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: 0.7054
-- Rewards/chosen: -0.0010
-- Rewards/rejected: 0.0163
-- Rewards/accuracies: 0.4726
-- Rewards/margins: -0.0174
-- Logps/rejected: -37.4962
-- Logps/chosen: -34.0358
-- Logits/rejected: -2.2386
-- Logits/chosen: -2.2435
+- Loss: 0.6982
+- Rewards/chosen: 0.0156
+- Rewards/rejected: 0.0171
+- Rewards/accuracies: 0.4983
+- Rewards/margins: -0.0015
+- Logps/rejected: -37.4976
+- Logps/chosen: -34.0172
+- Logits/rejected: -2.2388
+- Logits/chosen: -2.2437
 
 ## 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 |
 |:-------------:|:-----:|:----:|:---------------:|:--------------:|:----------------:|:------------------:|:---------------:|:--------------:|:------------:|:---------------:|:-------------:|
-| 0.6927        | 0.26  | 100  | 0.6927          | 0.0150         | 0.0088           | 0.5187             | 0.0062          | -37.5056       | -34.0159     | -2.2389         | -2.2438       |
-| 0.6949        | 0.52  | 200  | 0.6928          | 0.0118         | 0.0055           | 0.5274             | 0.0063          | -37.5098       | -34.0198     | -2.2387         | -2.2436       |
-| 0.6798        | 0.78  | 300  | 0.6981          | 0.0062         | 0.0101           | 0.4983             | -0.0039         | -37.5040       | -34.0268     | -2.2390         | -2.2439       |
+| 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       |
 
 
 ### 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",
     "v_proj",
-    "o_proj",
-    "q_proj",
-    "k_proj",
-    "up_proj"
+    "k_proj"
   ],
   "task_type": "CAUSAL_LM",
   "use_rslora": false

adapter_model.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:4241a5bac4adc39debfd72b4e4d2124e198b356cd95c7245382199b8383c449c
+oid sha256:b30a3614e90564741bd0b06f4b5d211ac9ab0d5d4651e82dec2b571e49a407d7
 size 176183216

all_results.json

@@ -1,21 +1,8 @@
 {
     "epoch": 1.0,
-    "eval_logits/chosen": -2.2434651851654053,
-    "eval_logits/rejected": -2.2385833263397217,
-    "eval_logps/chosen": -34.03584671020508,
-    "eval_logps/rejected": -37.496219635009766,
-    "eval_loss": 0.7054064273834229,
-    "eval_rewards/accuracies": 0.472591370344162,
-    "eval_rewards/chosen": -0.001035186112858355,
-    "eval_rewards/margins": -0.017356980592012405,
-    "eval_rewards/rejected": 0.01632179319858551,
-    "eval_runtime": 145.8229,
-    "eval_samples": 343,
-    "eval_samples_per_second": 2.352,
-    "eval_steps_per_second": 0.295,
-    "train_loss": 0.6928803004227675,
-    "train_runtime": 3255.7403,
+    "train_loss": 0.6916276120520257,
+    "train_runtime": 3257.6062,
     "train_samples": 3079,
-    "train_samples_per_second": 0.946,
+    "train_samples_per_second": 0.945,
     "train_steps_per_second": 0.118
 }

train_results.json

@@ -1,8 +1,8 @@
 {
     "epoch": 1.0,
-    "train_loss": 0.6928803004227675,
-    "train_runtime": 3255.7403,
+    "train_loss": 0.6916276120520257,
+    "train_runtime": 3257.6062,
     "train_samples": 3079,
-    "train_samples_per_second": 0.946,
+    "train_samples_per_second": 0.945,
     "train_steps_per_second": 0.118
 }

trainer_state.json

@@ -25,590 +25,590 @@
     {
       "epoch": 0.03,
       "learning_rate": 1.2820512820512818e-07,
-      "logits/chosen": -1.8660534620285034,
-      "logits/rejected": -1.870363712310791,
-      "logps/chosen": -37.0097770690918,
-      "logps/rejected": -33.656063079833984,
-      "loss": 0.6866,
-      "rewards/accuracies": 0.4861111044883728,
-      "rewards/chosen": -0.0025405611377209425,
-      "rewards/margins": 0.018248552456498146,
-      "rewards/rejected": -0.020789114758372307,
+      "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,
       "step": 10
     },
     {
       "epoch": 0.05,
       "learning_rate": 2.5641025641025636e-07,
-      "logits/chosen": -1.9975675344467163,
-      "logits/rejected": -2.0001988410949707,
-      "logps/chosen": -29.643625259399414,
-      "logps/rejected": -29.03875732421875,
-      "loss": 0.7064,
-      "rewards/accuracies": 0.4000000059604645,
-      "rewards/chosen": -0.0011415861081331968,
-      "rewards/margins": -0.02089018002152443,
-      "rewards/rejected": 0.019748592749238014,
+      "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,
       "step": 20
     },
     {
       "epoch": 0.08,
       "learning_rate": 3.8461538461538463e-07,
-      "logits/chosen": -1.919878363609314,
-      "logits/rejected": -1.9171901941299438,
-      "logps/chosen": -31.411758422851562,
-      "logps/rejected": -33.21895980834961,
-      "loss": 0.6945,
-      "rewards/accuracies": 0.48750001192092896,
-      "rewards/chosen": 0.0034693297930061817,
-      "rewards/margins": 0.003873053938150406,
-      "rewards/rejected": -0.0004037246108055115,
+      "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,
       "step": 30
     },
     {
       "epoch": 0.1,
       "learning_rate": 4.999896948438433e-07,
-      "logits/chosen": -2.0166752338409424,
-      "logits/rejected": -2.007917881011963,
-      "logps/chosen": -32.59661102294922,
-      "logps/rejected": -32.506187438964844,
-      "loss": 0.7064,
+      "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": -0.01584548130631447,
-      "rewards/margins": -0.02103354223072529,
-      "rewards/rejected": 0.005188059527426958,
+      "rewards/chosen": -0.00143451988697052,
+      "rewards/margins": -0.018809262663125992,
+      "rewards/rejected": 0.017374742776155472,
       "step": 40
     },
     {
       "epoch": 0.13,
       "learning_rate": 4.987541037542186e-07,
-      "logits/chosen": -1.8648042678833008,
-      "logits/rejected": -1.8540267944335938,
-      "logps/chosen": -33.56422805786133,
-      "logps/rejected": -35.431602478027344,
-      "loss": 0.7067,
-      "rewards/accuracies": 0.42500001192092896,
-      "rewards/chosen": -0.005400418769568205,
-      "rewards/margins": -0.02199491299688816,
-      "rewards/rejected": 0.01659449376165867,
+      "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,
       "step": 50
     },
     {
       "epoch": 0.16,
       "learning_rate": 4.954691471941118e-07,
-      "logits/chosen": -1.9453144073486328,
-      "logits/rejected": -1.9472593069076538,
-      "logps/chosen": -32.59089279174805,
-      "logps/rejected": -33.192405700683594,
-      "loss": 0.6844,
-      "rewards/accuracies": 0.4625000059604645,
-      "rewards/chosen": 0.007432769052684307,
-      "rewards/margins": 0.024254899471998215,
-      "rewards/rejected": -0.016822131350636482,
+      "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,
       "step": 60
     },
     {
       "epoch": 0.18,
       "learning_rate": 4.901618883413548e-07,
-      "logits/chosen": -2.079728364944458,
-      "logits/rejected": -2.0847129821777344,
-      "logps/chosen": -33.983558654785156,
-      "logps/rejected": -36.584938049316406,
-      "loss": 0.688,
-      "rewards/accuracies": 0.5375000238418579,
-      "rewards/chosen": 0.003943340387195349,
-      "rewards/margins": 0.016796352341771126,
-      "rewards/rejected": -0.01285301148891449,
+      "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,
       "step": 70
     },
     {
       "epoch": 0.21,
       "learning_rate": 4.828760511501322e-07,
-      "logits/chosen": -1.9422305822372437,
-      "logits/rejected": -1.9453967809677124,
-      "logps/chosen": -34.4206657409668,
-      "logps/rejected": -34.566123962402344,
-      "loss": 0.7053,
-      "rewards/accuracies": 0.4375,
-      "rewards/chosen": 0.005717824678868055,
-      "rewards/margins": -0.015306837856769562,
-      "rewards/rejected": 0.021024659276008606,
+      "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,
       "step": 80
     },
     {
       "epoch": 0.23,
       "learning_rate": 4.736716601303429e-07,
-      "logits/chosen": -1.950453519821167,
-      "logits/rejected": -1.9549684524536133,
-      "logps/chosen": -32.457984924316406,
-      "logps/rejected": -32.36473083496094,
-      "loss": 0.6897,
+      "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": 0.007545561995357275,
-      "rewards/margins": 0.012203911319375038,
-      "rewards/rejected": -0.004658351186662912,
+      "rewards/chosen": 0.0161209125071764,
+      "rewards/margins": 0.011405264027416706,
+      "rewards/rejected": 0.004715651273727417,
       "step": 90
     },
     {
       "epoch": 0.26,
       "learning_rate": 4.62624545834521e-07,
-      "logits/chosen": -2.048896312713623,
-      "logits/rejected": -2.0468926429748535,
-      "logps/chosen": -32.23190689086914,
-      "logps/rejected": -31.26666831970215,
-      "loss": 0.6927,
+      "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": 0.007141563110053539,
-      "rewards/margins": 0.007507171481847763,
-      "rewards/rejected": -0.0003656085464172065,
+      "rewards/chosen": 0.0037471160758286715,
+      "rewards/margins": 0.0015437646070495248,
+      "rewards/rejected": 0.002203352050855756,
       "step": 100
     },
     {
       "epoch": 0.26,
-      "eval_logits/chosen": -2.2437734603881836,
-      "eval_logits/rejected": -2.2388932704925537,
-      "eval_logps/chosen": -34.01585006713867,
-      "eval_logps/rejected": -37.50562286376953,
-      "eval_loss": 0.6927134394645691,
-      "eval_rewards/accuracies": 0.5186877250671387,
-      "eval_rewards/chosen": 0.014961617067456245,
-      "eval_rewards/margins": 0.006164299789816141,
-      "eval_rewards/rejected": 0.008797316811978817,
-      "eval_runtime": 146.2315,
-      "eval_samples_per_second": 2.346,
+      "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_steps_per_second": 0.294,
       "step": 100
     },
     {
       "epoch": 0.29,
       "learning_rate": 4.4982572012636904e-07,
-      "logits/chosen": -2.005382537841797,
-      "logits/rejected": -2.002962589263916,
-      "logps/chosen": -33.231327056884766,
-      "logps/rejected": -34.03110885620117,
-      "loss": 0.6912,
-      "rewards/accuracies": 0.5249999761581421,
-      "rewards/chosen": 0.009991822764277458,
-      "rewards/margins": 0.01138134766370058,
-      "rewards/rejected": -0.0013895228039473295,
+      "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,
       "step": 110
     },
     {
       "epoch": 0.31,
       "learning_rate": 4.353806263777677e-07,
-      "logits/chosen": -2.01668643951416,
-      "logits/rejected": -2.0083160400390625,
-      "logps/chosen": -32.46741485595703,
-      "logps/rejected": -32.169979095458984,
-      "loss": 0.7019,
-      "rewards/accuracies": 0.4375,
-      "rewards/chosen": -0.018008466809988022,
-      "rewards/margins": -0.011230994947254658,
-      "rewards/rejected": -0.006777471862733364,
+      "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,
       "step": 120
     },
     {
       "epoch": 0.34,
       "learning_rate": 4.194082707715275e-07,
-      "logits/chosen": -2.046079158782959,
-      "logits/rejected": -2.038038969039917,
-      "logps/chosen": -30.489391326904297,
-      "logps/rejected": -32.04558563232422,
-      "loss": 0.7042,
-      "rewards/accuracies": 0.4625000059604645,
-      "rewards/chosen": -0.010940661653876305,
-      "rewards/margins": -0.01687880977988243,
-      "rewards/rejected": 0.005938149057328701,
+      "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,
       "step": 130
     },
     {
       "epoch": 0.36,
       "learning_rate": 4.020402418666621e-07,
-      "logits/chosen": -1.9770376682281494,
-      "logits/rejected": -1.9873006343841553,
-      "logps/chosen": -31.392288208007812,
-      "logps/rejected": -32.56155014038086,
-      "loss": 0.6749,
-      "rewards/accuracies": 0.6625000238418579,
-      "rewards/chosen": 0.021898187696933746,
-      "rewards/margins": 0.043236203491687775,
-      "rewards/rejected": -0.02133801393210888,
+      "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,
       "step": 140
     },
     {
       "epoch": 0.39,
       "learning_rate": 3.8341962650351185e-07,
-      "logits/chosen": -1.8905258178710938,
-      "logits/rejected": -1.891603708267212,
-      "logps/chosen": -34.185401916503906,
-      "logps/rejected": -34.75876998901367,
-      "loss": 0.6933,
+      "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": 0.005832195747643709,
-      "rewards/margins": 0.007043843157589436,
-      "rewards/rejected": -0.0012116472935304046,
+      "rewards/chosen": 0.01495879702270031,
+      "rewards/margins": 0.024784717708826065,
+      "rewards/rejected": -0.00982591975480318,
       "step": 150
     },
     {
       "epoch": 0.42,
       "learning_rate": 3.636998309800572e-07,
-      "logits/chosen": -1.942525863647461,
-      "logits/rejected": -1.9390523433685303,
-      "logps/chosen": -36.15251922607422,
-      "logps/rejected": -32.721561431884766,
-      "loss": 0.6891,
-      "rewards/accuracies": 0.512499988079071,
-      "rewards/chosen": 0.01532670110464096,
-      "rewards/margins": 0.014371859841048717,
-      "rewards/rejected": 0.0009548395755700767,
+      "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,
       "step": 160
     },
     {
       "epoch": 0.44,
       "learning_rate": 3.430433172111807e-07,
-      "logits/chosen": -2.042137622833252,
-      "logits/rejected": -2.0347514152526855,
-      "logps/chosen": -33.78362274169922,
-      "logps/rejected": -31.34653091430664,
-      "loss": 0.6922,
-      "rewards/accuracies": 0.6000000238418579,
-      "rewards/chosen": 0.011391163803637028,
-      "rewards/margins": 0.007135935127735138,
-      "rewards/rejected": 0.004255227744579315,
+      "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,
       "step": 170
     },
     {
       "epoch": 0.47,
       "learning_rate": 3.216202642830543e-07,
-      "logits/chosen": -2.047314405441284,
-      "logits/rejected": -2.052593231201172,
-      "logps/chosen": -32.52313995361328,
-      "logps/rejected": -32.499427795410156,
-      "loss": 0.6811,
-      "rewards/accuracies": 0.550000011920929,
-      "rewards/chosen": 0.01731465384364128,
-      "rewards/margins": 0.028742337599396706,
-      "rewards/rejected": -0.011427680030465126,
+      "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,
       "step": 180
     },
     {
       "epoch": 0.49,
       "learning_rate": 2.9960716642946403e-07,
-      "logits/chosen": -2.048703908920288,
-      "logits/rejected": -2.0459342002868652,
-      "logps/chosen": -31.495092391967773,
-      "logps/rejected": -31.32244300842285,
-      "loss": 0.6967,
-      "rewards/accuracies": 0.4749999940395355,
-      "rewards/chosen": -0.004913434386253357,
-      "rewards/margins": -0.0015291940653696656,
-      "rewards/rejected": -0.0033842422999441624,
+      "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,
       "step": 190
     },
     {
       "epoch": 0.52,
       "learning_rate": 2.771853789806683e-07,
-      "logits/chosen": -1.9186630249023438,
-      "logits/rejected": -1.9233324527740479,
-      "logps/chosen": -31.604812622070312,
-      "logps/rejected": -32.791358947753906,
-      "loss": 0.6949,
-      "rewards/accuracies": 0.44999998807907104,
-      "rewards/chosen": 0.0011774369049817324,
-      "rewards/margins": 0.0015200242633000016,
-      "rewards/rejected": -0.00034258683444932103,
+      "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,
       "step": 200
     },
     {
       "epoch": 0.52,
-      "eval_logits/chosen": -2.2436025142669678,
-      "eval_logits/rejected": -2.2387301921844482,
-      "eval_logps/chosen": -34.01983642578125,
-      "eval_logps/rejected": -37.50978469848633,
-      "eval_loss": 0.6928001642227173,
-      "eval_rewards/accuracies": 0.5274086594581604,
-      "eval_rewards/chosen": 0.011776229366660118,
-      "eval_rewards/margins": 0.006309796124696732,
-      "eval_rewards/rejected": 0.0054664346389472485,
-      "eval_runtime": 145.9171,
-      "eval_samples_per_second": 2.351,
-      "eval_steps_per_second": 0.295,
+      "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,
       "step": 200
     },
     {
       "epoch": 0.55,
       "learning_rate": 2.5453962426402e-07,
-      "logits/chosen": -2.0316667556762695,
-      "logits/rejected": -2.0423641204833984,
-      "logps/chosen": -31.9429874420166,
-      "logps/rejected": -33.863956451416016,
-      "loss": 0.6846,
-      "rewards/accuracies": 0.512499988079071,
-      "rewards/chosen": 0.022652912884950638,
-      "rewards/margins": 0.023658016696572304,
-      "rewards/rejected": -0.0010051012504845858,
+      "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,
       "step": 210
     },
     {
       "epoch": 0.57,
       "learning_rate": 2.318564697655179e-07,
-      "logits/chosen": -1.9254566431045532,
-      "logits/rejected": -1.9402967691421509,
-      "logps/chosen": -30.087352752685547,
-      "logps/rejected": -31.55807113647461,
-      "loss": 0.6858,
-      "rewards/accuracies": 0.5375000238418579,
-      "rewards/chosen": 0.019350551068782806,
-      "rewards/margins": 0.019265536218881607,
-      "rewards/rejected": 8.501634147251025e-05,
+      "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,
       "step": 220
     },
     {
       "epoch": 0.6,
       "learning_rate": 2.093227910899832e-07,
-      "logits/chosen": -1.9833059310913086,
-      "logits/rejected": -1.9872699975967407,
-      "logps/chosen": -33.40017318725586,
-      "logps/rejected": -31.5333309173584,
-      "loss": 0.6963,
-      "rewards/accuracies": 0.4625000059604645,
-      "rewards/chosen": 0.013836635276675224,
-      "rewards/margins": 0.0010608620941638947,
-      "rewards/rejected": 0.012775774113833904,
+      "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,
       "step": 230
     },
     {
       "epoch": 0.62,
       "learning_rate": 1.8712423238279356e-07,
-      "logits/chosen": -1.983289122581482,
-      "logits/rejected": -1.9613111019134521,
-      "logps/chosen": -34.167720794677734,
-      "logps/rejected": -34.96076965332031,
-      "loss": 0.6978,
-      "rewards/accuracies": 0.550000011920929,
-      "rewards/chosen": -0.011179247871041298,
-      "rewards/margins": -0.0031959381885826588,
-      "rewards/rejected": -0.007983307354152203,
+      "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,
       "step": 240
     },
     {
       "epoch": 0.65,
       "learning_rate": 1.654436768970182e-07,
-      "logits/chosen": -2.0248379707336426,
-      "logits/rejected": -2.0215275287628174,
-      "logps/chosen": -32.91853713989258,
-      "logps/rejected": -36.22992706298828,
-      "loss": 0.6918,
-      "rewards/accuracies": 0.5249999761581421,
-      "rewards/chosen": 0.005225626286119223,
-      "rewards/margins": 0.009226549416780472,
-      "rewards/rejected": -0.004000924527645111,
+      "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,
       "step": 250
     },
     {
       "epoch": 0.68,
       "learning_rate": 1.444597403062196e-07,
-      "logits/chosen": -1.8912242650985718,
-      "logits/rejected": -1.8887859582901,
-      "logps/chosen": -34.17466735839844,
-      "logps/rejected": -35.498863220214844,
-      "loss": 0.6925,
-      "rewards/accuracies": 0.5874999761581421,
-      "rewards/chosen": 0.01098463125526905,
-      "rewards/margins": 0.006378169171512127,
-      "rewards/rejected": 0.004606460686773062,
+      "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,
       "step": 260
     },
     {
       "epoch": 0.7,
       "learning_rate": 1.2434529917578887e-07,
-      "logits/chosen": -1.8761816024780273,
-      "logits/rejected": -1.8736416101455688,
-      "logps/chosen": -34.40642166137695,
-      "logps/rejected": -31.732311248779297,
-      "loss": 0.7147,
-      "rewards/accuracies": 0.36250001192092896,
-      "rewards/chosen": -0.007056808564811945,
-      "rewards/margins": -0.03827430680394173,
-      "rewards/rejected": 0.03121749497950077,
+      "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,
       "step": 270
     },
     {
       "epoch": 0.73,
       "learning_rate": 1.0526606671603521e-07,
-      "logits/chosen": -1.9794687032699585,
-      "logits/rejected": -1.9688432216644287,
-      "logps/chosen": -35.32493591308594,
-      "logps/rejected": -31.86197853088379,
-      "loss": 0.6757,
-      "rewards/accuracies": 0.5874999761581421,
-      "rewards/chosen": 0.025365391746163368,
-      "rewards/margins": 0.04215434193611145,
-      "rewards/rejected": -0.016788948327302933,
+      "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,
       "step": 280
     },
     {
       "epoch": 0.75,
       "learning_rate": 8.737922755071453e-08,
-      "logits/chosen": -2.0760245323181152,
-      "logits/rejected": -2.0610060691833496,
-      "logps/chosen": -30.925411224365234,
-      "logps/rejected": -32.648170471191406,
-      "loss": 0.6978,
-      "rewards/accuracies": 0.4625000059604645,
-      "rewards/chosen": -0.0010284179588779807,
-      "rewards/margins": -0.003961210139095783,
-      "rewards/rejected": 0.0029327913653105497,
+      "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,
       "step": 290
     },
     {
       "epoch": 0.78,
       "learning_rate": 7.08321427484816e-08,
-      "logits/chosen": -1.946812629699707,
-      "logits/rejected": -1.9442851543426514,
-      "logps/chosen": -32.90645217895508,
-      "logps/rejected": -30.836864471435547,
-      "loss": 0.6798,
-      "rewards/accuracies": 0.5625,
-      "rewards/chosen": 0.016527386382222176,
-      "rewards/margins": 0.032571423798799515,
-      "rewards/rejected": -0.016044041141867638,
+      "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,
       "step": 300
     },
     {
       "epoch": 0.78,
-      "eval_logits/chosen": -2.243879556655884,
-      "eval_logits/rejected": -2.239006996154785,
-      "eval_logps/chosen": -34.026798248291016,
-      "eval_logps/rejected": -37.50396728515625,
-      "eval_loss": 0.6981266140937805,
+      "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": 0.006204271223396063,
-      "eval_rewards/margins": -0.003920448012650013,
-      "eval_rewards/rejected": 0.010124719701707363,
-      "eval_runtime": 145.9481,
-      "eval_samples_per_second": 2.35,
-      "eval_steps_per_second": 0.295,
+      "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,
       "step": 300
     },
     {
       "epoch": 0.81,
       "learning_rate": 5.576113578589034e-08,
-      "logits/chosen": -1.9286400079727173,
-      "logits/rejected": -1.92538583278656,
-      "logps/chosen": -31.571802139282227,
-      "logps/rejected": -33.7106819152832,
-      "loss": 0.6961,
+      "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": 0.025238346308469772,
-      "rewards/margins": 5.8975441788788885e-05,
-      "rewards/rejected": 0.025179371237754822,
+      "rewards/chosen": 0.026891669258475304,
+      "rewards/margins": -0.000918733305297792,
+      "rewards/rejected": 0.02781040407717228,
       "step": 310
     },
     {
       "epoch": 0.83,
       "learning_rate": 4.229036944380912e-08,
-      "logits/chosen": -1.9804341793060303,
-      "logits/rejected": -1.9681293964385986,
-      "logps/chosen": -34.57010269165039,
-      "logps/rejected": -33.552330017089844,
-      "loss": 0.6803,
-      "rewards/accuracies": 0.5,
-      "rewards/chosen": 0.020780155435204506,
-      "rewards/margins": 0.031573377549648285,
-      "rewards/rejected": -0.01079322025179863,
+      "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,
       "step": 320
     },
     {
       "epoch": 0.86,
       "learning_rate": 3.053082288996112e-08,
-      "logits/chosen": -2.0159571170806885,
-      "logits/rejected": -2.0144896507263184,
-      "logps/chosen": -33.452049255371094,
-      "logps/rejected": -32.46733856201172,
-      "loss": 0.687,
-      "rewards/accuracies": 0.5,
-      "rewards/chosen": 0.026994183659553528,
-      "rewards/margins": 0.017200354486703873,
-      "rewards/rejected": 0.009793824516236782,
+      "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,
       "step": 330
     },
     {
       "epoch": 0.88,
       "learning_rate": 2.05793773749158e-08,
-      "logits/chosen": -2.103339433670044,
-      "logits/rejected": -2.0875279903411865,
-      "logps/chosen": -34.17384719848633,
-      "logps/rejected": -33.08417510986328,
-      "loss": 0.7089,
-      "rewards/accuracies": 0.4375,
-      "rewards/chosen": 0.004437476396560669,
-      "rewards/margins": -0.02264832705259323,
-      "rewards/rejected": 0.02708580158650875,
+      "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,
       "step": 340
     },
     {
       "epoch": 0.91,
       "learning_rate": 1.251801807404168e-08,
-      "logits/chosen": -1.974784255027771,
-      "logits/rejected": -1.9738404750823975,
-      "logps/chosen": -33.24315643310547,
-      "logps/rejected": -32.44605255126953,
-      "loss": 0.6895,
-      "rewards/accuracies": 0.4625000059604645,
-      "rewards/chosen": 0.031151747331023216,
-      "rewards/margins": 0.01587487757205963,
-      "rewards/rejected": 0.015276871621608734,
+      "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,
       "step": 350
     },
     {
       "epoch": 0.94,
       "learning_rate": 6.41315865106129e-09,
-      "logits/chosen": -1.9305660724639893,
-      "logits/rejected": -1.940930724143982,
-      "logps/chosen": -32.193580627441406,
-      "logps/rejected": -35.30970764160156,
-      "loss": 0.6828,
-      "rewards/accuracies": 0.5375000238418579,
-      "rewards/chosen": 0.008886909112334251,
-      "rewards/margins": 0.025648051872849464,
-      "rewards/rejected": -0.016761142760515213,
+      "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,
       "step": 360
     },
     {
       "epoch": 0.96,
       "learning_rate": 2.3150941078050324e-09,
-      "logits/chosen": -2.069566249847412,
-      "logits/rejected": -2.0630078315734863,
-      "logps/chosen": -33.643856048583984,
-      "logps/rejected": -29.229293823242188,
-      "loss": 0.6922,
-      "rewards/accuracies": 0.5,
-      "rewards/chosen": 0.00649127596989274,
-      "rewards/margins": 0.008784732781350613,
-      "rewards/rejected": -0.0022934596054255962,
+      "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,
       "step": 370
     },
     {
       "epoch": 0.99,
       "learning_rate": 2.575864278703266e-10,
-      "logits/chosen": -1.929059386253357,
-      "logits/rejected": -1.931227445602417,
-      "logps/chosen": -34.235416412353516,
-      "logps/rejected": -30.882593154907227,
-      "loss": 0.6913,
-      "rewards/accuracies": 0.550000011920929,
-      "rewards/chosen": 0.0029094829224050045,
-      "rewards/margins": 0.010880164802074432,
-      "rewards/rejected": -0.00797068141400814,
+      "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,
       "step": 380
     },
     {
       "epoch": 1.0,
       "step": 385,
       "total_flos": 0.0,
-      "train_loss": 0.6928803004227675,
-      "train_runtime": 3255.7403,
-      "train_samples_per_second": 0.946,
+      "train_loss": 0.6916276120520257,
+      "train_runtime": 3257.6062,
+      "train_samples_per_second": 0.945,
       "train_steps_per_second": 0.118
     }
   ],

training_args.bin

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:dc7aa3a52fc0900b5ed63328278c9ca3dff8bd9d598f073a059f8c65d62bd827
+oid sha256:914e32cb321285a370aa5440b23d05260ae348cf6fed9440d5a1e634ebac11d9
 size 4984