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: 0.9988
-- Rewards/chosen: 0.0026
-- Rewards/rejected: 0.0013
-- Rewards/accuracies: 0.4925
-- Rewards/margins: 0.0013
-- Logps/rejected: -37.5141
-- Logps/chosen: -34.0293
-- Logits/rejected: -2.2384
-- Logits/chosen: -2.2433
+- Loss: 1.0020
+- Rewards/chosen: 0.0050
+- Rewards/rejected: 0.0068
+- Rewards/accuracies: 0.4635
+- Rewards/margins: -0.0019
+- Logps/rejected: -37.5052
+- Logps/chosen: -34.0263
+- 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 |
 |:-------------:|:-----:|:----:|:---------------:|:--------------:|:----------------:|:------------------:|:---------------:|:--------------:|:------------:|:---------------:|:-------------:|
-| 1.007         | 0.26  | 100  | 0.9963          | 0.0083         | 0.0045           | 0.5166             | 0.0039          | -37.5077       | -34.0179     | -2.2391         | -2.2440       |
-| 0.9725        | 0.52  | 200  | 1.0085          | 0.0008         | 0.0092           | 0.4992             | -0.0084         | -37.4982       | -34.0330     | -2.2385         | -2.2434       |
-| 0.982         | 0.78  | 300  | 1.0014          | 0.0061         | 0.0075           | 0.5104             | -0.0014         | -37.5016       | -34.0223     | -2.2389         | -2.2438       |
+| 1.0007        | 0.26  | 100  | 0.9989          | 0.0092         | 0.0079           | 0.5403             | 0.0013          | -37.5035       | -34.0192     | -2.2391         | -2.2440       |
+| 0.9898        | 0.52  | 200  | 0.9970          | 0.0124         | 0.0094           | 0.5307             | 0.0030          | -37.5009       | -34.0138     | -2.2391         | -2.2440       |
+| 0.9846        | 0.78  | 300  | 1.0020          | 0.0050         | 0.0068           | 0.4635             | -0.0019         | -37.5052       | -34.0263     | -2.2388         | -2.2437       |
 
 
 ### Framework versions

adapter_model.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:0f1a2ac2d3d83fe93faa89fedf9ab0b78af86ab25a03d190cceec07b1c464766
+oid sha256:51387effbb49cdadf9979c5a56ff61a06766e3483674b168cbccd22db7c79182
 size 176183216

all_results.json

@@ -1,20 +1,7 @@
 {
     "epoch": 1.0,
-    "eval_logits/chosen": -2.243324041366577,
-    "eval_logits/rejected": -2.2384440898895264,
-    "eval_logps/chosen": -34.02931213378906,
-    "eval_logps/rejected": -37.51406478881836,
-    "eval_loss": 0.9987560510635376,
-    "eval_rewards/accuracies": 0.4925249218940735,
-    "eval_rewards/chosen": 0.0026186639443039894,
-    "eval_rewards/margins": 0.0013388387160375714,
-    "eval_rewards/rejected": 0.0012798249954357743,
-    "eval_runtime": 145.7879,
-    "eval_samples": 343,
-    "eval_samples_per_second": 2.353,
-    "eval_steps_per_second": 0.295,
-    "train_loss": 0.9930156608680626,
-    "train_runtime": 3255.7181,
+    "train_loss": 0.9932638019710393,
+    "train_runtime": 3254.1177,
     "train_samples": 3079,
     "train_samples_per_second": 0.946,
     "train_steps_per_second": 0.118

train_results.json

@@ -1,7 +1,7 @@
 {
     "epoch": 1.0,
-    "train_loss": 0.9930156608680626,
-    "train_runtime": 3255.7181,
+    "train_loss": 0.9932638019710393,
+    "train_runtime": 3254.1177,
     "train_samples": 3079,
     "train_samples_per_second": 0.946,
     "train_steps_per_second": 0.118

trainer_state.json

@@ -25,589 +25,589 @@
     {
       "epoch": 0.03,
       "learning_rate": 1.2820512820512818e-07,
-      "logits/chosen": -1.8661351203918457,
-      "logits/rejected": -1.8704447746276855,
-      "logps/chosen": -36.994537353515625,
-      "logps/rejected": -33.65791320800781,
-      "loss": 0.98,
+      "logits/chosen": -1.866068720817566,
+      "logits/rejected": -1.8703863620758057,
+      "logps/chosen": -36.99105453491211,
+      "logps/rejected": -33.65992736816406,
+      "loss": 0.9728,
       "rewards/accuracies": 0.5138888955116272,
-      "rewards/chosen": 0.006032215431332588,
-      "rewards/margins": 0.019952965900301933,
-      "rewards/rejected": -0.013920750468969345,
+      "rewards/chosen": 0.009325359016656876,
+      "rewards/margins": 0.027240369468927383,
+      "rewards/rejected": -0.017915012314915657,
       "step": 10
     },
     {
       "epoch": 0.05,
       "learning_rate": 2.5641025641025636e-07,
-      "logits/chosen": -1.9972000122070312,
-      "logits/rejected": -1.9998512268066406,
-      "logps/chosen": -29.651025772094727,
-      "logps/rejected": -29.047237396240234,
-      "loss": 1.0125,
-      "rewards/accuracies": 0.42500001192092896,
-      "rewards/chosen": -0.004414811730384827,
-      "rewards/margins": -0.012516966089606285,
-      "rewards/rejected": 0.008102154359221458,
+      "logits/chosen": -1.9970887899398804,
+      "logits/rejected": -1.9997413158416748,
+      "logps/chosen": -29.63595199584961,
+      "logps/rejected": -29.070571899414062,
+      "loss": 0.992,
+      "rewards/accuracies": 0.5,
+      "rewards/chosen": 0.003744622226804495,
+      "rewards/margins": 0.008021654561161995,
+      "rewards/rejected": -0.004277031868696213,
       "step": 20
     },
     {
       "epoch": 0.08,
       "learning_rate": 3.8461538461538463e-07,
-      "logits/chosen": -1.919605016708374,
-      "logits/rejected": -1.9169189929962158,
-      "logps/chosen": -31.410541534423828,
-      "logps/rejected": -33.22766876220703,
-      "loss": 0.9926,
-      "rewards/accuracies": 0.550000011920929,
-      "rewards/chosen": 0.002776858163997531,
-      "rewards/margins": 0.007382071111351252,
-      "rewards/rejected": -0.004605212714523077,
+      "logits/chosen": -1.9195963144302368,
+      "logits/rejected": -1.9168994426727295,
+      "logps/chosen": -31.414175033569336,
+      "logps/rejected": -33.25041961669922,
+      "loss": 0.9797,
+      "rewards/accuracies": 0.5625,
+      "rewards/chosen": 0.0011531396303325891,
+      "rewards/margins": 0.02032935619354248,
+      "rewards/rejected": -0.019176214933395386,
       "step": 30
     },
     {
       "epoch": 0.1,
       "learning_rate": 4.999896948438433e-07,
-      "logits/chosen": -2.016796827316284,
-      "logits/rejected": -2.0080485343933105,
-      "logps/chosen": -32.59518814086914,
-      "logps/rejected": -32.49793243408203,
-      "loss": 1.0166,
+      "logits/chosen": -2.01688814163208,
+      "logits/rejected": -2.0081212520599365,
+      "logps/chosen": -32.583614349365234,
+      "logps/rejected": -32.48966979980469,
+      "loss": 1.0179,
       "rewards/accuracies": 0.4375,
-      "rewards/chosen": -0.00919206440448761,
-      "rewards/margins": -0.016562189906835556,
-      "rewards/rejected": 0.007370126433670521,
+      "rewards/chosen": -0.004085154738277197,
+      "rewards/margins": -0.017886672168970108,
+      "rewards/rejected": 0.013801517896354198,
       "step": 40
     },
     {
       "epoch": 0.13,
       "learning_rate": 4.987541037542186e-07,
-      "logits/chosen": -1.8643512725830078,
-      "logits/rejected": -1.853578805923462,
-      "logps/chosen": -33.55515670776367,
-      "logps/rejected": -35.426795959472656,
-      "loss": 1.0116,
-      "rewards/accuracies": 0.4749999940395355,
-      "rewards/chosen": 0.0011615634430199862,
-      "rewards/margins": -0.011613734066486359,
-      "rewards/rejected": 0.012775297276675701,
+      "logits/chosen": -1.8644500970840454,
+      "logits/rejected": -1.853655219078064,
+      "logps/chosen": -33.54804229736328,
+      "logps/rejected": -35.43767166137695,
+      "loss": 1.0031,
+      "rewards/accuracies": 0.4375,
+      "rewards/chosen": 0.005664288066327572,
+      "rewards/margins": -0.0031387731432914734,
+      "rewards/rejected": 0.00880306214094162,
       "step": 50
     },
     {
       "epoch": 0.16,
       "learning_rate": 4.954691471941118e-07,
-      "logits/chosen": -1.9452184438705444,
-      "logits/rejected": -1.9471704959869385,
-      "logps/chosen": -32.57328414916992,
-      "logps/rejected": -33.17564010620117,
-      "loss": 0.9844,
-      "rewards/accuracies": 0.5375000238418579,
-      "rewards/chosen": 0.013449115678668022,
-      "rewards/margins": 0.015578977763652802,
-      "rewards/rejected": -0.002129861619323492,
+      "logits/chosen": -1.9449504613876343,
+      "logits/rejected": -1.946890115737915,
+      "logps/chosen": -32.58049774169922,
+      "logps/rejected": -33.199378967285156,
+      "loss": 0.9714,
+      "rewards/accuracies": 0.612500011920929,
+      "rewards/chosen": 0.011811850592494011,
+      "rewards/margins": 0.02860717847943306,
+      "rewards/rejected": -0.016795331612229347,
       "step": 60
     },
     {
       "epoch": 0.18,
       "learning_rate": 4.901618883413548e-07,
-      "logits/chosen": -2.079322338104248,
-      "logits/rejected": -2.0843043327331543,
-      "logps/chosen": -33.998104095458984,
-      "logps/rejected": -36.586326599121094,
-      "loss": 0.9961,
-      "rewards/accuracies": 0.5625,
-      "rewards/chosen": -0.004809260368347168,
-      "rewards/margins": 0.003915563225746155,
-      "rewards/rejected": -0.008724823594093323,
+      "logits/chosen": -2.0795464515686035,
+      "logits/rejected": -2.0845184326171875,
+      "logps/chosen": -33.98798370361328,
+      "logps/rejected": -36.569732666015625,
+      "loss": 0.9992,
+      "rewards/accuracies": 0.48750001192092896,
+      "rewards/chosen": 0.0003018675488419831,
+      "rewards/margins": 0.0008206713828258216,
+      "rewards/rejected": -0.0005188033101148903,
       "step": 70
     },
     {
       "epoch": 0.21,
       "learning_rate": 4.828760511501322e-07,
-      "logits/chosen": -1.9419012069702148,
-      "logits/rejected": -1.9450546503067017,
-      "logps/chosen": -34.40644836425781,
-      "logps/rejected": -34.56013107299805,
-      "loss": 1.0055,
-      "rewards/accuracies": 0.5625,
-      "rewards/chosen": 0.010679833590984344,
-      "rewards/margins": -0.0054560392163693905,
-      "rewards/rejected": 0.016135873273015022,
+      "logits/chosen": -1.9425585269927979,
+      "logits/rejected": -1.945704698562622,
+      "logps/chosen": -34.408721923828125,
+      "logps/rejected": -34.581504821777344,
+      "loss": 0.9951,
+      "rewards/accuracies": 0.5375000238418579,
+      "rewards/chosen": 0.011452903971076012,
+      "rewards/margins": 0.0049109989777207375,
+      "rewards/rejected": 0.006541903130710125,
       "step": 80
     },
     {
       "epoch": 0.23,
       "learning_rate": 4.736716601303429e-07,
-      "logits/chosen": -1.9507659673690796,
-      "logits/rejected": -1.9552879333496094,
-      "logps/chosen": -32.45410919189453,
-      "logps/rejected": -32.34144973754883,
-      "loss": 1.0021,
-      "rewards/accuracies": 0.5249999761581421,
-      "rewards/chosen": 0.006657355930656195,
-      "rewards/margins": -0.002072322415187955,
-      "rewards/rejected": 0.008729678578674793,
+      "logits/chosen": -1.9505088329315186,
+      "logits/rejected": -1.9550243616104126,
+      "logps/chosen": -32.451255798339844,
+      "logps/rejected": -32.366790771484375,
+      "loss": 0.9856,
+      "rewards/accuracies": 0.512499988079071,
+      "rewards/chosen": 0.009697502478957176,
+      "rewards/margins": 0.014425436034798622,
+      "rewards/rejected": -0.0047279344871640205,
       "step": 90
     },
     {
       "epoch": 0.26,
       "learning_rate": 4.62624545834521e-07,
-      "logits/chosen": -2.0492916107177734,
-      "logits/rejected": -2.047295093536377,
-      "logps/chosen": -32.23640441894531,
-      "logps/rejected": -31.247751235961914,
-      "loss": 1.007,
-      "rewards/accuracies": 0.4375,
-      "rewards/chosen": 0.002216553781181574,
-      "rewards/margins": -0.007013450376689434,
-      "rewards/rejected": 0.009230004623532295,
+      "logits/chosen": -2.048947334289551,
+      "logits/rejected": -2.046935796737671,
+      "logps/chosen": -32.25363540649414,
+      "logps/rejected": -31.277917861938477,
+      "loss": 1.0007,
+      "rewards/accuracies": 0.44999998807907104,
+      "rewards/chosen": -0.007680465932935476,
+      "rewards/margins": -0.0006578800384886563,
+      "rewards/rejected": -0.007022587116807699,
       "step": 100
     },
     {
       "epoch": 0.26,
-      "eval_logits/chosen": -2.2440249919891357,
-      "eval_logits/rejected": -2.239149808883667,
-      "eval_logps/chosen": -34.017887115478516,
-      "eval_logps/rejected": -37.50767517089844,
-      "eval_loss": 0.9963483810424805,
-      "eval_rewards/accuracies": 0.5166113376617432,
-      "eval_rewards/chosen": 0.008332207798957825,
-      "eval_rewards/margins": 0.0038595683872699738,
-      "eval_rewards/rejected": 0.004472639411687851,
-      "eval_runtime": 146.1438,
-      "eval_samples_per_second": 2.347,
-      "eval_steps_per_second": 0.294,
+      "eval_logits/chosen": -2.243957042694092,
+      "eval_logits/rejected": -2.2390689849853516,
+      "eval_logps/chosen": -34.019203186035156,
+      "eval_logps/rejected": -37.50347900390625,
+      "eval_loss": 0.998858630657196,
+      "eval_rewards/accuracies": 0.5402824282646179,
+      "eval_rewards/chosen": 0.009209612384438515,
+      "eval_rewards/margins": 0.0013267018366605043,
+      "eval_rewards/rejected": 0.00788290984928608,
+      "eval_runtime": 145.9923,
+      "eval_samples_per_second": 2.349,
+      "eval_steps_per_second": 0.295,
       "step": 100
     },
     {
       "epoch": 0.29,
       "learning_rate": 4.4982572012636904e-07,
-      "logits/chosen": -2.005338191986084,
-      "logits/rejected": -2.002920627593994,
-      "logps/chosen": -33.22484588623047,
-      "logps/rejected": -34.01424026489258,
-      "loss": 0.9981,
-      "rewards/accuracies": 0.48750001192092896,
-      "rewards/chosen": 0.009485405869781971,
-      "rewards/margins": 0.001919901347719133,
-      "rewards/rejected": 0.007565504405647516,
+      "logits/chosen": -2.005302667617798,
+      "logits/rejected": -2.002894878387451,
+      "logps/chosen": -33.23944854736328,
+      "logps/rejected": -34.005104064941406,
+      "loss": 1.0119,
+      "rewards/accuracies": 0.44999998807907104,
+      "rewards/chosen": 0.002622986678034067,
+      "rewards/margins": -0.01194014959037304,
+      "rewards/rejected": 0.014563137665390968,
       "step": 110
     },
     {
       "epoch": 0.31,
       "learning_rate": 4.353806263777677e-07,
-      "logits/chosen": -2.0164568424224854,
-      "logits/rejected": -2.008070468902588,
-      "logps/chosen": -32.454193115234375,
-      "logps/rejected": -32.175106048583984,
-      "loss": 0.9978,
-      "rewards/accuracies": 0.4749999940395355,
-      "rewards/chosen": -0.004645171575248241,
-      "rewards/margins": 0.0021557719446718693,
-      "rewards/rejected": -0.006800943519920111,
+      "logits/chosen": -2.0167059898376465,
+      "logits/rejected": -2.0083398818969727,
+      "logps/chosen": -32.45713424682617,
+      "logps/rejected": -32.17597198486328,
+      "loss": 0.9987,
+      "rewards/accuracies": 0.44999998807907104,
+      "rewards/chosen": -0.0073361145332455635,
+      "rewards/margins": 0.0013455990701913834,
+      "rewards/rejected": -0.008681714534759521,
       "step": 120
     },
     {
       "epoch": 0.34,
       "learning_rate": 4.194082707715275e-07,
-      "logits/chosen": -2.046295642852783,
-      "logits/rejected": -2.03825044631958,
-      "logps/chosen": -30.491985321044922,
-      "logps/rejected": -32.04901885986328,
-      "loss": 1.0101,
-      "rewards/accuracies": 0.4625000059604645,
-      "rewards/chosen": -0.008134904317557812,
-      "rewards/margins": -0.010127933695912361,
-      "rewards/rejected": 0.001993028912693262,
+      "logits/chosen": -2.0464558601379395,
+      "logits/rejected": -2.038412094116211,
+      "logps/chosen": -30.482311248779297,
+      "logps/rejected": -32.04933166503906,
+      "loss": 1.0062,
+      "rewards/accuracies": 0.48750001192092896,
+      "rewards/chosen": -0.003956255037337542,
+      "rewards/margins": -0.006162940990179777,
+      "rewards/rejected": 0.002206685720011592,
       "step": 130
     },
     {
       "epoch": 0.36,
       "learning_rate": 4.020402418666621e-07,
-      "logits/chosen": -1.9772052764892578,
-      "logits/rejected": -1.9874699115753174,
-      "logps/chosen": -31.385913848876953,
-      "logps/rejected": -32.56010818481445,
-      "loss": 0.9705,
-      "rewards/accuracies": 0.5625,
-      "rewards/chosen": 0.016873350366950035,
-      "rewards/margins": 0.02948569692671299,
-      "rewards/rejected": -0.01261234562844038,
+      "logits/chosen": -1.9769861698150635,
+      "logits/rejected": -1.9872652292251587,
+      "logps/chosen": -31.415592193603516,
+      "logps/rejected": -32.5562858581543,
+      "loss": 0.9847,
+      "rewards/accuracies": 0.574999988079071,
+      "rewards/chosen": 0.0024404728319495916,
+      "rewards/margins": 0.015285758301615715,
+      "rewards/rejected": -0.01284528523683548,
       "step": 140
     },
     {
       "epoch": 0.39,
       "learning_rate": 3.8341962650351185e-07,
-      "logits/chosen": -1.8907018899917603,
-      "logits/rejected": -1.8917919397354126,
-      "logps/chosen": -34.191062927246094,
-      "logps/rejected": -34.757171630859375,
-      "loss": 0.9992,
-      "rewards/accuracies": 0.5,
-      "rewards/chosen": 0.0008138298871926963,
-      "rewards/margins": 0.0007720142602920532,
-      "rewards/rejected": 4.181563781457953e-05,
+      "logits/chosen": -1.890750527381897,
+      "logits/rejected": -1.8918195962905884,
+      "logps/chosen": -34.196632385253906,
+      "logps/rejected": -34.74388885498047,
+      "loss": 1.0104,
+      "rewards/accuracies": 0.4000000059604645,
+      "rewards/chosen": -0.002364098560065031,
+      "rewards/margins": -0.0103833619505167,
+      "rewards/rejected": 0.008019264787435532,
       "step": 150
     },
     {
       "epoch": 0.42,
       "learning_rate": 3.636998309800572e-07,
-      "logits/chosen": -1.9424282312393188,
-      "logits/rejected": -1.9389431476593018,
-      "logps/chosen": -36.157745361328125,
-      "logps/rejected": -32.73271942138672,
-      "loss": 0.988,
-      "rewards/accuracies": 0.5874999761581421,
-      "rewards/chosen": 0.0069676609709858894,
-      "rewards/margins": 0.011951583437621593,
-      "rewards/rejected": -0.004983922932296991,
+      "logits/chosen": -1.9424158334732056,
+      "logits/rejected": -1.9389365911483765,
+      "logps/chosen": -36.13062286376953,
+      "logps/rejected": -32.720314025878906,
+      "loss": 0.9768,
+      "rewards/accuracies": 0.637499988079071,
+      "rewards/chosen": 0.02463117241859436,
+      "rewards/margins": 0.023165332153439522,
+      "rewards/rejected": 0.0014658428262919188,
       "step": 160
     },
     {
       "epoch": 0.44,
       "learning_rate": 3.430433172111807e-07,
-      "logits/chosen": -2.0421340465545654,
-      "logits/rejected": -2.0347371101379395,
-      "logps/chosen": -33.798614501953125,
-      "logps/rejected": -31.370548248291016,
-      "loss": 0.991,
-      "rewards/accuracies": 0.5249999761581421,
-      "rewards/chosen": -0.00037657321081496775,
-      "rewards/margins": 0.008971447125077248,
-      "rewards/rejected": -0.00934801995754242,
+      "logits/chosen": -2.04191517829895,
+      "logits/rejected": -2.034518003463745,
+      "logps/chosen": -33.780067443847656,
+      "logps/rejected": -31.35748863220215,
+      "loss": 0.9859,
+      "rewards/accuracies": 0.5375000238418579,
+      "rewards/chosen": 0.010676576755940914,
+      "rewards/margins": 0.014060018584132195,
+      "rewards/rejected": -0.0033834422938525677,
       "step": 170
     },
     {
       "epoch": 0.47,
       "learning_rate": 3.216202642830543e-07,
-      "logits/chosen": -2.0471675395965576,
-      "logits/rejected": -2.0524353981018066,
-      "logps/chosen": -32.519779205322266,
-      "logps/rejected": -32.49123001098633,
-      "loss": 0.9845,
-      "rewards/accuracies": 0.48750001192092896,
-      "rewards/chosen": 0.012504413723945618,
-      "rewards/margins": 0.015548673458397388,
-      "rewards/rejected": -0.003044259501621127,
+      "logits/chosen": -2.0474770069122314,
+      "logits/rejected": -2.0527572631835938,
+      "logps/chosen": -32.51685333251953,
+      "logps/rejected": -32.48542022705078,
+      "loss": 0.9831,
+      "rewards/accuracies": 0.5,
+      "rewards/chosen": 0.01676076650619507,
+      "rewards/margins": 0.016928378492593765,
+      "rewards/rejected": -0.00016761067672632635,
       "step": 180
     },
     {
       "epoch": 0.49,
       "learning_rate": 2.9960716642946403e-07,
-      "logits/chosen": -2.0482120513916016,
-      "logits/rejected": -2.0454375743865967,
-      "logps/chosen": -31.485937118530273,
-      "logps/rejected": -31.332433700561523,
-      "loss": 0.9914,
-      "rewards/accuracies": 0.574999988079071,
-      "rewards/chosen": 0.0015064775943756104,
-      "rewards/margins": 0.008616042323410511,
-      "rewards/rejected": -0.007109564729034901,
+      "logits/chosen": -2.0486884117126465,
+      "logits/rejected": -2.045897960662842,
+      "logps/chosen": -31.490026473999023,
+      "logps/rejected": -31.336299896240234,
+      "loss": 0.9898,
+      "rewards/accuracies": 0.5,
+      "rewards/chosen": -0.0006444025784730911,
+      "rewards/margins": 0.010206506587564945,
+      "rewards/rejected": -0.010850909166038036,
       "step": 190
     },
     {
       "epoch": 0.52,
       "learning_rate": 2.771853789806683e-07,
-      "logits/chosen": -1.9187748432159424,
-      "logits/rejected": -1.9234533309936523,
-      "logps/chosen": -31.564071655273438,
-      "logps/rejected": -32.80377960205078,
-      "loss": 0.9725,
-      "rewards/accuracies": 0.625,
-      "rewards/chosen": 0.021104933694005013,
-      "rewards/margins": 0.027528975158929825,
-      "rewards/rejected": -0.006424039602279663,
+      "logits/chosen": -1.9185329675674438,
+      "logits/rejected": -1.9231984615325928,
+      "logps/chosen": -31.579875946044922,
+      "logps/rejected": -32.781455993652344,
+      "loss": 0.9898,
+      "rewards/accuracies": 0.574999988079071,
+      "rewards/chosen": 0.015846019610762596,
+      "rewards/margins": 0.010161412879824638,
+      "rewards/rejected": 0.005684606730937958,
       "step": 200
     },
     {
       "epoch": 0.52,
-      "eval_logits/chosen": -2.24338698387146,
-      "eval_logits/rejected": -2.2385094165802,
-      "eval_logps/chosen": -34.032955169677734,
-      "eval_logps/rejected": -37.49824142456055,
-      "eval_loss": 1.0085184574127197,
-      "eval_rewards/accuracies": 0.4991694688796997,
-      "eval_rewards/chosen": 0.0007996229687705636,
-      "eval_rewards/margins": -0.008389986120164394,
-      "eval_rewards/rejected": 0.009189609438180923,
-      "eval_runtime": 145.9067,
-      "eval_samples_per_second": 2.351,
+      "eval_logits/chosen": -2.2439582347869873,
+      "eval_logits/rejected": -2.239088296890259,
+      "eval_logps/chosen": -34.01380920410156,
+      "eval_logps/rejected": -37.500911712646484,
+      "eval_loss": 0.9969704151153564,
+      "eval_rewards/accuracies": 0.5307309031486511,
+      "eval_rewards/chosen": 0.01244510430842638,
+      "eval_rewards/margins": 0.0030194728169590235,
+      "eval_rewards/rejected": 0.009425631724298,
+      "eval_runtime": 145.8589,
+      "eval_samples_per_second": 2.352,
       "eval_steps_per_second": 0.295,
       "step": 200
     },
     {
       "epoch": 0.55,
       "learning_rate": 2.5453962426402e-07,
-      "logits/chosen": -2.0320358276367188,
-      "logits/rejected": -2.042719602584839,
-      "logps/chosen": -31.940053939819336,
-      "logps/rejected": -33.88956069946289,
-      "loss": 0.9709,
-      "rewards/accuracies": 0.5375000238418579,
-      "rewards/chosen": 0.015625199303030968,
-      "rewards/margins": 0.029055744409561157,
-      "rewards/rejected": -0.01343054324388504,
+      "logits/chosen": -2.0318081378936768,
+      "logits/rejected": -2.0424768924713135,
+      "logps/chosen": -31.95905113220215,
+      "logps/rejected": -33.87731170654297,
+      "loss": 0.9839,
+      "rewards/accuracies": 0.574999988079071,
+      "rewards/chosen": 0.007351105101406574,
+      "rewards/margins": 0.01611880213022232,
+      "rewards/rejected": -0.008767696097493172,
       "step": 210
     },
     {
       "epoch": 0.57,
       "learning_rate": 2.318564697655179e-07,
-      "logits/chosen": -1.9255237579345703,
-      "logits/rejected": -1.9403839111328125,
-      "logps/chosen": -30.091089248657227,
-      "logps/rejected": -31.54927635192871,
-      "loss": 0.9942,
-      "rewards/accuracies": 0.5,
-      "rewards/chosen": 0.010226013138890266,
-      "rewards/margins": 0.005776937119662762,
-      "rewards/rejected": 0.004449075553566217,
+      "logits/chosen": -1.9253685474395752,
+      "logits/rejected": -1.940233826637268,
+      "logps/chosen": -30.08770179748535,
+      "logps/rejected": -31.577301025390625,
+      "loss": 0.9742,
+      "rewards/accuracies": 0.612500011920929,
+      "rewards/chosen": 0.014303619973361492,
+      "rewards/margins": 0.025777745991945267,
+      "rewards/rejected": -0.011474122293293476,
       "step": 220
     },
     {
       "epoch": 0.6,
       "learning_rate": 2.093227910899832e-07,
-      "logits/chosen": -1.9833370447158813,
-      "logits/rejected": -1.9873158931732178,
-      "logps/chosen": -33.393131256103516,
-      "logps/rejected": -31.559478759765625,
-      "loss": 0.9827,
-      "rewards/accuracies": 0.512499988079071,
-      "rewards/chosen": 0.012168514542281628,
-      "rewards/margins": 0.017256852239370346,
-      "rewards/rejected": -0.005088338162750006,
+      "logits/chosen": -1.9832963943481445,
+      "logits/rejected": -1.9872655868530273,
+      "logps/chosen": -33.39131546020508,
+      "logps/rejected": -31.548086166381836,
+      "loss": 0.985,
+      "rewards/accuracies": 0.574999988079071,
+      "rewards/chosen": 0.015694385394454002,
+      "rewards/margins": 0.014964587986469269,
+      "rewards/rejected": 0.0007297966512851417,
       "step": 230
     },
     {
       "epoch": 0.62,
       "learning_rate": 1.8712423238279356e-07,
-      "logits/chosen": -1.9829031229019165,
-      "logits/rejected": -1.9609248638153076,
-      "logps/chosen": -34.161617279052734,
-      "logps/rejected": -34.973785400390625,
-      "loss": 0.9924,
-      "rewards/accuracies": 0.5,
-      "rewards/chosen": -0.003932595252990723,
-      "rewards/margins": 0.007563352584838867,
-      "rewards/rejected": -0.01149594783782959,
+      "logits/chosen": -1.9828859567642212,
+      "logits/rejected": -1.9609191417694092,
+      "logps/chosen": -34.16459655761719,
+      "logps/rejected": -34.95134735107422,
+      "loss": 1.0062,
+      "rewards/accuracies": 0.4124999940395355,
+      "rewards/chosen": -0.0065087564289569855,
+      "rewards/margins": -0.006178082898259163,
+      "rewards/rejected": -0.0003306727739982307,
       "step": 240
     },
     {
       "epoch": 0.65,
       "learning_rate": 1.654436768970182e-07,
-      "logits/chosen": -2.0244410037994385,
-      "logits/rejected": -2.0211281776428223,
-      "logps/chosen": -32.90896987915039,
-      "logps/rejected": -36.21443557739258,
-      "loss": 0.9972,
+      "logits/chosen": -2.0246593952178955,
+      "logits/rejected": -2.0213539600372314,
+      "logps/chosen": -32.9222412109375,
+      "logps/rejected": -36.228416442871094,
+      "loss": 0.9962,
       "rewards/accuracies": 0.4749999940395355,
-      "rewards/chosen": 0.008049068041145802,
-      "rewards/margins": 0.0028051852714270353,
-      "rewards/rejected": 0.0052438825368881226,
+      "rewards/chosen": 0.0016961356159299612,
+      "rewards/margins": 0.0037924889475107193,
+      "rewards/rejected": -0.0020963551942259073,
       "step": 250
     },
     {
       "epoch": 0.68,
       "learning_rate": 1.444597403062196e-07,
-      "logits/chosen": -1.8918297290802002,
-      "logits/rejected": -1.8893934488296509,
-      "logps/chosen": -34.178001403808594,
-      "logps/rejected": -35.52972412109375,
-      "loss": 0.9823,
-      "rewards/accuracies": 0.5874999761581421,
-      "rewards/chosen": 0.00519949197769165,
-      "rewards/margins": 0.017749812453985214,
-      "rewards/rejected": -0.012550321407616138,
+      "logits/chosen": -1.8918907642364502,
+      "logits/rejected": -1.8894577026367188,
+      "logps/chosen": -34.189327239990234,
+      "logps/rejected": -35.505496978759766,
+      "loss": 1.0,
+      "rewards/accuracies": 0.5,
+      "rewards/chosen": -0.0005569729837588966,
+      "rewards/margins": -2.999706157424953e-05,
+      "rewards/rejected": -0.0005269756657071412,
       "step": 260
     },
     {
       "epoch": 0.7,
       "learning_rate": 1.2434529917578887e-07,
-      "logits/chosen": -1.875998854637146,
-      "logits/rejected": -1.873445749282837,
-      "logps/chosen": -34.39997100830078,
-      "logps/rejected": -31.749908447265625,
-      "loss": 1.0119,
-      "rewards/accuracies": 0.4625000059604645,
-      "rewards/chosen": -0.0011831462616100907,
-      "rewards/margins": -0.011894017457962036,
-      "rewards/rejected": 0.01071087084710598,
+      "logits/chosen": -1.8758213520050049,
+      "logits/rejected": -1.8732786178588867,
+      "logps/chosen": -34.38149642944336,
+      "logps/rejected": -31.752059936523438,
+      "loss": 1.0019,
+      "rewards/accuracies": 0.550000011920929,
+      "rewards/chosen": 0.009660609066486359,
+      "rewards/margins": -0.0019017171580344439,
+      "rewards/rejected": 0.011562327854335308,
       "step": 270
     },
     {
       "epoch": 0.73,
       "learning_rate": 1.0526606671603521e-07,
-      "logits/chosen": -1.9797567129135132,
-      "logits/rejected": -1.9691368341445923,
-      "logps/chosen": -35.324066162109375,
-      "logps/rejected": -31.857070922851562,
-      "loss": 0.9757,
-      "rewards/accuracies": 0.5249999761581421,
-      "rewards/chosen": 0.016287770122289658,
-      "rewards/margins": 0.024326175451278687,
-      "rewards/rejected": -0.008038404397666454,
+      "logits/chosen": -1.9798027276992798,
+      "logits/rejected": -1.9691746234893799,
+      "logps/chosen": -35.328853607177734,
+      "logps/rejected": -31.868383407592773,
+      "loss": 0.9669,
+      "rewards/accuracies": 0.5375000238418579,
+      "rewards/chosen": 0.016671547666192055,
+      "rewards/margins": 0.03310702368617058,
+      "rewards/rejected": -0.016435474157333374,
       "step": 280
     },
     {
       "epoch": 0.75,
       "learning_rate": 8.737922755071453e-08,
-      "logits/chosen": -2.0752615928649902,
-      "logits/rejected": -2.06022310256958,
-      "logps/chosen": -30.904525756835938,
-      "logps/rejected": -32.66613006591797,
-      "loss": 0.9831,
-      "rewards/accuracies": 0.5249999761581421,
-      "rewards/chosen": 0.00980077963322401,
-      "rewards/margins": 0.016948198899626732,
-      "rewards/rejected": -0.007147419266402721,
+      "logits/chosen": -2.075648546218872,
+      "logits/rejected": -2.0605978965759277,
+      "logps/chosen": -30.926809310913086,
+      "logps/rejected": -32.621070861816406,
+      "loss": 1.0201,
+      "rewards/accuracies": 0.4124999940395355,
+      "rewards/chosen": -0.0016108205309137702,
+      "rewards/margins": -0.020070534199476242,
+      "rewards/rejected": 0.018459713086485863,
       "step": 290
     },
     {
       "epoch": 0.78,
       "learning_rate": 7.08321427484816e-08,
-      "logits/chosen": -1.946509599685669,
-      "logits/rejected": -1.9439618587493896,
-      "logps/chosen": -32.89145278930664,
-      "logps/rejected": -30.81719398498535,
-      "loss": 0.982,
-      "rewards/accuracies": 0.5375000238418579,
-      "rewards/chosen": 0.017829114571213722,
-      "rewards/margins": 0.018021035939455032,
-      "rewards/rejected": -0.00019192098989151418,
+      "logits/chosen": -1.9466993808746338,
+      "logits/rejected": -1.9441711902618408,
+      "logps/chosen": -32.90770721435547,
+      "logps/rejected": -30.82305335998535,
+      "loss": 0.9846,
+      "rewards/accuracies": 0.5249999761581421,
+      "rewards/chosen": 0.011643724516034126,
+      "rewards/margins": 0.015391023829579353,
+      "rewards/rejected": -0.0037472969852387905,
       "step": 300
     },
     {
       "epoch": 0.78,
-      "eval_logits/chosen": -2.2438137531280518,
-      "eval_logits/rejected": -2.2389373779296875,
-      "eval_logps/chosen": -34.02228546142578,
-      "eval_logps/rejected": -37.50155258178711,
-      "eval_loss": 1.0014283657073975,
-      "eval_rewards/accuracies": 0.5103820562362671,
-      "eval_rewards/chosen": 0.006133119110018015,
-      "eval_rewards/margins": -0.001401092391461134,
-      "eval_rewards/rejected": 0.007534211501479149,
-      "eval_runtime": 145.9391,
-      "eval_samples_per_second": 2.35,
+      "eval_logits/chosen": -2.243670701980591,
+      "eval_logits/rejected": -2.2387943267822266,
+      "eval_logps/chosen": -34.02626419067383,
+      "eval_logps/rejected": -37.50523376464844,
+      "eval_loss": 1.001994013786316,
+      "eval_rewards/accuracies": 0.4634551703929901,
+      "eval_rewards/chosen": 0.0049731116741895676,
+      "eval_rewards/margins": -0.0018590801628306508,
+      "eval_rewards/rejected": 0.006832191254943609,
+      "eval_runtime": 145.8336,
+      "eval_samples_per_second": 2.352,
       "eval_steps_per_second": 0.295,
       "step": 300
     },
     {
       "epoch": 0.81,
       "learning_rate": 5.576113578589034e-08,
-      "logits/chosen": -1.9287340641021729,
-      "logits/rejected": -1.9254703521728516,
-      "logps/chosen": -31.576568603515625,
-      "logps/rejected": -33.73060607910156,
-      "loss": 0.9924,
-      "rewards/accuracies": 0.5249999761581421,
-      "rewards/chosen": 0.013391390442848206,
-      "rewards/margins": 0.00761887151747942,
-      "rewards/rejected": 0.005772518925368786,
+      "logits/chosen": -1.9287983179092407,
+      "logits/rejected": -1.9255239963531494,
+      "logps/chosen": -31.573007583618164,
+      "logps/rejected": -33.742637634277344,
+      "loss": 0.9815,
+      "rewards/accuracies": 0.6000000238418579,
+      "rewards/chosen": 0.018206708133220673,
+      "rewards/margins": 0.018496429547667503,
+      "rewards/rejected": -0.0002897246740758419,
       "step": 310
     },
     {
       "epoch": 0.83,
       "learning_rate": 4.229036944380912e-08,
-      "logits/chosen": -1.980047583580017,
-      "logits/rejected": -1.9677263498306274,
-      "logps/chosen": -34.565635681152344,
-      "logps/rejected": -33.560508728027344,
-      "loss": 0.9739,
-      "rewards/accuracies": 0.5249999761581421,
-      "rewards/chosen": 0.015218675136566162,
-      "rewards/margins": 0.02605234459042549,
-      "rewards/rejected": -0.010833668522536755,
+      "logits/chosen": -1.9805524349212646,
+      "logits/rejected": -1.9682422876358032,
+      "logps/chosen": -34.57670211791992,
+      "logps/rejected": -33.561363220214844,
+      "loss": 0.9749,
+      "rewards/accuracies": 0.574999988079071,
+      "rewards/chosen": 0.011624744161963463,
+      "rewards/margins": 0.0251374039798975,
+      "rewards/rejected": -0.013512656092643738,
       "step": 320
     },
     {
       "epoch": 0.86,
       "learning_rate": 3.053082288996112e-08,
-      "logits/chosen": -2.0154004096984863,
-      "logits/rejected": -2.013948917388916,
-      "logps/chosen": -33.46211624145508,
-      "logps/rejected": -32.472103118896484,
-      "loss": 0.9919,
-      "rewards/accuracies": 0.5249999761581421,
-      "rewards/chosen": 0.011838942766189575,
-      "rewards/margins": 0.008100616745650768,
-      "rewards/rejected": 0.0037383257877081633,
+      "logits/chosen": -2.016091823577881,
+      "logits/rejected": -2.0146238803863525,
+      "logps/chosen": -33.47222137451172,
+      "logps/rejected": -32.46577835083008,
+      "loss": 1.0001,
+      "rewards/accuracies": 0.4749999940395355,
+      "rewards/chosen": 0.008143280632793903,
+      "rewards/margins": -0.0001393534184899181,
+      "rewards/rejected": 0.008282631635665894,
       "step": 330
     },
     {
       "epoch": 0.88,
       "learning_rate": 2.05793773749158e-08,
-      "logits/chosen": -2.1025614738464355,
-      "logits/rejected": -2.086764097213745,
-      "logps/chosen": -34.168495178222656,
-      "logps/rejected": -33.07737731933594,
-      "loss": 1.0149,
-      "rewards/accuracies": 0.42500001192092896,
-      "rewards/chosen": 0.005447447299957275,
-      "rewards/margins": -0.014881876297295094,
-      "rewards/rejected": 0.020329322665929794,
+      "logits/chosen": -2.1032156944274902,
+      "logits/rejected": -2.0874197483062744,
+      "logps/chosen": -34.174861907958984,
+      "logps/rejected": -33.088890075683594,
+      "loss": 1.0148,
+      "rewards/accuracies": 0.5249999761581421,
+      "rewards/chosen": 0.002716648392379284,
+      "rewards/margins": -0.014771336689591408,
+      "rewards/rejected": 0.017487986013293266,
       "step": 340
     },
     {
       "epoch": 0.91,
       "learning_rate": 1.251801807404168e-08,
-      "logits/chosen": -1.9740177392959595,
-      "logits/rejected": -1.9730663299560547,
-      "logps/chosen": -33.242271423339844,
-      "logps/rejected": -32.460182189941406,
-      "loss": 0.9826,
-      "rewards/accuracies": 0.550000011920929,
-      "rewards/chosen": 0.019913053140044212,
-      "rewards/margins": 0.017432674765586853,
-      "rewards/rejected": 0.0024803788401186466,
+      "logits/chosen": -1.9747329950332642,
+      "logits/rejected": -1.9737837314605713,
+      "logps/chosen": -33.25817108154297,
+      "logps/rejected": -32.459861755371094,
+      "loss": 0.9888,
+      "rewards/accuracies": 0.44999998807907104,
+      "rewards/chosen": 0.014356844127178192,
+      "rewards/margins": 0.01118617132306099,
+      "rewards/rejected": 0.003170671407133341,
       "step": 350
     },
     {
       "epoch": 0.94,
       "learning_rate": 6.41315865106129e-09,
-      "logits/chosen": -1.9304364919662476,
-      "logits/rejected": -1.9407978057861328,
-      "logps/chosen": -32.19029998779297,
-      "logps/rejected": -35.303955078125,
-      "loss": 0.9852,
-      "rewards/accuracies": 0.5874999761581421,
-      "rewards/chosen": 0.007193857338279486,
-      "rewards/margins": 0.014792281202971935,
-      "rewards/rejected": -0.00759842386469245,
+      "logits/chosen": -1.9303538799285889,
+      "logits/rejected": -1.94071364402771,
+      "logps/chosen": -32.20769500732422,
+      "logps/rejected": -35.273780822753906,
+      "loss": 1.0108,
+      "rewards/accuracies": 0.44999998807907104,
+      "rewards/chosen": -0.0018019669223576784,
+      "rewards/margins": -0.010786842554807663,
+      "rewards/rejected": 0.008984875865280628,
       "step": 360
     },
     {
       "epoch": 0.96,
       "learning_rate": 2.3150941078050324e-09,
-      "logits/chosen": -2.069343090057373,
-      "logits/rejected": -2.062788724899292,
-      "logps/chosen": -33.646358489990234,
-      "logps/rejected": -29.213809967041016,
-      "loss": 1.0035,
-      "rewards/accuracies": 0.4625000059604645,
-      "rewards/chosen": 0.0028047324158251286,
-      "rewards/margins": -0.0035038769710808992,
-      "rewards/rejected": 0.006308609154075384,
+      "logits/chosen": -2.069509744644165,
+      "logits/rejected": -2.0629658699035645,
+      "logps/chosen": -33.647117614746094,
+      "logps/rejected": -29.217060089111328,
+      "loss": 1.0027,
+      "rewards/accuracies": 0.48750001192092896,
+      "rewards/chosen": 0.002911838237196207,
+      "rewards/margins": -0.0027081891894340515,
+      "rewards/rejected": 0.005620025563985109,
       "step": 370
     },
     {
       "epoch": 0.99,
       "learning_rate": 2.575864278703266e-10,
-      "logits/chosen": -1.9290740489959717,
-      "logits/rejected": -1.9312365055084229,
-      "logps/chosen": -34.245155334472656,
-      "logps/rejected": -30.889087677001953,
-      "loss": 0.9948,
-      "rewards/accuracies": 0.574999988079071,
-      "rewards/chosen": -0.003053072141483426,
-      "rewards/margins": 0.005176535341888666,
-      "rewards/rejected": -0.008229607716202736,
+      "logits/chosen": -1.929449439048767,
+      "logits/rejected": -1.9315989017486572,
+      "logps/chosen": -34.25099563598633,
+      "logps/rejected": -30.906116485595703,
+      "loss": 0.9871,
+      "rewards/accuracies": 0.5249999761581421,
+      "rewards/chosen": -0.007165629416704178,
+      "rewards/margins": 0.012927901931107044,
+      "rewards/rejected": -0.020093530416488647,
       "step": 380
     },
     {
       "epoch": 1.0,
       "step": 385,
       "total_flos": 0.0,
-      "train_loss": 0.9930156608680626,
-      "train_runtime": 3255.7181,
+      "train_loss": 0.9932638019710393,
+      "train_runtime": 3254.1177,
       "train_samples_per_second": 0.946,
       "train_steps_per_second": 0.118
     }