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: 24.8075
-- Rewards/chosen: -0.0054
-- Rewards/rejected: -0.0092
-- Rewards/accuracies: 0.5133
-- Rewards/margins: 0.0038
-- Logps/rejected: -37.6085
-- Logps/chosen: -34.0885
-- Logits/rejected: -2.2234
-- Logits/chosen: -2.2282
+- Loss: 0.9961
+- Rewards/chosen: -0.0016
+- Rewards/rejected: -0.0055
+- Rewards/accuracies: 0.5594
+- Rewards/margins: 0.0039
+- Logps/rejected: -37.5717
+- Logps/chosen: -34.0508
+- Logits/rejected: -2.2195
+- Logits/chosen: -2.2243
 
 ## 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 |
 |:-------------:|:-----:|:----:|:---------------:|:--------------:|:----------------:|:------------------:|:---------------:|:--------------:|:------------:|:---------------:|:-------------:|
-| 23.9156       | 0.26  | 100  | 24.9323         | -0.0015        | -0.0033          | 0.5282             | 0.0018          | -37.5501       | -34.0500     | -2.2289         | -2.2338       |
-| 22.4879       | 0.52  | 200  | 24.7992         | -0.0025        | -0.0061          | 0.5598             | 0.0037          | -37.5780       | -34.0593     | -2.2258         | -2.2307       |
-| 19.8329       | 0.78  | 300  | 24.9557         | -0.0066        | -0.0089          | 0.5216             | 0.0023          | -37.6060       | -34.1010     | -2.2229         | -2.2278       |
+| 0.9851        | 0.26  | 100  | 0.9973          | 0.0013         | -0.0014          | 0.5311             | 0.0027          | -37.5310       | -34.0218     | -2.2257         | -2.2305       |
+| 0.9596        | 0.52  | 200  | 0.9944          | 0.0028         | -0.0028          | 0.5336             | 0.0056          | -37.5450       | -34.0069     | -2.2237         | -2.2285       |
+| 0.9305        | 0.78  | 300  | 0.9961          | -0.0016        | -0.0055          | 0.5594             | 0.0039          | -37.5717       | -34.0508     | -2.2195         | -2.2243       |
 
 
 ### Framework versions

adapter_model.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:076e169daae610676a71ed19e044c259301ea551232ff4f3bbefb889b8087358
+oid sha256:02f4916c6df444a6fa927d97ccaacdbdb852ca684cbeab3638897a126af5c4eb
 size 176183216

all_results.json

@@ -1,21 +1,8 @@
 {
     "epoch": 1.0,
-    "eval_logits/chosen": -2.228182792663574,
-    "eval_logits/rejected": -2.2233598232269287,
-    "eval_logps/chosen": -34.08854293823242,
-    "eval_logps/rejected": -37.60850143432617,
-    "eval_loss": 24.80753517150879,
-    "eval_rewards/accuracies": 0.5132890343666077,
-    "eval_rewards/chosen": -0.005399257410317659,
-    "eval_rewards/margins": 0.0037888663355261087,
-    "eval_rewards/rejected": -0.009188123047351837,
-    "eval_runtime": 145.6001,
-    "eval_samples": 343,
-    "eval_samples_per_second": 2.356,
-    "eval_steps_per_second": 0.295,
-    "train_loss": 22.846377404943688,
-    "train_runtime": 3252.1936,
+    "train_loss": 0.9734619264478808,
+    "train_runtime": 3253.307,
     "train_samples": 3079,
-    "train_samples_per_second": 0.947,
+    "train_samples_per_second": 0.946,
     "train_steps_per_second": 0.118
 }

train_results.json

@@ -1,8 +1,8 @@
 {
     "epoch": 1.0,
-    "train_loss": 22.846377404943688,
-    "train_runtime": 3252.1936,
+    "train_loss": 0.9734619264478808,
+    "train_runtime": 3253.307,
     "train_samples": 3079,
-    "train_samples_per_second": 0.947,
+    "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": 25.0,
+      "loss": 1.0,
       "rewards/accuracies": 0.0,
       "rewards/chosen": 0.0,
       "rewards/margins": 0.0,
@@ -25,155 +25,155 @@
     {
       "epoch": 0.03,
       "learning_rate": 1.282051282051282e-06,
-      "logits/chosen": -1.86653733253479,
-      "logits/rejected": -1.8708631992340088,
-      "logps/chosen": -36.98527526855469,
-      "logps/rejected": -33.67702865600586,
-      "loss": 24.351,
-      "rewards/accuracies": 0.5555555820465088,
-      "rewards/chosen": 0.0021321007516235113,
-      "rewards/margins": 0.006827179808169603,
-      "rewards/rejected": -0.004695079289376736,
+      "logits/chosen": -1.8667426109313965,
+      "logits/rejected": -1.8710602521896362,
+      "logps/chosen": -36.991912841796875,
+      "logps/rejected": -33.67206954956055,
+      "loss": 0.9943,
+      "rewards/accuracies": 0.5694444179534912,
+      "rewards/chosen": 0.0014687532093375921,
+      "rewards/margins": 0.005667862948030233,
+      "rewards/rejected": -0.004199109505861998,
       "step": 10
     },
     {
       "epoch": 0.05,
       "learning_rate": 2.564102564102564e-06,
-      "logits/chosen": -1.9982036352157593,
-      "logits/rejected": -2.0008480548858643,
-      "logps/chosen": -29.6201114654541,
-      "logps/rejected": -29.028329849243164,
-      "loss": 25.1673,
-      "rewards/accuracies": 0.42500001192092896,
-      "rewards/chosen": 0.002208721823990345,
-      "rewards/margins": -0.001302852644585073,
-      "rewards/rejected": 0.00351157458499074,
+      "logits/chosen": -1.9977442026138306,
+      "logits/rejected": -2.0003952980041504,
+      "logps/chosen": -29.659366607666016,
+      "logps/rejected": -29.05437660217285,
+      "loss": 1.0026,
+      "rewards/accuracies": 0.4124999940395355,
+      "rewards/chosen": -0.0017170545179396868,
+      "rewards/margins": -0.002623475855216384,
+      "rewards/rejected": 0.0009064216283150017,
       "step": 20
     },
     {
       "epoch": 0.08,
       "learning_rate": 3.846153846153847e-06,
-      "logits/chosen": -1.9204835891723633,
-      "logits/rejected": -1.917790174484253,
-      "logps/chosen": -31.419384002685547,
-      "logps/rejected": -33.234046936035156,
-      "loss": 24.926,
-      "rewards/accuracies": 0.5249999761581421,
-      "rewards/chosen": -0.00032898251083679497,
-      "rewards/margins": 0.001229552668519318,
-      "rewards/rejected": -0.0015585350338369608,
+      "logits/chosen": -1.920693039894104,
+      "logits/rejected": -1.91802179813385,
+      "logps/chosen": -31.39971351623535,
+      "logps/rejected": -33.21495819091797,
+      "loss": 0.9987,
+      "rewards/accuracies": 0.5625,
+      "rewards/chosen": 0.001637960784137249,
+      "rewards/margins": 0.0012880933936685324,
+      "rewards/rejected": 0.000349867797922343,
       "step": 30
     },
     {
       "epoch": 0.1,
       "learning_rate": 4.999896948438434e-06,
-      "logits/chosen": -2.0177228450775146,
-      "logits/rejected": -2.008981227874756,
-      "logps/chosen": -32.599449157714844,
-      "logps/rejected": -32.504905700683594,
-      "loss": 25.3506,
-      "rewards/accuracies": 0.42500001192092896,
-      "rewards/chosen": -0.0022641182877123356,
-      "rewards/margins": -0.0030405251309275627,
-      "rewards/rejected": 0.0007764073088765144,
+      "logits/chosen": -2.018057107925415,
+      "logits/rejected": -2.0093047618865967,
+      "logps/chosen": -32.565284729003906,
+      "logps/rejected": -32.50053405761719,
+      "loss": 1.0001,
+      "rewards/accuracies": 0.5375000238418579,
+      "rewards/chosen": 0.0011519074905663729,
+      "rewards/margins": -6.170915730763227e-05,
+      "rewards/rejected": 0.0012136166915297508,
       "step": 40
     },
     {
       "epoch": 0.13,
       "learning_rate": 4.987541037542187e-06,
-      "logits/chosen": -1.8622967004776,
-      "logits/rejected": -1.8515253067016602,
-      "logps/chosen": -33.548858642578125,
-      "logps/rejected": -35.451934814453125,
-      "loss": 24.9661,
-      "rewards/accuracies": 0.512499988079071,
-      "rewards/chosen": 0.0008619143627583981,
-      "rewards/margins": 0.0008210102096199989,
-      "rewards/rejected": 4.09040367230773e-05,
+      "logits/chosen": -1.8627817630767822,
+      "logits/rejected": -1.851999044418335,
+      "logps/chosen": -33.549964904785156,
+      "logps/rejected": -35.44340896606445,
+      "loss": 1.0001,
+      "rewards/accuracies": 0.4749999940395355,
+      "rewards/chosen": 0.0007515085162594914,
+      "rewards/margins": -0.00014221524179447442,
+      "rewards/rejected": 0.000893724150955677,
       "step": 50
     },
     {
       "epoch": 0.16,
       "learning_rate": 4.954691471941119e-06,
-      "logits/chosen": -1.941347360610962,
-      "logits/rejected": -1.9432964324951172,
-      "logps/chosen": -32.533042907714844,
-      "logps/rejected": -33.22543716430664,
-      "loss": 23.9783,
-      "rewards/accuracies": 0.6625000238418579,
-      "rewards/chosen": 0.006714025977998972,
-      "rewards/margins": 0.012119540013372898,
-      "rewards/rejected": -0.005405513569712639,
+      "logits/chosen": -1.9415528774261475,
+      "logits/rejected": -1.9434846639633179,
+      "logps/chosen": -32.53631591796875,
+      "logps/rejected": -33.215850830078125,
+      "loss": 0.9892,
+      "rewards/accuracies": 0.5625,
+      "rewards/chosen": 0.006386542227119207,
+      "rewards/margins": 0.010833840817213058,
+      "rewards/rejected": -0.004447298124432564,
       "step": 60
     },
     {
       "epoch": 0.18,
       "learning_rate": 4.901618883413549e-06,
-      "logits/chosen": -2.0721096992492676,
-      "logits/rejected": -2.077080249786377,
-      "logps/chosen": -33.98393249511719,
-      "logps/rejected": -36.645912170410156,
-      "loss": 24.3657,
+      "logits/chosen": -2.072315216064453,
+      "logits/rejected": -2.07728910446167,
+      "logps/chosen": -33.98168182373047,
+      "logps/rejected": -36.64748001098633,
+      "loss": 0.9915,
       "rewards/accuracies": 0.5625,
-      "rewards/chosen": 0.0004553424078039825,
-      "rewards/margins": 0.008159191347658634,
-      "rewards/rejected": -0.00770384818315506,
+      "rewards/chosen": 0.0006806664168834686,
+      "rewards/margins": 0.00854154396802187,
+      "rewards/rejected": -0.007860877551138401,
       "step": 70
     },
     {
       "epoch": 0.21,
       "learning_rate": 4.828760511501322e-06,
-      "logits/chosen": -1.9330637454986572,
-      "logits/rejected": -1.9362014532089233,
-      "logps/chosen": -34.31563949584961,
-      "logps/rejected": -34.64789581298828,
-      "loss": 23.5462,
-      "rewards/accuracies": 0.637499988079071,
-      "rewards/chosen": 0.011217002756893635,
-      "rewards/margins": 0.01676579937338829,
-      "rewards/rejected": -0.005548796150833368,
+      "logits/chosen": -1.932929277420044,
+      "logits/rejected": -1.9360615015029907,
+      "logps/chosen": -34.32741928100586,
+      "logps/rejected": -34.65910720825195,
+      "loss": 0.9833,
+      "rewards/accuracies": 0.625,
+      "rewards/chosen": 0.010038824751973152,
+      "rewards/margins": 0.01670856587588787,
+      "rewards/rejected": -0.006669741123914719,
       "step": 80
     },
     {
       "epoch": 0.23,
       "learning_rate": 4.7367166013034295e-06,
-      "logits/chosen": -1.9414646625518799,
-      "logits/rejected": -1.9459855556488037,
-      "logps/chosen": -32.386444091796875,
-      "logps/rejected": -32.36344528198242,
-      "loss": 24.298,
-      "rewards/accuracies": 0.637499988079071,
-      "rewards/chosen": 0.008097763173282146,
-      "rewards/margins": 0.008551515638828278,
-      "rewards/rejected": -0.0004537526401691139,
+      "logits/chosen": -1.9398362636566162,
+      "logits/rejected": -1.9443330764770508,
+      "logps/chosen": -32.37015914916992,
+      "logps/rejected": -32.354644775390625,
+      "loss": 0.9907,
+      "rewards/accuracies": 0.6625000238418579,
+      "rewards/chosen": 0.009726046584546566,
+      "rewards/margins": 0.00929970107972622,
+      "rewards/rejected": 0.0004263453301973641,
       "step": 90
     },
     {
       "epoch": 0.26,
       "learning_rate": 4.626245458345211e-06,
-      "logits/chosen": -2.038987636566162,
-      "logits/rejected": -2.03700590133667,
-      "logps/chosen": -32.161869049072266,
-      "logps/rejected": -31.31038475036621,
-      "loss": 23.9156,
-      "rewards/accuracies": 0.637499988079071,
-      "rewards/chosen": 0.007896887138485909,
-      "rewards/margins": 0.012314335443079472,
-      "rewards/rejected": -0.004417449701577425,
+      "logits/chosen": -2.036825656890869,
+      "logits/rejected": -2.0348496437072754,
+      "logps/chosen": -32.12454605102539,
+      "logps/rejected": -31.299402236938477,
+      "loss": 0.9851,
+      "rewards/accuracies": 0.625,
+      "rewards/chosen": 0.011629154905676842,
+      "rewards/margins": 0.014948204159736633,
+      "rewards/rejected": -0.003319049719721079,
       "step": 100
     },
     {
       "epoch": 0.26,
-      "eval_logits/chosen": -2.2337892055511475,
-      "eval_logits/rejected": -2.228935956954956,
-      "eval_logps/chosen": -34.049957275390625,
-      "eval_logps/rejected": -37.550052642822266,
-      "eval_loss": 24.932252883911133,
-      "eval_rewards/accuracies": 0.5282392501831055,
-      "eval_rewards/chosen": -0.001540520810522139,
-      "eval_rewards/margins": 0.0018024941673502326,
-      "eval_rewards/rejected": -0.0033430152107030153,
-      "eval_runtime": 145.9828,
+      "eval_logits/chosen": -2.230534791946411,
+      "eval_logits/rejected": -2.225694417953491,
+      "eval_logps/chosen": -34.02181625366211,
+      "eval_logps/rejected": -37.53096389770508,
+      "eval_loss": 0.9973288774490356,
+      "eval_rewards/accuracies": 0.531146228313446,
+      "eval_rewards/chosen": 0.0012735594063997269,
+      "eval_rewards/margins": 0.002707866718992591,
+      "eval_rewards/rejected": -0.0014343069633468986,
+      "eval_runtime": 145.9767,
       "eval_samples_per_second": 2.35,
       "eval_steps_per_second": 0.295,
       "step": 100
@@ -181,434 +181,434 @@
     {
       "epoch": 0.29,
       "learning_rate": 4.498257201263691e-06,
-      "logits/chosen": -1.9933013916015625,
-      "logits/rejected": -1.9909241199493408,
-      "logps/chosen": -33.13324737548828,
-      "logps/rejected": -34.06130599975586,
-      "loss": 24.1131,
+      "logits/chosen": -1.991271734237671,
+      "logits/rejected": -1.9888839721679688,
+      "logps/chosen": -33.108524322509766,
+      "logps/rejected": -33.99832534790039,
+      "loss": 0.9896,
       "rewards/accuracies": 0.612500011920929,
-      "rewards/chosen": 0.011056952178478241,
-      "rewards/margins": 0.01425047405064106,
-      "rewards/rejected": -0.003193522337824106,
+      "rewards/chosen": 0.013529380783438683,
+      "rewards/margins": 0.01042473316192627,
+      "rewards/rejected": 0.0031046485528349876,
       "step": 110
     },
     {
       "epoch": 0.31,
       "learning_rate": 4.353806263777678e-06,
-      "logits/chosen": -2.0050952434539795,
-      "logits/rejected": -1.9967548847198486,
-      "logps/chosen": -32.3184814453125,
-      "logps/rejected": -32.16117477416992,
-      "loss": 23.9805,
-      "rewards/accuracies": 0.6625000238418579,
-      "rewards/chosen": 0.012642547488212585,
-      "rewards/margins": 0.012609380297362804,
-      "rewards/rejected": 3.316625952720642e-05,
+      "logits/chosen": -2.0025076866149902,
+      "logits/rejected": -1.9941962957382202,
+      "logps/chosen": -32.313053131103516,
+      "logps/rejected": -32.1397705078125,
+      "loss": 0.989,
+      "rewards/accuracies": 0.6000000238418579,
+      "rewards/chosen": 0.013185250572860241,
+      "rewards/margins": 0.011011496186256409,
+      "rewards/rejected": 0.002173755317926407,
       "step": 120
     },
     {
       "epoch": 0.34,
       "learning_rate": 4.1940827077152755e-06,
-      "logits/chosen": -2.033102512359619,
-      "logits/rejected": -2.025144100189209,
-      "logps/chosen": -30.341167449951172,
-      "logps/rejected": -32.081993103027344,
-      "loss": 23.735,
-      "rewards/accuracies": 0.6499999761581421,
-      "rewards/chosen": 0.013454675674438477,
-      "rewards/margins": 0.016353728249669075,
-      "rewards/rejected": -0.0028990507125854492,
+      "logits/chosen": -2.030679941177368,
+      "logits/rejected": -2.022730827331543,
+      "logps/chosen": -30.313289642333984,
+      "logps/rejected": -32.024986267089844,
+      "loss": 0.9866,
+      "rewards/accuracies": 0.5249999761581421,
+      "rewards/chosen": 0.016243018209934235,
+      "rewards/margins": 0.013441057875752449,
+      "rewards/rejected": 0.00280195870436728,
       "step": 130
     },
     {
       "epoch": 0.36,
       "learning_rate": 4.0204024186666215e-06,
-      "logits/chosen": -1.9635530710220337,
-      "logits/rejected": -1.9737846851348877,
-      "logps/chosen": -31.221776962280273,
-      "logps/rejected": -32.5506477355957,
-      "loss": 23.1662,
-      "rewards/accuracies": 0.625,
-      "rewards/chosen": 0.019788404926657677,
-      "rewards/margins": 0.021364931017160416,
-      "rewards/rejected": -0.0015765223652124405,
+      "logits/chosen": -1.9600480794906616,
+      "logits/rejected": -1.9702802896499634,
+      "logps/chosen": -31.212570190429688,
+      "logps/rejected": -32.54707717895508,
+      "loss": 0.9781,
+      "rewards/accuracies": 0.637499988079071,
+      "rewards/chosen": 0.020709021016955376,
+      "rewards/margins": 0.02192843146622181,
+      "rewards/rejected": -0.0012194132432341576,
       "step": 140
     },
     {
       "epoch": 0.39,
       "learning_rate": 3.834196265035119e-06,
-      "logits/chosen": -1.875128984451294,
-      "logits/rejected": -1.876308798789978,
-      "logps/chosen": -33.92096710205078,
-      "logps/rejected": -34.793277740478516,
-      "loss": 22.4884,
-      "rewards/accuracies": 0.6000000238418579,
-      "rewards/chosen": 0.027172502130270004,
-      "rewards/margins": 0.030774494633078575,
-      "rewards/rejected": -0.00360199436545372,
+      "logits/chosen": -1.8716413974761963,
+      "logits/rejected": -1.8728001117706299,
+      "logps/chosen": -33.884464263916016,
+      "logps/rejected": -34.786155700683594,
+      "loss": 0.9663,
+      "rewards/accuracies": 0.625,
+      "rewards/chosen": 0.030822690576314926,
+      "rewards/margins": 0.03371291980147362,
+      "rewards/rejected": -0.0028902278281748295,
       "step": 150
     },
     {
       "epoch": 0.42,
       "learning_rate": 3.636998309800573e-06,
-      "logits/chosen": -1.9267343282699585,
-      "logits/rejected": -1.9233125448226929,
-      "logps/chosen": -36.03553009033203,
-      "logps/rejected": -32.72822570800781,
-      "loss": 23.8244,
-      "rewards/accuracies": 0.6499999761581421,
-      "rewards/chosen": 0.013614905066788197,
-      "rewards/margins": 0.014162260107696056,
-      "rewards/rejected": -0.0005473563214763999,
+      "logits/chosen": -1.923204779624939,
+      "logits/rejected": -1.9197826385498047,
+      "logps/chosen": -35.9670295715332,
+      "logps/rejected": -32.66301727294922,
+      "loss": 0.9855,
+      "rewards/accuracies": 0.625,
+      "rewards/chosen": 0.02046511135995388,
+      "rewards/margins": 0.014491458423435688,
+      "rewards/rejected": 0.00597365340217948,
       "step": 160
     },
     {
       "epoch": 0.44,
       "learning_rate": 3.4304331721118078e-06,
-      "logits/chosen": -2.0270447731018066,
-      "logits/rejected": -2.0197136402130127,
-      "logps/chosen": -33.48883819580078,
-      "logps/rejected": -31.42250633239746,
-      "loss": 21.7573,
+      "logits/chosen": -2.0242836475372314,
+      "logits/rejected": -2.0169827938079834,
+      "logps/chosen": -33.456817626953125,
+      "logps/rejected": -31.400043487548828,
+      "loss": 0.9611,
       "rewards/accuracies": 0.675000011920929,
-      "rewards/chosen": 0.030902724713087082,
-      "rewards/margins": 0.037968240678310394,
-      "rewards/rejected": -0.007065513636916876,
+      "rewards/chosen": 0.03410445898771286,
+      "rewards/margins": 0.038923610001802444,
+      "rewards/rejected": -0.004819151014089584,
       "step": 170
     },
     {
       "epoch": 0.47,
       "learning_rate": 3.2162026428305436e-06,
-      "logits/chosen": -2.0334360599517822,
-      "logits/rejected": -2.038677215576172,
-      "logps/chosen": -32.235355377197266,
-      "logps/rejected": -32.45377731323242,
-      "loss": 22.5681,
-      "rewards/accuracies": 0.6875,
-      "rewards/chosen": 0.030943090096116066,
-      "rewards/margins": 0.027806812897324562,
-      "rewards/rejected": 0.0031362746376544237,
+      "logits/chosen": -2.0301671028137207,
+      "logits/rejected": -2.0354058742523193,
+      "logps/chosen": -32.161399841308594,
+      "logps/rejected": -32.41175842285156,
+      "loss": 0.969,
+      "rewards/accuracies": 0.7124999761581421,
+      "rewards/chosen": 0.03833850100636482,
+      "rewards/margins": 0.031000157818198204,
+      "rewards/rejected": 0.007338343653827906,
       "step": 180
     },
     {
       "epoch": 0.49,
       "learning_rate": 2.996071664294641e-06,
-      "logits/chosen": -2.034454345703125,
-      "logits/rejected": -2.03169846534729,
-      "logps/chosen": -31.2611141204834,
-      "logps/rejected": -31.35662841796875,
-      "loss": 22.7646,
-      "rewards/accuracies": 0.675000011920929,
-      "rewards/chosen": 0.022784242406487465,
-      "rewards/margins": 0.026625871658325195,
-      "rewards/rejected": -0.0038416311144828796,
+      "logits/chosen": -2.0318570137023926,
+      "logits/rejected": -2.0290966033935547,
+      "logps/chosen": -31.263225555419922,
+      "logps/rejected": -31.29646873474121,
+      "loss": 0.9796,
+      "rewards/accuracies": 0.6000000238418579,
+      "rewards/chosen": 0.022572491317987442,
+      "rewards/margins": 0.0203980952501297,
+      "rewards/rejected": 0.0021743960678577423,
       "step": 190
     },
     {
       "epoch": 0.52,
       "learning_rate": 2.7718537898066833e-06,
-      "logits/chosen": -1.9043190479278564,
-      "logits/rejected": -1.9089492559432983,
-      "logps/chosen": -31.331798553466797,
-      "logps/rejected": -32.8171501159668,
-      "loss": 22.4879,
-      "rewards/accuracies": 0.699999988079071,
-      "rewards/chosen": 0.027448315173387527,
-      "rewards/margins": 0.030070018023252487,
-      "rewards/rejected": -0.002621703315526247,
+      "logits/chosen": -1.9017003774642944,
+      "logits/rejected": -1.906355619430542,
+      "logps/chosen": -31.227977752685547,
+      "logps/rejected": -32.81672286987305,
+      "loss": 0.9596,
+      "rewards/accuracies": 0.75,
+      "rewards/chosen": 0.03783063963055611,
+      "rewards/margins": 0.0404098778963089,
+      "rewards/rejected": -0.002579244552180171,
       "step": 200
     },
     {
       "epoch": 0.52,
-      "eval_logits/chosen": -2.2306509017944336,
-      "eval_logits/rejected": -2.2258191108703613,
-      "eval_logps/chosen": -34.05928039550781,
-      "eval_logps/rejected": -37.57795333862305,
-      "eval_loss": 24.79918098449707,
-      "eval_rewards/accuracies": 0.5598006844520569,
-      "eval_rewards/chosen": -0.0024726546835154295,
-      "eval_rewards/margins": 0.0036605680361390114,
-      "eval_rewards/rejected": -0.006133222486823797,
-      "eval_runtime": 145.6285,
-      "eval_samples_per_second": 2.355,
+      "eval_logits/chosen": -2.2284843921661377,
+      "eval_logits/rejected": -2.2236602306365967,
+      "eval_logps/chosen": -34.00685501098633,
+      "eval_logps/rejected": -37.54502868652344,
+      "eval_loss": 0.9944055080413818,
+      "eval_rewards/accuracies": 0.5336378812789917,
+      "eval_rewards/chosen": 0.0027695323806256056,
+      "eval_rewards/margins": 0.005610140040516853,
+      "eval_rewards/rejected": -0.002840606961399317,
+      "eval_runtime": 145.7295,
+      "eval_samples_per_second": 2.354,
       "eval_steps_per_second": 0.295,
       "step": 200
     },
     {
       "epoch": 0.55,
       "learning_rate": 2.5453962426402006e-06,
-      "logits/chosen": -2.0171916484832764,
-      "logits/rejected": -2.027843952178955,
-      "logps/chosen": -31.748519897460938,
-      "logps/rejected": -33.94535827636719,
-      "loss": 22.396,
-      "rewards/accuracies": 0.7124999761581421,
-      "rewards/chosen": 0.022278117015957832,
-      "rewards/margins": 0.030543917790055275,
-      "rewards/rejected": -0.008265801705420017,
+      "logits/chosen": -2.0140950679779053,
+      "logits/rejected": -2.024714946746826,
+      "logps/chosen": -31.731609344482422,
+      "logps/rejected": -33.924285888671875,
+      "loss": 0.9699,
+      "rewards/accuracies": 0.625,
+      "rewards/chosen": 0.02396966516971588,
+      "rewards/margins": 0.030128711834549904,
+      "rewards/rejected": -0.006159046199172735,
       "step": 210
     },
     {
       "epoch": 0.57,
       "learning_rate": 2.3185646976551794e-06,
-      "logits/chosen": -1.9094616174697876,
-      "logits/rejected": -1.9241745471954346,
-      "logps/chosen": -29.818714141845703,
-      "logps/rejected": -31.6313533782959,
-      "loss": 21.8415,
-      "rewards/accuracies": 0.737500011920929,
-      "rewards/chosen": 0.029282480478286743,
-      "rewards/margins": 0.0366000160574913,
-      "rewards/rejected": -0.007317539304494858,
+      "logits/chosen": -1.9061514139175415,
+      "logits/rejected": -1.9208866357803345,
+      "logps/chosen": -29.77530288696289,
+      "logps/rejected": -31.6021728515625,
+      "loss": 0.962,
+      "rewards/accuracies": 0.699999988079071,
+      "rewards/chosen": 0.03362376615405083,
+      "rewards/margins": 0.038023434579372406,
+      "rewards/rejected": -0.004399660509079695,
       "step": 220
     },
     {
       "epoch": 0.6,
       "learning_rate": 2.0932279108998323e-06,
-      "logits/chosen": -1.9662542343139648,
-      "logits/rejected": -1.9702177047729492,
-      "logps/chosen": -33.06914138793945,
-      "logps/rejected": -31.627248764038086,
-      "loss": 21.4931,
-      "rewards/accuracies": 0.7250000238418579,
-      "rewards/chosen": 0.03483276441693306,
-      "rewards/margins": 0.042627494782209396,
-      "rewards/rejected": -0.007794731762260199,
+      "logits/chosen": -1.9625619649887085,
+      "logits/rejected": -1.9665281772613525,
+      "logps/chosen": -33.06541061401367,
+      "logps/rejected": -31.63521385192871,
+      "loss": 0.9562,
+      "rewards/accuracies": 0.6625000238418579,
+      "rewards/chosen": 0.0352059006690979,
+      "rewards/margins": 0.04379696026444435,
+      "rewards/rejected": -0.008591057732701302,
       "step": 230
     },
     {
       "epoch": 0.62,
       "learning_rate": 1.8712423238279358e-06,
-      "logits/chosen": -1.964835524559021,
-      "logits/rejected": -1.9430347681045532,
-      "logps/chosen": -33.808719635009766,
-      "logps/rejected": -35.10743713378906,
-      "loss": 20.8354,
-      "rewards/accuracies": 0.7124999761581421,
-      "rewards/chosen": 0.034502916038036346,
-      "rewards/margins": 0.05016731098294258,
-      "rewards/rejected": -0.015664398670196533,
+      "logits/chosen": -1.960546851158142,
+      "logits/rejected": -1.938780426979065,
+      "logps/chosen": -33.803611755371094,
+      "logps/rejected": -35.09512710571289,
+      "loss": 0.9506,
+      "rewards/accuracies": 0.675000011920929,
+      "rewards/chosen": 0.03501402214169502,
+      "rewards/margins": 0.049447186291217804,
+      "rewards/rejected": -0.01443316787481308,
       "step": 240
     },
     {
       "epoch": 0.65,
       "learning_rate": 1.6544367689701824e-06,
-      "logits/chosen": -2.0063602924346924,
-      "logits/rejected": -2.0030438899993896,
-      "logps/chosen": -32.721397399902344,
-      "logps/rejected": -36.288963317871094,
-      "loss": 22.7702,
-      "rewards/accuracies": 0.612500011920929,
-      "rewards/chosen": 0.020366890355944633,
-      "rewards/margins": 0.02677120640873909,
-      "rewards/rejected": -0.006404316984117031,
+      "logits/chosen": -2.002559185028076,
+      "logits/rejected": -1.9992578029632568,
+      "logps/chosen": -32.70106506347656,
+      "logps/rejected": -36.2247428894043,
+      "loss": 0.9776,
+      "rewards/accuracies": 0.6499999761581421,
+      "rewards/chosen": 0.0224003903567791,
+      "rewards/margins": 0.022382449358701706,
+      "rewards/rejected": 1.793876253941562e-05,
       "step": 250
     },
     {
       "epoch": 0.68,
       "learning_rate": 1.4445974030621963e-06,
-      "logits/chosen": -1.8736040592193604,
-      "logits/rejected": -1.8712146282196045,
-      "logps/chosen": -33.984458923339844,
-      "logps/rejected": -35.55829620361328,
-      "loss": 22.8113,
-      "rewards/accuracies": 0.737500011920929,
-      "rewards/chosen": 0.020393703132867813,
-      "rewards/margins": 0.02576116845011711,
-      "rewards/rejected": -0.0053674690425395966,
+      "logits/chosen": -1.8695703744888306,
+      "logits/rejected": -1.867144227027893,
+      "logps/chosen": -33.96361541748047,
+      "logps/rejected": -35.52811050415039,
+      "loss": 0.9752,
+      "rewards/accuracies": 0.6625000238418579,
+      "rewards/chosen": 0.022478096187114716,
+      "rewards/margins": 0.024826791137456894,
+      "rewards/rejected": -0.0023486947175115347,
       "step": 260
     },
     {
       "epoch": 0.7,
       "learning_rate": 1.243452991757889e-06,
-      "logits/chosen": -1.8587570190429688,
-      "logits/rejected": -1.8563568592071533,
-      "logps/chosen": -34.19550704956055,
-      "logps/rejected": -31.841039657592773,
-      "loss": 22.7599,
-      "rewards/accuracies": 0.6875,
-      "rewards/chosen": 0.020208846777677536,
-      "rewards/margins": 0.027179840952157974,
-      "rewards/rejected": -0.00697099044919014,
+      "logits/chosen": -1.8547470569610596,
+      "logits/rejected": -1.8523423671722412,
+      "logps/chosen": -34.13903045654297,
+      "logps/rejected": -31.83770179748535,
+      "loss": 0.9675,
+      "rewards/accuracies": 0.6625000238418579,
+      "rewards/chosen": 0.02585705555975437,
+      "rewards/margins": 0.03249421715736389,
+      "rewards/rejected": -0.006637162063270807,
       "step": 270
     },
     {
       "epoch": 0.73,
       "learning_rate": 1.0526606671603523e-06,
-      "logits/chosen": -1.9618046283721924,
-      "logits/rejected": -1.9513158798217773,
-      "logps/chosen": -34.9988899230957,
-      "logps/rejected": -31.8876895904541,
-      "loss": 21.4726,
-      "rewards/accuracies": 0.699999988079071,
-      "rewards/chosen": 0.03577522560954094,
-      "rewards/margins": 0.040445104241371155,
-      "rewards/rejected": -0.0046698772348463535,
+      "logits/chosen": -1.9572479724884033,
+      "logits/rejected": -1.9467785358428955,
+      "logps/chosen": -34.97367858886719,
+      "logps/rejected": -31.858196258544922,
+      "loss": 0.96,
+      "rewards/accuracies": 0.7124999761581421,
+      "rewards/chosen": 0.03829622268676758,
+      "rewards/margins": 0.04001673310995102,
+      "rewards/rejected": -0.001720509259030223,
       "step": 280
     },
     {
       "epoch": 0.75,
       "learning_rate": 8.737922755071455e-07,
-      "logits/chosen": -2.057340621948242,
-      "logits/rejected": -2.042445659637451,
-      "logps/chosen": -30.718120574951172,
-      "logps/rejected": -32.65331268310547,
-      "loss": 23.4433,
+      "logits/chosen": -2.0524978637695312,
+      "logits/rejected": -2.0376133918762207,
+      "logps/chosen": -30.694583892822266,
+      "logps/rejected": -32.60517501831055,
+      "loss": 0.9817,
       "rewards/accuracies": 0.612500011920929,
-      "rewards/chosen": 0.020600156858563423,
-      "rewards/margins": 0.020748335868120193,
-      "rewards/rejected": -0.00014818087220191956,
+      "rewards/chosen": 0.022954344749450684,
+      "rewards/margins": 0.01828841306269169,
+      "rewards/rejected": 0.004665931686758995,
       "step": 290
     },
     {
       "epoch": 0.78,
       "learning_rate": 7.08321427484816e-07,
-      "logits/chosen": -1.928430199623108,
-      "logits/rejected": -1.9259220361709595,
-      "logps/chosen": -32.38062286376953,
-      "logps/rejected": -30.925670623779297,
-      "loss": 19.8329,
-      "rewards/accuracies": 0.800000011920929,
-      "rewards/chosen": 0.05464861914515495,
-      "rewards/margins": 0.06553500890731812,
-      "rewards/rejected": -0.010886380448937416,
+      "logits/chosen": -1.923266053199768,
+      "logits/rejected": -1.9207680225372314,
+      "logps/chosen": -32.310142517089844,
+      "logps/rejected": -30.895198822021484,
+      "loss": 0.9305,
+      "rewards/accuracies": 0.762499988079071,
+      "rewards/chosen": 0.061696797609329224,
+      "rewards/margins": 0.06953591853380203,
+      "rewards/rejected": -0.007839125581085682,
       "step": 300
     },
     {
       "epoch": 0.78,
-      "eval_logits/chosen": -2.227757215499878,
-      "eval_logits/rejected": -2.222930908203125,
-      "eval_logps/chosen": -34.10100555419922,
-      "eval_logps/rejected": -37.605979919433594,
-      "eval_loss": 24.95565414428711,
-      "eval_rewards/accuracies": 0.5215947031974792,
-      "eval_rewards/chosen": -0.006645896937698126,
-      "eval_rewards/margins": 0.0022900840267539024,
-      "eval_rewards/rejected": -0.008935980498790741,
-      "eval_runtime": 145.783,
-      "eval_samples_per_second": 2.353,
+      "eval_logits/chosen": -2.224316358566284,
+      "eval_logits/rejected": -2.2194955348968506,
+      "eval_logps/chosen": -34.0507698059082,
+      "eval_logps/rejected": -37.57167053222656,
+      "eval_loss": 0.9961203336715698,
+      "eval_rewards/accuracies": 0.559385359287262,
+      "eval_rewards/chosen": -0.001622139709070325,
+      "eval_rewards/margins": 0.0038830821868032217,
+      "eval_rewards/rejected": -0.005505221430212259,
+      "eval_runtime": 145.6991,
+      "eval_samples_per_second": 2.354,
       "eval_steps_per_second": 0.295,
       "step": 300
     },
     {
       "epoch": 0.81,
       "learning_rate": 5.576113578589035e-07,
-      "logits/chosen": -1.9124844074249268,
-      "logits/rejected": -1.9092193841934204,
-      "logps/chosen": -31.315048217773438,
-      "logps/rejected": -33.801414489746094,
-      "loss": 22.1909,
+      "logits/chosen": -1.9086971282958984,
+      "logits/rejected": -1.9054778814315796,
+      "logps/chosen": -31.2874755859375,
+      "logps/rejected": -33.80743408203125,
+      "loss": 0.9619,
       "rewards/accuracies": 0.7250000238418579,
-      "rewards/chosen": 0.028830338269472122,
-      "rewards/margins": 0.034757114946842194,
-      "rewards/rejected": -0.005926776677370071,
+      "rewards/chosen": 0.03158777207136154,
+      "rewards/margins": 0.03811599686741829,
+      "rewards/rejected": -0.006528225727379322,
       "step": 310
     },
     {
       "epoch": 0.83,
       "learning_rate": 4.229036944380913e-07,
-      "logits/chosen": -1.963141679763794,
-      "logits/rejected": -1.9509350061416626,
-      "logps/chosen": -34.32793426513672,
-      "logps/rejected": -33.663185119628906,
-      "loss": 21.758,
-      "rewards/accuracies": 0.6625000238418579,
-      "rewards/chosen": 0.026814326643943787,
-      "rewards/margins": 0.03924902528524399,
-      "rewards/rejected": -0.012434699572622776,
+      "logits/chosen": -1.958402395248413,
+      "logits/rejected": -1.946244478225708,
+      "logps/chosen": -34.26841735839844,
+      "logps/rejected": -33.66584014892578,
+      "loss": 0.9545,
+      "rewards/accuracies": 0.699999988079071,
+      "rewards/chosen": 0.03276572749018669,
+      "rewards/margins": 0.04546588286757469,
+      "rewards/rejected": -0.012700155377388,
       "step": 320
     },
     {
       "epoch": 0.86,
       "learning_rate": 3.053082288996112e-07,
-      "logits/chosen": -1.9982473850250244,
-      "logits/rejected": -1.9968185424804688,
-      "logps/chosen": -33.15186309814453,
-      "logps/rejected": -32.56055450439453,
-      "loss": 21.5888,
-      "rewards/accuracies": 0.7250000238418579,
-      "rewards/chosen": 0.03339282050728798,
-      "rewards/margins": 0.0414896085858345,
-      "rewards/rejected": -0.008096789941191673,
+      "logits/chosen": -1.9934791326522827,
+      "logits/rejected": -1.9920657873153687,
+      "logps/chosen": -33.07835006713867,
+      "logps/rejected": -32.51830291748047,
+      "loss": 0.9554,
+      "rewards/accuracies": 0.75,
+      "rewards/chosen": 0.040744177997112274,
+      "rewards/margins": 0.0446164533495903,
+      "rewards/rejected": -0.0038722690660506487,
       "step": 330
     },
     {
       "epoch": 0.88,
       "learning_rate": 2.0579377374915805e-07,
-      "logits/chosen": -2.084803819656372,
-      "logits/rejected": -2.069124698638916,
-      "logps/chosen": -33.807334899902344,
-      "logps/rejected": -33.127708435058594,
-      "loss": 21.7884,
-      "rewards/accuracies": 0.7124999761581421,
-      "rewards/chosen": 0.03720540553331375,
-      "rewards/margins": 0.0381726399064064,
-      "rewards/rejected": -0.0009672341984696686,
+      "logits/chosen": -2.0803451538085938,
+      "logits/rejected": -2.064704418182373,
+      "logps/chosen": -33.757957458496094,
+      "logps/rejected": -33.0684700012207,
+      "loss": 0.9628,
+      "rewards/accuracies": 0.675000011920929,
+      "rewards/chosen": 0.0421435609459877,
+      "rewards/margins": 0.037186361849308014,
+      "rewards/rejected": 0.004957201890647411,
       "step": 340
     },
     {
       "epoch": 0.91,
       "learning_rate": 1.2518018074041684e-07,
-      "logits/chosen": -1.9572391510009766,
-      "logits/rejected": -1.9564039707183838,
-      "logps/chosen": -32.83594512939453,
-      "logps/rejected": -32.575740814208984,
-      "loss": 20.605,
-      "rewards/accuracies": 0.7250000238418579,
-      "rewards/chosen": 0.04461529105901718,
-      "rewards/margins": 0.05567503720521927,
-      "rewards/rejected": -0.011059742420911789,
+      "logits/chosen": -1.9529346227645874,
+      "logits/rejected": -1.9521318674087524,
+      "logps/chosen": -32.80330276489258,
+      "logps/rejected": -32.498817443847656,
+      "loss": 0.9488,
+      "rewards/accuracies": 0.6875,
+      "rewards/chosen": 0.04787950962781906,
+      "rewards/margins": 0.051246631890535355,
+      "rewards/rejected": -0.0033671148121356964,
       "step": 350
     },
     {
       "epoch": 0.94,
       "learning_rate": 6.41315865106129e-08,
-      "logits/chosen": -1.9132989645004272,
-      "logits/rejected": -1.923563003540039,
-      "logps/chosen": -31.88213539123535,
-      "logps/rejected": -35.31838607788086,
-      "loss": 22.111,
+      "logits/chosen": -1.908622145652771,
+      "logits/rejected": -1.9189058542251587,
+      "logps/chosen": -31.85636329650879,
+      "logps/rejected": -35.29613494873047,
+      "loss": 0.9644,
       "rewards/accuracies": 0.6625000238418579,
-      "rewards/chosen": 0.03225552663207054,
-      "rewards/margins": 0.03521796688437462,
-      "rewards/rejected": -0.0029624411836266518,
+      "rewards/chosen": 0.03483304753899574,
+      "rewards/margins": 0.03557039424777031,
+      "rewards/rejected": -0.0007373450207524002,
       "step": 360
     },
     {
       "epoch": 0.96,
       "learning_rate": 2.3150941078050325e-08,
-      "logits/chosen": -2.0528197288513184,
-      "logits/rejected": -2.046360492706299,
-      "logps/chosen": -33.33294677734375,
-      "logps/rejected": -29.250751495361328,
-      "loss": 22.0205,
-      "rewards/accuracies": 0.6875,
-      "rewards/chosen": 0.03190212696790695,
-      "rewards/margins": 0.03433450311422348,
-      "rewards/rejected": -0.0024323747493326664,
+      "logits/chosen": -2.048346757888794,
+      "logits/rejected": -2.041914939880371,
+      "logps/chosen": -33.28711700439453,
+      "logps/rejected": -29.241107940673828,
+      "loss": 0.962,
+      "rewards/accuracies": 0.7124999761581421,
+      "rewards/chosen": 0.03648493438959122,
+      "rewards/margins": 0.03795299679040909,
+      "rewards/rejected": -0.001468065194785595,
       "step": 370
     },
     {
       "epoch": 0.99,
       "learning_rate": 2.575864278703266e-09,
-      "logits/chosen": -1.9122035503387451,
-      "logits/rejected": -1.9144351482391357,
-      "logps/chosen": -33.86780548095703,
-      "logps/rejected": -30.96420669555664,
-      "loss": 21.1141,
-      "rewards/accuracies": 0.800000011920929,
-      "rewards/chosen": 0.03712456300854683,
-      "rewards/margins": 0.046282440423965454,
-      "rewards/rejected": -0.009157879278063774,
+      "logits/chosen": -1.9080861806869507,
+      "logits/rejected": -1.910278558731079,
+      "logps/chosen": -33.861846923828125,
+      "logps/rejected": -30.95591163635254,
+      "loss": 0.954,
+      "rewards/accuracies": 0.6625000238418579,
+      "rewards/chosen": 0.03772079199552536,
+      "rewards/margins": 0.04604965075850487,
+      "rewards/rejected": -0.00832886528223753,
       "step": 380
     },
     {
       "epoch": 1.0,
       "step": 385,
       "total_flos": 0.0,
-      "train_loss": 22.846377404943688,
-      "train_runtime": 3252.1936,
-      "train_samples_per_second": 0.947,
+      "train_loss": 0.9734619264478808,
+      "train_runtime": 3253.307,
+      "train_samples_per_second": 0.946,
       "train_steps_per_second": 0.118
     }
   ],