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: 0.7359
-- Rewards/chosen: 0.0075
-- Rewards/rejected: 0.0073
-- Rewards/accuracies: 0.5311
-- Rewards/margins: 0.0003
-- Logps/rejected: -37.5045
-- Logps/chosen: -34.0220
-- Logits/rejected: -2.2382
-- Logits/chosen: -2.2431
+- Loss: 0.5652
+- Rewards/chosen: 0.0047
+- Rewards/rejected: 0.0129
+- Rewards/accuracies: 0.4809
+- Rewards/margins: -0.0083
+- Logps/rejected: -37.4981
+- Logps/chosen: -34.0279
+- Logits/rejected: -2.2391
+- Logits/chosen: -2.2439
 
 ## 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 |
 |:-------------:|:-----:|:----:|:---------------:|:--------------:|:----------------:|:------------------:|:---------------:|:--------------:|:------------:|:---------------:|:-------------:|
-| 0.7731        | 0.26  | 100  | 0.7157          | 0.0126         | 0.0053           | 0.5216             | 0.0073          | -37.5079       | -34.0136     | -2.2387         | -2.2436       |
-| 0.7072        | 0.52  | 200  | 0.7401          | 0.0040         | 0.0081           | 0.4963             | -0.0041         | -37.5031       | -34.0279     | -2.2387         | -2.2436       |
-| 0.6516        | 0.78  | 300  | 0.7443          | 0.0055         | 0.0094           | 0.4958             | -0.0039         | -37.5009       | -34.0254     | -2.2391         | -2.2439       |
+| 0.5589        | 0.26  | 100  | 0.5107          | 0.0178         | 0.0027           | 0.5573             | 0.0151          | -37.5127       | -34.0091     | -2.2393         | -2.2442       |
+| 0.522         | 0.52  | 200  | 0.5423          | 0.0040         | 0.0035           | 0.4689             | 0.0005          | -37.5116       | -34.0288     | -2.2385         | -2.2433       |
+| 0.4616        | 0.78  | 300  | 0.5652          | 0.0047         | 0.0129           | 0.4809             | -0.0083         | -37.4981       | -34.0279     | -2.2391         | -2.2439       |
 
 
 ### Framework versions

adapter_model.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:e41ff2836264cd148e707d27674ceb166bf309aad16e50617f9808693d4e8288
+oid sha256:5e6b2a60197aedd47c7b976865f1da979dedbbc97bd5ad68d342cc0acb779f92
 size 176183216

all_results.json

@@ -1,21 +1,8 @@
 {
     "epoch": 1.0,
-    "eval_logits/chosen": -2.24311900138855,
-    "eval_logits/rejected": -2.2382466793060303,
-    "eval_logps/chosen": -34.021976470947266,
-    "eval_logps/rejected": -37.50453567504883,
-    "eval_loss": 0.7359318137168884,
-    "eval_rewards/accuracies": 0.531146228313446,
-    "eval_rewards/chosen": 0.0075458260253071785,
-    "eval_rewards/margins": 0.0002943648141808808,
-    "eval_rewards/rejected": 0.007251461502164602,
-    "eval_runtime": 145.7677,
-    "eval_samples": 343,
-    "eval_samples_per_second": 2.353,
-    "eval_steps_per_second": 0.295,
-    "train_loss": 0.7160632399769572,
-    "train_runtime": 3256.7737,
+    "train_loss": 0.5317366457604743,
+    "train_runtime": 3254.062,
     "train_samples": 3079,
-    "train_samples_per_second": 0.945,
+    "train_samples_per_second": 0.946,
     "train_steps_per_second": 0.118
 }

train_results.json

@@ -1,8 +1,8 @@
 {
     "epoch": 1.0,
-    "train_loss": 0.7160632399769572,
-    "train_runtime": 3256.7737,
+    "train_loss": 0.5317366457604743,
+    "train_runtime": 3254.062,
     "train_samples": 3079,
-    "train_samples_per_second": 0.945,
+    "train_samples_per_second": 0.946,
     "train_steps_per_second": 0.118
 }

trainer_state.json

@@ -15,7 +15,7 @@
       "logits/rejected": -1.7377450466156006,
       "logps/chosen": -29.553977966308594,
       "logps/rejected": -42.813133239746094,
-      "loss": 0.6944,
+      "loss": 0.5102,
       "rewards/accuracies": 0.0,
       "rewards/chosen": 0.0,
       "rewards/margins": 0.0,
@@ -25,590 +25,590 @@
     {
       "epoch": 0.03,
       "learning_rate": 1.2820512820512818e-07,
-      "logits/chosen": -1.8663125038146973,
-      "logits/rejected": -1.8706269264221191,
-      "logps/chosen": -37.00217819213867,
-      "logps/rejected": -33.66637420654297,
-      "loss": 0.662,
-      "rewards/accuracies": 0.4861111044883728,
-      "rewards/chosen": 0.0026538786478340626,
-      "rewards/margins": 0.024431757628917694,
-      "rewards/rejected": -0.021777881309390068,
+      "logits/chosen": -1.8663597106933594,
+      "logits/rejected": -1.8706719875335693,
+      "logps/chosen": -36.990386962890625,
+      "logps/rejected": -33.658267974853516,
+      "loss": 0.4771,
+      "rewards/accuracies": 0.5416666865348816,
+      "rewards/chosen": 0.011347133666276932,
+      "rewards/margins": 0.031080076470971107,
+      "rewards/rejected": -0.019732946529984474,
       "step": 10
     },
     {
       "epoch": 0.05,
       "learning_rate": 2.5641025641025636e-07,
-      "logits/chosen": -1.997361183166504,
-      "logits/rejected": -2.0000081062316895,
-      "logps/chosen": -29.635391235351562,
-      "logps/rejected": -29.05219078063965,
-      "loss": 0.7317,
+      "logits/chosen": -1.9978214502334595,
+      "logits/rejected": -2.000457286834717,
+      "logps/chosen": -29.62484359741211,
+      "logps/rejected": -29.059850692749023,
+      "loss": 0.5261,
       "rewards/accuracies": 0.5249999761581421,
-      "rewards/chosen": 0.004081652499735355,
-      "rewards/margins": -0.002670821500942111,
-      "rewards/rejected": 0.006752474699169397,
+      "rewards/chosen": 0.012144992128014565,
+      "rewards/margins": 0.009631244465708733,
+      "rewards/rejected": 0.0025137457996606827,
       "step": 20
     },
     {
       "epoch": 0.08,
       "learning_rate": 3.8461538461538463e-07,
-      "logits/chosen": -1.9197700023651123,
-      "logits/rejected": -1.917066216468811,
-      "logps/chosen": -31.422348022460938,
-      "logps/rejected": -33.22368240356445,
-      "loss": 0.7319,
-      "rewards/accuracies": 0.512499988079071,
-      "rewards/chosen": -0.003753386437892914,
-      "rewards/margins": -0.0006187178078107536,
-      "rewards/rejected": -0.0031346683390438557,
+      "logits/chosen": -1.9197276830673218,
+      "logits/rejected": -1.9170366525650024,
+      "logps/chosen": -31.404308319091797,
+      "logps/rejected": -33.229034423828125,
+      "loss": 0.5076,
+      "rewards/accuracies": 0.574999988079071,
+      "rewards/chosen": 0.008251860737800598,
+      "rewards/margins": 0.015654325485229492,
+      "rewards/rejected": -0.007402463350445032,
       "step": 30
     },
     {
       "epoch": 0.1,
       "learning_rate": 4.999896948438433e-07,
-      "logits/chosen": -2.0170438289642334,
-      "logits/rejected": -2.0082850456237793,
-      "logps/chosen": -32.59685516357422,
-      "logps/rejected": -32.49561309814453,
-      "loss": 0.7894,
-      "rewards/accuracies": 0.4000000059604645,
-      "rewards/chosen": -0.012029164470732212,
-      "rewards/margins": -0.022262731567025185,
-      "rewards/rejected": 0.010233565233647823,
+      "logits/chosen": -2.0169289112091064,
+      "logits/rejected": -2.008176803588867,
+      "logps/chosen": -32.55079650878906,
+      "logps/rejected": -32.502708435058594,
+      "loss": 0.535,
+      "rewards/accuracies": 0.512499988079071,
+      "rewards/chosen": 0.018207356333732605,
+      "rewards/margins": 0.011235545389354229,
+      "rewards/rejected": 0.006971807684749365,
       "step": 40
     },
     {
       "epoch": 0.13,
       "learning_rate": 4.987541037542186e-07,
-      "logits/chosen": -1.8647209405899048,
-      "logits/rejected": -1.853936791419983,
-      "logps/chosen": -33.57290267944336,
-      "logps/rejected": -35.4411735534668,
-      "loss": 0.7695,
-      "rewards/accuracies": 0.38749998807907104,
-      "rewards/chosen": -0.009253564290702343,
-      "rewards/margins": -0.015956666320562363,
-      "rewards/rejected": 0.006703103426843882,
+      "logits/chosen": -1.86395263671875,
+      "logits/rejected": -1.8531732559204102,
+      "logps/chosen": -33.559814453125,
+      "logps/rejected": -35.43522262573242,
+      "loss": 0.5718,
+      "rewards/accuracies": 0.4625000059604645,
+      "rewards/chosen": -0.0016335565596818924,
+      "rewards/margins": -0.013616559095680714,
+      "rewards/rejected": 0.011983001604676247,
       "step": 50
     },
     {
       "epoch": 0.16,
       "learning_rate": 4.954691471941118e-07,
-      "logits/chosen": -1.9456017017364502,
-      "logits/rejected": -1.9475319385528564,
-      "logps/chosen": -32.578392028808594,
-      "logps/rejected": -33.17626190185547,
-      "loss": 0.6746,
-      "rewards/accuracies": 0.6000000238418579,
-      "rewards/chosen": 0.013076215982437134,
-      "rewards/margins": 0.0160076767206192,
-      "rewards/rejected": -0.00293146213516593,
+      "logits/chosen": -1.9451929330825806,
+      "logits/rejected": -1.9471458196640015,
+      "logps/chosen": -32.59247589111328,
+      "logps/rejected": -33.19312286376953,
+      "loss": 0.4982,
+      "rewards/accuracies": 0.574999988079071,
+      "rewards/chosen": 0.005393369123339653,
+      "rewards/margins": 0.020615221932530403,
+      "rewards/rejected": -0.015221851877868176,
       "step": 60
     },
     {
       "epoch": 0.18,
       "learning_rate": 4.901618883413548e-07,
-      "logits/chosen": -2.079268217086792,
-      "logits/rejected": -2.084235668182373,
-      "logps/chosen": -33.979801177978516,
-      "logps/rejected": -36.58202362060547,
-      "loss": 0.7098,
-      "rewards/accuracies": 0.5249999761581421,
-      "rewards/chosen": 0.005212540738284588,
-      "rewards/margins": 0.013104838319122791,
-      "rewards/rejected": -0.007892299443483353,
+      "logits/chosen": -2.0797886848449707,
+      "logits/rejected": -2.08477783203125,
+      "logps/chosen": -33.99976348876953,
+      "logps/rejected": -36.57415771484375,
+      "loss": 0.575,
+      "rewards/accuracies": 0.44999998807907104,
+      "rewards/chosen": -0.007893012836575508,
+      "rewards/margins": -0.004194633569568396,
+      "rewards/rejected": -0.0036983792670071125,
       "step": 70
     },
     {
       "epoch": 0.21,
       "learning_rate": 4.828760511501322e-07,
-      "logits/chosen": -1.9421241283416748,
-      "logits/rejected": -1.9452680349349976,
-      "logps/chosen": -34.39490509033203,
-      "logps/rejected": -34.577003479003906,
-      "loss": 0.7084,
-      "rewards/accuracies": 0.512499988079071,
-      "rewards/chosen": 0.019745096564292908,
-      "rewards/margins": 0.010502790100872517,
-      "rewards/rejected": 0.009242306463420391,
+      "logits/chosen": -1.9419677257537842,
+      "logits/rejected": -1.94512939453125,
+      "logps/chosen": -34.39982986450195,
+      "logps/rejected": -34.590721130371094,
+      "loss": 0.5441,
+      "rewards/accuracies": 0.5625,
+      "rewards/chosen": 0.019587691873311996,
+      "rewards/margins": 0.01840476132929325,
+      "rewards/rejected": 0.0011829293798655272,
       "step": 80
     },
     {
       "epoch": 0.23,
       "learning_rate": 4.736716601303429e-07,
-      "logits/chosen": -1.950796365737915,
-      "logits/rejected": -1.9553086757659912,
-      "logps/chosen": -32.45103073120117,
-      "logps/rejected": -32.358036041259766,
-      "loss": 0.7023,
-      "rewards/accuracies": 0.550000011920929,
-      "rewards/chosen": 0.009835213422775269,
-      "rewards/margins": 0.009310262277722359,
-      "rewards/rejected": 0.0005249513196758926,
+      "logits/chosen": -1.9507859945297241,
+      "logits/rejected": -1.9552862644195557,
+      "logps/chosen": -32.467891693115234,
+      "logps/rejected": -32.358272552490234,
+      "loss": 0.5391,
+      "rewards/accuracies": 0.5375000238418579,
+      "rewards/chosen": -0.00033024101867340505,
+      "rewards/margins": -0.0007759220898151398,
+      "rewards/rejected": 0.00044568348675966263,
       "step": 90
     },
     {
       "epoch": 0.26,
       "learning_rate": 4.62624545834521e-07,
-      "logits/chosen": -2.0491089820861816,
-      "logits/rejected": -2.0471129417419434,
-      "logps/chosen": -32.255008697509766,
-      "logps/rejected": -31.249691009521484,
-      "loss": 0.7731,
-      "rewards/accuracies": 0.4375,
-      "rewards/chosen": -0.008502892218530178,
-      "rewards/margins": -0.018415190279483795,
-      "rewards/rejected": 0.009912299923598766,
+      "logits/chosen": -2.049070358276367,
+      "logits/rejected": -2.047074794769287,
+      "logps/chosen": -32.25985336303711,
+      "logps/rejected": -31.274398803710938,
+      "loss": 0.5589,
+      "rewards/accuracies": 0.5,
+      "rewards/chosen": -0.013313899748027325,
+      "rewards/margins": -0.007583809085190296,
+      "rewards/rejected": -0.005730087868869305,
       "step": 100
     },
     {
       "epoch": 0.26,
-      "eval_logits/chosen": -2.2435789108276367,
-      "eval_logits/rejected": -2.2386937141418457,
-      "eval_logps/chosen": -34.01361083984375,
-      "eval_logps/rejected": -37.507869720458984,
-      "eval_loss": 0.7156652212142944,
-      "eval_rewards/accuracies": 0.5215947031974792,
-      "eval_rewards/chosen": 0.012563038617372513,
-      "eval_rewards/margins": 0.007312912028282881,
-      "eval_rewards/rejected": 0.005250126589089632,
-      "eval_runtime": 146.1753,
-      "eval_samples_per_second": 2.346,
+      "eval_logits/chosen": -2.244201421737671,
+      "eval_logits/rejected": -2.239315986633301,
+      "eval_logps/chosen": -34.009124755859375,
+      "eval_logps/rejected": -37.5127067565918,
+      "eval_loss": 0.5106843709945679,
+      "eval_rewards/accuracies": 0.5573089718818665,
+      "eval_rewards/chosen": 0.017796913161873817,
+      "eval_rewards/margins": 0.015057443641126156,
+      "eval_rewards/rejected": 0.0027394662611186504,
+      "eval_runtime": 146.0293,
+      "eval_samples_per_second": 2.349,
       "eval_steps_per_second": 0.294,
       "step": 100
     },
     {
       "epoch": 0.29,
       "learning_rate": 4.4982572012636904e-07,
-      "logits/chosen": -2.0054879188537598,
-      "logits/rejected": -2.0030770301818848,
-      "logps/chosen": -33.253334045410156,
-      "logps/rejected": -34.042930603027344,
-      "loss": 0.7285,
-      "rewards/accuracies": 0.5,
-      "rewards/chosen": -0.005710972007364035,
-      "rewards/margins": 0.0024225921370089054,
-      "rewards/rejected": -0.00813356600701809,
+      "logits/chosen": -2.005356550216675,
+      "logits/rejected": -2.0029516220092773,
+      "logps/chosen": -33.2292366027832,
+      "logps/rejected": -34.054996490478516,
+      "loss": 0.4957,
+      "rewards/accuracies": 0.5874999761581421,
+      "rewards/chosen": 0.01020804513245821,
+      "rewards/margins": 0.028141701593995094,
+      "rewards/rejected": -0.01793365553021431,
       "step": 110
     },
     {
       "epoch": 0.31,
       "learning_rate": 4.353806263777677e-07,
-      "logits/chosen": -2.0168392658233643,
-      "logits/rejected": -2.008460760116577,
-      "logps/chosen": -32.45271301269531,
-      "logps/rejected": -32.185306549072266,
-      "loss": 0.7003,
-      "rewards/accuracies": 0.512499988079071,
-      "rewards/chosen": -0.004686274565756321,
-      "rewards/margins": 0.009594673290848732,
-      "rewards/rejected": -0.014280945062637329,
+      "logits/chosen": -2.0166029930114746,
+      "logits/rejected": -2.0082242488861084,
+      "logps/chosen": -32.45597457885742,
+      "logps/rejected": -32.18632507324219,
+      "loss": 0.5282,
+      "rewards/accuracies": 0.550000011920929,
+      "rewards/chosen": -0.007746054325252771,
+      "rewards/margins": 0.009627602994441986,
+      "rewards/rejected": -0.01737365685403347,
       "step": 120
     },
     {
       "epoch": 0.34,
       "learning_rate": 4.194082707715275e-07,
-      "logits/chosen": -2.0460398197174072,
-      "logits/rejected": -2.0379834175109863,
-      "logps/chosen": -30.48455238342285,
-      "logps/rejected": -32.02381134033203,
-      "loss": 0.7899,
+      "logits/chosen": -2.046288251876831,
+      "logits/rejected": -2.038238525390625,
+      "logps/chosen": -30.492712020874023,
+      "logps/rejected": -32.042259216308594,
+      "loss": 0.5777,
       "rewards/accuracies": 0.44999998807907104,
-      "rewards/chosen": -0.005300796125084162,
-      "rewards/margins": -0.022818461060523987,
-      "rewards/rejected": 0.017517665401101112,
+      "rewards/chosen": -0.011898026801645756,
+      "rewards/margins": -0.01942119374871254,
+      "rewards/rejected": 0.007523166947066784,
       "step": 130
     },
     {
       "epoch": 0.36,
       "learning_rate": 4.020402418666621e-07,
-      "logits/chosen": -1.9768435955047607,
-      "logits/rejected": -1.9871145486831665,
-      "logps/chosen": -31.402883529663086,
-      "logps/rejected": -32.54676818847656,
-      "loss": 0.6858,
-      "rewards/accuracies": 0.5375000238418579,
-      "rewards/chosen": 0.010066190734505653,
-      "rewards/margins": 0.017196927219629288,
-      "rewards/rejected": -0.007130734622478485,
+      "logits/chosen": -1.976900339126587,
+      "logits/rejected": -1.9871864318847656,
+      "logps/chosen": -31.39472007751465,
+      "logps/rejected": -32.55016326904297,
+      "loss": 0.4864,
+      "rewards/accuracies": 0.5874999761581421,
+      "rewards/chosen": 0.01745815947651863,
+      "rewards/margins": 0.028155237436294556,
+      "rewards/rejected": -0.01069707702845335,
       "step": 140
     },
     {
       "epoch": 0.39,
       "learning_rate": 3.8341962650351185e-07,
-      "logits/chosen": -1.8907657861709595,
-      "logits/rejected": -1.891840934753418,
-      "logps/chosen": -34.175559997558594,
-      "logps/rejected": -34.763084411621094,
-      "loss": 0.702,
-      "rewards/accuracies": 0.5249999761581421,
-      "rewards/chosen": 0.010278819128870964,
-      "rewards/margins": 0.013775828294456005,
-      "rewards/rejected": -0.003497010562568903,
+      "logits/chosen": -1.8908016681671143,
+      "logits/rejected": -1.8918870687484741,
+      "logps/chosen": -34.199378967285156,
+      "logps/rejected": -34.751861572265625,
+      "loss": 0.5621,
+      "rewards/accuracies": 0.4625000059604645,
+      "rewards/chosen": -0.004678909666836262,
+      "rewards/margins": -0.008454290218651295,
+      "rewards/rejected": 0.0037753782235085964,
       "step": 150
     },
     {
       "epoch": 0.42,
       "learning_rate": 3.636998309800572e-07,
-      "logits/chosen": -1.94234299659729,
-      "logits/rejected": -1.9388611316680908,
-      "logps/chosen": -36.15486526489258,
-      "logps/rejected": -32.73242950439453,
-      "loss": 0.6916,
+      "logits/chosen": -1.9425878524780273,
+      "logits/rejected": -1.9391052722930908,
+      "logps/chosen": -36.136619567871094,
+      "logps/rejected": -32.712371826171875,
+      "loss": 0.5148,
       "rewards/accuracies": 0.5,
-      "rewards/chosen": 0.010085567831993103,
-      "rewards/margins": 0.015891926363110542,
-      "rewards/rejected": -0.005806358531117439,
+      "rewards/chosen": 0.02454109489917755,
+      "rewards/margins": 0.01727622002363205,
+      "rewards/rejected": 0.007264876272529364,
       "step": 160
     },
     {
       "epoch": 0.44,
       "learning_rate": 3.430433172111807e-07,
-      "logits/chosen": -2.0420398712158203,
-      "logits/rejected": -2.0346436500549316,
-      "logps/chosen": -33.77216720581055,
-      "logps/rejected": -31.360408782958984,
-      "loss": 0.6733,
-      "rewards/accuracies": 0.6000000238418579,
-      "rewards/chosen": 0.015415417030453682,
-      "rewards/margins": 0.020550500601530075,
-      "rewards/rejected": -0.005135083571076393,
+      "logits/chosen": -2.042212963104248,
+      "logits/rejected": -2.0348353385925293,
+      "logps/chosen": -33.78378677368164,
+      "logps/rejected": -31.365230560302734,
+      "loss": 0.5133,
+      "rewards/accuracies": 0.5249999761581421,
+      "rewards/chosen": 0.009854594245553017,
+      "rewards/margins": 0.01922372169792652,
+      "rewards/rejected": -0.009369125589728355,
       "step": 170
     },
     {
       "epoch": 0.47,
       "learning_rate": 3.216202642830543e-07,
-      "logits/chosen": -2.047333002090454,
-      "logits/rejected": -2.052605628967285,
-      "logps/chosen": -32.51974105834961,
-      "logps/rejected": -32.49883270263672,
-      "loss": 0.6574,
-      "rewards/accuracies": 0.574999988079071,
-      "rewards/chosen": 0.015028724446892738,
-      "rewards/margins": 0.023243196308612823,
-      "rewards/rejected": -0.008214469067752361,
+      "logits/chosen": -2.047489881515503,
+      "logits/rejected": -2.0527498722076416,
+      "logps/chosen": -32.54710388183594,
+      "logps/rejected": -32.50310134887695,
+      "loss": 0.5261,
+      "rewards/accuracies": 0.4625000059604645,
+      "rewards/chosen": -0.001622129580937326,
+      "rewards/margins": 0.010947163216769695,
+      "rewards/rejected": -0.01256929337978363,
       "step": 180
     },
     {
       "epoch": 0.49,
       "learning_rate": 2.9960716642946403e-07,
-      "logits/chosen": -2.0488243103027344,
-      "logits/rejected": -2.046041488647461,
-      "logps/chosen": -31.49191665649414,
-      "logps/rejected": -31.331462860107422,
-      "loss": 0.7128,
-      "rewards/accuracies": 0.512499988079071,
-      "rewards/chosen": -0.0017768737161532044,
-      "rewards/margins": 0.006172674708068371,
-      "rewards/rejected": -0.00794955063611269,
+      "logits/chosen": -2.0485682487487793,
+      "logits/rejected": -2.04580020904541,
+      "logps/chosen": -31.48910903930664,
+      "logps/rejected": -31.318958282470703,
+      "loss": 0.5482,
+      "rewards/accuracies": 0.550000011920929,
+      "rewards/chosen": -0.00010873023711610585,
+      "rewards/margins": 0.0004080161452293396,
+      "rewards/rejected": -0.000516746542416513,
       "step": 190
     },
     {
       "epoch": 0.52,
       "learning_rate": 2.771853789806683e-07,
-      "logits/chosen": -1.9186776876449585,
-      "logits/rejected": -1.9233558177947998,
-      "logps/chosen": -31.598682403564453,
-      "logps/rejected": -32.798866271972656,
-      "loss": 0.7072,
-      "rewards/accuracies": 0.512499988079071,
-      "rewards/chosen": 0.004562787711620331,
-      "rewards/margins": 0.009323189035058022,
-      "rewards/rejected": -0.004760399926453829,
+      "logits/chosen": -1.9184706211090088,
+      "logits/rejected": -1.9231430292129517,
+      "logps/chosen": -31.597286224365234,
+      "logps/rejected": -32.80516815185547,
+      "loss": 0.522,
+      "rewards/accuracies": 0.5,
+      "rewards/chosen": 0.006296842359006405,
+      "rewards/margins": 0.016262350603938103,
+      "rewards/rejected": -0.009965506382286549,
       "step": 200
     },
     {
       "epoch": 0.52,
-      "eval_logits/chosen": -2.2435684204101562,
-      "eval_logits/rejected": -2.2386868000030518,
-      "eval_logps/chosen": -34.02791213989258,
-      "eval_logps/rejected": -37.50313949584961,
-      "eval_loss": 0.7401300072669983,
-      "eval_rewards/accuracies": 0.49626249074935913,
-      "eval_rewards/chosen": 0.003984179813414812,
-      "eval_rewards/margins": -0.004102990496903658,
-      "eval_rewards/rejected": 0.00808717031031847,
-      "eval_runtime": 145.9773,
-      "eval_samples_per_second": 2.35,
+      "eval_logits/chosen": -2.2433319091796875,
+      "eval_logits/rejected": -2.238459825515747,
+      "eval_logps/chosen": -34.028812408447266,
+      "eval_logps/rejected": -37.51158905029297,
+      "eval_loss": 0.5422906875610352,
+      "eval_rewards/accuracies": 0.46885380148887634,
+      "eval_rewards/chosen": 0.00401716772466898,
+      "eval_rewards/margins": 0.0004931418807245791,
+      "eval_rewards/rejected": 0.0035240259021520615,
+      "eval_runtime": 145.7818,
+      "eval_samples_per_second": 2.353,
       "eval_steps_per_second": 0.295,
       "step": 200
     },
     {
       "epoch": 0.55,
       "learning_rate": 2.5453962426402e-07,
-      "logits/chosen": -2.0318801403045654,
-      "logits/rejected": -2.042569398880005,
-      "logps/chosen": -31.933334350585938,
-      "logps/rejected": -33.868324279785156,
-      "loss": 0.6562,
-      "rewards/accuracies": 0.512499988079071,
-      "rewards/chosen": 0.022781556472182274,
-      "rewards/margins": 0.026157278567552567,
-      "rewards/rejected": -0.003375719068571925,
+      "logits/chosen": -2.0319957733154297,
+      "logits/rejected": -2.0426838397979736,
+      "logps/chosen": -31.943634033203125,
+      "logps/rejected": -33.902008056640625,
+      "loss": 0.4594,
+      "rewards/accuracies": 0.612500011920929,
+      "rewards/chosen": 0.01936788484454155,
+      "rewards/margins": 0.046881090849637985,
+      "rewards/rejected": -0.027513209730386734,
       "step": 210
     },
     {
       "epoch": 0.57,
       "learning_rate": 2.318564697655179e-07,
-      "logits/chosen": -1.9248113632202148,
-      "logits/rejected": -1.9396663904190063,
-      "logps/chosen": -30.091327667236328,
-      "logps/rejected": -31.557119369506836,
-      "loss": 0.6992,
-      "rewards/accuracies": 0.550000011920929,
-      "rewards/chosen": 0.012127770110964775,
-      "rewards/margins": 0.011495334096252918,
-      "rewards/rejected": 0.000632436596788466,
+      "logits/chosen": -1.9252302646636963,
+      "logits/rejected": -1.9400790929794312,
+      "logps/chosen": -30.092998504638672,
+      "logps/rejected": -31.55881690979004,
+      "loss": 0.5269,
+      "rewards/accuracies": 0.5625,
+      "rewards/chosen": 0.012976284138858318,
+      "rewards/margins": 0.013424187898635864,
+      "rewards/rejected": -0.0004478988121263683,
       "step": 220
     },
     {
       "epoch": 0.6,
       "learning_rate": 2.093227910899832e-07,
-      "logits/chosen": -1.9829256534576416,
-      "logits/rejected": -1.9868942499160767,
-      "logps/chosen": -33.38996505737305,
-      "logps/rejected": -31.552440643310547,
-      "loss": 0.6919,
-      "rewards/accuracies": 0.5625,
-      "rewards/chosen": 0.016500946134328842,
-      "rewards/margins": 0.01838388293981552,
-      "rewards/rejected": -0.001882936805486679,
+      "logits/chosen": -1.983025312423706,
+      "logits/rejected": -1.9869884252548218,
+      "logps/chosen": -33.37274932861328,
+      "logps/rejected": -31.550548553466797,
+      "loss": 0.496,
+      "rewards/accuracies": 0.512499988079071,
+      "rewards/chosen": 0.03130333498120308,
+      "rewards/margins": 0.03217558190226555,
+      "rewards/rejected": -0.0008722454076632857,
       "step": 230
     },
     {
       "epoch": 0.62,
       "learning_rate": 1.8712423238279356e-07,
-      "logits/chosen": -1.9827773571014404,
-      "logits/rejected": -1.9607956409454346,
-      "logps/chosen": -34.12545394897461,
-      "logps/rejected": -34.96198272705078,
-      "loss": 0.6654,
-      "rewards/accuracies": 0.637499988079071,
-      "rewards/chosen": 0.01697678305208683,
-      "rewards/margins": 0.02368919551372528,
-      "rewards/rejected": -0.0067124143242836,
+      "logits/chosen": -1.983493447303772,
+      "logits/rejected": -1.9615182876586914,
+      "logps/chosen": -34.13421630859375,
+      "logps/rejected": -34.97159957885742,
+      "loss": 0.4915,
+      "rewards/accuracies": 0.512499988079071,
+      "rewards/chosen": 0.01367081981152296,
+      "rewards/margins": 0.02823723293840885,
+      "rewards/rejected": -0.014566412195563316,
       "step": 240
     },
     {
       "epoch": 0.65,
       "learning_rate": 1.654436768970182e-07,
-      "logits/chosen": -2.0244033336639404,
-      "logits/rejected": -2.021089553833008,
-      "logps/chosen": -32.914833068847656,
-      "logps/rejected": -36.214698791503906,
-      "loss": 0.7396,
-      "rewards/accuracies": 0.4749999940395355,
-      "rewards/chosen": 0.006139112636446953,
-      "rewards/margins": 6.052106527931755e-06,
-      "rewards/rejected": 0.006133060436695814,
+      "logits/chosen": -2.02471661567688,
+      "logits/rejected": -2.0214104652404785,
+      "logps/chosen": -32.94930648803711,
+      "logps/rejected": -36.21381759643555,
+      "loss": 0.6084,
+      "rewards/accuracies": 0.4000000059604645,
+      "rewards/chosen": -0.016968127340078354,
+      "rewards/margins": -0.02474220283329487,
+      "rewards/rejected": 0.007774075958877802,
       "step": 250
     },
     {
       "epoch": 0.68,
       "learning_rate": 1.444597403062196e-07,
-      "logits/chosen": -1.8913761377334595,
-      "logits/rejected": -1.8889095783233643,
-      "logps/chosen": -34.184478759765625,
-      "logps/rejected": -35.503787994384766,
-      "loss": 0.7346,
-      "rewards/accuracies": 0.5,
-      "rewards/chosen": 0.0023515145294368267,
-      "rewards/margins": 0.0018522702157497406,
-      "rewards/rejected": 0.0004992469912394881,
+      "logits/chosen": -1.8912674188613892,
+      "logits/rejected": -1.8888145685195923,
+      "logps/chosen": -34.208099365234375,
+      "logps/rejected": -35.50531005859375,
+      "loss": 0.5753,
+      "rewards/accuracies": 0.44999998807907104,
+      "rewards/chosen": -0.013791380450129509,
+      "rewards/margins": -0.013306483626365662,
+      "rewards/rejected": -0.00048489533946849406,
       "step": 260
     },
     {
       "epoch": 0.7,
       "learning_rate": 1.2434529917578887e-07,
-      "logits/chosen": -1.8754098415374756,
-      "logits/rejected": -1.8728771209716797,
-      "logps/chosen": -34.38546371459961,
-      "logps/rejected": -31.757741928100586,
-      "loss": 0.729,
+      "logits/chosen": -1.8766494989395142,
+      "logits/rejected": -1.8741207122802734,
+      "logps/chosen": -34.38150405883789,
+      "logps/rejected": -31.744686126708984,
+      "loss": 0.5631,
       "rewards/accuracies": 0.5249999761581421,
-      "rewards/chosen": 0.0072813271544873714,
-      "rewards/margins": -0.0008724328363314271,
-      "rewards/rejected": 0.008153757080435753,
+      "rewards/chosen": 0.011268051341176033,
+      "rewards/margins": -0.007382377982139587,
+      "rewards/rejected": 0.018650425598025322,
       "step": 270
     },
     {
       "epoch": 0.73,
       "learning_rate": 1.0526606671603521e-07,
-      "logits/chosen": -1.9793100357055664,
-      "logits/rejected": -1.9686877727508545,
-      "logps/chosen": -35.343257904052734,
-      "logps/rejected": -31.828807830810547,
-      "loss": 0.728,
-      "rewards/accuracies": 0.48750001192092896,
-      "rewards/chosen": 0.00802794098854065,
-      "rewards/margins": 0.0007186919683590531,
-      "rewards/rejected": 0.007309247739613056,
+      "logits/chosen": -1.9800630807876587,
+      "logits/rejected": -1.9694359302520752,
+      "logps/chosen": -35.32728958129883,
+      "logps/rejected": -31.843835830688477,
+      "loss": 0.4967,
+      "rewards/accuracies": 0.550000011920929,
+      "rewards/chosen": 0.02054680697619915,
+      "rewards/margins": 0.022537903860211372,
+      "rewards/rejected": -0.001991095719859004,
       "step": 280
     },
     {
       "epoch": 0.75,
       "learning_rate": 8.737922755071453e-08,
-      "logits/chosen": -2.0753531455993652,
-      "logits/rejected": -2.0603315830230713,
-      "logps/chosen": -30.929107666015625,
-      "logps/rejected": -32.63794708251953,
-      "loss": 0.7688,
+      "logits/chosen": -2.0754363536834717,
+      "logits/rejected": -2.060412645339966,
+      "logps/chosen": -30.902868270874023,
+      "logps/rejected": -32.63262939453125,
+      "loss": 0.5486,
       "rewards/accuracies": 0.4625000059604645,
-      "rewards/chosen": -0.0029905380215495825,
-      "rewards/margins": -0.0113234743475914,
-      "rewards/rejected": 0.008332937955856323,
+      "rewards/chosen": 0.014878431335091591,
+      "rewards/margins": 0.0014350058045238256,
+      "rewards/rejected": 0.013443423435091972,
       "step": 290
     },
     {
       "epoch": 0.78,
       "learning_rate": 7.08321427484816e-08,
-      "logits/chosen": -1.9466259479522705,
-      "logits/rejected": -1.9440994262695312,
-      "logps/chosen": -32.894100189208984,
-      "logps/rejected": -30.831714630126953,
-      "loss": 0.6516,
-      "rewards/accuracies": 0.5375000238418579,
-      "rewards/chosen": 0.019804339855909348,
-      "rewards/margins": 0.028747867792844772,
-      "rewards/rejected": -0.008943530730903149,
+      "logits/chosen": -1.9468863010406494,
+      "logits/rejected": -1.9443439245224,
+      "logps/chosen": -32.88282775878906,
+      "logps/rejected": -30.835247039794922,
+      "loss": 0.4616,
+      "rewards/accuracies": 0.637499988079071,
+      "rewards/chosen": 0.03099890425801277,
+      "rewards/margins": 0.04390609636902809,
+      "rewards/rejected": -0.012907189317047596,
       "step": 300
     },
     {
       "epoch": 0.78,
-      "eval_logits/chosen": -2.243934392929077,
-      "eval_logits/rejected": -2.2390530109405518,
-      "eval_logps/chosen": -34.02543640136719,
-      "eval_logps/rejected": -37.50093460083008,
-      "eval_loss": 0.7443073391914368,
-      "eval_rewards/accuracies": 0.4958471655845642,
-      "eval_rewards/chosen": 0.005467735230922699,
-      "eval_rewards/margins": -0.0039433627389371395,
-      "eval_rewards/rejected": 0.009411096572875977,
-      "eval_runtime": 145.938,
-      "eval_samples_per_second": 2.35,
+      "eval_logits/chosen": -2.243941307067871,
+      "eval_logits/rejected": -2.2390596866607666,
+      "eval_logps/chosen": -34.027870178222656,
+      "eval_logps/rejected": -37.49814224243164,
+      "eval_loss": 0.5651828050613403,
+      "eval_rewards/accuracies": 0.4808970093727112,
+      "eval_rewards/chosen": 0.004676156677305698,
+      "eval_rewards/margins": -0.008257162757217884,
+      "eval_rewards/rejected": 0.012933320365846157,
+      "eval_runtime": 145.8956,
+      "eval_samples_per_second": 2.351,
       "eval_steps_per_second": 0.295,
       "step": 300
     },
     {
       "epoch": 0.81,
       "learning_rate": 5.576113578589034e-08,
-      "logits/chosen": -1.9287230968475342,
-      "logits/rejected": -1.9254610538482666,
-      "logps/chosen": -31.579919815063477,
-      "logps/rejected": -33.720069885253906,
-      "loss": 0.7291,
-      "rewards/accuracies": 0.550000011920929,
-      "rewards/chosen": 0.014056472107768059,
-      "rewards/margins": 0.0008082970743998885,
-      "rewards/rejected": 0.013248175382614136,
+      "logits/chosen": -1.928396224975586,
+      "logits/rejected": -1.9251337051391602,
+      "logps/chosen": -31.597158432006836,
+      "logps/rejected": -33.751304626464844,
+      "loss": 0.5251,
+      "rewards/accuracies": 0.512499988079071,
+      "rewards/chosen": 0.004333490040153265,
+      "rewards/margins": 0.01073872484266758,
+      "rewards/rejected": -0.006405232939869165,
       "step": 310
     },
     {
       "epoch": 0.83,
       "learning_rate": 4.229036944380912e-08,
-      "logits/chosen": -1.9804229736328125,
-      "logits/rejected": -1.9681212902069092,
-      "logps/chosen": -34.572120666503906,
-      "logps/rejected": -33.557823181152344,
-      "loss": 0.6635,
-      "rewards/accuracies": 0.5625,
-      "rewards/chosen": 0.014373106881976128,
-      "rewards/margins": 0.025763820856809616,
-      "rewards/rejected": -0.01139071211218834,
+      "logits/chosen": -1.9806029796600342,
+      "logits/rejected": -1.9683090448379517,
+      "logps/chosen": -34.563812255859375,
+      "logps/rejected": -33.55340576171875,
+      "loss": 0.4803,
+      "rewards/accuracies": 0.5874999761581421,
+      "rewards/chosen": 0.022582050412893295,
+      "rewards/margins": 0.03277861326932907,
+      "rewards/rejected": -0.010196560993790627,
       "step": 320
     },
     {
       "epoch": 0.86,
       "learning_rate": 3.053082288996112e-08,
-      "logits/chosen": -2.0162224769592285,
-      "logits/rejected": -2.014760971069336,
-      "logps/chosen": -33.47685241699219,
-      "logps/rejected": -32.48881530761719,
-      "loss": 0.7084,
-      "rewards/accuracies": 0.5375000238418579,
-      "rewards/chosen": 0.005364089272916317,
-      "rewards/margins": 0.010902222245931625,
-      "rewards/rejected": -0.0055381315760314465,
+      "logits/chosen": -2.0163512229919434,
+      "logits/rejected": -2.0149030685424805,
+      "logps/chosen": -33.46595764160156,
+      "logps/rejected": -32.46282196044922,
+      "loss": 0.5443,
+      "rewards/accuracies": 0.48750001192092896,
+      "rewards/chosen": 0.013885289430618286,
+      "rewards/margins": 0.0021549216471612453,
+      "rewards/rejected": 0.011730367317795753,
       "step": 330
     },
     {
       "epoch": 0.88,
       "learning_rate": 2.05793773749158e-08,
-      "logits/chosen": -2.1029350757598877,
-      "logits/rejected": -2.0871243476867676,
-      "logps/chosen": -34.16062545776367,
-      "logps/rejected": -33.069419860839844,
-      "loss": 0.7783,
-      "rewards/accuracies": 0.48750001192092896,
-      "rewards/chosen": 0.011257719248533249,
-      "rewards/margins": -0.017912957817316055,
-      "rewards/rejected": 0.029170680791139603,
+      "logits/chosen": -2.1030476093292236,
+      "logits/rejected": -2.0872480869293213,
+      "logps/chosen": -34.173221588134766,
+      "logps/rejected": -33.08686447143555,
+      "loss": 0.5853,
+      "rewards/accuracies": 0.5,
+      "rewards/chosen": 0.004321468528360128,
+      "rewards/margins": -0.01749541237950325,
+      "rewards/rejected": 0.021816883236169815,
       "step": 340
     },
     {
       "epoch": 0.91,
       "learning_rate": 1.251801807404168e-08,
-      "logits/chosen": -1.9747432470321655,
-      "logits/rejected": -1.9738028049468994,
-      "logps/chosen": -33.259681701660156,
-      "logps/rejected": -32.44579315185547,
-      "loss": 0.7418,
-      "rewards/accuracies": 0.512499988079071,
-      "rewards/chosen": 0.013451650738716125,
-      "rewards/margins": 0.001838967902585864,
-      "rewards/rejected": 0.011612680740654469,
+      "logits/chosen": -1.9748103618621826,
+      "logits/rejected": -1.9738647937774658,
+      "logps/chosen": -33.254066467285156,
+      "logps/rejected": -32.4422721862793,
+      "loss": 0.5603,
+      "rewards/accuracies": 0.38749998807907104,
+      "rewards/chosen": 0.019621744751930237,
+      "rewards/margins": 0.0036095953546464443,
+      "rewards/rejected": 0.016012147068977356,
       "step": 350
     },
     {
       "epoch": 0.94,
       "learning_rate": 6.41315865106129e-09,
-      "logits/chosen": -1.9304958581924438,
-      "logits/rejected": -1.9408457279205322,
-      "logps/chosen": -32.20914840698242,
-      "logps/rejected": -35.29241180419922,
-      "loss": 0.739,
-      "rewards/accuracies": 0.5375000238418579,
-      "rewards/chosen": -0.002673571230843663,
-      "rewards/margins": -0.0004814237472601235,
-      "rewards/rejected": -0.0021921484731137753,
+      "logits/chosen": -1.9305530786514282,
+      "logits/rejected": -1.9409068822860718,
+      "logps/chosen": -32.2257080078125,
+      "logps/rejected": -35.29142379760742,
+      "loss": 0.5729,
+      "rewards/accuracies": 0.512499988079071,
+      "rewards/chosen": -0.0147119564935565,
+      "rewards/margins": -0.012842650525271893,
+      "rewards/rejected": -0.0018693048041313887,
       "step": 360
     },
     {
       "epoch": 0.96,
       "learning_rate": 2.3150941078050324e-09,
-      "logits/chosen": -2.0693862438201904,
-      "logits/rejected": -2.0628304481506348,
-      "logps/chosen": -33.65240478515625,
-      "logps/rejected": -29.189916610717773,
-      "loss": 0.7913,
-      "rewards/accuracies": 0.4375,
-      "rewards/chosen": -0.0002606172056403011,
-      "rewards/margins": -0.022166749462485313,
-      "rewards/rejected": 0.021906131878495216,
+      "logits/chosen": -2.069432497024536,
+      "logits/rejected": -2.0628814697265625,
+      "logps/chosen": -33.65316390991211,
+      "logps/rejected": -29.218481063842773,
+      "loss": 0.5602,
+      "rewards/accuracies": 0.48750001192092896,
+      "rewards/chosen": -0.000836326158605516,
+      "rewards/margins": -0.006397420074790716,
+      "rewards/rejected": 0.005561096128076315,
       "step": 370
     },
     {
       "epoch": 0.99,
       "learning_rate": 2.575864278703266e-10,
-      "logits/chosen": -1.9290335178375244,
-      "logits/rejected": -1.931208848953247,
-      "logps/chosen": -34.243492126464844,
-      "logps/rejected": -30.905447006225586,
-      "loss": 0.6779,
-      "rewards/accuracies": 0.550000011920929,
-      "rewards/chosen": -0.0026660130824893713,
-      "rewards/margins": 0.01702696830034256,
-      "rewards/rejected": -0.019692981615662575,
+      "logits/chosen": -1.9289767742156982,
+      "logits/rejected": -1.9311527013778687,
+      "logps/chosen": -34.243995666503906,
+      "logps/rejected": -30.892364501953125,
+      "loss": 0.5254,
+      "rewards/accuracies": 0.5249999761581421,
+      "rewards/chosen": -0.0034601963125169277,
+      "rewards/margins": 0.010354455560445786,
+      "rewards/rejected": -0.013814652338624,
       "step": 380
     },
     {
       "epoch": 1.0,
       "step": 385,
       "total_flos": 0.0,
-      "train_loss": 0.7160632399769572,
-      "train_runtime": 3256.7737,
-      "train_samples_per_second": 0.945,
+      "train_loss": 0.5317366457604743,
+      "train_runtime": 3254.062,
+      "train_samples_per_second": 0.946,
       "train_steps_per_second": 0.118
     }
   ],