Model save

README.md

@@ -1,13 +1,10 @@
 ---
 library_name: peft
 tags:
-- alignment-handbook
 - trl
 - dpo
 - generated_from_trainer
 base_model: NbAiLab/nb-gpt-j-6B-v2
-datasets:
-- hugodk-sch/aftonposten_title_prefs
 model-index:
 - name: aftonposten-6b-align-scan
   results: []
@@ -18,17 +15,17 @@ should probably proofread and complete it, then remove this comment. -->
 
 # aftonposten-6b-align-scan
 
-This model is a fine-tuned version of [data/ap-gpt-j-6b-sft-qlora-04-08](https://huggingface.co/data/ap-gpt-j-6b-sft-qlora-04-08) on the hugodk-sch/aftonposten_title_prefs dataset.
+This model is a fine-tuned version of [NbAiLab/nb-gpt-j-6B-v2](https://huggingface.co/NbAiLab/nb-gpt-j-6B-v2) on an unknown dataset.
 It achieves the following results on the evaluation set:
-- Loss: 0.6990
-- Rewards/chosen: 0.0098
-- Rewards/rejected: 0.0133
-- Rewards/accuracies: 0.4996
-- Rewards/margins: -0.0035
-- Logps/rejected: -37.5019
-- Logps/chosen: -34.0236
-- Logits/rejected: -2.2393
-- Logits/chosen: -2.2442
+- Loss: 0.5000
+- Rewards/chosen: 0.0003
+- Rewards/rejected: 0.0003
+- Rewards/accuracies: 0.5365
+- Rewards/margins: 0.0000
+- Logps/rejected: -37.4870
+- Logps/chosen: -34.0017
+- Logits/rejected: -2.2396
+- Logits/chosen: -2.2445
 
 ## 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 |
 |:-------------:|:-----:|:----:|:---------------:|:--------------:|:----------------:|:------------------:|:---------------:|:--------------:|:------------:|:---------------:|:-------------:|
-| 0.6951        | 0.26  | 100  | 0.6966          | 0.0084         | 0.0081           | 0.4954             | 0.0003          | -37.5076       | -34.0252     | -2.2389         | -2.2438       |
-| 0.6891        | 0.52  | 200  | 0.6947          | 0.0092         | 0.0046           | 0.4934             | 0.0046          | -37.5115       | -34.0243     | -2.2384         | -2.2432       |
-| 0.693         | 0.78  | 300  | 0.6982          | 0.0156         | 0.0171           | 0.4983             | -0.0015         | -37.4976       | -34.0172     | -2.2388         | -2.2437       |
+| 0.5           | 0.26  | 100  | 0.5000          | 0.0001         | 0.0002           | 0.5137             | -0.0002         | -37.4933       | -34.0268     | -2.2390         | -2.2439       |
+| 0.5           | 0.52  | 200  | 0.5000          | 0.0003         | 0.0002           | 0.5299             | 0.0001          | -37.4969       | -34.0030     | -2.2396         | -2.2444       |
+| 0.4999        | 0.78  | 300  | 0.5000          | 0.0003         | 0.0003           | 0.5365             | 0.0000          | -37.4870       | -34.0017     | -2.2396         | -2.2445       |
 
 
 ### Framework versions

adapter_config.json

@@ -19,13 +19,13 @@
   "rank_pattern": {},
   "revision": null,
   "target_modules": [
-    "up_proj",
     "q_proj",
     "gate_proj",
-    "o_proj",
-    "down_proj",
+    "up_proj",
+    "k_proj",
     "v_proj",
-    "k_proj"
+    "o_proj",
+    "down_proj"
   ],
   "task_type": "CAUSAL_LM",
   "use_rslora": false

adapter_model.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:b30a3614e90564741bd0b06f4b5d211ac9ab0d5d4651e82dec2b571e49a407d7
+oid sha256:3eca000156eaf07e8ac034ca10d58da5ad65e1f240e08fe64a18d57afb6fb4c4
 size 176183216

all_results.json

@@ -1,21 +1,8 @@
 {
     "epoch": 1.0,
-    "eval_logits/chosen": -2.244168758392334,
-    "eval_logits/rejected": -2.239292860031128,
-    "eval_logps/chosen": -34.02364730834961,
-    "eval_logps/rejected": -37.501853942871094,
-    "eval_loss": 0.6989675760269165,
-    "eval_rewards/accuracies": 0.49958473443984985,
-    "eval_rewards/chosen": 0.009816422127187252,
-    "eval_rewards/margins": -0.0034731123596429825,
-    "eval_rewards/rejected": 0.01328953169286251,
-    "eval_runtime": 145.7819,
-    "eval_samples": 343,
-    "eval_samples_per_second": 2.353,
-    "eval_steps_per_second": 0.295,
-    "train_loss": 0.6916276120520257,
-    "train_runtime": 3257.6062,
+    "train_loss": 0.4999783658362054,
+    "train_runtime": 3254.8055,
     "train_samples": 3079,
-    "train_samples_per_second": 0.945,
+    "train_samples_per_second": 0.946,
     "train_steps_per_second": 0.118
 }

train_results.json

@@ -1,8 +1,8 @@
 {
     "epoch": 1.0,
-    "train_loss": 0.6916276120520257,
-    "train_runtime": 3257.6062,
+    "train_loss": 0.4999783658362054,
+    "train_runtime": 3254.8055,
     "train_samples": 3079,
-    "train_samples_per_second": 0.945,
+    "train_samples_per_second": 0.946,
     "train_steps_per_second": 0.118
 }

trainer_state.json

@@ -15,7 +15,7 @@
       "logits/rejected": -1.7377450466156006,
       "logps/chosen": -29.553977966308594,
       "logps/rejected": -42.813133239746094,
-      "loss": 0.6931,
+      "loss": 0.5,
       "rewards/accuracies": 0.0,
       "rewards/chosen": 0.0,
       "rewards/margins": 0.0,
@@ -25,590 +25,590 @@
     {
       "epoch": 0.03,
       "learning_rate": 1.2820512820512818e-07,
-      "logits/chosen": -1.866059422492981,
-      "logits/rejected": -1.870389699935913,
-      "logps/chosen": -36.98250198364258,
-      "logps/rejected": -33.64270782470703,
-      "loss": 0.6804,
-      "rewards/accuracies": 0.4027777910232544,
-      "rewards/chosen": 0.021689780056476593,
-      "rewards/margins": 0.033059459179639816,
-      "rewards/rejected": -0.011369682848453522,
+      "logits/chosen": -1.8662651777267456,
+      "logits/rejected": -1.8705823421478271,
+      "logps/chosen": -37.01448440551758,
+      "logps/rejected": -33.66516876220703,
+      "loss": 0.5,
+      "rewards/accuracies": 0.5,
+      "rewards/chosen": -7.885291415732354e-05,
+      "rewards/margins": 0.0002720804768614471,
+      "rewards/rejected": -0.00035093340557068586,
       "step": 10
     },
     {
       "epoch": 0.05,
       "learning_rate": 2.5641025641025636e-07,
-      "logits/chosen": -1.9973828792572021,
-      "logits/rejected": -2.0000317096710205,
-      "logps/chosen": -29.631702423095703,
-      "logps/rejected": -29.05877113342285,
-      "loss": 0.6937,
-      "rewards/accuracies": 0.512499988079071,
-      "rewards/chosen": 0.009445475414395332,
-      "rewards/margins": 0.005240145605057478,
-      "rewards/rejected": 0.004205327946692705,
+      "logits/chosen": -1.9972978830337524,
+      "logits/rejected": -1.9999454021453857,
+      "logps/chosen": -29.626657485961914,
+      "logps/rejected": -29.065475463867188,
+      "loss": 0.5,
+      "rewards/accuracies": 0.550000011920929,
+      "rewards/chosen": 0.00015535543207079172,
+      "rewards/margins": 0.00017565980670042336,
+      "rewards/rejected": -2.0304325516917743e-05,
       "step": 20
     },
     {
       "epoch": 0.08,
       "learning_rate": 3.8461538461538463e-07,
-      "logits/chosen": -1.9198739528656006,
-      "logits/rejected": -1.9171788692474365,
-      "logps/chosen": -31.4038028717041,
-      "logps/rejected": -33.20899963378906,
-      "loss": 0.696,
-      "rewards/accuracies": 0.5,
-      "rewards/chosen": 0.01106296293437481,
-      "rewards/margins": 0.0025471807457506657,
-      "rewards/rejected": 0.008515783585608006,
+      "logits/chosen": -1.9197067022323608,
+      "logits/rejected": -1.917022466659546,
+      "logps/chosen": -31.42661476135254,
+      "logps/rejected": -33.22706985473633,
+      "loss": 0.5,
+      "rewards/accuracies": 0.5375000238418579,
+      "rewards/chosen": -0.0001051858052960597,
+      "rewards/margins": -1.9103928934782743e-05,
+      "rewards/rejected": -8.608187636127695e-05,
       "step": 30
     },
     {
       "epoch": 0.1,
       "learning_rate": 4.999896948438433e-07,
-      "logits/chosen": -2.0166726112365723,
-      "logits/rejected": -2.0079195499420166,
-      "logps/chosen": -32.57840347290039,
-      "logps/rejected": -32.493370056152344,
-      "loss": 0.7066,
-      "rewards/accuracies": 0.48750001192092896,
-      "rewards/chosen": -0.00143451988697052,
-      "rewards/margins": -0.018809262663125992,
-      "rewards/rejected": 0.017374742776155472,
+      "logits/chosen": -2.0163283348083496,
+      "logits/rejected": -2.0075771808624268,
+      "logps/chosen": -32.59123992919922,
+      "logps/rejected": -32.48072052001953,
+      "loss": 0.5001,
+      "rewards/accuracies": 0.4625000059604645,
+      "rewards/chosen": -0.00014431352610699832,
+      "rewards/margins": -0.00046383924200199544,
+      "rewards/rejected": 0.0003195256576873362,
       "step": 40
     },
     {
       "epoch": 0.13,
       "learning_rate": 4.987541037542186e-07,
-      "logits/chosen": -1.8648544549942017,
-      "logits/rejected": -1.8540741205215454,
-      "logps/chosen": -33.54566955566406,
-      "logps/rejected": -35.43162155151367,
-      "loss": 0.7004,
-      "rewards/accuracies": 0.4749999940395355,
-      "rewards/chosen": 0.01062955055385828,
-      "rewards/margins": -0.00802132673561573,
-      "rewards/rejected": 0.018650878220796585,
+      "logits/chosen": -1.8642585277557373,
+      "logits/rejected": -1.8534786701202393,
+      "logps/chosen": -33.55145263671875,
+      "logps/rejected": -35.415096282958984,
+      "loss": 0.5001,
+      "rewards/accuracies": 0.375,
+      "rewards/chosen": 6.026435949024744e-05,
+      "rewards/margins": -0.0003122163761872798,
+      "rewards/rejected": 0.00037248071748763323,
       "step": 50
     },
     {
       "epoch": 0.16,
       "learning_rate": 4.954691471941118e-07,
-      "logits/chosen": -1.9450620412826538,
-      "logits/rejected": -1.947021245956421,
-      "logps/chosen": -32.5836296081543,
-      "logps/rejected": -33.20051193237305,
-      "loss": 0.6761,
-      "rewards/accuracies": 0.5375000238418579,
-      "rewards/chosen": 0.014901289716362953,
-      "rewards/margins": 0.04112350940704346,
-      "rewards/rejected": -0.026222219690680504,
+      "logits/chosen": -1.9458900690078735,
+      "logits/rejected": -1.9478263854980469,
+      "logps/chosen": -32.57926559448242,
+      "logps/rejected": -33.194175720214844,
+      "loss": 0.4999,
+      "rewards/accuracies": 0.5625,
+      "rewards/chosen": 0.00020916396169923246,
+      "rewards/margins": 0.00043710897443816066,
+      "rewards/rejected": -0.00022794504184275866,
       "step": 60
     },
     {
       "epoch": 0.18,
       "learning_rate": 4.901618883413548e-07,
-      "logits/chosen": -2.079589366912842,
-      "logits/rejected": -2.0845787525177,
-      "logps/chosen": -33.9754524230957,
-      "logps/rejected": -36.5793342590332,
-      "loss": 0.6876,
-      "rewards/accuracies": 0.512499988079071,
-      "rewards/chosen": 0.01173111516982317,
-      "rewards/margins": 0.021146392449736595,
-      "rewards/rejected": -0.00941527634859085,
+      "logits/chosen": -2.0795435905456543,
+      "logits/rejected": -2.084535837173462,
+      "logps/chosen": -33.986812591552734,
+      "logps/rejected": -36.573219299316406,
+      "loss": 0.5,
+      "rewards/accuracies": 0.48750001192092896,
+      "rewards/chosen": 1.6729489288991317e-05,
+      "rewards/margins": 6.016326369717717e-05,
+      "rewards/rejected": -4.343378896010108e-05,
       "step": 70
     },
     {
       "epoch": 0.21,
       "learning_rate": 4.828760511501322e-07,
-      "logits/chosen": -1.9424660205841064,
-      "logits/rejected": -1.945634126663208,
-      "logps/chosen": -34.394561767578125,
-      "logps/rejected": -34.57395553588867,
-      "loss": 0.6921,
-      "rewards/accuracies": 0.5625,
-      "rewards/chosen": 0.029921507462859154,
-      "rewards/margins": 0.013314949348568916,
-      "rewards/rejected": 0.016606558114290237,
+      "logits/chosen": -1.9420961141586304,
+      "logits/rejected": -1.9452673196792603,
+      "logps/chosen": -34.41278839111328,
+      "logps/rejected": -34.570655822753906,
+      "loss": 0.5,
+      "rewards/accuracies": 0.512499988079071,
+      "rewards/chosen": 0.00015023796004243195,
+      "rewards/margins": -6.727596337441355e-05,
+      "rewards/rejected": 0.0002175139234168455,
       "step": 80
     },
     {
       "epoch": 0.23,
       "learning_rate": 4.736716601303429e-07,
-      "logits/chosen": -1.950564980506897,
-      "logits/rejected": -1.955082893371582,
-      "logps/chosen": -32.44951248168945,
-      "logps/rejected": -32.353668212890625,
-      "loss": 0.6909,
-      "rewards/accuracies": 0.5249999761581421,
-      "rewards/chosen": 0.0161209125071764,
-      "rewards/margins": 0.011405264027416706,
-      "rewards/rejected": 0.004715651273727417,
+      "logits/chosen": -1.9506244659423828,
+      "logits/rejected": -1.9551340341567993,
+      "logps/chosen": -32.45224380493164,
+      "logps/rejected": -32.344200134277344,
+      "loss": 0.5,
+      "rewards/accuracies": 0.512499988079071,
+      "rewards/chosen": 0.0001517904020147398,
+      "rewards/margins": 4.694242761615897e-06,
+      "rewards/rejected": 0.00014709614333696663,
       "step": 90
     },
     {
       "epoch": 0.26,
       "learning_rate": 4.62624545834521e-07,
-      "logits/chosen": -2.0489234924316406,
-      "logits/rejected": -2.046917200088501,
-      "logps/chosen": -32.236671447753906,
-      "logps/rejected": -31.263763427734375,
-      "loss": 0.6951,
-      "rewards/accuracies": 0.5,
-      "rewards/chosen": 0.0037471160758286715,
-      "rewards/margins": 0.0015437646070495248,
-      "rewards/rejected": 0.002203352050855756,
+      "logits/chosen": -2.048786163330078,
+      "logits/rejected": -2.046797275543213,
+      "logps/chosen": -32.23818588256836,
+      "logps/rejected": -31.283069610595703,
+      "loss": 0.5,
+      "rewards/accuracies": 0.5874999761581421,
+      "rewards/chosen": 2.6492711185710505e-05,
+      "rewards/margins": 0.0001950610603671521,
+      "rewards/rejected": -0.00016856836737133563,
       "step": 100
     },
     {
       "epoch": 0.26,
-      "eval_logits/chosen": -2.243756055831909,
-      "eval_logits/rejected": -2.238872766494751,
-      "eval_logps/chosen": -34.025177001953125,
-      "eval_logps/rejected": -37.50759506225586,
-      "eval_loss": 0.696557343006134,
-      "eval_rewards/accuracies": 0.49543190002441406,
-      "eval_rewards/chosen": 0.008434689603745937,
-      "eval_rewards/margins": 0.0003138432221021503,
-      "eval_rewards/rejected": 0.008120844140648842,
-      "eval_runtime": 146.3373,
-      "eval_samples_per_second": 2.344,
+      "eval_logits/chosen": -2.243879556655884,
+      "eval_logits/rejected": -2.2390010356903076,
+      "eval_logps/chosen": -34.026790618896484,
+      "eval_logps/rejected": -37.49333572387695,
+      "eval_loss": 0.5000349879264832,
+      "eval_rewards/accuracies": 0.5137043595314026,
+      "eval_rewards/chosen": 7.762637687847018e-05,
+      "eval_rewards/margins": -0.00015523387992288917,
+      "eval_rewards/rejected": 0.00023286025680135936,
+      "eval_runtime": 146.2199,
+      "eval_samples_per_second": 2.346,
       "eval_steps_per_second": 0.294,
       "step": 100
     },
     {
       "epoch": 0.29,
       "learning_rate": 4.4982572012636904e-07,
-      "logits/chosen": -2.0052237510681152,
-      "logits/rejected": -2.002803325653076,
-      "logps/chosen": -33.24525833129883,
-      "logps/rejected": -34.01074981689453,
-      "loss": 0.7064,
-      "rewards/accuracies": 0.48750001192092896,
-      "rewards/chosen": -0.0013012334238737822,
-      "rewards/margins": -0.018059223890304565,
-      "rewards/rejected": 0.016757991164922714,
+      "logits/chosen": -2.005380392074585,
+      "logits/rejected": -2.0029690265655518,
+      "logps/chosen": -33.23069381713867,
+      "logps/rejected": -34.01119613647461,
+      "loss": 0.5,
+      "rewards/accuracies": 0.5249999761581421,
+      "rewards/chosen": 0.0001312425301875919,
+      "rewards/margins": -5.051253901910968e-05,
+      "rewards/rejected": 0.00018175505101680756,
       "step": 110
     },
     {
       "epoch": 0.31,
       "learning_rate": 4.353806263777677e-07,
-      "logits/chosen": -2.016326665878296,
-      "logits/rejected": -2.007951498031616,
-      "logps/chosen": -32.478302001953125,
-      "logps/rejected": -32.186073303222656,
-      "loss": 0.7012,
-      "rewards/accuracies": 0.48750001192092896,
-      "rewards/chosen": -0.030056962743401527,
-      "rewards/margins": -0.007947373203933239,
-      "rewards/rejected": -0.022109590470790863,
+      "logits/chosen": -2.016681432723999,
+      "logits/rejected": -2.0083086490631104,
+      "logps/chosen": -32.43777847290039,
+      "logps/rejected": -32.1855354309082,
+      "loss": 0.5,
+      "rewards/accuracies": 0.5625,
+      "rewards/chosen": 7.132566679501906e-05,
+      "rewards/margins": 0.00031160234357230365,
+      "rewards/rejected": -0.0002402766840532422,
       "step": 120
     },
     {
       "epoch": 0.34,
       "learning_rate": 4.194082707715275e-07,
-      "logits/chosen": -2.045950412750244,
-      "logits/rejected": -2.037904739379883,
-      "logps/chosen": -30.491313934326172,
-      "logps/rejected": -32.0540771484375,
-      "loss": 0.7025,
-      "rewards/accuracies": 0.4375,
-      "rewards/chosen": -0.01403898186981678,
-      "rewards/margins": -0.013075167313218117,
-      "rewards/rejected": -0.0009638145565986633,
+      "logits/chosen": -2.04573392868042,
+      "logits/rejected": -2.0376956462860107,
+      "logps/chosen": -30.474706649780273,
+      "logps/rejected": -32.02512741088867,
+      "loss": 0.5001,
+      "rewards/accuracies": 0.44999998807907104,
+      "rewards/chosen": 1.0108102287631482e-05,
+      "rewards/margins": -0.0002687241358216852,
+      "rewards/rejected": 0.00027883221628144383,
       "step": 130
     },
     {
       "epoch": 0.36,
       "learning_rate": 4.020402418666621e-07,
-      "logits/chosen": -1.9765291213989258,
-      "logits/rejected": -1.986802101135254,
-      "logps/chosen": -31.39646339416504,
-      "logps/rejected": -32.522178649902344,
-      "loss": 0.6923,
-      "rewards/accuracies": 0.512499988079071,
-      "rewards/chosen": 0.02087983302772045,
-      "rewards/margins": 0.00944516807794571,
-      "rewards/rejected": 0.011434664018452168,
+      "logits/chosen": -1.9768989086151123,
+      "logits/rejected": -1.9871835708618164,
+      "logps/chosen": -31.406789779663086,
+      "logps/rejected": -32.54368209838867,
+      "loss": 0.5,
+      "rewards/accuracies": 0.550000011920929,
+      "rewards/chosen": 0.00012870995851699263,
+      "rewards/margins": 0.00021670905698556453,
+      "rewards/rejected": -8.799910574452952e-05,
       "step": 140
     },
     {
       "epoch": 0.39,
       "learning_rate": 3.8341962650351185e-07,
-      "logits/chosen": -1.8905407190322876,
-      "logits/rejected": -1.8916336297988892,
-      "logps/chosen": -34.17607116699219,
-      "logps/rejected": -34.76817321777344,
-      "loss": 0.6851,
-      "rewards/accuracies": 0.4749999940395355,
-      "rewards/chosen": 0.01495879702270031,
-      "rewards/margins": 0.024784717708826065,
-      "rewards/rejected": -0.00982591975480318,
+      "logits/chosen": -1.8907657861709595,
+      "logits/rejected": -1.8918464183807373,
+      "logps/chosen": -34.17363739013672,
+      "logps/rejected": -34.758506774902344,
+      "loss": 0.5,
+      "rewards/accuracies": 0.5375000238418579,
+      "rewards/chosen": 0.00019054378208238631,
+      "rewards/margins": 0.00020306630176492035,
+      "rewards/rejected": -1.2522534234449267e-05,
       "step": 150
     },
     {
       "epoch": 0.42,
       "learning_rate": 3.636998309800572e-07,
-      "logits/chosen": -1.942565679550171,
-      "logits/rejected": -1.9390919208526611,
-      "logps/chosen": -36.1614875793457,
-      "logps/rejected": -32.72848129272461,
-      "loss": 0.6907,
-      "rewards/accuracies": 0.4625000059604645,
-      "rewards/chosen": 0.00917207170277834,
-      "rewards/margins": 0.01433003693819046,
-      "rewards/rejected": -0.0051579661667346954,
+      "logits/chosen": -1.9421974420547485,
+      "logits/rejected": -1.9387152194976807,
+      "logps/chosen": -36.14430618286133,
+      "logps/rejected": -32.699371337890625,
+      "loss": 0.5,
+      "rewards/accuracies": 0.6000000238418579,
+      "rewards/chosen": 0.0002737110189627856,
+      "rewards/margins": 3.994655708083883e-05,
+      "rewards/rejected": 0.00023376448370981961,
       "step": 160
     },
     {
       "epoch": 0.44,
       "learning_rate": 3.430433172111807e-07,
-      "logits/chosen": -2.0418710708618164,
-      "logits/rejected": -2.034475564956665,
-      "logps/chosen": -33.76544952392578,
-      "logps/rejected": -31.358556747436523,
-      "loss": 0.6791,
-      "rewards/accuracies": 0.574999988079071,
-      "rewards/chosen": 0.02917439118027687,
-      "rewards/margins": 0.03521214798092842,
-      "rewards/rejected": -0.0060377540066838264,
+      "logits/chosen": -2.042232036590576,
+      "logits/rejected": -2.034842014312744,
+      "logps/chosen": -33.76560974121094,
+      "logps/rejected": -31.359146118164062,
+      "loss": 0.4999,
+      "rewards/accuracies": 0.550000011920929,
+      "rewards/chosen": 0.0003225207910872996,
+      "rewards/margins": 0.00039547347114421427,
+      "rewards/rejected": -7.295267278095707e-05,
       "step": 170
     },
     {
       "epoch": 0.47,
       "learning_rate": 3.216202642830543e-07,
-      "logits/chosen": -2.0476231575012207,
-      "logits/rejected": -2.0529019832611084,
-      "logps/chosen": -32.533023834228516,
-      "logps/rejected": -32.506805419921875,
-      "loss": 0.6814,
-      "rewards/accuracies": 0.5625,
-      "rewards/chosen": 0.010587882250547409,
-      "rewards/margins": 0.03008626028895378,
-      "rewards/rejected": -0.019498378038406372,
+      "logits/chosen": -2.047433376312256,
+      "logits/rejected": -2.052696704864502,
+      "logps/chosen": -32.51101303100586,
+      "logps/rejected": -32.49535369873047,
+      "loss": 0.4999,
+      "rewards/accuracies": 0.550000011920929,
+      "rewards/chosen": 0.00033775376505218446,
+      "rewards/margins": 0.00043991307029500604,
+      "rewards/rejected": -0.0001021592179313302,
       "step": 180
     },
     {
       "epoch": 0.49,
       "learning_rate": 2.9960716642946403e-07,
-      "logits/chosen": -2.048576593399048,
-      "logits/rejected": -2.0457961559295654,
-      "logps/chosen": -31.470510482788086,
-      "logps/rejected": -31.322284698486328,
-      "loss": 0.687,
-      "rewards/accuracies": 0.5625,
-      "rewards/chosen": 0.016599375754594803,
-      "rewards/margins": 0.020261235535144806,
-      "rewards/rejected": -0.0036618602462112904,
+      "logits/chosen": -2.048191785812378,
+      "logits/rejected": -2.0454134941101074,
+      "logps/chosen": -31.475040435791016,
+      "logps/rejected": -31.315017700195312,
+      "loss": 0.5,
+      "rewards/accuracies": 0.5249999761581421,
+      "rewards/chosen": 0.00013913081784266979,
+      "rewards/margins": 0.00010714305244619027,
+      "rewards/rejected": 3.1987779948394746e-05,
       "step": 190
     },
     {
       "epoch": 0.52,
       "learning_rate": 2.771853789806683e-07,
-      "logits/chosen": -1.9184268712997437,
-      "logits/rejected": -1.9230947494506836,
-      "logps/chosen": -31.57404136657715,
-      "logps/rejected": -32.775550842285156,
-      "loss": 0.6891,
-      "rewards/accuracies": 0.48750001192092896,
-      "rewards/chosen": 0.02901501953601837,
-      "rewards/margins": 0.015173261985182762,
-      "rewards/rejected": 0.013841753825545311,
+      "logits/chosen": -1.9189927577972412,
+      "logits/rejected": -1.9236657619476318,
+      "logps/chosen": -31.597900390625,
+      "logps/rejected": -32.774513244628906,
+      "loss": 0.5,
+      "rewards/accuracies": 0.5,
+      "rewards/chosen": 8.383009844692424e-05,
+      "rewards/margins": -8.035473001655191e-05,
+      "rewards/rejected": 0.0001641848066356033,
       "step": 200
     },
     {
       "epoch": 0.52,
-      "eval_logits/chosen": -2.243248462677002,
-      "eval_logits/rejected": -2.238374948501587,
-      "eval_logps/chosen": -34.02431869506836,
-      "eval_logps/rejected": -37.511497497558594,
-      "eval_loss": 0.6946919560432434,
-      "eval_rewards/accuracies": 0.49335551261901855,
-      "eval_rewards/chosen": 0.009210066869854927,
-      "eval_rewards/margins": 0.0045979218557477,
-      "eval_rewards/rejected": 0.004612144082784653,
-      "eval_runtime": 146.0514,
-      "eval_samples_per_second": 2.348,
-      "eval_steps_per_second": 0.294,
+      "eval_logits/chosen": -2.244427442550659,
+      "eval_logits/rejected": -2.2395591735839844,
+      "eval_logps/chosen": -34.002986907958984,
+      "eval_logps/rejected": -37.49692916870117,
+      "eval_loss": 0.4999743700027466,
+      "eval_rewards/accuracies": 0.529900312423706,
+      "eval_rewards/chosen": 0.00031563788070343435,
+      "eval_rewards/margins": 0.0001187350062537007,
+      "eval_rewards/rejected": 0.00019690288172569126,
+      "eval_runtime": 145.9131,
+      "eval_samples_per_second": 2.351,
+      "eval_steps_per_second": 0.295,
       "step": 200
     },
     {
       "epoch": 0.55,
       "learning_rate": 2.5453962426402e-07,
-      "logits/chosen": -2.0314748287200928,
-      "logits/rejected": -2.0421440601348877,
-      "logps/chosen": -31.954639434814453,
-      "logps/rejected": -33.87318801879883,
-      "loss": 0.6864,
-      "rewards/accuracies": 0.550000011920929,
-      "rewards/chosen": 0.014999927952885628,
-      "rewards/margins": 0.024442464113235474,
-      "rewards/rejected": -0.009442536160349846,
+      "logits/chosen": -2.031794786453247,
+      "logits/rejected": -2.0424611568450928,
+      "logps/chosen": -31.942611694335938,
+      "logps/rejected": -33.87322998046875,
+      "loss": 0.4999,
+      "rewards/accuracies": 0.512499988079071,
+      "rewards/chosen": 0.0002869277377612889,
+      "rewards/margins": 0.0003922684118151665,
+      "rewards/rejected": -0.00010534075408941135,
       "step": 210
     },
     {
       "epoch": 0.57,
       "learning_rate": 2.318564697655179e-07,
-      "logits/chosen": -1.9249544143676758,
-      "logits/rejected": -1.9398053884506226,
-      "logps/chosen": -30.072036743164062,
-      "logps/rejected": -31.55165672302246,
-      "loss": 0.6827,
-      "rewards/accuracies": 0.6000000238418579,
-      "rewards/chosen": 0.035550959408283234,
-      "rewards/margins": 0.029681822285056114,
-      "rewards/rejected": 0.005869132932275534,
+      "logits/chosen": -1.925797462463379,
+      "logits/rejected": -1.9406683444976807,
+      "logps/chosen": -30.079837799072266,
+      "logps/rejected": -31.56089210510254,
+      "loss": 0.4999,
+      "rewards/accuracies": 0.5874999761581421,
+      "rewards/chosen": 0.00031700366525910795,
+      "rewards/margins": 0.0003441544540692121,
+      "rewards/rejected": -2.715078335313592e-05,
       "step": 220
     },
     {
       "epoch": 0.6,
       "learning_rate": 2.093227910899832e-07,
-      "logits/chosen": -1.9830297231674194,
-      "logits/rejected": -1.9869968891143799,
-      "logps/chosen": -33.3900146484375,
-      "logps/rejected": -31.545156478881836,
-      "loss": 0.6872,
-      "rewards/accuracies": 0.5375000238418579,
-      "rewards/chosen": 0.024708259850740433,
-      "rewards/margins": 0.02097567543387413,
-      "rewards/rejected": 0.00373258744366467,
+      "logits/chosen": -1.9836803674697876,
+      "logits/rejected": -1.9876391887664795,
+      "logps/chosen": -33.396968841552734,
+      "logps/rejected": -31.5497989654541,
+      "loss": 0.5,
+      "rewards/accuracies": 0.550000011920929,
+      "rewards/chosen": 0.00020496857177931815,
+      "rewards/margins": 0.0002099033008562401,
+      "rewards/rejected": -4.934729986416642e-06,
       "step": 230
     },
     {
       "epoch": 0.62,
       "learning_rate": 1.8712423238279356e-07,
-      "logits/chosen": -1.9825174808502197,
-      "logits/rejected": -1.9605509042739868,
-      "logps/chosen": -34.16549301147461,
-      "logps/rejected": -34.945499420166016,
-      "loss": 0.7065,
-      "rewards/accuracies": 0.5,
-      "rewards/chosen": -0.010572168976068497,
-      "rewards/margins": -0.015337007120251656,
-      "rewards/rejected": 0.004764837212860584,
+      "logits/chosen": -1.9834476709365845,
+      "logits/rejected": -1.9614884853363037,
+      "logps/chosen": -34.1593017578125,
+      "logps/rejected": -34.974971771240234,
+      "loss": 0.5,
+      "rewards/accuracies": 0.550000011920929,
+      "rewards/chosen": -5.55333899683319e-05,
+      "rewards/margins": 0.00018626711971592158,
+      "rewards/rejected": -0.00024180053151212633,
       "step": 240
     },
     {
       "epoch": 0.65,
       "learning_rate": 1.654436768970182e-07,
-      "logits/chosen": -2.0242011547088623,
-      "logits/rejected": -2.02089262008667,
-      "logps/chosen": -32.91257858276367,
-      "logps/rejected": -36.22357177734375,
-      "loss": 0.6913,
-      "rewards/accuracies": 0.5,
-      "rewards/chosen": 0.01124012004584074,
-      "rewards/margins": 0.010020162910223007,
-      "rewards/rejected": 0.0012199539924040437,
+      "logits/chosen": -2.0252513885498047,
+      "logits/rejected": -2.021942615509033,
+      "logps/chosen": -32.891014099121094,
+      "logps/rejected": -36.209808349609375,
+      "loss": 0.4999,
+      "rewards/accuracies": 0.4749999940395355,
+      "rewards/chosen": 0.00034049636451527476,
+      "rewards/margins": 0.00018933510000351816,
+      "rewards/rejected": 0.00015116124995984137,
       "step": 250
     },
     {
       "epoch": 0.68,
       "learning_rate": 1.444597403062196e-07,
-      "logits/chosen": -1.8910300731658936,
-      "logits/rejected": -1.888593316078186,
-      "logps/chosen": -34.191619873046875,
-      "logps/rejected": -35.50851058959961,
-      "loss": 0.6975,
-      "rewards/accuracies": 0.6000000238418579,
-      "rewards/chosen": -0.0028982784133404493,
-      "rewards/margins": 0.0006028197822161019,
-      "rewards/rejected": -0.003501094877719879,
+      "logits/chosen": -1.891709327697754,
+      "logits/rejected": -1.889259934425354,
+      "logps/chosen": -34.162227630615234,
+      "logps/rejected": -35.52161407470703,
+      "loss": 0.4999,
+      "rewards/accuracies": 0.612500011920929,
+      "rewards/chosen": 0.0002616822312120348,
+      "rewards/margins": 0.0004316047707106918,
+      "rewards/rejected": -0.0001699226413620636,
       "step": 260
     },
     {
       "epoch": 0.7,
       "learning_rate": 1.2434529917578887e-07,
-      "logits/chosen": -1.8757693767547607,
-      "logits/rejected": -1.8732143640518188,
-      "logps/chosen": -34.38648223876953,
-      "logps/rejected": -31.753692626953125,
-      "loss": 0.6995,
-      "rewards/accuracies": 0.512499988079071,
-      "rewards/chosen": 0.010007266886532307,
-      "rewards/margins": -0.005866709630936384,
-      "rewards/rejected": 0.015873977914452553,
+      "logits/chosen": -1.8765720129013062,
+      "logits/rejected": -1.8740289211273193,
+      "logps/chosen": -34.38764190673828,
+      "logps/rejected": -31.73568344116211,
+      "loss": 0.5,
+      "rewards/accuracies": 0.44999998807907104,
+      "rewards/chosen": 9.958025475498289e-05,
+      "rewards/margins": -0.00025685896980576217,
+      "rewards/rejected": 0.00035643921000882983,
       "step": 270
     },
     {
       "epoch": 0.73,
       "learning_rate": 1.0526606671603521e-07,
-      "logits/chosen": -1.9795001745224,
-      "logits/rejected": -1.9688692092895508,
-      "logps/chosen": -35.32746887207031,
-      "logps/rejected": -31.84292221069336,
-      "loss": 0.6843,
-      "rewards/accuracies": 0.550000011920929,
-      "rewards/chosen": 0.026254409924149513,
-      "rewards/margins": 0.027990642935037613,
-      "rewards/rejected": -0.001736226724460721,
+      "logits/chosen": -1.980292558670044,
+      "logits/rejected": -1.9696547985076904,
+      "logps/chosen": -35.3267936706543,
+      "logps/rejected": -31.829299926757812,
+      "loss": 0.5,
+      "rewards/accuracies": 0.4749999940395355,
+      "rewards/chosen": 0.00029847928090021014,
+      "rewards/margins": 0.00018156312580686063,
+      "rewards/rejected": 0.00011691616236930713,
       "step": 280
     },
     {
       "epoch": 0.75,
       "learning_rate": 8.737922755071453e-08,
-      "logits/chosen": -2.0753533840179443,
-      "logits/rejected": -2.0603318214416504,
-      "logps/chosen": -30.902263641357422,
-      "logps/rejected": -32.65242385864258,
-      "loss": 0.6872,
-      "rewards/accuracies": 0.512499988079071,
-      "rewards/chosen": 0.019674303010106087,
-      "rewards/margins": 0.020201902836561203,
-      "rewards/rejected": -0.0005276011070236564,
+      "logits/chosen": -2.075955390930176,
+      "logits/rejected": -2.0609400272369385,
+      "logps/chosen": -30.922338485717773,
+      "logps/rejected": -32.651031494140625,
+      "loss": 0.5,
+      "rewards/accuracies": 0.4625000059604645,
+      "rewards/chosen": 1.7866250345832668e-05,
+      "rewards/margins": 9.798279279493727e-06,
+      "rewards/rejected": 8.067952876444906e-06,
       "step": 290
     },
     {
       "epoch": 0.78,
       "learning_rate": 7.08321427484816e-08,
-      "logits/chosen": -1.946671724319458,
-      "logits/rejected": -1.9441486597061157,
-      "logps/chosen": -32.906158447265625,
-      "logps/rejected": -30.805999755859375,
-      "loss": 0.693,
-      "rewards/accuracies": 0.4749999940395355,
-      "rewards/chosen": 0.018858108669519424,
-      "rewards/margins": 0.009128611534833908,
-      "rewards/rejected": 0.009729497134685516,
+      "logits/chosen": -1.9476518630981445,
+      "logits/rejected": -1.945123314857483,
+      "logps/chosen": -32.88804626464844,
+      "logps/rejected": -30.81081771850586,
+      "loss": 0.4999,
+      "rewards/accuracies": 0.4625000059604645,
+      "rewards/chosen": 0.00039066909812390804,
+      "rewards/margins": 0.0003307634615339339,
+      "rewards/rejected": 5.9905636589974165e-05,
       "step": 300
     },
     {
       "epoch": 0.78,
-      "eval_logits/chosen": -2.243694305419922,
-      "eval_logits/rejected": -2.2388274669647217,
-      "eval_logps/chosen": -34.017173767089844,
-      "eval_logps/rejected": -37.497615814208984,
-      "eval_loss": 0.6981561183929443,
-      "eval_rewards/accuracies": 0.49833887815475464,
-      "eval_rewards/chosen": 0.015641551464796066,
-      "eval_rewards/margins": -0.0014640000881627202,
-      "eval_rewards/rejected": 0.017105549573898315,
-      "eval_runtime": 146.0253,
-      "eval_samples_per_second": 2.349,
-      "eval_steps_per_second": 0.294,
+      "eval_logits/chosen": -2.2444517612457275,
+      "eval_logits/rejected": -2.239577054977417,
+      "eval_logps/chosen": -34.00169372558594,
+      "eval_logps/rejected": -37.48704528808594,
+      "eval_loss": 0.49999141693115234,
+      "eval_rewards/accuracies": 0.5365448594093323,
+      "eval_rewards/chosen": 0.0003285549464635551,
+      "eval_rewards/margins": 3.2781601476017386e-05,
+      "eval_rewards/rejected": 0.00029577332315966487,
+      "eval_runtime": 145.927,
+      "eval_samples_per_second": 2.35,
+      "eval_steps_per_second": 0.295,
       "step": 300
     },
     {
       "epoch": 0.81,
       "learning_rate": 5.576113578589034e-08,
-      "logits/chosen": -1.9287757873535156,
-      "logits/rejected": -1.9255115985870361,
-      "logps/chosen": -31.573471069335938,
-      "logps/rejected": -33.71125030517578,
-      "loss": 0.698,
-      "rewards/accuracies": 0.5375000238418579,
-      "rewards/chosen": 0.026891669258475304,
-      "rewards/margins": -0.000918733305297792,
-      "rewards/rejected": 0.02781040407717228,
+      "logits/chosen": -1.929610013961792,
+      "logits/rejected": -1.9263417720794678,
+      "logps/chosen": -31.55474281311035,
+      "logps/rejected": -33.74074935913086,
+      "loss": 0.4999,
+      "rewards/accuracies": 0.612500011920929,
+      "rewards/chosen": 0.0004860997141804546,
+      "rewards/margins": 0.0004720650613307953,
+      "rewards/rejected": 1.4034636478754692e-05,
       "step": 310
     },
     {
       "epoch": 0.83,
       "learning_rate": 4.229036944380912e-08,
-      "logits/chosen": -1.9807313680648804,
-      "logits/rejected": -1.968425989151001,
-      "logps/chosen": -34.54930877685547,
-      "logps/rejected": -33.57306671142578,
-      "loss": 0.6615,
-      "rewards/accuracies": 0.637499988079071,
-      "rewards/chosen": 0.04208679869771004,
-      "rewards/margins": 0.07289497554302216,
-      "rewards/rejected": -0.030808180570602417,
+      "logits/chosen": -1.980975866317749,
+      "logits/rejected": -1.9686771631240845,
+      "logps/chosen": -34.552978515625,
+      "logps/rejected": -33.54259490966797,
+      "loss": 0.4999,
+      "rewards/accuracies": 0.550000011920929,
+      "rewards/chosen": 0.0004309539217501879,
+      "rewards/margins": 0.00046852690866217017,
+      "rewards/rejected": -3.757303420570679e-05,
       "step": 320
     },
     {
       "epoch": 0.86,
       "learning_rate": 3.053082288996112e-08,
-      "logits/chosen": -2.0161099433898926,
-      "logits/rejected": -2.0146496295928955,
-      "logps/chosen": -33.46127700805664,
-      "logps/rejected": -32.4744873046875,
-      "loss": 0.6883,
-      "rewards/accuracies": 0.5375000238418579,
-      "rewards/chosen": 0.022065162658691406,
-      "rewards/margins": 0.01747960038483143,
-      "rewards/rejected": 0.004585559479892254,
+      "logits/chosen": -2.016475200653076,
+      "logits/rejected": -2.015024185180664,
+      "logps/chosen": -33.44532012939453,
+      "logps/rejected": -32.47471237182617,
+      "loss": 0.4999,
+      "rewards/accuracies": 0.512499988079071,
+      "rewards/chosen": 0.00040476821595802903,
+      "rewards/margins": 0.0003560716868378222,
+      "rewards/rejected": 4.869652912020683e-05,
       "step": 330
     },
     {
       "epoch": 0.88,
       "learning_rate": 2.05793773749158e-08,
-      "logits/chosen": -2.102802276611328,
-      "logits/rejected": -2.0869994163513184,
-      "logps/chosen": -34.16718292236328,
-      "logps/rejected": -33.08827590942383,
-      "loss": 0.706,
-      "rewards/accuracies": 0.48750001192092896,
-      "rewards/chosen": 0.010987209156155586,
-      "rewards/margins": -0.01579815149307251,
-      "rewards/rejected": 0.026785362511873245,
+      "logits/chosen": -2.10372257232666,
+      "logits/rejected": -2.0879228115081787,
+      "logps/chosen": -34.1622200012207,
+      "logps/rejected": -33.08952331542969,
+      "loss": 0.5,
+      "rewards/accuracies": 0.4124999940395355,
+      "rewards/chosen": 0.00017170429055113345,
+      "rewards/margins": -0.00011341056961100549,
+      "rewards/rejected": 0.00028511486016213894,
       "step": 340
     },
     {
       "epoch": 0.91,
       "learning_rate": 1.251801807404168e-08,
-      "logits/chosen": -1.97471022605896,
-      "logits/rejected": -1.9737637042999268,
-      "logps/chosen": -33.25645065307617,
-      "logps/rejected": -32.46539306640625,
-      "loss": 0.6861,
-      "rewards/accuracies": 0.512499988079071,
-      "rewards/chosen": 0.023080622777342796,
-      "rewards/margins": 0.023302335292100906,
-      "rewards/rejected": -0.00022171511955093592,
+      "logits/chosen": -1.9753808975219727,
+      "logits/rejected": -1.9744256734848022,
+      "logps/chosen": -33.2431755065918,
+      "logps/rejected": -32.457496643066406,
+      "loss": 0.4999,
+      "rewards/accuracies": 0.44999998807907104,
+      "rewards/chosen": 0.0003892013046424836,
+      "rewards/margins": 0.00031266576843336225,
+      "rewards/rejected": 7.653549255337566e-05,
       "step": 350
     },
     {
       "epoch": 0.94,
       "learning_rate": 6.41315865106129e-09,
-      "logits/chosen": -1.9304895401000977,
-      "logits/rejected": -1.9408460855484009,
-      "logps/chosen": -32.22315979003906,
-      "logps/rejected": -35.28199005126953,
-      "loss": 0.7085,
-      "rewards/accuracies": 0.42500001192092896,
-      "rewards/chosen": -0.01662321388721466,
-      "rewards/margins": -0.022716889157891273,
-      "rewards/rejected": 0.006093672942370176,
+      "logits/chosen": -1.9311378002166748,
+      "logits/rejected": -1.9414989948272705,
+      "logps/chosen": -32.21795654296875,
+      "logps/rejected": -35.26204299926758,
+      "loss": 0.5001,
+      "rewards/accuracies": 0.44999998807907104,
+      "rewards/chosen": -0.00013266984024085104,
+      "rewards/margins": -0.0003998771426267922,
+      "rewards/rejected": 0.00026720724417828023,
       "step": 360
     },
     {
       "epoch": 0.96,
       "learning_rate": 2.3150941078050324e-09,
-      "logits/chosen": -2.0692858695983887,
-      "logits/rejected": -2.0627408027648926,
-      "logps/chosen": -33.64545440673828,
-      "logps/rejected": -29.203441619873047,
-      "loss": 0.704,
-      "rewards/accuracies": 0.42500001192092896,
-      "rewards/chosen": 0.00586346909403801,
-      "rewards/margins": -0.014821496792137623,
-      "rewards/rejected": 0.020684964954853058,
+      "logits/chosen": -2.0701680183410645,
+      "logits/rejected": -2.063605546951294,
+      "logps/chosen": -33.66097640991211,
+      "logps/rejected": -29.19500732421875,
+      "loss": 0.5001,
+      "rewards/accuracies": 0.4000000059604645,
+      "rewards/chosen": -9.009057976072654e-05,
+      "rewards/margins": -0.0004042644868604839,
+      "rewards/rejected": 0.0003141739289276302,
       "step": 370
     },
     {
       "epoch": 0.99,
       "learning_rate": 2.575864278703266e-10,
-      "logits/chosen": -1.928911805152893,
-      "logits/rejected": -1.9310725927352905,
-      "logps/chosen": -34.24489212036133,
-      "logps/rejected": -30.92240333557129,
-      "loss": 0.6763,
-      "rewards/accuracies": 0.5249999761581421,
-      "rewards/chosen": -0.005258283112198114,
-      "rewards/margins": 0.03954260051250458,
-      "rewards/rejected": -0.044800881296396255,
+      "logits/chosen": -1.9299176931381226,
+      "logits/rejected": -1.9320815801620483,
+      "logps/chosen": -34.23712921142578,
+      "logps/rejected": -30.907180786132812,
+      "loss": 0.4999,
+      "rewards/accuracies": 0.550000011920929,
+      "rewards/chosen": 1.9172584870830178e-05,
+      "rewards/margins": 0.00036473083309829235,
+      "rewards/rejected": -0.0003455582191236317,
       "step": 380
     },
     {
       "epoch": 1.0,
       "step": 385,
       "total_flos": 0.0,
-      "train_loss": 0.6916276120520257,
-      "train_runtime": 3257.6062,
-      "train_samples_per_second": 0.945,
+      "train_loss": 0.4999783658362054,
+      "train_runtime": 3254.8055,
+      "train_samples_per_second": 0.946,
       "train_steps_per_second": 0.118
     }
   ],

training_args.bin

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:914e32cb321285a370aa5440b23d05260ae348cf6fed9440d5a1e634ebac11d9
+oid sha256:1b322081d1b31ac18d7549fcc597dd984caab9247d9431dcc16128652c9c9ae5
 size 4984