silviasapora's picture
Model save
2f79711 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": 1138.854736328125,
"learning_rate": 7.8125e-06,
"log_odds_chosen": -1.0697920322418213,
"log_odds_ratio": -11.011500358581543,
"logps/chosen": -22.719226837158203,
"logps/rejected": -21.649660110473633,
"loss": 450.84,
"nll_loss": 9.846151351928711,
"po_loss": 4.242598533630371,
"rewards/accuracies": 0.48124998807907104,
"rewards/chosen": -11.359613418579102,
"rewards/margins": -0.5347837209701538,
"rewards/rejected": -10.824830055236816,
"step": 5
},
{
"epoch": 0.09481481481481481,
"grad_norm": 597.58203125,
"learning_rate": 1.5625e-05,
"log_odds_chosen": -3.116546392440796,
"log_odds_ratio": -11.056388854980469,
"logps/chosen": -22.023601531982422,
"logps/rejected": -18.90738868713379,
"loss": 421.2396,
"nll_loss": 8.764582633972168,
"po_loss": 4.39915132522583,
"rewards/accuracies": 0.44999998807907104,
"rewards/chosen": -11.011800765991211,
"rewards/margins": -1.5581079721450806,
"rewards/rejected": -9.453694343566895,
"step": 10
},
{
"epoch": 0.14222222222222222,
"grad_norm": 462.572265625,
"learning_rate": 2.34375e-05,
"log_odds_chosen": -1.2942144870758057,
"log_odds_ratio": -11.329824447631836,
"logps/chosen": -22.003665924072266,
"logps/rejected": -20.707988739013672,
"loss": 393.115,
"nll_loss": 8.009772300720215,
"po_loss": 4.2750701904296875,
"rewards/accuracies": 0.515625,
"rewards/chosen": -11.001832962036133,
"rewards/margins": -0.6478384137153625,
"rewards/rejected": -10.353994369506836,
"step": 15
},
{
"epoch": 0.18962962962962962,
"grad_norm": 3452.147705078125,
"learning_rate": 3.125e-05,
"log_odds_chosen": -4.19545316696167,
"log_odds_ratio": -9.310150146484375,
"logps/chosen": -16.894468307495117,
"logps/rejected": -12.701272964477539,
"loss": 308.1191,
"nll_loss": 6.174555778503418,
"po_loss": 3.454167127609253,
"rewards/accuracies": 0.4281249940395355,
"rewards/chosen": -8.447234153747559,
"rewards/margins": -2.0965964794158936,
"rewards/rejected": -6.3506364822387695,
"step": 20
},
{
"epoch": 0.23703703703703705,
"grad_norm": 254.29014587402344,
"learning_rate": 3.90625e-05,
"log_odds_chosen": -0.08858927339315414,
"log_odds_ratio": -1.799390435218811,
"logps/chosen": -3.9111340045928955,
"logps/rejected": -3.804535388946533,
"loss": 90.7617,
"nll_loss": 2.6883437633514404,
"po_loss": 0.14795991778373718,
"rewards/accuracies": 0.5562499761581421,
"rewards/chosen": -1.9555670022964478,
"rewards/margins": -0.05329931899905205,
"rewards/rejected": -1.9022676944732666,
"step": 25
},
{
"epoch": 0.28444444444444444,
"grad_norm": 269.8670349121094,
"learning_rate": 4.6875e-05,
"log_odds_chosen": 0.05695251747965813,
"log_odds_ratio": -0.9221351742744446,
"logps/chosen": -1.9418389797210693,
"logps/rejected": -1.9890735149383545,
"loss": 65.438,
"nll_loss": 2.0685558319091797,
"po_loss": -0.023617293685674667,
"rewards/accuracies": 0.5406249761581421,
"rewards/chosen": -0.9709194898605347,
"rewards/margins": 0.023617293685674667,
"rewards/rejected": -0.9945367574691772,
"step": 30
},
{
"epoch": 0.33185185185185184,
"grad_norm": 265.45452880859375,
"learning_rate": 4.998613757348784e-05,
"log_odds_chosen": 0.3491572439670563,
"log_odds_ratio": -0.8370898962020874,
"logps/chosen": -1.8574031591415405,
"logps/rejected": -2.1753742694854736,
"loss": 57.9215,
"nll_loss": 1.967643141746521,
"po_loss": -0.1575944572687149,
"rewards/accuracies": 0.5406249761581421,
"rewards/chosen": -0.9287015795707703,
"rewards/margins": 0.15898552536964417,
"rewards/rejected": -1.0876871347427368,
"step": 35
},
{
"epoch": 0.37925925925925924,
"grad_norm": 194.72329711914062,
"learning_rate": 4.990147841143462e-05,
"log_odds_chosen": 0.4601282477378845,
"log_odds_ratio": -0.6781095266342163,
"logps/chosen": -1.6383177042007446,
"logps/rejected": -2.0322861671447754,
"loss": 54.2986,
"nll_loss": 1.8938146829605103,
"po_loss": -0.1969841569662094,
"rewards/accuracies": 0.643750011920929,
"rewards/chosen": -0.8191588521003723,
"rewards/margins": 0.1969841569662094,
"rewards/rejected": -1.0161430835723877,
"step": 40
},
{
"epoch": 0.4266666666666667,
"grad_norm": 421.30731201171875,
"learning_rate": 4.97401218720448e-05,
"log_odds_chosen": 1.2004860639572144,
"log_odds_ratio": -0.6465462446212769,
"logps/chosen": -1.7445472478866577,
"logps/rejected": -2.871616840362549,
"loss": 42.8079,
"nll_loss": 1.8902437686920166,
"po_loss": -0.5524969100952148,
"rewards/accuracies": 0.690625011920929,
"rewards/chosen": -0.8722736239433289,
"rewards/margins": 0.5635346174240112,
"rewards/rejected": -1.4358084201812744,
"step": 45
},
{
"epoch": 0.4740740740740741,
"grad_norm": 850.62548828125,
"learning_rate": 4.9502564938797946e-05,
"log_odds_chosen": 2.4571447372436523,
"log_odds_ratio": -1.3941503763198853,
"logps/chosen": -3.2160325050354004,
"logps/rejected": -5.586895942687988,
"loss": 68.2657,
"nll_loss": 3.1136634349823,
"po_loss": -0.9803593754768372,
"rewards/accuracies": 0.706250011920929,
"rewards/chosen": -1.6080162525177002,
"rewards/margins": 1.1854315996170044,
"rewards/rejected": -2.793447971343994,
"step": 50
},
{
"epoch": 0.5214814814814814,
"grad_norm": 3458.772216796875,
"learning_rate": 4.918953929490768e-05,
"log_odds_chosen": 3.8432812690734863,
"log_odds_ratio": -0.9803560972213745,
"logps/chosen": -2.737734317779541,
"logps/rejected": -6.494544982910156,
"loss": 47.9796,
"nll_loss": 2.612356662750244,
"po_loss": -1.1129950284957886,
"rewards/accuracies": 0.6875,
"rewards/chosen": -1.3688671588897705,
"rewards/margins": 1.8784055709838867,
"rewards/rejected": -3.247272491455078,
"step": 55
},
{
"epoch": 0.5688888888888889,
"grad_norm": 689.9036254882812,
"learning_rate": 4.88020090697132e-05,
"log_odds_chosen": 4.405730724334717,
"log_odds_ratio": -0.9279032945632935,
"logps/chosen": -2.4398763179779053,
"logps/rejected": -6.7704877853393555,
"loss": 34.5555,
"nll_loss": 2.4739956855773926,
"po_loss": -1.3941353559494019,
"rewards/accuracies": 0.675000011920929,
"rewards/chosen": -1.2199381589889526,
"rewards/margins": 2.1653053760528564,
"rewards/rejected": -3.3852438926696777,
"step": 60
},
{
"epoch": 0.6162962962962963,
"grad_norm": 627.5538330078125,
"learning_rate": 4.834116786912897e-05,
"log_odds_chosen": 3.3358917236328125,
"log_odds_ratio": -1.2476164102554321,
"logps/chosen": -3.2438666820526123,
"logps/rejected": -6.503542423248291,
"loss": 59.5781,
"nll_loss": 3.0055854320526123,
"po_loss": -1.143770456314087,
"rewards/accuracies": 0.65625,
"rewards/chosen": -1.6219333410263062,
"rewards/margins": 1.6298377513885498,
"rewards/rejected": -3.2517712116241455,
"step": 65
},
{
"epoch": 0.6637037037037037,
"grad_norm": 1038.5926513671875,
"learning_rate": 4.7808435099299045e-05,
"log_odds_chosen": 5.5194196701049805,
"log_odds_ratio": -1.1829580068588257,
"logps/chosen": -3.315161943435669,
"logps/rejected": -8.737940788269043,
"loss": 18.1436,
"nll_loss": 2.795347213745117,
"po_loss": -2.228358030319214,
"rewards/accuracies": 0.7437499761581421,
"rewards/chosen": -1.6575809717178345,
"rewards/margins": 2.7113900184631348,
"rewards/rejected": -4.3689703941345215,
"step": 70
},
{
"epoch": 0.7111111111111111,
"grad_norm": 957.5834350585938,
"learning_rate": 4.720545159477922e-05,
"log_odds_chosen": 3.6360058784484863,
"log_odds_ratio": -1.2177813053131104,
"logps/chosen": -2.6733596324920654,
"logps/rejected": -6.228394031524658,
"loss": 39.0704,
"nll_loss": 2.3507373332977295,
"po_loss": -1.1297887563705444,
"rewards/accuracies": 0.6812499761581421,
"rewards/chosen": -1.3366798162460327,
"rewards/margins": 1.7775170803070068,
"rewards/rejected": -3.114197015762329,
"step": 75
},
{
"epoch": 0.7585185185185185,
"grad_norm": 391.593505859375,
"learning_rate": 4.653407456471222e-05,
"log_odds_chosen": 4.495142459869385,
"log_odds_ratio": -1.3136035203933716,
"logps/chosen": -2.7727952003479004,
"logps/rejected": -7.165794372558594,
"loss": 31.409,
"nll_loss": 2.5395021438598633,
"po_loss": -1.5579702854156494,
"rewards/accuracies": 0.628125011920929,
"rewards/chosen": -1.3863976001739502,
"rewards/margins": 2.196500062942505,
"rewards/rejected": -3.582897186279297,
"step": 80
},
{
"epoch": 0.8059259259259259,
"grad_norm": 2327.972900390625,
"learning_rate": 4.579637187256222e-05,
"log_odds_chosen": 1.418734073638916,
"log_odds_ratio": -0.8357670903205872,
"logps/chosen": -1.7513229846954346,
"logps/rejected": -3.0980544090270996,
"loss": 41.072,
"nll_loss": 1.9310455322265625,
"po_loss": -0.6475440859794617,
"rewards/accuracies": 0.625,
"rewards/chosen": -0.8756614923477173,
"rewards/margins": 0.6733657121658325,
"rewards/rejected": -1.5490272045135498,
"step": 85
},
{
"epoch": 0.8533333333333334,
"grad_norm": 778.7096557617188,
"learning_rate": 4.499461566702685e-05,
"log_odds_chosen": 4.9186859130859375,
"log_odds_ratio": -0.8927431106567383,
"logps/chosen": -3.104053497314453,
"logps/rejected": -7.928048610687256,
"loss": 26.255,
"nll_loss": 2.6208713054656982,
"po_loss": -1.8004029989242554,
"rewards/accuracies": 0.706250011920929,
"rewards/chosen": -1.5520267486572266,
"rewards/margins": 2.4119973182678223,
"rewards/rejected": -3.964024305343628,
"step": 90
},
{
"epoch": 0.9007407407407407,
"grad_norm": 1113.090576171875,
"learning_rate": 4.413127538374411e-05,
"log_odds_chosen": 4.415733337402344,
"log_odds_ratio": -1.0066527128219604,
"logps/chosen": -2.505484104156494,
"logps/rejected": -6.809667110443115,
"loss": 15.2194,
"nll_loss": 2.1575310230255127,
"po_loss": -1.6819244623184204,
"rewards/accuracies": 0.6937500238418579,
"rewards/chosen": -1.252742052078247,
"rewards/margins": 2.1520915031433105,
"rewards/rejected": -3.4048335552215576,
"step": 95
},
{
"epoch": 0.9481481481481482,
"grad_norm": 5262.9931640625,
"learning_rate": 4.320901013934887e-05,
"log_odds_chosen": 4.908814430236816,
"log_odds_ratio": -1.2427480220794678,
"logps/chosen": -2.8461787700653076,
"logps/rejected": -7.642964839935303,
"loss": 17.6349,
"nll_loss": 2.5891518592834473,
"po_loss": -2.0380616188049316,
"rewards/accuracies": 0.699999988079071,
"rewards/chosen": -1.4230893850326538,
"rewards/margins": 2.398393154144287,
"rewards/rejected": -3.8214824199676514,
"step": 100
},
{
"epoch": 0.9955555555555555,
"grad_norm": 465.037841796875,
"learning_rate": 4.223066054130568e-05,
"log_odds_chosen": 2.4661428928375244,
"log_odds_ratio": -0.8429304957389832,
"logps/chosen": -1.9539477825164795,
"logps/rejected": -4.3218865394592285,
"loss": 27.8368,
"nll_loss": 2.0108437538146973,
"po_loss": -1.1409438848495483,
"rewards/accuracies": 0.690625011920929,
"rewards/chosen": -0.9769738912582397,
"rewards/margins": 1.1839690208435059,
"rewards/rejected": -2.1609432697296143,
"step": 105
},
{
"epoch": 1.037925925925926,
"grad_norm": 1398.3836669921875,
"learning_rate": 4.1199239938743797e-05,
"log_odds_chosen": 6.5568623542785645,
"log_odds_ratio": -1.0280228853225708,
"logps/chosen": -3.167757511138916,
"logps/rejected": -9.576342582702637,
"loss": 6.365,
"nll_loss": 2.601748466491699,
"po_loss": -2.3791961669921875,
"rewards/accuracies": 0.748251736164093,
"rewards/chosen": -1.583878755569458,
"rewards/margins": 3.2042922973632812,
"rewards/rejected": -4.788171291351318,
"step": 110
},
{
"epoch": 1.0853333333333333,
"grad_norm": 468.6627502441406,
"learning_rate": 4.0117925141242174e-05,
"log_odds_chosen": 7.842375755310059,
"log_odds_ratio": -0.8314769864082336,
"logps/chosen": -2.4292473793029785,
"logps/rejected": -10.092866897583008,
"loss": -10.5443,
"nll_loss": 2.201972007751465,
"po_loss": -2.5314807891845703,
"rewards/accuracies": 0.75,
"rewards/chosen": -1.2146236896514893,
"rewards/margins": 3.8318095207214355,
"rewards/rejected": -5.046433448791504,
"step": 115
},
{
"epoch": 1.1327407407407408,
"grad_norm": 1342.4088134765625,
"learning_rate": 3.899004663415084e-05,
"log_odds_chosen": 6.2360992431640625,
"log_odds_ratio": -0.7881489992141724,
"logps/chosen": -2.0156655311584473,
"logps/rejected": -8.078754425048828,
"loss": -3.799,
"nll_loss": 2.015702247619629,
"po_loss": -2.1344211101531982,
"rewards/accuracies": 0.7562500238418579,
"rewards/chosen": -1.0078327655792236,
"rewards/margins": 3.0315444469451904,
"rewards/rejected": -4.039377212524414,
"step": 120
},
{
"epoch": 1.1801481481481482,
"grad_norm": 780.4815673828125,
"learning_rate": 3.781907832058587e-05,
"log_odds_chosen": 6.6184797286987305,
"log_odds_ratio": -1.2134829759597778,
"logps/chosen": -2.7371468544006348,
"logps/rejected": -9.186931610107422,
"loss": 11.0535,
"nll_loss": 2.5944888591766357,
"po_loss": -2.249067783355713,
"rewards/accuracies": 0.75,
"rewards/chosen": -1.3685734272003174,
"rewards/margins": 3.2248923778533936,
"rewards/rejected": -4.593465805053711,
"step": 125
},
{
"epoch": 1.2275555555555555,
"grad_norm": 827.787353515625,
"learning_rate": 3.660862682169282e-05,
"log_odds_chosen": 8.269857406616211,
"log_odds_ratio": -1.018838882446289,
"logps/chosen": -2.7996513843536377,
"logps/rejected": -10.883502960205078,
"loss": -11.0712,
"nll_loss": 2.5594162940979004,
"po_loss": -2.9053916931152344,
"rewards/accuracies": 0.753125011920929,
"rewards/chosen": -1.3998256921768188,
"rewards/margins": 4.041925430297852,
"rewards/rejected": -5.441751480102539,
"step": 130
},
{
"epoch": 1.274962962962963,
"grad_norm": 602.3118896484375,
"learning_rate": 3.5362420368134356e-05,
"log_odds_chosen": 5.271297454833984,
"log_odds_ratio": -0.828192412853241,
"logps/chosen": -1.9964498281478882,
"logps/rejected": -7.077614784240723,
"loss": 1.6283,
"nll_loss": 2.024073362350464,
"po_loss": -1.973189353942871,
"rewards/accuracies": 0.7437499761581421,
"rewards/chosen": -0.9982249140739441,
"rewards/margins": 2.5405819416046143,
"rewards/rejected": -3.5388073921203613,
"step": 135
},
{
"epoch": 1.3223703703703704,
"grad_norm": 926.779052734375,
"learning_rate": 3.408429731701635e-05,
"log_odds_chosen": 5.869942665100098,
"log_odds_ratio": -0.8702658414840698,
"logps/chosen": -2.663573980331421,
"logps/rejected": -8.38515853881836,
"loss": 0.999,
"nll_loss": 2.1175601482391357,
"po_loss": -2.0863418579101562,
"rewards/accuracies": 0.7437499761581421,
"rewards/chosen": -1.3317869901657104,
"rewards/margins": 2.8607921600341797,
"rewards/rejected": -4.19257926940918,
"step": 140
},
{
"epoch": 1.3697777777777778,
"grad_norm": 1278.476806640625,
"learning_rate": 3.2778194329621104e-05,
"log_odds_chosen": 9.726934432983398,
"log_odds_ratio": -1.0185357332229614,
"logps/chosen": -3.2876574993133545,
"logps/rejected": -12.812705993652344,
"loss": -19.5231,
"nll_loss": 2.5718603134155273,
"po_loss": -3.181957483291626,
"rewards/accuracies": 0.793749988079071,
"rewards/chosen": -1.6438287496566772,
"rewards/margins": 4.762524604797363,
"rewards/rejected": -6.406352996826172,
"step": 145
},
{
"epoch": 1.417185185185185,
"grad_norm": 1008.686279296875,
"learning_rate": 3.144813424636031e-05,
"log_odds_chosen": 7.952484130859375,
"log_odds_ratio": -0.8371334075927734,
"logps/chosen": -2.3361740112304688,
"logps/rejected": -10.076719284057617,
"loss": -22.1519,
"nll_loss": 1.983311653137207,
"po_loss": -2.6755595207214355,
"rewards/accuracies": 0.7875000238418579,
"rewards/chosen": -1.1680870056152344,
"rewards/margins": 3.8702728748321533,
"rewards/rejected": -5.038359642028809,
"step": 150
},
{
"epoch": 1.4645925925925927,
"grad_norm": 789.317138671875,
"learning_rate": 3.0098213696293542e-05,
"log_odds_chosen": 8.040031433105469,
"log_odds_ratio": -1.153850793838501,
"logps/chosen": -2.9109256267547607,
"logps/rejected": -10.789738655090332,
"loss": 0.2885,
"nll_loss": 2.3872437477111816,
"po_loss": -2.378229856491089,
"rewards/accuracies": 0.71875,
"rewards/chosen": -1.4554628133773804,
"rewards/margins": 3.939406156539917,
"rewards/rejected": -5.394869327545166,
"step": 155
},
{
"epoch": 1.512,
"grad_norm": 951.3047485351562,
"learning_rate": 2.8732590479375165e-05,
"log_odds_chosen": 7.912862300872803,
"log_odds_ratio": -0.7059909105300903,
"logps/chosen": -2.1726765632629395,
"logps/rejected": -9.871898651123047,
"loss": -31.1045,
"nll_loss": 2.01442289352417,
"po_loss": -2.9864375591278076,
"rewards/accuracies": 0.768750011920929,
"rewards/chosen": -1.0863382816314697,
"rewards/margins": 3.8496108055114746,
"rewards/rejected": -4.935949325561523,
"step": 160
},
{
"epoch": 1.5594074074074074,
"grad_norm": 2454.81591796875,
"learning_rate": 2.7355470760292956e-05,
"log_odds_chosen": 9.038829803466797,
"log_odds_ratio": -1.0712902545928955,
"logps/chosen": -3.265695095062256,
"logps/rejected": -12.09699821472168,
"loss": -28.5012,
"nll_loss": 2.4790894985198975,
"po_loss": -3.3697502613067627,
"rewards/accuracies": 0.778124988079071,
"rewards/chosen": -1.632847547531128,
"rewards/margins": 4.415651798248291,
"rewards/rejected": -6.04849910736084,
"step": 165
},
{
"epoch": 1.6068148148148147,
"grad_norm": 1764.2366943359375,
"learning_rate": 2.597109611334169e-05,
"log_odds_chosen": 9.091514587402344,
"log_odds_ratio": -1.6616445779800415,
"logps/chosen": -4.427676200866699,
"logps/rejected": -13.332109451293945,
"loss": -3.2606,
"nll_loss": 3.0899510383605957,
"po_loss": -3.191845417022705,
"rewards/accuracies": 0.7718750238418579,
"rewards/chosen": -2.2138381004333496,
"rewards/margins": 4.452216625213623,
"rewards/rejected": -6.666054725646973,
"step": 170
},
{
"epoch": 1.6542222222222223,
"grad_norm": 1194.388916015625,
"learning_rate": 2.458373045823404e-05,
"log_odds_chosen": 7.1326398849487305,
"log_odds_ratio": -0.9163221120834351,
"logps/chosen": -2.580127716064453,
"logps/rejected": -9.50422477722168,
"loss": -18.0293,
"nll_loss": 2.2237651348114014,
"po_loss": -2.787180185317993,
"rewards/accuracies": 0.762499988079071,
"rewards/chosen": -1.2900638580322266,
"rewards/margins": 3.462048292160034,
"rewards/rejected": -4.75211238861084,
"step": 175
},
{
"epoch": 1.7016296296296296,
"grad_norm": 2008.02197265625,
"learning_rate": 2.3197646927086697e-05,
"log_odds_chosen": 7.623831748962402,
"log_odds_ratio": -0.8227715492248535,
"logps/chosen": -2.170476198196411,
"logps/rejected": -9.566916465759277,
"loss": -24.3356,
"nll_loss": 2.127624988555908,
"po_loss": -2.888112783432007,
"rewards/accuracies": 0.768750011920929,
"rewards/chosen": -1.0852380990982056,
"rewards/margins": 3.698220729827881,
"rewards/rejected": -4.783458232879639,
"step": 180
},
{
"epoch": 1.749037037037037,
"grad_norm": 2212.9755859375,
"learning_rate": 2.1817114703032176e-05,
"log_odds_chosen": 11.049670219421387,
"log_odds_ratio": -1.2048530578613281,
"logps/chosen": -3.1207997798919678,
"logps/rejected": -13.965868949890137,
"loss": -42.3188,
"nll_loss": 2.493201732635498,
"po_loss": -3.8156638145446777,
"rewards/accuracies": 0.78125,
"rewards/chosen": -1.5603998899459839,
"rewards/margins": 5.422535419464111,
"rewards/rejected": -6.982934474945068,
"step": 185
},
{
"epoch": 1.7964444444444445,
"grad_norm": 1651.638671875,
"learning_rate": 2.0446385870993467e-05,
"log_odds_chosen": 9.86381721496582,
"log_odds_ratio": -0.9923731088638306,
"logps/chosen": -2.9157769680023193,
"logps/rejected": -12.55068588256836,
"loss": -36.9979,
"nll_loss": 2.172974109649658,
"po_loss": -3.329157590866089,
"rewards/accuracies": 0.796875,
"rewards/chosen": -1.4578884840011597,
"rewards/margins": 4.8174543380737305,
"rewards/rejected": -6.27534294128418,
"step": 190
},
{
"epoch": 1.8438518518518519,
"grad_norm": 5688.65869140625,
"learning_rate": 1.9089682321121834e-05,
"log_odds_chosen": 8.133454322814941,
"log_odds_ratio": -1.3896712064743042,
"logps/chosen": -3.334374189376831,
"logps/rejected": -11.257940292358398,
"loss": -14.291,
"nll_loss": 2.494044303894043,
"po_loss": -2.9406371116638184,
"rewards/accuracies": 0.7906249761581421,
"rewards/chosen": -1.6671870946884155,
"rewards/margins": 3.961782932281494,
"rewards/rejected": -5.628970146179199,
"step": 195
},
{
"epoch": 1.8912592592592592,
"grad_norm": 4854.55810546875,
"learning_rate": 1.775118274523545e-05,
"log_odds_chosen": 8.46328353881836,
"log_odds_ratio": -2.0774197578430176,
"logps/chosen": -4.338870525360107,
"logps/rejected": -12.631368637084961,
"loss": 0.4968,
"nll_loss": 3.0006825923919678,
"po_loss": -2.985157012939453,
"rewards/accuracies": 0.737500011920929,
"rewards/chosen": -2.1694352626800537,
"rewards/margins": 4.1462483406066895,
"rewards/rejected": -6.3156843185424805,
"step": 200
},
{
"epoch": 1.9386666666666668,
"grad_norm": 1195.15283203125,
"learning_rate": 1.643500976631037e-05,
"log_odds_chosen": 8.545049667358398,
"log_odds_ratio": -1.2959668636322021,
"logps/chosen": -3.3477349281311035,
"logps/rejected": -11.675881385803223,
"loss": -24.4773,
"nll_loss": 2.649782419204712,
"po_loss": -3.414698362350464,
"rewards/accuracies": 0.784375011920929,
"rewards/chosen": -1.6738674640655518,
"rewards/margins": 4.1640729904174805,
"rewards/rejected": -5.837940692901611,
"step": 205
},
{
"epoch": 1.986074074074074,
"grad_norm": 1435.34326171875,
"learning_rate": 1.514521724066537e-05,
"log_odds_chosen": 7.497690677642822,
"log_odds_ratio": -1.6079471111297607,
"logps/chosen": -3.8709499835968018,
"logps/rejected": -11.176200866699219,
"loss": -8.5711,
"nll_loss": 2.686655282974243,
"po_loss": -2.9545013904571533,
"rewards/accuracies": 0.762499988079071,
"rewards/chosen": -1.9354749917984009,
"rewards/margins": 3.6526246070861816,
"rewards/rejected": -5.588100433349609,
"step": 210
},
{
"epoch": 2.0284444444444443,
"grad_norm": 784.1499633789062,
"learning_rate": 1.3885777771950348e-05,
"log_odds_chosen": 8.465600967407227,
"log_odds_ratio": -0.868893027305603,
"logps/chosen": -2.0467655658721924,
"logps/rejected": -10.241917610168457,
"loss": -37.4481,
"nll_loss": 1.8826793432235718,
"po_loss": -3.1920523643493652,
"rewards/accuracies": 0.7692307829856873,
"rewards/chosen": -1.0233827829360962,
"rewards/margins": 4.097576141357422,
"rewards/rejected": -5.1209588050842285,
"step": 215
},
{
"epoch": 2.075851851851852,
"grad_norm": 1654.237548828125,
"learning_rate": 1.2660570475395683e-05,
"log_odds_chosen": 8.654611587524414,
"log_odds_ratio": -1.1223597526550293,
"logps/chosen": -2.459672212600708,
"logps/rejected": -10.875265121459961,
"loss": -34.0465,
"nll_loss": 2.0625672340393066,
"po_loss": -3.126519203186035,
"rewards/accuracies": 0.765625,
"rewards/chosen": -1.229836106300354,
"rewards/margins": 4.207796573638916,
"rewards/rejected": -5.4376325607299805,
"step": 220
},
{
"epoch": 2.1232592592592594,
"grad_norm": 2589.6494140625,
"learning_rate": 1.1473369030008974e-05,
"log_odds_chosen": 15.148641586303711,
"log_odds_ratio": -1.0744845867156982,
"logps/chosen": -3.3443615436553955,
"logps/rejected": -18.203109741210938,
"loss": -73.6899,
"nll_loss": 2.520362377166748,
"po_loss": -4.8231706619262695,
"rewards/accuracies": 0.840624988079071,
"rewards/chosen": -1.6721807718276978,
"rewards/margins": 7.429374694824219,
"rewards/rejected": -9.101554870605469,
"step": 225
},
{
"epoch": 2.1706666666666665,
"grad_norm": 1232.9119873046875,
"learning_rate": 1.0327830055518842e-05,
"log_odds_chosen": 14.175605773925781,
"log_odds_ratio": -0.9750823974609375,
"logps/chosen": -3.1031196117401123,
"logps/rejected": -17.016685485839844,
"loss": -66.4024,
"nll_loss": 2.539062976837158,
"po_loss": -4.614138603210449,
"rewards/accuracies": 0.840624988079071,
"rewards/chosen": -1.5515598058700562,
"rewards/margins": 6.956783294677734,
"rewards/rejected": -8.508342742919922,
"step": 230
},
{
"epoch": 2.218074074074074,
"grad_norm": 2281.852294921875,
"learning_rate": 9.227481849865235e-06,
"log_odds_chosen": 13.702494621276855,
"log_odds_ratio": -0.6508721113204956,
"logps/chosen": -2.776512861251831,
"logps/rejected": -16.18694496154785,
"loss": -76.9486,
"nll_loss": 2.2473607063293457,
"po_loss": -4.652005672454834,
"rewards/accuracies": 0.831250011920929,
"rewards/chosen": -1.3882564306259155,
"rewards/margins": 6.705216407775879,
"rewards/rejected": -8.093472480773926,
"step": 235
},
{
"epoch": 2.2654814814814817,
"grad_norm": 1797.7962646484375,
"learning_rate": 8.175713521924978e-06,
"log_odds_chosen": 14.10627555847168,
"log_odds_ratio": -0.9793645739555359,
"logps/chosen": -3.2738196849823,
"logps/rejected": -17.115571975708008,
"loss": -71.8213,
"nll_loss": 2.362848997116089,
"po_loss": -4.607265472412109,
"rewards/accuracies": 0.859375,
"rewards/chosen": -1.63690984249115,
"rewards/margins": 6.92087459564209,
"rewards/rejected": -8.557785987854004,
"step": 240
},
{
"epoch": 2.3128888888888888,
"grad_norm": 2065.531005859375,
"learning_rate": 7.1757645529443665e-06,
"log_odds_chosen": 14.409799575805664,
"log_odds_ratio": -0.6093363165855408,
"logps/chosen": -2.691865921020508,
"logps/rejected": -16.832874298095703,
"loss": -86.8692,
"nll_loss": 2.2543296813964844,
"po_loss": -4.968992710113525,
"rewards/accuracies": 0.8374999761581421,
"rewards/chosen": -1.345932960510254,
"rewards/margins": 7.070504188537598,
"rewards/rejected": -8.416437149047852,
"step": 245
},
{
"epoch": 2.3602962962962963,
"grad_norm": 1674.9385986328125,
"learning_rate": 6.230714818829733e-06,
"log_odds_chosen": 14.006881713867188,
"log_odds_ratio": -0.6611822247505188,
"logps/chosen": -2.298327684402466,
"logps/rejected": -16.025455474853516,
"loss": -73.411,
"nll_loss": 2.0404696464538574,
"po_loss": -4.3345627784729,
"rewards/accuracies": 0.784375011920929,
"rewards/chosen": -1.149163842201233,
"rewards/margins": 6.863563537597656,
"rewards/rejected": -8.012727737426758,
"step": 250
},
{
"epoch": 2.407703703703704,
"grad_norm": 1458.6246337890625,
"learning_rate": 5.343475104027743e-06,
"log_odds_chosen": 14.005816459655762,
"log_odds_ratio": -0.6005424857139587,
"logps/chosen": -2.2070460319519043,
"logps/rejected": -15.93396282196045,
"loss": -76.6993,
"nll_loss": 1.9984962940216064,
"po_loss": -4.395350456237793,
"rewards/accuracies": 0.856249988079071,
"rewards/chosen": -1.1035230159759521,
"rewards/margins": 6.863458156585693,
"rewards/rejected": -7.966981410980225,
"step": 255
},
{
"epoch": 2.455111111111111,
"grad_norm": 1062.482666015625,
"learning_rate": 4.516778136213037e-06,
"log_odds_chosen": 13.041534423828125,
"log_odds_ratio": -0.6385210752487183,
"logps/chosen": -2.050607204437256,
"logps/rejected": -14.810098648071289,
"loss": -73.7325,
"nll_loss": 1.8901704549789429,
"po_loss": -4.19431209564209,
"rewards/accuracies": 0.862500011920929,
"rewards/chosen": -1.025303602218628,
"rewards/margins": 6.379745006561279,
"rewards/rejected": -7.4050493240356445,
"step": 260
},
{
"epoch": 2.5025185185185186,
"grad_norm": 2426.10986328125,
"learning_rate": 3.7531701693965554e-06,
"log_odds_chosen": 16.418107986450195,
"log_odds_ratio": -0.8002341985702515,
"logps/chosen": -2.38873028755188,
"logps/rejected": -18.518217086791992,
"loss": -90.9132,
"nll_loss": 2.055996894836426,
"po_loss": -4.897036075592041,
"rewards/accuracies": 0.84375,
"rewards/chosen": -1.19436514377594,
"rewards/margins": 8.064743995666504,
"rewards/rejected": -9.259108543395996,
"step": 265
},
{
"epoch": 2.549925925925926,
"grad_norm": 2285.034423828125,
"learning_rate": 3.055003141378948e-06,
"log_odds_chosen": 18.129314422607422,
"log_odds_ratio": -1.1312395334243774,
"logps/chosen": -3.370629072189331,
"logps/rejected": -21.226877212524414,
"loss": -80.7968,
"nll_loss": 2.6373114585876465,
"po_loss": -5.162210941314697,
"rewards/accuracies": 0.828125,
"rewards/chosen": -1.6853145360946655,
"rewards/margins": 8.928122520446777,
"rewards/rejected": -10.613438606262207,
"step": 270
},
{
"epoch": 2.5973333333333333,
"grad_norm": 2841.40380859375,
"learning_rate": 2.424427429704365e-06,
"log_odds_chosen": 16.357511520385742,
"log_odds_ratio": -1.1468102931976318,
"logps/chosen": -3.206864595413208,
"logps/rejected": -19.272470474243164,
"loss": -74.3951,
"nll_loss": 2.3466782569885254,
"po_loss": -4.671523571014404,
"rewards/accuracies": 0.856249988079071,
"rewards/chosen": -1.603432297706604,
"rewards/margins": 8.032803535461426,
"rewards/rejected": -9.636235237121582,
"step": 275
},
{
"epoch": 2.644740740740741,
"grad_norm": 1928.48876953125,
"learning_rate": 1.8633852284264508e-06,
"log_odds_chosen": 16.074771881103516,
"log_odds_ratio": -0.9569836854934692,
"logps/chosen": -2.4576973915100098,
"logps/rejected": -18.23931884765625,
"loss": -88.0982,
"nll_loss": 2.244922161102295,
"po_loss": -4.997992515563965,
"rewards/accuracies": 0.828125,
"rewards/chosen": -1.2288486957550049,
"rewards/margins": 7.890810489654541,
"rewards/rejected": -9.119659423828125,
"step": 280
},
{
"epoch": 2.6921481481481484,
"grad_norm": 1673.14990234375,
"learning_rate": 1.3736045660864034e-06,
"log_odds_chosen": 15.582837104797363,
"log_odds_ratio": -0.819076657295227,
"logps/chosen": -2.702998638153076,
"logps/rejected": -18.022306442260742,
"loss": -88.1281,
"nll_loss": 2.126950740814209,
"po_loss": -4.880953788757324,
"rewards/accuracies": 0.8374999761581421,
"rewards/chosen": -1.351499319076538,
"rewards/margins": 7.659654140472412,
"rewards/rejected": -9.011153221130371,
"step": 285
},
{
"epoch": 2.7395555555555555,
"grad_norm": 1918.1820068359375,
"learning_rate": 9.565939833279192e-07,
"log_odds_chosen": 14.058061599731445,
"log_odds_ratio": -1.0425951480865479,
"logps/chosen": -3.089294910430908,
"logps/rejected": -16.88263511657715,
"loss": -59.2928,
"nll_loss": 2.3471436500549316,
"po_loss": -4.200043678283691,
"rewards/accuracies": 0.8343750238418579,
"rewards/chosen": -1.544647455215454,
"rewards/margins": 6.896668434143066,
"rewards/rejected": -8.441317558288574,
"step": 290
},
{
"epoch": 2.786962962962963,
"grad_norm": 1549.3990478515625,
"learning_rate": 6.136378865420872e-07,
"log_odds_chosen": 14.449419975280762,
"log_odds_ratio": -0.8233845829963684,
"logps/chosen": -2.452683448791504,
"logps/rejected": -16.653139114379883,
"loss": -82.7587,
"nll_loss": 1.9743688106536865,
"po_loss": -4.560579299926758,
"rewards/accuracies": 0.8031250238418579,
"rewards/chosen": -1.226341724395752,
"rewards/margins": 7.100228309631348,
"rewards/rejected": -8.326569557189941,
"step": 295
},
{
"epoch": 2.83437037037037,
"grad_norm": 1157.6502685546875,
"learning_rate": 3.45792591853214e-07,
"log_odds_chosen": 15.727914810180664,
"log_odds_ratio": -0.5168970227241516,
"logps/chosen": -2.1314849853515625,
"logps/rejected": -17.58350944519043,
"loss": -96.6356,
"nll_loss": 1.8993088006973267,
"po_loss": -4.9191718101501465,
"rewards/accuracies": 0.8374999761581421,
"rewards/chosen": -1.0657424926757812,
"rewards/margins": 7.72601318359375,
"rewards/rejected": -8.791754722595215,
"step": 300
},
{
"epoch": 2.8817777777777778,
"grad_norm": 3823.998291015625,
"learning_rate": 1.538830716302092e-07,
"log_odds_chosen": 16.228670120239258,
"log_odds_ratio": -0.9718038439750671,
"logps/chosen": -2.4307305812835693,
"logps/rejected": -18.350786209106445,
"loss": -87.9496,
"nll_loss": 2.063523769378662,
"po_loss": -4.811949253082275,
"rewards/accuracies": 0.828125,
"rewards/chosen": -1.2153652906417847,
"rewards/margins": 7.960028171539307,
"rewards/rejected": -9.175393104553223,
"step": 305
},
{
"epoch": 2.9291851851851853,
"grad_norm": 2266.873779296875,
"learning_rate": 3.8500413544415025e-08,
"log_odds_chosen": 16.433456420898438,
"log_odds_ratio": -0.6722828149795532,
"logps/chosen": -2.7718887329101562,
"logps/rejected": -18.93569564819336,
"loss": -83.3821,
"nll_loss": 2.255375623703003,
"po_loss": -4.861066818237305,
"rewards/accuracies": 0.828125,
"rewards/chosen": -1.3859443664550781,
"rewards/margins": 8.081903457641602,
"rewards/rejected": -9.46784782409668,
"step": 310
},
{
"epoch": 2.9765925925925925,
"grad_norm": 1334.365234375,
"learning_rate": 0.0,
"log_odds_chosen": 15.443585395812988,
"log_odds_ratio": -0.6007210612297058,
"logps/chosen": -1.9952449798583984,
"logps/rejected": -17.120744705200195,
"loss": -98.7664,
"nll_loss": 1.7973524332046509,
"po_loss": -4.8838019371032715,
"rewards/accuracies": 0.8500000238418579,
"rewards/chosen": -0.9976224899291992,
"rewards/margins": 7.562750339508057,
"rewards/rejected": -8.560372352600098,
"step": 315
},
{
"epoch": 2.9765925925925925,
"step": 315,
"total_flos": 0.0,
"train_loss": 6.844926292177231,
"train_runtime": 9518.1794,
"train_samples_per_second": 2.128,
"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
}