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: 23.7465
-- Rewards/chosen: -0.0136
-- Rewards/rejected: -0.0343
-- Rewards/accuracies: 0.5801
-- Rewards/margins: 0.0207
-- Logps/rejected: -35.0411
-- Logps/chosen: -31.4188
-- Logits/rejected: -2.8155
-- Logits/chosen: -2.8182
+- Loss: 0.4705
+- Rewards/chosen: 0.0616
+- Rewards/rejected: 0.0107
+- Rewards/accuracies: 0.5370
+- Rewards/margins: 0.0508
+- Logps/rejected: -34.6847
+- Logps/chosen: -31.2055
+- Logits/rejected: -2.8071
+- Logits/chosen: -2.8098
 
 ## 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 |
 |:-------------:|:-----:|:----:|:---------------:|:--------------:|:----------------:|:------------------:|:---------------:|:--------------:|:------------:|:---------------:|:-------------:|
-| 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       |
+| 1.0237        | 0.26  | 100  | 0.4815          | 0.0235         | -0.0157          | 0.5486             | 0.0392          | -34.7178       | -31.2531     | -2.8136         | -2.8161       |
+| 2.9692        | 0.52  | 200  | 0.4973          | 0.0315         | -0.0118          | 0.5631             | 0.0433          | -34.7129       | -31.2431     | -2.8099         | -2.8121       |
+| 0.9535        | 0.78  | 300  | 0.4705          | 0.0616         | 0.0107           | 0.5370             | 0.0508          | -34.6847       | -31.2055     | -2.8071         | -2.8098       |
 
 
 ### Framework versions

adapter_model.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:3f70c3bc27f637db64790317a9f22ad0cb65faee6c0ee6d71fc2a86166518cc2
+oid sha256:d26b2e35065889ded962873c4dfb1e3fe51d0957dbd63814ca1cb40582f4a534
 size 671150064

all_results.json

@@ -1,20 +1,7 @@
 {
     "epoch": 1.0,
-    "eval_logits/chosen": -2.818209171295166,
-    "eval_logits/rejected": -2.815542459487915,
-    "eval_logps/chosen": -31.4188232421875,
-    "eval_logps/rejected": -35.04113006591797,
-    "eval_loss": 23.746549606323242,
-    "eval_rewards/accuracies": 0.5801494717597961,
-    "eval_rewards/chosen": -0.013637068681418896,
-    "eval_rewards/margins": 0.020660024136304855,
-    "eval_rewards/rejected": -0.03429709002375603,
-    "eval_runtime": 112.6126,
-    "eval_samples": 343,
-    "eval_samples_per_second": 3.046,
-    "eval_steps_per_second": 0.382,
-    "train_loss": 19.552858486423244,
-    "train_runtime": 2721.3646,
+    "train_loss": 1.1209261448352368,
+    "train_runtime": 2723.1709,
     "train_samples": 3079,
     "train_samples_per_second": 1.131,
     "train_steps_per_second": 0.141

train_results.json

@@ -1,7 +1,7 @@
 {
     "epoch": 1.0,
-    "train_loss": 19.552858486423244,
-    "train_runtime": 2721.3646,
+    "train_loss": 1.1209261448352368,
+    "train_runtime": 2723.1709,
     "train_samples": 3079,
     "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": 792.0,
+      "grad_norm": 99.0,
       "learning_rate": 1.282051282051282e-07,
       "logits/chosen": -2.7358343601226807,
       "logits/rejected": -2.7480404376983643,
       "logps/chosen": -27.35565757751465,
       "logps/rejected": -21.06114387512207,
-      "loss": 25.0,
+      "loss": 0.3906,
       "rewards/accuracies": 0.0,
       "rewards/chosen": 0.0,
       "rewards/margins": 0.0,
@@ -25,628 +25,628 @@
     },
     {
       "epoch": 0.03,
-      "grad_norm": 1384.0,
+      "grad_norm": 161.0,
       "learning_rate": 1.282051282051282e-06,
-      "logits/chosen": -3.0097055435180664,
-      "logits/rejected": -2.9980692863464355,
-      "logps/chosen": -33.19136047363281,
-      "logps/rejected": -31.962371826171875,
-      "loss": 25.1536,
+      "logits/chosen": -3.0094523429870605,
+      "logits/rejected": -2.998090982437134,
+      "logps/chosen": -33.19883728027344,
+      "logps/rejected": -31.973581314086914,
+      "loss": 0.4457,
       "rewards/accuracies": 0.4166666567325592,
-      "rewards/chosen": -0.0010418971069157124,
-      "rewards/margins": -0.001195515040308237,
-      "rewards/rejected": 0.00015361772966571152,
+      "rewards/chosen": -0.014314251020550728,
+      "rewards/margins": -0.006575056351721287,
+      "rewards/rejected": -0.007739194668829441,
       "step": 10
     },
     {
       "epoch": 0.05,
-      "grad_norm": 940.0,
+      "grad_norm": 128.0,
       "learning_rate": 2.564102564102564e-06,
-      "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,
+      "logits/chosen": -2.8992228507995605,
+      "logits/rejected": -2.8941798210144043,
+      "logps/chosen": -32.46299362182617,
+      "logps/rejected": -28.942882537841797,
+      "loss": 0.4894,
+      "rewards/accuracies": 0.4375,
+      "rewards/chosen": 0.009680476039648056,
+      "rewards/margins": -0.021505311131477356,
+      "rewards/rejected": 0.031185787171125412,
       "step": 20
     },
     {
       "epoch": 0.08,
-      "grad_norm": 872.0,
+      "grad_norm": 129.0,
       "learning_rate": 3.846153846153847e-06,
-      "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,
+      "logits/chosen": -3.097111225128174,
+      "logits/rejected": -3.109170436859131,
+      "logps/chosen": -32.810890197753906,
+      "logps/rejected": -30.153121948242188,
+      "loss": 0.4421,
+      "rewards/accuracies": 0.512499988079071,
+      "rewards/chosen": 0.07823446393013,
+      "rewards/margins": 0.04388953000307083,
+      "rewards/rejected": 0.03434494137763977,
       "step": 30
     },
     {
       "epoch": 0.1,
-      "grad_norm": 788.0,
+      "grad_norm": 200.0,
       "learning_rate": 4.999896948438434e-06,
-      "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,
+      "logits/chosen": -2.86364483833313,
+      "logits/rejected": -2.853883981704712,
+      "logps/chosen": -31.620784759521484,
+      "logps/rejected": -32.36823654174805,
+      "loss": 0.7104,
+      "rewards/accuracies": 0.6000000238418579,
+      "rewards/chosen": 0.1676289290189743,
+      "rewards/margins": 0.17625470459461212,
+      "rewards/rejected": -0.008625739254057407,
       "step": 40
     },
     {
       "epoch": 0.13,
-      "grad_norm": 536.0,
+      "grad_norm": 229.0,
       "learning_rate": 4.987541037542187e-06,
-      "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,
+      "logits/chosen": -2.886505603790283,
+      "logits/rejected": -2.884448528289795,
+      "logps/chosen": -29.601898193359375,
+      "logps/rejected": -30.147045135498047,
+      "loss": 0.8433,
+      "rewards/accuracies": 0.550000011920929,
+      "rewards/chosen": 0.11060991138219833,
+      "rewards/margins": 0.15709055960178375,
+      "rewards/rejected": -0.04648064821958542,
       "step": 50
     },
     {
       "epoch": 0.16,
-      "grad_norm": 784.0,
+      "grad_norm": 137.0,
       "learning_rate": 4.954691471941119e-06,
-      "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,
+      "logits/chosen": -2.9212472438812256,
+      "logits/rejected": -2.921980381011963,
+      "logps/chosen": -30.149127960205078,
+      "logps/rejected": -28.14251136779785,
+      "loss": 0.5196,
+      "rewards/accuracies": 0.625,
+      "rewards/chosen": -0.01548430323600769,
+      "rewards/margins": 0.13397768139839172,
+      "rewards/rejected": -0.1494619995355606,
       "step": 60
     },
     {
       "epoch": 0.18,
-      "grad_norm": 1600.0,
+      "grad_norm": 1104.0,
       "learning_rate": 4.901618883413549e-06,
-      "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,
+      "logits/chosen": -3.0032145977020264,
+      "logits/rejected": -3.010529041290283,
+      "logps/chosen": -29.345632553100586,
+      "logps/rejected": -31.078609466552734,
+      "loss": 1.3076,
+      "rewards/accuracies": 0.612500011920929,
+      "rewards/chosen": -0.0517343208193779,
+      "rewards/margins": 0.09771490097045898,
+      "rewards/rejected": -0.1494491994380951,
       "step": 70
     },
     {
       "epoch": 0.21,
-      "grad_norm": 1072.0,
+      "grad_norm": 186.0,
       "learning_rate": 4.828760511501322e-06,
-      "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,
+      "logits/chosen": -2.816901922225952,
+      "logits/rejected": -2.832296848297119,
+      "logps/chosen": -29.427043914794922,
+      "logps/rejected": -29.738506317138672,
+      "loss": 0.6608,
+      "rewards/accuracies": 0.5625,
+      "rewards/chosen": 0.0909457728266716,
+      "rewards/margins": 0.14232560992240906,
+      "rewards/rejected": -0.051379840821027756,
       "step": 80
     },
     {
       "epoch": 0.23,
-      "grad_norm": 740.0,
+      "grad_norm": 216.0,
       "learning_rate": 4.7367166013034295e-06,
-      "logits/chosen": -2.8982763290405273,
-      "logits/rejected": -2.8804843425750732,
-      "logps/chosen": -32.65864181518555,
-      "logps/rejected": -30.32712745666504,
-      "loss": 25.4545,
+      "logits/chosen": -2.902531623840332,
+      "logits/rejected": -2.884403705596924,
+      "logps/chosen": -32.63992691040039,
+      "logps/rejected": -30.051746368408203,
+      "loss": 3.8065,
       "rewards/accuracies": 0.699999988079071,
-      "rewards/chosen": 0.017779098823666573,
-      "rewards/margins": 0.04832718148827553,
-      "rewards/rejected": -0.030548080801963806,
+      "rewards/chosen": 0.15720424056053162,
+      "rewards/margins": 0.18128542602062225,
+      "rewards/rejected": -0.024081196635961533,
       "step": 90
     },
     {
       "epoch": 0.26,
-      "grad_norm": 720.0,
+      "grad_norm": 162.0,
       "learning_rate": 4.626245458345211e-06,
-      "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": 0.026112746447324753,
-      "rewards/margins": 0.034251101315021515,
-      "rewards/rejected": -0.008138353936374187,
+      "logits/chosen": -3.0053141117095947,
+      "logits/rejected": -3.0066592693328857,
+      "logps/chosen": -31.79007911682129,
+      "logps/rejected": -30.816768646240234,
+      "loss": 1.0237,
+      "rewards/accuracies": 0.612500011920929,
+      "rewards/chosen": 0.2134353667497635,
+      "rewards/margins": 0.21390971541404724,
+      "rewards/rejected": -0.000474362081149593,
       "step": 100
     },
     {
       "epoch": 0.26,
-      "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_logits/chosen": -2.8161001205444336,
+      "eval_logits/rejected": -2.8135948181152344,
+      "eval_logps/chosen": -31.253108978271484,
+      "eval_logps/rejected": -34.717830657958984,
+      "eval_loss": 0.4814997613430023,
+      "eval_rewards/accuracies": 0.5485880374908447,
+      "eval_rewards/chosen": 0.023475930094718933,
+      "eval_rewards/margins": 0.039213381707668304,
+      "eval_rewards/rejected": -0.015737449750304222,
+      "eval_runtime": 113.131,
       "eval_samples_per_second": 3.032,
       "eval_steps_per_second": 0.38,
       "step": 100
     },
     {
       "epoch": 0.29,
-      "grad_norm": 1008.0,
+      "grad_norm": 156.0,
       "learning_rate": 4.498257201263691e-06,
-      "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,
+      "logits/chosen": -2.962667942047119,
+      "logits/rejected": -2.9380180835723877,
+      "logps/chosen": -31.729461669921875,
+      "logps/rejected": -31.084197998046875,
+      "loss": 1.6614,
+      "rewards/accuracies": 0.737500011920929,
+      "rewards/chosen": 0.38986626267433167,
+      "rewards/margins": 0.33014416694641113,
+      "rewards/rejected": 0.05972205847501755,
       "step": 110
     },
     {
       "epoch": 0.31,
-      "grad_norm": 792.0,
+      "grad_norm": 100.5,
       "learning_rate": 4.353806263777678e-06,
-      "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,
+      "logits/chosen": -3.0431807041168213,
+      "logits/rejected": -3.072758197784424,
+      "logps/chosen": -28.905542373657227,
+      "logps/rejected": -34.141502380371094,
+      "loss": 0.822,
+      "rewards/accuracies": 0.637499988079071,
+      "rewards/chosen": 0.20668914914131165,
+      "rewards/margins": 0.20185013115406036,
+      "rewards/rejected": 0.004838997032493353,
       "step": 120
     },
     {
       "epoch": 0.34,
-      "grad_norm": 668.0,
+      "grad_norm": 73.5,
       "learning_rate": 4.1940827077152755e-06,
-      "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": 0.037400223314762115,
-      "rewards/margins": 0.05806100368499756,
-      "rewards/rejected": -0.020660774782299995,
+      "logits/chosen": -2.743009090423584,
+      "logits/rejected": -2.7388014793395996,
+      "logps/chosen": -28.7690486907959,
+      "logps/rejected": -30.156269073486328,
+      "loss": 0.7399,
+      "rewards/accuracies": 0.637499988079071,
+      "rewards/chosen": 0.16450510919094086,
+      "rewards/margins": 0.1824747771024704,
+      "rewards/rejected": -0.01796967163681984,
       "step": 130
     },
     {
       "epoch": 0.36,
-      "grad_norm": 624.0,
+      "grad_norm": 197.0,
       "learning_rate": 4.0204024186666215e-06,
-      "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,
+      "logits/chosen": -3.015079975128174,
+      "logits/rejected": -3.0121235847473145,
+      "logps/chosen": -27.263751983642578,
+      "logps/rejected": -31.766677856445312,
+      "loss": 0.8956,
+      "rewards/accuracies": 0.625,
+      "rewards/chosen": 0.17797937989234924,
+      "rewards/margins": 0.2659076154232025,
+      "rewards/rejected": -0.08792825043201447,
       "step": 140
     },
     {
       "epoch": 0.39,
-      "grad_norm": 668.0,
+      "grad_norm": 126.0,
       "learning_rate": 3.834196265035119e-06,
-      "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,
+      "logits/chosen": -2.807605028152466,
+      "logits/rejected": -2.8025310039520264,
+      "logps/chosen": -27.548370361328125,
+      "logps/rejected": -31.246013641357422,
+      "loss": 0.6864,
+      "rewards/accuracies": 0.699999988079071,
+      "rewards/chosen": 0.22170917689800262,
+      "rewards/margins": 0.23960641026496887,
+      "rewards/rejected": -0.01789720356464386,
       "step": 150
     },
     {
       "epoch": 0.42,
-      "grad_norm": 872.0,
+      "grad_norm": 97.0,
       "learning_rate": 3.636998309800573e-06,
-      "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,
+      "logits/chosen": -3.123861312866211,
+      "logits/rejected": -3.10616397857666,
+      "logps/chosen": -32.041046142578125,
+      "logps/rejected": -28.996301651000977,
+      "loss": 1.2758,
+      "rewards/accuracies": 0.612500011920929,
+      "rewards/chosen": 0.36053839325904846,
+      "rewards/margins": 0.3737615942955017,
+      "rewards/rejected": -0.013223287649452686,
       "step": 160
     },
     {
       "epoch": 0.44,
-      "grad_norm": 688.0,
+      "grad_norm": 150.0,
       "learning_rate": 3.4304331721118078e-06,
-      "logits/chosen": -2.9469733238220215,
-      "logits/rejected": -2.9541804790496826,
-      "logps/chosen": -29.33526039123535,
-      "logps/rejected": -31.71769142150879,
-      "loss": 17.9149,
+      "logits/chosen": -2.9405550956726074,
+      "logits/rejected": -2.947495222091675,
+      "logps/chosen": -29.35744857788086,
+      "logps/rejected": -31.04239845275879,
+      "loss": 1.0844,
       "rewards/accuracies": 0.7124999761581421,
-      "rewards/chosen": 0.0559355802834034,
-      "rewards/margins": 0.11775630712509155,
-      "rewards/rejected": -0.06182072311639786,
+      "rewards/chosen": 0.42973223328590393,
+      "rewards/margins": 0.3840641677379608,
+      "rewards/rejected": 0.045668020844459534,
       "step": 170
     },
     {
       "epoch": 0.47,
-      "grad_norm": 688.0,
+      "grad_norm": 87.5,
       "learning_rate": 3.2162026428305436e-06,
-      "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,
+      "logits/chosen": -2.7875921726226807,
+      "logits/rejected": -2.784825086593628,
+      "logps/chosen": -29.229150772094727,
+      "logps/rejected": -29.77825927734375,
+      "loss": 0.7375,
+      "rewards/accuracies": 0.6625000238418579,
+      "rewards/chosen": 0.35255879163742065,
+      "rewards/margins": 0.36062949895858765,
+      "rewards/rejected": -0.008070724084973335,
       "step": 180
     },
     {
       "epoch": 0.49,
-      "grad_norm": 496.0,
+      "grad_norm": 314.0,
       "learning_rate": 2.996071664294641e-06,
-      "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": 0.044930893927812576,
-      "rewards/margins": 0.08335400372743607,
-      "rewards/rejected": -0.03842311352491379,
+      "logits/chosen": -2.905453681945801,
+      "logits/rejected": -2.901573896408081,
+      "logps/chosen": -29.90896224975586,
+      "logps/rejected": -28.3870906829834,
+      "loss": 1.6507,
+      "rewards/accuracies": 0.5874999761581421,
+      "rewards/chosen": 0.22220328450202942,
+      "rewards/margins": 0.24585184454917908,
+      "rewards/rejected": -0.023648545145988464,
       "step": 190
     },
     {
       "epoch": 0.52,
-      "grad_norm": 446.0,
+      "grad_norm": 320.0,
       "learning_rate": 2.7718537898066833e-06,
-      "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,
+      "logits/chosen": -2.97483491897583,
+      "logits/rejected": -2.9631428718566895,
+      "logps/chosen": -33.094482421875,
+      "logps/rejected": -30.27695655822754,
+      "loss": 2.9692,
+      "rewards/accuracies": 0.6499999761581421,
+      "rewards/chosen": 0.48363131284713745,
+      "rewards/margins": 0.35792380571365356,
+      "rewards/rejected": 0.12570755183696747,
       "step": 200
     },
     {
       "epoch": 0.52,
-      "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,
+      "eval_logits/chosen": -2.812093496322632,
+      "eval_logits/rejected": -2.809893846511841,
+      "eval_logps/chosen": -31.243083953857422,
+      "eval_logps/rejected": -34.712913513183594,
+      "eval_loss": 0.4973182678222656,
+      "eval_rewards/accuracies": 0.5631229281425476,
+      "eval_rewards/chosen": 0.03149374574422836,
+      "eval_rewards/margins": 0.04329964146018028,
+      "eval_rewards/rejected": -0.011805894784629345,
+      "eval_runtime": 113.1074,
+      "eval_samples_per_second": 3.033,
+      "eval_steps_per_second": 0.38,
       "step": 200
     },
     {
       "epoch": 0.55,
-      "grad_norm": 756.0,
+      "grad_norm": 112.5,
       "learning_rate": 2.5453962426402006e-06,
-      "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,
+      "logits/chosen": -2.9090988636016846,
+      "logits/rejected": -2.9100005626678467,
+      "logps/chosen": -32.562644958496094,
+      "logps/rejected": -33.86018371582031,
+      "loss": 1.264,
+      "rewards/accuracies": 0.7875000238418579,
+      "rewards/chosen": 0.37660837173461914,
+      "rewards/margins": 0.375582754611969,
+      "rewards/rejected": 0.0010256364475935698,
       "step": 210
     },
     {
       "epoch": 0.57,
-      "grad_norm": 584.0,
+      "grad_norm": 204.0,
       "learning_rate": 2.3185646976551794e-06,
-      "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,
+      "logits/chosen": -2.8871023654937744,
+      "logits/rejected": -2.90246844291687,
+      "logps/chosen": -29.662246704101562,
+      "logps/rejected": -28.47257423400879,
+      "loss": 1.4961,
+      "rewards/accuracies": 0.7124999761581421,
+      "rewards/chosen": 0.4126364588737488,
+      "rewards/margins": 0.3680770695209503,
+      "rewards/rejected": 0.04455941915512085,
       "step": 220
     },
     {
       "epoch": 0.6,
-      "grad_norm": 644.0,
+      "grad_norm": 87.5,
       "learning_rate": 2.0932279108998323e-06,
-      "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,
+      "logits/chosen": -2.9373838901519775,
+      "logits/rejected": -2.941819667816162,
+      "logps/chosen": -30.841754913330078,
+      "logps/rejected": -31.728525161743164,
+      "loss": 0.7905,
+      "rewards/accuracies": 0.5375000238418579,
+      "rewards/chosen": 0.2480934113264084,
+      "rewards/margins": 0.2243928462266922,
+      "rewards/rejected": 0.02370055951178074,
       "step": 230
     },
     {
       "epoch": 0.62,
-      "grad_norm": 692.0,
+      "grad_norm": 73.0,
       "learning_rate": 1.8712423238279358e-06,
-      "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,
+      "logits/chosen": -2.987827777862549,
+      "logits/rejected": -2.9960155487060547,
+      "logps/chosen": -30.307302474975586,
+      "logps/rejected": -30.214122772216797,
+      "loss": 0.9428,
+      "rewards/accuracies": 0.762499988079071,
+      "rewards/chosen": 0.4780976176261902,
+      "rewards/margins": 0.46120333671569824,
+      "rewards/rejected": 0.016894282773137093,
       "step": 240
     },
     {
       "epoch": 0.65,
-      "grad_norm": 828.0,
+      "grad_norm": 296.0,
       "learning_rate": 1.6544367689701824e-06,
-      "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,
+      "logits/chosen": -2.818535566329956,
+      "logits/rejected": -2.8091442584991455,
+      "logps/chosen": -26.581512451171875,
+      "logps/rejected": -29.30862808227539,
+      "loss": 1.3618,
+      "rewards/accuracies": 0.625,
+      "rewards/chosen": 0.2981645464897156,
+      "rewards/margins": 0.11975376307964325,
+      "rewards/rejected": 0.17841079831123352,
       "step": 250
     },
     {
       "epoch": 0.68,
-      "grad_norm": 482.0,
+      "grad_norm": 127.5,
       "learning_rate": 1.4445974030621963e-06,
-      "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,
+      "logits/chosen": -2.802452802658081,
+      "logits/rejected": -2.8227944374084473,
+      "logps/chosen": -29.27554702758789,
+      "logps/rejected": -34.11919403076172,
+      "loss": 1.2344,
+      "rewards/accuracies": 0.7250000238418579,
+      "rewards/chosen": 0.4435352385044098,
+      "rewards/margins": 0.3996209502220154,
+      "rewards/rejected": 0.043914251029491425,
       "step": 260
     },
     {
       "epoch": 0.7,
-      "grad_norm": 792.0,
+      "grad_norm": 64.5,
       "learning_rate": 1.243452991757889e-06,
-      "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,
+      "logits/chosen": -2.944246768951416,
+      "logits/rejected": -2.9497275352478027,
+      "logps/chosen": -30.331872940063477,
+      "logps/rejected": -29.763620376586914,
+      "loss": 0.9017,
+      "rewards/accuracies": 0.7124999761581421,
+      "rewards/chosen": 0.3595869243144989,
+      "rewards/margins": 0.27827444672584534,
+      "rewards/rejected": 0.08131249994039536,
       "step": 270
     },
     {
       "epoch": 0.73,
-      "grad_norm": 552.0,
+      "grad_norm": 87.0,
       "learning_rate": 1.0526606671603523e-06,
-      "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,
+      "logits/chosen": -2.9589145183563232,
+      "logits/rejected": -2.9460854530334473,
+      "logps/chosen": -30.4505615234375,
+      "logps/rejected": -28.466442108154297,
+      "loss": 1.0356,
+      "rewards/accuracies": 0.625,
+      "rewards/chosen": 0.2169012576341629,
+      "rewards/margins": 0.08497779071331024,
+      "rewards/rejected": 0.13192343711853027,
       "step": 280
     },
     {
       "epoch": 0.75,
-      "grad_norm": 616.0,
+      "grad_norm": 137.0,
       "learning_rate": 8.737922755071455e-07,
-      "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,
+      "logits/chosen": -2.8869829177856445,
+      "logits/rejected": -2.868814706802368,
+      "logps/chosen": -32.20147705078125,
+      "logps/rejected": -30.42227554321289,
+      "loss": 1.5568,
+      "rewards/accuracies": 0.5874999761581421,
+      "rewards/chosen": 0.3478601574897766,
+      "rewards/margins": 0.2670307755470276,
+      "rewards/rejected": 0.08082933723926544,
       "step": 290
     },
     {
       "epoch": 0.78,
-      "grad_norm": 760.0,
+      "grad_norm": 362.0,
       "learning_rate": 7.08321427484816e-07,
-      "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,
+      "logits/chosen": -2.8908376693725586,
+      "logits/rejected": -2.887605905532837,
+      "logps/chosen": -31.88857078552246,
+      "logps/rejected": -27.47490882873535,
+      "loss": 0.9535,
+      "rewards/accuracies": 0.637499988079071,
+      "rewards/chosen": 0.2672681212425232,
+      "rewards/margins": 0.16432540118694305,
+      "rewards/rejected": 0.10294272005558014,
       "step": 300
     },
     {
       "epoch": 0.78,
-      "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,
+      "eval_logits/chosen": -2.8098394870758057,
+      "eval_logits/rejected": -2.807095527648926,
+      "eval_logps/chosen": -31.205495834350586,
+      "eval_logps/rejected": -34.68474197387695,
+      "eval_loss": 0.47054800391197205,
+      "eval_rewards/accuracies": 0.5369601845741272,
+      "eval_rewards/chosen": 0.0615648552775383,
+      "eval_rewards/margins": 0.050835300236940384,
+      "eval_rewards/rejected": 0.010729559697210789,
+      "eval_runtime": 112.9246,
+      "eval_samples_per_second": 3.037,
+      "eval_steps_per_second": 0.381,
       "step": 300
     },
     {
       "epoch": 0.81,
-      "grad_norm": 784.0,
+      "grad_norm": 139.0,
       "learning_rate": 5.576113578589035e-07,
-      "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,
+      "logits/chosen": -2.769951343536377,
+      "logits/rejected": -2.7882049083709717,
+      "logps/chosen": -28.826038360595703,
+      "logps/rejected": -30.570903778076172,
+      "loss": 1.2033,
+      "rewards/accuracies": 0.6625000238418579,
+      "rewards/chosen": 0.322704941034317,
+      "rewards/margins": 0.19382169842720032,
+      "rewards/rejected": 0.1288832128047943,
       "step": 310
     },
     {
       "epoch": 0.83,
-      "grad_norm": 524.0,
+      "grad_norm": 142.0,
       "learning_rate": 4.229036944380913e-07,
-      "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,
+      "logits/chosen": -3.0174190998077393,
+      "logits/rejected": -3.002908706665039,
+      "logps/chosen": -29.645282745361328,
+      "logps/rejected": -28.338443756103516,
+      "loss": 1.9393,
+      "rewards/accuracies": 0.7250000238418579,
+      "rewards/chosen": 0.4530436396598816,
+      "rewards/margins": 0.4687787890434265,
+      "rewards/rejected": -0.015735208988189697,
       "step": 320
     },
     {
       "epoch": 0.86,
-      "grad_norm": 360.0,
+      "grad_norm": 75.0,
       "learning_rate": 3.053082288996112e-07,
-      "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,
+      "logits/chosen": -2.931588649749756,
+      "logits/rejected": -2.9138498306274414,
+      "logps/chosen": -27.903301239013672,
+      "logps/rejected": -30.379735946655273,
+      "loss": 0.7854,
+      "rewards/accuracies": 0.762499988079071,
+      "rewards/chosen": 0.36701759696006775,
+      "rewards/margins": 0.35749000310897827,
+      "rewards/rejected": 0.009527605026960373,
       "step": 330
     },
     {
       "epoch": 0.88,
-      "grad_norm": 716.0,
+      "grad_norm": 101.5,
       "learning_rate": 2.0579377374915805e-07,
-      "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,
+      "logits/chosen": -3.1477503776550293,
+      "logits/rejected": -3.153796672821045,
+      "logps/chosen": -31.147018432617188,
+      "logps/rejected": -32.76397705078125,
+      "loss": 0.8634,
+      "rewards/accuracies": 0.8125,
+      "rewards/chosen": 0.39374879002571106,
+      "rewards/margins": 0.44812655448913574,
+      "rewards/rejected": -0.05437783524394035,
       "step": 340
     },
     {
       "epoch": 0.91,
-      "grad_norm": 552.0,
+      "grad_norm": 176.0,
       "learning_rate": 1.2518018074041684e-07,
-      "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": 0.1063508540391922,
-      "rewards/margins": 0.1436188519001007,
-      "rewards/rejected": -0.03726799413561821,
+      "logits/chosen": -3.023674249649048,
+      "logits/rejected": -3.0265414714813232,
+      "logps/chosen": -30.397380828857422,
+      "logps/rejected": -31.44447898864746,
+      "loss": 0.5848,
+      "rewards/accuracies": 0.675000011920929,
+      "rewards/chosen": 0.3276014029979706,
+      "rewards/margins": 0.27253925800323486,
+      "rewards/rejected": 0.05506212264299393,
       "step": 350
     },
     {
       "epoch": 0.94,
-      "grad_norm": 608.0,
+      "grad_norm": 262.0,
       "learning_rate": 6.41315865106129e-08,
-      "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,
+      "logits/chosen": -2.8648974895477295,
+      "logits/rejected": -2.866847515106201,
+      "logps/chosen": -28.0252628326416,
+      "logps/rejected": -29.79150390625,
+      "loss": 1.1652,
+      "rewards/accuracies": 0.7749999761581421,
+      "rewards/chosen": 0.49957290291786194,
+      "rewards/margins": 0.4680249094963074,
+      "rewards/rejected": 0.03154800459742546,
       "step": 360
     },
     {
       "epoch": 0.96,
-      "grad_norm": 984.0,
+      "grad_norm": 161.0,
       "learning_rate": 2.3150941078050325e-08,
-      "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": 0.10065505653619766,
-      "rewards/margins": 0.13729141652584076,
-      "rewards/rejected": -0.03663638234138489,
+      "logits/chosen": -2.937594175338745,
+      "logits/rejected": -2.936284065246582,
+      "logps/chosen": -30.407318115234375,
+      "logps/rejected": -32.029598236083984,
+      "loss": 0.8323,
+      "rewards/accuracies": 0.7250000238418579,
+      "rewards/chosen": 0.30346399545669556,
+      "rewards/margins": 0.38470739126205444,
+      "rewards/rejected": -0.0812433734536171,
       "step": 370
     },
     {
       "epoch": 0.99,
-      "grad_norm": 556.0,
+      "grad_norm": 520.0,
       "learning_rate": 2.575864278703266e-09,
-      "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,
+      "logits/chosen": -2.8965048789978027,
+      "logits/rejected": -2.879058837890625,
+      "logps/chosen": -28.52047348022461,
+      "logps/rejected": -27.761425018310547,
+      "loss": 0.7634,
+      "rewards/accuracies": 0.7250000238418579,
+      "rewards/chosen": 0.33964917063713074,
+      "rewards/margins": 0.23709365725517273,
+      "rewards/rejected": 0.10255551338195801,
       "step": 380
     },
     {
       "epoch": 1.0,
       "step": 385,
       "total_flos": 0.0,
-      "train_loss": 19.552858486423244,
-      "train_runtime": 2721.3646,
+      "train_loss": 1.1209261448352368,
+      "train_runtime": 2723.1709,
       "train_samples_per_second": 1.131,
       "train_steps_per_second": 0.141
     }