -3.028554916381836,
"logits/rejected": -2.980391025543213,
"logps/chosen": -0.8234447836875916,
"logps/rejected": -0.9135499000549316,
"loss": 1.1579,
"nll_loss": 1.0945292711257935,
"rewards/accuracies": 0.75,
"rewards/chosen": -0.08234447985887527,
"rewards/margins": 0.009010514244437218,
"rewards/rejected": -0.09135499596595764,
"step": 56
},
{
"epoch": 4.5600000000000005,
"grad_norm": 1.6086125373840332,
"learning_rate": 4.8e-07,
"log_odds_chosen": -0.02157880738377571,
"log_odds_ratio": -0.7184982895851135,
"logits/chosen": -2.8279547691345215,
"logits/rejected": -2.8562469482421875,
"logps/chosen": -0.8981537222862244,
"logps/rejected": -0.8426607847213745,
"loss": 1.1922,
"nll_loss": 1.1203769445419312,
"rewards/accuracies": 0.625,
"rewards/chosen": -0.08981536328792572,
"rewards/margins": -0.005549294408410788,
"rewards/rejected": -0.08426607400178909,
"step": 57
},
{
"epoch": 4.64,
"grad_norm": 2.1013498306274414,
"learning_rate": 3.2e-07,
"log_odds_chosen": 0.09676241874694824,
"log_odds_ratio": -0.6955091953277588,
"logits/chosen": -2.927985668182373,
"logits/rejected": -2.8649566173553467,
"logps/chosen": -0.7449172735214233,
"logps/rejected": -0.7649428844451904,
"loss": 1.1142,
"nll_loss": 1.0446666479110718,
"rewards/accuracies": 0.5,
"rewards/chosen": -0.07449173182249069,
"rewards/margins": 0.0020025642588734627,
"rewards/rejected": -0.07649429142475128,
"step": 58
},
{
"epoch": 4.72,
"grad_norm": 1.8846663236618042,
"learning_rate": 1.6e-07,
"log_odds_chosen": 0.09952077269554138,
"log_odds_ratio": -0.65927654504776,
"logits/chosen": -3.1487090587615967,
"logits/rejected": -3.0312399864196777,
"logps/chosen": -0.8334336280822754,
"logps/rejected": -0.9005411267280579,
"loss": 1.0835,
"nll_loss": 1.0175987482070923,
"rewards/accuracies": 0.75,
"rewards/chosen": -0.08334335684776306,
"rewards/margins": 0.006710751913487911,
"rewards/rejected": -0.09005410969257355,
"step": 59
},
{
"epoch": 4.8,
"grad_norm": 1.5581363439559937,
"learning_rate": 0.0,
"log_odds_chosen": 0.30142536759376526,
"log_odds_ratio": -0.5821777582168579,
"logits/chosen": -3.118028163909912,
"logits/rejected": -3.039905309677124,
"logps/chosen": -0.6326413154602051,
"logps/rejected": -0.7793558239936829,
"loss": 1.0224,
"nll_loss": 0.964179277420044,
"rewards/accuracies": 0.75,
"rewards/chosen": -0.06326413154602051,
"rewards/margins": 0.01467144675552845,
"rewards/rejected": -0.0779355838894844,
"step": 60
},
{
"epoch": 4.8,
"eval_log_odds_chosen": 0.03376197814941406,
"eval_log_odds_ratio": -0.6764087080955505,
"eval_logits/chosen": -3.361872434616089,
"eval_logits/rejected": -3.0813393592834473,
"eval_logps/chosen": -0.6450251936912537,
"eval_logps/rejected": -0.6612161993980408,
"eval_loss": 0.9213418960571289,
"eval_nll_loss": 0.8537010550498962,
"eval_rewards/accuracies": 1.0,
"eval_rewards/chosen": -0.0645025223493576,
"eval_rewards/margins": 0.001619100570678711,
"eval_rewards/rejected": -0.06612162292003632,
"eval_runtime": 3.9606,
"eval_samples_per_second": 0.252,
"eval_steps_per_second": 0.252,
"step": 60
}
],
"logging_steps": 1,
"max_steps": 60,
"num_input_tokens_seen": 0,
"num_train_epochs": 5,
"save_steps": 500,
"stateful_callbacks": {
"TrainerControl": {
"args": {
"should_epoch_stop": false,
"should_evaluate": false,
"should_log": false,
"should_save": true,
"should_training_stop": true
},
"attributes": {}
}
},
"total_flos": 0.0,
"train_batch_size": 2,
"trial_name": null,
"trial_params": null
}