{ "best_metric": 0.4541548192501068, "best_model_checkpoint": "./mistral/20-04-24-Weni-WeniGPT-Agents-Mistral-1.0.6-SFT-1.0.5-DPO_Experiment on DPO with other hyperparameters and best SFT model of WeniGPT-2_max_steps-366_batch_4_2024-04-20_ppid_9/checkpoint-180", "epoch": 2.926829268292683, "eval_steps": 30, "global_step": 180, "is_hyper_param_search": false, "is_local_process_zero": true, "is_world_process_zero": true, "log_history": [ { "epoch": 0.16, "grad_norm": 8.378021240234375, "learning_rate": 4.0909090909090915e-06, "logits/chosen": -1.830958604812622, "logits/rejected": -1.8507845401763916, "logps/chosen": -28.701984405517578, "logps/rejected": -54.28569793701172, "loss": 0.6924, "rewards/accuracies": 0.20000000298023224, "rewards/chosen": 0.0008967495523393154, "rewards/margins": 0.0014666033675894141, "rewards/rejected": -0.0005698538152500987, "step": 10 }, { "epoch": 0.33, "grad_norm": 5.193418502807617, "learning_rate": 4.887323943661972e-06, "logits/chosen": -1.7550897598266602, "logits/rejected": -1.770708680152893, "logps/chosen": -47.344207763671875, "logps/rejected": -64.0368423461914, "loss": 0.6852, "rewards/accuracies": 0.4000000059604645, "rewards/chosen": 0.017231885343790054, "rewards/margins": 0.01606021076440811, "rewards/rejected": 0.0011716745793819427, "step": 20 }, { "epoch": 0.49, "grad_norm": 7.308932304382324, "learning_rate": 4.746478873239437e-06, "logits/chosen": -1.781267762184143, "logits/rejected": -1.8114898204803467, "logps/chosen": -54.274559020996094, "logps/rejected": -95.20500183105469, "loss": 0.6635, "rewards/accuracies": 0.5, "rewards/chosen": 0.0641159638762474, "rewards/margins": 0.061691801995038986, "rewards/rejected": 0.0024241588544100523, "step": 30 }, { "epoch": 0.49, "eval_logits/chosen": -1.7831767797470093, "eval_logits/rejected": -1.8043663501739502, "eval_logps/chosen": -55.16960906982422, "eval_logps/rejected": -97.32585144042969, "eval_loss": 0.6523757576942444, "eval_rewards/accuracies": 0.4642857015132904, "eval_rewards/chosen": 0.09036973863840103, "eval_rewards/margins": 0.08673857897520065, "eval_rewards/rejected": 0.0036311547737568617, "eval_runtime": 8.141, "eval_samples_per_second": 3.439, "eval_steps_per_second": 1.72, "step": 30 }, { "epoch": 0.65, "grad_norm": 0.0, "learning_rate": 4.6056338028169015e-06, "logits/chosen": -1.889905333518982, "logits/rejected": -1.9024461507797241, "logps/chosen": -27.918941497802734, "logps/rejected": -42.093284606933594, "loss": 0.668, "rewards/accuracies": 0.25, "rewards/chosen": 0.054457180202007294, "rewards/margins": 0.0539846234023571, "rewards/rejected": 0.0004725646285805851, "step": 40 }, { "epoch": 0.81, "grad_norm": 8.53225326538086, "learning_rate": 4.464788732394367e-06, "logits/chosen": -1.8278567790985107, "logits/rejected": -1.849957823753357, "logps/chosen": -43.8238639831543, "logps/rejected": -68.02179718017578, "loss": 0.6358, "rewards/accuracies": 0.3499999940395355, "rewards/chosen": 0.13941256701946259, "rewards/margins": 0.13133978843688965, "rewards/rejected": 0.008072790689766407, "step": 50 }, { "epoch": 0.98, "grad_norm": 9.436968803405762, "learning_rate": 4.3239436619718315e-06, "logits/chosen": -1.805991768836975, "logits/rejected": -1.8437427282333374, "logps/chosen": -43.8873291015625, "logps/rejected": -95.2943115234375, "loss": 0.6026, "rewards/accuracies": 0.44999998807907104, "rewards/chosen": 0.18793432414531708, "rewards/margins": 0.21308371424674988, "rewards/rejected": -0.025149401277303696, "step": 60 }, { "epoch": 0.98, "eval_logits/chosen": -1.7877694368362427, "eval_logits/rejected": -1.8098936080932617, "eval_logps/chosen": -53.567203521728516, "eval_logps/rejected": -97.33795928955078, "eval_loss": 0.5890871286392212, "eval_rewards/accuracies": 0.4642857015132904, "eval_rewards/chosen": 0.25061002373695374, "eval_rewards/margins": 0.2481890469789505, "eval_rewards/rejected": 0.002420984674245119, "eval_runtime": 8.1404, "eval_samples_per_second": 3.44, "eval_steps_per_second": 1.72, "step": 60 }, { "epoch": 1.14, "grad_norm": 0.0, "learning_rate": 4.183098591549296e-06, "logits/chosen": -1.8344879150390625, "logits/rejected": -1.8489716053009033, "logps/chosen": -40.38930892944336, "logps/rejected": -60.9084358215332, "loss": 0.6031, "rewards/accuracies": 0.375, "rewards/chosen": 0.19739331305027008, "rewards/margins": 0.22638121247291565, "rewards/rejected": -0.028987903147935867, "step": 70 }, { "epoch": 1.3, "grad_norm": 5.49536657333374, "learning_rate": 4.042253521126761e-06, "logits/chosen": -1.7903095483779907, "logits/rejected": -1.8362411260604858, "logps/chosen": -44.288116455078125, "logps/rejected": -90.21073913574219, "loss": 0.5357, "rewards/accuracies": 0.4749999940395355, "rewards/chosen": 0.34061312675476074, "rewards/margins": 0.40679749846458435, "rewards/rejected": -0.06618441641330719, "step": 80 }, { "epoch": 1.46, "grad_norm": 13.401692390441895, "learning_rate": 3.901408450704225e-06, "logits/chosen": -1.8004281520843506, "logits/rejected": -1.8247934579849243, "logps/chosen": -42.32465362548828, "logps/rejected": -70.9749984741211, "loss": 0.5387, "rewards/accuracies": 0.4749999940395355, "rewards/chosen": 0.3678433299064636, "rewards/margins": 0.4186524450778961, "rewards/rejected": -0.05080908536911011, "step": 90 }, { "epoch": 1.46, "eval_logits/chosen": -1.7943389415740967, "eval_logits/rejected": -1.8181126117706299, "eval_logps/chosen": -51.677486419677734, "eval_logps/rejected": -97.63689422607422, "eval_loss": 0.529485821723938, "eval_rewards/accuracies": 0.4642857015132904, "eval_rewards/chosen": 0.4395819306373596, "eval_rewards/margins": 0.4670555889606476, "eval_rewards/rejected": -0.027473628520965576, "eval_runtime": 8.1412, "eval_samples_per_second": 3.439, "eval_steps_per_second": 1.72, "step": 90 }, { "epoch": 1.63, "grad_norm": 5.040858745574951, "learning_rate": 3.7605633802816903e-06, "logits/chosen": -1.8601042032241821, "logits/rejected": -1.8790462017059326, "logps/chosen": -43.77570343017578, "logps/rejected": -70.64997863769531, "loss": 0.5466, "rewards/accuracies": 0.4000000059604645, "rewards/chosen": 0.36673134565353394, "rewards/margins": 0.42903366684913635, "rewards/rejected": -0.06230226159095764, "step": 100 }, { "epoch": 1.79, "grad_norm": 11.182683944702148, "learning_rate": 3.6197183098591553e-06, "logits/chosen": -1.8602203130722046, "logits/rejected": -1.8786903619766235, "logps/chosen": -29.601736068725586, "logps/rejected": -66.1338882446289, "loss": 0.6003, "rewards/accuracies": 0.2750000059604645, "rewards/chosen": 0.3122637867927551, "rewards/margins": 0.2756831645965576, "rewards/rejected": 0.03658062964677811, "step": 110 }, { "epoch": 1.95, "grad_norm": 3.9169583320617676, "learning_rate": 3.47887323943662e-06, "logits/chosen": -1.8304624557495117, "logits/rejected": -1.8451646566390991, "logps/chosen": -31.413599014282227, "logps/rejected": -56.841880798339844, "loss": 0.6033, "rewards/accuracies": 0.2750000059604645, "rewards/chosen": 0.21899382770061493, "rewards/margins": 0.2744571566581726, "rewards/rejected": -0.05546332150697708, "step": 120 }, { "epoch": 1.95, "eval_logits/chosen": -1.80086350440979, "eval_logits/rejected": -1.8260576725006104, "eval_logps/chosen": -50.32191848754883, "eval_logps/rejected": -98.02101135253906, "eval_loss": 0.49604225158691406, "eval_rewards/accuracies": 0.4642857015132904, "eval_rewards/chosen": 0.5751391053199768, "eval_rewards/margins": 0.6410244107246399, "eval_rewards/rejected": -0.0658852607011795, "eval_runtime": 8.1445, "eval_samples_per_second": 3.438, "eval_steps_per_second": 1.719, "step": 120 }, { "epoch": 2.11, "grad_norm": 1.4047716856002808, "learning_rate": 3.338028169014085e-06, "logits/chosen": -1.8776130676269531, "logits/rejected": -1.8995519876480103, "logps/chosen": -22.69371795654297, "logps/rejected": -53.5282096862793, "loss": 0.5611, "rewards/accuracies": 0.25, "rewards/chosen": 0.35938918590545654, "rewards/margins": 0.5045264959335327, "rewards/rejected": -0.14513733983039856, "step": 130 }, { "epoch": 2.28, "grad_norm": 0.7528722882270813, "learning_rate": 3.1971830985915496e-06, "logits/chosen": -1.8126357793807983, "logits/rejected": -1.832371711730957, "logps/chosen": -38.33379364013672, "logps/rejected": -67.96979522705078, "loss": 0.5142, "rewards/accuracies": 0.375, "rewards/chosen": 0.5568062663078308, "rewards/margins": 0.6818712949752808, "rewards/rejected": -0.12506499886512756, "step": 140 }, { "epoch": 2.44, "grad_norm": 3.405579090118408, "learning_rate": 3.056338028169014e-06, "logits/chosen": -1.8196109533309937, "logits/rejected": -1.8556429147720337, "logps/chosen": -36.78864669799805, "logps/rejected": -83.05890655517578, "loss": 0.5042, "rewards/accuracies": 0.4000000059604645, "rewards/chosen": 0.542107105255127, "rewards/margins": 0.6411095857620239, "rewards/rejected": -0.09900249540805817, "step": 150 }, { "epoch": 2.44, "eval_logits/chosen": -1.805869698524475, "eval_logits/rejected": -1.8330577611923218, "eval_logps/chosen": -49.10601043701172, "eval_logps/rejected": -98.84068298339844, "eval_loss": 0.4709201455116272, "eval_rewards/accuracies": 0.4642857015132904, "eval_rewards/chosen": 0.6967297196388245, "eval_rewards/margins": 0.8445812463760376, "eval_rewards/rejected": -0.1478516012430191, "eval_runtime": 8.1382, "eval_samples_per_second": 3.441, "eval_steps_per_second": 1.72, "step": 150 }, { "epoch": 2.6, "grad_norm": 7.778740882873535, "learning_rate": 2.915492957746479e-06, "logits/chosen": -1.848589301109314, "logits/rejected": -1.8790754079818726, "logps/chosen": -36.49171447753906, "logps/rejected": -72.55968475341797, "loss": 0.4927, "rewards/accuracies": 0.44999998807907104, "rewards/chosen": 0.49555450677871704, "rewards/margins": 0.6891830563545227, "rewards/rejected": -0.1936284601688385, "step": 160 }, { "epoch": 2.76, "grad_norm": 4.058627605438232, "learning_rate": 2.774647887323944e-06, "logits/chosen": -1.812421441078186, "logits/rejected": -1.8415311574935913, "logps/chosen": -45.62999725341797, "logps/rejected": -87.85527038574219, "loss": 0.4541, "rewards/accuracies": 0.4749999940395355, "rewards/chosen": 0.7084562182426453, "rewards/margins": 0.9553689956665039, "rewards/rejected": -0.24691279232501984, "step": 170 }, { "epoch": 2.93, "grad_norm": 0.0, "learning_rate": 2.6338028169014084e-06, "logits/chosen": -1.8475942611694336, "logits/rejected": -1.8678725957870483, "logps/chosen": -40.53328323364258, "logps/rejected": -64.86616516113281, "loss": 0.5087, "rewards/accuracies": 0.375, "rewards/chosen": 0.5022943019866943, "rewards/margins": 0.7252141833305359, "rewards/rejected": -0.22291991114616394, "step": 180 }, { "epoch": 2.93, "eval_logits/chosen": -1.8136398792266846, "eval_logits/rejected": -1.8424787521362305, "eval_logps/chosen": -48.19547653198242, "eval_logps/rejected": -99.7900161743164, "eval_loss": 0.4541548192501068, "eval_rewards/accuracies": 0.4642857015132904, "eval_rewards/chosen": 0.7877826690673828, "eval_rewards/margins": 1.0305674076080322, "eval_rewards/rejected": -0.24278469383716583, "eval_runtime": 8.1397, "eval_samples_per_second": 3.44, "eval_steps_per_second": 1.72, "step": 180 } ], "logging_steps": 10, "max_steps": 366, "num_input_tokens_seen": 0, "num_train_epochs": 6, "save_steps": 90, "total_flos": 0.0, "train_batch_size": 2, "trial_name": null, "trial_params": null }