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: 0.9947
-- Rewards/chosen: -0.0004
-- Rewards/rejected: -0.0057
-- Rewards/accuracies: 0.5710
-- Rewards/margins: 0.0053
-- Logps/rejected: -37.5734
-- Logps/chosen: -34.0383
-- Logits/rejected: -2.2197
-- Logits/chosen: -2.2245
+- Loss: 0.6909
+- Rewards/chosen: -0.0119
+- Rewards/rejected: -0.0186
+- Rewards/accuracies: 0.4950
+- Rewards/margins: 0.0067
+- Logps/rejected: -37.6097
+- Logps/chosen: -34.0940
+- Logits/rejected: -2.2214
+- Logits/chosen: -2.2262
 
 ## 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 |
 |:-------------:|:-----:|:----:|:---------------:|:--------------:|:----------------:|:------------------:|:---------------:|:--------------:|:------------:|:---------------:|:-------------:|
-| 0.9851        | 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       |
+| 0.6813        | 0.26  | 100  | 0.6915          | 0.0010         | -0.0036          | 0.5220             | 0.0046          | -37.5345       | -34.0295     | -2.2270         | -2.2319       |
+| 0.6632        | 0.52  | 200  | 0.6888          | -0.0034        | -0.0137          | 0.5660             | 0.0103          | -37.5852       | -34.0514     | -2.2240         | -2.2288       |
+| 0.6327        | 0.78  | 300  | 0.6909          | -0.0119        | -0.0186          | 0.4950             | 0.0067          | -37.6097       | -34.0940     | -2.2214         | -2.2262       |
 
 
 ### Framework versions

adapter_model.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:397afa8bdefda8302c86928dc7ee1a1a66f4213a276c2ea31f11a5742544b838
+oid sha256:395008a5337e9476a1dc1f5375da135344d6f0ec95ea8b2cbda1a750fc60ba5f
 size 176183216

all_results.json

@@ -1,21 +1,8 @@
 {
     "epoch": 1.0,
-    "eval_logits/chosen": -2.22452712059021,
-    "eval_logits/rejected": -2.219710350036621,
-    "eval_logps/chosen": -34.03827667236328,
-    "eval_logps/rejected": -37.573402404785156,
-    "eval_loss": 0.9946814179420471,
-    "eval_rewards/accuracies": 0.5710132718086243,
-    "eval_rewards/chosen": -0.00037224299740046263,
-    "eval_rewards/margins": 0.005306063685566187,
-    "eval_rewards/rejected": -0.005678307265043259,
-    "eval_runtime": 145.3784,
-    "eval_samples": 343,
-    "eval_samples_per_second": 2.359,
-    "eval_steps_per_second": 0.296,
-    "train_loss": 0.9734619264478808,
-    "train_runtime": 3253.307,
+    "train_loss": 0.6689951481757226,
+    "train_runtime": 3249.3574,
     "train_samples": 3079,
-    "train_samples_per_second": 0.946,
+    "train_samples_per_second": 0.948,
     "train_steps_per_second": 0.118
 }

train_results.json

@@ -1,8 +1,8 @@
 {
     "epoch": 1.0,
-    "train_loss": 0.9734619264478808,
-    "train_runtime": 3253.307,
+    "train_loss": 0.6689951481757226,
+    "train_runtime": 3249.3574,
     "train_samples": 3079,
-    "train_samples_per_second": 0.946,
+    "train_samples_per_second": 0.948,
     "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": 1.0,
+      "loss": 0.6931,
       "rewards/accuracies": 0.0,
       "rewards/chosen": 0.0,
       "rewards/margins": 0.0,
@@ -25,590 +25,590 @@
     {
       "epoch": 0.03,
       "learning_rate": 1.282051282051282e-06,
-      "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,
+      "logits/chosen": -1.8665436506271362,
+      "logits/rejected": -1.8708542585372925,
+      "logps/chosen": -36.982460021972656,
+      "logps/rejected": -33.656585693359375,
+      "loss": 0.6882,
+      "rewards/accuracies": 0.5138888955116272,
+      "rewards/chosen": 0.0048276386223733425,
+      "rewards/margins": 0.010130541399121284,
+      "rewards/rejected": -0.005302901845425367,
       "step": 10
     },
     {
       "epoch": 0.05,
       "learning_rate": 2.564102564102564e-06,
-      "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,
+      "logits/chosen": -1.9975639581680298,
+      "logits/rejected": -2.000220537185669,
+      "logps/chosen": -29.6324462890625,
+      "logps/rejected": -29.057437896728516,
+      "loss": 0.693,
+      "rewards/accuracies": 0.512499988079071,
+      "rewards/chosen": 0.0019495869055390358,
+      "rewards/margins": 0.0007479515625163913,
+      "rewards/rejected": 0.0012016354594379663,
       "step": 20
     },
     {
       "epoch": 0.08,
       "learning_rate": 3.846153846153847e-06,
-      "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,
+      "logits/chosen": -1.9208921194076538,
+      "logits/rejected": -1.918195128440857,
+      "logps/chosen": -31.420969009399414,
+      "logps/rejected": -33.22509765625,
+      "loss": 0.6932,
+      "rewards/accuracies": 0.48750001192092896,
+      "rewards/chosen": -0.000974868715275079,
+      "rewards/margins": 0.0003514128620736301,
+      "rewards/rejected": -0.0013262818101793528,
       "step": 30
     },
     {
       "epoch": 0.1,
       "learning_rate": 4.999896948438434e-06,
-      "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,
+      "logits/chosen": -2.018131971359253,
+      "logits/rejected": -2.0093750953674316,
+      "logps/chosen": -32.56100082397461,
+      "logps/rejected": -32.525699615478516,
+      "loss": 0.6905,
+      "rewards/accuracies": 0.512499988079071,
+      "rewards/chosen": 0.0031610552687197924,
+      "rewards/margins": 0.0057663023471832275,
+      "rewards/rejected": -0.002605247776955366,
       "step": 40
     },
     {
       "epoch": 0.13,
       "learning_rate": 4.987541037542187e-06,
-      "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,
+      "logits/chosen": -1.8627980947494507,
+      "logits/rejected": -1.852043867111206,
+      "logps/chosen": -33.57464599609375,
+      "logps/rejected": -35.459083557128906,
+      "loss": 0.6945,
+      "rewards/accuracies": 0.4625000059604645,
+      "rewards/chosen": -0.003433463629335165,
+      "rewards/margins": -0.0020852875895798206,
+      "rewards/rejected": -0.0013481763890013099,
       "step": 50
     },
     {
       "epoch": 0.16,
       "learning_rate": 4.954691471941119e-06,
-      "logits/chosen": -1.9415528774261475,
-      "logits/rejected": -1.9434846639633179,
-      "logps/chosen": -32.53631591796875,
-      "logps/rejected": -33.215850830078125,
-      "loss": 0.9892,
+      "logits/chosen": -1.9412128925323486,
+      "logits/rejected": -1.9431606531143188,
+      "logps/chosen": -32.51996612548828,
+      "logps/rejected": -33.226768493652344,
+      "loss": 0.6806,
       "rewards/accuracies": 0.5625,
-      "rewards/chosen": 0.006386542227119207,
-      "rewards/margins": 0.010833840817213058,
-      "rewards/rejected": -0.004447298124432564,
+      "rewards/chosen": 0.01604365184903145,
+      "rewards/margins": 0.02712194062769413,
+      "rewards/rejected": -0.011078287847340107,
       "step": 60
     },
     {
       "epoch": 0.18,
       "learning_rate": 4.901618883413549e-06,
-      "logits/chosen": -2.072315216064453,
-      "logits/rejected": -2.07728910446167,
-      "logps/chosen": -33.98168182373047,
-      "logps/rejected": -36.64748001098633,
-      "loss": 0.9915,
+      "logits/chosen": -2.072383165359497,
+      "logits/rejected": -2.0773634910583496,
+      "logps/chosen": -33.964813232421875,
+      "logps/rejected": -36.62574005126953,
+      "loss": 0.6861,
       "rewards/accuracies": 0.5625,
-      "rewards/chosen": 0.0006806664168834686,
-      "rewards/margins": 0.00854154396802187,
-      "rewards/rejected": -0.007860877551138401,
+      "rewards/chosen": 0.004735334776341915,
+      "rewards/margins": 0.016109289601445198,
+      "rewards/rejected": -0.011373954825103283,
       "step": 70
     },
     {
       "epoch": 0.21,
       "learning_rate": 4.828760511501322e-06,
-      "logits/chosen": -1.932929277420044,
-      "logits/rejected": -1.9360615015029907,
-      "logps/chosen": -34.32741928100586,
-      "logps/rejected": -34.65910720825195,
-      "loss": 0.9833,
+      "logits/chosen": -1.9332096576690674,
+      "logits/rejected": -1.9363422393798828,
+      "logps/chosen": -34.310150146484375,
+      "logps/rejected": -34.669891357421875,
+      "loss": 0.6749,
       "rewards/accuracies": 0.625,
-      "rewards/chosen": 0.010038824751973152,
-      "rewards/margins": 0.01670856587588787,
-      "rewards/rejected": -0.006669741123914719,
+      "rewards/chosen": 0.023532329127192497,
+      "rewards/margins": 0.03902902454137802,
+      "rewards/rejected": -0.015496693551540375,
       "step": 80
     },
     {
       "epoch": 0.23,
       "learning_rate": 4.7367166013034295e-06,
-      "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,
+      "logits/chosen": -1.9401609897613525,
+      "logits/rejected": -1.944685935974121,
+      "logps/chosen": -32.39472961425781,
+      "logps/rejected": -32.360809326171875,
+      "loss": 0.6865,
+      "rewards/accuracies": 0.625,
+      "rewards/chosen": 0.01453828252851963,
+      "rewards/margins": 0.014918209984898567,
+      "rewards/rejected": -0.000379929319024086,
       "step": 90
     },
     {
       "epoch": 0.26,
       "learning_rate": 4.626245458345211e-06,
-      "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,
+      "logits/chosen": -2.0377113819122314,
+      "logits/rejected": -2.035726308822632,
+      "logps/chosen": -32.14560317993164,
+      "logps/rejected": -31.297948837280273,
+      "loss": 0.6813,
+      "rewards/accuracies": 0.6000000238418579,
+      "rewards/chosen": 0.01904645375907421,
+      "rewards/margins": 0.0253940187394619,
+      "rewards/rejected": -0.006347564049065113,
       "step": 100
     },
     {
       "epoch": 0.26,
-      "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_logits/chosen": -2.231863498687744,
+      "eval_logits/rejected": -2.227025270462036,
+      "eval_logps/chosen": -34.02951431274414,
+      "eval_logps/rejected": -37.534454345703125,
+      "eval_loss": 0.6914929151535034,
+      "eval_rewards/accuracies": 0.5220099687576294,
+      "eval_rewards/chosen": 0.0010076783364638686,
+      "eval_rewards/margins": 0.004574176389724016,
+      "eval_rewards/rejected": -0.0035664979368448257,
+      "eval_runtime": 145.7925,
+      "eval_samples_per_second": 2.353,
       "eval_steps_per_second": 0.295,
       "step": 100
     },
     {
       "epoch": 0.29,
       "learning_rate": 4.498257201263691e-06,
-      "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": 0.013529380783438683,
-      "rewards/margins": 0.01042473316192627,
-      "rewards/rejected": 0.0031046485528349876,
+      "logits/chosen": -1.992661476135254,
+      "logits/rejected": -1.9902756214141846,
+      "logps/chosen": -33.13728713989258,
+      "logps/rejected": -34.051490783691406,
+      "loss": 0.6834,
+      "rewards/accuracies": 0.6499999761581421,
+      "rewards/chosen": 0.02130572497844696,
+      "rewards/margins": 0.025729293003678322,
+      "rewards/rejected": -0.004423565696924925,
       "step": 110
     },
     {
       "epoch": 0.31,
       "learning_rate": 4.353806263777678e-06,
-      "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,
+      "logits/chosen": -2.0037741661071777,
+      "logits/rejected": -1.995439887046814,
+      "logps/chosen": -32.33051681518555,
+      "logps/rejected": -32.13713836669922,
+      "loss": 0.6853,
+      "rewards/accuracies": 0.5874999761581421,
+      "rewards/chosen": 0.022878289222717285,
+      "rewards/margins": 0.01800495944917202,
+      "rewards/rejected": 0.004873327445238829,
       "step": 120
     },
     {
       "epoch": 0.34,
       "learning_rate": 4.1940827077152755e-06,
-      "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,
+      "logits/chosen": -2.032257318496704,
+      "logits/rejected": -2.024292230606079,
+      "logps/chosen": -30.306774139404297,
+      "logps/rejected": -32.081024169921875,
+      "loss": 0.6754,
+      "rewards/accuracies": 0.6499999761581421,
+      "rewards/chosen": 0.03378837928175926,
+      "rewards/margins": 0.03939158096909523,
+      "rewards/rejected": -0.00560319609940052,
       "step": 130
     },
     {
       "epoch": 0.36,
       "learning_rate": 4.0204024186666215e-06,
-      "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,
+      "logits/chosen": -1.9630043506622314,
+      "logits/rejected": -1.973233938217163,
+      "logps/chosen": -31.224035263061523,
+      "logps/rejected": -32.56679916381836,
+      "loss": 0.6718,
+      "rewards/accuracies": 0.675000011920929,
+      "rewards/chosen": 0.0391254760324955,
+      "rewards/margins": 0.045508723706007004,
+      "rewards/rejected": -0.006383246276527643,
       "step": 140
     },
     {
       "epoch": 0.39,
       "learning_rate": 3.834196265035119e-06,
-      "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,
+      "logits/chosen": -1.874066710472107,
+      "logits/rejected": -1.8752658367156982,
+      "logps/chosen": -33.90449905395508,
+      "logps/rejected": -34.78742218017578,
+      "loss": 0.6642,
+      "rewards/accuracies": 0.6000000238418579,
+      "rewards/chosen": 0.05763882398605347,
+      "rewards/margins": 0.06367169320583344,
+      "rewards/rejected": -0.006032869219779968,
       "step": 150
     },
     {
       "epoch": 0.42,
       "learning_rate": 3.636998309800573e-06,
-      "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,
+      "logits/chosen": -1.9251247644424438,
+      "logits/rejected": -1.9217197895050049,
+      "logps/chosen": -36.002262115478516,
+      "logps/rejected": -32.70859909057617,
+      "loss": 0.6788,
+      "rewards/accuracies": 0.6000000238418579,
+      "rewards/chosen": 0.03388362005352974,
+      "rewards/margins": 0.031053191050887108,
+      "rewards/rejected": 0.0028304329607635736,
       "step": 160
     },
     {
       "epoch": 0.44,
       "learning_rate": 3.4304331721118078e-06,
-      "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": 0.03410445898771286,
-      "rewards/margins": 0.038923610001802444,
-      "rewards/rejected": -0.004819151014089584,
+      "logits/chosen": -2.025562286376953,
+      "logits/rejected": -2.018228530883789,
+      "logps/chosen": -33.46672439575195,
+      "logps/rejected": -31.409042358398438,
+      "loss": 0.657,
+      "rewards/accuracies": 0.699999988079071,
+      "rewards/chosen": 0.06622789800167084,
+      "rewards/margins": 0.07766623049974442,
+      "rewards/rejected": -0.011438337154686451,
       "step": 170
     },
     {
       "epoch": 0.47,
       "learning_rate": 3.2162026428305436e-06,
-      "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,
+      "logits/chosen": -2.0323550701141357,
+      "logits/rejected": -2.037605047225952,
+      "logps/chosen": -32.23218536376953,
+      "logps/rejected": -32.439491271972656,
+      "loss": 0.6682,
+      "rewards/accuracies": 0.675000011920929,
+      "rewards/chosen": 0.06252055615186691,
+      "rewards/margins": 0.05339093878865242,
+      "rewards/rejected": 0.009129621088504791,
       "step": 180
     },
     {
       "epoch": 0.49,
       "learning_rate": 2.996071664294641e-06,
-      "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,
+      "logits/chosen": -2.0334537029266357,
+      "logits/rejected": -2.030702590942383,
+      "logps/chosen": -31.266775131225586,
+      "logps/rejected": -31.347997665405273,
+      "loss": 0.67,
+      "rewards/accuracies": 0.675000011920929,
+      "rewards/chosen": 0.04443611577153206,
+      "rewards/margins": 0.050393205136060715,
+      "rewards/rejected": -0.005957084707915783,
       "step": 190
     },
     {
       "epoch": 0.52,
       "learning_rate": 2.7718537898066833e-06,
-      "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,
+      "logits/chosen": -1.9037307500839233,
+      "logits/rejected": -1.9083878993988037,
+      "logps/chosen": -31.30868148803711,
+      "logps/rejected": -32.823123931884766,
+      "loss": 0.6632,
+      "rewards/accuracies": 0.6875,
+      "rewards/chosen": 0.05951983854174614,
+      "rewards/margins": 0.0659579336643219,
+      "rewards/rejected": -0.006438094191253185,
       "step": 200
     },
     {
       "epoch": 0.52,
-      "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_logits/chosen": -2.22880482673645,
+      "eval_logits/rejected": -2.2239701747894287,
+      "eval_logps/chosen": -34.05141067504883,
+      "eval_logps/rejected": -37.58522033691406,
+      "eval_loss": 0.6888246536254883,
+      "eval_rewards/accuracies": 0.5660299062728882,
+      "eval_rewards/chosen": -0.003371814964339137,
+      "eval_rewards/margins": 0.01034836657345295,
+      "eval_rewards/rejected": -0.013720180839300156,
+      "eval_runtime": 145.665,
+      "eval_samples_per_second": 2.355,
       "eval_steps_per_second": 0.295,
       "step": 200
     },
     {
       "epoch": 0.55,
       "learning_rate": 2.5453962426402006e-06,
-      "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,
+      "logits/chosen": -2.016369342803955,
+      "logits/rejected": -2.027027130126953,
+      "logps/chosen": -31.75009536743164,
+      "logps/rejected": -33.965911865234375,
+      "loss": 0.6631,
+      "rewards/accuracies": 0.6875,
+      "rewards/chosen": 0.04424174129962921,
+      "rewards/margins": 0.06488426774740219,
+      "rewards/rejected": -0.020642530173063278,
       "step": 210
     },
     {
       "epoch": 0.57,
       "learning_rate": 2.3185646976551794e-06,
-      "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,
+      "logits/chosen": -1.909157156944275,
+      "logits/rejected": -1.9239362478256226,
+      "logps/chosen": -29.8031005859375,
+      "logps/rejected": -31.647253036499023,
+      "loss": 0.6564,
+      "rewards/accuracies": 0.7124999761581421,
+      "rewards/chosen": 0.06168792396783829,
+      "rewards/margins": 0.07950346171855927,
+      "rewards/rejected": -0.017815548926591873,
       "step": 220
     },
     {
       "epoch": 0.6,
       "learning_rate": 2.0932279108998323e-06,
-      "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,
+      "logits/chosen": -1.9656283855438232,
+      "logits/rejected": -1.9696086645126343,
+      "logps/chosen": -33.09147644042969,
+      "logps/rejected": -31.6539306640625,
+      "loss": 0.6543,
+      "rewards/accuracies": 0.699999988079071,
+      "rewards/chosen": 0.06519898772239685,
+      "rewards/margins": 0.08612470328807831,
+      "rewards/rejected": -0.02092570997774601,
       "step": 230
     },
     {
       "epoch": 0.62,
       "learning_rate": 1.8712423238279358e-06,
-      "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,
+      "logits/chosen": -1.963721513748169,
+      "logits/rejected": -1.9419059753417969,
+      "logps/chosen": -33.84326934814453,
+      "logps/rejected": -35.13850784301758,
+      "loss": 0.6479,
+      "rewards/accuracies": 0.737500011920929,
+      "rewards/chosen": 0.06209472939372063,
+      "rewards/margins": 0.0996372401714325,
+      "rewards/rejected": -0.03754251450300217,
       "step": 240
     },
     {
       "epoch": 0.65,
       "learning_rate": 1.6544367689701824e-06,
-      "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,
+      "logits/chosen": -2.004868268966675,
+      "logits/rejected": -2.0015549659729004,
+      "logps/chosen": -32.729896545410156,
+      "logps/rejected": -36.27812957763672,
+      "loss": 0.6706,
+      "rewards/accuracies": 0.5874999761581421,
+      "rewards/chosen": 0.03903444483876228,
+      "rewards/margins": 0.04967557638883591,
+      "rewards/rejected": -0.010641133412718773,
       "step": 250
     },
     {
       "epoch": 0.68,
       "learning_rate": 1.4445974030621963e-06,
-      "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,
+      "logits/chosen": -1.8725497722625732,
+      "logits/rejected": -1.8701589107513428,
+      "logps/chosen": -33.99323654174805,
+      "logps/rejected": -35.557159423828125,
+      "loss": 0.6706,
+      "rewards/accuracies": 0.699999988079071,
+      "rewards/chosen": 0.039031852036714554,
+      "rewards/margins": 0.04953894019126892,
+      "rewards/rejected": -0.010507088154554367,
       "step": 260
     },
     {
       "epoch": 0.7,
       "learning_rate": 1.243452991757889e-06,
-      "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,
+      "logits/chosen": -1.8571383953094482,
+      "logits/rejected": -1.854741096496582,
+      "logps/chosen": -34.20746612548828,
+      "logps/rejected": -31.83416748046875,
+      "loss": 0.6708,
+      "rewards/accuracies": 0.6000000238418579,
+      "rewards/chosen": 0.0380266010761261,
+      "rewards/margins": 0.05059366300702095,
+      "rewards/rejected": -0.01256705541163683,
       "step": 270
     },
     {
       "epoch": 0.73,
       "learning_rate": 1.0526606671603523e-06,
-      "logits/chosen": -1.9572479724884033,
-      "logits/rejected": -1.9467785358428955,
-      "logps/chosen": -34.97367858886719,
-      "logps/rejected": -31.858196258544922,
-      "loss": 0.96,
+      "logits/chosen": -1.9600715637207031,
+      "logits/rejected": -1.9495933055877686,
+      "logps/chosen": -34.99934387207031,
+      "logps/rejected": -31.900096893310547,
+      "loss": 0.6544,
       "rewards/accuracies": 0.7124999761581421,
-      "rewards/chosen": 0.03829622268676758,
-      "rewards/margins": 0.04001673310995102,
-      "rewards/rejected": -0.001720509259030223,
+      "rewards/chosen": 0.07145937532186508,
+      "rewards/margins": 0.08328022807836533,
+      "rewards/rejected": -0.011820845305919647,
       "step": 280
     },
     {
       "epoch": 0.75,
       "learning_rate": 8.737922755071455e-07,
-      "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": 0.022954344749450684,
-      "rewards/margins": 0.01828841306269169,
-      "rewards/rejected": 0.004665931686758995,
+      "logits/chosen": -2.0559241771698,
+      "logits/rejected": -2.04101300239563,
+      "logps/chosen": -30.744888305664062,
+      "logps/rejected": -32.62079620361328,
+      "loss": 0.6809,
+      "rewards/accuracies": 0.5375000238418579,
+      "rewards/chosen": 0.035847391933202744,
+      "rewards/margins": 0.029639026150107384,
+      "rewards/rejected": 0.006208371836692095,
       "step": 290
     },
     {
       "epoch": 0.78,
       "learning_rate": 7.08321427484816e-07,
-      "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,
+      "logits/chosen": -1.9259971380233765,
+      "logits/rejected": -1.9234883785247803,
+      "logps/chosen": -32.35750961303711,
+      "logps/rejected": -30.92331314086914,
+      "loss": 0.6327,
+      "rewards/accuracies": 0.7875000238418579,
+      "rewards/chosen": 0.11392021179199219,
+      "rewards/margins": 0.13522081077098846,
+      "rewards/rejected": -0.021300604566931725,
       "step": 300
     },
     {
       "epoch": 0.78,
-      "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,
+      "eval_logits/chosen": -2.2262163162231445,
+      "eval_logits/rejected": -2.2213830947875977,
+      "eval_logps/chosen": -34.093963623046875,
+      "eval_logps/rejected": -37.60972213745117,
+      "eval_loss": 0.6908519268035889,
+      "eval_rewards/accuracies": 0.49501657485961914,
+      "eval_rewards/chosen": -0.011881927028298378,
+      "eval_rewards/margins": 0.006738076452165842,
+      "eval_rewards/rejected": -0.018620004877448082,
+      "eval_runtime": 145.4585,
+      "eval_samples_per_second": 2.358,
+      "eval_steps_per_second": 0.296,
       "step": 300
     },
     {
       "epoch": 0.81,
       "learning_rate": 5.576113578589035e-07,
-      "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": 0.03158777207136154,
-      "rewards/margins": 0.03811599686741829,
-      "rewards/rejected": -0.006528225727379322,
+      "logits/chosen": -1.9112951755523682,
+      "logits/rejected": -1.9080655574798584,
+      "logps/chosen": -31.314626693725586,
+      "logps/rejected": -33.85969543457031,
+      "loss": 0.6566,
+      "rewards/accuracies": 0.762499988079071,
+      "rewards/chosen": 0.05774437263607979,
+      "rewards/margins": 0.08125325292348862,
+      "rewards/rejected": -0.023508887737989426,
       "step": 310
     },
     {
       "epoch": 0.83,
       "learning_rate": 4.229036944380913e-07,
-      "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,
+      "logits/chosen": -1.960738182067871,
+      "logits/rejected": -1.9485470056533813,
+      "logps/chosen": -34.32622528076172,
+      "logps/rejected": -33.65083312988281,
+      "loss": 0.6586,
+      "rewards/accuracies": 0.637499988079071,
+      "rewards/chosen": 0.0539703294634819,
+      "rewards/margins": 0.0763682946562767,
+      "rewards/rejected": -0.022397976368665695,
       "step": 320
     },
     {
       "epoch": 0.86,
       "learning_rate": 3.053082288996112e-07,
-      "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,
+      "logits/chosen": -1.9961421489715576,
+      "logits/rejected": -1.9947240352630615,
+      "logps/chosen": -33.134159088134766,
+      "logps/rejected": -32.56267547607422,
+      "loss": 0.6536,
+      "rewards/accuracies": 0.762499988079071,
+      "rewards/chosen": 0.07032694667577744,
+      "rewards/margins": 0.08694546669721603,
+      "rewards/rejected": -0.0166185162961483,
       "step": 330
     },
     {
       "epoch": 0.88,
       "learning_rate": 2.0579377374915805e-07,
-      "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,
+      "logits/chosen": -2.083061933517456,
+      "logits/rejected": -2.0673813819885254,
+      "logps/chosen": -33.78687286376953,
+      "logps/rejected": -33.12049102783203,
+      "loss": 0.6568,
+      "rewards/accuracies": 0.75,
+      "rewards/chosen": 0.07850398868322372,
+      "rewards/margins": 0.07899539172649384,
+      "rewards/rejected": -0.0004914018791168928,
       "step": 340
     },
     {
       "epoch": 0.91,
       "learning_rate": 1.2518018074041684e-07,
-      "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,
+      "logits/chosen": -1.9557991027832031,
+      "logits/rejected": -1.9549639225006104,
+      "logps/chosen": -32.84474563598633,
+      "logps/rejected": -32.53619384765625,
+      "loss": 0.648,
+      "rewards/accuracies": 0.675000011920929,
+      "rewards/chosen": 0.0874704122543335,
+      "rewards/margins": 0.10167930275201797,
+      "rewards/rejected": -0.014208881184458733,
       "step": 350
     },
     {
       "epoch": 0.94,
       "learning_rate": 6.41315865106129e-08,
-      "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": 0.03483304753899574,
-      "rewards/margins": 0.03557039424777031,
-      "rewards/rejected": -0.0007373450207524002,
+      "logits/chosen": -1.9108169078826904,
+      "logits/rejected": -1.9210844039916992,
+      "logps/chosen": -31.8836612701416,
+      "logps/rejected": -35.36156463623047,
+      "loss": 0.6572,
+      "rewards/accuracies": 0.6875,
+      "rewards/chosen": 0.06420587003231049,
+      "rewards/margins": 0.07876714318990707,
+      "rewards/rejected": -0.014561265707015991,
       "step": 360
     },
     {
       "epoch": 0.96,
       "learning_rate": 2.3150941078050325e-08,
-      "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,
+      "logits/chosen": -2.0503978729248047,
+      "logits/rejected": -2.043942928314209,
+      "logps/chosen": -33.33507537841797,
+      "logps/rejected": -29.28571128845215,
+      "loss": 0.6582,
+      "rewards/accuracies": 0.75,
+      "rewards/chosen": 0.06337906420230865,
+      "rewards/margins": 0.07523626834154129,
+      "rewards/rejected": -0.011857211589813232,
       "step": 370
     },
     {
       "epoch": 0.99,
       "learning_rate": 2.575864278703266e-09,
-      "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,
+      "logits/chosen": -1.9104009866714478,
+      "logits/rejected": -1.9126161336898804,
+      "logps/chosen": -33.88220977783203,
+      "logps/rejected": -30.987438201904297,
+      "loss": 0.65,
+      "rewards/accuracies": 0.7250000238418579,
+      "rewards/chosen": 0.07136790454387665,
+      "rewards/margins": 0.09433047473430634,
+      "rewards/rejected": -0.02296256460249424,
       "step": 380
     },
     {
       "epoch": 1.0,
       "step": 385,
       "total_flos": 0.0,
-      "train_loss": 0.9734619264478808,
-      "train_runtime": 3253.307,
-      "train_samples_per_second": 0.946,
+      "train_loss": 0.6689951481757226,
+      "train_runtime": 3249.3574,
+      "train_samples_per_second": 0.948,
       "train_steps_per_second": 0.118
     }
   ],