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: 0.6907
-- Rewards/chosen: -0.0131
-- Rewards/rejected: -0.0201
-- Rewards/accuracies: 0.5133
-- Rewards/margins: 0.0070
-- Logps/rejected: -37.6172
-- Logps/chosen: -34.1001
-- Logits/rejected: -2.2213
-- Logits/chosen: -2.2261
+- Loss: 0.4980
+- Rewards/chosen: 0.0533
+- Rewards/rejected: 0.0449
+- Rewards/accuracies: 0.5158
+- Rewards/margins: 0.0084
+- Logps/rejected: -37.2920
+- Logps/chosen: -33.7681
+- Logits/rejected: -2.2264
+- Logits/chosen: -2.2312
 
 ## 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 |
 |:-------------:|:-----:|:----:|:---------------:|:--------------:|:----------------:|:------------------:|:---------------:|:--------------:|:------------:|:---------------:|:-------------:|
-| 0.6813        | 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       |
+| 0.4947        | 0.26  | 100  | 0.4994          | 0.0239         | 0.0208           | 0.5216             | 0.0031          | -37.4126       | -33.9151     | -2.2299         | -2.2348       |
+| 0.4825        | 0.52  | 200  | 0.4974          | 0.0511         | 0.0397           | 0.5544             | 0.0113          | -37.3179       | -33.7792     | -2.2274         | -2.2322       |
+| 0.4669        | 0.78  | 300  | 0.4980          | 0.0533         | 0.0449           | 0.5158             | 0.0084          | -37.2920       | -33.7681     | -2.2264         | -2.2312       |
 
 
 ### Framework versions

adapter_model.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:5fe3cf7d65d359cffa8bd5e777721b4103dc423a79361130841e72ce3085a9a9
+oid sha256:cbe40ba78b8d64b2ad247d9d8b649621cc40ab032cb206a1fa7ea2cb6bd7cb78
 size 176183216

all_results.json

@@ -1,21 +1,8 @@
 {
     "epoch": 1.0,
-    "eval_logits/chosen": -2.2261412143707275,
-    "eval_logits/rejected": -2.2213146686553955,
-    "eval_logps/chosen": -34.100101470947266,
-    "eval_logps/rejected": -37.61717224121094,
-    "eval_loss": 0.6907125115394592,
-    "eval_rewards/accuracies": 0.5132890343666077,
-    "eval_rewards/chosen": -0.013110256753861904,
-    "eval_rewards/margins": 0.007000547368079424,
-    "eval_rewards/rejected": -0.020110804587602615,
-    "eval_runtime": 145.4751,
-    "eval_samples": 343,
-    "eval_samples_per_second": 2.358,
-    "eval_steps_per_second": 0.296,
-    "train_loss": 0.6689951481757226,
-    "train_runtime": 3249.3574,
+    "train_loss": 0.4876757522682091,
+    "train_runtime": 3250.1859,
     "train_samples": 3079,
-    "train_samples_per_second": 0.948,
+    "train_samples_per_second": 0.947,
     "train_steps_per_second": 0.118
 }

train_results.json

@@ -1,8 +1,8 @@
 {
     "epoch": 1.0,
-    "train_loss": 0.6689951481757226,
-    "train_runtime": 3249.3574,
+    "train_loss": 0.4876757522682091,
+    "train_runtime": 3250.1859,
     "train_samples": 3079,
-    "train_samples_per_second": 0.948,
+    "train_samples_per_second": 0.947,
     "train_steps_per_second": 0.118
 }

trainer_state.json

@@ -15,7 +15,7 @@
       "logits/rejected": -1.7377450466156006,
       "logps/chosen": -29.553977966308594,
       "logps/rejected": -42.813133239746094,
-      "loss": 0.6931,
+      "loss": 0.5,
       "rewards/accuracies": 0.0,
       "rewards/chosen": 0.0,
       "rewards/margins": 0.0,
@@ -25,155 +25,155 @@
     {
       "epoch": 0.03,
       "learning_rate": 1.282051282051282e-06,
-      "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,
+      "logits/chosen": -1.8665586709976196,
+      "logits/rejected": -1.8708692789077759,
+      "logps/chosen": -37.00250244140625,
+      "logps/rejected": -33.66969299316406,
+      "loss": 0.4985,
+      "rewards/accuracies": 0.5416666865348816,
+      "rewards/chosen": 0.0008193479152396321,
+      "rewards/margins": 0.008743342012166977,
+      "rewards/rejected": -0.007923995144665241,
       "step": 10
     },
     {
       "epoch": 0.05,
       "learning_rate": 2.564102564102564e-06,
-      "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,
+      "logits/chosen": -1.9974254369735718,
+      "logits/rejected": -2.0000810623168945,
+      "logps/chosen": -29.634906768798828,
+      "logps/rejected": -29.027408599853516,
+      "loss": 0.5009,
+      "rewards/accuracies": 0.38749998807907104,
+      "rewards/chosen": 0.001457492122426629,
+      "rewards/margins": -0.005749998614192009,
+      "rewards/rejected": 0.007207490503787994,
       "step": 20
     },
     {
       "epoch": 0.08,
       "learning_rate": 3.846153846153847e-06,
-      "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,
+      "logits/chosen": -1.9204790592193604,
+      "logits/rejected": -1.9177839756011963,
+      "logps/chosen": -31.412555694580078,
+      "logps/rejected": -33.24369812011719,
+      "loss": 0.4992,
+      "rewards/accuracies": 0.550000011920929,
+      "rewards/chosen": 0.0007082058000378311,
+      "rewards/margins": 0.005755766294896603,
+      "rewards/rejected": -0.005047560669481754,
       "step": 30
     },
     {
       "epoch": 0.1,
       "learning_rate": 4.999896948438434e-06,
-      "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,
+      "logits/chosen": -2.0162367820739746,
+      "logits/rejected": -2.007521629333496,
+      "logps/chosen": -32.55222702026367,
+      "logps/rejected": -32.50428771972656,
+      "loss": 0.4992,
+      "rewards/accuracies": 0.5,
+      "rewards/chosen": 0.004915344063192606,
+      "rewards/margins": 0.0032389431726187468,
+      "rewards/rejected": 0.0016764007741585374,
       "step": 40
     },
     {
       "epoch": 0.13,
       "learning_rate": 4.987541037542187e-06,
-      "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,
+      "logits/chosen": -1.8627817630767822,
+      "logits/rejected": -1.8519961833953857,
+      "logps/chosen": -33.52722930908203,
+      "logps/rejected": -35.41474151611328,
+      "loss": 0.5002,
+      "rewards/accuracies": 0.4749999940395355,
+      "rewards/chosen": 0.0060504418797791,
+      "rewards/margins": -0.001469915034249425,
+      "rewards/rejected": 0.007520356681197882,
       "step": 50
     },
     {
       "epoch": 0.16,
       "learning_rate": 4.954691471941119e-06,
-      "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": 0.01604365184903145,
-      "rewards/margins": 0.02712194062769413,
-      "rewards/rejected": -0.011078287847340107,
+      "logits/chosen": -1.9401956796646118,
+      "logits/rejected": -1.9421344995498657,
+      "logps/chosen": -32.53112030029297,
+      "logps/rejected": -33.1898307800293,
+      "loss": 0.4964,
+      "rewards/accuracies": 0.6000000238418579,
+      "rewards/chosen": 0.01381197851151228,
+      "rewards/margins": 0.01750265061855316,
+      "rewards/rejected": -0.00369067071005702,
       "step": 60
     },
     {
       "epoch": 0.18,
       "learning_rate": 4.901618883413549e-06,
-      "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": 0.004735334776341915,
-      "rewards/margins": 0.016109289601445198,
-      "rewards/rejected": -0.011373954825103283,
+      "logits/chosen": -2.071833610534668,
+      "logits/rejected": -2.076794147491455,
+      "logps/chosen": -33.93788528442383,
+      "logps/rejected": -36.575801849365234,
+      "loss": 0.4976,
+      "rewards/accuracies": 0.5874999761581421,
+      "rewards/chosen": 0.010120250284671783,
+      "rewards/margins": 0.011506280861794949,
+      "rewards/rejected": -0.0013860296458005905,
       "step": 70
     },
     {
       "epoch": 0.21,
       "learning_rate": 4.828760511501322e-06,
-      "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": 0.023532329127192497,
-      "rewards/margins": 0.03902902454137802,
-      "rewards/rejected": -0.015496693551540375,
+      "logits/chosen": -1.9325587749481201,
+      "logits/rejected": -1.9356613159179688,
+      "logps/chosen": -34.26636505126953,
+      "logps/rejected": -34.54140090942383,
+      "loss": 0.4947,
+      "rewards/accuracies": 0.550000011920929,
+      "rewards/chosen": 0.032289423048496246,
+      "rewards/margins": 0.022087663412094116,
+      "rewards/rejected": 0.01020175963640213,
       "step": 80
     },
     {
       "epoch": 0.23,
       "learning_rate": 4.7367166013034295e-06,
-      "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,
+      "logits/chosen": -1.942940354347229,
+      "logits/rejected": -1.9474563598632812,
+      "logps/chosen": -32.32842254638672,
+      "logps/rejected": -32.28204345703125,
+      "loss": 0.4965,
+      "rewards/accuracies": 0.574999988079071,
+      "rewards/chosen": 0.027799556031823158,
+      "rewards/margins": 0.012426125817000866,
+      "rewards/rejected": 0.015373429283499718,
       "step": 90
     },
     {
       "epoch": 0.26,
       "learning_rate": 4.626245458345211e-06,
-      "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,
+      "logits/chosen": -2.0405964851379395,
+      "logits/rejected": -2.0386147499084473,
+      "logps/chosen": -32.07122039794922,
+      "logps/rejected": -31.215023040771484,
+      "loss": 0.4947,
+      "rewards/accuracies": 0.574999988079071,
+      "rewards/chosen": 0.033922821283340454,
+      "rewards/margins": 0.02368539571762085,
+      "rewards/rejected": 0.010237427428364754,
       "step": 100
     },
     {
       "epoch": 0.26,
-      "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_logits/chosen": -2.2347991466522217,
+      "eval_logits/rejected": -2.229947805404663,
+      "eval_logps/chosen": -33.91511917114258,
+      "eval_logps/rejected": -37.412628173828125,
+      "eval_loss": 0.499397873878479,
+      "eval_rewards/accuracies": 0.5215947031974792,
+      "eval_rewards/chosen": 0.023886699229478836,
+      "eval_rewards/margins": 0.0030889539048075676,
+      "eval_rewards/rejected": 0.020797746255993843,
+      "eval_runtime": 145.7619,
       "eval_samples_per_second": 2.353,
       "eval_steps_per_second": 0.295,
       "step": 100
@@ -181,434 +181,434 @@
     {
       "epoch": 0.29,
       "learning_rate": 4.498257201263691e-06,
-      "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,
+      "logits/chosen": -1.9959571361541748,
+      "logits/rejected": -1.99361252784729,
+      "logps/chosen": -32.979209899902344,
+      "logps/rejected": -33.898841857910156,
+      "loss": 0.4925,
+      "rewards/accuracies": 0.612500011920929,
+      "rewards/chosen": 0.052921634167432785,
+      "rewards/margins": 0.026816055178642273,
+      "rewards/rejected": 0.026105573400855064,
       "step": 110
     },
     {
       "epoch": 0.31,
       "learning_rate": 4.353806263777678e-06,
-      "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,
+      "logits/chosen": -2.007497787475586,
+      "logits/rejected": -1.9991832971572876,
+      "logps/chosen": -32.195396423339844,
+      "logps/rejected": -31.9866886138916,
+      "loss": 0.4967,
+      "rewards/accuracies": 0.5249999761581421,
+      "rewards/chosen": 0.04990261048078537,
+      "rewards/margins": 0.01493864320218563,
+      "rewards/rejected": 0.03496397286653519,
       "step": 120
     },
     {
       "epoch": 0.34,
       "learning_rate": 4.1940827077152755e-06,
-      "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,
+      "logits/chosen": -2.0352938175201416,
+      "logits/rejected": -2.027338743209839,
+      "logps/chosen": -30.16824722290039,
+      "logps/rejected": -31.9173526763916,
+      "loss": 0.4919,
+      "rewards/accuracies": 0.625,
+      "rewards/chosen": 0.061494071036577225,
+      "rewards/margins": 0.034362830221652985,
+      "rewards/rejected": 0.027131233364343643,
       "step": 130
     },
     {
       "epoch": 0.36,
       "learning_rate": 4.0204024186666215e-06,
-      "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,
+      "logits/chosen": -1.9656566381454468,
+      "logits/rejected": -1.9758468866348267,
+      "logps/chosen": -31.05475425720215,
+      "logps/rejected": -32.390235900878906,
+      "loss": 0.4893,
+      "rewards/accuracies": 0.637499988079071,
+      "rewards/chosen": 0.0729818269610405,
+      "rewards/margins": 0.04405337944626808,
+      "rewards/rejected": 0.028928453102707863,
       "step": 140
     },
     {
       "epoch": 0.39,
       "learning_rate": 3.834196265035119e-06,
-      "logits/chosen": -1.874066710472107,
-      "logits/rejected": -1.8752658367156982,
-      "logps/chosen": -33.90449905395508,
-      "logps/rejected": -34.78742218017578,
-      "loss": 0.6642,
+      "logits/chosen": -1.877623200416565,
+      "logits/rejected": -1.8787848949432373,
+      "logps/chosen": -33.642906188964844,
+      "logps/rejected": -34.57593536376953,
+      "loss": 0.4826,
       "rewards/accuracies": 0.6000000238418579,
-      "rewards/chosen": 0.05763882398605347,
-      "rewards/margins": 0.06367169320583344,
-      "rewards/rejected": -0.006032869219779968,
+      "rewards/chosen": 0.10995662212371826,
+      "rewards/margins": 0.07369254529476166,
+      "rewards/rejected": 0.0362640880048275,
       "step": 150
     },
     {
       "epoch": 0.42,
       "learning_rate": 3.636998309800573e-06,
-      "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,
+      "logits/chosen": -1.9285871982574463,
+      "logits/rejected": -1.9251912832260132,
+      "logps/chosen": -35.79336929321289,
+      "logps/rejected": -32.49341583251953,
+      "loss": 0.4925,
+      "rewards/accuracies": 0.574999988079071,
+      "rewards/chosen": 0.07566188275814056,
+      "rewards/margins": 0.029793858528137207,
+      "rewards/rejected": 0.045868031680583954,
       "step": 160
     },
     {
       "epoch": 0.44,
       "learning_rate": 3.4304331721118078e-06,
-      "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,
+      "logits/chosen": -2.0295512676239014,
+      "logits/rejected": -2.0222458839416504,
+      "logps/chosen": -33.23942947387695,
+      "logps/rejected": -31.19607925415039,
+      "loss": 0.4806,
+      "rewards/accuracies": 0.675000011920929,
+      "rewards/chosen": 0.11168631166219711,
+      "rewards/margins": 0.08053232729434967,
+      "rewards/rejected": 0.0311539676040411,
       "step": 170
     },
     {
       "epoch": 0.47,
       "learning_rate": 3.2162026428305436e-06,
-      "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,
+      "logits/chosen": -2.0365030765533447,
+      "logits/rejected": -2.0416781902313232,
+      "logps/chosen": -31.986160278320312,
+      "logps/rejected": -32.17144012451172,
+      "loss": 0.4878,
+      "rewards/accuracies": 0.637499988079071,
+      "rewards/chosen": 0.11172527074813843,
+      "rewards/margins": 0.048985324800014496,
+      "rewards/rejected": 0.06273995339870453,
       "step": 180
     },
     {
       "epoch": 0.49,
       "learning_rate": 2.996071664294641e-06,
-      "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,
+      "logits/chosen": -2.0367565155029297,
+      "logits/rejected": -2.0340359210968018,
+      "logps/chosen": -31.0958194732666,
+      "logps/rejected": -31.11456298828125,
+      "loss": 0.4905,
+      "rewards/accuracies": 0.637499988079071,
+      "rewards/chosen": 0.07862688601016998,
+      "rewards/margins": 0.03789640590548515,
+      "rewards/rejected": 0.04073048755526543,
       "step": 190
     },
     {
       "epoch": 0.52,
       "learning_rate": 2.7718537898066833e-06,
-      "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,
+      "logits/chosen": -1.90771484375,
+      "logits/rejected": -1.912388801574707,
+      "logps/chosen": -31.08389663696289,
+      "logps/rejected": -32.62942123413086,
+      "loss": 0.4825,
+      "rewards/accuracies": 0.699999988079071,
+      "rewards/chosen": 0.10447730123996735,
+      "rewards/margins": 0.07217548787593842,
+      "rewards/rejected": 0.032301802188158035,
       "step": 200
     },
     {
       "epoch": 0.52,
-      "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_logits/chosen": -2.2322466373443604,
+      "eval_logits/rejected": -2.2274229526519775,
+      "eval_logps/chosen": -33.779205322265625,
+      "eval_logps/rejected": -37.31794357299805,
+      "eval_loss": 0.4973558187484741,
+      "eval_rewards/accuracies": 0.5544019937515259,
+      "eval_rewards/chosen": 0.05106903612613678,
+      "eval_rewards/margins": 0.011334729380905628,
+      "eval_rewards/rejected": 0.039734311401844025,
+      "eval_runtime": 145.5488,
+      "eval_samples_per_second": 2.357,
       "eval_steps_per_second": 0.295,
       "step": 200
     },
     {
       "epoch": 0.55,
       "learning_rate": 2.5453962426402006e-06,
-      "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,
+      "logits/chosen": -2.0204057693481445,
+      "logits/rejected": -2.0309972763061523,
+      "logps/chosen": -31.54058837890625,
+      "logps/rejected": -33.71467590332031,
+      "loss": 0.4864,
+      "rewards/accuracies": 0.6000000238418579,
+      "rewards/chosen": 0.08614270389080048,
+      "rewards/margins": 0.0565376803278923,
+      "rewards/rejected": 0.029605034738779068,
       "step": 210
     },
     {
       "epoch": 0.57,
       "learning_rate": 2.3185646976551794e-06,
-      "logits/chosen": -1.909157156944275,
-      "logits/rejected": -1.9239362478256226,
-      "logps/chosen": -29.8031005859375,
-      "logps/rejected": -31.647253036499023,
-      "loss": 0.6564,
+      "logits/chosen": -1.9130290746688843,
+      "logits/rejected": -1.9277244806289673,
+      "logps/chosen": -29.63765525817871,
+      "logps/rejected": -31.437763214111328,
+      "loss": 0.4831,
       "rewards/accuracies": 0.7124999761581421,
-      "rewards/chosen": 0.06168792396783829,
-      "rewards/margins": 0.07950346171855927,
-      "rewards/rejected": -0.017815548926591873,
+      "rewards/chosen": 0.09477666765451431,
+      "rewards/margins": 0.07069384306669235,
+      "rewards/rejected": 0.02408282831311226,
       "step": 220
     },
     {
       "epoch": 0.6,
       "learning_rate": 2.0932279108998323e-06,
-      "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,
+      "logits/chosen": -1.9711300134658813,
+      "logits/rejected": -1.9751195907592773,
+      "logps/chosen": -32.86473846435547,
+      "logps/rejected": -31.40615463256836,
+      "loss": 0.4801,
+      "rewards/accuracies": 0.7124999761581421,
+      "rewards/chosen": 0.11054597795009613,
+      "rewards/margins": 0.0819164589047432,
+      "rewards/rejected": 0.02862953022122383,
       "step": 230
     },
     {
       "epoch": 0.62,
       "learning_rate": 1.8712423238279358e-06,
-      "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,
+      "logits/chosen": -1.9691890478134155,
+      "logits/rejected": -1.9474375247955322,
+      "logps/chosen": -33.62578201293945,
+      "logps/rejected": -34.88764953613281,
+      "loss": 0.4781,
+      "rewards/accuracies": 0.699999988079071,
+      "rewards/chosen": 0.10559363663196564,
+      "rewards/margins": 0.0929645225405693,
+      "rewards/rejected": 0.012629099190235138,
       "step": 240
     },
     {
       "epoch": 0.65,
       "learning_rate": 1.6544367689701824e-06,
-      "logits/chosen": -2.004868268966675,
-      "logits/rejected": -2.0015549659729004,
-      "logps/chosen": -32.729896545410156,
-      "logps/rejected": -36.27812957763672,
-      "loss": 0.6706,
+      "logits/chosen": -2.010312557220459,
+      "logits/rejected": -2.007035732269287,
+      "logps/chosen": -32.49589920043945,
+      "logps/rejected": -35.9986686706543,
+      "loss": 0.49,
       "rewards/accuracies": 0.5874999761581421,
-      "rewards/chosen": 0.03903444483876228,
-      "rewards/margins": 0.04967557638883591,
-      "rewards/rejected": -0.010641133412718773,
+      "rewards/chosen": 0.08583381026983261,
+      "rewards/margins": 0.040582992136478424,
+      "rewards/rejected": 0.04525081440806389,
       "step": 250
     },
     {
       "epoch": 0.68,
       "learning_rate": 1.4445974030621963e-06,
-      "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,
+      "logits/chosen": -1.8778746128082275,
+      "logits/rejected": -1.8754326105117798,
+      "logps/chosen": -33.7269287109375,
+      "logps/rejected": -35.267066955566406,
+      "loss": 0.4893,
+      "rewards/accuracies": 0.574999988079071,
+      "rewards/chosen": 0.09229358285665512,
+      "rewards/margins": 0.044782862067222595,
+      "rewards/rejected": 0.047510724514722824,
       "step": 260
     },
     {
       "epoch": 0.7,
       "learning_rate": 1.243452991757889e-06,
-      "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,
+      "logits/chosen": -1.8628604412078857,
+      "logits/rejected": -1.8603498935699463,
+      "logps/chosen": -33.917144775390625,
+      "logps/rejected": -31.5814151763916,
+      "loss": 0.4863,
+      "rewards/accuracies": 0.637499988079071,
+      "rewards/chosen": 0.09609059244394302,
+      "rewards/margins": 0.058107007294893265,
+      "rewards/rejected": 0.03798357769846916,
       "step": 270
     },
     {
       "epoch": 0.73,
       "learning_rate": 1.0526606671603523e-06,
-      "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": 0.07145937532186508,
-      "rewards/margins": 0.08328022807836533,
-      "rewards/rejected": -0.011820845305919647,
+      "logits/chosen": -1.9663734436035156,
+      "logits/rejected": -1.9559385776519775,
+      "logps/chosen": -34.753868103027344,
+      "logps/rejected": -31.635868072509766,
+      "loss": 0.4806,
+      "rewards/accuracies": 0.625,
+      "rewards/chosen": 0.1205538734793663,
+      "rewards/margins": 0.07952861487865448,
+      "rewards/rejected": 0.04102526605129242,
       "step": 280
     },
     {
       "epoch": 0.75,
       "learning_rate": 8.737922755071455e-07,
-      "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,
+      "logits/chosen": -2.062568187713623,
+      "logits/rejected": -2.047731399536133,
+      "logps/chosen": -30.416040420532227,
+      "logps/rejected": -32.359901428222656,
+      "loss": 0.4895,
+      "rewards/accuracies": 0.5874999761581421,
+      "rewards/chosen": 0.1016167551279068,
+      "rewards/margins": 0.04322956129908562,
+      "rewards/rejected": 0.05838719755411148,
       "step": 290
     },
     {
       "epoch": 0.78,
       "learning_rate": 7.08321427484816e-07,
-      "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,
+      "logits/chosen": -1.93375563621521,
+      "logits/rejected": -1.9313066005706787,
+      "logps/chosen": -32.096317291259766,
+      "logps/rejected": -30.669086456298828,
+      "loss": 0.4669,
+      "rewards/accuracies": 0.7250000238418579,
+      "rewards/chosen": 0.16615837812423706,
+      "rewards/margins": 0.13661329448223114,
+      "rewards/rejected": 0.029545078054070473,
       "step": 300
     },
     {
       "epoch": 0.78,
-      "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,
+      "eval_logits/chosen": -2.231231927871704,
+      "eval_logits/rejected": -2.226423978805542,
+      "eval_logps/chosen": -33.768104553222656,
+      "eval_logps/rejected": -37.29201889038086,
+      "eval_loss": 0.49796417355537415,
+      "eval_rewards/accuracies": 0.5157807469367981,
+      "eval_rewards/chosen": 0.05328937619924545,
+      "eval_rewards/margins": 0.008368566632270813,
+      "eval_rewards/rejected": 0.04492080584168434,
+      "eval_runtime": 145.91,
+      "eval_samples_per_second": 2.351,
+      "eval_steps_per_second": 0.295,
       "step": 300
     },
     {
       "epoch": 0.81,
       "learning_rate": 5.576113578589035e-07,
-      "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,
+      "logits/chosen": -1.9169145822525024,
+      "logits/rejected": -1.9137403964996338,
+      "logps/chosen": -31.04461097717285,
+      "logps/rejected": -33.57860565185547,
+      "loss": 0.4809,
+      "rewards/accuracies": 0.75,
+      "rewards/chosen": 0.11174802482128143,
+      "rewards/margins": 0.07903869450092316,
+      "rewards/rejected": 0.03270933777093887,
       "step": 310
     },
     {
       "epoch": 0.83,
       "learning_rate": 4.229036944380913e-07,
-      "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,
+      "logits/chosen": -1.9684025049209595,
+      "logits/rejected": -1.956284523010254,
+      "logps/chosen": -34.03736114501953,
+      "logps/rejected": -33.44590377807617,
+      "loss": 0.4778,
+      "rewards/accuracies": 0.762499988079071,
+      "rewards/chosen": 0.11174283176660538,
+      "rewards/margins": 0.09315498173236847,
+      "rewards/rejected": 0.018587838858366013,
       "step": 320
     },
     {
       "epoch": 0.86,
       "learning_rate": 3.053082288996112e-07,
-      "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,
+      "logits/chosen": -2.003418445587158,
+      "logits/rejected": -2.0020651817321777,
+      "logps/chosen": -32.87415313720703,
+      "logps/rejected": -32.24354553222656,
+      "loss": 0.4816,
+      "rewards/accuracies": 0.675000011920929,
+      "rewards/chosen": 0.12232844531536102,
+      "rewards/margins": 0.07512133568525314,
+      "rewards/rejected": 0.04720713198184967,
       "step": 330
     },
     {
       "epoch": 0.88,
       "learning_rate": 2.0579377374915805e-07,
-      "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,
+      "logits/chosen": -2.090510845184326,
+      "logits/rejected": -2.0748770236968994,
+      "logps/chosen": -33.48200607299805,
+      "logps/rejected": -32.82493209838867,
+      "loss": 0.4807,
+      "rewards/accuracies": 0.699999988079071,
+      "rewards/chosen": 0.13947655260562897,
+      "rewards/margins": 0.0808553546667099,
+      "rewards/rejected": 0.05862119793891907,
       "step": 340
     },
     {
       "epoch": 0.91,
       "learning_rate": 1.2518018074041684e-07,
-      "logits/chosen": -1.9557991027832031,
-      "logits/rejected": -1.9549639225006104,
-      "logps/chosen": -32.84474563598633,
-      "logps/rejected": -32.53619384765625,
-      "loss": 0.648,
+      "logits/chosen": -1.9624559879302979,
+      "logits/rejected": -1.9616531133651733,
+      "logps/chosen": -32.56671142578125,
+      "logps/rejected": -32.222007751464844,
+      "loss": 0.4774,
       "rewards/accuracies": 0.675000011920929,
-      "rewards/chosen": 0.0874704122543335,
-      "rewards/margins": 0.10167930275201797,
-      "rewards/rejected": -0.014208881184458733,
+      "rewards/chosen": 0.14307641983032227,
+      "rewards/margins": 0.09444761276245117,
+      "rewards/rejected": 0.04862881451845169,
       "step": 350
     },
     {
       "epoch": 0.94,
       "learning_rate": 6.41315865106129e-08,
-      "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,
+      "logits/chosen": -1.9191839694976807,
+      "logits/rejected": -1.9294923543930054,
+      "logps/chosen": -31.610462188720703,
+      "logps/rejected": -35.02184295654297,
+      "loss": 0.4841,
+      "rewards/accuracies": 0.675000011920929,
+      "rewards/chosen": 0.11884615570306778,
+      "rewards/margins": 0.06546333432197571,
+      "rewards/rejected": 0.05338282510638237,
       "step": 360
     },
     {
       "epoch": 0.96,
       "learning_rate": 2.3150941078050325e-08,
-      "logits/chosen": -2.0503978729248047,
-      "logits/rejected": -2.043942928314209,
-      "logps/chosen": -33.33507537841797,
-      "logps/rejected": -29.28571128845215,
-      "loss": 0.6582,
+      "logits/chosen": -2.0576701164245605,
+      "logits/rejected": -2.051162004470825,
+      "logps/chosen": -33.00985336303711,
+      "logps/rejected": -28.984561920166016,
+      "loss": 0.4803,
       "rewards/accuracies": 0.75,
-      "rewards/chosen": 0.06337906420230865,
-      "rewards/margins": 0.07523626834154129,
-      "rewards/rejected": -0.011857211589813232,
+      "rewards/chosen": 0.12842342257499695,
+      "rewards/margins": 0.08005058020353317,
+      "rewards/rejected": 0.04837283492088318,
       "step": 370
     },
     {
       "epoch": 0.99,
       "learning_rate": 2.575864278703266e-09,
-      "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,
+      "logits/chosen": -1.9184468984603882,
+      "logits/rejected": -1.9206174612045288,
+      "logps/chosen": -33.628562927246094,
+      "logps/rejected": -30.75335121154785,
+      "loss": 0.4766,
+      "rewards/accuracies": 0.75,
+      "rewards/chosen": 0.12209834903478622,
+      "rewards/margins": 0.09824265539646149,
+      "rewards/rejected": 0.023855695500969887,
       "step": 380
     },
     {
       "epoch": 1.0,
       "step": 385,
       "total_flos": 0.0,
-      "train_loss": 0.6689951481757226,
-      "train_runtime": 3249.3574,
-      "train_samples_per_second": 0.948,
+      "train_loss": 0.4876757522682091,
+      "train_runtime": 3250.1859,
+      "train_samples_per_second": 0.947,
       "train_steps_per_second": 0.118
     }
   ],