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: 0.9846
-- Rewards/chosen: -0.0137
-- Rewards/rejected: -0.0292
-- Rewards/accuracies: 0.5137
-- Rewards/margins: 0.0155
-- Logps/rejected: -37.5751
-- Logps/chosen: -34.0619
-- Logits/rejected: -2.2177
-- Logits/chosen: -2.2225
+- Loss: 0.6872
+- Rewards/chosen: -0.0004
+- Rewards/rejected: -0.0276
+- Rewards/accuracies: 0.5598
+- Rewards/margins: 0.0272
+- Logps/rejected: -37.5627
+- Logps/chosen: -34.0353
+- Logits/rejected: -2.2221
+- Logits/chosen: -2.2269
 
 ## 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 |
 |:-------------:|:-----:|:----:|:---------------:|:--------------:|:----------------:|:------------------:|:---------------:|:--------------:|:------------:|:---------------:|:-------------:|
-| 0.937         | 0.26  | 100  | 0.9921          | 0.0008         | -0.0073          | 0.5282             | 0.0081          | -37.5312       | -34.0329     | -2.2273         | -2.2322       |
-| 0.8332        | 0.52  | 200  | 0.9839          | -0.0016        | -0.0178          | 0.5307             | 0.0162          | -37.5522       | -34.0377     | -2.2233         | -2.2281       |
-| 0.7284        | 0.78  | 300  | 0.9828          | -0.0112        | -0.0284          | 0.5395             | 0.0172          | -37.5734       | -34.0569     | -2.2180         | -2.2228       |
+| 0.6595        | 0.26  | 100  | 0.6900          | 0.0170         | 0.0010           | 0.5253             | 0.0160          | -37.5150       | -34.0063     | -2.2287         | -2.2336       |
+| 0.6189        | 0.52  | 200  | 0.6895          | 0.0107         | -0.0106          | 0.5341             | 0.0213          | -37.5344       | -34.0167     | -2.2259         | -2.2307       |
+| 0.5582        | 0.78  | 300  | 0.6872          | -0.0004        | -0.0276          | 0.5598             | 0.0272          | -37.5627       | -34.0353     | -2.2221         | -2.2269       |
 
 
 ### Framework versions

adapter_model.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:b65d3844e391459fd84fa544e30dfe2778fe36d1c2060026037d3aa3b322e51d
+oid sha256:f6c4f495af7befc307023bb08f0b927438805515f61de16035bd2f7ed7db500c
 size 176183216

all_results.json

@@ -1,20 +1,7 @@
 {
     "epoch": 1.0,
-    "eval_logits/chosen": -2.2225441932678223,
-    "eval_logits/rejected": -2.217723846435547,
-    "eval_logps/chosen": -34.06193161010742,
-    "eval_logps/rejected": -37.57505416870117,
-    "eval_loss": 0.9846408367156982,
-    "eval_rewards/accuracies": 0.5137043595314026,
-    "eval_rewards/chosen": -0.013691178523004055,
-    "eval_rewards/margins": 0.015524487011134624,
-    "eval_rewards/rejected": -0.02921566367149353,
-    "eval_runtime": 145.6349,
-    "eval_samples": 343,
-    "eval_samples_per_second": 2.355,
-    "eval_steps_per_second": 0.295,
-    "train_loss": 0.8806782920639237,
-    "train_runtime": 3251.4322,
+    "train_loss": 0.6364745784115482,
+    "train_runtime": 3251.6479,
     "train_samples": 3079,
     "train_samples_per_second": 0.947,
     "train_steps_per_second": 0.118

train_results.json

@@ -1,7 +1,7 @@
 {
     "epoch": 1.0,
-    "train_loss": 0.8806782920639237,
-    "train_runtime": 3251.4322,
+    "train_loss": 0.6364745784115482,
+    "train_runtime": 3251.6479,
     "train_samples": 3079,
     "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": 1.0,
+      "loss": 0.6931,
       "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": -1.8664768934249878,
-      "logits/rejected": -1.8707948923110962,
-      "logps/chosen": -36.978759765625,
-      "logps/rejected": -33.655174255371094,
-      "loss": 0.9735,
-      "rewards/accuracies": 0.5555555820465088,
-      "rewards/chosen": 0.01391848549246788,
-      "rewards/margins": 0.026468267664313316,
-      "rewards/rejected": -0.012549781240522861,
+      "logits/chosen": -1.8668782711029053,
+      "logits/rejected": -1.8712005615234375,
+      "logps/chosen": -36.98646545410156,
+      "logps/rejected": -33.67870330810547,
+      "loss": 0.6747,
+      "rewards/accuracies": 0.5833333134651184,
+      "rewards/chosen": 0.012081857770681381,
+      "rewards/margins": 0.04126200079917908,
+      "rewards/rejected": -0.029180139303207397,
       "step": 10
     },
     {
       "epoch": 0.05,
       "learning_rate": 2.564102564102564e-06,
-      "logits/chosen": -1.9980016946792603,
-      "logits/rejected": -2.000636339187622,
-      "logps/chosen": -29.645702362060547,
-      "logps/rejected": -29.045913696289062,
-      "loss": 1.0105,
-      "rewards/accuracies": 0.38749998807907104,
-      "rewards/chosen": -0.0017539471155032516,
-      "rewards/margins": -0.010518571361899376,
-      "rewards/rejected": 0.00876462459564209,
+      "logits/chosen": -1.9979515075683594,
+      "logits/rejected": -2.0005881786346436,
+      "logps/chosen": -29.662744522094727,
+      "logps/rejected": -29.051654815673828,
+      "loss": 0.7042,
+      "rewards/accuracies": 0.4000000059604645,
+      "rewards/chosen": -0.01232814695686102,
+      "rewards/margins": -0.019403135403990746,
+      "rewards/rejected": 0.007074988447129726,
       "step": 20
     },
     {
       "epoch": 0.08,
       "learning_rate": 3.846153846153847e-06,
-      "logits/chosen": -1.920964002609253,
-      "logits/rejected": -1.9182822704315186,
-      "logps/chosen": -31.409515380859375,
-      "logps/rejected": -33.23186492919922,
-      "loss": 0.99,
-      "rewards/accuracies": 0.5249999761581421,
-      "rewards/chosen": 0.0032895475160330534,
-      "rewards/margins": 0.009993741288781166,
-      "rewards/rejected": -0.006704193539917469,
+      "logits/chosen": -1.921088457107544,
+      "logits/rejected": -1.9184081554412842,
+      "logps/chosen": -31.383258819580078,
+      "logps/rejected": -33.23828887939453,
+      "loss": 0.6794,
+      "rewards/accuracies": 0.5874999761581421,
+      "rewards/chosen": 0.019701367244124413,
+      "rewards/margins": 0.03159697726368904,
+      "rewards/rejected": -0.011895612813532352,
       "step": 30
     },
     {
       "epoch": 0.1,
       "learning_rate": 4.999896948438434e-06,
-      "logits/chosen": -2.0180246829986572,
-      "logits/rejected": -2.009291410446167,
-      "logps/chosen": -32.584468841552734,
-      "logps/rejected": -32.50810623168945,
-      "loss": 1.0061,
-      "rewards/accuracies": 0.44999998807907104,
-      "rewards/chosen": -0.0038314342964440584,
-      "rewards/margins": -0.006112524773925543,
-      "rewards/rejected": 0.002281090710312128,
+      "logits/chosen": -2.017509937286377,
+      "logits/rejected": -2.0087647438049316,
+      "logps/chosen": -32.577518463134766,
+      "logps/rejected": -32.509830474853516,
+      "loss": 0.6964,
+      "rewards/accuracies": 0.48750001192092896,
+      "rewards/chosen": -0.00042539089918136597,
+      "rewards/margins": -0.0021285698749125004,
+      "rewards/rejected": 0.001703177229501307,
       "step": 40
     },
     {
       "epoch": 0.13,
       "learning_rate": 4.987541037542187e-06,
-      "logits/chosen": -1.8629133701324463,
-      "logits/rejected": -1.852142572402954,
-      "logps/chosen": -33.5595703125,
-      "logps/rejected": -35.455909729003906,
-      "loss": 0.9993,
-      "rewards/accuracies": 0.42500001192092896,
-      "rewards/chosen": -0.0010461390484124422,
-      "rewards/margins": 0.0007355377310886979,
-      "rewards/rejected": -0.0017816766630858183,
+      "logits/chosen": -1.8623021841049194,
+      "logits/rejected": -1.8515303134918213,
+      "logps/chosen": -33.56303787231445,
+      "logps/rejected": -35.47795867919922,
+      "loss": 0.6896,
+      "rewards/accuracies": 0.5249999761581421,
+      "rewards/chosen": -0.0033326249103993177,
+      "rewards/margins": 0.012034483253955841,
+      "rewards/rejected": -0.015367108397185802,
       "step": 50
     },
     {
       "epoch": 0.16,
       "learning_rate": 4.954691471941119e-06,
-      "logits/chosen": -1.9417911767959595,
-      "logits/rejected": -1.9437233209609985,
-      "logps/chosen": -32.56698989868164,
-      "logps/rejected": -33.20896911621094,
-      "loss": 0.9691,
-      "rewards/accuracies": 0.5625,
-      "rewards/chosen": 0.01659708470106125,
-      "rewards/margins": 0.03539115563035011,
-      "rewards/rejected": -0.018794070929288864,
+      "logits/chosen": -1.9411529302597046,
+      "logits/rejected": -1.943098783493042,
+      "logps/chosen": -32.549232482910156,
+      "logps/rejected": -33.20621109008789,
+      "loss": 0.6747,
+      "rewards/accuracies": 0.5249999761581421,
+      "rewards/chosen": 0.030568569898605347,
+      "rewards/margins": 0.051466990262269974,
+      "rewards/rejected": -0.02089841663837433,
       "step": 60
     },
     {
       "epoch": 0.18,
       "learning_rate": 4.901618883413549e-06,
-      "logits/chosen": -2.0726003646850586,
-      "logits/rejected": -2.077578544616699,
-      "logps/chosen": -33.98097610473633,
-      "logps/rejected": -36.61317825317383,
-      "loss": 0.9741,
-      "rewards/accuracies": 0.5625,
-      "rewards/chosen": 0.0037565291859209538,
-      "rewards/margins": 0.02590828575193882,
-      "rewards/rejected": -0.022151757031679153,
+      "logits/chosen": -2.0727076530456543,
+      "logits/rejected": -2.07767391204834,
+      "logps/chosen": -33.96394729614258,
+      "logps/rejected": -36.61058807373047,
+      "loss": 0.6813,
+      "rewards/accuracies": 0.5249999761581421,
+      "rewards/chosen": 0.014724211767315865,
+      "rewards/margins": 0.03975607082247734,
+      "rewards/rejected": -0.025031859055161476,
       "step": 70
     },
     {
       "epoch": 0.21,
       "learning_rate": 4.828760511501322e-06,
-      "logits/chosen": -1.9339592456817627,
-      "logits/rejected": -1.9370877742767334,
-      "logps/chosen": -34.31932830810547,
-      "logps/rejected": -34.65187454223633,
-      "loss": 0.9169,
-      "rewards/accuracies": 0.6499999761581421,
-      "rewards/chosen": 0.05424312874674797,
-      "rewards/margins": 0.08397696167230606,
-      "rewards/rejected": -0.029733825474977493,
+      "logits/chosen": -1.9330183267593384,
+      "logits/rejected": -1.936171293258667,
+      "logps/chosen": -34.318397521972656,
+      "logps/rejected": -34.61701583862305,
+      "loss": 0.6634,
+      "rewards/accuracies": 0.550000011920929,
+      "rewards/chosen": 0.06564848870038986,
+      "rewards/margins": 0.08041460067033768,
+      "rewards/rejected": -0.014766111969947815,
       "step": 80
     },
     {
       "epoch": 0.23,
       "learning_rate": 4.7367166013034295e-06,
-      "logits/chosen": -1.9409306049346924,
-      "logits/rejected": -1.9454383850097656,
-      "logps/chosen": -32.3923454284668,
-      "logps/rejected": -32.371299743652344,
-      "loss": 0.9563,
-      "rewards/accuracies": 0.5625,
-      "rewards/chosen": 0.037537358701229095,
-      "rewards/margins": 0.043734077364206314,
-      "rewards/rejected": -0.006196716334670782,
+      "logits/chosen": -1.9412405490875244,
+      "logits/rejected": -1.945755958557129,
+      "logps/chosen": -32.374977111816406,
+      "logps/rejected": -32.33773422241211,
+      "loss": 0.6787,
+      "rewards/accuracies": 0.6499999761581421,
+      "rewards/chosen": 0.05546582490205765,
+      "rewards/margins": 0.04276125878095627,
+      "rewards/rejected": 0.012704563327133656,
       "step": 90
     },
     {
       "epoch": 0.26,
       "learning_rate": 4.626245458345211e-06,
-      "logits/chosen": -2.037719964981079,
-      "logits/rejected": -2.0357348918914795,
-      "logps/chosen": -32.144508361816406,
-      "logps/rejected": -31.295955657958984,
-      "loss": 0.937,
-      "rewards/accuracies": 0.637499988079071,
-      "rewards/chosen": 0.04816383868455887,
-      "rewards/margins": 0.06303636729717255,
-      "rewards/rejected": -0.014872525818645954,
+      "logits/chosen": -2.0381903648376465,
+      "logits/rejected": -2.0362119674682617,
+      "logps/chosen": -32.13701629638672,
+      "logps/rejected": -31.295801162719727,
+      "loss": 0.6595,
+      "rewards/accuracies": 0.6499999761581421,
+      "rewards/chosen": 0.06229018047451973,
+      "rewards/margins": 0.08004496991634369,
+      "rewards/rejected": -0.017754793167114258,
       "step": 100
     },
     {
       "epoch": 0.26,
-      "eval_logits/chosen": -2.2321972846984863,
-      "eval_logits/rejected": -2.2273480892181396,
-      "eval_logps/chosen": -34.032928466796875,
-      "eval_logps/rejected": -37.53125,
-      "eval_loss": 0.9921085834503174,
-      "eval_rewards/accuracies": 0.5282392501831055,
-      "eval_rewards/chosen": 0.000813809223473072,
-      "eval_rewards/margins": 0.008130725473165512,
-      "eval_rewards/rejected": -0.007316915784031153,
-      "eval_runtime": 145.8006,
+      "eval_logits/chosen": -2.2335634231567383,
+      "eval_logits/rejected": -2.228721857070923,
+      "eval_logps/chosen": -34.00627899169922,
+      "eval_logps/rejected": -37.514984130859375,
+      "eval_loss": 0.6900185346603394,
+      "eval_rewards/accuracies": 0.5253322720527649,
+      "eval_rewards/chosen": 0.016964510083198547,
+      "eval_rewards/margins": 0.015982570126652718,
+      "eval_rewards/rejected": 0.0009819410042837262,
+      "eval_runtime": 145.7969,
       "eval_samples_per_second": 2.353,
       "eval_steps_per_second": 0.295,
       "step": 100
@@ -181,433 +181,433 @@
     {
       "epoch": 0.29,
       "learning_rate": 4.498257201263691e-06,
-      "logits/chosen": -1.9925286769866943,
-      "logits/rejected": -1.9901431798934937,
-      "logps/chosen": -33.11371994018555,
-      "logps/rejected": -34.028839111328125,
-      "loss": 0.9352,
-      "rewards/accuracies": 0.6000000238418579,
-      "rewards/chosen": 0.06504794210195541,
-      "rewards/margins": 0.06478062272071838,
-      "rewards/rejected": 0.0002673208655323833,
+      "logits/chosen": -1.9929395914077759,
+      "logits/rejected": -1.99057936668396,
+      "logps/chosen": -33.06591796875,
+      "logps/rejected": -34.01008605957031,
+      "loss": 0.6678,
+      "rewards/accuracies": 0.675000011920929,
+      "rewards/chosen": 0.10673947632312775,
+      "rewards/margins": 0.09516827017068863,
+      "rewards/rejected": 0.011571208015084267,
       "step": 110
     },
     {
       "epoch": 0.31,
       "learning_rate": 4.353806263777678e-06,
-      "logits/chosen": -2.0043208599090576,
-      "logits/rejected": -1.9959907531738281,
-      "logps/chosen": -32.34038162231445,
-      "logps/rejected": -32.120628356933594,
-      "loss": 0.9686,
-      "rewards/accuracies": 0.5249999761581421,
-      "rewards/chosen": 0.05226168781518936,
-      "rewards/margins": 0.03182316944003105,
-      "rewards/rejected": 0.02043851651251316,
+      "logits/chosen": -2.0045437812805176,
+      "logits/rejected": -1.9962146282196045,
+      "logps/chosen": -32.33503341674805,
+      "logps/rejected": -32.12450408935547,
+      "loss": 0.68,
+      "rewards/accuracies": 0.5625,
+      "rewards/chosen": 0.06592197716236115,
+      "rewards/margins": 0.043721526861190796,
+      "rewards/rejected": 0.02220045030117035,
       "step": 120
     },
     {
       "epoch": 0.34,
       "learning_rate": 4.1940827077152755e-06,
-      "logits/chosen": -2.031864881515503,
-      "logits/rejected": -2.023920774459839,
-      "logps/chosen": -30.3155460357666,
-      "logps/rejected": -32.07514190673828,
-      "loss": 0.9237,
-      "rewards/accuracies": 0.574999988079071,
-      "rewards/chosen": 0.08008603006601334,
-      "rewards/margins": 0.09115206450223923,
-      "rewards/rejected": -0.011066052131354809,
+      "logits/chosen": -2.032952308654785,
+      "logits/rejected": -2.024994134902954,
+      "logps/chosen": -30.302433013916016,
+      "logps/rejected": -32.04313278198242,
+      "loss": 0.6609,
+      "rewards/accuracies": 0.612500011920929,
+      "rewards/chosen": 0.10397056490182877,
+      "rewards/margins": 0.09804753214120865,
+      "rewards/rejected": 0.0059230271726846695,
       "step": 130
     },
     {
       "epoch": 0.36,
       "learning_rate": 4.0204024186666215e-06,
-      "logits/chosen": -1.9618200063705444,
-      "logits/rejected": -1.9720451831817627,
-      "logps/chosen": -31.228313446044922,
-      "logps/rejected": -32.562313079833984,
-      "loss": 0.8906,
-      "rewards/accuracies": 0.675000011920929,
-      "rewards/chosen": 0.09567264467477798,
-      "rewards/margins": 0.10938718169927597,
-      "rewards/rejected": -0.01371453981846571,
+      "logits/chosen": -1.9635902643203735,
+      "logits/rejected": -1.9738080501556396,
+      "logps/chosen": -31.201534271240234,
+      "logps/rejected": -32.556739807128906,
+      "loss": 0.6343,
+      "rewards/accuracies": 0.6875,
+      "rewards/chosen": 0.13087674975395203,
+      "rewards/margins": 0.1439923495054245,
+      "rewards/rejected": -0.013115609996020794,
       "step": 140
     },
     {
       "epoch": 0.39,
       "learning_rate": 3.834196265035119e-06,
-      "logits/chosen": -1.8728482723236084,
-      "logits/rejected": -1.874018669128418,
-      "logps/chosen": -33.88795852661133,
-      "logps/rejected": -34.77352523803711,
-      "loss": 0.852,
-      "rewards/accuracies": 0.637499988079071,
-      "rewards/chosen": 0.15236759185791016,
-      "rewards/margins": 0.16050228476524353,
-      "rewards/rejected": -0.00813469011336565,
+      "logits/chosen": -1.8755298852920532,
+      "logits/rejected": -1.876691222190857,
+      "logps/chosen": -33.89242172241211,
+      "logps/rejected": -34.753570556640625,
+      "loss": 0.6272,
+      "rewards/accuracies": 0.612500011920929,
+      "rewards/chosen": 0.1801605522632599,
+      "rewards/margins": 0.17795029282569885,
+      "rewards/rejected": 0.002210266888141632,
       "step": 150
     },
     {
       "epoch": 0.42,
       "learning_rate": 3.636998309800573e-06,
-      "logits/chosen": -1.923365592956543,
-      "logits/rejected": -1.9199578762054443,
-      "logps/chosen": -35.98258590698242,
-      "logps/rejected": -32.678192138671875,
-      "loss": 0.9277,
-      "rewards/accuracies": 0.625,
-      "rewards/chosen": 0.09454482793807983,
-      "rewards/margins": 0.07226438820362091,
-      "rewards/rejected": 0.02228044718503952,
+      "logits/chosen": -1.9265025854110718,
+      "logits/rejected": -1.9230976104736328,
+      "logps/chosen": -35.98413848876953,
+      "logps/rejected": -32.69154357910156,
+      "loss": 0.6574,
+      "rewards/accuracies": 0.574999988079071,
+      "rewards/chosen": 0.11252401769161224,
+      "rewards/margins": 0.09379850327968597,
+      "rewards/rejected": 0.01872551441192627,
       "step": 160
     },
     {
       "epoch": 0.44,
       "learning_rate": 3.4304331721118078e-06,
-      "logits/chosen": -2.0240864753723145,
-      "logits/rejected": -2.0167737007141113,
-      "logps/chosen": -33.471466064453125,
-      "logps/rejected": -31.40371322631836,
-      "loss": 0.8154,
-      "rewards/accuracies": 0.6625000238418579,
-      "rewards/chosen": 0.16319802403450012,
-      "rewards/margins": 0.18913087248802185,
-      "rewards/rejected": -0.025932863354682922,
+      "logits/chosen": -2.0262577533721924,
+      "logits/rejected": -2.0189335346221924,
+      "logps/chosen": -33.45969772338867,
+      "logps/rejected": -31.372516632080078,
+      "loss": 0.6076,
+      "rewards/accuracies": 0.675000011920929,
+      "rewards/chosen": 0.20289841294288635,
+      "rewards/margins": 0.21529710292816162,
+      "rewards/rejected": -0.012398697435855865,
       "step": 170
     },
     {
       "epoch": 0.47,
       "learning_rate": 3.2162026428305436e-06,
-      "logits/chosen": -2.0294885635375977,
-      "logits/rejected": -2.034759044647217,
-      "logps/chosen": -32.19446563720703,
-      "logps/rejected": -32.4196891784668,
-      "loss": 0.8617,
-      "rewards/accuracies": 0.6875,
-      "rewards/chosen": 0.17516280710697174,
-      "rewards/margins": 0.1424373984336853,
-      "rewards/rejected": 0.03272542357444763,
+      "logits/chosen": -2.0338973999023438,
+      "logits/rejected": -2.0391509532928467,
+      "logps/chosen": -32.194793701171875,
+      "logps/rejected": -32.42069625854492,
+      "loss": 0.6221,
+      "rewards/accuracies": 0.737500011920929,
+      "rewards/chosen": 0.20999665558338165,
+      "rewards/margins": 0.17132976651191711,
+      "rewards/rejected": 0.03866690397262573,
       "step": 180
     },
     {
       "epoch": 0.49,
       "learning_rate": 2.996071664294641e-06,
-      "logits/chosen": -2.0309464931488037,
-      "logits/rejected": -2.028186798095703,
-      "logps/chosen": -31.26650619506836,
-      "logps/rejected": -31.340648651123047,
-      "loss": 0.8776,
-      "rewards/accuracies": 0.6875,
-      "rewards/chosen": 0.1112222671508789,
-      "rewards/margins": 0.1224404126405716,
-      "rewards/rejected": -0.011218142695724964,
+      "logits/chosen": -2.035045862197876,
+      "logits/rejected": -2.032278060913086,
+      "logps/chosen": -31.230976104736328,
+      "logps/rejected": -31.29391860961914,
+      "loss": 0.6403,
+      "rewards/accuracies": 0.637499988079071,
+      "rewards/chosen": 0.1547853648662567,
+      "rewards/margins": 0.14020755887031555,
+      "rewards/rejected": 0.014577840454876423,
       "step": 190
     },
     {
       "epoch": 0.52,
       "learning_rate": 2.7718537898066833e-06,
-      "logits/chosen": -1.9006068706512451,
-      "logits/rejected": -1.9052391052246094,
-      "logps/chosen": -31.295886993408203,
-      "logps/rejected": -32.81926345825195,
-      "loss": 0.8332,
+      "logits/chosen": -1.9042125940322876,
+      "logits/rejected": -1.9088417291641235,
+      "logps/chosen": -31.285167694091797,
+      "logps/rejected": -32.79944610595703,
+      "loss": 0.6189,
       "rewards/accuracies": 0.737500011920929,
-      "rewards/chosen": 0.15519794821739197,
-      "rewards/margins": 0.16936282813549042,
-      "rewards/rejected": -0.01416487991809845,
+      "rewards/chosen": 0.192669078707695,
+      "rewards/margins": 0.1977730393409729,
+      "rewards/rejected": -0.0051039643585681915,
       "step": 200
     },
     {
       "epoch": 0.52,
-      "eval_logits/chosen": -2.2281434535980225,
-      "eval_logits/rejected": -2.223317861557007,
-      "eval_logps/chosen": -34.03767013549805,
-      "eval_logps/rejected": -37.552223205566406,
-      "eval_loss": 0.983903169631958,
-      "eval_rewards/accuracies": 0.5307309031486511,
-      "eval_rewards/chosen": -0.001558887423016131,
-      "eval_rewards/margins": 0.016243286430835724,
-      "eval_rewards/rejected": -0.017802175134420395,
-      "eval_runtime": 145.6923,
-      "eval_samples_per_second": 2.354,
+      "eval_logits/chosen": -2.230701446533203,
+      "eval_logits/rejected": -2.2258734703063965,
+      "eval_logps/chosen": -34.0167350769043,
+      "eval_logps/rejected": -37.53435134887695,
+      "eval_loss": 0.6894960403442383,
+      "eval_rewards/accuracies": 0.5340532064437866,
+      "eval_rewards/chosen": 0.010691030882298946,
+      "eval_rewards/margins": 0.02133062295615673,
+      "eval_rewards/rejected": -0.010639593005180359,
+      "eval_runtime": 145.5559,
+      "eval_samples_per_second": 2.356,
       "eval_steps_per_second": 0.295,
       "step": 200
     },
     {
       "epoch": 0.55,
       "learning_rate": 2.5453962426402006e-06,
-      "logits/chosen": -2.013721466064453,
-      "logits/rejected": -2.0243561267852783,
-      "logps/chosen": -31.78135108947754,
-      "logps/rejected": -33.93622970581055,
-      "loss": 0.8758,
+      "logits/chosen": -2.018108367919922,
+      "logits/rejected": -2.0287561416625977,
+      "logps/chosen": -31.745798110961914,
+      "logps/rejected": -33.90629577636719,
+      "loss": 0.6305,
       "rewards/accuracies": 0.675000011920929,
-      "rewards/chosen": 0.09497732669115067,
-      "rewards/margins": 0.1317446529865265,
-      "rewards/rejected": -0.036767326295375824,
+      "rewards/chosen": 0.13530384004116058,
+      "rewards/margins": 0.16146349906921387,
+      "rewards/rejected": -0.026159662753343582,
       "step": 210
     },
     {
       "epoch": 0.57,
       "learning_rate": 2.3185646976551794e-06,
-      "logits/chosen": -1.9061295986175537,
-      "logits/rejected": -1.9209282398223877,
-      "logps/chosen": -29.796966552734375,
-      "logps/rejected": -31.621959686279297,
-      "loss": 0.8123,
-      "rewards/accuracies": 0.7124999761581421,
-      "rewards/chosen": 0.15728648006916046,
-      "rewards/margins": 0.18917784094810486,
-      "rewards/rejected": -0.03189137205481529,
+      "logits/chosen": -1.9105422496795654,
+      "logits/rejected": -1.9253056049346924,
+      "logps/chosen": -29.849069595336914,
+      "logps/rejected": -31.58382797241211,
+      "loss": 0.6251,
+      "rewards/accuracies": 0.6625000238418579,
+      "rewards/chosen": 0.15748274326324463,
+      "rewards/margins": 0.17287404835224152,
+      "rewards/rejected": -0.015391310676932335,
       "step": 220
     },
     {
       "epoch": 0.6,
       "learning_rate": 2.0932279108998323e-06,
-      "logits/chosen": -1.9621093273162842,
-      "logits/rejected": -1.9660680294036865,
-      "logps/chosen": -33.10811233520508,
-      "logps/rejected": -31.64517593383789,
-      "loss": 0.8222,
-      "rewards/accuracies": 0.675000011920929,
-      "rewards/chosen": 0.15467897057533264,
-      "rewards/margins": 0.20261602103710175,
-      "rewards/rejected": -0.04793703928589821,
+      "logits/chosen": -1.966581106185913,
+      "logits/rejected": -1.9705555438995361,
+      "logps/chosen": -33.059837341308594,
+      "logps/rejected": -31.605152130126953,
+      "loss": 0.6,
+      "rewards/accuracies": 0.7124999761581421,
+      "rewards/chosen": 0.21457910537719727,
+      "rewards/margins": 0.2480895221233368,
+      "rewards/rejected": -0.033510446548461914,
       "step": 230
     },
     {
       "epoch": 0.62,
       "learning_rate": 1.8712423238279358e-06,
-      "logits/chosen": -1.9597173929214478,
-      "logits/rejected": -1.9379403591156006,
-      "logps/chosen": -33.83074188232422,
-      "logps/rejected": -35.11478042602539,
-      "loss": 0.7755,
-      "rewards/accuracies": 0.7124999761581421,
-      "rewards/chosen": 0.16150443255901337,
-      "rewards/margins": 0.24349848926067352,
-      "rewards/rejected": -0.08199407160282135,
+      "logits/chosen": -1.9644801616668701,
+      "logits/rejected": -1.9426641464233398,
+      "logps/chosen": -33.819881439208984,
+      "logps/rejected": -35.105430603027344,
+      "loss": 0.5826,
+      "rewards/accuracies": 0.7250000238418579,
+      "rewards/chosen": 0.20031993091106415,
+      "rewards/margins": 0.2930986285209656,
+      "rewards/rejected": -0.09277870506048203,
       "step": 240
     },
     {
       "epoch": 0.65,
       "learning_rate": 1.6544367689701824e-06,
-      "logits/chosen": -2.001469135284424,
-      "logits/rejected": -1.998147964477539,
-      "logps/chosen": -32.717472076416016,
-      "logps/rejected": -36.230865478515625,
-      "loss": 0.8965,
-      "rewards/accuracies": 0.5375000238418579,
-      "rewards/chosen": 0.10379841178655624,
-      "rewards/margins": 0.10677065700292587,
-      "rewards/rejected": -0.0029722482431679964,
+      "logits/chosen": -2.005814552307129,
+      "logits/rejected": -2.002516746520996,
+      "logps/chosen": -32.663055419921875,
+      "logps/rejected": -36.233436584472656,
+      "loss": 0.632,
+      "rewards/accuracies": 0.6625000238418579,
+      "rewards/chosen": 0.15720662474632263,
+      "rewards/margins": 0.16231317818164825,
+      "rewards/rejected": -0.0051065413281321526,
       "step": 250
     },
     {
       "epoch": 0.68,
       "learning_rate": 1.4445974030621963e-06,
-      "logits/chosen": -1.8687664270401,
-      "logits/rejected": -1.8663409948349,
-      "logps/chosen": -33.96764373779297,
-      "logps/rejected": -35.5003776550293,
-      "loss": 0.8917,
-      "rewards/accuracies": 0.637499988079071,
-      "rewards/chosen": 0.11037871986627579,
-      "rewards/margins": 0.10825767368078232,
-      "rewards/rejected": 0.0021210461854934692,
+      "logits/chosen": -1.8735284805297852,
+      "logits/rejected": -1.8710968494415283,
+      "logps/chosen": -33.964805603027344,
+      "logps/rejected": -35.50111389160156,
+      "loss": 0.6428,
+      "rewards/accuracies": 0.6499999761581421,
+      "rewards/chosen": 0.13415411114692688,
+      "rewards/margins": 0.13204893469810486,
+      "rewards/rejected": 0.0021051731891930103,
       "step": 260
     },
     {
       "epoch": 0.7,
       "learning_rate": 1.243452991757889e-06,
-      "logits/chosen": -1.8529716730117798,
-      "logits/rejected": -1.8505760431289673,
-      "logps/chosen": -34.18817901611328,
-      "logps/rejected": -31.831729888916016,
-      "loss": 0.8701,
-      "rewards/accuracies": 0.574999988079071,
-      "rewards/chosen": 0.1047089546918869,
-      "rewards/margins": 0.1349087655544281,
-      "rewards/rejected": -0.0301998071372509,
+      "logits/chosen": -1.8585532903671265,
+      "logits/rejected": -1.8561296463012695,
+      "logps/chosen": -34.17288589477539,
+      "logps/rejected": -31.792491912841797,
+      "loss": 0.6391,
+      "rewards/accuracies": 0.6875,
+      "rewards/chosen": 0.13482868671417236,
+      "rewards/margins": 0.14752644300460815,
+      "rewards/rejected": -0.012697766534984112,
       "step": 270
     },
     {
       "epoch": 0.73,
       "learning_rate": 1.0526606671603523e-06,
-      "logits/chosen": -1.9559904336929321,
-      "logits/rejected": -1.9455276727676392,
-      "logps/chosen": -34.98174285888672,
-      "logps/rejected": -31.862768173217773,
-      "loss": 0.8017,
-      "rewards/accuracies": 0.7250000238418579,
-      "rewards/chosen": 0.18744894862174988,
-      "rewards/margins": 0.1983364075422287,
-      "rewards/rejected": -0.010887444019317627,
+      "logits/chosen": -1.9624073505401611,
+      "logits/rejected": -1.9519150257110596,
+      "logps/chosen": -34.95975875854492,
+      "logps/rejected": -31.84967041015625,
+      "loss": 0.5947,
+      "rewards/accuracies": 0.7749999761581421,
+      "rewards/chosen": 0.2381301373243332,
+      "rewards/margins": 0.2433377206325531,
+      "rewards/rejected": -0.005207589361816645,
       "step": 280
     },
     {
       "epoch": 0.75,
       "learning_rate": 8.737922755071455e-07,
-      "logits/chosen": -2.0518603324890137,
-      "logits/rejected": -2.0369694232940674,
-      "logps/chosen": -30.698474884033203,
-      "logps/rejected": -32.61089324951172,
-      "loss": 0.9111,
-      "rewards/accuracies": 0.612500011920929,
-      "rewards/chosen": 0.11282563209533691,
-      "rewards/margins": 0.09235484153032303,
-      "rewards/rejected": 0.02047080174088478,
+      "logits/chosen": -2.056440830230713,
+      "logits/rejected": -2.0415446758270264,
+      "logps/chosen": -30.704137802124023,
+      "logps/rejected": -32.5970458984375,
+      "loss": 0.6654,
+      "rewards/accuracies": 0.5625,
+      "rewards/chosen": 0.13198992609977722,
+      "rewards/margins": 0.09911760687828064,
+      "rewards/rejected": 0.03287229686975479,
       "step": 290
     },
     {
       "epoch": 0.78,
       "learning_rate": 7.08321427484816e-07,
-      "logits/chosen": -1.9218803644180298,
-      "logits/rejected": -1.9193775653839111,
-      "logps/chosen": -32.38130569458008,
-      "logps/rejected": -30.906925201416016,
-      "loss": 0.7284,
-      "rewards/accuracies": 0.800000011920929,
-      "rewards/chosen": 0.27290263772010803,
-      "rewards/margins": 0.3179616630077362,
-      "rewards/rejected": -0.04505903273820877,
+      "logits/chosen": -1.927122712135315,
+      "logits/rejected": -1.924602746963501,
+      "logps/chosen": -32.37379455566406,
+      "logps/rejected": -30.879268646240234,
+      "loss": 0.5582,
+      "rewards/accuracies": 0.7124999761581421,
+      "rewards/chosen": 0.3319894075393677,
+      "rewards/margins": 0.3694665729999542,
+      "rewards/rejected": -0.03747714310884476,
       "step": 300
     },
     {
       "epoch": 0.78,
-      "eval_logits/chosen": -2.222837209701538,
-      "eval_logits/rejected": -2.218010663986206,
-      "eval_logps/chosen": -34.05692672729492,
-      "eval_logps/rejected": -37.573402404785156,
-      "eval_loss": 0.9828158617019653,
-      "eval_rewards/accuracies": 0.5394518375396729,
-      "eval_rewards/chosen": -0.011187080293893814,
-      "eval_rewards/margins": 0.017203383147716522,
-      "eval_rewards/rejected": -0.028390465304255486,
-      "eval_runtime": 145.822,
-      "eval_samples_per_second": 2.352,
+      "eval_logits/chosen": -2.226900815963745,
+      "eval_logits/rejected": -2.2220816612243652,
+      "eval_logps/chosen": -34.03526306152344,
+      "eval_logps/rejected": -37.56266403198242,
+      "eval_loss": 0.6872054934501648,
+      "eval_rewards/accuracies": 0.5598006844520569,
+      "eval_rewards/chosen": -0.0004257837135810405,
+      "eval_rewards/margins": 0.027201363816857338,
+      "eval_rewards/rejected": -0.027627145871520042,
+      "eval_runtime": 145.6667,
+      "eval_samples_per_second": 2.355,
       "eval_steps_per_second": 0.295,
       "step": 300
     },
     {
       "epoch": 0.81,
       "learning_rate": 5.576113578589035e-07,
-      "logits/chosen": -1.9078595638275146,
-      "logits/rejected": -1.90461003780365,
-      "logps/chosen": -31.29986000061035,
-      "logps/rejected": -33.81876754760742,
-      "loss": 0.8259,
-      "rewards/accuracies": 0.737500011920929,
-      "rewards/chosen": 0.15174436569213867,
-      "rewards/margins": 0.19005274772644043,
-      "rewards/rejected": -0.03830839321017265,
+      "logits/chosen": -1.9124987125396729,
+      "logits/rejected": -1.9092620611190796,
+      "logps/chosen": -31.285348892211914,
+      "logps/rejected": -33.75069046020508,
+      "loss": 0.6188,
+      "rewards/accuracies": 0.675000011920929,
+      "rewards/chosen": 0.19080111384391785,
+      "rewards/margins": 0.1959228217601776,
+      "rewards/rejected": -0.0051217032596468925,
       "step": 310
     },
     {
       "epoch": 0.83,
       "learning_rate": 4.229036944380913e-07,
-      "logits/chosen": -1.957113265991211,
-      "logits/rejected": -1.9449272155761719,
-      "logps/chosen": -34.27944564819336,
-      "logps/rejected": -33.6619873046875,
-      "loss": 0.7968,
-      "rewards/accuracies": 0.6499999761581421,
-      "rewards/chosen": 0.15831336379051208,
-      "rewards/margins": 0.2198868989944458,
-      "rewards/rejected": -0.06157354637980461,
+      "logits/chosen": -1.9616212844848633,
+      "logits/rejected": -1.9494349956512451,
+      "logps/chosen": -34.30357360839844,
+      "logps/rejected": -33.64542770385742,
+      "loss": 0.5999,
+      "rewards/accuracies": 0.6875,
+      "rewards/chosen": 0.17549821734428406,
+      "rewards/margins": 0.23945149779319763,
+      "rewards/rejected": -0.06395327299833298,
       "step": 320
     },
     {
       "epoch": 0.86,
       "learning_rate": 3.053082288996112e-07,
-      "logits/chosen": -1.9918349981307983,
-      "logits/rejected": -1.9904123544692993,
-      "logps/chosen": -33.15383529663086,
-      "logps/rejected": -32.51991653442383,
-      "loss": 0.8186,
-      "rewards/accuracies": 0.6875,
-      "rewards/chosen": 0.16597984731197357,
-      "rewards/margins": 0.18614766001701355,
-      "rewards/rejected": -0.020167803391814232,
+      "logits/chosen": -1.997859239578247,
+      "logits/rejected": -1.9964158535003662,
+      "logps/chosen": -33.130615234375,
+      "logps/rejected": -32.511531829833984,
+      "loss": 0.6073,
+      "rewards/accuracies": 0.7124999761581421,
+      "rewards/chosen": 0.21310487389564514,
+      "rewards/margins": 0.23227325081825256,
+      "rewards/rejected": -0.01916835829615593,
       "step": 330
     },
     {
       "epoch": 0.88,
       "learning_rate": 2.0579377374915805e-07,
-      "logits/chosen": -2.0784294605255127,
-      "logits/rejected": -2.0627849102020264,
-      "logps/chosen": -33.7851448059082,
-      "logps/rejected": -33.072959899902344,
-      "loss": 0.8271,
-      "rewards/accuracies": 0.675000011920929,
-      "rewards/chosen": 0.1971231997013092,
-      "rewards/margins": 0.17458374798297882,
-      "rewards/rejected": 0.022539447993040085,
+      "logits/chosen": -2.083962917327881,
+      "logits/rejected": -2.06827449798584,
+      "logps/chosen": -33.732425689697266,
+      "logps/rejected": -33.07551193237305,
+      "loss": 0.5985,
+      "rewards/accuracies": 0.762499988079071,
+      "rewards/chosen": 0.2681804299354553,
+      "rewards/margins": 0.24266552925109863,
+      "rewards/rejected": 0.025514895096421242,
       "step": 340
     },
     {
       "epoch": 0.91,
       "learning_rate": 1.2518018074041684e-07,
-      "logits/chosen": -1.9518101215362549,
-      "logits/rejected": -1.9509942531585693,
-      "logps/chosen": -32.82368850708008,
-      "logps/rejected": -32.53012466430664,
-      "loss": 0.7648,
-      "rewards/accuracies": 0.6875,
-      "rewards/chosen": 0.22920279204845428,
-      "rewards/margins": 0.2616921663284302,
-      "rewards/rejected": -0.03248937055468559,
+      "logits/chosen": -1.956364631652832,
+      "logits/rejected": -1.9555118083953857,
+      "logps/chosen": -32.79610824584961,
+      "logps/rejected": -32.512969970703125,
+      "loss": 0.5758,
+      "rewards/accuracies": 0.7250000238418579,
+      "rewards/chosen": 0.2915918231010437,
+      "rewards/margins": 0.3202829957008362,
+      "rewards/rejected": -0.028691178187727928,
       "step": 350
     },
     {
       "epoch": 0.94,
       "learning_rate": 6.41315865106129e-08,
-      "logits/chosen": -1.9068868160247803,
-      "logits/rejected": -1.91714608669281,
-      "logps/chosen": -31.853801727294922,
-      "logps/rejected": -35.30488967895508,
-      "loss": 0.8248,
-      "rewards/accuracies": 0.637499988079071,
-      "rewards/chosen": 0.17544391751289368,
-      "rewards/margins": 0.18350856006145477,
-      "rewards/rejected": -0.008064627647399902,
+      "logits/chosen": -1.9118818044662476,
+      "logits/rejected": -1.9221813678741455,
+      "logps/chosen": -31.855304718017578,
+      "logps/rejected": -35.32331848144531,
+      "loss": 0.6036,
+      "rewards/accuracies": 0.7250000238418579,
+      "rewards/chosen": 0.2096320390701294,
+      "rewards/margins": 0.23036828637123108,
+      "rewards/rejected": -0.02073623239994049,
       "step": 360
     },
     {
       "epoch": 0.96,
       "learning_rate": 2.3150941078050325e-08,
-      "logits/chosen": -2.046671152114868,
-      "logits/rejected": -2.040235996246338,
-      "logps/chosen": -33.29608917236328,
-      "logps/rejected": -29.24057388305664,
-      "loss": 0.8215,
-      "rewards/accuracies": 0.699999988079071,
-      "rewards/chosen": 0.17794036865234375,
-      "rewards/margins": 0.1850142776966095,
-      "rewards/rejected": -0.007073921151459217,
+      "logits/chosen": -2.0520732402801514,
+      "logits/rejected": -2.0456154346466064,
+      "logps/chosen": -33.327049255371094,
+      "logps/rejected": -29.233760833740234,
+      "loss": 0.6124,
+      "rewards/accuracies": 0.7124999761581421,
+      "rewards/chosen": 0.19495204091072083,
+      "rewards/margins": 0.19935496151447296,
+      "rewards/rejected": -0.004402949940413237,
       "step": 370
     },
     {
       "epoch": 0.99,
       "learning_rate": 2.575864278703266e-09,
-      "logits/chosen": -1.9064124822616577,
-      "logits/rejected": -1.908599853515625,
-      "logps/chosen": -33.871421813964844,
-      "logps/rejected": -30.974201202392578,
-      "loss": 0.7824,
-      "rewards/accuracies": 0.7124999761581421,
-      "rewards/chosen": 0.18381480872631073,
-      "rewards/margins": 0.23460063338279724,
-      "rewards/rejected": -0.05078582838177681,
+      "logits/chosen": -1.911370038986206,
+      "logits/rejected": -1.9135433435440063,
+      "logps/chosen": -33.83781814575195,
+      "logps/rejected": -30.931133270263672,
+      "loss": 0.5854,
+      "rewards/accuracies": 0.762499988079071,
+      "rewards/chosen": 0.2407412976026535,
+      "rewards/margins": 0.27584755420684814,
+      "rewards/rejected": -0.035106249153614044,
       "step": 380
     },
     {
       "epoch": 1.0,
       "step": 385,
       "total_flos": 0.0,
-      "train_loss": 0.8806782920639237,
-      "train_runtime": 3251.4322,
+      "train_loss": 0.6364745784115482,
+      "train_runtime": 3251.6479,
       "train_samples_per_second": 0.947,
       "train_steps_per_second": 0.118
     }