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: 0.6981
-- Rewards/chosen: -0.0558
-- Rewards/rejected: -0.0768
-- Rewards/accuracies: 0.5453
-- Rewards/margins: 0.0210
-- Logps/rejected: -37.6020
-- Logps/chosen: -34.0965
-- Logits/rejected: -2.2248
-- Logits/chosen: -2.2297
+- Loss: 0.4904
+- Rewards/chosen: 0.2570
+- Rewards/rejected: 0.2171
+- Rewards/accuracies: 0.5220
+- Rewards/margins: 0.0399
+- Logps/rejected: -37.2754
+- Logps/chosen: -33.7490
+- Logits/rejected: -2.2259
+- Logits/chosen: -2.2308
 
 ## 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 |
 |:-------------:|:-----:|:----:|:---------------:|:--------------:|:----------------:|:------------------:|:---------------:|:--------------:|:------------:|:---------------:|:-------------:|
-| 0.6464        | 0.26  | 100  | 0.6903          | -0.0054        | -0.0300          | 0.5685             | 0.0246          | -37.5500       | -34.0405     | -2.2291         | -2.2340       |
-| 0.5931        | 0.52  | 200  | 0.6980          | -0.0346        | -0.0543          | 0.5158             | 0.0196          | -37.5769       | -34.0730     | -2.2267         | -2.2316       |
-| 0.5301        | 0.78  | 300  | 0.6973          | -0.0555        | -0.0750          | 0.5390             | 0.0195          | -37.6000       | -34.0962     | -2.2243         | -2.2292       |
+| 0.4639        | 0.26  | 100  | 0.4939          | 0.1483         | 0.1205           | 0.5627             | 0.0278          | -37.3828       | -33.8698     | -2.2305         | -2.2354       |
+| 0.4308        | 0.52  | 200  | 0.4894          | 0.2606         | 0.2153           | 0.5544             | 0.0453          | -37.2774       | -33.7450     | -2.2290         | -2.2338       |
+| 0.374         | 0.78  | 300  | 0.4904          | 0.2570         | 0.2171           | 0.5220             | 0.0399          | -37.2754       | -33.7490     | -2.2259         | -2.2308       |
 
 
 ### Framework versions

adapter_model.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:9addaa12487d43b81005b34259b6437cc0db548ca618137c666f8e8fd220c95a
+oid sha256:18e50d4be9a44117ac2d82a5aa7a6b4ea5cc21668165b759ff991307589f83b4
 size 176183216

all_results.json

@@ -1,21 +1,8 @@
 {
     "epoch": 1.0,
-    "eval_logits/chosen": -2.22967529296875,
-    "eval_logits/rejected": -2.2248213291168213,
-    "eval_logps/chosen": -34.096519470214844,
-    "eval_logps/rejected": -37.601966857910156,
-    "eval_loss": 0.6981170177459717,
-    "eval_rewards/accuracies": 0.545265793800354,
-    "eval_rewards/chosen": -0.0557732917368412,
-    "eval_rewards/margins": 0.021040165796875954,
-    "eval_rewards/rejected": -0.07681345194578171,
-    "eval_runtime": 145.6397,
-    "eval_samples": 343,
-    "eval_samples_per_second": 2.355,
-    "eval_steps_per_second": 0.295,
-    "train_loss": 0.6175476637753573,
-    "train_runtime": 3252.7839,
+    "train_loss": 0.447351616079157,
+    "train_runtime": 3253.4458,
     "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": 0.6175476637753573,
-    "train_runtime": 3252.7839,
+    "train_loss": 0.447351616079157,
+    "train_runtime": 3253.4458,
     "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": 0.6931,
+      "loss": 0.5,
       "rewards/accuracies": 0.0,
       "rewards/chosen": 0.0,
       "rewards/margins": 0.0,
@@ -25,590 +25,590 @@
     {
       "epoch": 0.03,
       "learning_rate": 1.282051282051282e-06,
-      "logits/chosen": -1.866413950920105,
-      "logits/rejected": -1.8707411289215088,
-      "logps/chosen": -36.98916244506836,
-      "logps/rejected": -33.67436981201172,
-      "loss": 0.6701,
-      "rewards/accuracies": 0.5416666865348816,
-      "rewards/chosen": 0.01569323241710663,
-      "rewards/margins": 0.05555717274546623,
-      "rewards/rejected": -0.039863936603069305,
+      "logits/chosen": -1.8665987253189087,
+      "logits/rejected": -1.8709272146224976,
+      "logps/chosen": -36.985595703125,
+      "logps/rejected": -33.68160629272461,
+      "loss": 0.4886,
+      "rewards/accuracies": 0.5694444179534912,
+      "rewards/chosen": 0.018904482945799828,
+      "rewards/margins": 0.06528304517269135,
+      "rewards/rejected": -0.04637856408953667,
       "step": 10
     },
     {
       "epoch": 0.05,
       "learning_rate": 2.564102564102564e-06,
-      "logits/chosen": -1.9979650974273682,
-      "logits/rejected": -2.0006086826324463,
-      "logps/chosen": -29.624820709228516,
-      "logps/rejected": -29.0762939453125,
-      "loss": 0.6837,
-      "rewards/accuracies": 0.5874999761581421,
-      "rewards/chosen": 0.01563635841012001,
-      "rewards/margins": 0.027204299345612526,
-      "rewards/rejected": -0.01156794372946024,
+      "logits/chosen": -1.997780203819275,
+      "logits/rejected": -2.000434398651123,
+      "logps/chosen": -29.643661499023438,
+      "logps/rejected": -29.043325424194336,
+      "loss": 0.5031,
+      "rewards/accuracies": 0.4375,
+      "rewards/chosen": -0.001316396868787706,
+      "rewards/margins": -0.019422104582190514,
+      "rewards/rejected": 0.018105709925293922,
       "step": 20
     },
     {
       "epoch": 0.08,
       "learning_rate": 3.846153846153847e-06,
-      "logits/chosen": -1.921021819114685,
-      "logits/rejected": -1.9183374643325806,
-      "logps/chosen": -31.40532875061035,
-      "logps/rejected": -33.23241424560547,
-      "loss": 0.6877,
+      "logits/chosen": -1.9207446575164795,
+      "logits/rejected": -1.918060064315796,
+      "logps/chosen": -31.41064453125,
+      "logps/rejected": -33.227088928222656,
+      "loss": 0.4976,
       "rewards/accuracies": 0.5249999761581421,
-      "rewards/chosen": 0.00968973059207201,
-      "rewards/margins": 0.022251319140195847,
-      "rewards/rejected": -0.012561586685478687,
+      "rewards/chosen": 0.004905471112579107,
+      "rewards/margins": 0.012669263407588005,
+      "rewards/rejected": -0.007763790898025036,
       "step": 30
     },
     {
       "epoch": 0.1,
       "learning_rate": 4.999896948438434e-06,
-      "logits/chosen": -2.0176353454589844,
-      "logits/rejected": -2.008906364440918,
-      "logps/chosen": -32.574256896972656,
-      "logps/rejected": -32.53368377685547,
-      "loss": 0.6874,
-      "rewards/accuracies": 0.512499988079071,
-      "rewards/chosen": 0.0022967704571783543,
-      "rewards/margins": 0.02120940014719963,
-      "rewards/rejected": -0.018912632018327713,
+      "logits/chosen": -2.017446517944336,
+      "logits/rejected": -2.0087125301361084,
+      "logps/chosen": -32.553016662597656,
+      "logps/rejected": -32.50551986694336,
+      "loss": 0.4977,
+      "rewards/accuracies": 0.5,
+      "rewards/chosen": 0.021415216848254204,
+      "rewards/margins": 0.014982220716774464,
+      "rewards/rejected": 0.006432999856770039,
       "step": 40
     },
     {
       "epoch": 0.13,
       "learning_rate": 4.987541037542187e-06,
-      "logits/chosen": -1.8619186878204346,
-      "logits/rejected": -1.85114324092865,
-      "logps/chosen": -33.55537414550781,
-      "logps/rejected": -35.45675277709961,
-      "loss": 0.6957,
-      "rewards/accuracies": 0.48750001192092896,
-      "rewards/chosen": 0.001892436295747757,
-      "rewards/margins": 0.005858602002263069,
-      "rewards/rejected": -0.003966164775192738,
+      "logits/chosen": -1.8627235889434814,
+      "logits/rejected": -1.851959228515625,
+      "logps/chosen": -33.5064697265625,
+      "logps/rejected": -35.43267059326172,
+      "loss": 0.4951,
+      "rewards/accuracies": 0.550000011920929,
+      "rewards/chosen": 0.0459100641310215,
+      "rewards/margins": 0.02820250764489174,
+      "rewards/rejected": 0.017707552760839462,
       "step": 50
     },
     {
       "epoch": 0.16,
       "learning_rate": 4.954691471941119e-06,
-      "logits/chosen": -1.9400945901870728,
-      "logits/rejected": -1.9420464038848877,
-      "logps/chosen": -32.56509780883789,
-      "logps/rejected": -33.2406120300293,
-      "loss": 0.6632,
+      "logits/chosen": -1.9425691366195679,
+      "logits/rejected": -1.94449782371521,
+      "logps/chosen": -32.46650695800781,
+      "logps/rejected": -33.15652847290039,
+      "loss": 0.4765,
       "rewards/accuracies": 0.6499999761581421,
-      "rewards/chosen": 0.031578924506902695,
-      "rewards/margins": 0.09388783574104309,
-      "rewards/rejected": -0.062308914959430695,
+      "rewards/chosen": 0.12030963599681854,
+      "rewards/margins": 0.10694190114736557,
+      "rewards/rejected": 0.013367725536227226,
       "step": 60
     },
     {
       "epoch": 0.18,
       "learning_rate": 4.901618883413549e-06,
-      "logits/chosen": -2.0712790489196777,
-      "logits/rejected": -2.0762436389923096,
-      "logps/chosen": -33.981910705566406,
-      "logps/rejected": -36.62363815307617,
-      "loss": 0.6833,
-      "rewards/accuracies": 0.5249999761581421,
-      "rewards/chosen": 0.005918038543313742,
-      "rewards/margins": 0.05520814657211304,
-      "rewards/rejected": -0.04929010197520256,
+      "logits/chosen": -2.073408842086792,
+      "logits/rejected": -2.078367233276367,
+      "logps/chosen": -33.917694091796875,
+      "logps/rejected": -36.547218322753906,
+      "loss": 0.4901,
+      "rewards/accuracies": 0.5375000238418579,
+      "rewards/chosen": 0.06371410191059113,
+      "rewards/margins": 0.04422418028116226,
+      "rewards/rejected": 0.019489921629428864,
       "step": 70
     },
     {
       "epoch": 0.21,
       "learning_rate": 4.828760511501322e-06,
-      "logits/chosen": -1.9327905178070068,
-      "logits/rejected": -1.935909628868103,
-      "logps/chosen": -34.32685470581055,
-      "logps/rejected": -34.65606689453125,
-      "loss": 0.639,
-      "rewards/accuracies": 0.6625000238418579,
-      "rewards/chosen": 0.09085920453071594,
-      "rewards/margins": 0.14815348386764526,
-      "rewards/rejected": -0.057294271886348724,
+      "logits/chosen": -1.9349607229232788,
+      "logits/rejected": -1.9380786418914795,
+      "logps/chosen": -34.223785400390625,
+      "logps/rejected": -34.53069305419922,
+      "loss": 0.4713,
+      "rewards/accuracies": 0.612500011920929,
+      "rewards/chosen": 0.1836203634738922,
+      "rewards/margins": 0.12807974219322205,
+      "rewards/rejected": 0.05554063245654106,
       "step": 80
     },
     {
       "epoch": 0.23,
       "learning_rate": 4.7367166013034295e-06,
-      "logits/chosen": -1.9414918422698975,
-      "logits/rejected": -1.946007490158081,
-      "logps/chosen": -32.406803131103516,
-      "logps/rejected": -32.36021041870117,
-      "loss": 0.6792,
-      "rewards/accuracies": 0.6000000238418579,
-      "rewards/chosen": 0.054556868970394135,
-      "rewards/margins": 0.05573350936174393,
-      "rewards/rejected": -0.0011766403913497925,
+      "logits/chosen": -1.9439691305160522,
+      "logits/rejected": -1.9484784603118896,
+      "logps/chosen": -32.27050018310547,
+      "logps/rejected": -32.26476287841797,
+      "loss": 0.4779,
+      "rewards/accuracies": 0.637499988079071,
+      "rewards/chosen": 0.1772255003452301,
+      "rewards/margins": 0.09250012785196304,
+      "rewards/rejected": 0.08472537249326706,
       "step": 90
     },
     {
       "epoch": 0.26,
       "learning_rate": 4.626245458345211e-06,
-      "logits/chosen": -2.039034128189087,
-      "logits/rejected": -2.0370402336120605,
-      "logps/chosen": -32.172786712646484,
-      "logps/rejected": -31.333194732666016,
-      "loss": 0.6464,
-      "rewards/accuracies": 0.625,
-      "rewards/chosen": 0.06124376505613327,
-      "rewards/margins": 0.12152798473834991,
-      "rewards/rejected": -0.06028420478105545,
+      "logits/chosen": -2.0411603450775146,
+      "logits/rejected": -2.039163112640381,
+      "logps/chosen": -31.98573875427246,
+      "logps/rejected": -31.193227767944336,
+      "loss": 0.4639,
+      "rewards/accuracies": 0.6625000238418579,
+      "rewards/chosen": 0.22958631813526154,
+      "rewards/margins": 0.16390272974967957,
+      "rewards/rejected": 0.06568360328674316,
       "step": 100
     },
     {
       "epoch": 0.26,
-      "eval_logits/chosen": -2.2339773178100586,
-      "eval_logits/rejected": -2.229137420654297,
-      "eval_logps/chosen": -34.04054641723633,
-      "eval_logps/rejected": -37.549957275390625,
-      "eval_loss": 0.6902773976325989,
-      "eval_rewards/accuracies": 0.5685215592384338,
-      "eval_rewards/chosen": -0.005393954925239086,
-      "eval_rewards/margins": 0.024608083069324493,
-      "eval_rewards/rejected": -0.030002037063241005,
-      "eval_runtime": 146.034,
-      "eval_samples_per_second": 2.349,
-      "eval_steps_per_second": 0.294,
+      "eval_logits/chosen": -2.235391855239868,
+      "eval_logits/rejected": -2.2305493354797363,
+      "eval_logps/chosen": -33.869815826416016,
+      "eval_logps/rejected": -37.382774353027344,
+      "eval_loss": 0.4939241409301758,
+      "eval_rewards/accuracies": 0.5627076625823975,
+      "eval_rewards/chosen": 0.1482628434896469,
+      "eval_rewards/margins": 0.02780282311141491,
+      "eval_rewards/rejected": 0.12046003341674805,
+      "eval_runtime": 145.9747,
+      "eval_samples_per_second": 2.35,
+      "eval_steps_per_second": 0.295,
       "step": 100
     },
     {
       "epoch": 0.29,
       "learning_rate": 4.498257201263691e-06,
-      "logits/chosen": -1.994192123413086,
-      "logits/rejected": -1.9918158054351807,
-      "logps/chosen": -33.142940521240234,
-      "logps/rejected": -34.01188278198242,
-      "loss": 0.6911,
-      "rewards/accuracies": 0.5375000238418579,
-      "rewards/chosen": 0.09078876674175262,
-      "rewards/margins": 0.07505009323358536,
-      "rewards/rejected": 0.015738680958747864,
+      "logits/chosen": -1.997287392616272,
+      "logits/rejected": -1.9949369430541992,
+      "logps/chosen": -32.96843719482422,
+      "logps/rejected": -33.866310119628906,
+      "loss": 0.4739,
+      "rewards/accuracies": 0.5249999761581421,
+      "rewards/chosen": 0.24784216284751892,
+      "rewards/margins": 0.10108550637960434,
+      "rewards/rejected": 0.14675670862197876,
       "step": 110
     },
     {
       "epoch": 0.31,
       "learning_rate": 4.353806263777678e-06,
-      "logits/chosen": -2.0053954124450684,
-      "logits/rejected": -1.997046709060669,
-      "logps/chosen": -32.33894348144531,
-      "logps/rejected": -32.1308708190918,
-      "loss": 0.6746,
-      "rewards/accuracies": 0.5625,
-      "rewards/chosen": 0.09536493569612503,
-      "rewards/margins": 0.06779730319976807,
-      "rewards/rejected": 0.027567636221647263,
+      "logits/chosen": -2.008091688156128,
+      "logits/rejected": -1.9997599124908447,
+      "logps/chosen": -32.20352554321289,
+      "logps/rejected": -31.995223999023438,
+      "loss": 0.485,
+      "rewards/accuracies": 0.5375000238418579,
+      "rewards/chosen": 0.2172430008649826,
+      "rewards/margins": 0.06759083271026611,
+      "rewards/rejected": 0.1496521681547165,
       "step": 120
     },
     {
       "epoch": 0.34,
       "learning_rate": 4.1940827077152755e-06,
-      "logits/chosen": -2.0336387157440186,
-      "logits/rejected": -2.025650978088379,
-      "logps/chosen": -30.345691680908203,
-      "logps/rejected": -32.078697204589844,
-      "loss": 0.6527,
-      "rewards/accuracies": 0.612500011920929,
-      "rewards/chosen": 0.11702337116003036,
-      "rewards/margins": 0.14014457166194916,
-      "rewards/rejected": -0.023121213540434837,
+      "logits/chosen": -2.035614490509033,
+      "logits/rejected": -2.027682304382324,
+      "logps/chosen": -30.1588077545166,
+      "logps/rejected": -31.886260986328125,
+      "loss": 0.4717,
+      "rewards/accuracies": 0.550000011920929,
+      "rewards/chosen": 0.2852162718772888,
+      "rewards/margins": 0.1351451873779297,
+      "rewards/rejected": 0.15007111430168152,
       "step": 130
     },
     {
       "epoch": 0.36,
       "learning_rate": 4.0204024186666215e-06,
-      "logits/chosen": -1.9642337560653687,
-      "logits/rejected": -1.9744552373886108,
-      "logps/chosen": -31.243911743164062,
-      "logps/rejected": -32.590267181396484,
-      "loss": 0.6171,
-      "rewards/accuracies": 0.6625000238418579,
-      "rewards/chosen": 0.1581769436597824,
-      "rewards/margins": 0.20802685618400574,
-      "rewards/rejected": -0.04984992742538452,
+      "logits/chosen": -1.965490698814392,
+      "logits/rejected": -1.9756921529769897,
+      "logps/chosen": -31.065088272094727,
+      "logps/rejected": -32.42934036254883,
+      "loss": 0.4482,
+      "rewards/accuracies": 0.6875,
+      "rewards/chosen": 0.3191176950931549,
+      "rewards/margins": 0.22413134574890137,
+      "rewards/rejected": 0.09498633444309235,
       "step": 140
     },
     {
       "epoch": 0.39,
       "learning_rate": 3.834196265035119e-06,
-      "logits/chosen": -1.876604437828064,
-      "logits/rejected": -1.8777605295181274,
-      "logps/chosen": -33.938690185546875,
-      "logps/rejected": -34.807891845703125,
-      "loss": 0.6043,
+      "logits/chosen": -1.8782259225845337,
+      "logits/rejected": -1.8793823719024658,
+      "logps/chosen": -33.68832778930664,
+      "logps/rejected": -34.58278274536133,
+      "loss": 0.4367,
       "rewards/accuracies": 0.612500011920929,
-      "rewards/chosen": 0.22860188782215118,
-      "rewards/margins": 0.2741745412349701,
-      "rewards/rejected": -0.0455726757645607,
+      "rewards/chosen": 0.45392999053001404,
+      "rewards/margins": 0.2969031035900116,
+      "rewards/rejected": 0.15702682733535767,
       "step": 150
     },
     {
       "epoch": 0.42,
       "learning_rate": 3.636998309800573e-06,
-      "logits/chosen": -1.9282041788101196,
-      "logits/rejected": -1.9247684478759766,
-      "logps/chosen": -36.02125930786133,
-      "logps/rejected": -32.71831130981445,
-      "loss": 0.6454,
-      "rewards/accuracies": 0.612500011920929,
-      "rewards/chosen": 0.13537634909152985,
-      "rewards/margins": 0.13137592375278473,
-      "rewards/rejected": 0.004000450484454632,
+      "logits/chosen": -1.9295704364776611,
+      "logits/rejected": -1.9261808395385742,
+      "logps/chosen": -35.74212646484375,
+      "logps/rejected": -32.51028060913086,
+      "loss": 0.4538,
+      "rewards/accuracies": 0.6499999761581421,
+      "rewards/chosen": 0.38659390807151794,
+      "rewards/margins": 0.1953679323196411,
+      "rewards/rejected": 0.1912260353565216,
       "step": 160
     },
     {
       "epoch": 0.44,
       "learning_rate": 3.4304331721118078e-06,
-      "logits/chosen": -2.029125928878784,
-      "logits/rejected": -2.0217747688293457,
-      "logps/chosen": -33.49839401245117,
-      "logps/rejected": -31.400177001953125,
-      "loss": 0.5828,
-      "rewards/accuracies": 0.637499988079071,
-      "rewards/chosen": 0.26951926946640015,
-      "rewards/margins": 0.3130132555961609,
-      "rewards/rejected": -0.04349397122859955,
+      "logits/chosen": -2.031176805496216,
+      "logits/rejected": -2.023855686187744,
+      "logps/chosen": -33.24225616455078,
+      "logps/rejected": -31.193195343017578,
+      "loss": 0.42,
+      "rewards/accuracies": 0.675000011920929,
+      "rewards/chosen": 0.5000473260879517,
+      "rewards/margins": 0.3572581112384796,
+      "rewards/rejected": 0.1427893042564392,
       "step": 170
     },
     {
       "epoch": 0.47,
       "learning_rate": 3.2162026428305436e-06,
-      "logits/chosen": -2.0355944633483887,
-      "logits/rejected": -2.040832042694092,
-      "logps/chosen": -32.235923767089844,
-      "logps/rejected": -32.460418701171875,
-      "loss": 0.5943,
-      "rewards/accuracies": 0.699999988079071,
-      "rewards/chosen": 0.2779761850833893,
-      "rewards/margins": 0.2557251751422882,
-      "rewards/rejected": 0.02225096896290779,
+      "logits/chosen": -2.038222074508667,
+      "logits/rejected": -2.0434067249298096,
+      "logps/chosen": -31.95560646057129,
+      "logps/rejected": -32.17836380004883,
+      "loss": 0.4404,
+      "rewards/accuracies": 0.6875,
+      "rewards/chosen": 0.5302629470825195,
+      "rewards/margins": 0.2541634440422058,
+      "rewards/rejected": 0.2760995924472809,
       "step": 180
     },
     {
       "epoch": 0.49,
       "learning_rate": 2.996071664294641e-06,
-      "logits/chosen": -2.0362112522125244,
-      "logits/rejected": -2.0334599018096924,
-      "logps/chosen": -31.269250869750977,
-      "logps/rejected": -31.325435638427734,
-      "loss": 0.6245,
-      "rewards/accuracies": 0.637499988079071,
-      "rewards/chosen": 0.19773444533348083,
-      "rewards/margins": 0.20423230528831482,
-      "rewards/rejected": -0.0064978525042533875,
+      "logits/chosen": -2.0387401580810547,
+      "logits/rejected": -2.036006450653076,
+      "logps/chosen": -31.0674991607666,
+      "logps/rejected": -31.083877563476562,
+      "loss": 0.4607,
+      "rewards/accuracies": 0.6499999761581421,
+      "rewards/chosen": 0.3793107569217682,
+      "rewards/margins": 0.16841106116771698,
+      "rewards/rejected": 0.21089968085289001,
       "step": 190
     },
     {
       "epoch": 0.52,
       "learning_rate": 2.7718537898066833e-06,
-      "logits/chosen": -1.9060389995574951,
-      "logits/rejected": -1.9106788635253906,
-      "logps/chosen": -31.306299209594727,
-      "logps/rejected": -32.81407165527344,
-      "loss": 0.5931,
-      "rewards/accuracies": 0.7124999761581421,
-      "rewards/chosen": 0.2699825167655945,
-      "rewards/margins": 0.2908058166503906,
-      "rewards/rejected": -0.02082330361008644,
+      "logits/chosen": -1.9085681438446045,
+      "logits/rejected": -1.9132543802261353,
+      "logps/chosen": -31.083459854125977,
+      "logps/rejected": -32.602638244628906,
+      "loss": 0.4308,
+      "rewards/accuracies": 0.6875,
+      "rewards/chosen": 0.4705420434474945,
+      "rewards/margins": 0.301074743270874,
+      "rewards/rejected": 0.16946731507778168,
       "step": 200
     },
     {
       "epoch": 0.52,
-      "eval_logits/chosen": -2.231553792953491,
-      "eval_logits/rejected": -2.2267112731933594,
-      "eval_logps/chosen": -34.07304763793945,
-      "eval_logps/rejected": -37.57693862915039,
-      "eval_loss": 0.6979728937149048,
-      "eval_rewards/accuracies": 0.5157807469367981,
-      "eval_rewards/chosen": -0.03464451804757118,
-      "eval_rewards/margins": 0.019641490653157234,
-      "eval_rewards/rejected": -0.054286014288663864,
-      "eval_runtime": 145.8095,
-      "eval_samples_per_second": 2.352,
+      "eval_logits/chosen": -2.2338287830352783,
+      "eval_logits/rejected": -2.229020357131958,
+      "eval_logps/chosen": -33.7449951171875,
+      "eval_logps/rejected": -37.27743911743164,
+      "eval_loss": 0.48942965269088745,
+      "eval_rewards/accuracies": 0.5544019937515259,
+      "eval_rewards/chosen": 0.26059985160827637,
+      "eval_rewards/margins": 0.0453372597694397,
+      "eval_rewards/rejected": 0.21526260673999786,
+      "eval_runtime": 145.8953,
+      "eval_samples_per_second": 2.351,
       "eval_steps_per_second": 0.295,
       "step": 200
     },
     {
       "epoch": 0.55,
       "learning_rate": 2.5453962426402006e-06,
-      "logits/chosen": -2.018519163131714,
-      "logits/rejected": -2.0291810035705566,
-      "logps/chosen": -31.742992401123047,
-      "logps/rejected": -33.946937561035156,
-      "loss": 0.5902,
-      "rewards/accuracies": 0.675000011920929,
-      "rewards/chosen": 0.2054794579744339,
-      "rewards/margins": 0.2812942862510681,
-      "rewards/rejected": -0.07581482082605362,
+      "logits/chosen": -2.0216596126556396,
+      "logits/rejected": -2.032275915145874,
+      "logps/chosen": -31.500268936157227,
+      "logps/rejected": -33.663352966308594,
+      "loss": 0.4458,
+      "rewards/accuracies": 0.612500011920929,
+      "rewards/chosen": 0.42393389344215393,
+      "rewards/margins": 0.2445230931043625,
+      "rewards/rejected": 0.17941072583198547,
       "step": 210
     },
     {
       "epoch": 0.57,
       "learning_rate": 2.3185646976551794e-06,
-      "logits/chosen": -1.911586046218872,
-      "logits/rejected": -1.9263393878936768,
-      "logps/chosen": -29.84616470336914,
-      "logps/rejected": -31.615009307861328,
-      "loss": 0.5879,
-      "rewards/accuracies": 0.762499988079071,
-      "rewards/chosen": 0.23883743584156036,
-      "rewards/margins": 0.2899848222732544,
-      "rewards/rejected": -0.051147449761629105,
+      "logits/chosen": -1.9136396646499634,
+      "logits/rejected": -1.928344964981079,
+      "logps/chosen": -29.588964462280273,
+      "logps/rejected": -31.396224975585938,
+      "loss": 0.4269,
+      "rewards/accuracies": 0.699999988079071,
+      "rewards/chosen": 0.4703185558319092,
+      "rewards/margins": 0.3245617151260376,
+      "rewards/rejected": 0.1457568258047104,
       "step": 220
     },
     {
       "epoch": 0.6,
       "learning_rate": 2.0932279108998323e-06,
-      "logits/chosen": -1.9677941799163818,
-      "logits/rejected": -1.9717823266983032,
-      "logps/chosen": -33.100074768066406,
-      "logps/rejected": -31.62213134765625,
-      "loss": 0.5748,
-      "rewards/accuracies": 0.737500011920929,
-      "rewards/chosen": 0.28565075993537903,
-      "rewards/margins": 0.3511958718299866,
-      "rewards/rejected": -0.06554517149925232,
+      "logits/chosen": -1.970298171043396,
+      "logits/rejected": -1.974283218383789,
+      "logps/chosen": -32.81959915161133,
+      "logps/rejected": -31.408565521240234,
+      "loss": 0.4109,
+      "rewards/accuracies": 0.699999988079071,
+      "rewards/chosen": 0.5380831956863403,
+      "rewards/margins": 0.4114208221435547,
+      "rewards/rejected": 0.12666237354278564,
       "step": 230
     },
     {
       "epoch": 0.62,
       "learning_rate": 1.8712423238279358e-06,
-      "logits/chosen": -1.9661725759506226,
-      "logits/rejected": -1.944300651550293,
-      "logps/chosen": -33.841453552246094,
-      "logps/rejected": -35.11375045776367,
-      "loss": 0.5473,
-      "rewards/accuracies": 0.7875000238418579,
-      "rewards/chosen": 0.2810631990432739,
-      "rewards/margins": 0.4277234673500061,
-      "rewards/rejected": -0.14666026830673218,
+      "logits/chosen": -1.9695065021514893,
+      "logits/rejected": -1.9477574825286865,
+      "logps/chosen": -33.58247756958008,
+      "logps/rejected": -34.828121185302734,
+      "loss": 0.4129,
+      "rewards/accuracies": 0.6875,
+      "rewards/chosen": 0.5141419172286987,
+      "rewards/margins": 0.403735876083374,
+      "rewards/rejected": 0.11040612310171127,
       "step": 240
     },
     {
       "epoch": 0.65,
       "learning_rate": 1.6544367689701824e-06,
-      "logits/chosen": -2.007416009902954,
-      "logits/rejected": -2.0040948390960693,
-      "logps/chosen": -32.70330810546875,
-      "logps/rejected": -36.29412841796875,
-      "loss": 0.5992,
-      "rewards/accuracies": 0.6875,
-      "rewards/chosen": 0.1995842456817627,
-      "rewards/margins": 0.2618715763092041,
-      "rewards/rejected": -0.06228730082511902,
+      "logits/chosen": -2.0098202228546143,
+      "logits/rejected": -2.0065340995788574,
+      "logps/chosen": -32.43529510498047,
+      "logps/rejected": -35.97461700439453,
+      "loss": 0.4514,
+      "rewards/accuracies": 0.5625,
+      "rewards/chosen": 0.44079580903053284,
+      "rewards/margins": 0.2155180424451828,
+      "rewards/rejected": 0.22527781128883362,
       "step": 250
     },
     {
       "epoch": 0.68,
       "learning_rate": 1.4445974030621963e-06,
-      "logits/chosen": -1.8749721050262451,
-      "logits/rejected": -1.8725513219833374,
-      "logps/chosen": -34.00068664550781,
-      "logps/rejected": -35.53888702392578,
-      "loss": 0.6254,
-      "rewards/accuracies": 0.6875,
-      "rewards/chosen": 0.16894161701202393,
-      "rewards/margins": 0.1997825801372528,
-      "rewards/rejected": -0.030840963125228882,
+      "logits/chosen": -1.8770506381988525,
+      "logits/rejected": -1.8746120929718018,
+      "logps/chosen": -33.7199821472168,
+      "logps/rejected": -35.28092575073242,
+      "loss": 0.4498,
+      "rewards/accuracies": 0.6625000238418579,
+      "rewards/chosen": 0.42157459259033203,
+      "rewards/margins": 0.2202514111995697,
+      "rewards/rejected": 0.2013232260942459,
       "step": 260
     },
     {
       "epoch": 0.7,
       "learning_rate": 1.243452991757889e-06,
-      "logits/chosen": -1.8600317239761353,
-      "logits/rejected": -1.8576066493988037,
-      "logps/chosen": -34.1875,
-      "logps/rejected": -31.8159122467041,
-      "loss": 0.616,
-      "rewards/accuracies": 0.7124999761581421,
-      "rewards/chosen": 0.1890900433063507,
-      "rewards/margins": 0.22921428084373474,
-      "rewards/rejected": -0.04012420028448105,
+      "logits/chosen": -1.8618510961532593,
+      "logits/rejected": -1.8593294620513916,
+      "logps/chosen": -33.92017364501953,
+      "logps/rejected": -31.6002197265625,
+      "loss": 0.4397,
+      "rewards/accuracies": 0.6625000238418579,
+      "rewards/chosen": 0.42968273162841797,
+      "rewards/margins": 0.27568089962005615,
+      "rewards/rejected": 0.15400180220603943,
       "step": 270
     },
     {
       "epoch": 0.73,
       "learning_rate": 1.0526606671603523e-06,
-      "logits/chosen": -1.9631398916244507,
-      "logits/rejected": -1.9526073932647705,
-      "logps/chosen": -35.023719787597656,
-      "logps/rejected": -31.869693756103516,
-      "loss": 0.5782,
+      "logits/chosen": -1.9657011032104492,
+      "logits/rejected": -1.9552650451660156,
+      "logps/chosen": -34.72232437133789,
+      "logps/rejected": -31.632369995117188,
+      "loss": 0.4114,
       "rewards/accuracies": 0.699999988079071,
-      "rewards/chosen": 0.29963088035583496,
-      "rewards/margins": 0.32546472549438477,
-      "rewards/rejected": -0.025833839550614357,
+      "rewards/chosen": 0.5708868503570557,
+      "rewards/margins": 0.3831265866756439,
+      "rewards/rejected": 0.18776027858257294,
       "step": 280
     },
     {
       "epoch": 0.75,
       "learning_rate": 8.737922755071455e-07,
-      "logits/chosen": -2.0582926273345947,
-      "logits/rejected": -2.0433640480041504,
-      "logps/chosen": -30.733753204345703,
-      "logps/rejected": -32.67460632324219,
-      "loss": 0.6392,
-      "rewards/accuracies": 0.637499988079071,
-      "rewards/chosen": 0.17133468389511108,
-      "rewards/margins": 0.19182677567005157,
-      "rewards/rejected": -0.020492086187005043,
+      "logits/chosen": -2.0614376068115234,
+      "logits/rejected": -2.046600341796875,
+      "logps/chosen": -30.400625228881836,
+      "logps/rejected": -32.34136199951172,
+      "loss": 0.456,
+      "rewards/accuracies": 0.625,
+      "rewards/chosen": 0.4711507260799408,
+      "rewards/margins": 0.19172403216362,
+      "rewards/rejected": 0.2794266939163208,
       "step": 290
     },
     {
       "epoch": 0.78,
       "learning_rate": 7.08321427484816e-07,
-      "logits/chosen": -1.929610013961792,
-      "logits/rejected": -1.9270601272583008,
-      "logps/chosen": -32.42620086669922,
-      "logps/rejected": -30.873455047607422,
-      "loss": 0.5301,
+      "logits/chosen": -1.9332094192504883,
+      "logits/rejected": -1.9307467937469482,
+      "logps/chosen": -32.10976028442383,
+      "logps/rejected": -30.661523818969727,
+      "loss": 0.374,
       "rewards/accuracies": 0.737500011920929,
-      "rewards/chosen": 0.450817346572876,
-      "rewards/margins": 0.5018006563186646,
-      "rewards/rejected": -0.050983332097530365,
+      "rewards/chosen": 0.7356175184249878,
+      "rewards/margins": 0.5958597660064697,
+      "rewards/rejected": 0.13975778222084045,
       "step": 300
     },
     {
       "epoch": 0.78,
-      "eval_logits/chosen": -2.229154348373413,
-      "eval_logits/rejected": -2.2243051528930664,
-      "eval_logps/chosen": -34.09621810913086,
-      "eval_logps/rejected": -37.59999084472656,
-      "eval_loss": 0.6972895860671997,
-      "eval_rewards/accuracies": 0.5390365719795227,
-      "eval_rewards/chosen": -0.05550166219472885,
-      "eval_rewards/margins": 0.019528048112988472,
-      "eval_rewards/rejected": -0.07502970844507217,
-      "eval_runtime": 145.7792,
-      "eval_samples_per_second": 2.353,
+      "eval_logits/chosen": -2.2307660579681396,
+      "eval_logits/rejected": -2.22594952583313,
+      "eval_logps/chosen": -33.74896240234375,
+      "eval_logps/rejected": -37.275413513183594,
+      "eval_loss": 0.49038100242614746,
+      "eval_rewards/accuracies": 0.5220099687576294,
+      "eval_rewards/chosen": 0.2570302486419678,
+      "eval_rewards/margins": 0.039943769574165344,
+      "eval_rewards/rejected": 0.21708647906780243,
+      "eval_runtime": 145.8077,
+      "eval_samples_per_second": 2.352,
       "eval_steps_per_second": 0.295,
       "step": 300
     },
     {
       "epoch": 0.81,
       "learning_rate": 5.576113578589035e-07,
-      "logits/chosen": -1.9142345190048218,
-      "logits/rejected": -1.9109809398651123,
-      "logps/chosen": -31.33791732788086,
-      "logps/rejected": -33.82014465332031,
-      "loss": 0.5861,
-      "rewards/accuracies": 0.75,
-      "rewards/chosen": 0.23888680338859558,
-      "rewards/margins": 0.30907896161079407,
-      "rewards/rejected": -0.07019217312335968,
+      "logits/chosen": -1.9164836406707764,
+      "logits/rejected": -1.913336992263794,
+      "logps/chosen": -31.014041900634766,
+      "logps/rejected": -33.548377990722656,
+      "loss": 0.4218,
+      "rewards/accuracies": 0.762499988079071,
+      "rewards/chosen": 0.5303782224655151,
+      "rewards/margins": 0.3559793531894684,
+      "rewards/rejected": 0.1743989735841751,
       "step": 310
     },
     {
       "epoch": 0.83,
       "learning_rate": 4.229036944380913e-07,
-      "logits/chosen": -1.9650068283081055,
-      "logits/rejected": -1.9527791738510132,
-      "logps/chosen": -34.34791946411133,
-      "logps/rejected": -33.650447845458984,
-      "loss": 0.5828,
-      "rewards/accuracies": 0.6499999761581421,
-      "rewards/chosen": 0.22334297001361847,
-      "rewards/margins": 0.3237887918949127,
-      "rewards/rejected": -0.10044582933187485,
+      "logits/chosen": -1.967760682106018,
+      "logits/rejected": -1.955615758895874,
+      "logps/chosen": -34.05602264404297,
+      "logps/rejected": -33.42683410644531,
+      "loss": 0.4147,
+      "rewards/accuracies": 0.737500011920929,
+      "rewards/chosen": 0.48604661226272583,
+      "rewards/margins": 0.385240375995636,
+      "rewards/rejected": 0.10080619156360626,
       "step": 320
     },
     {
       "epoch": 0.86,
       "learning_rate": 3.053082288996112e-07,
-      "logits/chosen": -2.00040602684021,
-      "logits/rejected": -1.9989902973175049,
-      "logps/chosen": -33.210105895996094,
-      "logps/rejected": -32.56142807006836,
-      "loss": 0.5803,
-      "rewards/accuracies": 0.699999988079071,
-      "rewards/chosen": 0.24812059104442596,
-      "rewards/margins": 0.32178014516830444,
-      "rewards/rejected": -0.07365953922271729,
+      "logits/chosen": -2.003138780593872,
+      "logits/rejected": -2.001786470413208,
+      "logps/chosen": -32.86919403076172,
+      "logps/rejected": -32.247493743896484,
+      "loss": 0.4259,
+      "rewards/accuracies": 0.7124999761581421,
+      "rewards/chosen": 0.5549365878105164,
+      "rewards/margins": 0.3460560441017151,
+      "rewards/rejected": 0.20888061821460724,
       "step": 330
     },
     {
       "epoch": 0.88,
       "learning_rate": 2.0579377374915805e-07,
-      "logits/chosen": -2.0870866775512695,
-      "logits/rejected": -2.0713772773742676,
-      "logps/chosen": -33.80995178222656,
-      "logps/rejected": -33.120697021484375,
-      "loss": 0.5723,
-      "rewards/accuracies": 0.75,
-      "rewards/chosen": 0.3324963450431824,
-      "rewards/margins": 0.33488941192626953,
-      "rewards/rejected": -0.0023930787574499846,
+      "logits/chosen": -2.0895984172821045,
+      "logits/rejected": -2.073963165283203,
+      "logps/chosen": -33.46659469604492,
+      "logps/rejected": -32.82307815551758,
+      "loss": 0.4171,
+      "rewards/accuracies": 0.6875,
+      "rewards/chosen": 0.6415189504623413,
+      "rewards/margins": 0.37605759501457214,
+      "rewards/rejected": 0.26546135544776917,
       "step": 340
     },
     {
       "epoch": 0.91,
       "learning_rate": 1.2518018074041684e-07,
-      "logits/chosen": -1.959240198135376,
-      "logits/rejected": -1.9583876132965088,
-      "logps/chosen": -32.863216400146484,
-      "logps/rejected": -32.54397201538086,
-      "loss": 0.5523,
+      "logits/chosen": -1.9623206853866577,
+      "logits/rejected": -1.9614824056625366,
+      "logps/chosen": -32.549312591552734,
+      "logps/rejected": -32.24496078491211,
+      "loss": 0.4,
       "rewards/accuracies": 0.7124999761581421,
-      "rewards/chosen": 0.37699171900749207,
-      "rewards/margins": 0.44793614745140076,
-      "rewards/rejected": -0.0709443911910057,
+      "rewards/chosen": 0.6595046520233154,
+      "rewards/margins": 0.46133819222450256,
+      "rewards/rejected": 0.19816650450229645,
       "step": 350
     },
     {
       "epoch": 0.94,
       "learning_rate": 6.41315865106129e-08,
-      "logits/chosen": -1.9147189855575562,
-      "logits/rejected": -1.9250224828720093,
-      "logps/chosen": -31.902795791625977,
-      "logps/rejected": -35.3552131652832,
-      "loss": 0.5743,
-      "rewards/accuracies": 0.675000011920929,
-      "rewards/chosen": 0.2717086672782898,
-      "rewards/margins": 0.33151620626449585,
-      "rewards/rejected": -0.05980752781033516,
+      "logits/chosen": -1.9189163446426392,
+      "logits/rejected": -1.9292027950286865,
+      "logps/chosen": -31.60186767578125,
+      "logps/rejected": -34.987525939941406,
+      "loss": 0.4401,
+      "rewards/accuracies": 0.6499999761581421,
+      "rewards/chosen": 0.5425392985343933,
+      "rewards/margins": 0.27143171429634094,
+      "rewards/rejected": 0.27110758423805237,
       "step": 360
     },
     {
       "epoch": 0.96,
       "learning_rate": 2.3150941078050325e-08,
-      "logits/chosen": -2.05413556098938,
-      "logits/rejected": -2.047651767730713,
-      "logps/chosen": -33.377376556396484,
-      "logps/rejected": -29.2799072265625,
-      "loss": 0.5801,
-      "rewards/accuracies": 0.7250000238418579,
-      "rewards/chosen": 0.24713313579559326,
-      "rewards/margins": 0.2952673137187958,
-      "rewards/rejected": -0.04813414067029953,
+      "logits/chosen": -2.0577945709228516,
+      "logits/rejected": -2.051274061203003,
+      "logps/chosen": -33.05121612548828,
+      "logps/rejected": -28.990320205688477,
+      "loss": 0.4233,
+      "rewards/accuracies": 0.699999988079071,
+      "rewards/chosen": 0.5406805276870728,
+      "rewards/margins": 0.3281847834587097,
+      "rewards/rejected": 0.21249575912952423,
       "step": 370
     },
     {
       "epoch": 0.99,
       "learning_rate": 2.575864278703266e-09,
-      "logits/chosen": -1.9141871929168701,
-      "logits/rejected": -1.9163949489593506,
-      "logps/chosen": -33.87698745727539,
-      "logps/rejected": -30.976858139038086,
-      "loss": 0.5489,
-      "rewards/accuracies": 0.699999988079071,
-      "rewards/chosen": 0.32585546374320984,
-      "rewards/margins": 0.4196627140045166,
-      "rewards/rejected": -0.09380728751420975,
+      "logits/chosen": -1.917284607887268,
+      "logits/rejected": -1.9194421768188477,
+      "logps/chosen": -33.59749984741211,
+      "logps/rejected": -30.708057403564453,
+      "loss": 0.4061,
+      "rewards/accuracies": 0.762499988079071,
+      "rewards/chosen": 0.5773938298225403,
+      "rewards/margins": 0.429278701543808,
+      "rewards/rejected": 0.14811506867408752,
       "step": 380
     },
     {
       "epoch": 1.0,
       "step": 385,
       "total_flos": 0.0,
-      "train_loss": 0.6175476637753573,
-      "train_runtime": 3252.7839,
-      "train_samples_per_second": 0.947,
+      "train_loss": 0.447351616079157,
+      "train_runtime": 3253.4458,
+      "train_samples_per_second": 0.946,
       "train_steps_per_second": 0.118
     }
   ],