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: 1.0817
-- Rewards/chosen: 0.0048
-- Rewards/rejected: -0.0009
-- Rewards/accuracies: 0.5104
-- Rewards/margins: 0.0057
-- Logps/rejected: -37.5184
-- Logps/chosen: -34.0250
-- Logits/rejected: -2.2281
-- Logits/chosen: -2.2329
+- Loss: 0.9828
+- Rewards/chosen: -0.0112
+- Rewards/rejected: -0.0284
+- Rewards/accuracies: 0.5395
+- Rewards/margins: 0.0172
+- Logps/rejected: -37.5734
+- Logps/chosen: -34.0569
+- Logits/rejected: -2.2180
+- Logits/chosen: -2.2228
 
 ## 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 |
 |:-------------:|:-----:|:----:|:---------------:|:--------------:|:----------------:|:------------------:|:---------------:|:--------------:|:------------:|:---------------:|:-------------:|
-| 0.8581        | 0.26  | 100  | 1.0730          | 0.0088         | 0.0069           | 0.5199             | 0.0019          | -37.5028       | -34.0170     | -2.2326         | -2.2374       |
-| 0.899         | 0.52  | 200  | 1.0716          | -0.0015        | -0.0088          | 0.5071             | 0.0072          | -37.5342       | -34.0376     | -2.2306         | -2.2355       |
-| 0.6105        | 0.78  | 300  | 1.0315          | 0.0102         | -0.0099          | 0.5602             | 0.0201          | -37.5365       | -34.0141     | -2.2289         | -2.2338       |
+| 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       |
 
 
 ### Framework versions

adapter_model.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:8417328de29b6b5905745ef824b91a690031b7aaa0073a0a0ba98affc3e9bbbf
+oid sha256:0e9c83cfb196d121fd47bbe89f821e8f2c260e9d270801c4b868c49d6d3ab09c
 size 176183216

all_results.json

@@ -1,20 +1,7 @@
 {
     "epoch": 1.0,
-    "eval_logits/chosen": -2.232933521270752,
-    "eval_logits/rejected": -2.2280824184417725,
-    "eval_logps/chosen": -34.02503967285156,
-    "eval_logps/rejected": -37.51841735839844,
-    "eval_loss": 1.0816813707351685,
-    "eval_rewards/accuracies": 0.5103820562362671,
-    "eval_rewards/chosen": 0.004756772890686989,
-    "eval_rewards/margins": 0.005652438849210739,
-    "eval_rewards/rejected": -0.0008956658421084285,
-    "eval_runtime": 145.5936,
-    "eval_samples": 343,
-    "eval_samples_per_second": 2.356,
-    "eval_steps_per_second": 0.295,
-    "train_loss": 0.8446982934877469,
-    "train_runtime": 3252.3399,
+    "train_loss": 0.8806782920639237,
+    "train_runtime": 3251.4322,
     "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": 0.8446982934877469,
-    "train_runtime": 3252.3399,
+    "train_loss": 0.8806782920639237,
+    "train_runtime": 3251.4322,
     "train_samples": 3079,
     "train_samples_per_second": 0.947,
     "train_steps_per_second": 0.118

trainer_state.json

@@ -25,589 +25,589 @@
     {
       "epoch": 0.03,
       "learning_rate": 1.282051282051282e-06,
-      "logits/chosen": -1.866492748260498,
-      "logits/rejected": -1.87080979347229,
-      "logps/chosen": -36.97657775878906,
-      "logps/rejected": -33.65824890136719,
-      "loss": 0.9236,
-      "rewards/accuracies": 0.5277777910232544,
-      "rewards/chosen": 0.015009618364274502,
-      "rewards/margins": 0.02909613959491253,
-      "rewards/rejected": -0.014086521230638027,
+      "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,
       "step": 10
     },
     {
       "epoch": 0.05,
       "learning_rate": 2.564102564102564e-06,
-      "logits/chosen": -1.9977840185165405,
-      "logits/rejected": -2.000425100326538,
-      "logps/chosen": -29.640512466430664,
-      "logps/rejected": -29.048751831054688,
-      "loss": 1.0528,
-      "rewards/accuracies": 0.4375,
-      "rewards/chosen": 0.0008407801506109536,
-      "rewards/margins": -0.0065057664178311825,
-      "rewards/rejected": 0.007346546743065119,
+      "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,
       "step": 20
     },
     {
       "epoch": 0.08,
       "learning_rate": 3.846153846153847e-06,
-      "logits/chosen": -1.9210313558578491,
-      "logits/rejected": -1.9183330535888672,
-      "logps/chosen": -31.377187728881836,
-      "logps/rejected": -33.214942932128906,
-      "loss": 0.976,
-      "rewards/accuracies": 0.550000011920929,
-      "rewards/chosen": 0.019451653584837914,
-      "rewards/margins": 0.01769269071519375,
-      "rewards/rejected": 0.001758962869644165,
+      "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,
       "step": 30
     },
     {
       "epoch": 0.1,
       "learning_rate": 4.999896948438434e-06,
-      "logits/chosen": -2.0173258781433105,
-      "logits/rejected": -2.008592128753662,
-      "logps/chosen": -32.55642318725586,
-      "logps/rejected": -32.49436569213867,
-      "loss": 1.0336,
-      "rewards/accuracies": 0.4749999940395355,
-      "rewards/chosen": 0.010191624984145164,
-      "rewards/margins": 0.001039800001308322,
-      "rewards/rejected": 0.009151825681328773,
+      "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,
       "step": 40
     },
     {
       "epoch": 0.13,
       "learning_rate": 4.987541037542187e-06,
-      "logits/chosen": -1.8626506328582764,
-      "logits/rejected": -1.8518873453140259,
-      "logps/chosen": -33.54867172241211,
-      "logps/rejected": -35.45621109008789,
-      "loss": 1.0318,
-      "rewards/accuracies": 0.5,
-      "rewards/chosen": 0.004401391837745905,
-      "rewards/margins": 0.006334079895168543,
-      "rewards/rejected": -0.001932688057422638,
+      "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,
       "step": 50
     },
     {
       "epoch": 0.16,
       "learning_rate": 4.954691471941119e-06,
-      "logits/chosen": -1.940718650817871,
-      "logits/rejected": -1.9426720142364502,
-      "logps/chosen": -32.53395462036133,
-      "logps/rejected": -33.20496368408203,
-      "loss": 0.9445,
-      "rewards/accuracies": 0.6499999761581421,
-      "rewards/chosen": 0.03311632573604584,
-      "rewards/margins": 0.04990752786397934,
-      "rewards/rejected": -0.016791202127933502,
+      "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,
       "step": 60
     },
     {
       "epoch": 0.18,
       "learning_rate": 4.901618883413549e-06,
-      "logits/chosen": -2.0724740028381348,
-      "logits/rejected": -2.077458381652832,
-      "logps/chosen": -33.9911994934082,
-      "logps/rejected": -36.61388397216797,
-      "loss": 1.1022,
-      "rewards/accuracies": 0.5249999761581421,
-      "rewards/chosen": -0.001358801149763167,
-      "rewards/margins": 0.021143654361367226,
-      "rewards/rejected": -0.022502455860376358,
+      "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,
       "step": 70
     },
     {
       "epoch": 0.21,
       "learning_rate": 4.828760511501322e-06,
-      "logits/chosen": -1.936197280883789,
-      "logits/rejected": -1.9393237829208374,
-      "logps/chosen": -34.28167724609375,
-      "logps/rejected": -34.63819885253906,
-      "loss": 0.8186,
-      "rewards/accuracies": 0.699999988079071,
-      "rewards/chosen": 0.07306724786758423,
-      "rewards/margins": 0.09596274793148041,
-      "rewards/rejected": -0.02289549633860588,
+      "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,
       "step": 80
     },
     {
       "epoch": 0.23,
       "learning_rate": 4.7367166013034295e-06,
-      "logits/chosen": -1.9451515674591064,
-      "logits/rejected": -1.949669599533081,
-      "logps/chosen": -32.39059066772461,
-      "logps/rejected": -32.346839904785156,
-      "loss": 1.0017,
-      "rewards/accuracies": 0.637499988079071,
-      "rewards/chosen": 0.03841521218419075,
-      "rewards/margins": 0.03238191828131676,
-      "rewards/rejected": 0.006033292505890131,
+      "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,
       "step": 90
     },
     {
       "epoch": 0.26,
       "learning_rate": 4.626245458345211e-06,
-      "logits/chosen": -2.042168617248535,
-      "logits/rejected": -2.040160655975342,
-      "logps/chosen": -32.12788772583008,
-      "logps/rejected": -31.280298233032227,
-      "loss": 0.8581,
-      "rewards/accuracies": 0.612500011920929,
-      "rewards/chosen": 0.05647433549165726,
-      "rewards/margins": 0.06351961940526962,
-      "rewards/rejected": -0.007045289967209101,
+      "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,
       "step": 100
     },
     {
       "epoch": 0.26,
-      "eval_logits/chosen": -2.237440824508667,
-      "eval_logits/rejected": -2.232595682144165,
-      "eval_logps/chosen": -34.017024993896484,
-      "eval_logps/rejected": -37.50282287597656,
-      "eval_loss": 1.0730067491531372,
-      "eval_rewards/accuracies": 0.5199335813522339,
-      "eval_rewards/chosen": 0.00876238290220499,
-      "eval_rewards/margins": 0.001864485559053719,
-      "eval_rewards/rejected": 0.0068978965282440186,
-      "eval_runtime": 146.0042,
-      "eval_samples_per_second": 2.349,
+      "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_samples_per_second": 2.353,
       "eval_steps_per_second": 0.295,
       "step": 100
     },
     {
       "epoch": 0.29,
       "learning_rate": 4.498257201263691e-06,
-      "logits/chosen": -1.9984264373779297,
-      "logits/rejected": -1.9960410594940186,
-      "logps/chosen": -33.10862350463867,
-      "logps/rejected": -34.00126266479492,
-      "loss": 1.1665,
-      "rewards/accuracies": 0.5874999761581421,
-      "rewards/chosen": 0.0675957053899765,
-      "rewards/margins": 0.05354113504290581,
-      "rewards/rejected": 0.014054578728973866,
+      "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,
       "step": 110
     },
     {
       "epoch": 0.31,
       "learning_rate": 4.353806263777678e-06,
-      "logits/chosen": -2.0099892616271973,
-      "logits/rejected": -2.001642942428589,
-      "logps/chosen": -32.312686920166016,
-      "logps/rejected": -32.10304641723633,
-      "loss": 0.9851,
-      "rewards/accuracies": 0.5375000238418579,
-      "rewards/chosen": 0.06610965728759766,
-      "rewards/margins": 0.03688037022948265,
-      "rewards/rejected": 0.029229288920760155,
+      "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,
       "step": 120
     },
     {
       "epoch": 0.34,
       "learning_rate": 4.1940827077152755e-06,
-      "logits/chosen": -2.0365984439849854,
-      "logits/rejected": -2.0286123752593994,
-      "logps/chosen": -30.3278865814209,
-      "logps/rejected": -32.04685592651367,
-      "loss": 0.9819,
-      "rewards/accuracies": 0.625,
-      "rewards/chosen": 0.07391555607318878,
-      "rewards/margins": 0.07084138691425323,
-      "rewards/rejected": 0.0030741647351533175,
+      "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,
       "step": 130
     },
     {
       "epoch": 0.36,
       "learning_rate": 4.0204024186666215e-06,
-      "logits/chosen": -1.9668314456939697,
-      "logits/rejected": -1.9770755767822266,
-      "logps/chosen": -31.2120304107666,
-      "logps/rejected": -32.57902526855469,
-      "loss": 0.8178,
-      "rewards/accuracies": 0.7250000238418579,
-      "rewards/chosen": 0.10381648689508438,
-      "rewards/margins": 0.12588787078857422,
-      "rewards/rejected": -0.02207140065729618,
+      "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,
       "step": 140
     },
     {
       "epoch": 0.39,
       "learning_rate": 3.834196265035119e-06,
-      "logits/chosen": -1.8799388408660889,
-      "logits/rejected": -1.881087064743042,
-      "logps/chosen": -33.97100830078125,
-      "logps/rejected": -34.84876251220703,
-      "loss": 0.8351,
-      "rewards/accuracies": 0.6625000238418579,
-      "rewards/chosen": 0.11084038019180298,
-      "rewards/margins": 0.15659382939338684,
-      "rewards/rejected": -0.04575346037745476,
+      "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,
       "step": 150
     },
     {
       "epoch": 0.42,
       "learning_rate": 3.636998309800573e-06,
-      "logits/chosen": -1.9326753616333008,
-      "logits/rejected": -1.9292488098144531,
-      "logps/chosen": -36.040306091308594,
-      "logps/rejected": -32.739051818847656,
-      "loss": 0.8539,
-      "rewards/accuracies": 0.612500011920929,
-      "rewards/chosen": 0.0656842365860939,
-      "rewards/margins": 0.07383431494235992,
-      "rewards/rejected": -0.008150083012878895,
+      "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,
       "step": 160
     },
     {
       "epoch": 0.44,
       "learning_rate": 3.4304331721118078e-06,
-      "logits/chosen": -2.0338892936706543,
-      "logits/rejected": -2.026510238647461,
-      "logps/chosen": -33.518821716308594,
-      "logps/rejected": -31.37355613708496,
-      "loss": 0.7141,
-      "rewards/accuracies": 0.7250000238418579,
-      "rewards/chosen": 0.13951851427555084,
-      "rewards/margins": 0.15037165582180023,
-      "rewards/rejected": -0.010853144340217113,
+      "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,
       "step": 170
     },
     {
       "epoch": 0.47,
       "learning_rate": 3.2162026428305436e-06,
-      "logits/chosen": -2.039998769760132,
-      "logits/rejected": -2.045238733291626,
-      "logps/chosen": -32.28400421142578,
-      "logps/rejected": -32.450523376464844,
-      "loss": 0.7652,
-      "rewards/accuracies": 0.675000011920929,
-      "rewards/chosen": 0.13039177656173706,
-      "rewards/margins": 0.11308407783508301,
-      "rewards/rejected": 0.017307698726654053,
+      "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,
       "step": 180
     },
     {
       "epoch": 0.49,
       "learning_rate": 2.996071664294641e-06,
-      "logits/chosen": -2.0408711433410645,
-      "logits/rejected": -2.0380892753601074,
-      "logps/chosen": -31.287479400634766,
-      "logps/rejected": -31.33124351501465,
-      "loss": 0.8251,
-      "rewards/accuracies": 0.699999988079071,
-      "rewards/chosen": 0.10073776543140411,
-      "rewards/margins": 0.10725078731775284,
-      "rewards/rejected": -0.0065130265429615974,
+      "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,
       "step": 190
     },
     {
       "epoch": 0.52,
       "learning_rate": 2.7718537898066833e-06,
-      "logits/chosen": -1.9110311269760132,
-      "logits/rejected": -1.915704369544983,
-      "logps/chosen": -31.336145401000977,
-      "logps/rejected": -32.791221618652344,
-      "loss": 0.899,
-      "rewards/accuracies": 0.6625000238418579,
-      "rewards/chosen": 0.13506832718849182,
-      "rewards/margins": 0.13521215319633484,
-      "rewards/rejected": -0.0001438349427189678,
+      "logits/chosen": -1.9006068706512451,
+      "logits/rejected": -1.9052391052246094,
+      "logps/chosen": -31.295886993408203,
+      "logps/rejected": -32.81926345825195,
+      "loss": 0.8332,
+      "rewards/accuracies": 0.737500011920929,
+      "rewards/chosen": 0.15519794821739197,
+      "rewards/margins": 0.16936282813549042,
+      "rewards/rejected": -0.01416487991809845,
       "step": 200
     },
     {
       "epoch": 0.52,
-      "eval_logits/chosen": -2.2354750633239746,
-      "eval_logits/rejected": -2.2306265830993652,
-      "eval_logps/chosen": -34.03763198852539,
-      "eval_logps/rejected": -37.534156799316406,
-      "eval_loss": 1.0715795755386353,
-      "eval_rewards/accuracies": 0.5070598125457764,
-      "eval_rewards/chosen": -0.0015398082323372364,
-      "eval_rewards/margins": 0.007228231523185968,
-      "eval_rewards/rejected": -0.008768039755523205,
-      "eval_runtime": 145.8306,
-      "eval_samples_per_second": 2.352,
+      "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_steps_per_second": 0.295,
       "step": 200
     },
     {
       "epoch": 0.55,
       "learning_rate": 2.5453962426402006e-06,
-      "logits/chosen": -2.0229578018188477,
-      "logits/rejected": -2.033618450164795,
-      "logps/chosen": -31.776050567626953,
-      "logps/rejected": -33.90400314331055,
-      "loss": 0.7465,
-      "rewards/accuracies": 0.7250000238418579,
-      "rewards/chosen": 0.09762789309024811,
-      "rewards/margins": 0.11827856302261353,
-      "rewards/rejected": -0.020650675520300865,
+      "logits/chosen": -2.013721466064453,
+      "logits/rejected": -2.0243561267852783,
+      "logps/chosen": -31.78135108947754,
+      "logps/rejected": -33.93622970581055,
+      "loss": 0.8758,
+      "rewards/accuracies": 0.675000011920929,
+      "rewards/chosen": 0.09497732669115067,
+      "rewards/margins": 0.1317446529865265,
+      "rewards/rejected": -0.036767326295375824,
       "step": 210
     },
     {
       "epoch": 0.57,
       "learning_rate": 2.3185646976551794e-06,
-      "logits/chosen": -1.9135267734527588,
-      "logits/rejected": -1.9282987117767334,
-      "logps/chosen": -29.896175384521484,
-      "logps/rejected": -31.5633487701416,
-      "loss": 0.75,
-      "rewards/accuracies": 0.675000011920929,
-      "rewards/chosen": 0.10768184810876846,
-      "rewards/margins": 0.11026783287525177,
-      "rewards/rejected": -0.002585983369499445,
+      "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,
       "step": 220
     },
     {
       "epoch": 0.6,
       "learning_rate": 2.0932279108998323e-06,
-      "logits/chosen": -1.9711973667144775,
-      "logits/rejected": -1.9751732349395752,
-      "logps/chosen": -33.15174102783203,
-      "logps/rejected": -31.605077743530273,
-      "loss": 0.7191,
-      "rewards/accuracies": 0.7250000238418579,
-      "rewards/chosen": 0.1328631192445755,
-      "rewards/margins": 0.1607515811920166,
-      "rewards/rejected": -0.027888456359505653,
+      "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,
       "step": 230
     },
     {
       "epoch": 0.62,
       "learning_rate": 1.8712423238279358e-06,
-      "logits/chosen": -1.969957709312439,
-      "logits/rejected": -1.9480478763580322,
-      "logps/chosen": -33.9122200012207,
-      "logps/rejected": -35.02121353149414,
-      "loss": 0.6955,
-      "rewards/accuracies": 0.75,
-      "rewards/chosen": 0.12076646089553833,
-      "rewards/margins": 0.15597540140151978,
-      "rewards/rejected": -0.03520893678069115,
+      "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,
       "step": 240
     },
     {
       "epoch": 0.65,
       "learning_rate": 1.6544367689701824e-06,
-      "logits/chosen": -2.0103070735931396,
-      "logits/rejected": -2.006990909576416,
-      "logps/chosen": -32.72673797607422,
-      "logps/rejected": -36.23841094970703,
-      "loss": 0.8178,
-      "rewards/accuracies": 0.7250000238418579,
-      "rewards/chosen": 0.09916610270738602,
-      "rewards/margins": 0.10590960830450058,
-      "rewards/rejected": -0.006743511650711298,
+      "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,
       "step": 250
     },
     {
       "epoch": 0.68,
       "learning_rate": 1.4445974030621963e-06,
-      "logits/chosen": -1.8776795864105225,
-      "logits/rejected": -1.875245451927185,
-      "logps/chosen": -34.003971099853516,
-      "logps/rejected": -35.510765075683594,
-      "loss": 0.889,
-      "rewards/accuracies": 0.6625000238418579,
-      "rewards/chosen": 0.09221391379833221,
-      "rewards/margins": 0.09528535604476929,
-      "rewards/rejected": -0.0030714483000338078,
+      "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,
       "step": 260
     },
     {
       "epoch": 0.7,
       "learning_rate": 1.243452991757889e-06,
-      "logits/chosen": -1.863521933555603,
-      "logits/rejected": -1.8610206842422485,
-      "logps/chosen": -34.20132827758789,
-      "logps/rejected": -31.76943016052246,
-      "loss": 0.8268,
-      "rewards/accuracies": 0.7124999761581421,
-      "rewards/chosen": 0.09813599288463593,
-      "rewards/margins": 0.09718601405620575,
-      "rewards/rejected": 0.000949984765611589,
+      "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,
       "step": 270
     },
     {
       "epoch": 0.73,
       "learning_rate": 1.0526606671603523e-06,
-      "logits/chosen": -1.9676616191864014,
-      "logits/rejected": -1.957082748413086,
-      "logps/chosen": -35.020606994628906,
-      "logps/rejected": -31.831247329711914,
-      "loss": 0.6669,
-      "rewards/accuracies": 0.762499988079071,
-      "rewards/chosen": 0.168013796210289,
-      "rewards/margins": 0.16314153373241425,
-      "rewards/rejected": 0.00487226527184248,
+      "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,
       "step": 280
     },
     {
       "epoch": 0.75,
       "learning_rate": 8.737922755071455e-07,
-      "logits/chosen": -2.0636610984802246,
-      "logits/rejected": -2.0486764907836914,
-      "logps/chosen": -30.683029174804688,
-      "logps/rejected": -32.61827850341797,
-      "loss": 0.894,
-      "rewards/accuracies": 0.6625000238418579,
-      "rewards/chosen": 0.12054909765720367,
-      "rewards/margins": 0.10377003997564316,
-      "rewards/rejected": 0.016779040917754173,
+      "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,
       "step": 290
     },
     {
       "epoch": 0.78,
       "learning_rate": 7.08321427484816e-07,
-      "logits/chosen": -1.9357779026031494,
-      "logits/rejected": -1.9332023859024048,
-      "logps/chosen": -32.54056930541992,
-      "logps/rejected": -30.850332260131836,
-      "loss": 0.6105,
-      "rewards/accuracies": 0.737500011920929,
-      "rewards/chosen": 0.1932690441608429,
-      "rewards/margins": 0.21003055572509766,
-      "rewards/rejected": -0.01676151715219021,
+      "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,
       "step": 300
     },
     {
       "epoch": 0.78,
-      "eval_logits/chosen": -2.233783006668091,
-      "eval_logits/rejected": -2.228926658630371,
-      "eval_logps/chosen": -34.01414108276367,
-      "eval_logps/rejected": -37.53648376464844,
-      "eval_loss": 1.0314662456512451,
-      "eval_rewards/accuracies": 0.560215950012207,
-      "eval_rewards/chosen": 0.010205330327153206,
-      "eval_rewards/margins": 0.020134516060352325,
-      "eval_rewards/rejected": -0.009929186664521694,
-      "eval_runtime": 145.6316,
-      "eval_samples_per_second": 2.355,
+      "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_steps_per_second": 0.295,
       "step": 300
     },
     {
       "epoch": 0.81,
       "learning_rate": 5.576113578589035e-07,
-      "logits/chosen": -1.9184128046035767,
-      "logits/rejected": -1.9151279926300049,
-      "logps/chosen": -31.325061798095703,
-      "logps/rejected": -33.77220916748047,
-      "loss": 0.7283,
-      "rewards/accuracies": 0.75,
-      "rewards/chosen": 0.13914386928081512,
-      "rewards/margins": 0.15417365729808807,
-      "rewards/rejected": -0.01502978801727295,
+      "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,
       "step": 310
     },
     {
       "epoch": 0.83,
       "learning_rate": 4.229036944380913e-07,
-      "logits/chosen": -1.968726396560669,
-      "logits/rejected": -1.9564485549926758,
-      "logps/chosen": -34.366207122802734,
-      "logps/rejected": -33.61689376831055,
-      "loss": 0.6804,
-      "rewards/accuracies": 0.699999988079071,
-      "rewards/chosen": 0.11493507772684097,
-      "rewards/margins": 0.1539594680070877,
-      "rewards/rejected": -0.03902440145611763,
+      "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,
       "step": 320
     },
     {
       "epoch": 0.86,
       "learning_rate": 3.053082288996112e-07,
-      "logits/chosen": -2.0041136741638184,
-      "logits/rejected": -2.002657651901245,
-      "logps/chosen": -33.25464630126953,
-      "logps/rejected": -32.49077606201172,
-      "loss": 0.7724,
-      "rewards/accuracies": 0.75,
-      "rewards/chosen": 0.11557211726903915,
-      "rewards/margins": 0.12116815894842148,
-      "rewards/rejected": -0.005596047732979059,
+      "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,
       "step": 330
     },
     {
       "epoch": 0.88,
       "learning_rate": 2.0579377374915805e-07,
-      "logits/chosen": -2.0917208194732666,
-      "logits/rejected": -2.0759525299072266,
-      "logps/chosen": -33.83209991455078,
-      "logps/rejected": -33.08992004394531,
-      "loss": 0.7606,
-      "rewards/accuracies": 0.75,
-      "rewards/chosen": 0.17364642024040222,
-      "rewards/margins": 0.15958845615386963,
-      "rewards/rejected": 0.014057991094887257,
+      "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,
       "step": 340
     },
     {
       "epoch": 0.91,
       "learning_rate": 1.2518018074041684e-07,
-      "logits/chosen": -1.9635206460952759,
-      "logits/rejected": -1.9625988006591797,
-      "logps/chosen": -32.91681671142578,
-      "logps/rejected": -32.46485137939453,
-      "loss": 0.7913,
-      "rewards/accuracies": 0.7124999761581421,
-      "rewards/chosen": 0.18264031410217285,
-      "rewards/margins": 0.182492196559906,
-      "rewards/rejected": 0.00014809667482040823,
+      "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,
       "step": 350
     },
     {
       "epoch": 0.94,
       "learning_rate": 6.41315865106129e-08,
-      "logits/chosen": -1.9189882278442383,
-      "logits/rejected": -1.9293220043182373,
-      "logps/chosen": -31.9406681060791,
-      "logps/rejected": -35.306640625,
-      "loss": 0.7574,
-      "rewards/accuracies": 0.699999988079071,
-      "rewards/chosen": 0.1320098638534546,
-      "rewards/margins": 0.1409510374069214,
-      "rewards/rejected": -0.008941170759499073,
+      "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,
       "step": 360
     },
     {
       "epoch": 0.96,
       "learning_rate": 2.3150941078050325e-08,
-      "logits/chosen": -2.0582680702209473,
-      "logits/rejected": -2.051753520965576,
-      "logps/chosen": -33.39839553833008,
-      "logps/rejected": -29.18343734741211,
-      "loss": 0.7663,
-      "rewards/accuracies": 0.675000011920929,
-      "rewards/chosen": 0.12678876519203186,
-      "rewards/margins": 0.1052960604429245,
-      "rewards/rejected": 0.02149270847439766,
+      "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,
       "step": 370
     },
     {
       "epoch": 0.99,
       "learning_rate": 2.575864278703266e-09,
-      "logits/chosen": -1.9175922870635986,
-      "logits/rejected": -1.9197555780410767,
-      "logps/chosen": -33.878089904785156,
-      "logps/rejected": -30.871530532836914,
-      "loss": 0.6999,
-      "rewards/accuracies": 0.737500011920929,
-      "rewards/chosen": 0.18047982454299927,
-      "rewards/margins": 0.17993128299713135,
-      "rewards/rejected": 0.0005485474830493331,
+      "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,
       "step": 380
     },
     {
       "epoch": 1.0,
       "step": 385,
       "total_flos": 0.0,
-      "train_loss": 0.8446982934877469,
-      "train_runtime": 3252.3399,
+      "train_loss": 0.8806782920639237,
+      "train_runtime": 3251.4322,
       "train_samples_per_second": 0.947,
       "train_steps_per_second": 0.118
     }