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: 2495.9221
-- Rewards/chosen: -0.0002
-- Rewards/rejected: -0.0006
-- Rewards/accuracies: 0.5299
+- Loss: 0.9996
+- Rewards/chosen: -0.0005
+- Rewards/rejected: -0.0009
+- Rewards/accuracies: 0.5594
 - Rewards/margins: 0.0004
-- Logps/rejected: -37.5814
-- Logps/chosen: -34.0562
-- Logits/rejected: -2.2213
-- Logits/chosen: -2.2261
+- Logps/rejected: -37.6063
+- Logps/chosen: -34.0836
+- Logits/rejected: -2.2166
+- Logits/chosen: -2.2214
 
 ## 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 |
 |:-------------:|:-----:|:----:|:---------------:|:--------------:|:----------------:|:------------------:|:---------------:|:--------------:|:------------:|:---------------:|:-------------:|
-| 2484.2775     | 0.26  | 100  | 2498.1597       | 0.0002         | -0.0000          | 0.5565             | 0.0002          | -37.5204       | -34.0187     | -2.2275         | -2.2323       |
-| 2464.7859     | 0.52  | 200  | 2496.0791       | 0.0000         | -0.0004          | 0.5544             | 0.0004          | -37.5574       | -34.0338     | -2.2242         | -2.2291       |
-| 2430.5529     | 0.78  | 300  | 2495.6680       | -0.0003        | -0.0007          | 0.5390             | 0.0005          | -37.5906       | -34.0629     | -2.2213         | -2.2262       |
+| 0.9987        | 0.26  | 100  | 0.9998          | -0.0002        | -0.0004          | 0.5336             | 0.0002          | -37.5528       | -34.0499     | -2.2264         | -2.2313       |
+| 0.9965        | 0.52  | 200  | 0.9996          | -0.0003        | -0.0006          | 0.5071             | 0.0004          | -37.5790       | -34.0618     | -2.2204         | -2.2252       |
+| 0.9925        | 0.78  | 300  | 0.9996          | -0.0005        | -0.0009          | 0.5594             | 0.0004          | -37.6063       | -34.0836     | -2.2166         | -2.2214       |
 
 
 ### Framework versions

adapter_config.json

@@ -19,13 +19,13 @@
   "rank_pattern": {},
   "revision": null,
   "target_modules": [
+    "down_proj",
     "k_proj",
+    "gate_proj",
     "v_proj",
-    "up_proj",
     "o_proj",
     "q_proj",
-    "down_proj",
-    "gate_proj"
+    "up_proj"
   ],
   "task_type": "CAUSAL_LM",
   "use_rslora": false

adapter_model.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:7f0c4ea6df6e7aaa31e727d2dd4d83424991aee1c595d8217c3ac1603db10c9a
+oid sha256:505158d62e646b45acae6dcb03907e61bf5ef1ca9bfc13b3f569925e0663f614
 size 176183216

all_results.json

@@ -1,21 +1,8 @@
 {
     "epoch": 1.0,
-    "eval_logits/chosen": -2.226093053817749,
-    "eval_logits/rejected": -2.221263885498047,
-    "eval_logps/chosen": -34.05622100830078,
-    "eval_logps/rejected": -37.58139419555664,
-    "eval_loss": 2495.922119140625,
-    "eval_rewards/accuracies": 0.529900312423706,
-    "eval_rewards/chosen": -0.00021668746194336563,
-    "eval_rewards/margins": 0.00043102249037474394,
-    "eval_rewards/rejected": -0.0006477099377661943,
-    "eval_runtime": 145.39,
-    "eval_samples": 343,
-    "eval_samples_per_second": 2.359,
-    "eval_steps_per_second": 0.296,
-    "train_loss": 2474.2954646915587,
-    "train_runtime": 3252.174,
+    "train_loss": 0.9973225085766284,
+    "train_runtime": 3253.1307,
     "train_samples": 3079,
-    "train_samples_per_second": 0.947,
+    "train_samples_per_second": 0.946,
     "train_steps_per_second": 0.118
 }

train_results.json

@@ -1,8 +1,8 @@
 {
     "epoch": 1.0,
-    "train_loss": 2474.2954646915587,
-    "train_runtime": 3252.174,
+    "train_loss": 0.9973225085766284,
+    "train_runtime": 3253.1307,
     "train_samples": 3079,
-    "train_samples_per_second": 0.947,
+    "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": 2500.0,
+      "loss": 1.0,
       "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": -1.866841197013855,
-      "logits/rejected": -1.871166467666626,
-      "logps/chosen": -36.98617172241211,
-      "logps/rejected": -33.65531539916992,
-      "loss": 2495.4616,
-      "rewards/accuracies": 0.5,
-      "rewards/chosen": 0.00020427265553735197,
-      "rewards/margins": 0.00045667175436392426,
-      "rewards/rejected": -0.0002523990988265723,
+      "logits/chosen": -1.8664803504943848,
+      "logits/rejected": -1.8707994222640991,
+      "logps/chosen": -36.978511810302734,
+      "logps/rejected": -33.66939163208008,
+      "loss": 0.9993,
+      "rewards/accuracies": 0.5694444179534912,
+      "rewards/chosen": 0.00028087408281862736,
+      "rewards/margins": 0.0006740752141922712,
+      "rewards/rejected": -0.00039320107316598296,
       "step": 10
     },
     {
       "epoch": 0.05,
       "learning_rate": 2.564102564102564e-06,
-      "logits/chosen": -1.997936487197876,
-      "logits/rejected": -2.0005903244018555,
-      "logps/chosen": -29.64678382873535,
-      "logps/rejected": -29.045034408569336,
-      "loss": 2502.3262,
-      "rewards/accuracies": 0.36250001192092896,
-      "rewards/chosen": -4.586054274113849e-05,
-      "rewards/margins": -0.00022994528990238905,
-      "rewards/rejected": 0.0001840847689891234,
+      "logits/chosen": -1.9984451532363892,
+      "logits/rejected": -2.0010995864868164,
+      "logps/chosen": -29.63176918029785,
+      "logps/rejected": -29.05954933166504,
+      "loss": 0.9999,
+      "rewards/accuracies": 0.48750001192092896,
+      "rewards/chosen": 0.00010425634536659345,
+      "rewards/margins": 6.528960511786863e-05,
+      "rewards/rejected": 3.8966707506915554e-05,
       "step": 20
     },
     {
       "epoch": 0.08,
       "learning_rate": 3.846153846153847e-06,
-      "logits/chosen": -1.9207321405410767,
-      "logits/rejected": -1.9180399179458618,
-      "logps/chosen": -31.407222747802734,
-      "logps/rejected": -33.223663330078125,
-      "loss": 2498.6508,
-      "rewards/accuracies": 0.5625,
-      "rewards/chosen": 8.869935118127614e-05,
-      "rewards/margins": 0.00014070476754568517,
-      "rewards/rejected": -5.200541272643022e-05,
+      "logits/chosen": -1.9210799932479858,
+      "logits/rejected": -1.9183847904205322,
+      "logps/chosen": -31.414783477783203,
+      "logps/rejected": -33.19659423828125,
+      "loss": 1.0002,
+      "rewards/accuracies": 0.4749999940395355,
+      "rewards/chosen": 1.3138540452928282e-05,
+      "rewards/margins": -0.00020548875909298658,
+      "rewards/rejected": 0.00021862727589905262,
       "step": 30
     },
     {
       "epoch": 0.1,
       "learning_rate": 4.999896948438434e-06,
-      "logits/chosen": -2.0177226066589355,
-      "logits/rejected": -2.0089757442474365,
-      "logps/chosen": -32.58082962036133,
-      "logps/rejected": -32.527244567871094,
-      "loss": 2498.9926,
-      "rewards/accuracies": 0.5,
-      "rewards/chosen": -4.022592111141421e-05,
-      "rewards/margins": 0.00010554380423855036,
-      "rewards/rejected": -0.00014576970716007054,
+      "logits/chosen": -2.0177221298217773,
+      "logits/rejected": -2.008965492248535,
+      "logps/chosen": -32.57322311401367,
+      "logps/rejected": -32.500308990478516,
+      "loss": 1.0001,
+      "rewards/accuracies": 0.4749999940395355,
+      "rewards/chosen": 3.584356090868823e-05,
+      "rewards/margins": -8.777440234553069e-05,
+      "rewards/rejected": 0.00012361796689219773,
       "step": 40
     },
     {
       "epoch": 0.13,
       "learning_rate": 4.987541037542187e-06,
-      "logits/chosen": -1.8629518747329712,
-      "logits/rejected": -1.8521617650985718,
-      "logps/chosen": -33.5596923828125,
-      "logps/rejected": -35.45528793334961,
-      "loss": 2499.9863,
-      "rewards/accuracies": 0.4749999940395355,
-      "rewards/chosen": -2.213427796959877e-05,
-      "rewards/margins": 7.289124368980993e-06,
-      "rewards/rejected": -2.9423434170894325e-05,
+      "logits/chosen": -1.8622690439224243,
+      "logits/rejected": -1.851509690284729,
+      "logps/chosen": -33.547603607177734,
+      "logps/rejected": -35.463592529296875,
+      "loss": 0.9998,
+      "rewards/accuracies": 0.5,
+      "rewards/chosen": 9.876764670480043e-05,
+      "rewards/margins": 0.0002112251240760088,
+      "rewards/rejected": -0.00011245747737120837,
       "step": 50
     },
     {
       "epoch": 0.16,
       "learning_rate": 4.954691471941119e-06,
-      "logits/chosen": -1.9416849613189697,
-      "logits/rejected": -1.9436241388320923,
-      "logps/chosen": -32.546897888183594,
-      "logps/rejected": -33.21548843383789,
-      "loss": 2490.4672,
-      "rewards/accuracies": 0.5375000238418579,
-      "rewards/chosen": 0.0005328331026248634,
-      "rewards/margins": 0.00097393908072263,
-      "rewards/rejected": -0.00044110597809776664,
+      "logits/chosen": -1.9400131702423096,
+      "logits/rejected": -1.9419806003570557,
+      "logps/chosen": -32.52842330932617,
+      "logps/rejected": -33.22877883911133,
+      "loss": 0.9987,
+      "rewards/accuracies": 0.6499999761581421,
+      "rewards/chosen": 0.0007175664068199694,
+      "rewards/margins": 0.0012915965635329485,
+      "rewards/rejected": -0.00057403021492064,
       "step": 60
     },
     {
       "epoch": 0.18,
       "learning_rate": 4.901618883413549e-06,
-      "logits/chosen": -2.072330951690674,
-      "logits/rejected": -2.0772910118103027,
-      "logps/chosen": -34.00098419189453,
-      "logps/rejected": -36.63383102416992,
-      "loss": 2494.9414,
-      "rewards/accuracies": 0.550000011920929,
-      "rewards/chosen": -0.00012498130672611296,
-      "rewards/margins": 0.0005246406653895974,
-      "rewards/rejected": -0.0006496219430118799,
+      "logits/chosen": -2.070552349090576,
+      "logits/rejected": -2.0755274295806885,
+      "logps/chosen": -34.00461959838867,
+      "logps/rejected": -36.64922332763672,
+      "loss": 0.9994,
+      "rewards/accuracies": 0.612500011920929,
+      "rewards/chosen": -0.00016131921438500285,
+      "rewards/margins": 0.0006422021542675793,
+      "rewards/rejected": -0.0008035213686525822,
       "step": 70
     },
     {
       "epoch": 0.21,
       "learning_rate": 4.828760511501322e-06,
-      "logits/chosen": -1.9325841665267944,
-      "logits/rejected": -1.9357010126113892,
-      "logps/chosen": -34.33161163330078,
-      "logps/rejected": -34.630489349365234,
-      "loss": 2486.8059,
-      "rewards/accuracies": 0.5249999761581421,
-      "rewards/chosen": 0.0009619827615097165,
-      "rewards/margins": 0.0013428140664473176,
-      "rewards/rejected": -0.0003808312467299402,
+      "logits/chosen": -1.9306777715682983,
+      "logits/rejected": -1.9338254928588867,
+      "logps/chosen": -34.32624816894531,
+      "logps/rejected": -34.661468505859375,
+      "loss": 0.9983,
+      "rewards/accuracies": 0.612500011920929,
+      "rewards/chosen": 0.0010156143689528108,
+      "rewards/margins": 0.0017062196275219321,
+      "rewards/rejected": -0.0006906053749844432,
       "step": 80
     },
     {
       "epoch": 0.23,
       "learning_rate": 4.7367166013034295e-06,
-      "logits/chosen": -1.9400427341461182,
-      "logits/rejected": -1.9445598125457764,
-      "logps/chosen": -32.36492156982422,
-      "logps/rejected": -32.34357452392578,
-      "loss": 2491.4584,
-      "rewards/accuracies": 0.6000000238418579,
-      "rewards/chosen": 0.001024983124807477,
-      "rewards/margins": 0.0008716614102013409,
-      "rewards/rejected": 0.00015332190378103405,
+      "logits/chosen": -1.9389193058013916,
+      "logits/rejected": -1.9434226751327515,
+      "logps/chosen": -32.38957214355469,
+      "logps/rejected": -32.348140716552734,
+      "loss": 0.9993,
+      "rewards/accuracies": 0.5874999761581421,
+      "rewards/chosen": 0.0007785108755342662,
+      "rewards/margins": 0.0006708315922878683,
+      "rewards/rejected": 0.00010767912317533046,
       "step": 90
     },
     {
       "epoch": 0.26,
       "learning_rate": 4.626245458345211e-06,
-      "logits/chosen": -2.037466526031494,
-      "logits/rejected": -2.0354855060577393,
-      "logps/chosen": -32.11969757080078,
-      "logps/rejected": -31.30398178100586,
-      "loss": 2484.2775,
-      "rewards/accuracies": 0.675000011920929,
-      "rewards/chosen": 0.001211356371641159,
-      "rewards/margins": 0.0015890670474618673,
-      "rewards/rejected": -0.00037771055940538645,
+      "logits/chosen": -2.0358777046203613,
+      "logits/rejected": -2.0339014530181885,
+      "logps/chosen": -32.13254165649414,
+      "logps/rejected": -31.29019546508789,
+      "loss": 0.9987,
+      "rewards/accuracies": 0.612500011920929,
+      "rewards/chosen": 0.0010829826351255178,
+      "rewards/margins": 0.001322855008766055,
+      "rewards/rejected": -0.0002398724900558591,
       "step": 100
     },
     {
       "epoch": 0.26,
-      "eval_logits/chosen": -2.232342481613159,
-      "eval_logits/rejected": -2.2275006771087646,
-      "eval_logps/chosen": -34.01866149902344,
-      "eval_logps/rejected": -37.52037811279297,
-      "eval_loss": 2498.15966796875,
-      "eval_rewards/accuracies": 0.5564784407615662,
-      "eval_rewards/chosen": 0.00015893821546342224,
-      "eval_rewards/margins": 0.0001965187693713233,
-      "eval_rewards/rejected": -3.7580521166091785e-05,
-      "eval_runtime": 146.0331,
+      "eval_logits/chosen": -2.2312774658203125,
+      "eval_logits/rejected": -2.226422071456909,
+      "eval_logps/chosen": -34.04991149902344,
+      "eval_logps/rejected": -37.55283737182617,
+      "eval_loss": 0.9997907280921936,
+      "eval_rewards/accuracies": 0.5336378812789917,
+      "eval_rewards/chosen": -0.0001535558985779062,
+      "eval_rewards/margins": 0.00020861340453848243,
+      "eval_rewards/rejected": -0.00036216925946064293,
+      "eval_runtime": 146.0254,
       "eval_samples_per_second": 2.349,
       "eval_steps_per_second": 0.294,
       "step": 100
@@ -181,311 +181,311 @@
     {
       "epoch": 0.29,
       "learning_rate": 4.498257201263691e-06,
-      "logits/chosen": -1.991633415222168,
-      "logits/rejected": -1.9892610311508179,
-      "logps/chosen": -33.10456085205078,
-      "logps/rejected": -34.01618194580078,
-      "loss": 2488.0367,
-      "rewards/accuracies": 0.6000000238418579,
-      "rewards/chosen": 0.0013925316743552685,
-      "rewards/margins": 0.0012606054078787565,
-      "rewards/rejected": 0.00013192615006119013,
+      "logits/chosen": -1.9907060861587524,
+      "logits/rejected": -1.9883339405059814,
+      "logps/chosen": -33.13169860839844,
+      "logps/rejected": -34.033958435058594,
+      "loss": 0.9988,
+      "rewards/accuracies": 0.6875,
+      "rewards/chosen": 0.0011211589444428682,
+      "rewards/margins": 0.0011670273961499333,
+      "rewards/rejected": -4.586850991472602e-05,
       "step": 110
     },
     {
       "epoch": 0.31,
       "learning_rate": 4.353806263777678e-06,
-      "logits/chosen": -2.003302812576294,
-      "logits/rejected": -1.994974136352539,
-      "logps/chosen": -32.31616973876953,
-      "logps/rejected": -32.14063262939453,
-      "loss": 2489.4971,
-      "rewards/accuracies": 0.5874999761581421,
-      "rewards/chosen": 0.001287340302951634,
-      "rewards/margins": 0.0010786365019157529,
-      "rewards/rejected": 0.00020870394655503333,
+      "logits/chosen": -2.002023458480835,
+      "logits/rejected": -1.993699312210083,
+      "logps/chosen": -32.341697692871094,
+      "logps/rejected": -32.16511917114258,
+      "loss": 0.9989,
+      "rewards/accuracies": 0.612500011920929,
+      "rewards/chosen": 0.0010320657165721059,
+      "rewards/margins": 0.001068194629624486,
+      "rewards/rejected": -3.612901855376549e-05,
       "step": 120
     },
     {
       "epoch": 0.34,
       "learning_rate": 4.1940827077152755e-06,
-      "logits/chosen": -2.0306484699249268,
-      "logits/rejected": -2.022704601287842,
-      "logps/chosen": -30.306324005126953,
-      "logps/rejected": -32.04903793334961,
-      "loss": 2483.8781,
-      "rewards/accuracies": 0.6000000238418579,
-      "rewards/chosen": 0.0016939423512667418,
-      "rewards/margins": 0.0016542377416044474,
-      "rewards/rejected": 3.9704824303044006e-05,
+      "logits/chosen": -2.028505802154541,
+      "logits/rejected": -2.020526885986328,
+      "logps/chosen": -30.3519287109375,
+      "logps/rejected": -32.101314544677734,
+      "loss": 0.9983,
+      "rewards/accuracies": 0.612500011920929,
+      "rewards/chosen": 0.0012378758983686566,
+      "rewards/margins": 0.001720982021652162,
+      "rewards/rejected": -0.0004831062688026577,
       "step": 130
     },
     {
       "epoch": 0.36,
       "learning_rate": 4.0204024186666215e-06,
-      "logits/chosen": -1.9617973566055298,
-      "logits/rejected": -1.9720312356948853,
-      "logps/chosen": -31.230310440063477,
-      "logps/rejected": -32.547096252441406,
-      "loss": 2480.1322,
-      "rewards/accuracies": 0.574999988079071,
-      "rewards/chosen": 0.0018935112748295069,
-      "rewards/margins": 0.0020156968384981155,
-      "rewards/rejected": -0.00012218570918776095,
+      "logits/chosen": -1.9588673114776611,
+      "logits/rejected": -1.9690834283828735,
+      "logps/chosen": -31.205490112304688,
+      "logps/rejected": -32.55961608886719,
+      "loss": 0.9976,
+      "rewards/accuracies": 0.6499999761581421,
+      "rewards/chosen": 0.0021417266689240932,
+      "rewards/margins": 0.0023890691809356213,
+      "rewards/rejected": -0.00024734257021918893,
       "step": 140
     },
     {
       "epoch": 0.39,
       "learning_rate": 3.834196265035119e-06,
-      "logits/chosen": -1.8725357055664062,
-      "logits/rejected": -1.8737138509750366,
-      "logps/chosen": -33.889976501464844,
-      "logps/rejected": -34.795631408691406,
-      "loss": 2466.4881,
-      "rewards/accuracies": 0.6499999761581421,
-      "rewards/chosen": 0.003027186496183276,
-      "rewards/margins": 0.0034109093248844147,
-      "rewards/rejected": -0.0003837232361547649,
+      "logits/chosen": -1.8695415258407593,
+      "logits/rejected": -1.8707062005996704,
+      "logps/chosen": -33.88127899169922,
+      "logps/rejected": -34.7686653137207,
+      "loss": 0.9968,
+      "rewards/accuracies": 0.574999988079071,
+      "rewards/chosen": 0.0031141345389187336,
+      "rewards/margins": 0.003228238318115473,
+      "rewards/rejected": -0.00011410393926780671,
       "step": 150
     },
     {
       "epoch": 0.42,
       "learning_rate": 3.636998309800573e-06,
-      "logits/chosen": -1.9241313934326172,
-      "logits/rejected": -1.9207313060760498,
-      "logps/chosen": -35.98552322387695,
-      "logps/rejected": -32.693538665771484,
-      "loss": 2484.5627,
-      "rewards/accuracies": 0.612500011920929,
-      "rewards/chosen": 0.0018615787848830223,
-      "rewards/margins": 0.0015694532776251435,
-      "rewards/rejected": 0.00029212533263489604,
+      "logits/chosen": -1.9212032556533813,
+      "logits/rejected": -1.9178003072738647,
+      "logps/chosen": -35.99773406982422,
+      "logps/rejected": -32.705848693847656,
+      "loss": 0.9984,
+      "rewards/accuracies": 0.6499999761581421,
+      "rewards/chosen": 0.0017394202295690775,
+      "rewards/margins": 0.0015704210381954908,
+      "rewards/rejected": 0.00016899927868507802,
       "step": 160
     },
     {
       "epoch": 0.44,
       "learning_rate": 3.4304331721118078e-06,
-      "logits/chosen": -2.023880958557129,
-      "logits/rejected": -2.0165772438049316,
-      "logps/chosen": -33.457122802734375,
-      "logps/rejected": -31.414859771728516,
-      "loss": 2460.2227,
-      "rewards/accuracies": 0.7124999761581421,
-      "rewards/chosen": 0.0034074243158102036,
-      "rewards/margins": 0.004037545528262854,
-      "rewards/rejected": -0.0006301216781139374,
+      "logits/chosen": -2.0206995010375977,
+      "logits/rejected": -2.0133931636810303,
+      "logps/chosen": -33.504085540771484,
+      "logps/rejected": -31.432220458984375,
+      "loss": 0.9963,
+      "rewards/accuracies": 0.637499988079071,
+      "rewards/chosen": 0.0029377774335443974,
+      "rewards/margins": 0.003741443855687976,
+      "rewards/rejected": -0.0008036663057282567,
       "step": 170
     },
     {
       "epoch": 0.47,
       "learning_rate": 3.2162026428305436e-06,
-      "logits/chosen": -2.030813455581665,
-      "logits/rejected": -2.0360608100891113,
-      "logps/chosen": -32.20356750488281,
-      "logps/rejected": -32.4092903137207,
-      "loss": 2473.8211,
-      "rewards/accuracies": 0.699999988079071,
-      "rewards/chosen": 0.0034122199285775423,
-      "rewards/margins": 0.0026536998338997364,
-      "rewards/rejected": 0.0007585205021314323,
+      "logits/chosen": -2.0269291400909424,
+      "logits/rejected": -2.032160997390747,
+      "logps/chosen": -32.24355697631836,
+      "logps/rejected": -32.431182861328125,
+      "loss": 0.9975,
+      "rewards/accuracies": 0.625,
+      "rewards/chosen": 0.003012270200997591,
+      "rewards/margins": 0.0024727012496441603,
+      "rewards/rejected": 0.0005395688931457698,
       "step": 180
     },
     {
       "epoch": 0.49,
       "learning_rate": 2.996071664294641e-06,
-      "logits/chosen": -2.0314080715179443,
-      "logits/rejected": -2.0286362171173096,
-      "logps/chosen": -31.27242088317871,
-      "logps/rejected": -31.320995330810547,
-      "loss": 2478.5072,
-      "rewards/accuracies": 0.6625000238418579,
-      "rewards/chosen": 0.0021653182338923216,
-      "rewards/margins": 0.0021931403316557407,
-      "rewards/rejected": -2.7821719413623214e-05,
+      "logits/chosen": -2.027367115020752,
+      "logits/rejected": -2.0246078968048096,
+      "logps/chosen": -31.290613174438477,
+      "logps/rejected": -31.361133575439453,
+      "loss": 0.9976,
+      "rewards/accuracies": 0.625,
+      "rewards/chosen": 0.0019833946134895086,
+      "rewards/margins": 0.002412599278613925,
+      "rewards/rejected": -0.00042920451960526407,
       "step": 190
     },
     {
       "epoch": 0.52,
       "learning_rate": 2.7718537898066833e-06,
-      "logits/chosen": -1.9025766849517822,
-      "logits/rejected": -1.9072151184082031,
-      "logps/chosen": -31.255901336669922,
-      "logps/rejected": -32.79901885986328,
-      "loss": 2464.7859,
-      "rewards/accuracies": 0.6625000238418579,
-      "rewards/chosen": 0.0035038013011217117,
-      "rewards/margins": 0.0035846265964210033,
-      "rewards/rejected": -8.082549902610481e-05,
+      "logits/chosen": -1.8985168933868408,
+      "logits/rejected": -1.903148889541626,
+      "logps/chosen": -31.30405616760254,
+      "logps/rejected": -32.838443756103516,
+      "loss": 0.9965,
+      "rewards/accuracies": 0.7250000238418579,
+      "rewards/chosen": 0.0030222723726183176,
+      "rewards/margins": 0.0034973658621311188,
+      "rewards/rejected": -0.0004750936641357839,
       "step": 200
     },
     {
       "epoch": 0.52,
-      "eval_logits/chosen": -2.229053497314453,
-      "eval_logits/rejected": -2.224233865737915,
-      "eval_logps/chosen": -34.033775329589844,
-      "eval_logps/rejected": -37.55736541748047,
-      "eval_loss": 2496.0791015625,
-      "eval_rewards/accuracies": 0.5544019937515259,
-      "eval_rewards/chosen": 7.761791493976489e-06,
-      "eval_rewards/margins": 0.00041520988452248275,
-      "eval_rewards/rejected": -0.00040744812577031553,
-      "eval_runtime": 145.4716,
-      "eval_samples_per_second": 2.358,
-      "eval_steps_per_second": 0.296,
+      "eval_logits/chosen": -2.225177764892578,
+      "eval_logits/rejected": -2.2203547954559326,
+      "eval_logps/chosen": -34.06184387207031,
+      "eval_logps/rejected": -37.579010009765625,
+      "eval_loss": 0.9996482133865356,
+      "eval_rewards/accuracies": 0.5070598125457764,
+      "eval_rewards/chosen": -0.0002729461120907217,
+      "eval_rewards/margins": 0.00035095339990220964,
+      "eval_rewards/rejected": -0.0006238995119929314,
+      "eval_runtime": 145.704,
+      "eval_samples_per_second": 2.354,
+      "eval_steps_per_second": 0.295,
       "step": 200
     },
     {
       "epoch": 0.55,
       "learning_rate": 2.5453962426402006e-06,
-      "logits/chosen": -2.0153257846832275,
-      "logits/rejected": -2.0259604454040527,
-      "logps/chosen": -31.77630615234375,
-      "logps/rejected": -33.9268798828125,
-      "loss": 2474.5629,
+      "logits/chosen": -2.011120557785034,
+      "logits/rejected": -2.021751880645752,
+      "logps/chosen": -31.745685577392578,
+      "logps/rejected": -33.96772003173828,
+      "loss": 0.9967,
       "rewards/accuracies": 0.637499988079071,
-      "rewards/chosen": 0.001949988305568695,
-      "rewards/margins": 0.0025917969178408384,
-      "rewards/rejected": -0.0006418084958568215,
+      "rewards/chosen": 0.0022561827208846807,
+      "rewards/margins": 0.003306365106254816,
+      "rewards/rejected": -0.0010501822689548135,
       "step": 210
     },
     {
       "epoch": 0.57,
       "learning_rate": 2.3185646976551794e-06,
-      "logits/chosen": -1.907790184020996,
-      "logits/rejected": -1.9225451946258545,
-      "logps/chosen": -29.77730941772461,
-      "logps/rejected": -31.612323760986328,
-      "loss": 2461.7975,
-      "rewards/accuracies": 0.75,
-      "rewards/chosen": 0.0033422994893044233,
-      "rewards/margins": 0.0038837480824440718,
-      "rewards/rejected": -0.0005414488259702921,
+      "logits/chosen": -1.903857946395874,
+      "logits/rejected": -1.918621301651001,
+      "logps/chosen": -29.797290802001953,
+      "logps/rejected": -31.628814697265625,
+      "loss": 0.9962,
+      "rewards/accuracies": 0.7124999761581421,
+      "rewards/chosen": 0.0031425058841705322,
+      "rewards/margins": 0.0038488968275487423,
+      "rewards/rejected": -0.0007063907687552273,
       "step": 220
     },
     {
       "epoch": 0.6,
       "learning_rate": 2.0932279108998323e-06,
-      "logits/chosen": -1.9650691747665405,
-      "logits/rejected": -1.9690206050872803,
-      "logps/chosen": -33.07447052001953,
-      "logps/rejected": -31.645030975341797,
-      "loss": 2457.1672,
-      "rewards/accuracies": 0.6875,
-      "rewards/chosen": 0.003429980482906103,
-      "rewards/margins": 0.004387288354337215,
-      "rewards/rejected": -0.0009573075803928077,
+      "logits/chosen": -1.9593979120254517,
+      "logits/rejected": -1.9633464813232422,
+      "logps/chosen": -33.067623138427734,
+      "logps/rejected": -31.64206886291504,
+      "loss": 0.9956,
+      "rewards/accuracies": 0.675000011920929,
+      "rewards/chosen": 0.003498472273349762,
+      "rewards/margins": 0.004426136147230864,
+      "rewards/rejected": -0.0009276636992581189,
       "step": 230
     },
     {
       "epoch": 0.62,
       "learning_rate": 1.8712423238279358e-06,
-      "logits/chosen": -1.9625787734985352,
-      "logits/rejected": -1.9408048391342163,
-      "logps/chosen": -33.812347412109375,
-      "logps/rejected": -35.121795654296875,
-      "loss": 2449.7848,
-      "rewards/accuracies": 0.675000011920929,
-      "rewards/chosen": 0.0034140206407755613,
-      "rewards/margins": 0.005124006420373917,
-      "rewards/rejected": -0.0017099861288443208,
+      "logits/chosen": -1.9572632312774658,
+      "logits/rejected": -1.9354908466339111,
+      "logps/chosen": -33.843727111816406,
+      "logps/rejected": -35.1453742980957,
+      "loss": 0.995,
+      "rewards/accuracies": 0.699999988079071,
+      "rewards/chosen": 0.0031002266332507133,
+      "rewards/margins": 0.0050460235215723515,
+      "rewards/rejected": -0.0019457967719063163,
       "step": 240
     },
     {
       "epoch": 0.65,
       "learning_rate": 1.6544367689701824e-06,
-      "logits/chosen": -2.003685712814331,
-      "logits/rejected": -2.000408172607422,
-      "logps/chosen": -32.71784210205078,
-      "logps/rejected": -36.25305938720703,
-      "loss": 2476.9369,
-      "rewards/accuracies": 0.612500011920929,
-      "rewards/chosen": 0.002072283299639821,
-      "rewards/margins": 0.0023536235094070435,
-      "rewards/rejected": -0.00028134050080552697,
+      "logits/chosen": -1.9997854232788086,
+      "logits/rejected": -1.9964803457260132,
+      "logps/chosen": -32.75019454956055,
+      "logps/rejected": -36.28661346435547,
+      "loss": 0.9976,
+      "rewards/accuracies": 0.637499988079071,
+      "rewards/chosen": 0.0017487213481217623,
+      "rewards/margins": 0.0023655896075069904,
+      "rewards/rejected": -0.0006168682011775672,
       "step": 250
     },
     {
       "epoch": 0.68,
       "learning_rate": 1.4445974030621963e-06,
-      "logits/chosen": -1.8708940744400024,
-      "logits/rejected": -1.8684980869293213,
-      "logps/chosen": -33.97399139404297,
-      "logps/rejected": -35.522247314453125,
-      "loss": 2477.3357,
-      "rewards/accuracies": 0.625,
-      "rewards/chosen": 0.0021441043354570866,
-      "rewards/margins": 0.002320351079106331,
-      "rewards/rejected": -0.00017624672909732908,
+      "logits/chosen": -1.8673791885375977,
+      "logits/rejected": -1.8649587631225586,
+      "logps/chosen": -34.018226623535156,
+      "logps/rejected": -35.539276123046875,
+      "loss": 0.998,
+      "rewards/accuracies": 0.612500011920929,
+      "rewards/chosen": 0.001701725646853447,
+      "rewards/margins": 0.002048287307843566,
+      "rewards/rejected": -0.0003465614281594753,
       "step": 260
     },
     {
       "epoch": 0.7,
       "learning_rate": 1.243452991757889e-06,
-      "logits/chosen": -1.8561140298843384,
-      "logits/rejected": -1.8537418842315674,
-      "logps/chosen": -34.15688705444336,
-      "logps/rejected": -31.835697174072266,
-      "loss": 2470.1545,
-      "rewards/accuracies": 0.637499988079071,
-      "rewards/chosen": 0.0024070844519883394,
-      "rewards/margins": 0.003050738014280796,
-      "rewards/rejected": -0.0006436532130464911,
+      "logits/chosen": -1.8522275686264038,
+      "logits/rejected": -1.849872350692749,
+      "logps/chosen": -34.16339874267578,
+      "logps/rejected": -31.845317840576172,
+      "loss": 0.9969,
+      "rewards/accuracies": 0.612500011920929,
+      "rewards/chosen": 0.0023420110810548067,
+      "rewards/margins": 0.0030819105450063944,
+      "rewards/rejected": -0.0007398994639515877,
       "step": 270
     },
     {
       "epoch": 0.73,
       "learning_rate": 1.0526606671603523e-06,
-      "logits/chosen": -1.959524154663086,
-      "logits/rejected": -1.9490426778793335,
-      "logps/chosen": -34.99895477294922,
-      "logps/rejected": -31.8908634185791,
-      "loss": 2459.8076,
-      "rewards/accuracies": 0.6875,
-      "rewards/chosen": 0.003576862858608365,
-      "rewards/margins": 0.004075545351952314,
-      "rewards/rejected": -0.0004986823769286275,
+      "logits/chosen": -1.9549518823623657,
+      "logits/rejected": -1.94447922706604,
+      "logps/chosen": -35.027687072753906,
+      "logps/rejected": -31.895471572875977,
+      "loss": 0.9962,
+      "rewards/accuracies": 0.699999988079071,
+      "rewards/chosen": 0.003289591521024704,
+      "rewards/margins": 0.003834384260699153,
+      "rewards/rejected": -0.0005447928560897708,
       "step": 280
     },
     {
       "epoch": 0.75,
       "learning_rate": 8.737922755071455e-07,
-      "logits/chosen": -2.0554285049438477,
-      "logits/rejected": -2.0405356884002686,
-      "logps/chosen": -30.697057723999023,
-      "logps/rejected": -32.610191345214844,
-      "loss": 2482.0641,
+      "logits/chosen": -2.0498766899108887,
+      "logits/rejected": -2.034980297088623,
+      "logps/chosen": -30.72440528869629,
+      "logps/rejected": -32.658695220947266,
+      "loss": 0.9979,
       "rewards/accuracies": 0.637499988079071,
-      "rewards/chosen": 0.0022706831805408,
-      "rewards/margins": 0.0018542330944910645,
-      "rewards/rejected": 0.0004164502606727183,
+      "rewards/chosen": 0.0019971781875938177,
+      "rewards/margins": 0.002065772656351328,
+      "rewards/rejected": -6.859400309622288e-05,
       "step": 290
     },
     {
       "epoch": 0.78,
       "learning_rate": 7.08321427484816e-07,
-      "logits/chosen": -1.9246467351913452,
-      "logits/rejected": -1.922141671180725,
-      "logps/chosen": -32.302886962890625,
-      "logps/rejected": -30.90523338317871,
-      "loss": 2430.5529,
-      "rewards/accuracies": 0.7875000238418579,
-      "rewards/chosen": 0.0062422603368759155,
-      "rewards/margins": 0.007126508746296167,
-      "rewards/rejected": -0.0008842485258355737,
+      "logits/chosen": -1.9201946258544922,
+      "logits/rejected": -1.9177051782608032,
+      "logps/chosen": -32.3183479309082,
+      "logps/rejected": -30.95510482788086,
+      "loss": 0.9925,
+      "rewards/accuracies": 0.7749999761581421,
+      "rewards/chosen": 0.006087628658860922,
+      "rewards/margins": 0.007470599375665188,
+      "rewards/rejected": -0.0013829706003889441,
       "step": 300
     },
     {
       "epoch": 0.78,
-      "eval_logits/chosen": -2.22615647315979,
-      "eval_logits/rejected": -2.2213311195373535,
-      "eval_logps/chosen": -34.06288146972656,
-      "eval_logps/rejected": -37.59059524536133,
-      "eval_loss": 2495.66796875,
-      "eval_rewards/accuracies": 0.5390365719795227,
-      "eval_rewards/chosen": -0.0002833307080436498,
-      "eval_rewards/margins": 0.00045641581527888775,
-      "eval_rewards/rejected": -0.0007397464942187071,
-      "eval_runtime": 145.8977,
+      "eval_logits/chosen": -2.2213804721832275,
+      "eval_logits/rejected": -2.216555118560791,
+      "eval_logps/chosen": -34.083614349365234,
+      "eval_logps/rejected": -37.60634994506836,
+      "eval_loss": 0.999591052532196,
+      "eval_rewards/accuracies": 0.559385359287262,
+      "eval_rewards/chosen": -0.0004906260874122381,
+      "eval_rewards/margins": 0.00040669209556654096,
+      "eval_rewards/rejected": -0.0008973181829787791,
+      "eval_runtime": 145.8707,
       "eval_samples_per_second": 2.351,
       "eval_steps_per_second": 0.295,
       "step": 300
@@ -493,122 +493,122 @@
     {
       "epoch": 0.81,
       "learning_rate": 5.576113578589035e-07,
-      "logits/chosen": -1.9106557369232178,
-      "logits/rejected": -1.9074223041534424,
-      "logps/chosen": -31.308679580688477,
-      "logps/rejected": -33.81306838989258,
-      "loss": 2464.2926,
-      "rewards/accuracies": 0.7124999761581421,
-      "rewards/chosen": 0.0029467204585671425,
-      "rewards/margins": 0.0036559212021529675,
-      "rewards/rejected": -0.0007092010346241295,
+      "logits/chosen": -1.9060642719268799,
+      "logits/rejected": -1.9028133153915405,
+      "logps/chosen": -31.319162368774414,
+      "logps/rejected": -33.85043716430664,
+      "loss": 0.9961,
+      "rewards/accuracies": 0.737500011920929,
+      "rewards/chosen": 0.002841859357431531,
+      "rewards/margins": 0.003924719989299774,
+      "rewards/rejected": -0.0010828599333763123,
       "step": 310
     },
     {
       "epoch": 0.83,
       "learning_rate": 4.229036944380913e-07,
-      "logits/chosen": -1.9598299264907837,
-      "logits/rejected": -1.9476732015609741,
-      "logps/chosen": -34.26856231689453,
-      "logps/rejected": -33.67915725708008,
-      "loss": 2454.0646,
+      "logits/chosen": -1.9553836584091187,
+      "logits/rejected": -1.9432109594345093,
+      "logps/chosen": -34.27588653564453,
+      "logps/rejected": -33.672359466552734,
+      "loss": 0.9955,
       "rewards/accuracies": 0.699999988079071,
-      "rewards/chosen": 0.0032751329708844423,
-      "rewards/margins": 0.004678304307162762,
-      "rewards/rejected": -0.0014031711034476757,
+      "rewards/chosen": 0.0032019000500440598,
+      "rewards/margins": 0.004537059459835291,
+      "rewards/rejected": -0.0013351596426218748,
       "step": 320
     },
     {
       "epoch": 0.86,
       "learning_rate": 3.053082288996112e-07,
-      "logits/chosen": -1.9953035116195679,
-      "logits/rejected": -1.9938862323760986,
-      "logps/chosen": -33.125816345214844,
-      "logps/rejected": -32.51322555541992,
-      "loss": 2461.5437,
-      "rewards/accuracies": 0.7124999761581421,
-      "rewards/chosen": 0.003599814372137189,
-      "rewards/margins": 0.003936268854886293,
-      "rewards/rejected": -0.00033645430812612176,
+      "logits/chosen": -1.9905990362167358,
+      "logits/rejected": -1.9891618490219116,
+      "logps/chosen": -33.116233825683594,
+      "logps/rejected": -32.55724334716797,
+      "loss": 0.9955,
+      "rewards/accuracies": 0.737500011920929,
+      "rewards/chosen": 0.0036955769173800945,
+      "rewards/margins": 0.004472161643207073,
+      "rewards/rejected": -0.0007765850750729442,
       "step": 330
     },
     {
       "epoch": 0.88,
       "learning_rate": 2.0579377374915805e-07,
-      "logits/chosen": -2.0824451446533203,
-      "logits/rejected": -2.066790819168091,
-      "logps/chosen": -33.772789001464844,
-      "logps/rejected": -33.076927185058594,
-      "loss": 2464.0691,
-      "rewards/accuracies": 0.6625000238418579,
-      "rewards/chosen": 0.004065990447998047,
-      "rewards/margins": 0.0036548797506839037,
-      "rewards/rejected": 0.00041111078462563455,
+      "logits/chosen": -2.0769362449645996,
+      "logits/rejected": -2.0613036155700684,
+      "logps/chosen": -33.791297912597656,
+      "logps/rejected": -33.12422180175781,
+      "loss": 0.9961,
+      "rewards/accuracies": 0.699999988079071,
+      "rewards/chosen": 0.003880967851728201,
+      "rewards/margins": 0.00394281093031168,
+      "rewards/rejected": -6.184288213262334e-05,
       "step": 340
     },
     {
       "epoch": 0.91,
       "learning_rate": 1.2518018074041684e-07,
-      "logits/chosen": -1.9551340341567993,
-      "logits/rejected": -1.9542996883392334,
-      "logps/chosen": -32.82428741455078,
-      "logps/rejected": -32.49939727783203,
-      "loss": 2452.0605,
-      "rewards/accuracies": 0.675000011920929,
-      "rewards/chosen": 0.004578094929456711,
-      "rewards/margins": 0.004920572973787785,
-      "rewards/rejected": -0.000342478568200022,
+      "logits/chosen": -1.950060248374939,
+      "logits/rejected": -1.9492241144180298,
+      "logps/chosen": -32.82404327392578,
+      "logps/rejected": -32.50709915161133,
+      "loss": 0.995,
+      "rewards/accuracies": 0.7124999761581421,
+      "rewards/chosen": 0.004580510314553976,
+      "rewards/margins": 0.005000022705644369,
+      "rewards/rejected": -0.000419511750806123,
       "step": 350
     },
     {
       "epoch": 0.94,
       "learning_rate": 6.41315865106129e-08,
-      "logits/chosen": -1.9097168445587158,
-      "logits/rejected": -1.9199819564819336,
-      "logps/chosen": -31.839696884155273,
-      "logps/rejected": -35.30548858642578,
-      "loss": 2462.577,
+      "logits/chosen": -1.9050449132919312,
+      "logits/rejected": -1.915305733680725,
+      "logps/chosen": -31.87860679626465,
+      "logps/rejected": -35.34981155395508,
+      "loss": 0.9961,
       "rewards/accuracies": 0.6875,
-      "rewards/chosen": 0.003649922786280513,
-      "rewards/margins": 0.0038172483909875154,
-      "rewards/rejected": -0.00016732477524783462,
+      "rewards/chosen": 0.0032608420588076115,
+      "rewards/margins": 0.0038713677786290646,
+      "rewards/rejected": -0.0006105261854827404,
       "step": 360
     },
     {
       "epoch": 0.96,
       "learning_rate": 2.3150941078050325e-08,
-      "logits/chosen": -2.0509915351867676,
-      "logits/rejected": -2.0445377826690674,
-      "logps/chosen": -33.30391311645508,
-      "logps/rejected": -29.264400482177734,
-      "loss": 2461.9752,
+      "logits/chosen": -2.04546856880188,
+      "logits/rejected": -2.039043426513672,
+      "logps/chosen": -33.336219787597656,
+      "logps/rejected": -29.269311904907227,
+      "loss": 0.9964,
       "rewards/accuracies": 0.6625000238418579,
-      "rewards/chosen": 0.003480545710772276,
-      "rewards/margins": 0.0038603146094828844,
-      "rewards/rejected": -0.0003797690733335912,
+      "rewards/chosen": 0.0031574335880577564,
+      "rewards/margins": 0.003586276201531291,
+      "rewards/rejected": -0.0004288425261620432,
       "step": 370
     },
     {
       "epoch": 0.99,
       "learning_rate": 2.575864278703266e-09,
-      "logits/chosen": -1.9099807739257812,
-      "logits/rejected": -1.9121906757354736,
-      "logps/chosen": -33.859439849853516,
-      "logps/rejected": -30.962020874023438,
-      "loss": 2453.9484,
-      "rewards/accuracies": 0.675000011920929,
-      "rewards/chosen": 0.0037960794288665056,
-      "rewards/margins": 0.004689999856054783,
-      "rewards/rejected": -0.0008939205436035991,
+      "logits/chosen": -1.905160665512085,
+      "logits/rejected": -1.907360315322876,
+      "logps/chosen": -33.86741256713867,
+      "logps/rejected": -30.982807159423828,
+      "loss": 0.9952,
+      "rewards/accuracies": 0.7250000238418579,
+      "rewards/chosen": 0.0037163912784308195,
+      "rewards/margins": 0.004818186163902283,
+      "rewards/rejected": -0.0011017953511327505,
       "step": 380
     },
     {
       "epoch": 1.0,
       "step": 385,
       "total_flos": 0.0,
-      "train_loss": 2474.2954646915587,
-      "train_runtime": 3252.174,
-      "train_samples_per_second": 0.947,
+      "train_loss": 0.9973225085766284,
+      "train_runtime": 3253.1307,
+      "train_samples_per_second": 0.946,
       "train_steps_per_second": 0.118
     }
   ],

training_args.bin

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:1e84cda7a2012b9a986ed1d795f0f3267d79261f8b36bd8847caa82552ae37dc
+oid sha256:443ad34f1047430b0fb8bdd7d8c9c9eabc4d3c2dda8c21886ec6362e1a17c30c
 size 4984