Model save

README.md

@@ -1,11 +1,8 @@
 ---
 base_model: deepseek-ai/deepseek-math-7b-base
-datasets:
-- zfz1/my_preference_gsm8k_deepseek
 library_name: peft
 license: other
 tags:
-- alignment-handbook
 - trl
 - orpo
 - generated_from_trainer
@@ -17,23 +14,10 @@ model-index:
 <!-- This model card has been generated automatically according to the information the Trainer had access to. You
 should probably proofread and complete it, then remove this comment. -->
 
-[<img src="https://raw.githubusercontent.com/wandb/assets/main/wandb-github-badge-28.svg" alt="Visualize in Weights & Biases" width="200" height="32"/>](https://wandb.ai/thuzfz1/huggingface/runs/zs68qfsi)
+[<img src="https://raw.githubusercontent.com/wandb/assets/main/wandb-github-badge-28.svg" alt="Visualize in Weights & Biases" width="200" height="32"/>](https://wandb.ai/thuzfz1/huggingface/runs/z4q90sz1)
 # deepseek-8b-orpo-lora
 
-This model is a fine-tuned version of [deepseek-ai/deepseek-math-7b-base](https://huggingface.co/deepseek-ai/deepseek-math-7b-base) on the zfz1/my_preference_gsm8k_deepseek dataset.
-It achieves the following results on the evaluation set:
-- Loss: 1.1456
-- Rewards/chosen: -0.0726
-- Rewards/rejected: -0.0757
-- Rewards/accuracies: 0.5647
-- Rewards/margins: 0.0031
-- Logps/rejected: -0.7569
-- Logps/chosen: -0.7262
-- Logits/rejected: 30.3288
-- Logits/chosen: 30.4427
-- Nll Loss: 1.0719
-- Log Odds Ratio: -0.6909
-- Log Odds Chosen: 0.0553
+This model is a fine-tuned version of [deepseek-ai/deepseek-math-7b-base](https://huggingface.co/deepseek-ai/deepseek-math-7b-base) on an unknown dataset.
 
 ## Model description
 
@@ -52,7 +36,7 @@ More information needed
 ### Training hyperparameters
 
 The following hyperparameters were used during training:
-- learning_rate: 3e-06
+- learning_rate: 0.0001
 - train_batch_size: 16
 - eval_batch_size: 16
 - seed: 43

all_results.json

@@ -1,25 +1,9 @@
 {
     "epoch": 1.9936102236421727,
-    "eval_log_odds_chosen": 0.05526316165924072,
-    "eval_log_odds_ratio": -0.6908599138259888,
-    "eval_logits/chosen": 30.44268798828125,
-    "eval_logits/rejected": 30.32880210876465,
-    "eval_logps/chosen": -0.7262152433395386,
-    "eval_logps/rejected": -0.7568790316581726,
-    "eval_loss": 1.1456295251846313,
-    "eval_nll_loss": 1.071852684020996,
-    "eval_rewards/accuracies": 0.5647059082984924,
-    "eval_rewards/chosen": -0.07262152433395386,
-    "eval_rewards/margins": 0.0030663711950182915,
-    "eval_rewards/rejected": -0.07568789273500443,
-    "eval_runtime": 247.9026,
-    "eval_samples": 5398,
-    "eval_samples_per_second": 21.775,
-    "eval_steps_per_second": 0.343,
     "total_flos": 0.0,
-    "train_loss": 1.3676127867820935,
-    "train_runtime": 5524.9246,
+    "train_loss": 0.7846963420892373,
+    "train_runtime": 5577.3844,
     "train_samples": 20000,
-    "train_samples_per_second": 7.24,
+    "train_samples_per_second": 7.172,
     "train_steps_per_second": 0.056
 }

train_results.json

@@ -1,9 +1,9 @@
 {
     "epoch": 1.9936102236421727,
     "total_flos": 0.0,
-    "train_loss": 1.3676127867820935,
-    "train_runtime": 5524.9246,
+    "train_loss": 0.7846963420892373,
+    "train_runtime": 5577.3844,
     "train_samples": 20000,
-    "train_samples_per_second": 7.24,
+    "train_samples_per_second": 7.172,
     "train_steps_per_second": 0.056
 }

trainer_state.json

@@ -10,569 +10,569 @@
   "log_history": [
     {
       "epoch": 0.06389776357827476,
-      "grad_norm": 0.341777060295131,
-      "learning_rate": 9.375e-07,
-      "log_odds_chosen": 0.05486620217561722,
-      "log_odds_ratio": -0.6975381374359131,
-      "logits/chosen": 33.646114349365234,
-      "logits/rejected": 33.55038833618164,
-      "logps/chosen": -0.9772504568099976,
-      "logps/rejected": -1.0204073190689087,
-      "loss": 1.6735,
-      "nll_loss": 1.5879063606262207,
-      "rewards/accuracies": 0.543749988079071,
-      "rewards/chosen": -0.09772505611181259,
-      "rewards/margins": 0.004315672442317009,
-      "rewards/rejected": -0.10204073041677475,
+      "grad_norm": 0.4242139016276009,
+      "learning_rate": 3.125e-05,
+      "log_odds_chosen": 0.05670202523469925,
+      "log_odds_ratio": -0.6963189840316772,
+      "logits/chosen": 33.65892791748047,
+      "logits/rejected": 33.56386947631836,
+      "logps/chosen": -0.9726333618164062,
+      "logps/rejected": -1.0167349576950073,
+      "loss": 1.6695,
+      "nll_loss": 1.5828170776367188,
+      "rewards/accuracies": 0.5406249761581421,
+      "rewards/chosen": -0.09726335108280182,
+      "rewards/margins": 0.004410145338624716,
+      "rewards/rejected": -0.10167349874973297,
       "step": 10
     },
     {
       "epoch": 0.12779552715654952,
-      "grad_norm": 0.357474094120966,
-      "learning_rate": 1.875e-06,
-      "log_odds_chosen": 0.054913729429244995,
-      "log_odds_ratio": -0.6985839009284973,
-      "logits/chosen": 34.386383056640625,
-      "logits/rejected": 33.98283386230469,
-      "logps/chosen": -0.9762754440307617,
-      "logps/rejected": -1.0154238939285278,
-      "loss": 1.668,
-      "nll_loss": 1.6016145944595337,
-      "rewards/accuracies": 0.515625,
-      "rewards/chosen": -0.09762755781412125,
-      "rewards/margins": 0.003914830274879932,
-      "rewards/rejected": -0.10154237598180771,
+      "grad_norm": 0.527757575829653,
+      "learning_rate": 6.25e-05,
+      "log_odds_chosen": 0.060249220579862595,
+      "log_odds_ratio": -0.6938394904136658,
+      "logits/chosen": 34.7052001953125,
+      "logits/rejected": 34.30677032470703,
+      "logps/chosen": -0.8907906413078308,
+      "logps/rejected": -0.9303584098815918,
+      "loss": 1.4944,
+      "nll_loss": 1.4315097332000732,
+      "rewards/accuracies": 0.546875,
+      "rewards/chosen": -0.08907906711101532,
+      "rewards/margins": 0.003956770058721304,
+      "rewards/rejected": -0.09303583949804306,
       "step": 20
     },
     {
       "epoch": 0.19169329073482427,
-      "grad_norm": 0.3765211315354108,
-      "learning_rate": 2.8125e-06,
-      "log_odds_chosen": 0.06909728795289993,
-      "log_odds_ratio": -0.6911064386367798,
-      "logits/chosen": 33.32851028442383,
-      "logits/rejected": 32.749427795410156,
-      "logps/chosen": -0.9813788533210754,
-      "logps/rejected": -1.0349732637405396,
-      "loss": 1.6757,
-      "nll_loss": 1.6085166931152344,
-      "rewards/accuracies": 0.5249999761581421,
-      "rewards/chosen": -0.09813789278268814,
-      "rewards/margins": 0.005359448026865721,
-      "rewards/rejected": -0.10349734127521515,
+      "grad_norm": 0.4014606132018809,
+      "learning_rate": 9.375e-05,
+      "log_odds_chosen": 0.04292842000722885,
+      "log_odds_ratio": -0.698261022567749,
+      "logits/chosen": 30.987014770507812,
+      "logits/rejected": 30.439056396484375,
+      "logps/chosen": -0.6581910252571106,
+      "logps/rejected": -0.6830389499664307,
+      "loss": 1.0721,
+      "nll_loss": 1.006190538406372,
+      "rewards/accuracies": 0.5375000238418579,
+      "rewards/chosen": -0.06581910699605942,
+      "rewards/margins": 0.002484795870259404,
+      "rewards/rejected": -0.0683038979768753,
       "step": 30
     },
     {
       "epoch": 0.25559105431309903,
-      "grad_norm": 0.39833625263223615,
-      "learning_rate": 2.993961440992859e-06,
-      "log_odds_chosen": 0.11841567605733871,
-      "log_odds_ratio": -0.6757725477218628,
-      "logits/chosen": 33.839942932128906,
-      "logits/rejected": 33.162986755371094,
-      "logps/chosen": -0.9702426791191101,
-      "logps/rejected": -1.0581519603729248,
-      "loss": 1.664,
-      "nll_loss": 1.6022984981536865,
-      "rewards/accuracies": 0.6031249761581421,
-      "rewards/chosen": -0.09702426195144653,
-      "rewards/margins": 0.008790932595729828,
-      "rewards/rejected": -0.10581519454717636,
+      "grad_norm": 0.09410150706534143,
+      "learning_rate": 9.979871469976196e-05,
+      "log_odds_chosen": 0.07667034864425659,
+      "log_odds_ratio": -0.6961523294448853,
+      "logits/chosen": 28.151453018188477,
+      "logits/rejected": 27.650014877319336,
+      "logps/chosen": -0.43666666746139526,
+      "logps/rejected": -0.46118512749671936,
+      "loss": 0.8277,
+      "nll_loss": 0.7572250366210938,
+      "rewards/accuracies": 0.5687500238418579,
+      "rewards/chosen": -0.043666668236255646,
+      "rewards/margins": 0.0024518456775695086,
+      "rewards/rejected": -0.046118512749671936,
       "step": 40
     },
     {
       "epoch": 0.3194888178913738,
-      "grad_norm": 0.45952260963075464,
-      "learning_rate": 2.9695130976348534e-06,
-      "log_odds_chosen": 0.09267185628414154,
-      "log_odds_ratio": -0.684535801410675,
-      "logits/chosen": 34.052207946777344,
-      "logits/rejected": 32.802547454833984,
-      "logps/chosen": -0.9801093935966492,
-      "logps/rejected": -1.0446739196777344,
-      "loss": 1.6582,
-      "nll_loss": 1.5952296257019043,
-      "rewards/accuracies": 0.565625011920929,
-      "rewards/chosen": -0.09801094233989716,
-      "rewards/margins": 0.0064564356580376625,
-      "rewards/rejected": -0.10446737706661224,
+      "grad_norm": 0.09768125665941661,
+      "learning_rate": 9.898376992116179e-05,
+      "log_odds_chosen": 0.06605090945959091,
+      "log_odds_ratio": -0.7164387106895447,
+      "logits/chosen": 29.047494888305664,
+      "logits/rejected": 27.90814781188965,
+      "logps/chosen": -0.37932097911834717,
+      "logps/rejected": -0.39148497581481934,
+      "loss": 0.7778,
+      "nll_loss": 0.7044206261634827,
+      "rewards/accuracies": 0.5375000238418579,
+      "rewards/chosen": -0.03793209791183472,
+      "rewards/margins": 0.0012163992505520582,
+      "rewards/rejected": -0.03914849832653999,
       "step": 50
     },
     {
       "epoch": 0.38338658146964855,
-      "grad_norm": 0.49504068932785444,
-      "learning_rate": 2.9265847744427307e-06,
-      "log_odds_chosen": 0.06481704860925674,
-      "log_odds_ratio": -0.698478102684021,
-      "logits/chosen": 32.32544708251953,
-      "logits/rejected": 32.51976776123047,
-      "logps/chosen": -0.9770153164863586,
-      "logps/rejected": -1.0272716283798218,
-      "loss": 1.6387,
-      "nll_loss": 1.572589635848999,
-      "rewards/accuracies": 0.5687500238418579,
-      "rewards/chosen": -0.0977015346288681,
-      "rewards/margins": 0.0050256275571882725,
-      "rewards/rejected": -0.10272715240716934,
+      "grad_norm": 0.07343907441931204,
+      "learning_rate": 9.755282581475769e-05,
+      "log_odds_chosen": 0.08987905830144882,
+      "log_odds_ratio": -0.7063173055648804,
+      "logits/chosen": 28.47494125366211,
+      "logits/rejected": 28.794668197631836,
+      "logps/chosen": -0.3645591139793396,
+      "logps/rejected": -0.38652682304382324,
+      "loss": 0.7576,
+      "nll_loss": 0.6854395866394043,
+      "rewards/accuracies": 0.559374988079071,
+      "rewards/chosen": -0.0364559181034565,
+      "rewards/margins": 0.0021967687644064426,
+      "rewards/rejected": -0.038652680814266205,
       "step": 60
     },
     {
       "epoch": 0.4472843450479233,
-      "grad_norm": 0.5262433549192981,
-      "learning_rate": 2.865716319988224e-06,
-      "log_odds_chosen": 0.030134931206703186,
-      "log_odds_ratio": -0.7101849317550659,
-      "logits/chosen": 33.02894973754883,
-      "logits/rejected": 33.309715270996094,
-      "logps/chosen": -0.9551456570625305,
-      "logps/rejected": -0.9776851534843445,
-      "loss": 1.6289,
-      "nll_loss": 1.5476250648498535,
-      "rewards/accuracies": 0.5531250238418579,
-      "rewards/chosen": -0.09551456570625305,
-      "rewards/margins": 0.0022539461497217417,
-      "rewards/rejected": -0.09776850789785385,
+      "grad_norm": 0.06951733417216101,
+      "learning_rate": 9.552387733294081e-05,
+      "log_odds_chosen": 0.1109728068113327,
+      "log_odds_ratio": -0.7004402875900269,
+      "logits/chosen": 29.863407135009766,
+      "logits/rejected": 30.27783203125,
+      "logps/chosen": -0.35166820883750916,
+      "logps/rejected": -0.3710673451423645,
+      "loss": 0.7621,
+      "nll_loss": 0.6829724907875061,
+      "rewards/accuracies": 0.559374988079071,
+      "rewards/chosen": -0.035166822373867035,
+      "rewards/margins": 0.0019399106968194246,
+      "rewards/rejected": -0.03710673004388809,
       "step": 70
     },
     {
       "epoch": 0.5111821086261981,
-      "grad_norm": 0.5188886669640671,
-      "learning_rate": 2.7876731904027993e-06,
-      "log_odds_chosen": 0.08591620624065399,
-      "log_odds_ratio": -0.6787055730819702,
-      "logits/chosen": 34.61581039428711,
-      "logits/rejected": 34.63213348388672,
-      "logps/chosen": -0.9702480435371399,
-      "logps/rejected": -1.0235538482666016,
-      "loss": 1.5921,
-      "nll_loss": 1.5600519180297852,
-      "rewards/accuracies": 0.5625,
-      "rewards/chosen": -0.0970248132944107,
-      "rewards/margins": 0.005330582614988089,
-      "rewards/rejected": -0.10235539823770523,
+      "grad_norm": 0.06675882825106033,
+      "learning_rate": 9.292243968009331e-05,
+      "log_odds_chosen": 0.22688157856464386,
+      "log_odds_ratio": -0.6433924436569214,
+      "logits/chosen": 31.850915908813477,
+      "logits/rejected": 32.051448822021484,
+      "logps/chosen": -0.35137858986854553,
+      "logps/rejected": -0.4119800925254822,
+      "loss": 0.7548,
+      "nll_loss": 0.6953305602073669,
+      "rewards/accuracies": 0.640625,
+      "rewards/chosen": -0.03513786196708679,
+      "rewards/margins": 0.006060150917619467,
+      "rewards/rejected": -0.0411980114877224,
       "step": 80
     },
     {
       "epoch": 0.5750798722044729,
-      "grad_norm": 0.5939864943475555,
-      "learning_rate": 2.6934368233226715e-06,
-      "log_odds_chosen": 0.0763852447271347,
-      "log_odds_ratio": -0.6891772150993347,
-      "logits/chosen": 34.269248962402344,
-      "logits/rejected": 34.00069046020508,
-      "logps/chosen": -0.9529930949211121,
-      "logps/rejected": -1.0018848180770874,
-      "loss": 1.586,
-      "nll_loss": 1.5087509155273438,
-      "rewards/accuracies": 0.5687500238418579,
-      "rewards/chosen": -0.09529931843280792,
-      "rewards/margins": 0.004889167379587889,
-      "rewards/rejected": -0.1001884788274765,
+      "grad_norm": 0.07132675829804414,
+      "learning_rate": 8.978122744408906e-05,
+      "log_odds_chosen": 0.20013344287872314,
+      "log_odds_ratio": -0.6580983400344849,
+      "logits/chosen": 32.63883972167969,
+      "logits/rejected": 32.50263595581055,
+      "logps/chosen": -0.35542625188827515,
+      "logps/rejected": -0.41074585914611816,
+      "loss": 0.7549,
+      "nll_loss": 0.6860114932060242,
+      "rewards/accuracies": 0.6312500238418579,
+      "rewards/chosen": -0.03554262965917587,
+      "rewards/margins": 0.0055319564417004585,
+      "rewards/rejected": -0.04107458516955376,
       "step": 90
     },
     {
       "epoch": 0.6389776357827476,
-      "grad_norm": 0.6337044866684244,
-      "learning_rate": 2.584192295741087e-06,
-      "log_odds_chosen": 0.057517312467098236,
-      "log_odds_ratio": -0.7008931040763855,
-      "logits/chosen": 32.44559860229492,
-      "logits/rejected": 32.7852783203125,
-      "logps/chosen": -0.9386932253837585,
-      "logps/rejected": -0.9813023805618286,
-      "loss": 1.5499,
-      "nll_loss": 1.4884004592895508,
-      "rewards/accuracies": 0.5562499761581421,
-      "rewards/chosen": -0.09386932104825974,
-      "rewards/margins": 0.004260920453816652,
-      "rewards/rejected": -0.0981302410364151,
+      "grad_norm": 0.09149445807314589,
+      "learning_rate": 8.613974319136958e-05,
+      "log_odds_chosen": 0.2682611346244812,
+      "log_odds_ratio": -0.6336508989334106,
+      "logits/chosen": 32.11505889892578,
+      "logits/rejected": 32.7309455871582,
+      "logps/chosen": -0.35972970724105835,
+      "logps/rejected": -0.4426427483558655,
+      "loss": 0.753,
+      "nll_loss": 0.6944054365158081,
+      "rewards/accuracies": 0.653124988079071,
+      "rewards/chosen": -0.035972971469163895,
+      "rewards/margins": 0.008291301317512989,
+      "rewards/rejected": -0.04426427185535431,
       "step": 100
     },
     {
       "epoch": 0.7028753993610224,
-      "grad_norm": 0.658610355052368,
-      "learning_rate": 2.461313420977536e-06,
-      "log_odds_chosen": 0.06078052520751953,
-      "log_odds_ratio": -0.694057822227478,
-      "logits/chosen": 34.87937927246094,
-      "logits/rejected": 34.10810470581055,
-      "logps/chosen": -0.9328826665878296,
-      "logps/rejected": -0.9768407940864563,
-      "loss": 1.5083,
-      "nll_loss": 1.4501432180404663,
-      "rewards/accuracies": 0.543749988079071,
-      "rewards/chosen": -0.09328827261924744,
-      "rewards/margins": 0.004395808558911085,
-      "rewards/rejected": -0.09768407791852951,
+      "grad_norm": 0.08926217777865493,
+      "learning_rate": 8.20437806992512e-05,
+      "log_odds_chosen": 0.2868707776069641,
+      "log_odds_ratio": -0.6233163475990295,
+      "logits/chosen": 35.566490173339844,
+      "logits/rejected": 34.921634674072266,
+      "logps/chosen": -0.3582688868045807,
+      "logps/rejected": -0.4581901431083679,
+      "loss": 0.744,
+      "nll_loss": 0.6832225918769836,
+      "rewards/accuracies": 0.6343749761581421,
+      "rewards/chosen": -0.03582689166069031,
+      "rewards/margins": 0.00999213196337223,
+      "rewards/rejected": -0.04581902176141739,
       "step": 110
     },
     {
       "epoch": 0.7667731629392971,
-      "grad_norm": 0.7021451570051785,
-      "learning_rate": 2.3263454721781537e-06,
-      "log_odds_chosen": 0.012948527932167053,
-      "log_odds_ratio": -0.7222181558609009,
-      "logits/chosen": 32.433719635009766,
-      "logits/rejected": 32.53590774536133,
-      "logps/chosen": -0.9092488288879395,
-      "logps/rejected": -0.9225506782531738,
-      "loss": 1.4673,
-      "nll_loss": 1.4067564010620117,
-      "rewards/accuracies": 0.4937500059604645,
-      "rewards/chosen": -0.09092487394809723,
-      "rewards/margins": 0.0013302009319886565,
-      "rewards/rejected": -0.09225507825613022,
+      "grad_norm": 0.14837704688840048,
+      "learning_rate": 7.754484907260513e-05,
+      "log_odds_chosen": 0.30330824851989746,
+      "log_odds_ratio": -0.6247963905334473,
+      "logits/chosen": 34.01176834106445,
+      "logits/rejected": 34.39574432373047,
+      "logps/chosen": -0.35339340567588806,
+      "logps/rejected": -0.457806259393692,
+      "loss": 0.7419,
+      "nll_loss": 0.6851586103439331,
+      "rewards/accuracies": 0.6187499761581421,
+      "rewards/chosen": -0.035339340567588806,
+      "rewards/margins": 0.01044128555804491,
+      "rewards/rejected": -0.045780621469020844,
       "step": 120
     },
     {
       "epoch": 0.8306709265175719,
-      "grad_norm": 0.6227085414802311,
-      "learning_rate": 2.18098574960932e-06,
-      "log_odds_chosen": 0.08671535551548004,
-      "log_odds_ratio": -0.6804493069648743,
-      "logits/chosen": 33.88345718383789,
-      "logits/rejected": 33.001914978027344,
-      "logps/chosen": -0.8342811465263367,
-      "logps/rejected": -0.8885458707809448,
-      "loss": 1.3999,
-      "nll_loss": 1.3158425092697144,
-      "rewards/accuracies": 0.581250011920929,
-      "rewards/chosen": -0.08342811465263367,
-      "rewards/margins": 0.005426469258964062,
-      "rewards/rejected": -0.0888545885682106,
+      "grad_norm": 0.10699968939091851,
+      "learning_rate": 7.269952498697734e-05,
+      "log_odds_chosen": 0.3560214638710022,
+      "log_odds_ratio": -0.6051545143127441,
+      "logits/chosen": 36.380924224853516,
+      "logits/rejected": 35.524024963378906,
+      "logps/chosen": -0.355410635471344,
+      "logps/rejected": -0.47948652505874634,
+      "loss": 0.7296,
+      "nll_loss": 0.6667423844337463,
+      "rewards/accuracies": 0.668749988079071,
+      "rewards/chosen": -0.03554106876254082,
+      "rewards/margins": 0.01240758690983057,
+      "rewards/rejected": -0.04794865846633911,
       "step": 130
     },
     {
       "epoch": 0.8945686900958466,
-      "grad_norm": 0.6264864351444569,
-      "learning_rate": 2.027062236122014e-06,
-      "log_odds_chosen": 0.04714164510369301,
-      "log_odds_ratio": -0.6999929547309875,
-      "logits/chosen": 31.832406997680664,
-      "logits/rejected": 31.350921630859375,
-      "logps/chosen": -0.8387459516525269,
-      "logps/rejected": -0.876442551612854,
-      "loss": 1.3491,
-      "nll_loss": 1.2940082550048828,
-      "rewards/accuracies": 0.534375011920929,
-      "rewards/chosen": -0.08387459814548492,
-      "rewards/margins": 0.0037696503568440676,
-      "rewards/rejected": -0.08764425665140152,
+      "grad_norm": 0.1720595584929413,
+      "learning_rate": 6.756874120406714e-05,
+      "log_odds_chosen": 0.5141419172286987,
+      "log_odds_ratio": -0.5401668548583984,
+      "logits/chosen": 34.641624450683594,
+      "logits/rejected": 34.34906768798828,
+      "logps/chosen": -0.347829669713974,
+      "logps/rejected": -0.5514229536056519,
+      "loss": 0.7143,
+      "nll_loss": 0.6645579934120178,
+      "rewards/accuracies": 0.731249988079071,
+      "rewards/chosen": -0.03478296846151352,
+      "rewards/margins": 0.020359333604574203,
+      "rewards/rejected": -0.05514230206608772,
       "step": 140
     },
     {
       "epoch": 0.9584664536741214,
-      "grad_norm": 0.5655314348380543,
-      "learning_rate": 1.866510609206841e-06,
-      "log_odds_chosen": 0.044148243963718414,
-      "log_odds_ratio": -0.7019084692001343,
-      "logits/chosen": 33.6451416015625,
-      "logits/rejected": 33.04177474975586,
-      "logps/chosen": -0.8360700607299805,
-      "logps/rejected": -0.8655373454093933,
-      "loss": 1.3242,
-      "nll_loss": 1.2609388828277588,
-      "rewards/accuracies": 0.53125,
-      "rewards/chosen": -0.08360700309276581,
-      "rewards/margins": 0.0029467367567121983,
-      "rewards/rejected": -0.08655373752117157,
+      "grad_norm": 0.12630063541107736,
+      "learning_rate": 6.22170203068947e-05,
+      "log_odds_chosen": 0.5517296195030212,
+      "log_odds_ratio": -0.543639600276947,
+      "logits/chosen": 36.98326873779297,
+      "logits/rejected": 36.52146530151367,
+      "logps/chosen": -0.3562212884426117,
+      "logps/rejected": -0.578203022480011,
+      "loss": 0.7162,
+      "nll_loss": 0.6639058589935303,
+      "rewards/accuracies": 0.731249988079071,
+      "rewards/chosen": -0.03562213480472565,
+      "rewards/margins": 0.02219817042350769,
+      "rewards/rejected": -0.05782030150294304,
       "step": 150
     },
     {
       "epoch": 1.0223642172523961,
-      "grad_norm": 0.5571454673635674,
-      "learning_rate": 1.7013498987264833e-06,
-      "log_odds_chosen": 0.06723789870738983,
-      "log_odds_ratio": -0.6888397932052612,
-      "logits/chosen": 33.27342987060547,
-      "logits/rejected": 32.30434036254883,
-      "logps/chosen": -0.7994817495346069,
-      "logps/rejected": -0.8369095921516418,
-      "loss": 1.2899,
-      "nll_loss": 1.211586594581604,
-      "rewards/accuracies": 0.565625011920929,
-      "rewards/chosen": -0.07994817197322845,
-      "rewards/margins": 0.0037427886854857206,
-      "rewards/rejected": -0.08369095623493195,
+      "grad_norm": 0.1403636887974213,
+      "learning_rate": 5.6711663290882776e-05,
+      "log_odds_chosen": 0.6598159074783325,
+      "log_odds_ratio": -0.5024036169052124,
+      "logits/chosen": 36.992469787597656,
+      "logits/rejected": 35.901676177978516,
+      "logps/chosen": -0.33446019887924194,
+      "logps/rejected": -0.5835962891578674,
+      "loss": 0.7105,
+      "nll_loss": 0.657503604888916,
+      "rewards/accuracies": 0.7562500238418579,
+      "rewards/chosen": -0.033446017652750015,
+      "rewards/margins": 0.0249136071652174,
+      "rewards/rejected": -0.05835963040590286,
       "step": 160
     },
     {
       "epoch": 1.0862619808306708,
-      "grad_norm": 0.5333966997862959,
-      "learning_rate": 1.5336570964437077e-06,
-      "log_odds_chosen": 0.038995109498500824,
-      "log_odds_ratio": -0.6969125866889954,
-      "logits/chosen": 32.00867462158203,
-      "logits/rejected": 31.789836883544922,
-      "logps/chosen": -0.8085753321647644,
-      "logps/rejected": -0.8331602215766907,
-      "loss": 1.2863,
-      "nll_loss": 1.2334020137786865,
-      "rewards/accuracies": 0.503125011920929,
-      "rewards/chosen": -0.0808575376868248,
-      "rewards/margins": 0.0024584876373410225,
-      "rewards/rejected": -0.08331602811813354,
+      "grad_norm": 0.14799945317167665,
+      "learning_rate": 5.112190321479026e-05,
+      "log_odds_chosen": 0.6551031470298767,
+      "log_odds_ratio": -0.5045552849769592,
+      "logits/chosen": 35.59492874145508,
+      "logits/rejected": 35.479576110839844,
+      "logps/chosen": -0.36466851830482483,
+      "logps/rejected": -0.6384136080741882,
+      "loss": 0.7194,
+      "nll_loss": 0.6874456405639648,
+      "rewards/accuracies": 0.765625,
+      "rewards/chosen": -0.03646685555577278,
+      "rewards/margins": 0.027374515309929848,
+      "rewards/rejected": -0.06384135782718658,
       "step": 170
     },
     {
       "epoch": 1.1501597444089458,
-      "grad_norm": 0.5317464882215207,
-      "learning_rate": 1.3655410366448499e-06,
-      "log_odds_chosen": 0.10975570976734161,
-      "log_odds_ratio": -0.6656275987625122,
-      "logits/chosen": 32.76184844970703,
-      "logits/rejected": 32.081581115722656,
-      "logps/chosen": -0.7589792013168335,
-      "logps/rejected": -0.8160526156425476,
-      "loss": 1.2451,
-      "nll_loss": 1.1476625204086304,
-      "rewards/accuracies": 0.6000000238418579,
-      "rewards/chosen": -0.0758979320526123,
-      "rewards/margins": 0.0057073310017585754,
-      "rewards/rejected": -0.08160526305437088,
+      "grad_norm": 0.4144450852693044,
+      "learning_rate": 4.551803455482833e-05,
+      "log_odds_chosen": 0.8569015264511108,
+      "log_odds_ratio": -0.43813252449035645,
+      "logits/chosen": 36.478363037109375,
+      "logits/rejected": 35.625038146972656,
+      "logps/chosen": -0.3283715844154358,
+      "logps/rejected": -0.6759421229362488,
+      "loss": 0.7011,
+      "nll_loss": 0.6358424425125122,
+      "rewards/accuracies": 0.809374988079071,
+      "rewards/chosen": -0.0328371599316597,
+      "rewards/margins": 0.03475705534219742,
+      "rewards/rejected": -0.06759421527385712,
       "step": 180
     },
     {
       "epoch": 1.2140575079872205,
-      "grad_norm": 0.5186911330716555,
-      "learning_rate": 1.199115876325091e-06,
-      "log_odds_chosen": 0.04539128392934799,
-      "log_odds_ratio": -0.6990563869476318,
-      "logits/chosen": 33.35085678100586,
-      "logits/rejected": 33.3908805847168,
-      "logps/chosen": -0.7675826549530029,
-      "logps/rejected": -0.7923057675361633,
-      "loss": 1.2217,
-      "nll_loss": 1.161102294921875,
-      "rewards/accuracies": 0.53125,
-      "rewards/chosen": -0.07675826549530029,
-      "rewards/margins": 0.0024723131209611893,
-      "rewards/rejected": -0.07923058420419693,
+      "grad_norm": 0.1393106135709619,
+      "learning_rate": 3.9970529210836366e-05,
+      "log_odds_chosen": 0.7694636583328247,
+      "log_odds_ratio": -0.48203492164611816,
+      "logits/chosen": 37.65850067138672,
+      "logits/rejected": 37.32286071777344,
+      "logps/chosen": -0.34162312746047974,
+      "logps/rejected": -0.6649552583694458,
+      "loss": 0.696,
+      "nll_loss": 0.6590245962142944,
+      "rewards/accuracies": 0.78125,
+      "rewards/chosen": -0.034162312746047974,
+      "rewards/margins": 0.032333213835954666,
+      "rewards/rejected": -0.06649552285671234,
       "step": 190
     },
     {
       "epoch": 1.2779552715654952,
-      "grad_norm": 0.5411315465201,
-      "learning_rate": 1.036474508437579e-06,
-      "log_odds_chosen": 0.07649532705545425,
-      "log_odds_ratio": -0.6858269572257996,
-      "logits/chosen": 30.826059341430664,
-      "logits/rejected": 30.824636459350586,
-      "logps/chosen": -0.786870002746582,
-      "logps/rejected": -0.8327391743659973,
-      "loss": 1.215,
-      "nll_loss": 1.165637493133545,
-      "rewards/accuracies": 0.578125,
-      "rewards/chosen": -0.07868699729442596,
-      "rewards/margins": 0.004586914554238319,
-      "rewards/rejected": -0.08327391743659973,
+      "grad_norm": 0.18676893862042468,
+      "learning_rate": 3.4549150281252636e-05,
+      "log_odds_chosen": 0.8638985753059387,
+      "log_odds_ratio": -0.4615907073020935,
+      "logits/chosen": 34.856327056884766,
+      "logits/rejected": 34.688533782958984,
+      "logps/chosen": -0.3576621413230896,
+      "logps/rejected": -0.74553382396698,
+      "loss": 0.6981,
+      "nll_loss": 0.6673237085342407,
+      "rewards/accuracies": 0.78125,
+      "rewards/chosen": -0.03576621413230896,
+      "rewards/margins": 0.03878717124462128,
+      "rewards/rejected": -0.07455337792634964,
       "step": 200
     },
     {
       "epoch": 1.34185303514377,
-      "grad_norm": 0.5587119516125117,
-      "learning_rate": 8.796622425502193e-07,
-      "log_odds_chosen": 0.04922567307949066,
-      "log_odds_ratio": -0.6928130984306335,
-      "logits/chosen": 32.06999969482422,
-      "logits/rejected": 31.100805282592773,
-      "logps/chosen": -0.760046124458313,
-      "logps/rejected": -0.7879078388214111,
-      "loss": 1.2018,
-      "nll_loss": 1.1223350763320923,
-      "rewards/accuracies": 0.5218750238418579,
-      "rewards/chosen": -0.07600460201501846,
-      "rewards/margins": 0.00278617930598557,
-      "rewards/rejected": -0.07879078388214111,
+      "grad_norm": 0.2564156817901977,
+      "learning_rate": 2.932207475167398e-05,
+      "log_odds_chosen": 0.8325332403182983,
+      "log_odds_ratio": -0.45953792333602905,
+      "logits/chosen": 36.798072814941406,
+      "logits/rejected": 35.09117889404297,
+      "logps/chosen": -0.3448064923286438,
+      "logps/rejected": -0.7093031406402588,
+      "loss": 0.6982,
+      "nll_loss": 0.6432278752326965,
+      "rewards/accuracies": 0.8062499761581421,
+      "rewards/chosen": -0.03448065370321274,
+      "rewards/margins": 0.03644966334104538,
+      "rewards/rejected": -0.07093031704425812,
       "step": 210
     },
     {
       "epoch": 1.4057507987220448,
-      "grad_norm": 0.5532006875711314,
-      "learning_rate": 7.30651083891141e-07,
-      "log_odds_chosen": 0.09409201890230179,
-      "log_odds_ratio": -0.6719040870666504,
-      "logits/chosen": 31.775470733642578,
-      "logits/rejected": 31.210159301757812,
-      "logps/chosen": -0.7408252954483032,
-      "logps/rejected": -0.7944933772087097,
-      "loss": 1.1848,
-      "nll_loss": 1.1082121133804321,
-      "rewards/accuracies": 0.6000000238418579,
-      "rewards/chosen": -0.07408253848552704,
-      "rewards/margins": 0.005366811528801918,
-      "rewards/rejected": -0.07944934070110321,
+      "grad_norm": 0.1531688356816753,
+      "learning_rate": 2.43550361297047e-05,
+      "log_odds_chosen": 1.0281397104263306,
+      "log_odds_ratio": -0.4170387387275696,
+      "logits/chosen": 35.599212646484375,
+      "logits/rejected": 34.3337287902832,
+      "logps/chosen": -0.3171806335449219,
+      "logps/rejected": -0.7794741988182068,
+      "loss": 0.6881,
+      "nll_loss": 0.6331400275230408,
+      "rewards/accuracies": 0.84375,
+      "rewards/chosen": -0.03171805664896965,
+      "rewards/margins": 0.04622935503721237,
+      "rewards/rejected": -0.07794742286205292,
       "step": 220
     },
     {
       "epoch": 1.4696485623003195,
-      "grad_norm": 0.5478964489714558,
-      "learning_rate": 5.913149342387704e-07,
-      "log_odds_chosen": 0.022008871659636497,
-      "log_odds_ratio": -0.7079219818115234,
-      "logits/chosen": 30.458974838256836,
-      "logits/rejected": 30.637874603271484,
-      "logps/chosen": -0.7661860585212708,
-      "logps/rejected": -0.783053994178772,
-      "loss": 1.1753,
-      "nll_loss": 1.1201671361923218,
-      "rewards/accuracies": 0.5406249761581421,
-      "rewards/chosen": -0.07661859691143036,
-      "rewards/margins": 0.00168679840862751,
-      "rewards/rejected": -0.07830540090799332,
+      "grad_norm": 0.13237811516695533,
+      "learning_rate": 1.971049780795901e-05,
+      "log_odds_chosen": 0.9918270111083984,
+      "log_odds_ratio": -0.440161794424057,
+      "logits/chosen": 34.80529022216797,
+      "logits/rejected": 34.433349609375,
+      "logps/chosen": -0.3528688848018646,
+      "logps/rejected": -0.8333184123039246,
+      "loss": 0.684,
+      "nll_loss": 0.653908908367157,
+      "rewards/accuracies": 0.8125,
+      "rewards/chosen": -0.03528688848018646,
+      "rewards/margins": 0.04804495349526405,
+      "rewards/rejected": -0.08333183825016022,
       "step": 230
     },
     {
       "epoch": 1.5335463258785942,
-      "grad_norm": 0.5614636317534148,
-      "learning_rate": 4.63406026519703e-07,
-      "log_odds_chosen": 0.12717078626155853,
-      "log_odds_ratio": -0.6581142544746399,
-      "logits/chosen": 31.024799346923828,
-      "logits/rejected": 30.477502822875977,
-      "logps/chosen": -0.7121531367301941,
-      "logps/rejected": -0.7798753380775452,
-      "loss": 1.1701,
-      "nll_loss": 1.073432207107544,
-      "rewards/accuracies": 0.596875011920929,
-      "rewards/chosen": -0.07121531665325165,
-      "rewards/margins": 0.006772211752831936,
-      "rewards/rejected": -0.07798753678798676,
+      "grad_norm": 0.17618952870677607,
+      "learning_rate": 1.544686755065677e-05,
+      "log_odds_chosen": 1.110446810722351,
+      "log_odds_ratio": -0.39400768280029297,
+      "logits/chosen": 35.90544891357422,
+      "logits/rejected": 34.70677185058594,
+      "logps/chosen": -0.3103191554546356,
+      "logps/rejected": -0.8037627935409546,
+      "loss": 0.6899,
+      "nll_loss": 0.6227988004684448,
+      "rewards/accuracies": 0.824999988079071,
+      "rewards/chosen": -0.03103191778063774,
+      "rewards/margins": 0.04934436455368996,
+      "rewards/rejected": -0.0803762823343277,
       "step": 240
     },
     {
       "epoch": 1.5974440894568689,
-      "grad_norm": 0.5532326330060124,
-      "learning_rate": 3.4853288946298335e-07,
-      "log_odds_chosen": 0.031235849484801292,
-      "log_odds_ratio": -0.7052245140075684,
-      "logits/chosen": 31.759140014648438,
-      "logits/rejected": 31.2039737701416,
-      "logps/chosen": -0.7514272332191467,
-      "logps/rejected": -0.7712705135345459,
-      "loss": 1.1628,
-      "nll_loss": 1.0986436605453491,
-      "rewards/accuracies": 0.518750011920929,
-      "rewards/chosen": -0.07514272630214691,
-      "rewards/margins": 0.0019843343179672956,
-      "rewards/rejected": -0.07712705433368683,
+      "grad_norm": 0.17643350651397863,
+      "learning_rate": 1.1617762982099446e-05,
+      "log_odds_chosen": 1.0278904438018799,
+      "log_odds_ratio": -0.40579158067703247,
+      "logits/chosen": 36.86786651611328,
+      "logits/rejected": 35.534324645996094,
+      "logps/chosen": -0.34132060408592224,
+      "logps/rejected": -0.8093164563179016,
+      "loss": 0.6839,
+      "nll_loss": 0.6438942551612854,
+      "rewards/accuracies": 0.8218749761581421,
+      "rewards/chosen": -0.034132059663534164,
+      "rewards/margins": 0.04679957777261734,
+      "rewards/rejected": -0.0809316486120224,
       "step": 250
     },
     {
       "epoch": 1.6613418530351438,
-      "grad_norm": 0.5579681347407102,
-      "learning_rate": 2.48140119418046e-07,
-      "log_odds_chosen": 0.09361070394515991,
-      "log_odds_ratio": -0.6759673953056335,
-      "logits/chosen": 30.127365112304688,
-      "logits/rejected": 29.955150604248047,
-      "logps/chosen": -0.7377598881721497,
-      "logps/rejected": -0.789117157459259,
-      "loss": 1.1595,
-      "nll_loss": 1.0882426500320435,
-      "rewards/accuracies": 0.581250011920929,
-      "rewards/chosen": -0.0737759917974472,
-      "rewards/margins": 0.0051357327029109,
-      "rewards/rejected": -0.07891170680522919,
+      "grad_norm": 0.20335250536853633,
+      "learning_rate": 8.271337313934869e-06,
+      "log_odds_chosen": 1.0964257717132568,
+      "log_odds_ratio": -0.40502405166625977,
+      "logits/chosen": 35.01730728149414,
+      "logits/rejected": 33.87272262573242,
+      "logps/chosen": -0.35952043533325195,
+      "logps/rejected": -0.8931809663772583,
+      "loss": 0.6875,
+      "nll_loss": 0.6542503833770752,
+      "rewards/accuracies": 0.8500000238418579,
+      "rewards/chosen": -0.035952042788267136,
+      "rewards/margins": 0.0533660463988781,
+      "rewards/rejected": -0.08931808918714523,
       "step": 260
     },
     {
       "epoch": 1.7252396166134185,
-      "grad_norm": 0.5388500509178578,
-      "learning_rate": 1.634902137174483e-07,
-      "log_odds_chosen": 0.06861741840839386,
-      "log_odds_ratio": -0.6897167563438416,
-      "logits/chosen": 31.801860809326172,
-      "logits/rejected": 30.827133178710938,
-      "logps/chosen": -0.7297223210334778,
-      "logps/rejected": -0.7675324082374573,
-      "loss": 1.15,
-      "nll_loss": 1.091435194015503,
-      "rewards/accuracies": 0.5562499761581421,
-      "rewards/chosen": -0.07297223806381226,
-      "rewards/margins": 0.003781010629609227,
-      "rewards/rejected": -0.07675323635339737,
+      "grad_norm": 0.1417159132341433,
+      "learning_rate": 5.449673790581611e-06,
+      "log_odds_chosen": 1.1632859706878662,
+      "log_odds_ratio": -0.39327362179756165,
+      "logits/chosen": 37.34235763549805,
+      "logits/rejected": 35.14636993408203,
+      "logps/chosen": -0.3363361358642578,
+      "logps/rejected": -0.8959217071533203,
+      "loss": 0.6837,
+      "nll_loss": 0.6559703350067139,
+      "rewards/accuracies": 0.8531249761581421,
+      "rewards/chosen": -0.03363361582159996,
+      "rewards/margins": 0.055958546698093414,
+      "rewards/rejected": -0.08959217369556427,
       "step": 270
     },
     {
       "epoch": 1.7891373801916934,
-      "grad_norm": 0.5597110218019701,
-      "learning_rate": 9.564769404039419e-08,
-      "log_odds_chosen": 0.10912013053894043,
-      "log_odds_ratio": -0.6706396341323853,
-      "logits/chosen": 30.64389419555664,
-      "logits/rejected": 30.014019012451172,
-      "logps/chosen": -0.7340375185012817,
-      "logps/rejected": -0.7948423624038696,
-      "loss": 1.1518,
-      "nll_loss": 1.0911478996276855,
-      "rewards/accuracies": 0.590624988079071,
-      "rewards/chosen": -0.0734037533402443,
-      "rewards/margins": 0.00608047703281045,
-      "rewards/rejected": -0.0794842392206192,
+      "grad_norm": 0.39828151397493416,
+      "learning_rate": 3.18825646801314e-06,
+      "log_odds_chosen": 1.0620207786560059,
+      "log_odds_ratio": -0.41589412093162537,
+      "logits/chosen": 35.572288513183594,
+      "logits/rejected": 34.087669372558594,
+      "logps/chosen": -0.3491634726524353,
+      "logps/rejected": -0.8419939875602722,
+      "loss": 0.6856,
+      "nll_loss": 0.6521117091178894,
+      "rewards/accuracies": 0.8500000238418579,
+      "rewards/chosen": -0.03491634875535965,
+      "rewards/margins": 0.04928305745124817,
+      "rewards/rejected": -0.08419940620660782,
       "step": 280
     },
     {
       "epoch": 1.8530351437699681,
-      "grad_norm": 0.5432963139659418,
-      "learning_rate": 4.546571943496969e-08,
-      "log_odds_chosen": 0.071634940803051,
-      "log_odds_ratio": -0.6880494356155396,
-      "logits/chosen": 31.523670196533203,
-      "logits/rejected": 31.051345825195312,
-      "logps/chosen": -0.7210476398468018,
-      "logps/rejected": -0.7614242434501648,
-      "loss": 1.1457,
-      "nll_loss": 1.078147292137146,
-      "rewards/accuracies": 0.53125,
-      "rewards/chosen": -0.07210476696491241,
-      "rewards/margins": 0.0040376619435846806,
-      "rewards/rejected": -0.07614242285490036,
+      "grad_norm": 0.17410175622655502,
+      "learning_rate": 1.5155239811656563e-06,
+      "log_odds_chosen": 1.1792383193969727,
+      "log_odds_ratio": -0.3946213126182556,
+      "logits/chosen": 36.629905700683594,
+      "logits/rejected": 35.32917404174805,
+      "logps/chosen": -0.33119866251945496,
+      "logps/rejected": -0.899772047996521,
+      "loss": 0.6837,
+      "nll_loss": 0.6425603628158569,
+      "rewards/accuracies": 0.815625011920929,
+      "rewards/chosen": -0.033119868487119675,
+      "rewards/margins": 0.05685734748840332,
+      "rewards/rejected": -0.0899772122502327,
       "step": 290
     },
     {
       "epoch": 1.9169329073482428,
-      "grad_norm": 0.5550233900085839,
-      "learning_rate": 1.357535734809795e-08,
-      "log_odds_chosen": 0.08968226611614227,
-      "log_odds_ratio": -0.6760362386703491,
-      "logits/chosen": 30.686954498291016,
-      "logits/rejected": 29.28194236755371,
-      "logps/chosen": -0.7248884439468384,
-      "logps/rejected": -0.7753577828407288,
-      "loss": 1.1474,
-      "nll_loss": 1.0696710348129272,
-      "rewards/accuracies": 0.5531250238418579,
-      "rewards/chosen": -0.07248884439468384,
-      "rewards/margins": 0.005046932026743889,
-      "rewards/rejected": -0.07753578573465347,
+      "grad_norm": 0.13374493386406422,
+      "learning_rate": 4.52511911603265e-07,
+      "log_odds_chosen": 1.0675084590911865,
+      "log_odds_ratio": -0.42029696702957153,
+      "logits/chosen": 36.18767547607422,
+      "logits/rejected": 33.20973205566406,
+      "logps/chosen": -0.34383276104927063,
+      "logps/rejected": -0.8607551455497742,
+      "loss": 0.6844,
+      "nll_loss": 0.6432604193687439,
+      "rewards/accuracies": 0.809374988079071,
+      "rewards/chosen": -0.034383274614810944,
+      "rewards/margins": 0.05169224739074707,
+      "rewards/rejected": -0.08607552200555801,
       "step": 300
     },
     {
       "epoch": 1.9808306709265175,
-      "grad_norm": 0.49758643306678957,
-      "learning_rate": 3.77647586240204e-10,
-      "log_odds_chosen": 0.048125725239515305,
-      "log_odds_ratio": -0.6949858665466309,
-      "logits/chosen": 30.8554744720459,
-      "logits/rejected": 30.65484046936035,
-      "logps/chosen": -0.7389064431190491,
-      "logps/rejected": -0.7689411640167236,
-      "loss": 1.1435,
-      "nll_loss": 1.0751014947891235,
-      "rewards/accuracies": 0.534375011920929,
-      "rewards/chosen": -0.07389064878225327,
-      "rewards/margins": 0.0030034759547561407,
-      "rewards/rejected": -0.0768941268324852,
+      "grad_norm": 0.1377772193573712,
+      "learning_rate": 1.2588252874673468e-08,
+      "log_odds_chosen": 1.0935267210006714,
+      "log_odds_ratio": -0.41092976927757263,
+      "logits/chosen": 36.207176208496094,
+      "logits/rejected": 34.82603073120117,
+      "logps/chosen": -0.344975084066391,
+      "logps/rejected": -0.8865535855293274,
+      "loss": 0.6781,
+      "nll_loss": 0.6345787644386292,
+      "rewards/accuracies": 0.8062499761581421,
+      "rewards/chosen": -0.03449751064181328,
+      "rewards/margins": 0.05415785312652588,
+      "rewards/rejected": -0.08865536749362946,
       "step": 310
     },
     {
       "epoch": 1.9936102236421727,
       "step": 312,
       "total_flos": 0.0,
-      "train_loss": 1.3676127867820935,
-      "train_runtime": 5524.9246,
-      "train_samples_per_second": 7.24,
+      "train_loss": 0.7846963420892373,
+      "train_runtime": 5577.3844,
+      "train_samples_per_second": 7.172,
       "train_steps_per_second": 0.056
     }
   ],