silviasapora's picture
Model save
47fdd84 verified
{
"best_metric": null,
"best_model_checkpoint": null,
"epoch": 2.9765925925925925,
"eval_steps": 500,
"global_step": 315,
"is_hyper_param_search": false,
"is_local_process_zero": true,
"is_world_process_zero": true,
"log_history": [
{
"epoch": 0.047407407407407405,
"grad_norm": 1064.504150390625,
"learning_rate": 7.8125e-06,
"log_odds_chosen": -1.0592396259307861,
"log_odds_ratio": -11.0020170211792,
"logps/chosen": -22.70431900024414,
"logps/rejected": -21.645305633544922,
"loss": 450.5827,
"nll_loss": 9.83934497833252,
"po_loss": 4.241364479064941,
"rewards/accuracies": 0.4781250059604645,
"rewards/chosen": -11.35215950012207,
"rewards/margins": -0.5295066237449646,
"rewards/rejected": -10.822652816772461,
"step": 5
},
{
"epoch": 0.09481481481481481,
"grad_norm": 610.734619140625,
"learning_rate": 1.5625e-05,
"log_odds_chosen": -3.1254055500030518,
"log_odds_ratio": -11.053281784057617,
"logps/chosen": -22.00815200805664,
"logps/rejected": -18.883081436157227,
"loss": 422.0994,
"nll_loss": 8.754120826721191,
"po_loss": 4.43648624420166,
"rewards/accuracies": 0.44999998807907104,
"rewards/chosen": -11.00407600402832,
"rewards/margins": -1.5625368356704712,
"rewards/rejected": -9.441540718078613,
"step": 10
},
{
"epoch": 0.14222222222222222,
"grad_norm": 600.33056640625,
"learning_rate": 2.34375e-05,
"log_odds_chosen": -1.30233633518219,
"log_odds_ratio": -11.293824195861816,
"logps/chosen": -21.929264068603516,
"logps/rejected": -20.62543487548828,
"loss": 392.0485,
"nll_loss": 7.978579044342041,
"po_loss": 4.2729363441467285,
"rewards/accuracies": 0.515625,
"rewards/chosen": -10.964632034301758,
"rewards/margins": -0.6519135236740112,
"rewards/rejected": -10.31271743774414,
"step": 15
},
{
"epoch": 0.18962962962962962,
"grad_norm": 3273.544677734375,
"learning_rate": 3.125e-05,
"log_odds_chosen": -3.9110000133514404,
"log_odds_ratio": -8.825319290161133,
"logps/chosen": -16.157787322998047,
"logps/rejected": -12.249563217163086,
"loss": 296.8894,
"nll_loss": 5.963077545166016,
"po_loss": 3.314715623855591,
"rewards/accuracies": 0.44062501192092896,
"rewards/chosen": -8.078893661499023,
"rewards/margins": -1.954110860824585,
"rewards/rejected": -6.124781608581543,
"step": 20
},
{
"epoch": 0.23703703703703705,
"grad_norm": 309.38031005859375,
"learning_rate": 3.90625e-05,
"log_odds_chosen": -0.052033402025699615,
"log_odds_ratio": -1.5327208042144775,
"logps/chosen": -3.302628755569458,
"logps/rejected": -3.230381488800049,
"loss": 83.2066,
"nll_loss": 2.531467914581299,
"po_loss": 0.06873763352632523,
"rewards/accuracies": 0.5562499761581421,
"rewards/chosen": -1.651314377784729,
"rewards/margins": -0.036123599857091904,
"rewards/rejected": -1.6151907444000244,
"step": 25
},
{
"epoch": 0.28444444444444444,
"grad_norm": 1114.235107421875,
"learning_rate": 4.6875e-05,
"log_odds_chosen": 0.07359764724969864,
"log_odds_ratio": -0.942546546459198,
"logps/chosen": -1.9597572088241577,
"logps/rejected": -2.0194180011749268,
"loss": 64.8739,
"nll_loss": 2.057140827178955,
"po_loss": -0.029830310493707657,
"rewards/accuracies": 0.543749988079071,
"rewards/chosen": -0.9798786044120789,
"rewards/margins": 0.029830310493707657,
"rewards/rejected": -1.0097090005874634,
"step": 30
},
{
"epoch": 0.33185185185185184,
"grad_norm": 388.6631774902344,
"learning_rate": 4.998613757348784e-05,
"log_odds_chosen": 0.3581487536430359,
"log_odds_ratio": -0.841637909412384,
"logps/chosen": -1.9078718423843384,
"logps/rejected": -2.2348716259002686,
"loss": 58.6657,
"nll_loss": 1.9908554553985596,
"po_loss": -0.15755358338356018,
"rewards/accuracies": 0.543749988079071,
"rewards/chosen": -0.9539359211921692,
"rewards/margins": 0.1634998619556427,
"rewards/rejected": -1.1174358129501343,
"step": 35
},
{
"epoch": 0.37925925925925924,
"grad_norm": 215.6595458984375,
"learning_rate": 4.990147841143462e-05,
"log_odds_chosen": 0.49525055289268494,
"log_odds_ratio": -0.6980496644973755,
"logps/chosen": -1.7104499340057373,
"logps/rejected": -2.14109206199646,
"loss": 55.0966,
"nll_loss": 1.9370899200439453,
"po_loss": -0.21532109379768372,
"rewards/accuracies": 0.65625,
"rewards/chosen": -0.8552249670028687,
"rewards/margins": 0.21532109379768372,
"rewards/rejected": -1.07054603099823,
"step": 40
},
{
"epoch": 0.4266666666666667,
"grad_norm": 490.9734191894531,
"learning_rate": 4.97401218720448e-05,
"log_odds_chosen": 1.0681084394454956,
"log_odds_ratio": -0.748261034488678,
"logps/chosen": -1.9336668252944946,
"logps/rejected": -2.933429718017578,
"loss": 45.8169,
"nll_loss": 1.9130370616912842,
"po_loss": -0.4812597632408142,
"rewards/accuracies": 0.671875,
"rewards/chosen": -0.9668334126472473,
"rewards/margins": 0.4998815655708313,
"rewards/rejected": -1.466714859008789,
"step": 45
},
{
"epoch": 0.4740740740740741,
"grad_norm": 687.996826171875,
"learning_rate": 4.9502564938797946e-05,
"log_odds_chosen": 1.5803465843200684,
"log_odds_ratio": -0.8952202796936035,
"logps/chosen": -2.2172279357910156,
"logps/rejected": -3.716336488723755,
"loss": 51.598,
"nll_loss": 2.2828214168548584,
"po_loss": -0.6703845262527466,
"rewards/accuracies": 0.6937500238418579,
"rewards/chosen": -1.1086139678955078,
"rewards/margins": 0.7495540976524353,
"rewards/rejected": -1.8581682443618774,
"step": 50
},
{
"epoch": 0.5214814814814814,
"grad_norm": 4070.76123046875,
"learning_rate": 4.918953929490768e-05,
"log_odds_chosen": 2.9766428470611572,
"log_odds_ratio": -0.919791042804718,
"logps/chosen": -2.4137585163116455,
"logps/rejected": -5.309506416320801,
"loss": 37.3237,
"nll_loss": 2.3748691082000732,
"po_loss": -1.2085039615631104,
"rewards/accuracies": 0.6656249761581421,
"rewards/chosen": -1.2068792581558228,
"rewards/margins": 1.4478741884231567,
"rewards/rejected": -2.6547532081604004,
"step": 55
},
{
"epoch": 0.5688888888888889,
"grad_norm": 569.17724609375,
"learning_rate": 4.88020090697132e-05,
"log_odds_chosen": 5.00119686126709,
"log_odds_ratio": -1.5456047058105469,
"logps/chosen": -6.171454906463623,
"logps/rejected": -11.093535423278809,
"loss": 103.9391,
"nll_loss": 4.901269912719727,
"po_loss": -1.653172492980957,
"rewards/accuracies": 0.721875011920929,
"rewards/chosen": -3.0857274532318115,
"rewards/margins": 2.461040496826172,
"rewards/rejected": -5.546767711639404,
"step": 60
},
{
"epoch": 0.6162962962962963,
"grad_norm": 614.133544921875,
"learning_rate": 4.834116786912897e-05,
"log_odds_chosen": 1.351292371749878,
"log_odds_ratio": -1.3231507539749146,
"logps/chosen": -2.553697109222412,
"logps/rejected": -3.8612685203552246,
"loss": 63.6228,
"nll_loss": 2.556739091873169,
"po_loss": -0.5685266256332397,
"rewards/accuracies": 0.59375,
"rewards/chosen": -1.276848554611206,
"rewards/margins": 0.6537858843803406,
"rewards/rejected": -1.9306342601776123,
"step": 65
},
{
"epoch": 0.6637037037037037,
"grad_norm": 386.0463562011719,
"learning_rate": 4.7808435099299045e-05,
"log_odds_chosen": 2.089937210083008,
"log_odds_ratio": -0.670052170753479,
"logps/chosen": -1.7493152618408203,
"logps/rejected": -3.7693347930908203,
"loss": 34.9169,
"nll_loss": 1.8565772771835327,
"po_loss": -0.7654256820678711,
"rewards/accuracies": 0.6937500238418579,
"rewards/chosen": -0.8746576309204102,
"rewards/margins": 1.010009765625,
"rewards/rejected": -1.8846673965454102,
"step": 70
},
{
"epoch": 0.7111111111111111,
"grad_norm": 3206.72705078125,
"learning_rate": 4.720545159477922e-05,
"log_odds_chosen": 2.741664409637451,
"log_odds_ratio": -1.1608259677886963,
"logps/chosen": -3.1741726398468018,
"logps/rejected": -5.838489532470703,
"loss": 55.9823,
"nll_loss": 2.62483811378479,
"po_loss": -0.8753900527954102,
"rewards/accuracies": 0.6968749761581421,
"rewards/chosen": -1.5870863199234009,
"rewards/margins": 1.3321585655212402,
"rewards/rejected": -2.9192447662353516,
"step": 75
},
{
"epoch": 0.7585185185185185,
"grad_norm": 412.6952209472656,
"learning_rate": 4.653407456471222e-05,
"log_odds_chosen": 1.9728152751922607,
"log_odds_ratio": -0.9150087237358093,
"logps/chosen": -2.0698251724243164,
"logps/rejected": -3.9561798572540283,
"loss": 41.4899,
"nll_loss": 2.0737736225128174,
"po_loss": -0.7772130370140076,
"rewards/accuracies": 0.653124988079071,
"rewards/chosen": -1.0349125862121582,
"rewards/margins": 0.9431774020195007,
"rewards/rejected": -1.9780899286270142,
"step": 80
},
{
"epoch": 0.8059259259259259,
"grad_norm": 1100.900634765625,
"learning_rate": 4.579637187256222e-05,
"log_odds_chosen": 2.417811870574951,
"log_odds_ratio": -0.9469805955886841,
"logps/chosen": -2.1366405487060547,
"logps/rejected": -4.464547157287598,
"loss": 36.3301,
"nll_loss": 2.163440704345703,
"po_loss": -1.0281257629394531,
"rewards/accuracies": 0.6812499761581421,
"rewards/chosen": -1.0683202743530273,
"rewards/margins": 1.163953185081482,
"rewards/rejected": -2.232273578643799,
"step": 85
},
{
"epoch": 0.8533333333333334,
"grad_norm": 785.2890625,
"learning_rate": 4.499461566702685e-05,
"log_odds_chosen": 3.7706730365753174,
"log_odds_ratio": -1.2985801696777344,
"logps/chosen": -3.598789691925049,
"logps/rejected": -7.259984016418457,
"loss": 42.6548,
"nll_loss": 2.9308180809020996,
"po_loss": -1.597855567932129,
"rewards/accuracies": 0.690625011920929,
"rewards/chosen": -1.7993948459625244,
"rewards/margins": 1.830596923828125,
"rewards/rejected": -3.6299920082092285,
"step": 90
},
{
"epoch": 0.9007407407407407,
"grad_norm": 3162.0009765625,
"learning_rate": 4.413127538374411e-05,
"log_odds_chosen": 2.615981340408325,
"log_odds_ratio": -1.0526683330535889,
"logps/chosen": -2.3276286125183105,
"logps/rejected": -4.851325988769531,
"loss": 31.8051,
"nll_loss": 2.035627841949463,
"po_loss": -1.0417195558547974,
"rewards/accuracies": 0.6499999761581421,
"rewards/chosen": -1.1638143062591553,
"rewards/margins": 1.2618486881256104,
"rewards/rejected": -2.4256629943847656,
"step": 95
},
{
"epoch": 0.9481481481481482,
"grad_norm": 571.5673217773438,
"learning_rate": 4.320901013934887e-05,
"log_odds_chosen": 4.09200382232666,
"log_odds_ratio": -1.3848493099212646,
"logps/chosen": -3.318943500518799,
"logps/rejected": -7.310236930847168,
"loss": 37.0821,
"nll_loss": 2.7883472442626953,
"po_loss": -1.629533052444458,
"rewards/accuracies": 0.6812499761581421,
"rewards/chosen": -1.6594717502593994,
"rewards/margins": 1.9956468343734741,
"rewards/rejected": -3.655118465423584,
"step": 100
},
{
"epoch": 0.9955555555555555,
"grad_norm": 251.86697387695312,
"learning_rate": 4.223066054130568e-05,
"log_odds_chosen": 3.148193359375,
"log_odds_ratio": -0.6689270734786987,
"logps/chosen": -1.6926807165145874,
"logps/rejected": -4.753729820251465,
"loss": 24.9079,
"nll_loss": 1.8303325176239014,
"po_loss": -1.05196213722229,
"rewards/accuracies": 0.6781250238418579,
"rewards/chosen": -0.8463403582572937,
"rewards/margins": 1.5305248498916626,
"rewards/rejected": -2.3768649101257324,
"step": 105
},
{
"epoch": 1.037925925925926,
"grad_norm": 235.2350616455078,
"learning_rate": 4.1199239938743797e-05,
"log_odds_chosen": 3.255824565887451,
"log_odds_ratio": -0.6982910633087158,
"logps/chosen": -1.715062141418457,
"logps/rejected": -4.823176383972168,
"loss": 23.0323,
"nll_loss": 1.716023325920105,
"po_loss": -0.9106981754302979,
"rewards/accuracies": 0.7307692170143127,
"rewards/chosen": -0.8575310707092285,
"rewards/margins": 1.5540571212768555,
"rewards/rejected": -2.411588191986084,
"step": 110
},
{
"epoch": 1.0853333333333333,
"grad_norm": 457.39080810546875,
"learning_rate": 4.0117925141242174e-05,
"log_odds_chosen": 5.342764854431152,
"log_odds_ratio": -0.9483366012573242,
"logps/chosen": -2.369607448577881,
"logps/rejected": -7.539316654205322,
"loss": 17.0777,
"nll_loss": 2.0790820121765137,
"po_loss": -1.5454037189483643,
"rewards/accuracies": 0.7593749761581421,
"rewards/chosen": -1.1848037242889404,
"rewards/margins": 2.5848546028137207,
"rewards/rejected": -3.769658327102661,
"step": 115
},
{
"epoch": 1.1327407407407408,
"grad_norm": 478.9156799316406,
"learning_rate": 3.899004663415084e-05,
"log_odds_chosen": 4.5327959060668945,
"log_odds_ratio": -0.7982605695724487,
"logps/chosen": -2.2072598934173584,
"logps/rejected": -6.586657524108887,
"loss": 12.6686,
"nll_loss": 1.9718843698501587,
"po_loss": -1.5759899616241455,
"rewards/accuracies": 0.7406250238418579,
"rewards/chosen": -1.1036299467086792,
"rewards/margins": 2.1896984577178955,
"rewards/rejected": -3.2933287620544434,
"step": 120
},
{
"epoch": 1.1801481481481482,
"grad_norm": 760.638916015625,
"learning_rate": 3.781907832058587e-05,
"log_odds_chosen": 4.349984645843506,
"log_odds_ratio": -1.0327547788619995,
"logps/chosen": -2.662911891937256,
"logps/rejected": -6.870619297027588,
"loss": 32.4605,
"nll_loss": 2.5304787158966064,
"po_loss": -1.5160895586013794,
"rewards/accuracies": 0.7406250238418579,
"rewards/chosen": -1.331455945968628,
"rewards/margins": 2.103853702545166,
"rewards/rejected": -3.435309648513794,
"step": 125
},
{
"epoch": 1.2275555555555555,
"grad_norm": 1151.731201171875,
"learning_rate": 3.660862682169282e-05,
"log_odds_chosen": 6.421099662780762,
"log_odds_ratio": -1.0163640975952148,
"logps/chosen": -2.875323534011841,
"logps/rejected": -9.088385581970215,
"loss": -4.7801,
"nll_loss": 2.3773765563964844,
"po_loss": -2.5267534255981445,
"rewards/accuracies": 0.762499988079071,
"rewards/chosen": -1.4376617670059204,
"rewards/margins": 3.1065304279327393,
"rewards/rejected": -4.544192790985107,
"step": 130
},
{
"epoch": 1.274962962962963,
"grad_norm": 535.7034912109375,
"learning_rate": 3.5362420368134356e-05,
"log_odds_chosen": 3.287085771560669,
"log_odds_ratio": -0.6112745404243469,
"logps/chosen": -1.6216014623641968,
"logps/rejected": -4.728818893432617,
"loss": 9.9669,
"nll_loss": 1.7199676036834717,
"po_loss": -1.4085030555725098,
"rewards/accuracies": 0.7437499761581421,
"rewards/chosen": -0.8108007311820984,
"rewards/margins": 1.553608775138855,
"rewards/rejected": -2.3644094467163086,
"step": 135
},
{
"epoch": 1.3223703703703704,
"grad_norm": 1074.8087158203125,
"learning_rate": 3.408429731701635e-05,
"log_odds_chosen": 3.7467429637908936,
"log_odds_ratio": -0.7782642841339111,
"logps/chosen": -2.253432512283325,
"logps/rejected": -5.834208011627197,
"loss": 18.4068,
"nll_loss": 2.049558401107788,
"po_loss": -1.4743449687957764,
"rewards/accuracies": 0.734375,
"rewards/chosen": -1.1267162561416626,
"rewards/margins": 1.790387749671936,
"rewards/rejected": -2.9171040058135986,
"step": 140
},
{
"epoch": 1.3697777777777778,
"grad_norm": 2433.367919921875,
"learning_rate": 3.2778194329621104e-05,
"log_odds_chosen": 6.9977288246154785,
"log_odds_ratio": -0.8477832078933716,
"logps/chosen": -2.8123180866241455,
"logps/rejected": -9.62822151184082,
"loss": -7.1341,
"nll_loss": 2.3402633666992188,
"po_loss": -2.5632026195526123,
"rewards/accuracies": 0.7718750238418579,
"rewards/chosen": -1.4061590433120728,
"rewards/margins": 3.407951831817627,
"rewards/rejected": -4.81411075592041,
"step": 145
},
{
"epoch": 1.417185185185185,
"grad_norm": 1035.2568359375,
"learning_rate": 3.144813424636031e-05,
"log_odds_chosen": 7.531359672546387,
"log_odds_ratio": -0.9453743696212769,
"logps/chosen": -2.4776880741119385,
"logps/rejected": -9.81294059753418,
"loss": -8.8755,
"nll_loss": 2.127579689025879,
"po_loss": -2.404938220977783,
"rewards/accuracies": 0.762499988079071,
"rewards/chosen": -1.2388440370559692,
"rewards/margins": 3.667625904083252,
"rewards/rejected": -4.90647029876709,
"step": 150
},
{
"epoch": 1.4645925925925927,
"grad_norm": 1741.77978515625,
"learning_rate": 3.0098213696293542e-05,
"log_odds_chosen": 4.925074577331543,
"log_odds_ratio": -1.4115755558013916,
"logps/chosen": -2.6048836708068848,
"logps/rejected": -7.376564025878906,
"loss": 23.5026,
"nll_loss": 2.325968027114868,
"po_loss": -1.5915124416351318,
"rewards/accuracies": 0.6968749761581421,
"rewards/chosen": -1.3024418354034424,
"rewards/margins": 2.3858399391174316,
"rewards/rejected": -3.688282012939453,
"step": 155
},
{
"epoch": 1.512,
"grad_norm": 1471.9173583984375,
"learning_rate": 2.8732590479375165e-05,
"log_odds_chosen": 6.550325870513916,
"log_odds_ratio": -0.9376071691513062,
"logps/chosen": -2.9662528038024902,
"logps/rejected": -9.300813674926758,
"loss": -3.3757,
"nll_loss": 2.572293758392334,
"po_loss": -2.677783250808716,
"rewards/accuracies": 0.762499988079071,
"rewards/chosen": -1.4831264019012451,
"rewards/margins": 3.1672801971435547,
"rewards/rejected": -4.650406837463379,
"step": 160
},
{
"epoch": 1.5594074074074074,
"grad_norm": 562.9378051757812,
"learning_rate": 2.7355470760292956e-05,
"log_odds_chosen": 5.505670547485352,
"log_odds_ratio": -0.8015021085739136,
"logps/chosen": -2.046816349029541,
"logps/rejected": -7.327232360839844,
"loss": -4.0531,
"nll_loss": 2.0913338661193848,
"po_loss": -2.217992067337036,
"rewards/accuracies": 0.768750011920929,
"rewards/chosen": -1.0234081745147705,
"rewards/margins": 2.6402080059051514,
"rewards/rejected": -3.663616180419922,
"step": 165
},
{
"epoch": 1.6068148148148147,
"grad_norm": 1370.6552734375,
"learning_rate": 2.597109611334169e-05,
"log_odds_chosen": 9.098596572875977,
"log_odds_ratio": -1.227461576461792,
"logps/chosen": -3.2510433197021484,
"logps/rejected": -12.151667594909668,
"loss": -13.6098,
"nll_loss": 2.593132495880127,
"po_loss": -3.018439769744873,
"rewards/accuracies": 0.7749999761581421,
"rewards/chosen": -1.6255216598510742,
"rewards/margins": 4.450311660766602,
"rewards/rejected": -6.075833797454834,
"step": 170
},
{
"epoch": 1.6542222222222223,
"grad_norm": 1004.2781982421875,
"learning_rate": 2.458373045823404e-05,
"log_odds_chosen": 6.2596330642700195,
"log_odds_ratio": -0.9497385025024414,
"logps/chosen": -2.4776344299316406,
"logps/rejected": -8.515233039855957,
"loss": -1.5757,
"nll_loss": 2.2205429077148438,
"po_loss": -2.2697837352752686,
"rewards/accuracies": 0.7718750238418579,
"rewards/chosen": -1.2388172149658203,
"rewards/margins": 3.018799066543579,
"rewards/rejected": -4.2576165199279785,
"step": 175
},
{
"epoch": 1.7016296296296296,
"grad_norm": 3024.766845703125,
"learning_rate": 2.3197646927086697e-05,
"log_odds_chosen": 7.638102054595947,
"log_odds_ratio": -1.0293067693710327,
"logps/chosen": -3.2467892169952393,
"logps/rejected": -10.673727989196777,
"loss": -11.2032,
"nll_loss": 2.4954748153686523,
"po_loss": -2.845574378967285,
"rewards/accuracies": 0.768750011920929,
"rewards/chosen": -1.6233946084976196,
"rewards/margins": 3.7134690284729004,
"rewards/rejected": -5.336863994598389,
"step": 180
},
{
"epoch": 1.749037037037037,
"grad_norm": 976.9481201171875,
"learning_rate": 2.1817114703032176e-05,
"log_odds_chosen": 7.674588203430176,
"log_odds_ratio": -1.1676676273345947,
"logps/chosen": -3.3499820232391357,
"logps/rejected": -10.849446296691895,
"loss": -16.751,
"nll_loss": 2.3611953258514404,
"po_loss": -2.8846640586853027,
"rewards/accuracies": 0.7593749761581421,
"rewards/chosen": -1.6749910116195679,
"rewards/margins": 3.749732255935669,
"rewards/rejected": -5.424723148345947,
"step": 185
},
{
"epoch": 1.7964444444444445,
"grad_norm": 1403.0980224609375,
"learning_rate": 2.0446385870993467e-05,
"log_odds_chosen": 8.678237915039062,
"log_odds_ratio": -0.6416040062904358,
"logps/chosen": -2.349724054336548,
"logps/rejected": -10.788393020629883,
"loss": -39.35,
"nll_loss": 1.945783257484436,
"po_loss": -3.175469398498535,
"rewards/accuracies": 0.815625011920929,
"rewards/chosen": -1.174862027168274,
"rewards/margins": 4.219334602355957,
"rewards/rejected": -5.394196510314941,
"step": 190
},
{
"epoch": 1.8438518518518519,
"grad_norm": 2194.176025390625,
"learning_rate": 1.9089682321121834e-05,
"log_odds_chosen": 11.073338508605957,
"log_odds_ratio": -1.7082958221435547,
"logps/chosen": -4.391510486602783,
"logps/rejected": -15.270744323730469,
"loss": -2.4223,
"nll_loss": 3.056588649749756,
"po_loss": -3.132284641265869,
"rewards/accuracies": 0.7718750238418579,
"rewards/chosen": -2.1957552433013916,
"rewards/margins": 5.439617156982422,
"rewards/rejected": -7.635372161865234,
"step": 195
},
{
"epoch": 1.8912592592592592,
"grad_norm": 1372.590576171875,
"learning_rate": 1.775118274523545e-05,
"log_odds_chosen": 7.143240928649902,
"log_odds_ratio": -1.3253802061080933,
"logps/chosen": -3.159841537475586,
"logps/rejected": -10.11519718170166,
"loss": -3.8252,
"nll_loss": 2.4054999351501465,
"po_loss": -2.525035858154297,
"rewards/accuracies": 0.715624988079071,
"rewards/chosen": -1.579920768737793,
"rewards/margins": 3.477677583694458,
"rewards/rejected": -5.05759859085083,
"step": 200
},
{
"epoch": 1.9386666666666668,
"grad_norm": 1031.638916015625,
"learning_rate": 1.643500976631037e-05,
"log_odds_chosen": 5.937264442443848,
"log_odds_ratio": -0.8166869878768921,
"logps/chosen": -2.044642210006714,
"logps/rejected": -7.761441230773926,
"loss": -10.1847,
"nll_loss": 1.9633842706680298,
"po_loss": -2.281655788421631,
"rewards/accuracies": 0.7906249761581421,
"rewards/chosen": -1.022321105003357,
"rewards/margins": 2.8583993911743164,
"rewards/rejected": -3.880720615386963,
"step": 205
},
{
"epoch": 1.986074074074074,
"grad_norm": 1609.1983642578125,
"learning_rate": 1.514521724066537e-05,
"log_odds_chosen": 7.844486236572266,
"log_odds_ratio": -1.5023125410079956,
"logps/chosen": -3.9588305950164795,
"logps/rejected": -11.614171981811523,
"loss": -3.4843,
"nll_loss": 2.696570873260498,
"po_loss": -2.8054568767547607,
"rewards/accuracies": 0.746874988079071,
"rewards/chosen": -1.9794152975082397,
"rewards/margins": 3.827669858932495,
"rewards/rejected": -5.807085990905762,
"step": 210
},
{
"epoch": 2.0284444444444443,
"grad_norm": 1158.2188720703125,
"learning_rate": 1.3885777771950348e-05,
"log_odds_chosen": 10.403746604919434,
"log_odds_ratio": -1.2075212001800537,
"logps/chosen": -3.1166441440582275,
"logps/rejected": -13.249801635742188,
"loss": -41.5962,
"nll_loss": 2.3673253059387207,
"po_loss": -3.821737766265869,
"rewards/accuracies": 0.7867133021354675,
"rewards/chosen": -1.5583220720291138,
"rewards/margins": 5.066579341888428,
"rewards/rejected": -6.624900817871094,
"step": 215
},
{
"epoch": 2.075851851851852,
"grad_norm": 1111.02734375,
"learning_rate": 1.2660570475395683e-05,
"log_odds_chosen": 8.10044002532959,
"log_odds_ratio": -1.2627264261245728,
"logps/chosen": -3.747706890106201,
"logps/rejected": -11.619832992553711,
"loss": -19.3237,
"nll_loss": 2.5686912536621094,
"po_loss": -3.172555923461914,
"rewards/accuracies": 0.765625,
"rewards/chosen": -1.8738534450531006,
"rewards/margins": 3.936063289642334,
"rewards/rejected": -5.8099164962768555,
"step": 220
},
{
"epoch": 2.1232592592592594,
"grad_norm": 2720.02490234375,
"learning_rate": 1.1473369030008974e-05,
"log_odds_chosen": 11.025764465332031,
"log_odds_ratio": -0.7701746821403503,
"logps/chosen": -3.573103666305542,
"logps/rejected": -14.34417724609375,
"loss": -58.0462,
"nll_loss": 2.4860126972198486,
"po_loss": -4.299956798553467,
"rewards/accuracies": 0.8374999761581421,
"rewards/chosen": -1.786551833152771,
"rewards/margins": 5.385537147521973,
"rewards/rejected": -7.172088623046875,
"step": 225
},
{
"epoch": 2.1706666666666665,
"grad_norm": 1029.630126953125,
"learning_rate": 1.0327830055518842e-05,
"log_odds_chosen": 10.32060432434082,
"log_odds_ratio": -0.8015488386154175,
"logps/chosen": -3.26597261428833,
"logps/rejected": -13.332249641418457,
"loss": -49.913,
"nll_loss": 2.330512762069702,
"po_loss": -3.890293836593628,
"rewards/accuracies": 0.8062499761581421,
"rewards/chosen": -1.632986307144165,
"rewards/margins": 5.033139228820801,
"rewards/rejected": -6.6661248207092285,
"step": 230
},
{
"epoch": 2.218074074074074,
"grad_norm": 1995.1678466796875,
"learning_rate": 9.227481849865235e-06,
"log_odds_chosen": 11.123466491699219,
"log_odds_ratio": -0.6873899102210999,
"logps/chosen": -2.032088041305542,
"logps/rejected": -12.853292465209961,
"loss": -69.5192,
"nll_loss": 1.9619925022125244,
"po_loss": -4.1344685554504395,
"rewards/accuracies": 0.8187500238418579,
"rewards/chosen": -1.016044020652771,
"rewards/margins": 5.410602569580078,
"rewards/rejected": -6.4266462326049805,
"step": 235
},
{
"epoch": 2.2654814814814817,
"grad_norm": 2002.389892578125,
"learning_rate": 8.175713521924978e-06,
"log_odds_chosen": 12.202537536621094,
"log_odds_ratio": -0.6559484004974365,
"logps/chosen": -2.459198236465454,
"logps/rejected": -14.35517692565918,
"loss": -66.0372,
"nll_loss": 2.1631743907928467,
"po_loss": -4.226839065551758,
"rewards/accuracies": 0.859375,
"rewards/chosen": -1.229599118232727,
"rewards/margins": 5.947988986968994,
"rewards/rejected": -7.17758846282959,
"step": 240
},
{
"epoch": 2.3128888888888888,
"grad_norm": 1811.5279541015625,
"learning_rate": 7.1757645529443665e-06,
"log_odds_chosen": 12.961212158203125,
"log_odds_ratio": -0.48759812116622925,
"logps/chosen": -2.229421615600586,
"logps/rejected": -14.900075912475586,
"loss": -85.5446,
"nll_loss": 1.9955413341522217,
"po_loss": -4.668810844421387,
"rewards/accuracies": 0.8531249761581421,
"rewards/chosen": -1.114710807800293,
"rewards/margins": 6.335326671600342,
"rewards/rejected": -7.450037956237793,
"step": 245
},
{
"epoch": 2.3602962962962963,
"grad_norm": 1374.3148193359375,
"learning_rate": 6.230714818829733e-06,
"log_odds_chosen": 13.162447929382324,
"log_odds_ratio": -0.579684853553772,
"logps/chosen": -2.2635672092437744,
"logps/rejected": -15.118783950805664,
"loss": -78.0242,
"nll_loss": 1.9523729085922241,
"po_loss": -4.39063024520874,
"rewards/accuracies": 0.824999988079071,
"rewards/chosen": -1.1317836046218872,
"rewards/margins": 6.427608489990234,
"rewards/rejected": -7.559391975402832,
"step": 250
},
{
"epoch": 2.407703703703704,
"grad_norm": 1782.409423828125,
"learning_rate": 5.343475104027743e-06,
"log_odds_chosen": 12.180627822875977,
"log_odds_ratio": -0.7269414663314819,
"logps/chosen": -2.7276577949523926,
"logps/rejected": -14.642044067382812,
"loss": -61.2515,
"nll_loss": 2.155933380126953,
"po_loss": -4.07004451751709,
"rewards/accuracies": 0.831250011920929,
"rewards/chosen": -1.3638288974761963,
"rewards/margins": 5.957192897796631,
"rewards/rejected": -7.321022033691406,
"step": 255
},
{
"epoch": 2.455111111111111,
"grad_norm": 1750.225341796875,
"learning_rate": 4.516778136213037e-06,
"log_odds_chosen": 12.096226692199707,
"log_odds_ratio": -0.6728618741035461,
"logps/chosen": -2.2444777488708496,
"logps/rejected": -14.059532165527344,
"loss": -72.1379,
"nll_loss": 1.928434133529663,
"po_loss": -4.182744026184082,
"rewards/accuracies": 0.84375,
"rewards/chosen": -1.1222388744354248,
"rewards/margins": 5.907527923583984,
"rewards/rejected": -7.029766082763672,
"step": 260
},
{
"epoch": 2.5025185185185186,
"grad_norm": 1919.8836669921875,
"learning_rate": 3.7531701693965554e-06,
"log_odds_chosen": 11.939556121826172,
"log_odds_ratio": -0.5605658888816833,
"logps/chosen": -2.554964780807495,
"logps/rejected": -14.228480339050293,
"loss": -72.1107,
"nll_loss": 2.096379518508911,
"po_loss": -4.3498382568359375,
"rewards/accuracies": 0.84375,
"rewards/chosen": -1.2774823904037476,
"rewards/margins": 5.836758136749268,
"rewards/rejected": -7.1142401695251465,
"step": 265
},
{
"epoch": 2.549925925925926,
"grad_norm": 3865.446044921875,
"learning_rate": 3.055003141378948e-06,
"log_odds_chosen": 13.130399703979492,
"log_odds_ratio": -0.7432569265365601,
"logps/chosen": -2.868731737136841,
"logps/rejected": -15.705291748046875,
"loss": -71.6455,
"nll_loss": 2.3375840187072754,
"po_loss": -4.576504230499268,
"rewards/accuracies": 0.8374999761581421,
"rewards/chosen": -1.4343658685684204,
"rewards/margins": 6.418280124664307,
"rewards/rejected": -7.8526458740234375,
"step": 270
},
{
"epoch": 2.5973333333333333,
"grad_norm": 1720.9822998046875,
"learning_rate": 2.424427429704365e-06,
"log_odds_chosen": 12.388298034667969,
"log_odds_ratio": -0.7917383313179016,
"logps/chosen": -2.539780378341675,
"logps/rejected": -14.631436347961426,
"loss": -74.1033,
"nll_loss": 2.096407651901245,
"po_loss": -4.412135124206543,
"rewards/accuracies": 0.862500011920929,
"rewards/chosen": -1.2698901891708374,
"rewards/margins": 6.045827865600586,
"rewards/rejected": -7.315718173980713,
"step": 275
},
{
"epoch": 2.644740740740741,
"grad_norm": 1882.84033203125,
"learning_rate": 1.8633852284264508e-06,
"log_odds_chosen": 13.360041618347168,
"log_odds_ratio": -0.7875075936317444,
"logps/chosen": -2.570291757583618,
"logps/rejected": -15.660143852233887,
"loss": -80.8755,
"nll_loss": 2.1218926906585693,
"po_loss": -4.649253845214844,
"rewards/accuracies": 0.840624988079071,
"rewards/chosen": -1.285145878791809,
"rewards/margins": 6.544926643371582,
"rewards/rejected": -7.830071926116943,
"step": 280
},
{
"epoch": 2.6921481481481484,
"grad_norm": 2601.312255859375,
"learning_rate": 1.3736045660864034e-06,
"log_odds_chosen": 12.91968059539795,
"log_odds_ratio": -0.935935378074646,
"logps/chosen": -3.055326461791992,
"logps/rejected": -15.716508865356445,
"loss": -68.9803,
"nll_loss": 2.237934112548828,
"po_loss": -4.393568515777588,
"rewards/accuracies": 0.846875011920929,
"rewards/chosen": -1.527663230895996,
"rewards/margins": 6.33059024810791,
"rewards/rejected": -7.858254432678223,
"step": 285
},
{
"epoch": 2.7395555555555555,
"grad_norm": 2271.935302734375,
"learning_rate": 9.565939833279192e-07,
"log_odds_chosen": 12.334630966186523,
"log_odds_ratio": -0.5848469734191895,
"logps/chosen": -3.0595030784606934,
"logps/rejected": -15.129420280456543,
"loss": -64.562,
"nll_loss": 2.4212985038757324,
"po_loss": -4.438861846923828,
"rewards/accuracies": 0.8374999761581421,
"rewards/chosen": -1.5297515392303467,
"rewards/margins": 6.034958362579346,
"rewards/rejected": -7.5647101402282715,
"step": 290
},
{
"epoch": 2.786962962962963,
"grad_norm": 1566.753173828125,
"learning_rate": 6.136378865420872e-07,
"log_odds_chosen": 11.438375473022461,
"log_odds_ratio": -0.9271749258041382,
"logps/chosen": -2.976278781890869,
"logps/rejected": -14.156593322753906,
"loss": -57.0123,
"nll_loss": 2.327953577041626,
"po_loss": -4.109589576721191,
"rewards/accuracies": 0.8187500238418579,
"rewards/chosen": -1.4881393909454346,
"rewards/margins": 5.5901570320129395,
"rewards/rejected": -7.078296661376953,
"step": 295
},
{
"epoch": 2.83437037037037,
"grad_norm": 2821.74609375,
"learning_rate": 3.45792591853214e-07,
"log_odds_chosen": 11.852496147155762,
"log_odds_ratio": -0.5676802396774292,
"logps/chosen": -2.4296622276306152,
"logps/rejected": -14.001643180847168,
"loss": -73.6596,
"nll_loss": 1.9556903839111328,
"po_loss": -4.2575531005859375,
"rewards/accuracies": 0.815625011920929,
"rewards/chosen": -1.2148311138153076,
"rewards/margins": 5.785989761352539,
"rewards/rejected": -7.000821590423584,
"step": 300
},
{
"epoch": 2.8817777777777778,
"grad_norm": 2135.990966796875,
"learning_rate": 1.538830716302092e-07,
"log_odds_chosen": 12.745733261108398,
"log_odds_ratio": -0.7597203850746155,
"logps/chosen": -2.1967830657958984,
"logps/rejected": -14.625683784484863,
"loss": -78.7585,
"nll_loss": 1.9788535833358765,
"po_loss": -4.440057277679443,
"rewards/accuracies": 0.8500000238418579,
"rewards/chosen": -1.0983915328979492,
"rewards/margins": 6.214449882507324,
"rewards/rejected": -7.312841892242432,
"step": 305
},
{
"epoch": 2.9291851851851853,
"grad_norm": 1398.578369140625,
"learning_rate": 3.8500413544415025e-08,
"log_odds_chosen": 13.022871017456055,
"log_odds_ratio": -0.4309860169887543,
"logps/chosen": -2.221106767654419,
"logps/rejected": -14.942235946655273,
"loss": -86.5137,
"nll_loss": 1.9662892818450928,
"po_loss": -4.66984224319458,
"rewards/accuracies": 0.871874988079071,
"rewards/chosen": -1.1105533838272095,
"rewards/margins": 6.360564708709717,
"rewards/rejected": -7.471117973327637,
"step": 310
},
{
"epoch": 2.9765925925925925,
"grad_norm": 1498.622314453125,
"learning_rate": 0.0,
"log_odds_chosen": 11.79759407043457,
"log_odds_ratio": -0.305752694606781,
"logps/chosen": -1.7462921142578125,
"logps/rejected": -13.216029167175293,
"loss": -82.7571,
"nll_loss": 1.7008321285247803,
"po_loss": -4.286990642547607,
"rewards/accuracies": 0.859375,
"rewards/chosen": -0.8731460571289062,
"rewards/margins": 5.734868049621582,
"rewards/rejected": -6.6080145835876465,
"step": 315
},
{
"epoch": 2.9765925925925925,
"step": 315,
"total_flos": 0.0,
"train_loss": 16.270644454350546,
"train_runtime": 9547.5166,
"train_samples_per_second": 2.121,
"train_steps_per_second": 0.033
}
],
"logging_steps": 5,
"max_steps": 315,
"num_input_tokens_seen": 0,
"num_train_epochs": 3,
"save_steps": 100000,
"stateful_callbacks": {
"TrainerControl": {
"args": {
"should_epoch_stop": false,
"should_evaluate": false,
"should_log": false,
"should_save": true,
"should_training_stop": true
},
"attributes": {}
}
},
"total_flos": 0.0,
"train_batch_size": 1,
"trial_name": null,
"trial_params": null
}