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: 0.9781
-- Rewards/chosen: -0.0159
-- Rewards/rejected: -0.0383
-- Rewards/accuracies: 0.5170
-- Rewards/margins: 0.0224
-- Logps/rejected: -37.5644
-- Logps/chosen: -34.0544
-- Logits/rejected: -2.2184
-- Logits/chosen: -2.2232
+- Loss: 0.6973
+- Rewards/chosen: -0.0555
+- Rewards/rejected: -0.0750
+- Rewards/accuracies: 0.5390
+- Rewards/margins: 0.0195
+- Logps/rejected: -37.6000
+- Logps/chosen: -34.0962
+- Logits/rejected: -2.2243
+- Logits/chosen: -2.2292
 
 ## 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 |
 |:-------------:|:-----:|:----:|:---------------:|:--------------:|:----------------:|:------------------:|:---------------:|:--------------:|:------------:|:---------------:|:-------------:|
-| 0.9007        | 0.26  | 100  | 0.9692          | 0.0124         | -0.0188          | 0.5752             | 0.0312          | -37.5401       | -34.0190     | -2.2262         | -2.2310       |
-| 0.7243        | 0.52  | 200  | 0.9691          | 0.0136         | -0.0181          | 0.5656             | 0.0317          | -37.5392       | -34.0175     | -2.2232         | -2.2280       |
-| 0.6515        | 0.78  | 300  | 0.9644          | -0.0061        | -0.0440          | 0.5602             | 0.0378          | -37.5716       | -34.0422     | -2.2186         | -2.2234       |
+| 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       |
 
 
 ### Framework versions

adapter_model.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:64cfd5abbbb41907ae32f148bf01549e450f38c8c806cac936b2152f636757da
+oid sha256:955ddc6d94a67cda7538a1b702c976860689f0909cc722268fc8280507e6d239
 size 176183216

all_results.json

@@ -1,20 +1,7 @@
 {
     "epoch": 1.0,
-    "eval_logits/chosen": -2.2231976985931396,
-    "eval_logits/rejected": -2.218374013900757,
-    "eval_logps/chosen": -34.05440139770508,
-    "eval_logps/rejected": -37.5644416809082,
-    "eval_loss": 0.9781046509742737,
-    "eval_rewards/accuracies": 0.5170266032218933,
-    "eval_rewards/chosen": -0.01588033325970173,
-    "eval_rewards/margins": 0.02237490564584732,
-    "eval_rewards/rejected": -0.0382552333176136,
-    "eval_runtime": 145.5341,
-    "eval_samples": 343,
-    "eval_samples_per_second": 2.357,
-    "eval_steps_per_second": 0.295,
-    "train_loss": 0.8298485310046704,
-    "train_runtime": 3249.7236,
+    "train_loss": 0.6175476637753573,
+    "train_runtime": 3252.7839,
     "train_samples": 3079,
     "train_samples_per_second": 0.947,
     "train_steps_per_second": 0.118

train_results.json

@@ -1,7 +1,7 @@
 {
     "epoch": 1.0,
-    "train_loss": 0.8298485310046704,
-    "train_runtime": 3249.7236,
+    "train_loss": 0.6175476637753573,
+    "train_runtime": 3252.7839,
     "train_samples": 3079,
     "train_samples_per_second": 0.947,
     "train_steps_per_second": 0.118

trainer_state.json

@@ -15,7 +15,7 @@
       "logits/rejected": -1.7377450466156006,
       "logps/chosen": -29.553977966308594,
       "logps/rejected": -42.813133239746094,
-      "loss": 1.0,
+      "loss": 0.6931,
       "rewards/accuracies": 0.0,
       "rewards/chosen": 0.0,
       "rewards/margins": 0.0,
@@ -25,155 +25,155 @@
     {
       "epoch": 0.03,
       "learning_rate": 1.282051282051282e-06,
-      "logits/chosen": -1.8667426109313965,
-      "logits/rejected": -1.8710602521896362,
-      "logps/chosen": -36.991912841796875,
-      "logps/rejected": -33.67206954956055,
-      "loss": 0.9547,
-      "rewards/accuracies": 0.5694444179534912,
-      "rewards/chosen": 0.011750025674700737,
-      "rewards/margins": 0.04534290358424187,
-      "rewards/rejected": -0.03359287604689598,
+      "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,
       "step": 10
     },
     {
       "epoch": 0.05,
       "learning_rate": 2.564102564102564e-06,
-      "logits/chosen": -1.9977442026138306,
-      "logits/rejected": -2.0003952980041504,
-      "logps/chosen": -29.659366607666016,
-      "logps/rejected": -29.05437660217285,
-      "loss": 1.021,
-      "rewards/accuracies": 0.4124999940395355,
-      "rewards/chosen": -0.013736436143517494,
-      "rewards/margins": -0.02098780684173107,
-      "rewards/rejected": 0.007251373026520014,
+      "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,
       "step": 20
     },
     {
       "epoch": 0.08,
       "learning_rate": 3.846153846153847e-06,
-      "logits/chosen": -1.920693039894104,
-      "logits/rejected": -1.91802179813385,
-      "logps/chosen": -31.39971351623535,
-      "logps/rejected": -33.21495819091797,
-      "loss": 0.9897,
-      "rewards/accuracies": 0.5625,
-      "rewards/chosen": 0.013103686273097992,
-      "rewards/margins": 0.010304747149348259,
-      "rewards/rejected": 0.002798942383378744,
+      "logits/chosen": -1.921021819114685,
+      "logits/rejected": -1.9183374643325806,
+      "logps/chosen": -31.40532875061035,
+      "logps/rejected": -33.23241424560547,
+      "loss": 0.6877,
+      "rewards/accuracies": 0.5249999761581421,
+      "rewards/chosen": 0.00968973059207201,
+      "rewards/margins": 0.022251319140195847,
+      "rewards/rejected": -0.012561586685478687,
       "step": 30
     },
     {
       "epoch": 0.1,
       "learning_rate": 4.999896948438434e-06,
-      "logits/chosen": -2.018057107925415,
-      "logits/rejected": -2.0093047618865967,
-      "logps/chosen": -32.565284729003906,
-      "logps/rejected": -32.50053405761719,
-      "loss": 1.0005,
-      "rewards/accuracies": 0.5375000238418579,
-      "rewards/chosen": 0.009215259924530983,
-      "rewards/margins": -0.0004936732584610581,
-      "rewards/rejected": 0.009708933532238007,
+      "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,
       "step": 40
     },
     {
       "epoch": 0.13,
       "learning_rate": 4.987541037542187e-06,
-      "logits/chosen": -1.8627817630767822,
-      "logits/rejected": -1.851999044418335,
-      "logps/chosen": -33.549964904785156,
-      "logps/rejected": -35.44340896606445,
-      "loss": 1.0011,
-      "rewards/accuracies": 0.4749999940395355,
-      "rewards/chosen": 0.0060120681300759315,
-      "rewards/margins": -0.0011377219343557954,
-      "rewards/rejected": 0.007149793207645416,
+      "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,
       "step": 50
     },
     {
       "epoch": 0.16,
       "learning_rate": 4.954691471941119e-06,
-      "logits/chosen": -1.9416770935058594,
-      "logits/rejected": -1.9436094760894775,
-      "logps/chosen": -32.53351593017578,
-      "logps/rejected": -33.217529296875,
-      "loss": 0.9247,
-      "rewards/accuracies": 0.5625,
-      "rewards/chosen": 0.05333293229341507,
-      "rewards/margins": 0.09025315940380096,
-      "rewards/rejected": -0.036920223385095596,
+      "logits/chosen": -1.9400945901870728,
+      "logits/rejected": -1.9420464038848877,
+      "logps/chosen": -32.56509780883789,
+      "logps/rejected": -33.2406120300293,
+      "loss": 0.6632,
+      "rewards/accuracies": 0.6499999761581421,
+      "rewards/chosen": 0.031578924506902695,
+      "rewards/margins": 0.09388783574104309,
+      "rewards/rejected": -0.062308914959430695,
       "step": 60
     },
     {
       "epoch": 0.18,
       "learning_rate": 4.901618883413549e-06,
-      "logits/chosen": -2.072779655456543,
-      "logits/rejected": -2.077756881713867,
-      "logps/chosen": -34.002342224121094,
-      "logps/rejected": -36.633216857910156,
-      "loss": 0.9596,
+      "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": -0.011083832010626793,
-      "rewards/margins": 0.040393490344285965,
-      "rewards/rejected": -0.05147732421755791,
+      "rewards/chosen": 0.005918038543313742,
+      "rewards/margins": 0.05520814657211304,
+      "rewards/rejected": -0.04929010197520256,
       "step": 70
     },
     {
       "epoch": 0.21,
       "learning_rate": 4.828760511501322e-06,
-      "logits/chosen": -1.9329026937484741,
-      "logits/rejected": -1.936031699180603,
-      "logps/chosen": -34.33915710449219,
-      "logps/rejected": -34.65736770629883,
-      "loss": 0.8892,
-      "rewards/accuracies": 0.5625,
-      "rewards/chosen": 0.07092130184173584,
-      "rewards/margins": 0.12288935482501984,
-      "rewards/rejected": -0.051968056708574295,
+      "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,
       "step": 80
     },
     {
       "epoch": 0.23,
       "learning_rate": 4.7367166013034295e-06,
-      "logits/chosen": -1.9402366876602173,
-      "logits/rejected": -1.9447383880615234,
-      "logps/chosen": -32.383628845214844,
-      "logps/rejected": -32.351661682128906,
-      "loss": 0.9388,
-      "rewards/accuracies": 0.574999988079071,
-      "rewards/chosen": 0.06703362613916397,
-      "rewards/margins": 0.06123671680688858,
-      "rewards/rejected": 0.0057969121262431145,
+      "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,
       "step": 90
     },
     {
       "epoch": 0.26,
       "learning_rate": 4.626245458345211e-06,
-      "logits/chosen": -2.0364508628845215,
-      "logits/rejected": -2.0344765186309814,
-      "logps/chosen": -32.12981033325195,
-      "logps/rejected": -31.279254913330078,
-      "loss": 0.9007,
-      "rewards/accuracies": 0.612500011920929,
-      "rewards/chosen": 0.08881844580173492,
-      "rewards/margins": 0.09925105422735214,
-      "rewards/rejected": -0.010432596318423748,
+      "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,
       "step": 100
     },
     {
       "epoch": 0.26,
-      "eval_logits/chosen": -2.231006622314453,
-      "eval_logits/rejected": -2.2261619567871094,
-      "eval_logps/chosen": -34.019046783447266,
-      "eval_logps/rejected": -37.540069580078125,
-      "eval_loss": 0.96916264295578,
-      "eval_rewards/accuracies": 0.5751661062240601,
-      "eval_rewards/chosen": 0.012404282577335835,
-      "eval_rewards/margins": 0.031165316700935364,
-      "eval_rewards/rejected": -0.018761036917567253,
-      "eval_runtime": 146.0189,
+      "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,
       "step": 100
@@ -181,311 +181,311 @@
     {
       "epoch": 0.29,
       "learning_rate": 4.498257201263691e-06,
-      "logits/chosen": -1.991776466369629,
-      "logits/rejected": -1.9893957376480103,
-      "logps/chosen": -33.122779846191406,
-      "logps/rejected": -33.99274444580078,
-      "loss": 0.9495,
-      "rewards/accuracies": 0.6000000238418579,
-      "rewards/chosen": 0.09682749211788177,
-      "rewards/margins": 0.0675249844789505,
-      "rewards/rejected": 0.029302507638931274,
+      "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,
       "step": 110
     },
     {
       "epoch": 0.31,
       "learning_rate": 4.353806263777678e-06,
-      "logits/chosen": -2.003622531890869,
-      "logits/rejected": -1.9952924251556396,
-      "logps/chosen": -32.31382369995117,
-      "logps/rejected": -32.122901916503906,
-      "loss": 0.9341,
-      "rewards/accuracies": 0.5874999761581421,
-      "rewards/chosen": 0.10486602783203125,
-      "rewards/margins": 0.07398126274347305,
-      "rewards/rejected": 0.030884763225913048,
+      "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,
       "step": 120
     },
     {
       "epoch": 0.34,
       "learning_rate": 4.1940827077152755e-06,
-      "logits/chosen": -2.0311331748962402,
-      "logits/rejected": -2.0231704711914062,
-      "logps/chosen": -30.32816505432129,
-      "logps/rejected": -32.052425384521484,
-      "loss": 0.9159,
-      "rewards/accuracies": 0.574999988079071,
-      "rewards/chosen": 0.11803986132144928,
-      "rewards/margins": 0.11757204681634903,
-      "rewards/rejected": 0.0004678152617998421,
+      "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,
       "step": 130
     },
     {
       "epoch": 0.36,
       "learning_rate": 4.0204024186666215e-06,
-      "logits/chosen": -1.9613821506500244,
-      "logits/rejected": -1.9716154336929321,
-      "logps/chosen": -31.22269630432129,
-      "logps/rejected": -32.548851013183594,
-      "loss": 0.8406,
-      "rewards/accuracies": 0.6499999761581421,
-      "rewards/chosen": 0.15757359564304352,
-      "rewards/margins": 0.16875064373016357,
-      "rewards/rejected": -0.01117704901844263,
+      "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,
       "step": 140
     },
     {
       "epoch": 0.39,
       "learning_rate": 3.834196265035119e-06,
-      "logits/chosen": -1.8720792531967163,
-      "logits/rejected": -1.8732519149780273,
-      "logps/chosen": -33.89937973022461,
-      "logps/rejected": -34.796844482421875,
-      "loss": 0.7907,
-      "rewards/accuracies": 0.5874999761581421,
-      "rewards/chosen": 0.23464930057525635,
-      "rewards/margins": 0.2663186490535736,
-      "rewards/rejected": -0.03166933357715607,
+      "logits/chosen": -1.876604437828064,
+      "logits/rejected": -1.8777605295181274,
+      "logps/chosen": -33.938690185546875,
+      "logps/rejected": -34.807891845703125,
+      "loss": 0.6043,
+      "rewards/accuracies": 0.612500011920929,
+      "rewards/chosen": 0.22860188782215118,
+      "rewards/margins": 0.2741745412349701,
+      "rewards/rejected": -0.0455726757645607,
       "step": 150
     },
     {
       "epoch": 0.42,
       "learning_rate": 3.636998309800573e-06,
-      "logits/chosen": -1.923437476158142,
-      "logits/rejected": -1.9200271368026733,
-      "logps/chosen": -35.97971725463867,
-      "logps/rejected": -32.6976203918457,
-      "loss": 0.8704,
-      "rewards/accuracies": 0.625,
-      "rewards/chosen": 0.1535695493221283,
-      "rewards/margins": 0.1334635317325592,
-      "rewards/rejected": 0.02010601945221424,
+      "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,
       "step": 160
     },
     {
       "epoch": 0.44,
       "learning_rate": 3.4304331721118078e-06,
-      "logits/chosen": -2.0249733924865723,
-      "logits/rejected": -2.017641067504883,
-      "logps/chosen": -33.441261291503906,
-      "logps/rejected": -31.389623641967773,
-      "loss": 0.7234,
-      "rewards/accuracies": 0.7124999761581421,
-      "rewards/chosen": 0.2852834165096283,
-      "rewards/margins": 0.31550443172454834,
-      "rewards/rejected": -0.03022097982466221,
+      "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,
       "step": 170
     },
     {
       "epoch": 0.47,
       "learning_rate": 3.2162026428305436e-06,
-      "logits/chosen": -2.031703472137451,
-      "logits/rejected": -2.036947727203369,
-      "logps/chosen": -32.208961486816406,
-      "logps/rejected": -32.41345977783203,
-      "loss": 0.8097,
+      "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": 0.26866063475608826,
-      "rewards/margins": 0.21131709218025208,
-      "rewards/rejected": 0.05734356492757797,
+      "rewards/chosen": 0.2779761850833893,
+      "rewards/margins": 0.2557251751422882,
+      "rewards/rejected": 0.02225096896290779,
       "step": 180
     },
     {
       "epoch": 0.49,
       "learning_rate": 2.996071664294641e-06,
-      "logits/chosen": -2.032174587249756,
-      "logits/rejected": -2.029404401779175,
-      "logps/chosen": -31.245046615600586,
-      "logps/rejected": -31.299081802368164,
-      "loss": 0.8381,
-      "rewards/accuracies": 0.625,
-      "rewards/chosen": 0.19512517750263214,
-      "rewards/margins": 0.17981843650341034,
-      "rewards/rejected": 0.015306718647480011,
+      "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,
       "step": 190
     },
     {
       "epoch": 0.52,
       "learning_rate": 2.7718537898066833e-06,
-      "logits/chosen": -1.9025481939315796,
-      "logits/rejected": -1.9071909189224243,
-      "logps/chosen": -31.24801254272461,
-      "logps/rejected": -32.809139251708984,
-      "loss": 0.7243,
-      "rewards/accuracies": 0.762499988079071,
-      "rewards/chosen": 0.28661441802978516,
-      "rewards/margins": 0.3011789917945862,
-      "rewards/rejected": -0.014564569108188152,
+      "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,
       "step": 200
     },
     {
       "epoch": 0.52,
-      "eval_logits/chosen": -2.227985382080078,
-      "eval_logits/rejected": -2.2231621742248535,
-      "eval_logps/chosen": -34.01750946044922,
-      "eval_logps/rejected": -37.539207458496094,
-      "eval_loss": 0.9690985083580017,
-      "eval_rewards/accuracies": 0.565614640712738,
-      "eval_rewards/chosen": 0.013635948300361633,
-      "eval_rewards/margins": 0.03170585632324219,
-      "eval_rewards/rejected": -0.018069909885525703,
-      "eval_runtime": 145.4584,
-      "eval_samples_per_second": 2.358,
-      "eval_steps_per_second": 0.296,
+      "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_steps_per_second": 0.295,
       "step": 200
     },
     {
       "epoch": 0.55,
       "learning_rate": 2.5453962426402006e-06,
-      "logits/chosen": -2.0139317512512207,
-      "logits/rejected": -2.0245535373687744,
-      "logps/chosen": -31.7435359954834,
-      "logps/rejected": -33.9241828918457,
-      "loss": 0.7947,
-      "rewards/accuracies": 0.6499999761581421,
-      "rewards/chosen": 0.1822136789560318,
-      "rewards/margins": 0.23139998316764832,
-      "rewards/rejected": -0.04918632656335831,
+      "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,
       "step": 210
     },
     {
       "epoch": 0.57,
       "learning_rate": 2.3185646976551794e-06,
-      "logits/chosen": -1.9063589572906494,
-      "logits/rejected": -1.9211170673370361,
-      "logps/chosen": -29.79791831970215,
-      "logps/rejected": -31.584829330444336,
-      "loss": 0.7478,
+      "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": 0.2508983910083771,
-      "rewards/margins": 0.2722209095954895,
-      "rewards/rejected": -0.02132250741124153,
+      "rewards/chosen": 0.23883743584156036,
+      "rewards/margins": 0.2899848222732544,
+      "rewards/rejected": -0.051147449761629105,
       "step": 220
     },
     {
       "epoch": 0.6,
       "learning_rate": 2.0932279108998323e-06,
-      "logits/chosen": -1.9628435373306274,
-      "logits/rejected": -1.9668251276016235,
-      "logps/chosen": -33.099082946777344,
-      "logps/rejected": -31.59071922302246,
-      "loss": 0.7716,
-      "rewards/accuracies": 0.675000011920929,
-      "rewards/chosen": 0.25470608472824097,
-      "rewards/margins": 0.28783971071243286,
-      "rewards/rejected": -0.033133648335933685,
+      "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,
       "step": 230
     },
     {
       "epoch": 0.62,
       "learning_rate": 1.8712423238279358e-06,
-      "logits/chosen": -1.9614289999008179,
-      "logits/rejected": -1.939612627029419,
-      "logps/chosen": -33.833091735839844,
-      "logps/rejected": -35.08003234863281,
-      "loss": 0.7052,
-      "rewards/accuracies": 0.6875,
-      "rewards/chosen": 0.25653010606765747,
-      "rewards/margins": 0.3599211871623993,
-      "rewards/rejected": -0.10339111089706421,
+      "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,
       "step": 240
     },
     {
       "epoch": 0.65,
       "learning_rate": 1.6544367689701824e-06,
-      "logits/chosen": -2.0032362937927246,
-      "logits/rejected": -1.999916434288025,
-      "logps/chosen": -32.70505142211914,
-      "logps/rejected": -36.224037170410156,
-      "loss": 0.8429,
-      "rewards/accuracies": 0.6499999761581421,
-      "rewards/chosen": 0.17601460218429565,
-      "rewards/margins": 0.17530463635921478,
-      "rewards/rejected": 0.0007099613430909812,
+      "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,
       "step": 250
     },
     {
       "epoch": 0.68,
       "learning_rate": 1.4445974030621963e-06,
-      "logits/chosen": -1.8701452016830444,
-      "logits/rejected": -1.8677012920379639,
-      "logps/chosen": -33.97339630126953,
-      "logps/rejected": -35.50096130371094,
-      "loss": 0.8408,
+      "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": 0.17200371623039246,
-      "rewards/margins": 0.16907431185245514,
-      "rewards/rejected": 0.002929417882114649,
+      "rewards/chosen": 0.16894161701202393,
+      "rewards/margins": 0.1997825801372528,
+      "rewards/rejected": -0.030840963125228882,
       "step": 260
     },
     {
       "epoch": 0.7,
       "learning_rate": 1.243452991757889e-06,
-      "logits/chosen": -1.8542945384979248,
-      "logits/rejected": -1.851894736289978,
-      "logps/chosen": -34.1937370300293,
-      "logps/rejected": -31.8076114654541,
-      "loss": 0.8124,
-      "rewards/accuracies": 0.6625000238418579,
-      "rewards/chosen": 0.1630934774875641,
-      "rewards/margins": 0.1921185553073883,
-      "rewards/rejected": -0.029025081545114517,
+      "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,
       "step": 270
     },
     {
       "epoch": 0.73,
       "learning_rate": 1.0526606671603523e-06,
-      "logits/chosen": -1.9576352834701538,
-      "logits/rejected": -1.9471737146377563,
-      "logps/chosen": -34.961952209472656,
-      "logps/rejected": -31.824676513671875,
-      "loss": 0.7162,
-      "rewards/accuracies": 0.75,
-      "rewards/chosen": 0.31575411558151245,
-      "rewards/margins": 0.30270126461982727,
-      "rewards/rejected": 0.013052871450781822,
+      "logits/chosen": -1.9631398916244507,
+      "logits/rejected": -1.9526073932647705,
+      "logps/chosen": -35.023719787597656,
+      "logps/rejected": -31.869693756103516,
+      "loss": 0.5782,
+      "rewards/accuracies": 0.699999988079071,
+      "rewards/chosen": 0.29963088035583496,
+      "rewards/margins": 0.32546472549438477,
+      "rewards/rejected": -0.025833839550614357,
       "step": 280
     },
     {
       "epoch": 0.75,
       "learning_rate": 8.737922755071455e-07,
-      "logits/chosen": -2.052537202835083,
-      "logits/rejected": -2.037666082382202,
-      "logps/chosen": -30.679821014404297,
-      "logps/rejected": -32.591346740722656,
-      "loss": 0.8779,
-      "rewards/accuracies": 0.6000000238418579,
-      "rewards/chosen": 0.1954430788755417,
-      "rewards/margins": 0.14705480635166168,
-      "rewards/rejected": 0.048388272523880005,
+      "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,
       "step": 290
     },
     {
       "epoch": 0.78,
       "learning_rate": 7.08321427484816e-07,
-      "logits/chosen": -1.9234060049057007,
-      "logits/rejected": -1.9208656549453735,
-      "logps/chosen": -32.422889709472656,
-      "logps/rejected": -30.887353897094727,
-      "loss": 0.6515,
-      "rewards/accuracies": 0.75,
-      "rewards/chosen": 0.4033745229244232,
-      "rewards/margins": 0.45980948209762573,
-      "rewards/rejected": -0.056434907019138336,
+      "logits/chosen": -1.929610013961792,
+      "logits/rejected": -1.9270601272583008,
+      "logps/chosen": -32.42620086669922,
+      "logps/rejected": -30.873455047607422,
+      "loss": 0.5301,
+      "rewards/accuracies": 0.737500011920929,
+      "rewards/chosen": 0.450817346572876,
+      "rewards/margins": 0.5018006563186646,
+      "rewards/rejected": -0.050983332097530365,
       "step": 300
     },
     {
       "epoch": 0.78,
-      "eval_logits/chosen": -2.2234299182891846,
-      "eval_logits/rejected": -2.218611001968384,
-      "eval_logps/chosen": -34.04222106933594,
-      "eval_logps/rejected": -37.57155990600586,
-      "eval_loss": 0.9644458889961243,
-      "eval_rewards/accuracies": 0.560215950012207,
-      "eval_rewards/chosen": -0.006133326329290867,
-      "eval_rewards/margins": 0.03781980276107788,
-      "eval_rewards/rejected": -0.04395313188433647,
-      "eval_runtime": 145.7457,
+      "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_steps_per_second": 0.295,
       "step": 300
@@ -493,121 +493,121 @@
     {
       "epoch": 0.81,
       "learning_rate": 5.576113578589035e-07,
-      "logits/chosen": -1.9086744785308838,
-      "logits/rejected": -1.905432939529419,
-      "logps/chosen": -31.29376220703125,
-      "logps/rejected": -33.765647888183594,
-      "loss": 0.7696,
-      "rewards/accuracies": 0.7124999761581421,
-      "rewards/chosen": 0.24766740202903748,
-      "rewards/margins": 0.2664671242237091,
-      "rewards/rejected": -0.01879967749118805,
+      "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,
       "step": 310
     },
     {
       "epoch": 0.83,
       "learning_rate": 4.229036944380913e-07,
-      "logits/chosen": -1.958169937133789,
-      "logits/rejected": -1.9459987878799438,
-      "logps/chosen": -34.305747985839844,
-      "logps/rejected": -33.66083526611328,
-      "loss": 0.7197,
-      "rewards/accuracies": 0.6875,
-      "rewards/chosen": 0.23226289451122284,
-      "rewards/margins": 0.32985854148864746,
-      "rewards/rejected": -0.0975956842303276,
+      "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,
       "step": 320
     },
     {
       "epoch": 0.86,
       "learning_rate": 3.053082288996112e-07,
-      "logits/chosen": -1.9929885864257812,
-      "logits/rejected": -1.9915746450424194,
-      "logps/chosen": -33.1512336730957,
-      "logps/rejected": -32.53498077392578,
-      "loss": 0.7279,
-      "rewards/accuracies": 0.75,
-      "rewards/chosen": 0.26764601469039917,
-      "rewards/margins": 0.3119625747203827,
-      "rewards/rejected": -0.04431656002998352,
+      "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,
       "step": 330
     },
     {
       "epoch": 0.88,
       "learning_rate": 2.0579377374915805e-07,
-      "logits/chosen": -2.0791096687316895,
-      "logits/rejected": -2.063427686691284,
-      "logps/chosen": -33.73707580566406,
-      "logps/rejected": -33.045745849609375,
-      "loss": 0.7446,
-      "rewards/accuracies": 0.6875,
-      "rewards/chosen": 0.3538528084754944,
-      "rewards/margins": 0.2960215210914612,
-      "rewards/rejected": 0.05783123895525932,
+      "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,
       "step": 340
     },
     {
       "epoch": 0.91,
       "learning_rate": 1.2518018074041684e-07,
-      "logits/chosen": -1.952497124671936,
-      "logits/rejected": -1.951664686203003,
-      "logps/chosen": -32.81846618652344,
-      "logps/rejected": -32.528770446777344,
-      "loss": 0.6539,
-      "rewards/accuracies": 0.75,
-      "rewards/chosen": 0.37090426683425903,
-      "rewards/margins": 0.42179951071739197,
-      "rewards/rejected": -0.05089529603719711,
+      "logits/chosen": -1.959240198135376,
+      "logits/rejected": -1.9583876132965088,
+      "logps/chosen": -32.863216400146484,
+      "logps/rejected": -32.54397201538086,
+      "loss": 0.5523,
+      "rewards/accuracies": 0.7124999761581421,
+      "rewards/chosen": 0.37699171900749207,
+      "rewards/margins": 0.44793614745140076,
+      "rewards/rejected": -0.0709443911910057,
       "step": 350
     },
     {
       "epoch": 0.94,
       "learning_rate": 6.41315865106129e-08,
-      "logits/chosen": -1.907370924949646,
-      "logits/rejected": -1.9176517724990845,
-      "logps/chosen": -31.86977195739746,
-      "logps/rejected": -35.2895393371582,
-      "loss": 0.7562,
+      "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": 0.2679324746131897,
-      "rewards/margins": 0.26855653524398804,
-      "rewards/rejected": -0.0006240725633688271,
+      "rewards/chosen": 0.2717086672782898,
+      "rewards/margins": 0.33151620626449585,
+      "rewards/rejected": -0.05980752781033516,
       "step": 360
     },
     {
       "epoch": 0.96,
       "learning_rate": 2.3150941078050325e-08,
-      "logits/chosen": -2.047156572341919,
-      "logits/rejected": -2.0407111644744873,
-      "logps/chosen": -33.288124084472656,
-      "logps/rejected": -29.224395751953125,
-      "loss": 0.7344,
-      "rewards/accuracies": 0.699999988079071,
-      "rewards/chosen": 0.2910745441913605,
-      "rewards/margins": 0.28944963216781616,
-      "rewards/rejected": 0.0016249760519713163,
+      "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,
       "step": 370
     },
     {
       "epoch": 0.99,
       "learning_rate": 2.575864278703266e-09,
-      "logits/chosen": -1.9066269397735596,
-      "logits/rejected": -1.9088102579116821,
-      "logps/chosen": -33.847068786621094,
-      "logps/rejected": -30.924524307250977,
-      "loss": 0.7185,
-      "rewards/accuracies": 0.7250000238418579,
-      "rewards/chosen": 0.31358352303504944,
-      "rewards/margins": 0.35510215163230896,
-      "rewards/rejected": -0.04151865094900131,
+      "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,
       "step": 380
     },
     {
       "epoch": 1.0,
       "step": 385,
       "total_flos": 0.0,
-      "train_loss": 0.8298485310046704,
-      "train_runtime": 3249.7236,
+      "train_loss": 0.6175476637753573,
+      "train_runtime": 3252.7839,
       "train_samples_per_second": 0.947,
       "train_steps_per_second": 0.118
     }