Model save

README.md

@@ -15,17 +15,17 @@ should probably proofread and complete it, then remove this comment. -->
 
 # zephyr-7b-dpo-full
 
-This model is a fine-tuned version of [HuggingFaceH4/mistral-7b-sft-beta](https://huggingface.co/HuggingFaceH4/mistral-7b-sft-beta) on an unknown dataset.
+This model is a fine-tuned version of [HuggingFaceH4/mistral-7b-sft-beta](https://huggingface.co/HuggingFaceH4/mistral-7b-sft-beta) on the None dataset.
 It achieves the following results on the evaluation set:
-- Loss: 0.5694
-- Rewards/chosen: -2.1945
-- Rewards/rejected: -3.1035
-- Rewards/accuracies: 0.7383
-- Rewards/margins: 0.9091
-- Logps/rejected: -567.7068
-- Logps/chosen: -476.4866
-- Logits/rejected: -2.0444
-- Logits/chosen: -2.0871
+- Loss: 0.4276
+- Rewards/chosen: -0.5735
+- Rewards/rejected: -2.0942
+- Rewards/accuracies: 0.7031
+- Rewards/margins: 1.5207
+- Logps/rejected: -466.7732
+- Logps/chosen: -314.3910
+- Logits/rejected: -2.1887
+- Logits/chosen: -2.2631
 
 ## Model description
 
@@ -47,7 +47,7 @@ The following hyperparameters were used during training:
 - learning_rate: 5e-07
 - train_batch_size: 8
 - eval_batch_size: 8
-- seed: 42
+- seed: 1
 - distributed_type: multi-GPU
 - num_devices: 8
 - gradient_accumulation_steps: 2
@@ -62,15 +62,15 @@ 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 |
 |:-------------:|:-----:|:----:|:---------------:|:--------------:|:----------------:|:------------------:|:---------------:|:--------------:|:------------:|:---------------:|:-------------:|
-| 0.4319        | 0.21  | 100  | 0.6499          | -1.4335        | -1.7747          | 0.6328             | 0.3412          | -434.8237      | -400.3860    | -2.2766         | -2.3001       |
-| 0.3523        | 0.42  | 200  | 0.5954          | -2.0216        | -2.7375          | 0.7188             | 0.7159          | -531.1034      | -459.1997    | -2.0559         | -2.0889       |
-| 0.3611        | 0.63  | 300  | 0.5743          | -1.9913        | -2.8389          | 0.7188             | 0.8476          | -541.2482      | -456.1738    | -2.0873         | -2.1299       |
-| 0.3321        | 0.84  | 400  | 0.5694          | -2.1945        | -3.1035          | 0.7383             | 0.9091          | -567.7068      | -476.4866    | -2.0444         | -2.0871       |
+| 0.4633        | 0.21  | 100  | 0.4652          | 0.0851         | -0.1487          | 0.7031             | 0.2338          | -272.2198      | -248.5258    | -2.7842         | -2.7997       |
+| 0.4346        | 0.42  | 200  | 0.4392          | -0.4511        | -1.4431          | 0.7188             | 0.9920          | -401.6668      | -302.1487    | -2.4410         | -2.4793       |
+| 0.4312        | 0.63  | 300  | 0.4311          | -0.7178        | -2.2532          | 0.7148             | 1.5354          | -482.6770      | -328.8193    | -2.2068         | -2.2801       |
+| 0.4292        | 0.84  | 400  | 0.4276          | -0.5735        | -2.0942          | 0.7031             | 1.5207          | -466.7732      | -314.3910    | -2.1887         | -2.2631       |
 
 
 ### Framework versions
 
 - Transformers 4.35.2
 - Pytorch 2.1.2+cu121
-- Datasets 2.19.1
+- Datasets 2.14.6
 - Tokenizers 0.14.1

all_results.json

@@ -1,8 +1,8 @@
 {
     "epoch": 1.0,
-    "train_loss": 0.390623705157675,
-    "train_runtime": 4433.9669,
-    "train_samples": 61133,
-    "train_samples_per_second": 13.787,
-    "train_steps_per_second": 0.108
+    "train_loss": 0.4418607356658042,
+    "train_runtime": 4329.7925,
+    "train_samples": 61134,
+    "train_samples_per_second": 14.119,
+    "train_steps_per_second": 0.11
 }

model-00001-of-00003.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:1ac7f5951ba58240da40017294a3f524b60736e2052049a80955d47bb73f6dd7
+oid sha256:d15f51d564bae9c9340fcabd0095af395f85c00df443f1ec1391d987c2a50332
 size 4943162336

model-00002-of-00003.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:df1d3db1fc5d58035765a9c4d75823cd49fcd1d84f8706c8ad477d7f7c10c74b
+oid sha256:159889776157ee667d75bce24b683c2ec720ce0c6fe20d483894325edebdeaa9
 size 4999819336

model-00003-of-00003.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:58534372317f409edd57eefc022e856443567520e3ef5a5df9c6638ecbcfe4ef
+oid sha256:dbe15d56222f5ea2c904c1aea147b3b8167bf5f8df49582cf84cbd1760ab2921
 size 4540516344

train_results.json

@@ -1,8 +1,8 @@
 {
     "epoch": 1.0,
-    "train_loss": 0.390623705157675,
-    "train_runtime": 4433.9669,
-    "train_samples": 61133,
-    "train_samples_per_second": 13.787,
-    "train_steps_per_second": 0.108
+    "train_loss": 0.4418607356658042,
+    "train_runtime": 4329.7925,
+    "train_samples": 61134,
+    "train_samples_per_second": 14.119,
+    "train_steps_per_second": 0.11
 }

trainer_state.json

@@ -11,11 +11,11 @@
     {
       "epoch": 0.0,
       "learning_rate": 1.0416666666666666e-08,
-      "logits/chosen": -2.7338109016418457,
-      "logits/rejected": -2.710578203201294,
-      "logps/chosen": -528.6826782226562,
-      "logps/rejected": -158.67160034179688,
-      "loss": 0.6931,
+      "logits/chosen": -2.847970962524414,
+      "logits/rejected": -2.79160213470459,
+      "logps/chosen": -284.9612731933594,
+      "logps/rejected": -276.45928955078125,
+      "loss": 0.5,
       "rewards/accuracies": 0.0,
       "rewards/chosen": 0.0,
       "rewards/margins": 0.0,
@@ -25,733 +25,733 @@
     {
       "epoch": 0.02,
       "learning_rate": 1.0416666666666667e-07,
-      "logits/chosen": -2.7741899490356445,
-      "logits/rejected": -2.7445850372314453,
-      "logps/chosen": -372.50982666015625,
-      "logps/rejected": -129.0567169189453,
-      "loss": 0.6928,
-      "rewards/accuracies": 0.5486111044883728,
-      "rewards/chosen": 0.0008477116352878511,
-      "rewards/margins": 0.0011047045700252056,
-      "rewards/rejected": -0.00025699284742586315,
+      "logits/chosen": -2.7547006607055664,
+      "logits/rejected": -2.752781391143799,
+      "logps/chosen": -249.9474334716797,
+      "logps/rejected": -223.04925537109375,
+      "loss": 0.5,
+      "rewards/accuracies": 0.4166666567325592,
+      "rewards/chosen": 3.489381015242543e-06,
+      "rewards/margins": 1.664027695369441e-05,
+      "rewards/rejected": -1.3150915037840605e-05,
       "step": 10
     },
     {
       "epoch": 0.04,
       "learning_rate": 2.0833333333333333e-07,
-      "logits/chosen": -2.7527709007263184,
-      "logits/rejected": -2.729275703430176,
-      "logps/chosen": -367.230712890625,
-      "logps/rejected": -137.4402313232422,
-      "loss": 0.683,
-      "rewards/accuracies": 0.762499988079071,
-      "rewards/chosen": 0.010893024504184723,
-      "rewards/margins": 0.02107730321586132,
-      "rewards/rejected": -0.010184278711676598,
+      "logits/chosen": -2.7449228763580322,
+      "logits/rejected": -2.7454206943511963,
+      "logps/chosen": -257.4685974121094,
+      "logps/rejected": -247.5232391357422,
+      "loss": 0.4998,
+      "rewards/accuracies": 0.5062500238418579,
+      "rewards/chosen": -0.00013088509149383754,
+      "rewards/margins": 0.0007947316626086831,
+      "rewards/rejected": -0.000925616710446775,
       "step": 20
     },
     {
       "epoch": 0.06,
       "learning_rate": 3.1249999999999997e-07,
-      "logits/chosen": -2.6821541786193848,
-      "logits/rejected": -2.6682209968566895,
-      "logps/chosen": -345.0670166015625,
-      "logps/rejected": -137.2961883544922,
-      "loss": 0.6356,
-      "rewards/accuracies": 0.7562500238418579,
-      "rewards/chosen": 0.0399758443236351,
-      "rewards/margins": 0.09227657318115234,
-      "rewards/rejected": -0.05230072885751724,
+      "logits/chosen": -2.803709030151367,
+      "logits/rejected": -2.756469488143921,
+      "logps/chosen": -298.9071350097656,
+      "logps/rejected": -260.29119873046875,
+      "loss": 0.4979,
+      "rewards/accuracies": 0.65625,
+      "rewards/chosen": 0.01669885963201523,
+      "rewards/margins": 0.008692960254848003,
+      "rewards/rejected": 0.008005897514522076,
       "step": 30
     },
     {
       "epoch": 0.08,
       "learning_rate": 4.1666666666666667e-07,
-      "logits/chosen": -2.5740535259246826,
-      "logits/rejected": -2.5633654594421387,
-      "logps/chosen": -314.03118896484375,
-      "logps/rejected": -128.52047729492188,
-      "loss": 0.5788,
-      "rewards/accuracies": 0.737500011920929,
-      "rewards/chosen": 0.0424940325319767,
-      "rewards/margins": 0.26952430605888367,
-      "rewards/rejected": -0.22703024744987488,
+      "logits/chosen": -2.771388292312622,
+      "logits/rejected": -2.7593226432800293,
+      "logps/chosen": -250.7498016357422,
+      "logps/rejected": -268.94195556640625,
+      "loss": 0.4933,
+      "rewards/accuracies": 0.6875,
+      "rewards/chosen": 0.05684065818786621,
+      "rewards/margins": 0.025426805019378662,
+      "rewards/rejected": 0.03141385689377785,
       "step": 40
     },
     {
       "epoch": 0.1,
       "learning_rate": 4.999733114418725e-07,
-      "logits/chosen": -2.5243449211120605,
-      "logits/rejected": -2.512850046157837,
-      "logps/chosen": -384.26727294921875,
-      "logps/rejected": -183.08828735351562,
-      "loss": 0.5176,
-      "rewards/accuracies": 0.831250011920929,
-      "rewards/chosen": -0.009876979514956474,
-      "rewards/margins": 0.5483093857765198,
-      "rewards/rejected": -0.5581863522529602,
+      "logits/chosen": -2.7790515422821045,
+      "logits/rejected": -2.752472400665283,
+      "logps/chosen": -274.9833679199219,
+      "logps/rejected": -246.99960327148438,
+      "loss": 0.4866,
+      "rewards/accuracies": 0.65625,
+      "rewards/chosen": 0.07096005976200104,
+      "rewards/margins": 0.06392381340265274,
+      "rewards/rejected": 0.007036241702735424,
       "step": 50
     },
     {
       "epoch": 0.13,
       "learning_rate": 4.990398100856366e-07,
-      "logits/chosen": -2.452897548675537,
-      "logits/rejected": -2.436417818069458,
-      "logps/chosen": -404.2002868652344,
-      "logps/rejected": -219.8968505859375,
-      "loss": 0.4859,
-      "rewards/accuracies": 0.824999988079071,
-      "rewards/chosen": -0.01602819748222828,
-      "rewards/margins": 0.8050289154052734,
-      "rewards/rejected": -0.8210570216178894,
+      "logits/chosen": -2.7575879096984863,
+      "logits/rejected": -2.7460415363311768,
+      "logps/chosen": -271.18243408203125,
+      "logps/rejected": -244.96578979492188,
+      "loss": 0.48,
+      "rewards/accuracies": 0.668749988079071,
+      "rewards/chosen": 0.04702044278383255,
+      "rewards/margins": 0.09620000422000885,
+      "rewards/rejected": -0.0491795614361763,
       "step": 60
     },
     {
       "epoch": 0.15,
       "learning_rate": 4.967775735898179e-07,
-      "logits/chosen": -2.3969461917877197,
-      "logits/rejected": -2.367985248565674,
-      "logps/chosen": -372.53216552734375,
-      "logps/rejected": -244.3639373779297,
-      "loss": 0.4452,
-      "rewards/accuracies": 0.768750011920929,
-      "rewards/chosen": -0.21240440011024475,
-      "rewards/margins": 0.8786581754684448,
-      "rewards/rejected": -1.0910624265670776,
+      "logits/chosen": -2.807112216949463,
+      "logits/rejected": -2.7648634910583496,
+      "logps/chosen": -260.88311767578125,
+      "logps/rejected": -237.0400390625,
+      "loss": 0.4761,
+      "rewards/accuracies": 0.65625,
+      "rewards/chosen": 0.06252235174179077,
+      "rewards/margins": 0.11351621150970459,
+      "rewards/rejected": -0.050993870943784714,
       "step": 70
     },
     {
       "epoch": 0.17,
       "learning_rate": 4.931986719649298e-07,
-      "logits/chosen": -2.3214633464813232,
-      "logits/rejected": -2.2642316818237305,
-      "logps/chosen": -382.7598571777344,
-      "logps/rejected": -267.8367004394531,
-      "loss": 0.4444,
-      "rewards/accuracies": 0.75,
-      "rewards/chosen": -0.5137729048728943,
-      "rewards/margins": 0.947833240032196,
-      "rewards/rejected": -1.4616062641143799,
+      "logits/chosen": -2.808347225189209,
+      "logits/rejected": -2.7839419841766357,
+      "logps/chosen": -251.40426635742188,
+      "logps/rejected": -277.35076904296875,
+      "loss": 0.4692,
+      "rewards/accuracies": 0.643750011920929,
+      "rewards/chosen": 0.08948676288127899,
+      "rewards/margins": 0.14001944661140442,
+      "rewards/rejected": -0.05053266882896423,
       "step": 80
     },
     {
       "epoch": 0.19,
       "learning_rate": 4.883222001996351e-07,
-      "logits/chosen": -2.374678134918213,
-      "logits/rejected": -2.3153045177459717,
-      "logps/chosen": -416.29443359375,
-      "logps/rejected": -304.565673828125,
-      "loss": 0.4029,
-      "rewards/accuracies": 0.7749999761581421,
-      "rewards/chosen": -0.4262094497680664,
-      "rewards/margins": 1.292773723602295,
-      "rewards/rejected": -1.7189830541610718,
+      "logits/chosen": -2.8350749015808105,
+      "logits/rejected": -2.813979387283325,
+      "logps/chosen": -263.50994873046875,
+      "logps/rejected": -260.364501953125,
+      "loss": 0.4675,
+      "rewards/accuracies": 0.7562500238418579,
+      "rewards/chosen": 0.11856864392757416,
+      "rewards/margins": 0.21022722125053406,
+      "rewards/rejected": -0.0916585922241211,
       "step": 90
     },
     {
       "epoch": 0.21,
       "learning_rate": 4.821741763807186e-07,
-      "logits/chosen": -2.3450849056243896,
-      "logits/rejected": -2.2891693115234375,
-      "logps/chosen": -364.70013427734375,
-      "logps/rejected": -251.80361938476562,
-      "loss": 0.4319,
-      "rewards/accuracies": 0.8125,
-      "rewards/chosen": -0.2382417917251587,
-      "rewards/margins": 1.069990873336792,
-      "rewards/rejected": -1.3082325458526611,
+      "logits/chosen": -2.825021266937256,
+      "logits/rejected": -2.794950485229492,
+      "logps/chosen": -281.08648681640625,
+      "logps/rejected": -276.0872497558594,
+      "loss": 0.4633,
+      "rewards/accuracies": 0.6937500238418579,
+      "rewards/chosen": 0.07878530770540237,
+      "rewards/margins": 0.21750935912132263,
+      "rewards/rejected": -0.13872405886650085,
       "step": 100
     },
     {
       "epoch": 0.21,
-      "eval_logits/chosen": -2.3001105785369873,
-      "eval_logits/rejected": -2.2766027450561523,
-      "eval_logps/chosen": -400.3859558105469,
-      "eval_logps/rejected": -434.82366943359375,
-      "eval_loss": 0.6498657464981079,
-      "eval_rewards/accuracies": 0.6328125,
-      "eval_rewards/chosen": -1.433463454246521,
-      "eval_rewards/margins": 0.3412409722805023,
-      "eval_rewards/rejected": -1.7747045755386353,
-      "eval_runtime": 54.2454,
-      "eval_samples_per_second": 36.869,
-      "eval_steps_per_second": 0.59,
+      "eval_logits/chosen": -2.7996602058410645,
+      "eval_logits/rejected": -2.784224033355713,
+      "eval_logps/chosen": -248.5258026123047,
+      "eval_logps/rejected": -272.21978759765625,
+      "eval_loss": 0.46518972516059875,
+      "eval_rewards/accuracies": 0.703125,
+      "eval_rewards/chosen": 0.0851384699344635,
+      "eval_rewards/margins": 0.23380354046821594,
+      "eval_rewards/rejected": -0.14866507053375244,
+      "eval_runtime": 53.2131,
+      "eval_samples_per_second": 37.585,
+      "eval_steps_per_second": 0.601,
       "step": 100
     },
     {
       "epoch": 0.23,
       "learning_rate": 4.747874028753375e-07,
-      "logits/chosen": -2.233649730682373,
-      "logits/rejected": -2.1401028633117676,
-      "logps/chosen": -503.4501953125,
-      "logps/rejected": -361.48541259765625,
-      "loss": 0.4161,
-      "rewards/accuracies": 0.800000011920929,
-      "rewards/chosen": -0.7022782564163208,
-      "rewards/margins": 1.533695936203003,
-      "rewards/rejected": -2.2359743118286133,
+      "logits/chosen": -2.7745070457458496,
+      "logits/rejected": -2.7703776359558105,
+      "logps/chosen": -253.75216674804688,
+      "logps/rejected": -293.00921630859375,
+      "loss": 0.4569,
+      "rewards/accuracies": 0.75,
+      "rewards/chosen": 0.10511144250631332,
+      "rewards/margins": 0.30809086561203003,
+      "rewards/rejected": -0.2029794454574585,
       "step": 110
     },
     {
       "epoch": 0.25,
       "learning_rate": 4.662012913161997e-07,
-      "logits/chosen": -2.273224353790283,
-      "logits/rejected": -2.2325010299682617,
-      "logps/chosen": -407.72186279296875,
-      "logps/rejected": -321.96258544921875,
-      "loss": 0.3851,
-      "rewards/accuracies": 0.8187500238418579,
-      "rewards/chosen": -0.5440581440925598,
-      "rewards/margins": 1.4105150699615479,
-      "rewards/rejected": -1.9545730352401733,
+      "logits/chosen": -2.787919521331787,
+      "logits/rejected": -2.75307559967041,
+      "logps/chosen": -267.96954345703125,
+      "logps/rejected": -277.1288757324219,
+      "loss": 0.458,
+      "rewards/accuracies": 0.699999988079071,
+      "rewards/chosen": -0.004297289066016674,
+      "rewards/margins": 0.2931981682777405,
+      "rewards/rejected": -0.29749545454978943,
       "step": 120
     },
     {
       "epoch": 0.27,
       "learning_rate": 4.5646165232345103e-07,
-      "logits/chosen": -2.2552895545959473,
-      "logits/rejected": -2.164522647857666,
-      "logps/chosen": -436.8758239746094,
-      "logps/rejected": -313.2684631347656,
-      "loss": 0.3915,
-      "rewards/accuracies": 0.800000011920929,
-      "rewards/chosen": -0.6610016822814941,
-      "rewards/margins": 1.3072267770767212,
-      "rewards/rejected": -1.9682285785675049,
+      "logits/chosen": -2.7546322345733643,
+      "logits/rejected": -2.722233533859253,
+      "logps/chosen": -298.22113037109375,
+      "logps/rejected": -277.7261962890625,
+      "loss": 0.4523,
+      "rewards/accuracies": 0.731249988079071,
+      "rewards/chosen": -0.06089673191308975,
+      "rewards/margins": 0.31393277645111084,
+      "rewards/rejected": -0.37482950091362,
       "step": 130
     },
     {
       "epoch": 0.29,
       "learning_rate": 4.456204510851956e-07,
-      "logits/chosen": -2.194746494293213,
-      "logits/rejected": -2.107067108154297,
-      "logps/chosen": -472.96380615234375,
-      "logps/rejected": -399.2961730957031,
-      "loss": 0.3906,
-      "rewards/accuracies": 0.831250011920929,
-      "rewards/chosen": -0.8916561007499695,
-      "rewards/margins": 1.7341525554656982,
-      "rewards/rejected": -2.6258087158203125,
+      "logits/chosen": -2.6280975341796875,
+      "logits/rejected": -2.6091978549957275,
+      "logps/chosen": -270.306884765625,
+      "logps/rejected": -291.79815673828125,
+      "loss": 0.4476,
+      "rewards/accuracies": 0.643750011920929,
+      "rewards/chosen": -0.21947817504405975,
+      "rewards/margins": 0.5008858442306519,
+      "rewards/rejected": -0.720363974571228,
       "step": 140
     },
     {
       "epoch": 0.31,
       "learning_rate": 4.337355301007335e-07,
-      "logits/chosen": -2.1862947940826416,
-      "logits/rejected": -2.0885581970214844,
-      "logps/chosen": -470.4183654785156,
-      "logps/rejected": -381.9734191894531,
-      "loss": 0.377,
-      "rewards/accuracies": 0.8125,
-      "rewards/chosen": -0.8573510050773621,
-      "rewards/margins": 1.6041991710662842,
-      "rewards/rejected": -2.461550235748291,
+      "logits/chosen": -2.6492972373962402,
+      "logits/rejected": -2.63801908493042,
+      "logps/chosen": -279.10015869140625,
+      "logps/rejected": -306.9041442871094,
+      "loss": 0.4449,
+      "rewards/accuracies": 0.731249988079071,
+      "rewards/chosen": -0.13866892457008362,
+      "rewards/margins": 0.5307237505912781,
+      "rewards/rejected": -0.6693927049636841,
       "step": 150
     },
     {
       "epoch": 0.33,
       "learning_rate": 4.2087030056579986e-07,
-      "logits/chosen": -2.149928092956543,
-      "logits/rejected": -2.0291361808776855,
-      "logps/chosen": -473.9505310058594,
-      "logps/rejected": -384.2547607421875,
-      "loss": 0.3711,
-      "rewards/accuracies": 0.8125,
-      "rewards/chosen": -0.9108888506889343,
-      "rewards/margins": 1.6116889715194702,
-      "rewards/rejected": -2.5225777626037598,
+      "logits/chosen": -2.618921995162964,
+      "logits/rejected": -2.594566822052002,
+      "logps/chosen": -299.5556335449219,
+      "logps/rejected": -344.01458740234375,
+      "loss": 0.4427,
+      "rewards/accuracies": 0.6625000238418579,
+      "rewards/chosen": -0.38575804233551025,
+      "rewards/margins": 0.7390523552894592,
+      "rewards/rejected": -1.1248104572296143,
       "step": 160
     },
     {
       "epoch": 0.36,
       "learning_rate": 4.070934040463998e-07,
-      "logits/chosen": -2.1031107902526855,
-      "logits/rejected": -1.9537290334701538,
-      "logps/chosen": -552.788330078125,
-      "logps/rejected": -459.04718017578125,
-      "loss": 0.3672,
-      "rewards/accuracies": 0.831250011920929,
-      "rewards/chosen": -1.2671574354171753,
-      "rewards/margins": 2.013477087020874,
-      "rewards/rejected": -3.2806344032287598,
+      "logits/chosen": -2.5482430458068848,
+      "logits/rejected": -2.517625331878662,
+      "logps/chosen": -306.1454162597656,
+      "logps/rejected": -347.05078125,
+      "loss": 0.4363,
+      "rewards/accuracies": 0.6812499761581421,
+      "rewards/chosen": -0.3748893141746521,
+      "rewards/margins": 0.5893394351005554,
+      "rewards/rejected": -0.9642287492752075,
       "step": 170
     },
     {
       "epoch": 0.38,
       "learning_rate": 3.9247834624635404e-07,
-      "logits/chosen": -2.116580009460449,
-      "logits/rejected": -2.005300283432007,
-      "logps/chosen": -485.86346435546875,
-      "logps/rejected": -424.8414611816406,
-      "loss": 0.3752,
-      "rewards/accuracies": 0.8062499761581421,
-      "rewards/chosen": -1.1004022359848022,
-      "rewards/margins": 1.7027782201766968,
-      "rewards/rejected": -2.803180456161499,
+      "logits/chosen": -2.5718603134155273,
+      "logits/rejected": -2.5104472637176514,
+      "logps/chosen": -316.1801452636719,
+      "logps/rejected": -380.98211669921875,
+      "loss": 0.4341,
+      "rewards/accuracies": 0.71875,
+      "rewards/chosen": -0.39730843901634216,
+      "rewards/margins": 0.9330652356147766,
+      "rewards/rejected": -1.3303735256195068,
       "step": 180
     },
     {
       "epoch": 0.4,
       "learning_rate": 3.7710310482256523e-07,
-      "logits/chosen": -2.0973212718963623,
-      "logits/rejected": -1.9831724166870117,
-      "logps/chosen": -553.8712158203125,
-      "logps/rejected": -454.5481872558594,
-      "loss": 0.3748,
-      "rewards/accuracies": 0.8062499761581421,
-      "rewards/chosen": -1.1362320184707642,
-      "rewards/margins": 2.0223002433776855,
-      "rewards/rejected": -3.1585323810577393,
+      "logits/chosen": -2.409097909927368,
+      "logits/rejected": -2.360410690307617,
+      "logps/chosen": -367.5240783691406,
+      "logps/rejected": -451.7735290527344,
+      "loss": 0.4347,
+      "rewards/accuracies": 0.7250000238418579,
+      "rewards/chosen": -1.103116750717163,
+      "rewards/margins": 1.0464074611663818,
+      "rewards/rejected": -2.149524211883545,
       "step": 190
     },
     {
       "epoch": 0.42,
       "learning_rate": 3.610497133404795e-07,
-      "logits/chosen": -2.0881752967834473,
-      "logits/rejected": -1.980513572692871,
-      "logps/chosen": -542.2076416015625,
-      "logps/rejected": -463.62890625,
-      "loss": 0.3523,
-      "rewards/accuracies": 0.8187500238418579,
-      "rewards/chosen": -1.2750364542007446,
-      "rewards/margins": 2.0321054458618164,
-      "rewards/rejected": -3.3071417808532715,
+      "logits/chosen": -2.536165714263916,
+      "logits/rejected": -2.4360404014587402,
+      "logps/chosen": -383.1653137207031,
+      "logps/rejected": -465.27032470703125,
+      "loss": 0.4346,
+      "rewards/accuracies": 0.7562500238418579,
+      "rewards/chosen": -0.4858878254890442,
+      "rewards/margins": 1.3677942752838135,
+      "rewards/rejected": -1.8536821603775024,
       "step": 200
     },
     {
       "epoch": 0.42,
-      "eval_logits/chosen": -2.0889179706573486,
-      "eval_logits/rejected": -2.0558736324310303,
-      "eval_logps/chosen": -459.19970703125,
-      "eval_logps/rejected": -531.1033935546875,
-      "eval_loss": 0.5953767895698547,
+      "eval_logits/chosen": -2.4792861938476562,
+      "eval_logits/rejected": -2.441007137298584,
+      "eval_logps/chosen": -302.1487121582031,
+      "eval_logps/rejected": -401.66680908203125,
+      "eval_loss": 0.4391615688800812,
       "eval_rewards/accuracies": 0.71875,
-      "eval_rewards/chosen": -2.0216004848480225,
-      "eval_rewards/margins": 0.715901255607605,
-      "eval_rewards/rejected": -2.737501621246338,
-      "eval_runtime": 53.4074,
-      "eval_samples_per_second": 37.448,
-      "eval_steps_per_second": 0.599,
+      "eval_rewards/chosen": -0.4510906934738159,
+      "eval_rewards/margins": 0.9920450448989868,
+      "eval_rewards/rejected": -1.4431358575820923,
+      "eval_runtime": 53.2202,
+      "eval_samples_per_second": 37.58,
+      "eval_steps_per_second": 0.601,
       "step": 200
     },
     {
       "epoch": 0.44,
       "learning_rate": 3.4440382358952115e-07,
-      "logits/chosen": -2.1334011554718018,
-      "logits/rejected": -2.029670476913452,
-      "logps/chosen": -484.1700744628906,
-      "logps/rejected": -433.7367248535156,
-      "loss": 0.3484,
-      "rewards/accuracies": 0.831250011920929,
-      "rewards/chosen": -1.0799142122268677,
-      "rewards/margins": 1.8150622844696045,
-      "rewards/rejected": -2.894976854324341,
+      "logits/chosen": -2.425490617752075,
+      "logits/rejected": -2.3446462154388428,
+      "logps/chosen": -332.01141357421875,
+      "logps/rejected": -393.47320556640625,
+      "loss": 0.4404,
+      "rewards/accuracies": 0.7124999761581421,
+      "rewards/chosen": -0.648047685623169,
+      "rewards/margins": 0.8823984265327454,
+      "rewards/rejected": -1.5304460525512695,
       "step": 210
     },
     {
       "epoch": 0.46,
       "learning_rate": 3.272542485937368e-07,
-      "logits/chosen": -2.18733549118042,
-      "logits/rejected": -2.02341890335083,
-      "logps/chosen": -521.8484497070312,
-      "logps/rejected": -446.3534240722656,
-      "loss": 0.3472,
-      "rewards/accuracies": 0.8812500238418579,
-      "rewards/chosen": -0.9639819264411926,
-      "rewards/margins": 2.202716827392578,
-      "rewards/rejected": -3.166698932647705,
+      "logits/chosen": -2.3783254623413086,
+      "logits/rejected": -2.3185007572174072,
+      "logps/chosen": -336.74871826171875,
+      "logps/rejected": -391.41021728515625,
+      "loss": 0.4377,
+      "rewards/accuracies": 0.71875,
+      "rewards/chosen": -0.42188310623168945,
+      "rewards/margins": 1.0818744897842407,
+      "rewards/rejected": -1.5037577152252197,
       "step": 220
     },
     {
       "epoch": 0.48,
       "learning_rate": 3.096924887558854e-07,
-      "logits/chosen": -2.1084365844726562,
-      "logits/rejected": -2.022747278213501,
-      "logps/chosen": -504.220947265625,
-      "logps/rejected": -441.27667236328125,
-      "loss": 0.3541,
-      "rewards/accuracies": 0.8187500238418579,
-      "rewards/chosen": -0.9997168779373169,
-      "rewards/margins": 1.8952099084854126,
-      "rewards/rejected": -2.8949267864227295,
+      "logits/chosen": -2.3234615325927734,
+      "logits/rejected": -2.269005298614502,
+      "logps/chosen": -353.1332092285156,
+      "logps/rejected": -496.53936767578125,
+      "loss": 0.4305,
+      "rewards/accuracies": 0.71875,
+      "rewards/chosen": -0.7132310271263123,
+      "rewards/margins": 1.3435031175613403,
+      "rewards/rejected": -2.056734085083008,
       "step": 230
     },
     {
       "epoch": 0.5,
       "learning_rate": 2.9181224366319943e-07,
-      "logits/chosen": -2.111999034881592,
-      "logits/rejected": -2.023984432220459,
-      "logps/chosen": -526.4149169921875,
-      "logps/rejected": -442.402099609375,
-      "loss": 0.3521,
-      "rewards/accuracies": 0.824999988079071,
-      "rewards/chosen": -1.2106670141220093,
-      "rewards/margins": 1.8976614475250244,
-      "rewards/rejected": -3.108328342437744,
+      "logits/chosen": -2.3222591876983643,
+      "logits/rejected": -2.2519450187683105,
+      "logps/chosen": -344.47113037109375,
+      "logps/rejected": -446.96856689453125,
+      "loss": 0.4332,
+      "rewards/accuracies": 0.699999988079071,
+      "rewards/chosen": -0.9076830744743347,
+      "rewards/margins": 0.976728618144989,
+      "rewards/rejected": -1.8844118118286133,
       "step": 240
     },
     {
       "epoch": 0.52,
       "learning_rate": 2.7370891215954565e-07,
-      "logits/chosen": -2.089487075805664,
-      "logits/rejected": -2.001037120819092,
-      "logps/chosen": -484.54559326171875,
-      "logps/rejected": -447.81463623046875,
-      "loss": 0.3631,
-      "rewards/accuracies": 0.831250011920929,
-      "rewards/chosen": -1.2301665544509888,
-      "rewards/margins": 1.9752994775772095,
-      "rewards/rejected": -3.2054660320281982,
+      "logits/chosen": -2.2862257957458496,
+      "logits/rejected": -2.1678268909454346,
+      "logps/chosen": -376.39666748046875,
+      "logps/rejected": -490.06866455078125,
+      "loss": 0.432,
+      "rewards/accuracies": 0.7250000238418579,
+      "rewards/chosen": -1.1747140884399414,
+      "rewards/margins": 1.37172532081604,
+      "rewards/rejected": -2.5464394092559814,
       "step": 250
     },
     {
       "epoch": 0.54,
       "learning_rate": 2.55479083351317e-07,
-      "logits/chosen": -2.2092173099517822,
-      "logits/rejected": -2.1207237243652344,
-      "logps/chosen": -484.1170959472656,
-      "logps/rejected": -414.44134521484375,
-      "loss": 0.3554,
-      "rewards/accuracies": 0.8187500238418579,
-      "rewards/chosen": -1.0310567617416382,
-      "rewards/margins": 1.7033233642578125,
-      "rewards/rejected": -2.7343802452087402,
+      "logits/chosen": -2.3139841556549072,
+      "logits/rejected": -2.2681527137756348,
+      "logps/chosen": -337.32501220703125,
+      "logps/rejected": -411.294921875,
+      "loss": 0.4362,
+      "rewards/accuracies": 0.6499999761581421,
+      "rewards/chosen": -0.6871622800827026,
+      "rewards/margins": 0.9410949945449829,
+      "rewards/rejected": -1.628257155418396,
       "step": 260
     },
     {
       "epoch": 0.56,
       "learning_rate": 2.3722002126275822e-07,
-      "logits/chosen": -2.128089427947998,
-      "logits/rejected": -2.0141799449920654,
-      "logps/chosen": -475.9017028808594,
-      "logps/rejected": -440.27142333984375,
-      "loss": 0.3495,
-      "rewards/accuracies": 0.8500000238418579,
-      "rewards/chosen": -1.2480560541152954,
-      "rewards/margins": 2.013627529144287,
-      "rewards/rejected": -3.261683940887451,
+      "logits/chosen": -2.2898974418640137,
+      "logits/rejected": -2.2423269748687744,
+      "logps/chosen": -349.0024108886719,
+      "logps/rejected": -429.98712158203125,
+      "loss": 0.4249,
+      "rewards/accuracies": 0.7437499761581421,
+      "rewards/chosen": -0.7414897084236145,
+      "rewards/margins": 1.0027714967727661,
+      "rewards/rejected": -1.7442611455917358,
       "step": 270
     },
     {
       "epoch": 0.59,
       "learning_rate": 2.19029145890313e-07,
-      "logits/chosen": -2.1510426998138428,
-      "logits/rejected": -2.0308613777160645,
-      "logps/chosen": -480.66485595703125,
-      "logps/rejected": -439.12530517578125,
-      "loss": 0.3473,
-      "rewards/accuracies": 0.8374999761581421,
-      "rewards/chosen": -1.2660208940505981,
-      "rewards/margins": 1.8633956909179688,
-      "rewards/rejected": -3.1294167041778564,
+      "logits/chosen": -2.261748790740967,
+      "logits/rejected": -2.18464732170105,
+      "logps/chosen": -326.8984680175781,
+      "logps/rejected": -411.0516662597656,
+      "loss": 0.4357,
+      "rewards/accuracies": 0.7124999761581421,
+      "rewards/chosen": -0.8549286127090454,
+      "rewards/margins": 0.9535172581672668,
+      "rewards/rejected": -1.808445930480957,
       "step": 280
     },
     {
       "epoch": 0.61,
       "learning_rate": 2.0100351342479216e-07,
-      "logits/chosen": -2.1496615409851074,
-      "logits/rejected": -2.032989025115967,
-      "logps/chosen": -499.2440490722656,
-      "logps/rejected": -448.85107421875,
-      "loss": 0.3588,
-      "rewards/accuracies": 0.8374999761581421,
-      "rewards/chosen": -0.9003239870071411,
-      "rewards/margins": 2.171739101409912,
-      "rewards/rejected": -3.0720629692077637,
+      "logits/chosen": -2.2318782806396484,
+      "logits/rejected": -2.1214816570281982,
+      "logps/chosen": -343.24822998046875,
+      "logps/rejected": -468.28515625,
+      "loss": 0.4302,
+      "rewards/accuracies": 0.7124999761581421,
+      "rewards/chosen": -0.9663702845573425,
+      "rewards/margins": 1.3740543127059937,
+      "rewards/rejected": -2.3404245376586914,
       "step": 290
     },
     {
       "epoch": 0.63,
       "learning_rate": 1.8323929841460178e-07,
-      "logits/chosen": -2.158599376678467,
-      "logits/rejected": -2.0707848072052,
-      "logps/chosen": -469.2913513183594,
-      "logps/rejected": -447.3359375,
-      "loss": 0.3611,
-      "rewards/accuracies": 0.824999988079071,
-      "rewards/chosen": -1.2566159963607788,
-      "rewards/margins": 1.9593569040298462,
-      "rewards/rejected": -3.215972900390625,
+      "logits/chosen": -2.2208759784698486,
+      "logits/rejected": -2.1360106468200684,
+      "logps/chosen": -395.91851806640625,
+      "logps/rejected": -485.6800842285156,
+      "loss": 0.4312,
+      "rewards/accuracies": 0.6312500238418579,
+      "rewards/chosen": -1.3964992761611938,
+      "rewards/margins": 0.8867896795272827,
+      "rewards/rejected": -2.2832889556884766,
       "step": 300
     },
     {
       "epoch": 0.63,
-      "eval_logits/chosen": -2.1299195289611816,
-      "eval_logits/rejected": -2.0872669219970703,
-      "eval_logps/chosen": -456.173828125,
-      "eval_logps/rejected": -541.2481689453125,
-      "eval_loss": 0.5743047595024109,
-      "eval_rewards/accuracies": 0.71875,
-      "eval_rewards/chosen": -1.9913413524627686,
-      "eval_rewards/margins": 0.847608208656311,
-      "eval_rewards/rejected": -2.838949680328369,
-      "eval_runtime": 53.3455,
-      "eval_samples_per_second": 37.491,
-      "eval_steps_per_second": 0.6,
+      "eval_logits/chosen": -2.2800676822662354,
+      "eval_logits/rejected": -2.2067744731903076,
+      "eval_logps/chosen": -328.81927490234375,
+      "eval_logps/rejected": -482.677001953125,
+      "eval_loss": 0.43114951252937317,
+      "eval_rewards/accuracies": 0.71484375,
+      "eval_rewards/chosen": -0.7177965044975281,
+      "eval_rewards/margins": 1.5354408025741577,
+      "eval_rewards/rejected": -2.253237009048462,
+      "eval_runtime": 53.2401,
+      "eval_samples_per_second": 37.566,
+      "eval_steps_per_second": 0.601,
       "step": 300
     },
     {
       "epoch": 0.65,
       "learning_rate": 1.6583128063291573e-07,
-      "logits/chosen": -2.1583058834075928,
-      "logits/rejected": -2.0590124130249023,
-      "logps/chosen": -517.3380737304688,
-      "logps/rejected": -457.45928955078125,
-      "loss": 0.3338,
-      "rewards/accuracies": 0.8687499761581421,
-      "rewards/chosen": -1.147152066230774,
-      "rewards/margins": 2.0138449668884277,
-      "rewards/rejected": -3.160996913909912,
+      "logits/chosen": -2.1827075481414795,
+      "logits/rejected": -2.0488195419311523,
+      "logps/chosen": -345.99652099609375,
+      "logps/rejected": -477.7587890625,
+      "loss": 0.4233,
+      "rewards/accuracies": 0.768750011920929,
+      "rewards/chosen": -0.7281765937805176,
+      "rewards/margins": 1.5777080059051514,
+      "rewards/rejected": -2.305884599685669,
       "step": 310
     },
     {
       "epoch": 0.67,
       "learning_rate": 1.488723393865766e-07,
-      "logits/chosen": -2.0327324867248535,
-      "logits/rejected": -1.9210631847381592,
-      "logps/chosen": -525.4856567382812,
-      "logps/rejected": -502.88934326171875,
-      "loss": 0.3473,
-      "rewards/accuracies": 0.824999988079071,
-      "rewards/chosen": -1.5122380256652832,
-      "rewards/margins": 2.099067449569702,
-      "rewards/rejected": -3.6113052368164062,
+      "logits/chosen": -2.2175135612487793,
+      "logits/rejected": -2.1210460662841797,
+      "logps/chosen": -377.0932312011719,
+      "logps/rejected": -452.9117126464844,
+      "loss": 0.4274,
+      "rewards/accuracies": 0.731249988079071,
+      "rewards/chosen": -0.807108998298645,
+      "rewards/margins": 1.3679430484771729,
+      "rewards/rejected": -2.1750521659851074,
       "step": 320
     },
     {
       "epoch": 0.69,
       "learning_rate": 1.3245295796480788e-07,
-      "logits/chosen": -2.1876847743988037,
-      "logits/rejected": -2.0464909076690674,
-      "logps/chosen": -545.4427490234375,
-      "logps/rejected": -492.4483947753906,
-      "loss": 0.3278,
-      "rewards/accuracies": 0.84375,
-      "rewards/chosen": -1.2790089845657349,
-      "rewards/margins": 2.3831381797790527,
-      "rewards/rejected": -3.662147045135498,
+      "logits/chosen": -2.2537410259246826,
+      "logits/rejected": -2.1571004390716553,
+      "logps/chosen": -308.38360595703125,
+      "logps/rejected": -401.64312744140625,
+      "loss": 0.4228,
+      "rewards/accuracies": 0.75,
+      "rewards/chosen": -0.5617040395736694,
+      "rewards/margins": 1.1184189319610596,
+      "rewards/rejected": -1.680122971534729,
       "step": 330
     },
     {
       "epoch": 0.71,
       "learning_rate": 1.1666074087171627e-07,
-      "logits/chosen": -2.108828544616699,
-      "logits/rejected": -1.9729408025741577,
-      "logps/chosen": -533.9615478515625,
-      "logps/rejected": -490.66851806640625,
-      "loss": 0.3636,
-      "rewards/accuracies": 0.8374999761581421,
-      "rewards/chosen": -1.4988614320755005,
-      "rewards/margins": 2.179060459136963,
-      "rewards/rejected": -3.677922010421753,
+      "logits/chosen": -2.2037549018859863,
+      "logits/rejected": -2.134270191192627,
+      "logps/chosen": -321.118896484375,
+      "logps/rejected": -394.3348693847656,
+      "loss": 0.4314,
+      "rewards/accuracies": 0.6875,
+      "rewards/chosen": -0.69875568151474,
+      "rewards/margins": 0.8253641128540039,
+      "rewards/rejected": -1.5241197347640991,
       "step": 340
     },
     {
       "epoch": 0.73,
       "learning_rate": 1.0157994641835734e-07,
-      "logits/chosen": -2.1237919330596924,
-      "logits/rejected": -1.9951225519180298,
-      "logps/chosen": -511.7247619628906,
-      "logps/rejected": -494.952880859375,
-      "loss": 0.3334,
-      "rewards/accuracies": 0.84375,
-      "rewards/chosen": -1.251680612564087,
-      "rewards/margins": 2.2906720638275146,
-      "rewards/rejected": -3.5423526763916016,
+      "logits/chosen": -2.179765224456787,
+      "logits/rejected": -2.1164839267730713,
+      "logps/chosen": -306.08087158203125,
+      "logps/rejected": -439.74481201171875,
+      "loss": 0.4208,
+      "rewards/accuracies": 0.6812499761581421,
+      "rewards/chosen": -0.8013069033622742,
+      "rewards/margins": 1.1891696453094482,
+      "rewards/rejected": -1.9904766082763672,
       "step": 350
     },
     {
       "epoch": 0.75,
       "learning_rate": 8.729103716819111e-08,
-      "logits/chosen": -2.1752729415893555,
-      "logits/rejected": -2.0284740924835205,
-      "logps/chosen": -543.2025146484375,
-      "logps/rejected": -459.12493896484375,
-      "loss": 0.3591,
-      "rewards/accuracies": 0.8500000238418579,
-      "rewards/chosen": -1.0760266780853271,
-      "rewards/margins": 2.3620123863220215,
-      "rewards/rejected": -3.4380390644073486,
+      "logits/chosen": -2.1958014965057373,
+      "logits/rejected": -2.0591156482696533,
+      "logps/chosen": -342.7279357910156,
+      "logps/rejected": -487.24237060546875,
+      "loss": 0.4171,
+      "rewards/accuracies": 0.7562500238418579,
+      "rewards/chosen": -0.754597544670105,
+      "rewards/margins": 1.7244411706924438,
+      "rewards/rejected": -2.4790384769439697,
       "step": 360
     },
     {
       "epoch": 0.77,
       "learning_rate": 7.387025063449081e-08,
-      "logits/chosen": -2.1319737434387207,
-      "logits/rejected": -2.020498275756836,
-      "logps/chosen": -488.7276306152344,
-      "logps/rejected": -446.2596740722656,
-      "loss": 0.3274,
-      "rewards/accuracies": 0.8187500238418579,
-      "rewards/chosen": -1.2242181301116943,
-      "rewards/margins": 2.015012502670288,
-      "rewards/rejected": -3.2392303943634033,
+      "logits/chosen": -2.1464810371398926,
+      "logits/rejected": -2.069282054901123,
+      "logps/chosen": -320.1985778808594,
+      "logps/rejected": -539.7567138671875,
+      "loss": 0.4171,
+      "rewards/accuracies": 0.7437499761581421,
+      "rewards/chosen": -0.6495109796524048,
+      "rewards/margins": 2.205376148223877,
+      "rewards/rejected": -2.8548874855041504,
       "step": 370
     },
     {
       "epoch": 0.79,
       "learning_rate": 6.138919252022435e-08,
-      "logits/chosen": -2.1206445693969727,
-      "logits/rejected": -1.98409104347229,
-      "logps/chosen": -502.968994140625,
-      "logps/rejected": -463.2781677246094,
-      "loss": 0.3248,
-      "rewards/accuracies": 0.8500000238418579,
-      "rewards/chosen": -1.1951261758804321,
-      "rewards/margins": 2.3360695838928223,
-      "rewards/rejected": -3.531195878982544,
+      "logits/chosen": -2.2096333503723145,
+      "logits/rejected": -2.132408380508423,
+      "logps/chosen": -332.7948303222656,
+      "logps/rejected": -451.5855407714844,
+      "loss": 0.4228,
+      "rewards/accuracies": 0.7124999761581421,
+      "rewards/chosen": -0.6696550846099854,
+      "rewards/margins": 1.47793447971344,
+      "rewards/rejected": -2.147589683532715,
       "step": 380
     },
     {
       "epoch": 0.82,
       "learning_rate": 4.991445467064689e-08,
-      "logits/chosen": -2.160815715789795,
-      "logits/rejected": -2.0336337089538574,
-      "logps/chosen": -521.4240112304688,
-      "logps/rejected": -455.9097595214844,
-      "loss": 0.3419,
-      "rewards/accuracies": 0.824999988079071,
-      "rewards/chosen": -1.1749980449676514,
-      "rewards/margins": 2.0022196769714355,
-      "rewards/rejected": -3.177217721939087,
+      "logits/chosen": -2.203737735748291,
+      "logits/rejected": -2.1293728351593018,
+      "logps/chosen": -320.31732177734375,
+      "logps/rejected": -470.7608337402344,
+      "loss": 0.4213,
+      "rewards/accuracies": 0.731249988079071,
+      "rewards/chosen": -0.6311790347099304,
+      "rewards/margins": 1.6567703485488892,
+      "rewards/rejected": -2.287949323654175,
       "step": 390
     },
     {
       "epoch": 0.84,
       "learning_rate": 3.9507259776993954e-08,
-      "logits/chosen": -2.0740323066711426,
-      "logits/rejected": -1.9285573959350586,
-      "logps/chosen": -533.530029296875,
-      "logps/rejected": -476.70166015625,
-      "loss": 0.3321,
-      "rewards/accuracies": 0.8687499761581421,
-      "rewards/chosen": -1.1504842042922974,
-      "rewards/margins": 2.290919780731201,
-      "rewards/rejected": -3.441403865814209,
+      "logits/chosen": -2.2450804710388184,
+      "logits/rejected": -2.1665468215942383,
+      "logps/chosen": -374.5340576171875,
+      "logps/rejected": -481.30609130859375,
+      "loss": 0.4292,
+      "rewards/accuracies": 0.762499988079071,
+      "rewards/chosen": -0.743740439414978,
+      "rewards/margins": 1.3402055501937866,
+      "rewards/rejected": -2.0839457511901855,
       "step": 400
     },
     {
       "epoch": 0.84,
-      "eval_logits/chosen": -2.0871171951293945,
-      "eval_logits/rejected": -2.04443621635437,
-      "eval_logps/chosen": -476.4866027832031,
-      "eval_logps/rejected": -567.706787109375,
-      "eval_loss": 0.5693503618240356,
-      "eval_rewards/accuracies": 0.73828125,
-      "eval_rewards/chosen": -2.194469451904297,
-      "eval_rewards/margins": 0.909065306186676,
-      "eval_rewards/rejected": -3.103534698486328,
-      "eval_runtime": 53.3645,
-      "eval_samples_per_second": 37.478,
-      "eval_steps_per_second": 0.6,
+      "eval_logits/chosen": -2.2630538940429688,
+      "eval_logits/rejected": -2.188692331314087,
+      "eval_logps/chosen": -314.3909912109375,
+      "eval_logps/rejected": -466.773193359375,
+      "eval_loss": 0.4275889992713928,
+      "eval_rewards/accuracies": 0.703125,
+      "eval_rewards/chosen": -0.5735137462615967,
+      "eval_rewards/margins": 1.5206854343414307,
+      "eval_rewards/rejected": -2.0941991806030273,
+      "eval_runtime": 53.2545,
+      "eval_samples_per_second": 37.555,
+      "eval_steps_per_second": 0.601,
       "step": 400
     },
     {
       "epoch": 0.86,
       "learning_rate": 3.022313472693447e-08,
-      "logits/chosen": -2.1464505195617676,
-      "logits/rejected": -1.9985647201538086,
-      "logps/chosen": -569.2177124023438,
-      "logps/rejected": -501.0960388183594,
-      "loss": 0.3418,
-      "rewards/accuracies": 0.8812500238418579,
-      "rewards/chosen": -1.199937343597412,
-      "rewards/margins": 2.462070941925049,
-      "rewards/rejected": -3.662008762359619,
+      "logits/chosen": -2.195916175842285,
+      "logits/rejected": -2.140748977661133,
+      "logps/chosen": -333.9373474121094,
+      "logps/rejected": -417.89898681640625,
+      "loss": 0.4219,
+      "rewards/accuracies": 0.6625000238418579,
+      "rewards/chosen": -0.792479932308197,
+      "rewards/margins": 1.0618550777435303,
+      "rewards/rejected": -1.8543351888656616,
       "step": 410
     },
     {
       "epoch": 0.88,
       "learning_rate": 2.2111614344599684e-08,
-      "logits/chosen": -2.01700758934021,
-      "logits/rejected": -1.9420589208602905,
-      "logps/chosen": -496.57281494140625,
-      "logps/rejected": -473.21502685546875,
-      "loss": 0.3415,
-      "rewards/accuracies": 0.8187500238418579,
-      "rewards/chosen": -1.3657633066177368,
-      "rewards/margins": 2.07015323638916,
-      "rewards/rejected": -3.4359164237976074,
+      "logits/chosen": -2.290001153945923,
+      "logits/rejected": -2.15373158454895,
+      "logps/chosen": -333.3667297363281,
+      "logps/rejected": -482.655029296875,
+      "loss": 0.4199,
+      "rewards/accuracies": 0.78125,
+      "rewards/chosen": -0.5566810369491577,
+      "rewards/margins": 1.7732408046722412,
+      "rewards/rejected": -2.3299221992492676,
       "step": 420
     },
     {
       "epoch": 0.9,
       "learning_rate": 1.521597710086439e-08,
-      "logits/chosen": -2.0969324111938477,
-      "logits/rejected": -1.9792184829711914,
-      "logps/chosen": -480.86492919921875,
-      "logps/rejected": -453.40216064453125,
-      "loss": 0.3243,
-      "rewards/accuracies": 0.8374999761581421,
-      "rewards/chosen": -1.4078073501586914,
-      "rewards/margins": 1.961418867111206,
-      "rewards/rejected": -3.3692264556884766,
+      "logits/chosen": -2.263404130935669,
+      "logits/rejected": -2.170976161956787,
+      "logps/chosen": -358.19171142578125,
+      "logps/rejected": -446.0741271972656,
+      "loss": 0.4195,
+      "rewards/accuracies": 0.7250000238418579,
+      "rewards/chosen": -0.8323851823806763,
+      "rewards/margins": 1.1718313694000244,
+      "rewards/rejected": -2.004216432571411,
       "step": 430
     },
     {
       "epoch": 0.92,
       "learning_rate": 9.57301420397924e-09,
-      "logits/chosen": -2.0999059677124023,
-      "logits/rejected": -1.9967210292816162,
-      "logps/chosen": -514.0860595703125,
-      "logps/rejected": -497.3169860839844,
-      "loss": 0.3191,
-      "rewards/accuracies": 0.8812500238418579,
-      "rewards/chosen": -1.4036227464675903,
-      "rewards/margins": 2.1624085903167725,
-      "rewards/rejected": -3.5660316944122314,
+      "logits/chosen": -2.1995832920074463,
+      "logits/rejected": -2.125598907470703,
+      "logps/chosen": -349.9782409667969,
+      "logps/rejected": -470.9933166503906,
+      "loss": 0.4187,
+      "rewards/accuracies": 0.699999988079071,
+      "rewards/chosen": -0.7171262502670288,
+      "rewards/margins": 1.590043306350708,
+      "rewards/rejected": -2.3071694374084473,
       "step": 440
     },
     {
       "epoch": 0.94,
       "learning_rate": 5.212833302556258e-09,
-      "logits/chosen": -2.1073479652404785,
-      "logits/rejected": -1.98806631565094,
-      "logps/chosen": -508.578125,
-      "logps/rejected": -465.4120178222656,
-      "loss": 0.3281,
-      "rewards/accuracies": 0.862500011920929,
-      "rewards/chosen": -1.237577199935913,
-      "rewards/margins": 2.141489028930664,
-      "rewards/rejected": -3.379065990447998,
+      "logits/chosen": -2.2377769947052,
+      "logits/rejected": -2.1389012336730957,
+      "logps/chosen": -341.77386474609375,
+      "logps/rejected": -469.19329833984375,
+      "loss": 0.4221,
+      "rewards/accuracies": 0.6875,
+      "rewards/chosen": -0.6515626907348633,
+      "rewards/margins": 1.3916916847229004,
+      "rewards/rejected": -2.0432541370391846,
       "step": 450
     },
     {
       "epoch": 0.96,
       "learning_rate": 2.158697848236607e-09,
-      "logits/chosen": -2.0812013149261475,
-      "logits/rejected": -1.983912706375122,
-      "logps/chosen": -520.6010131835938,
-      "logps/rejected": -473.6297302246094,
-      "loss": 0.3436,
-      "rewards/accuracies": 0.8500000238418579,
-      "rewards/chosen": -1.305751085281372,
-      "rewards/margins": 1.9923391342163086,
-      "rewards/rejected": -3.2980899810791016,
+      "logits/chosen": -2.1782374382019043,
+      "logits/rejected": -2.113285541534424,
+      "logps/chosen": -329.5958557128906,
+      "logps/rejected": -473.4051818847656,
+      "loss": 0.4183,
+      "rewards/accuracies": 0.75,
+      "rewards/chosen": -0.6429389715194702,
+      "rewards/margins": 1.62165105342865,
+      "rewards/rejected": -2.264590263366699,
       "step": 460
     },
     {
       "epoch": 0.98,
       "learning_rate": 4.269029751107489e-10,
-      "logits/chosen": -2.1194050312042236,
-      "logits/rejected": -1.9787203073501587,
-      "logps/chosen": -526.4293212890625,
-      "logps/rejected": -478.50537109375,
-      "loss": 0.3464,
-      "rewards/accuracies": 0.893750011920929,
-      "rewards/chosen": -1.139508843421936,
-      "rewards/margins": 2.3456408977508545,
-      "rewards/rejected": -3.485149383544922,
+      "logits/chosen": -2.2372963428497314,
+      "logits/rejected": -2.1125941276550293,
+      "logps/chosen": -343.3780517578125,
+      "logps/rejected": -446.16363525390625,
+      "loss": 0.4191,
+      "rewards/accuracies": 0.737500011920929,
+      "rewards/chosen": -0.5751535892486572,
+      "rewards/margins": 1.529253363609314,
+      "rewards/rejected": -2.1044068336486816,
       "step": 470
     },
     {
       "epoch": 1.0,
       "step": 478,
       "total_flos": 0.0,
-      "train_loss": 0.390623705157675,
-      "train_runtime": 4433.9669,
-      "train_samples_per_second": 13.787,
-      "train_steps_per_second": 0.108
+      "train_loss": 0.4418607356658042,
+      "train_runtime": 4329.7925,
+      "train_samples_per_second": 14.119,
+      "train_steps_per_second": 0.11
     }
   ],
   "logging_steps": 10,