Model save

README.md

@@ -1,13 +1,10 @@
 ---
 library_name: peft
 tags:
-- alignment-handbook
 - trl
 - dpo
 - generated_from_trainer
 base_model: NorLLM-AI/NorMistral-7B
-datasets:
-- hugodk-sch/aftonposten_title_prefs
 model-index:
 - name: norllm-ai-normistral-7b-align-scan
   results: []
@@ -18,17 +15,17 @@ should probably proofread and complete it, then remove this comment. -->
 
 # norllm-ai-normistral-7b-align-scan
 
-This model is a fine-tuned version of [data/norllm-ai-normistral-7b-sft-qlora](https://huggingface.co/data/norllm-ai-normistral-7b-sft-qlora) on the hugodk-sch/aftonposten_title_prefs dataset.
+This model is a fine-tuned version of [NorLLM-AI/NorMistral-7B](https://huggingface.co/NorLLM-AI/NorMistral-7B) on the None dataset.
 It achieves the following results on the evaluation set:
-- Loss: 1.6120
-- Rewards/chosen: 0.0138
-- Rewards/rejected: -0.0163
-- Rewards/accuracies: 0.5689
-- Rewards/margins: 0.0302
-- Logps/rejected: -34.7390
-- Logps/chosen: -31.2479
-- Logits/rejected: -2.8036
-- Logits/chosen: -2.8061
+- Loss: 23.8556
+- Rewards/chosen: -0.0119
+- Rewards/rejected: -0.0312
+- Rewards/accuracies: 0.5718
+- Rewards/margins: 0.0193
+- Logps/rejected: -35.0104
+- Logps/chosen: -31.4016
+- Logits/rejected: -2.8159
+- Logits/chosen: -2.8182
 
 ## Model description
 
@@ -63,9 +60,9 @@ The following hyperparameters were used during training:
 
 | Training Loss | Epoch | Step | Validation Loss | Rewards/chosen | Rewards/rejected | Rewards/accuracies | Rewards/margins | Logps/rejected | Logps/chosen | Logits/rejected | Logits/chosen |
 |:-------------:|:-----:|:----:|:---------------:|:--------------:|:----------------:|:------------------:|:---------------:|:--------------:|:------------:|:---------------:|:-------------:|
-| 1.7948        | 0.26  | 100  | 1.6241          | 0.0102         | -0.0177          | 0.5685             | 0.0278          | -34.7424       | -31.2571     | -2.8147         | -2.8176       |
-| 2.6629        | 0.52  | 200  | 1.5649          | 0.0182         | -0.0143          | 0.5457             | 0.0325          | -34.7338       | -31.2369     | -2.8095         | -2.8121       |
-| 1.5067        | 0.78  | 300  | 1.6904          | 0.0126         | -0.0056          | 0.5573             | 0.0181          | -34.7121       | -31.2510     | -2.8041         | -2.8065       |
+| 22.8693       | 0.26  | 100  | 23.9885         | 0.0019         | -0.0147          | 0.6130             | 0.0166          | -34.8449       | -31.2633     | -2.8087         | -2.8112       |
+| 19.5946       | 0.52  | 200  | 23.7575         | -0.0117        | -0.0312          | 0.5718             | 0.0194          | -35.0097       | -31.3997     | -2.8172         | -2.8197       |
+| 18.7484       | 0.78  | 300  | 23.8556         | -0.0119        | -0.0312          | 0.5718             | 0.0193          | -35.0104       | -31.4016     | -2.8159         | -2.8182       |
 
 
 ### Framework versions

adapter_model.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:1265a2cb88e14ec5b7d0e625745426baa5544bb288da5f771e85f568d54e358a
+oid sha256:34b472dac9e58f7bfa1ec51a7ed4c6bc55a86d934339b9f80e3740f27f568ab8
 size 671150064

all_results.json

@@ -1,21 +1,8 @@
 {
     "epoch": 1.0,
-    "eval_logits/chosen": -2.806074619293213,
-    "eval_logits/rejected": -2.803609848022461,
-    "eval_logps/chosen": -31.247922897338867,
-    "eval_logps/rejected": -34.73902893066406,
-    "eval_loss": 1.6120386123657227,
-    "eval_rewards/accuracies": 0.5689368844032288,
-    "eval_rewards/chosen": 0.01381184346973896,
-    "eval_rewards/margins": 0.03016132488846779,
-    "eval_rewards/rejected": -0.016349485144019127,
-    "eval_runtime": 112.7318,
-    "eval_samples": 343,
-    "eval_samples_per_second": 3.043,
-    "eval_steps_per_second": 0.381,
-    "train_loss": 1.7377146052075672,
-    "train_runtime": 2719.6001,
+    "train_loss": 19.552858486423244,
+    "train_runtime": 2721.3646,
     "train_samples": 3079,
-    "train_samples_per_second": 1.132,
-    "train_steps_per_second": 0.142
+    "train_samples_per_second": 1.131,
+    "train_steps_per_second": 0.141
 }

train_results.json

@@ -1,8 +1,8 @@
 {
     "epoch": 1.0,
-    "train_loss": 1.7377146052075672,
-    "train_runtime": 2719.6001,
+    "train_loss": 19.552858486423244,
+    "train_runtime": 2721.3646,
     "train_samples": 3079,
-    "train_samples_per_second": 1.132,
-    "train_steps_per_second": 0.142
+    "train_samples_per_second": 1.131,
+    "train_steps_per_second": 0.141
 }

trainer_state.json

@@ -10,13 +10,13 @@
   "log_history": [
     {
       "epoch": 0.0,
-      "grad_norm": 198.0,
+      "grad_norm": 792.0,
       "learning_rate": 1.282051282051282e-07,
       "logits/chosen": -2.7358343601226807,
       "logits/rejected": -2.7480404376983643,
       "logps/chosen": -27.35565757751465,
       "logps/rejected": -21.06114387512207,
-      "loss": 1.5625,
+      "loss": 25.0,
       "rewards/accuracies": 0.0,
       "rewards/chosen": 0.0,
       "rewards/margins": 0.0,
@@ -25,630 +25,630 @@
     },
     {
       "epoch": 0.03,
-      "grad_norm": 306.0,
+      "grad_norm": 1384.0,
       "learning_rate": 1.282051282051282e-06,
-      "logits/chosen": -3.009331703186035,
-      "logits/rejected": -2.9980063438415527,
-      "logps/chosen": -33.193904876708984,
-      "logps/rejected": -32.01276779174805,
-      "loss": 1.5179,
-      "rewards/accuracies": 0.5416666865348816,
-      "rewards/chosen": -0.005185974761843681,
-      "rewards/margins": 0.01435802225023508,
-      "rewards/rejected": -0.019543997943401337,
+      "logits/chosen": -3.0097055435180664,
+      "logits/rejected": -2.9980692863464355,
+      "logps/chosen": -33.19136047363281,
+      "logps/rejected": -31.962371826171875,
+      "loss": 25.1536,
+      "rewards/accuracies": 0.4166666567325592,
+      "rewards/chosen": -0.0010418971069157124,
+      "rewards/margins": -0.001195515040308237,
+      "rewards/rejected": 0.00015361772966571152,
       "step": 10
     },
     {
       "epoch": 0.05,
-      "grad_norm": 268.0,
+      "grad_norm": 940.0,
       "learning_rate": 2.564102564102564e-06,
-      "logits/chosen": -2.8995871543884277,
-      "logits/rejected": -2.894530773162842,
-      "logps/chosen": -32.46534729003906,
-      "logps/rejected": -28.974437713623047,
-      "loss": 1.6084,
-      "rewards/accuracies": 0.5249999761581421,
-      "rewards/chosen": 0.0038995600771158934,
-      "rewards/margins": 0.0009288033470511436,
-      "rewards/rejected": 0.002970754634588957,
+      "logits/chosen": -2.899199962615967,
+      "logits/rejected": -2.893846273422241,
+      "logps/chosen": -32.49421691894531,
+      "logps/rejected": -28.94989013671875,
+      "loss": 25.5664,
+      "rewards/accuracies": 0.44999998807907104,
+      "rewards/chosen": -0.0019120589131489396,
+      "rewards/margins": -0.005109794903546572,
+      "rewards/rejected": 0.0031977363396435976,
       "step": 20
     },
     {
       "epoch": 0.08,
-      "grad_norm": 221.0,
+      "grad_norm": 872.0,
       "learning_rate": 3.846153846153847e-06,
-      "logits/chosen": -3.097817897796631,
-      "logits/rejected": -3.1089720726013184,
-      "logps/chosen": -32.821983337402344,
-      "logps/rejected": -30.1259708404541,
-      "loss": 1.6124,
-      "rewards/accuracies": 0.4625000059604645,
-      "rewards/chosen": 0.0346810482442379,
-      "rewards/margins": 0.006648649461567402,
-      "rewards/rejected": 0.028032397851347923,
+      "logits/chosen": -3.0976572036743164,
+      "logits/rejected": -3.1090171337127686,
+      "logps/chosen": -32.78887176513672,
+      "logps/rejected": -30.141361236572266,
+      "loss": 24.4486,
+      "rewards/accuracies": 0.5625,
+      "rewards/chosen": 0.011981222778558731,
+      "rewards/margins": 0.006511862389743328,
+      "rewards/rejected": 0.005469360388815403,
       "step": 30
     },
     {
       "epoch": 0.1,
-      "grad_norm": 225.0,
+      "grad_norm": 788.0,
       "learning_rate": 4.999896948438434e-06,
-      "logits/chosen": -2.8635919094085693,
-      "logits/rejected": -2.8542821407318115,
-      "logps/chosen": -31.561412811279297,
-      "logps/rejected": -32.368797302246094,
-      "loss": 1.5659,
-      "rewards/accuracies": 0.5625,
-      "rewards/chosen": 0.10756368935108185,
-      "rewards/margins": 0.11209961026906967,
-      "rewards/rejected": -0.004535915795713663,
+      "logits/chosen": -2.8625879287719727,
+      "logits/rejected": -2.853339195251465,
+      "logps/chosen": -31.531911849975586,
+      "logps/rejected": -32.36937713623047,
+      "loss": 22.726,
+      "rewards/accuracies": 0.5874999761581421,
+      "rewards/chosen": 0.02984103001654148,
+      "rewards/margins": 0.031033098697662354,
+      "rewards/rejected": -0.001192068331874907,
       "step": 40
     },
     {
       "epoch": 0.13,
-      "grad_norm": 268.0,
+      "grad_norm": 536.0,
       "learning_rate": 4.987541037542187e-06,
-      "logits/chosen": -2.886489152908325,
-      "logits/rejected": -2.884337902069092,
-      "logps/chosen": -29.50502586364746,
-      "logps/rejected": -30.12764549255371,
-      "loss": 1.6791,
-      "rewards/accuracies": 0.5874999761581421,
-      "rewards/chosen": 0.0940544530749321,
-      "rewards/margins": 0.10953446477651596,
-      "rewards/rejected": -0.015479998663067818,
+      "logits/chosen": -2.880608558654785,
+      "logits/rejected": -2.8780317306518555,
+      "logps/chosen": -29.381378173828125,
+      "logps/rejected": -30.077362060546875,
+      "loss": 23.1409,
+      "rewards/accuracies": 0.5249999761581421,
+      "rewards/chosen": 0.035878658294677734,
+      "rewards/margins": 0.034720782190561295,
+      "rewards/rejected": 0.0011578750563785434,
       "step": 50
     },
     {
       "epoch": 0.16,
-      "grad_norm": 217.0,
+      "grad_norm": 784.0,
       "learning_rate": 4.954691471941119e-06,
-      "logits/chosen": -2.9178414344787598,
-      "logits/rejected": -2.9194931983947754,
-      "logps/chosen": -30.02047348022461,
-      "logps/rejected": -28.11154556274414,
-      "loss": 1.3051,
-      "rewards/accuracies": 0.699999988079071,
-      "rewards/chosen": 0.04371771961450577,
-      "rewards/margins": 0.10606314986944199,
-      "rewards/rejected": -0.06234544515609741,
+      "logits/chosen": -2.9093430042266846,
+      "logits/rejected": -2.910740375518799,
+      "logps/chosen": -29.860986709594727,
+      "logps/rejected": -28.04913902282715,
+      "loss": 22.1643,
+      "rewards/accuracies": 0.7124999761581421,
+      "rewards/chosen": 0.02687840536236763,
+      "rewards/margins": 0.036224402487277985,
+      "rewards/rejected": -0.00934599619358778,
       "step": 60
     },
     {
       "epoch": 0.18,
-      "grad_norm": 1032.0,
+      "grad_norm": 1600.0,
       "learning_rate": 4.901618883413549e-06,
-      "logits/chosen": -3.0002479553222656,
-      "logits/rejected": -3.007493495941162,
-      "logps/chosen": -29.312774658203125,
-      "logps/rejected": -31.054119110107422,
-      "loss": 2.1422,
-      "rewards/accuracies": 0.5625,
-      "rewards/chosen": -0.012723004445433617,
-      "rewards/margins": 0.05220501497387886,
-      "rewards/rejected": -0.06492801755666733,
+      "logits/chosen": -2.9914920330047607,
+      "logits/rejected": -2.99847674369812,
+      "logps/chosen": -29.173349380493164,
+      "logps/rejected": -30.959280014038086,
+      "loss": 24.455,
+      "rewards/accuracies": 0.6000000238418579,
+      "rewards/chosen": 0.01076198648661375,
+      "rewards/margins": 0.017510101199150085,
+      "rewards/rejected": -0.006748115178197622,
       "step": 70
     },
     {
       "epoch": 0.21,
-      "grad_norm": 326.0,
+      "grad_norm": 1072.0,
       "learning_rate": 4.828760511501322e-06,
-      "logits/chosen": -2.814807415008545,
-      "logits/rejected": -2.8305888175964355,
-      "logps/chosen": -29.402179718017578,
-      "logps/rejected": -29.830846786499023,
-      "loss": 1.5698,
-      "rewards/accuracies": 0.612500011920929,
-      "rewards/chosen": 0.05541967228055,
-      "rewards/margins": 0.11804421246051788,
-      "rewards/rejected": -0.06262453645467758,
+      "logits/chosen": -2.8084774017333984,
+      "logits/rejected": -2.823390007019043,
+      "logps/chosen": -29.325841903686523,
+      "logps/rejected": -29.870441436767578,
+      "loss": 22.0145,
+      "rewards/accuracies": 0.637499988079071,
+      "rewards/chosen": 0.021488705649971962,
+      "rewards/margins": 0.041104547679424286,
+      "rewards/rejected": -0.01961584761738777,
       "step": 80
     },
     {
       "epoch": 0.23,
-      "grad_norm": 322.0,
+      "grad_norm": 740.0,
       "learning_rate": 4.7367166013034295e-06,
-      "logits/chosen": -2.9036896228790283,
-      "logits/rejected": -2.885866641998291,
-      "logps/chosen": -32.615840911865234,
-      "logps/rejected": -30.156829833984375,
-      "loss": 5.9533,
+      "logits/chosen": -2.8982763290405273,
+      "logits/rejected": -2.8804843425750732,
+      "logps/chosen": -32.65864181518555,
+      "logps/rejected": -30.32712745666504,
+      "loss": 25.4545,
       "rewards/accuracies": 0.699999988079071,
-      "rewards/chosen": 0.08823554217815399,
-      "rewards/margins": 0.1423090249300003,
-      "rewards/rejected": -0.054073482751846313,
+      "rewards/chosen": 0.017779098823666573,
+      "rewards/margins": 0.04832718148827553,
+      "rewards/rejected": -0.030548080801963806,
       "step": 90
     },
     {
       "epoch": 0.26,
-      "grad_norm": 177.0,
+      "grad_norm": 720.0,
       "learning_rate": 4.626245458345211e-06,
-      "logits/chosen": -3.007761001586914,
-      "logits/rejected": -3.0078930854797363,
-      "logps/chosen": -31.708675384521484,
-      "logps/rejected": -30.79708480834961,
-      "loss": 1.7948,
+      "logits/chosen": -3.0048553943634033,
+      "logits/rejected": -3.004960298538208,
+      "logps/chosen": -31.79574966430664,
+      "logps/rejected": -30.897558212280273,
+      "loss": 22.8693,
       "rewards/accuracies": 0.625,
-      "rewards/chosen": 0.1392797976732254,
-      "rewards/margins": 0.13164444267749786,
-      "rewards/rejected": 0.007635355927050114,
+      "rewards/chosen": 0.026112746447324753,
+      "rewards/margins": 0.034251101315021515,
+      "rewards/rejected": -0.008138353936374187,
       "step": 100
     },
     {
       "epoch": 0.26,
-      "eval_logits/chosen": -2.8176231384277344,
-      "eval_logits/rejected": -2.814725399017334,
-      "eval_logps/chosen": -31.257057189941406,
-      "eval_logps/rejected": -34.74238204956055,
-      "eval_loss": 1.6240766048431396,
-      "eval_rewards/accuracies": 0.5685215592384338,
-      "eval_rewards/chosen": 0.010158155113458633,
-      "eval_rewards/margins": 0.027848951518535614,
-      "eval_rewards/rejected": -0.01769079640507698,
-      "eval_runtime": 112.9872,
-      "eval_samples_per_second": 3.036,
-      "eval_steps_per_second": 0.381,
+      "eval_logits/chosen": -2.8112006187438965,
+      "eval_logits/rejected": -2.8086960315704346,
+      "eval_logps/chosen": -31.26334571838379,
+      "eval_logps/rejected": -34.84490203857422,
+      "eval_loss": 23.988525390625,
+      "eval_rewards/accuracies": 0.6129568219184875,
+      "eval_rewards/chosen": 0.0019106707768514752,
+      "eval_rewards/margins": 0.016585489735007286,
+      "eval_rewards/rejected": -0.014674817211925983,
+      "eval_runtime": 113.1272,
+      "eval_samples_per_second": 3.032,
+      "eval_steps_per_second": 0.38,
       "step": 100
     },
     {
       "epoch": 0.29,
-      "grad_norm": 251.0,
+      "grad_norm": 1008.0,
       "learning_rate": 4.498257201263691e-06,
-      "logits/chosen": -2.964388847351074,
-      "logits/rejected": -2.9403655529022217,
-      "logps/chosen": -31.717798233032227,
-      "logps/rejected": -31.185800552368164,
-      "loss": 1.9523,
-      "rewards/accuracies": 0.762499988079071,
-      "rewards/chosen": 0.19960005581378937,
-      "rewards/margins": 0.21038112044334412,
-      "rewards/rejected": -0.010781067423522472,
+      "logits/chosen": -2.9598517417907715,
+      "logits/rejected": -2.9358468055725098,
+      "logps/chosen": -31.7904109954834,
+      "logps/rejected": -31.376705169677734,
+      "loss": 21.3073,
+      "rewards/accuracies": 0.7250000238418579,
+      "rewards/chosen": 0.042638350278139114,
+      "rewards/margins": 0.06442411243915558,
+      "rewards/rejected": -0.02178577333688736,
       "step": 110
     },
     {
       "epoch": 0.31,
-      "grad_norm": 182.0,
+      "grad_norm": 792.0,
       "learning_rate": 4.353806263777678e-06,
-      "logits/chosen": -3.0449271202087402,
-      "logits/rejected": -3.074241876602173,
-      "logps/chosen": -28.759485244750977,
-      "logps/rejected": -34.1177864074707,
-      "loss": 1.6223,
-      "rewards/accuracies": 0.637499988079071,
-      "rewards/chosen": 0.1617673933506012,
-      "rewards/margins": 0.1498604714870453,
-      "rewards/rejected": 0.01190691627562046,
+      "logits/chosen": -3.0428009033203125,
+      "logits/rejected": -3.071193218231201,
+      "logps/chosen": -28.65635108947754,
+      "logps/rejected": -34.34435272216797,
+      "loss": 20.586,
+      "rewards/accuracies": 0.6875,
+      "rewards/chosen": 0.05075501650571823,
+      "rewards/margins": 0.07043538987636566,
+      "rewards/rejected": -0.01968037523329258,
       "step": 120
     },
     {
       "epoch": 0.34,
-      "grad_norm": 153.0,
+      "grad_norm": 668.0,
       "learning_rate": 4.1940827077152755e-06,
-      "logits/chosen": -2.745648145675659,
-      "logits/rejected": -2.7412946224212646,
-      "logps/chosen": -28.58340835571289,
-      "logps/rejected": -30.125900268554688,
-      "loss": 1.9846,
+      "logits/chosen": -2.744912624359131,
+      "logits/rejected": -2.740849494934082,
+      "logps/chosen": -28.600677490234375,
+      "logps/rejected": -30.34041404724121,
+      "loss": 21.4152,
       "rewards/accuracies": 0.6875,
-      "rewards/chosen": 0.15650880336761475,
-      "rewards/margins": 0.1533457338809967,
-      "rewards/rejected": 0.0031630918383598328,
+      "rewards/chosen": 0.037400223314762115,
+      "rewards/margins": 0.05806100368499756,
+      "rewards/rejected": -0.020660774782299995,
       "step": 130
     },
     {
       "epoch": 0.36,
-      "grad_norm": 128.0,
+      "grad_norm": 624.0,
       "learning_rate": 4.0204024186666215e-06,
-      "logits/chosen": -3.016637086868286,
-      "logits/rejected": -3.014425277709961,
-      "logps/chosen": -27.189367294311523,
-      "logps/rejected": -31.768428802490234,
-      "loss": 1.6015,
-      "rewards/accuracies": 0.6000000238418579,
-      "rewards/chosen": 0.1187439113855362,
-      "rewards/margins": 0.16340722143650055,
-      "rewards/rejected": -0.04466330260038376,
+      "logits/chosen": -3.0207009315490723,
+      "logits/rejected": -3.0183839797973633,
+      "logps/chosen": -27.157459259033203,
+      "logps/rejected": -31.912410736083984,
+      "loss": 22.1262,
+      "rewards/accuracies": 0.5874999761581421,
+      "rewards/chosen": 0.03287653252482414,
+      "rewards/margins": 0.05844072252511978,
+      "rewards/rejected": -0.02556418441236019,
       "step": 140
     },
     {
       "epoch": 0.39,
-      "grad_norm": 191.0,
+      "grad_norm": 668.0,
       "learning_rate": 3.834196265035119e-06,
-      "logits/chosen": -2.812203884124756,
-      "logits/rejected": -2.8067917823791504,
-      "logps/chosen": -27.404495239257812,
-      "logps/rejected": -31.38730239868164,
-      "loss": 1.2187,
-      "rewards/accuracies": 0.737500011920929,
-      "rewards/chosen": 0.16840532422065735,
-      "rewards/margins": 0.2338680922985077,
-      "rewards/rejected": -0.06546276807785034,
+      "logits/chosen": -2.8160903453826904,
+      "logits/rejected": -2.8107659816741943,
+      "logps/chosen": -27.24477195739746,
+      "logps/rejected": -31.60110092163086,
+      "loss": 18.6765,
+      "rewards/accuracies": 0.675000011920929,
+      "rewards/chosen": 0.05807359144091606,
+      "rewards/margins": 0.09581951797008514,
+      "rewards/rejected": -0.03774593397974968,
       "step": 150
     },
     {
       "epoch": 0.42,
-      "grad_norm": 223.0,
+      "grad_norm": 872.0,
       "learning_rate": 3.636998309800573e-06,
-      "logits/chosen": -3.127948522567749,
-      "logits/rejected": -3.1103968620300293,
-      "logps/chosen": -31.860748291015625,
-      "logps/rejected": -29.207168579101562,
-      "loss": 1.9596,
-      "rewards/accuracies": 0.762499988079071,
-      "rewards/chosen": 0.2523886263370514,
-      "rewards/margins": 0.3433467745780945,
-      "rewards/rejected": -0.09095814824104309,
+      "logits/chosen": -3.134445905685425,
+      "logits/rejected": -3.1172776222229004,
+      "logps/chosen": -31.69814109802246,
+      "logps/rejected": -29.522869110107422,
+      "loss": 16.4703,
+      "rewards/accuracies": 0.75,
+      "rewards/chosen": 0.07935762405395508,
+      "rewards/margins": 0.1336670219898224,
+      "rewards/rejected": -0.054309405386447906,
       "step": 160
     },
     {
       "epoch": 0.44,
-      "grad_norm": 112.0,
+      "grad_norm": 688.0,
       "learning_rate": 3.4304331721118078e-06,
-      "logits/chosen": -2.9414854049682617,
-      "logits/rejected": -2.9481499195098877,
-      "logps/chosen": -29.37969970703125,
-      "logps/rejected": -31.226505279541016,
-      "loss": 1.5997,
-      "rewards/accuracies": 0.762499988079071,
-      "rewards/chosen": 0.205966979265213,
-      "rewards/margins": 0.25677576661109924,
-      "rewards/rejected": -0.05080880969762802,
+      "logits/chosen": -2.9469733238220215,
+      "logits/rejected": -2.9541804790496826,
+      "logps/chosen": -29.33526039123535,
+      "logps/rejected": -31.71769142150879,
+      "loss": 17.9149,
+      "rewards/accuracies": 0.7124999761581421,
+      "rewards/chosen": 0.0559355802834034,
+      "rewards/margins": 0.11775630712509155,
+      "rewards/rejected": -0.06182072311639786,
       "step": 170
     },
     {
       "epoch": 0.47,
-      "grad_norm": 118.5,
+      "grad_norm": 688.0,
       "learning_rate": 3.2162026428305436e-06,
-      "logits/chosen": -2.789107322692871,
-      "logits/rejected": -2.7866973876953125,
-      "logps/chosen": -29.164358139038086,
-      "logps/rejected": -29.882699966430664,
-      "loss": 1.5371,
-      "rewards/accuracies": 0.6625000238418579,
-      "rewards/chosen": 0.2021961212158203,
-      "rewards/margins": 0.24800649285316467,
-      "rewards/rejected": -0.045810386538505554,
+      "logits/chosen": -2.798689126968384,
+      "logits/rejected": -2.7966549396514893,
+      "logps/chosen": -29.22808265686035,
+      "logps/rejected": -30.211177825927734,
+      "loss": 19.5177,
+      "rewards/accuracies": 0.699999988079071,
+      "rewards/chosen": 0.04417658969759941,
+      "rewards/margins": 0.088477224111557,
+      "rewards/rejected": -0.044300634413957596,
       "step": 180
     },
     {
       "epoch": 0.49,
-      "grad_norm": 143.0,
+      "grad_norm": 496.0,
       "learning_rate": 2.996071664294641e-06,
-      "logits/chosen": -2.9042458534240723,
-      "logits/rejected": -2.900311231613159,
-      "logps/chosen": -29.820514678955078,
-      "logps/rejected": -28.412799835205078,
-      "loss": 2.882,
+      "logits/chosen": -2.9117610454559326,
+      "logits/rejected": -2.9078633785247803,
+      "logps/chosen": -29.73740577697754,
+      "logps/rejected": -28.741764068603516,
+      "loss": 22.4526,
       "rewards/accuracies": 0.675000011920929,
-      "rewards/chosen": 0.14647993445396423,
-      "rewards/margins": 0.1685885637998581,
-      "rewards/rejected": -0.022108638659119606,
+      "rewards/chosen": 0.044930893927812576,
+      "rewards/margins": 0.08335400372743607,
+      "rewards/rejected": -0.03842311352491379,
       "step": 190
     },
     {
       "epoch": 0.52,
-      "grad_norm": 200.0,
+      "grad_norm": 446.0,
       "learning_rate": 2.7718537898066833e-06,
-      "logits/chosen": -2.9734153747558594,
-      "logits/rejected": -2.9612646102905273,
-      "logps/chosen": -32.977813720703125,
-      "logps/rejected": -30.30489730834961,
-      "loss": 2.6629,
-      "rewards/accuracies": 0.6625000238418579,
-      "rewards/chosen": 0.288481742143631,
-      "rewards/margins": 0.23680701851844788,
-      "rewards/rejected": 0.05167470499873161,
+      "logits/chosen": -2.9809372425079346,
+      "logits/rejected": -2.9683079719543457,
+      "logps/chosen": -32.74761962890625,
+      "logps/rejected": -30.57464027404785,
+      "loss": 19.5946,
+      "rewards/accuracies": 0.6875,
+      "rewards/chosen": 0.09513962268829346,
+      "rewards/margins": 0.10919489711523056,
+      "rewards/rejected": -0.01405528374016285,
       "step": 200
     },
     {
       "epoch": 0.52,
-      "eval_logits/chosen": -2.812067985534668,
-      "eval_logits/rejected": -2.8095240592956543,
-      "eval_logps/chosen": -31.236892700195312,
-      "eval_logps/rejected": -34.73383712768555,
-      "eval_loss": 1.5648894309997559,
-      "eval_rewards/accuracies": 0.5456810593605042,
-      "eval_rewards/chosen": 0.018223287537693977,
-      "eval_rewards/margins": 0.032496776431798935,
-      "eval_rewards/rejected": -0.014273490756750107,
-      "eval_runtime": 112.7253,
-      "eval_samples_per_second": 3.043,
+      "eval_logits/chosen": -2.8196890354156494,
+      "eval_logits/rejected": -2.817232370376587,
+      "eval_logps/chosen": -31.39967155456543,
+      "eval_logps/rejected": -35.00971221923828,
+      "eval_loss": 23.757455825805664,
+      "eval_rewards/accuracies": 0.5718438625335693,
+      "eval_rewards/chosen": -0.011722045950591564,
+      "eval_rewards/margins": 0.019433511421084404,
+      "eval_rewards/rejected": -0.03115556202828884,
+      "eval_runtime": 112.9888,
+      "eval_samples_per_second": 3.036,
       "eval_steps_per_second": 0.381,
       "step": 200
     },
     {
       "epoch": 0.55,
-      "grad_norm": 214.0,
+      "grad_norm": 756.0,
       "learning_rate": 2.5453962426402006e-06,
-      "logits/chosen": -2.9090185165405273,
-      "logits/rejected": -2.910029888153076,
-      "logps/chosen": -32.432430267333984,
-      "logps/rejected": -33.925514221191406,
-      "loss": 1.49,
-      "rewards/accuracies": 0.7875000238418579,
-      "rewards/chosen": 0.2403920441865921,
-      "rewards/margins": 0.26600998640060425,
-      "rewards/rejected": -0.025617951527237892,
+      "logits/chosen": -2.9178690910339355,
+      "logits/rejected": -2.9186930656433105,
+      "logps/chosen": -32.30227279663086,
+      "logps/rejected": -34.28144073486328,
+      "loss": 17.7864,
+      "rewards/accuracies": 0.7250000238418579,
+      "rewards/chosen": 0.0731135681271553,
+      "rewards/margins": 0.1151106134057045,
+      "rewards/rejected": -0.041997045278549194,
       "step": 210
     },
     {
       "epoch": 0.57,
-      "grad_norm": 202.0,
+      "grad_norm": 584.0,
       "learning_rate": 2.3185646976551794e-06,
-      "logits/chosen": -2.8856301307678223,
-      "logits/rejected": -2.9012038707733154,
-      "logps/chosen": -29.458789825439453,
-      "logps/rejected": -28.523406982421875,
-      "loss": 2.1585,
-      "rewards/accuracies": 0.7124999761581421,
-      "rewards/chosen": 0.2876998782157898,
-      "rewards/margins": 0.28575605154037476,
-      "rewards/rejected": 0.001943854964338243,
+      "logits/chosen": -2.8974623680114746,
+      "logits/rejected": -2.9132611751556396,
+      "logps/chosen": -29.29262924194336,
+      "logps/rejected": -28.876379013061523,
+      "loss": 18.1243,
+      "rewards/accuracies": 0.699999988079071,
+      "rewards/chosen": 0.08854089677333832,
+      "rewards/margins": 0.12335184961557388,
+      "rewards/rejected": -0.03481093421578407,
       "step": 220
     },
     {
       "epoch": 0.6,
-      "grad_norm": 150.0,
+      "grad_norm": 644.0,
       "learning_rate": 2.0932279108998323e-06,
-      "logits/chosen": -2.9351718425750732,
-      "logits/rejected": -2.939762592315674,
-      "logps/chosen": -30.686193466186523,
-      "logps/rejected": -31.73343276977539,
-      "loss": 1.6516,
-      "rewards/accuracies": 0.6625000238418579,
-      "rewards/chosen": 0.18626943230628967,
-      "rewards/margins": 0.1763811856508255,
-      "rewards/rejected": 0.00988826435059309,
+      "logits/chosen": -2.9454588890075684,
+      "logits/rejected": -2.9493844509124756,
+      "logps/chosen": -30.615680694580078,
+      "logps/rejected": -32.01018142700195,
+      "loss": 20.5727,
+      "rewards/accuracies": 0.699999988079071,
+      "rewards/chosen": 0.05361897870898247,
+      "rewards/margins": 0.07882187515497208,
+      "rewards/rejected": -0.02520289644598961,
       "step": 230
     },
     {
       "epoch": 0.62,
-      "grad_norm": 117.0,
+      "grad_norm": 692.0,
       "learning_rate": 1.8712423238279358e-06,
-      "logits/chosen": -2.98549485206604,
-      "logits/rejected": -2.9937961101531982,
-      "logps/chosen": -30.101831436157227,
-      "logps/rejected": -30.255285263061523,
-      "loss": 0.8256,
-      "rewards/accuracies": 0.862500011920929,
-      "rewards/chosen": 0.32123786211013794,
-      "rewards/margins": 0.329254686832428,
-      "rewards/rejected": -0.008016781881451607,
+      "logits/chosen": -3.000526189804077,
+      "logits/rejected": -3.007200241088867,
+      "logps/chosen": -29.994653701782227,
+      "logps/rejected": -30.485973358154297,
+      "loss": 16.6669,
+      "rewards/accuracies": 0.800000011920929,
+      "rewards/chosen": 0.09102725982666016,
+      "rewards/margins": 0.11610045284032822,
+      "rewards/rejected": -0.02507319487631321,
       "step": 240
     },
     {
       "epoch": 0.65,
-      "grad_norm": 350.0,
+      "grad_norm": 828.0,
       "learning_rate": 1.6544367689701824e-06,
-      "logits/chosen": -2.817945718765259,
-      "logits/rejected": -2.808492660522461,
-      "logps/chosen": -26.3896427154541,
-      "logps/rejected": -29.31142807006836,
-      "loss": 1.6788,
-      "rewards/accuracies": 0.6625000238418579,
-      "rewards/chosen": 0.22583059966564178,
-      "rewards/margins": 0.13774575293064117,
-      "rewards/rejected": 0.0880848616361618,
+      "logits/chosen": -2.8318004608154297,
+      "logits/rejected": -2.821593761444092,
+      "logps/chosen": -26.242666244506836,
+      "logps/rejected": -29.69923973083496,
+      "loss": 19.8026,
+      "rewards/accuracies": 0.762499988079071,
+      "rewards/chosen": 0.07115494459867477,
+      "rewards/margins": 0.08791451901197433,
+      "rewards/rejected": -0.016759568825364113,
       "step": 250
     },
     {
       "epoch": 0.68,
-      "grad_norm": 105.5,
+      "grad_norm": 482.0,
       "learning_rate": 1.4445974030621963e-06,
-      "logits/chosen": -2.7999472618103027,
-      "logits/rejected": -2.8214704990386963,
-      "logps/chosen": -29.094181060791016,
-      "logps/rejected": -34.21991729736328,
-      "loss": 1.111,
-      "rewards/accuracies": 0.8374999761581421,
-      "rewards/chosen": 0.294312059879303,
-      "rewards/margins": 0.3126428425312042,
-      "rewards/rejected": -0.01833074726164341,
+      "logits/chosen": -2.813377857208252,
+      "logits/rejected": -2.8331103324890137,
+      "logps/chosen": -28.772018432617188,
+      "logps/rejected": -34.57004165649414,
+      "loss": 16.4634,
+      "rewards/accuracies": 0.800000011920929,
+      "rewards/chosen": 0.1057942658662796,
+      "rewards/margins": 0.14538946747779846,
+      "rewards/rejected": -0.039595186710357666,
       "step": 260
     },
     {
       "epoch": 0.7,
-      "grad_norm": 124.0,
+      "grad_norm": 792.0,
       "learning_rate": 1.243452991757889e-06,
-      "logits/chosen": -2.942542791366577,
-      "logits/rejected": -2.9476306438446045,
-      "logps/chosen": -30.103805541992188,
-      "logps/rejected": -29.8814640045166,
-      "loss": 1.3753,
-      "rewards/accuracies": 0.7875000238418579,
-      "rewards/chosen": 0.27102118730545044,
-      "rewards/margins": 0.27750203013420105,
-      "rewards/rejected": -0.0064808703027665615,
+      "logits/chosen": -2.955620527267456,
+      "logits/rejected": -2.9620866775512695,
+      "logps/chosen": -30.066015243530273,
+      "logps/rejected": -30.39284324645996,
+      "loss": 16.2985,
+      "rewards/accuracies": 0.8374999761581421,
+      "rewards/chosen": 0.071534164249897,
+      "rewards/margins": 0.1242922693490982,
+      "rewards/rejected": -0.0527581050992012,
       "step": 270
     },
     {
       "epoch": 0.73,
-      "grad_norm": 96.0,
+      "grad_norm": 552.0,
       "learning_rate": 1.0526606671603523e-06,
-      "logits/chosen": -2.9563868045806885,
-      "logits/rejected": -2.9434051513671875,
-      "logps/chosen": -30.231460571289062,
-      "logps/rejected": -28.510696411132812,
-      "loss": 1.8713,
-      "rewards/accuracies": 0.6875,
-      "rewards/chosen": 0.196088969707489,
-      "rewards/margins": 0.1478276252746582,
-      "rewards/rejected": 0.0482613667845726,
+      "logits/chosen": -2.9709465503692627,
+      "logits/rejected": -2.957740306854248,
+      "logps/chosen": -30.085498809814453,
+      "logps/rejected": -28.866928100585938,
+      "loss": 21.3586,
+      "rewards/accuracies": 0.7875000238418579,
+      "rewards/chosen": 0.06361865997314453,
+      "rewards/margins": 0.08717648684978485,
+      "rewards/rejected": -0.02355783060193062,
       "step": 280
     },
     {
       "epoch": 0.75,
-      "grad_norm": 176.0,
+      "grad_norm": 616.0,
       "learning_rate": 8.737922755071455e-07,
-      "logits/chosen": -2.884552478790283,
-      "logits/rejected": -2.8662493228912354,
-      "logps/chosen": -31.870113372802734,
-      "logps/rejected": -30.56744384765625,
-      "loss": 1.5277,
-      "rewards/accuracies": 0.7749999761581421,
-      "rewards/chosen": 0.3064747452735901,
-      "rewards/margins": 0.32412776350975037,
-      "rewards/rejected": -0.017653033137321472,
+      "logits/chosen": -2.89894437789917,
+      "logits/rejected": -2.881261110305786,
+      "logps/chosen": -31.43949317932129,
+      "logps/rejected": -31.19020652770996,
+      "loss": 13.7665,
+      "rewards/accuracies": 0.824999988079071,
+      "rewards/chosen": 0.11968123912811279,
+      "rewards/margins": 0.1863711178302765,
+      "rewards/rejected": -0.0666898787021637,
       "step": 290
     },
     {
       "epoch": 0.78,
-      "grad_norm": 372.0,
+      "grad_norm": 760.0,
       "learning_rate": 7.08321427484816e-07,
-      "logits/chosen": -2.889061689376831,
-      "logits/rejected": -2.885744571685791,
-      "logps/chosen": -31.58562660217285,
-      "logps/rejected": -27.505157470703125,
-      "loss": 1.5067,
-      "rewards/accuracies": 0.7250000238418579,
-      "rewards/chosen": 0.254812091588974,
-      "rewards/margins": 0.21544210612773895,
-      "rewards/rejected": 0.039369963109493256,
+      "logits/chosen": -2.9006505012512207,
+      "logits/rejected": -2.896925449371338,
+      "logps/chosen": -31.2346134185791,
+      "logps/rejected": -27.882694244384766,
+      "loss": 18.7484,
+      "rewards/accuracies": 0.7124999761581421,
+      "rewards/chosen": 0.0988042801618576,
+      "rewards/margins": 0.12671498954296112,
+      "rewards/rejected": -0.027910714969038963,
       "step": 300
     },
     {
       "epoch": 0.78,
-      "eval_logits/chosen": -2.80647873878479,
-      "eval_logits/rejected": -2.8040761947631836,
-      "eval_logps/chosen": -31.251041412353516,
-      "eval_logps/rejected": -34.71211624145508,
-      "eval_loss": 1.6903845071792603,
-      "eval_rewards/accuracies": 0.5573089718818665,
-      "eval_rewards/chosen": 0.012563018128275871,
-      "eval_rewards/margins": 0.018146870657801628,
-      "eval_rewards/rejected": -0.005583854392170906,
-      "eval_runtime": 113.035,
-      "eval_samples_per_second": 3.034,
+      "eval_logits/chosen": -2.818225860595703,
+      "eval_logits/rejected": -2.81588077545166,
+      "eval_logps/chosen": -31.401628494262695,
+      "eval_logps/rejected": -35.010406494140625,
+      "eval_loss": 23.85563087463379,
+      "eval_rewards/accuracies": 0.5718438625335693,
+      "eval_rewards/chosen": -0.011917698197066784,
+      "eval_rewards/margins": 0.019307522103190422,
+      "eval_rewards/rejected": -0.03122522123157978,
+      "eval_runtime": 113.0254,
+      "eval_samples_per_second": 3.035,
       "eval_steps_per_second": 0.38,
       "step": 300
     },
     {
       "epoch": 0.81,
-      "grad_norm": 129.0,
+      "grad_norm": 784.0,
       "learning_rate": 5.576113578589035e-07,
-      "logits/chosen": -2.7671875953674316,
-      "logits/rejected": -2.785642147064209,
-      "logps/chosen": -28.604686737060547,
-      "logps/rejected": -30.709314346313477,
-      "loss": 1.2664,
-      "rewards/accuracies": 0.7875000238418579,
-      "rewards/chosen": 0.24989323318004608,
-      "rewards/margins": 0.24081555008888245,
-      "rewards/rejected": 0.009077723138034344,
+      "logits/chosen": -2.781919002532959,
+      "logits/rejected": -2.7987494468688965,
+      "logps/chosen": -28.510242462158203,
+      "logps/rejected": -31.248886108398438,
+      "loss": 17.0117,
+      "rewards/accuracies": 0.8500000238418579,
+      "rewards/chosen": 0.07191774249076843,
+      "rewards/margins": 0.1236056461930275,
+      "rewards/rejected": -0.051687903702259064,
       "step": 310
     },
     {
       "epoch": 0.83,
-      "grad_norm": 92.5,
+      "grad_norm": 524.0,
       "learning_rate": 4.229036944380913e-07,
-      "logits/chosen": -3.014265537261963,
-      "logits/rejected": -2.9993736743927,
-      "logps/chosen": -29.781946182250977,
-      "logps/rejected": -28.38212013244629,
-      "loss": 1.4826,
-      "rewards/accuracies": 0.800000011920929,
-      "rewards/chosen": 0.17185598611831665,
-      "rewards/margins": 0.19719351828098297,
-      "rewards/rejected": -0.025337522849440575,
+      "logits/chosen": -3.0269644260406494,
+      "logits/rejected": -3.012279748916626,
+      "logps/chosen": -29.241586685180664,
+      "logps/rejected": -28.835968017578125,
+      "loss": 15.086,
+      "rewards/accuracies": 0.8374999761581421,
+      "rewards/chosen": 0.09699983894824982,
+      "rewards/margins": 0.1487189531326294,
+      "rewards/rejected": -0.051719121634960175,
       "step": 320
     },
     {
       "epoch": 0.86,
-      "grad_norm": 163.0,
+      "grad_norm": 360.0,
       "learning_rate": 3.053082288996112e-07,
-      "logits/chosen": -2.9294309616088867,
-      "logits/rejected": -2.9114794731140137,
-      "logps/chosen": -27.724899291992188,
-      "logps/rejected": -30.527271270751953,
-      "loss": 1.2966,
-      "rewards/accuracies": 0.8125,
-      "rewards/chosen": 0.2548690140247345,
-      "rewards/margins": 0.30911877751350403,
-      "rewards/rejected": -0.054249756038188934,
+      "logits/chosen": -2.9412882328033447,
+      "logits/rejected": -2.9239907264709473,
+      "logps/chosen": -27.266399383544922,
+      "logps/rejected": -31.0549259185791,
+      "loss": 14.0005,
+      "rewards/accuracies": 0.862500011920929,
+      "rewards/chosen": 0.10956673324108124,
+      "rewards/margins": 0.1758948117494583,
+      "rewards/rejected": -0.06632807105779648,
       "step": 330
     },
     {
       "epoch": 0.88,
-      "grad_norm": 191.0,
+      "grad_norm": 716.0,
       "learning_rate": 2.0579377374915805e-07,
-      "logits/chosen": -3.144012928009033,
-      "logits/rejected": -3.1506927013397217,
-      "logps/chosen": -31.059814453125,
-      "logps/rejected": -32.82073211669922,
-      "loss": 1.5694,
-      "rewards/accuracies": 0.762499988079071,
-      "rewards/chosen": 0.23175635933876038,
-      "rewards/margins": 0.2816471457481384,
-      "rewards/rejected": -0.04989078640937805,
+      "logits/chosen": -3.158176898956299,
+      "logits/rejected": -3.163879871368408,
+      "logps/chosen": -30.825565338134766,
+      "logps/rejected": -33.42596435546875,
+      "loss": 15.6936,
+      "rewards/accuracies": 0.7875000238418579,
+      "rewards/chosen": 0.08136396110057831,
+      "rewards/margins": 0.15435989201068878,
+      "rewards/rejected": -0.07299593091011047,
       "step": 340
     },
     {
       "epoch": 0.91,
-      "grad_norm": 112.5,
+      "grad_norm": 552.0,
       "learning_rate": 1.2518018074041684e-07,
-      "logits/chosen": -3.021515130996704,
-      "logits/rejected": -3.02475643157959,
-      "logps/chosen": -30.08322525024414,
-      "logps/rejected": -31.63114356994629,
-      "loss": 1.3735,
+      "logits/chosen": -3.034761428833008,
+      "logits/rejected": -3.038020372390747,
+      "logps/chosen": -29.743371963500977,
+      "logps/rejected": -31.885990142822266,
+      "loss": 17.0818,
       "rewards/accuracies": 0.7875000238418579,
-      "rewards/chosen": 0.2894637882709503,
-      "rewards/margins": 0.3365970253944397,
-      "rewards/rejected": -0.04713326320052147,
+      "rewards/chosen": 0.1063508540391922,
+      "rewards/margins": 0.1436188519001007,
+      "rewards/rejected": -0.03726799413561821,
       "step": 350
     },
     {
       "epoch": 0.94,
-      "grad_norm": 215.0,
+      "grad_norm": 608.0,
       "learning_rate": 6.41315865106129e-08,
-      "logits/chosen": -2.8619635105133057,
-      "logits/rejected": -2.8640754222869873,
-      "logps/chosen": -27.785686492919922,
-      "logps/rejected": -29.807357788085938,
-      "loss": 1.4796,
-      "rewards/accuracies": 0.8125,
-      "rewards/chosen": 0.34561628103256226,
-      "rewards/margins": 0.33618348836898804,
-      "rewards/rejected": 0.00943276472389698,
+      "logits/chosen": -2.8735318183898926,
+      "logits/rejected": -2.8751091957092285,
+      "logps/chosen": -27.433670043945312,
+      "logps/rejected": -30.212413787841797,
+      "loss": 15.0018,
+      "rewards/accuracies": 0.800000011920929,
+      "rewards/chosen": 0.12160570919513702,
+      "rewards/margins": 0.1597532331943512,
+      "rewards/rejected": -0.038147542625665665,
       "step": 360
     },
     {
       "epoch": 0.96,
-      "grad_norm": 127.5,
+      "grad_norm": 984.0,
       "learning_rate": 2.3150941078050325e-08,
-      "logits/chosen": -2.9357876777648926,
-      "logits/rejected": -2.9344675540924072,
-      "logps/chosen": -30.16611671447754,
-      "logps/rejected": -32.088932037353516,
-      "loss": 1.1067,
+      "logits/chosen": -2.950293779373169,
+      "logits/rejected": -2.947977066040039,
+      "logps/chosen": -29.780099868774414,
+      "logps/rejected": -32.294410705566406,
+      "loss": 17.366,
       "rewards/accuracies": 0.800000011920929,
-      "rewards/chosen": 0.24821257591247559,
-      "rewards/margins": 0.31256765127182007,
-      "rewards/rejected": -0.06435510516166687,
+      "rewards/chosen": 0.10065505653619766,
+      "rewards/margins": 0.13729141652584076,
+      "rewards/rejected": -0.03663638234138489,
       "step": 370
     },
     {
       "epoch": 0.99,
-      "grad_norm": 552.0,
+      "grad_norm": 556.0,
       "learning_rate": 2.575864278703266e-09,
-      "logits/chosen": -2.89326548576355,
-      "logits/rejected": -2.8758132457733154,
-      "logps/chosen": -28.290267944335938,
-      "logps/rejected": -27.925792694091797,
-      "loss": 1.6812,
-      "rewards/accuracies": 0.762499988079071,
-      "rewards/chosen": 0.2619072496891022,
-      "rewards/margins": 0.27637776732444763,
-      "rewards/rejected": -0.01447053998708725,
+      "logits/chosen": -2.906440258026123,
+      "logits/rejected": -2.8900303840637207,
+      "logps/chosen": -28.174755096435547,
+      "logps/rejected": -28.421972274780273,
+      "loss": 16.0238,
+      "rewards/accuracies": 0.8500000238418579,
+      "rewards/chosen": 0.07702798396348953,
+      "rewards/margins": 0.13026344776153564,
+      "rewards/rejected": -0.05323547124862671,
       "step": 380
     },
     {
       "epoch": 1.0,
       "step": 385,
       "total_flos": 0.0,
-      "train_loss": 1.7377146052075672,
-      "train_runtime": 2719.6001,
-      "train_samples_per_second": 1.132,
-      "train_steps_per_second": 0.142
+      "train_loss": 19.552858486423244,
+      "train_runtime": 2721.3646,
+      "train_samples_per_second": 1.131,
+      "train_steps_per_second": 0.141
     }
   ],
   "logging_steps": 10,