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: 0.9960
-- Rewards/chosen: 0.0097
-- Rewards/rejected: 0.0053
-- Rewards/accuracies: 0.5141
-- Rewards/margins: 0.0044
-- Logps/rejected: -37.5100
-- Logps/chosen: -34.0224
-- Logits/rejected: -2.2385
-- Logits/chosen: -2.2434
+- Loss: 1.0120
+- Rewards/chosen: 0.0035
+- Rewards/rejected: 0.0155
+- Rewards/accuracies: 0.4726
+- Rewards/margins: -0.0120
+- Logps/rejected: -37.4994
+- Logps/chosen: -34.0306
+- Logits/rejected: -2.2388
+- Logits/chosen: -2.2437
 
 ## 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 |
 |:-------------:|:-----:|:----:|:---------------:|:--------------:|:----------------:|:------------------:|:---------------:|:--------------:|:------------:|:---------------:|:-------------:|
-| 1.0026        | 0.26  | 100  | 1.0066          | 0.0091         | 0.0155           | 0.4983             | -0.0064         | -37.4972       | -34.0232     | -2.2384         | -2.2433       |
-| 0.9923        | 0.52  | 200  | 0.9864          | 0.0127         | -0.0007          | 0.5328             | 0.0134          | -37.5175       | -34.0187     | -2.2389         | -2.2438       |
-| 0.9775        | 0.78  | 300  | 0.9998          | 0.0135         | 0.0132           | 0.4904             | 0.0004          | -37.5002       | -34.0177     | -2.2386         | -2.2435       |
+| 0.9823        | 0.26  | 100  | 0.9982          | 0.0116         | 0.0097           | 0.5104             | 0.0019          | -37.5058       | -34.0216     | -2.2391         | -2.2440       |
+| 0.9696        | 0.52  | 200  | 1.0014          | 0.0116         | 0.0128           | 0.4693             | -0.0012         | -37.5024       | -34.0217     | -2.2390         | -2.2439       |
+| 0.9581        | 0.78  | 300  | 1.0120          | 0.0035         | 0.0155           | 0.4726             | -0.0120         | -37.4994       | -34.0306     | -2.2388         | -2.2437       |
 
 
 ### Framework versions

adapter_model.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:1df4cb76dc42b0eb8cc1e7764c3730bed662ce4db821e8573a54ef4df89941bb
+oid sha256:a1e7f79a4fe0164d309e418b64bfbaabd64069413317820121adc83c26c731cb
 size 176183216

all_results.json

@@ -1,20 +1,7 @@
 {
     "epoch": 1.0,
-    "eval_logits/chosen": -2.2434139251708984,
-    "eval_logits/rejected": -2.2385354042053223,
-    "eval_logps/chosen": -34.0224494934082,
-    "eval_logps/rejected": -37.51002883911133,
-    "eval_loss": 0.9960440397262573,
-    "eval_rewards/accuracies": 0.5141196250915527,
-    "eval_rewards/chosen": 0.0096820630133152,
-    "eval_rewards/margins": 0.004409347660839558,
-    "eval_rewards/rejected": 0.005272718146443367,
-    "eval_runtime": 145.6562,
-    "eval_samples": 343,
-    "eval_samples_per_second": 2.355,
-    "eval_steps_per_second": 0.295,
-    "train_loss": 0.9929689927534624,
-    "train_runtime": 3255.1859,
+    "train_loss": 0.9887851405453373,
+    "train_runtime": 3254.6365,
     "train_samples": 3079,
     "train_samples_per_second": 0.946,
     "train_steps_per_second": 0.118

train_results.json

@@ -1,7 +1,7 @@
 {
     "epoch": 1.0,
-    "train_loss": 0.9929689927534624,
-    "train_runtime": 3255.1859,
+    "train_loss": 0.9887851405453373,
+    "train_runtime": 3254.6365,
     "train_samples": 3079,
     "train_samples_per_second": 0.946,
     "train_steps_per_second": 0.118

trainer_state.json

@@ -25,589 +25,589 @@
     {
       "epoch": 0.03,
       "learning_rate": 1.2820512820512818e-07,
-      "logits/chosen": -1.866065502166748,
-      "logits/rejected": -1.8703795671463013,
-      "logps/chosen": -36.988380432128906,
-      "logps/rejected": -33.66728210449219,
-      "loss": 0.9557,
-      "rewards/accuracies": 0.5694444179534912,
-      "rewards/chosen": 0.014575847424566746,
-      "rewards/margins": 0.04434243589639664,
-      "rewards/rejected": -0.02976658195257187,
+      "logits/chosen": -1.8661272525787354,
+      "logits/rejected": -1.8704373836517334,
+      "logps/chosen": -36.993247985839844,
+      "logps/rejected": -33.67000961303711,
+      "loss": 0.952,
+      "rewards/accuracies": 0.5138888955116272,
+      "rewards/chosen": 0.012012440711259842,
+      "rewards/margins": 0.047955650836229324,
+      "rewards/rejected": -0.035943206399679184,
       "step": 10
     },
     {
       "epoch": 0.05,
       "learning_rate": 2.5641025641025636e-07,
-      "logits/chosen": -1.997332215309143,
-      "logits/rejected": -1.999983549118042,
-      "logps/chosen": -29.625896453857422,
-      "logps/rejected": -29.035802841186523,
-      "loss": 1.0091,
-      "rewards/accuracies": 0.4375,
-      "rewards/chosen": 0.013037634082138538,
-      "rewards/margins": -0.009074507281184196,
-      "rewards/rejected": 0.02211214043200016,
+      "logits/chosen": -1.9975776672363281,
+      "logits/rejected": -2.000213146209717,
+      "logps/chosen": -29.647293090820312,
+      "logps/rejected": -29.056921005249023,
+      "loss": 1.0105,
+      "rewards/accuracies": 0.5,
+      "rewards/chosen": -0.00459072832018137,
+      "rewards/margins": -0.010462107136845589,
+      "rewards/rejected": 0.005871377885341644,
       "step": 20
     },
     {
       "epoch": 0.08,
       "learning_rate": 3.8461538461538463e-07,
-      "logits/chosen": -1.9199495315551758,
-      "logits/rejected": -1.917249321937561,
-      "logps/chosen": -31.421478271484375,
-      "logps/rejected": -33.2115364074707,
-      "loss": 1.0098,
-      "rewards/accuracies": 0.4625000059604645,
-      "rewards/chosen": -0.004308355040848255,
-      "rewards/margins": -0.009847955778241158,
-      "rewards/rejected": 0.005539602134376764,
+      "logits/chosen": -1.920013189315796,
+      "logits/rejected": -1.917337417602539,
+      "logps/chosen": -31.4055233001709,
+      "logps/rejected": -33.22685623168945,
+      "loss": 0.9829,
+      "rewards/accuracies": 0.5625,
+      "rewards/chosen": 0.009516814723610878,
+      "rewards/margins": 0.01707320474088192,
+      "rewards/rejected": -0.007556394673883915,
       "step": 30
     },
     {
       "epoch": 0.1,
       "learning_rate": 4.999896948438433e-07,
-      "logits/chosen": -2.0169284343719482,
-      "logits/rejected": -2.008178949356079,
-      "logps/chosen": -32.59435272216797,
-      "logps/rejected": -32.49193572998047,
-      "loss": 1.0306,
-      "rewards/accuracies": 0.44999998807907104,
-      "rewards/chosen": -0.014034454710781574,
-      "rewards/margins": -0.030623534694314003,
-      "rewards/rejected": 0.016589080914855003,
+      "logits/chosen": -2.0167124271392822,
+      "logits/rejected": -2.0079712867736816,
+      "logps/chosen": -32.590545654296875,
+      "logps/rejected": -32.499671936035156,
+      "loss": 1.0241,
+      "rewards/accuracies": 0.4375,
+      "rewards/chosen": -0.01236567273736,
+      "rewards/margins": -0.024063030257821083,
+      "rewards/rejected": 0.011697360314428806,
       "step": 40
     },
     {
       "epoch": 0.13,
       "learning_rate": 4.987541037542186e-07,
-      "logits/chosen": -1.86457097530365,
-      "logits/rejected": -1.8537908792495728,
-      "logps/chosen": -33.56566619873047,
-      "logps/rejected": -35.423240661621094,
-      "loss": 1.0298,
+      "logits/chosen": -1.8641105890274048,
+      "logits/rejected": -1.8533456325531006,
+      "logps/chosen": -33.56208038330078,
+      "logps/rejected": -35.437400817871094,
+      "loss": 1.0176,
       "rewards/accuracies": 0.4375,
-      "rewards/chosen": -0.006550998892635107,
-      "rewards/margins": -0.029834842309355736,
-      "rewards/rejected": 0.02328384295105934,
+      "rewards/chosen": -0.004139312542974949,
+      "rewards/margins": -0.017586929723620415,
+      "rewards/rejected": 0.013447612524032593,
       "step": 50
     },
     {
       "epoch": 0.16,
       "learning_rate": 4.954691471941118e-07,
-      "logits/chosen": -1.9449050426483154,
-      "logits/rejected": -1.9468472003936768,
-      "logps/chosen": -32.59955596923828,
-      "logps/rejected": -33.1828498840332,
-      "loss": 0.9903,
-      "rewards/accuracies": 0.512499988079071,
-      "rewards/chosen": 0.0005022765835747123,
-      "rewards/margins": 0.009675229899585247,
-      "rewards/rejected": -0.00917295552790165,
+      "logits/chosen": -1.9452192783355713,
+      "logits/rejected": -1.9471514225006104,
+      "logps/chosen": -32.587867736816406,
+      "logps/rejected": -33.20810317993164,
+      "loss": 0.9559,
+      "rewards/accuracies": 0.5874999761581421,
+      "rewards/chosen": 0.01108462829142809,
+      "rewards/margins": 0.04413585737347603,
+      "rewards/rejected": -0.033051230013370514,
       "step": 60
     },
     {
       "epoch": 0.18,
       "learning_rate": 4.901618883413548e-07,
-      "logits/chosen": -2.079878330230713,
-      "logits/rejected": -2.084862232208252,
-      "logps/chosen": -33.98878860473633,
-      "logps/rejected": -36.574462890625,
-      "loss": 0.9958,
-      "rewards/accuracies": 0.512499988079071,
-      "rewards/chosen": -0.00024233898147940636,
-      "rewards/margins": 0.004231657367199659,
-      "rewards/rejected": -0.004473996348679066,
+      "logits/chosen": -2.0798697471618652,
+      "logits/rejected": -2.084836483001709,
+      "logps/chosen": -33.99953079223633,
+      "logps/rejected": -36.59130859375,
+      "loss": 0.9897,
+      "rewards/accuracies": 0.4749999940395355,
+      "rewards/chosen": -0.009940117597579956,
+      "rewards/margins": 0.010251833125948906,
+      "rewards/rejected": -0.020191945135593414,
       "step": 70
     },
     {
       "epoch": 0.21,
       "learning_rate": 4.828760511501322e-07,
-      "logits/chosen": -1.9425569772720337,
-      "logits/rejected": -1.9457191228866577,
-      "logps/chosen": -34.40068054199219,
-      "logps/rejected": -34.5762939453125,
-      "loss": 0.9912,
-      "rewards/accuracies": 0.5249999761581421,
-      "rewards/chosen": 0.021701809018850327,
-      "rewards/margins": 0.00881027989089489,
-      "rewards/rejected": 0.012891530990600586,
+      "logits/chosen": -1.9422235488891602,
+      "logits/rejected": -1.9453849792480469,
+      "logps/chosen": -34.411861419677734,
+      "logps/rejected": -34.599830627441406,
+      "loss": 0.979,
+      "rewards/accuracies": 0.5625,
+      "rewards/chosen": 0.014353242702782154,
+      "rewards/margins": 0.02102992869913578,
+      "rewards/rejected": -0.006676685996353626,
       "step": 80
     },
     {
       "epoch": 0.23,
       "learning_rate": 4.736716601303429e-07,
-      "logits/chosen": -1.9507396221160889,
-      "logits/rejected": -1.9552500247955322,
-      "logps/chosen": -32.460357666015625,
-      "logps/rejected": -32.354434967041016,
-      "loss": 0.9979,
-      "rewards/accuracies": 0.512499988079071,
-      "rewards/chosen": 0.005654479376971722,
-      "rewards/margins": 0.002074276329949498,
-      "rewards/rejected": 0.0035802037455141544,
+      "logits/chosen": -1.950791358947754,
+      "logits/rejected": -1.9553003311157227,
+      "logps/chosen": -32.45661163330078,
+      "logps/rejected": -32.35947799682617,
+      "loss": 0.9898,
+      "rewards/accuracies": 0.5375000238418579,
+      "rewards/chosen": 0.009730497375130653,
+      "rewards/margins": 0.010243801400065422,
+      "rewards/rejected": -0.0005133040249347687,
       "step": 90
     },
     {
       "epoch": 0.26,
       "learning_rate": 4.62624545834521e-07,
-      "logits/chosen": -2.0492873191833496,
-      "logits/rejected": -2.0472888946533203,
-      "logps/chosen": -32.23810958862305,
-      "logps/rejected": -31.260278701782227,
-      "loss": 1.0026,
-      "rewards/accuracies": 0.4625000059604645,
-      "rewards/chosen": 0.0021828364115208387,
-      "rewards/margins": -0.002563956892117858,
-      "rewards/rejected": 0.004746791906654835,
+      "logits/chosen": -2.0489749908447266,
+      "logits/rejected": -2.046968460083008,
+      "logps/chosen": -32.233360290527344,
+      "logps/rejected": -31.278451919555664,
+      "loss": 0.9823,
+      "rewards/accuracies": 0.512499988079071,
+      "rewards/chosen": 0.006729811429977417,
+      "rewards/margins": 0.01774778962135315,
+      "rewards/rejected": -0.011017980054020882,
       "step": 100
     },
     {
       "epoch": 0.26,
-      "eval_logits/chosen": -2.243312358856201,
-      "eval_logits/rejected": -2.238436222076416,
-      "eval_logps/chosen": -34.023216247558594,
-      "eval_logps/rejected": -37.49723434448242,
-      "eval_loss": 1.0065712928771973,
-      "eval_rewards/accuracies": 0.49833887815475464,
-      "eval_rewards/chosen": 0.009070919826626778,
-      "eval_rewards/margins": -0.006438401062041521,
-      "eval_rewards/rejected": 0.015509321354329586,
-      "eval_runtime": 146.1739,
-      "eval_samples_per_second": 2.347,
+      "eval_logits/chosen": -2.243974447250366,
+      "eval_logits/rejected": -2.239091634750366,
+      "eval_logps/chosen": -34.021610260009766,
+      "eval_logps/rejected": -37.50578689575195,
+      "eval_loss": 0.9981797933578491,
+      "eval_rewards/accuracies": 0.5103820562362671,
+      "eval_rewards/chosen": 0.011647163890302181,
+      "eval_rewards/margins": 0.0018998426385223866,
+      "eval_rewards/rejected": 0.009747318923473358,
+      "eval_runtime": 146.0735,
+      "eval_samples_per_second": 2.348,
       "eval_steps_per_second": 0.294,
       "step": 100
     },
     {
       "epoch": 0.29,
       "learning_rate": 4.4982572012636904e-07,
-      "logits/chosen": -2.005169153213501,
-      "logits/rejected": -2.0027499198913574,
-      "logps/chosen": -33.2365837097168,
-      "logps/rejected": -34.01953125,
-      "loss": 1.0021,
-      "rewards/accuracies": 0.4625000059604645,
-      "rewards/chosen": 0.005784005858004093,
-      "rewards/margins": -0.002089007291942835,
-      "rewards/rejected": 0.007873011752963066,
+      "logits/chosen": -2.0054612159729004,
+      "logits/rejected": -2.003042221069336,
+      "logps/chosen": -33.2550048828125,
+      "logps/rejected": -34.02202224731445,
+      "loss": 1.0167,
+      "rewards/accuracies": 0.44999998807907104,
+      "rewards/chosen": -0.010068392381072044,
+      "rewards/margins": -0.016683781519532204,
+      "rewards/rejected": 0.006615391466766596,
       "step": 110
     },
     {
       "epoch": 0.31,
       "learning_rate": 4.353806263777677e-07,
-      "logits/chosen": -2.0166728496551514,
-      "logits/rejected": -2.0083022117614746,
-      "logps/chosen": -32.457847595214844,
-      "logps/rejected": -32.18357467651367,
-      "loss": 0.9927,
-      "rewards/accuracies": 0.550000011920929,
-      "rewards/chosen": -0.010353903286159039,
-      "rewards/margins": 0.0072973729111254215,
-      "rewards/rejected": -0.017651278525590897,
+      "logits/chosen": -2.016511917114258,
+      "logits/rejected": -2.008146047592163,
+      "logps/chosen": -32.45395278930664,
+      "logps/rejected": -32.17681121826172,
+      "loss": 0.9944,
+      "rewards/accuracies": 0.5249999761581421,
+      "rewards/chosen": -0.008141009137034416,
+      "rewards/margins": 0.00563241308555007,
+      "rewards/rejected": -0.013773423619568348,
       "step": 120
     },
     {
       "epoch": 0.34,
       "learning_rate": 4.194082707715275e-07,
-      "logits/chosen": -2.0462448596954346,
-      "logits/rejected": -2.038203001022339,
-      "logps/chosen": -30.475027084350586,
-      "logps/rejected": -32.046302795410156,
-      "loss": 1.0048,
-      "rewards/accuracies": 0.48750001192092896,
-      "rewards/chosen": 0.0005506269517354667,
-      "rewards/margins": -0.004812211729586124,
-      "rewards/rejected": 0.005362839438021183,
+      "logits/chosen": -2.0463106632232666,
+      "logits/rejected": -2.0382590293884277,
+      "logps/chosen": -30.476207733154297,
+      "logps/rejected": -32.020484924316406,
+      "loss": 1.0297,
+      "rewards/accuracies": 0.3499999940395355,
+      "rewards/chosen": -0.00044600124238058925,
+      "rewards/margins": -0.029714178293943405,
+      "rewards/rejected": 0.02926817536354065,
       "step": 130
     },
     {
       "epoch": 0.36,
       "learning_rate": 4.020402418666621e-07,
-      "logits/chosen": -1.976986289024353,
-      "logits/rejected": -1.9872560501098633,
-      "logps/chosen": -31.407278060913086,
-      "logps/rejected": -32.543296813964844,
-      "loss": 0.9834,
-      "rewards/accuracies": 0.48750001192092896,
-      "rewards/chosen": 0.009907384403049946,
-      "rewards/margins": 0.016641682013869286,
-      "rewards/rejected": -0.006734299007803202,
+      "logits/chosen": -1.9767357110977173,
+      "logits/rejected": -1.986999273300171,
+      "logps/chosen": -31.381155014038086,
+      "logps/rejected": -32.54108428955078,
+      "loss": 0.9598,
+      "rewards/accuracies": 0.5375000238418579,
+      "rewards/chosen": 0.034655503928661346,
+      "rewards/margins": 0.04024083539843559,
+      "rewards/rejected": -0.005585335195064545,
       "step": 140
     },
     {
       "epoch": 0.39,
       "learning_rate": 3.8341962650351185e-07,
-      "logits/chosen": -1.8905508518218994,
-      "logits/rejected": -1.891632080078125,
-      "logps/chosen": -34.20501708984375,
-      "logps/rejected": -34.77235412597656,
-      "loss": 0.9978,
-      "rewards/accuracies": 0.4625000059604645,
-      "rewards/chosen": -0.009861050173640251,
-      "rewards/margins": 0.0022162157110869884,
-      "rewards/rejected": -0.012077265419065952,
+      "logits/chosen": -1.8906068801879883,
+      "logits/rejected": -1.8916860818862915,
+      "logps/chosen": -34.20075225830078,
+      "logps/rejected": -34.7618522644043,
+      "loss": 1.0031,
+      "rewards/accuracies": 0.4124999940395355,
+      "rewards/chosen": -0.007252261973917484,
+      "rewards/margins": -0.0031175222247838974,
+      "rewards/rejected": -0.004134741611778736,
       "step": 150
     },
     {
       "epoch": 0.42,
       "learning_rate": 3.636998309800572e-07,
-      "logits/chosen": -1.94281005859375,
-      "logits/rejected": -1.939327597618103,
-      "logps/chosen": -36.144107818603516,
-      "logps/rejected": -32.72822570800781,
-      "loss": 0.9736,
-      "rewards/accuracies": 0.5375000238418579,
-      "rewards/chosen": 0.02205902896821499,
-      "rewards/margins": 0.026437148451805115,
-      "rewards/rejected": -0.004378115758299828,
+      "logits/chosen": -1.942251443862915,
+      "logits/rejected": -1.9387743473052979,
+      "logps/chosen": -36.145790100097656,
+      "logps/rejected": -32.73196029663086,
+      "loss": 0.9684,
+      "rewards/accuracies": 0.512499988079071,
+      "rewards/chosen": 0.023301539942622185,
+      "rewards/margins": 0.031586963683366776,
+      "rewards/rejected": -0.008285422809422016,
       "step": 160
     },
     {
       "epoch": 0.44,
       "learning_rate": 3.430433172111807e-07,
-      "logits/chosen": -2.0420708656311035,
-      "logits/rejected": -2.0346803665161133,
-      "logps/chosen": -33.771934509277344,
-      "logps/rejected": -31.371145248413086,
-      "loss": 0.9638,
-      "rewards/accuracies": 0.5375000238418579,
-      "rewards/chosen": 0.020745161920785904,
-      "rewards/margins": 0.036180444061756134,
-      "rewards/rejected": -0.015435280278325081,
+      "logits/chosen": -2.041952610015869,
+      "logits/rejected": -2.0345559120178223,
+      "logps/chosen": -33.81856155395508,
+      "logps/rejected": -31.359323501586914,
+      "loss": 1.0119,
+      "rewards/accuracies": 0.4749999940395355,
+      "rewards/chosen": -0.018632039427757263,
+      "rewards/margins": -0.011905943043529987,
+      "rewards/rejected": -0.006726093590259552,
       "step": 170
     },
     {
       "epoch": 0.47,
       "learning_rate": 3.216202642830543e-07,
-      "logits/chosen": -2.0475738048553467,
-      "logits/rejected": -2.0528526306152344,
-      "logps/chosen": -32.524593353271484,
-      "logps/rejected": -32.510643005371094,
-      "loss": 0.9634,
-      "rewards/accuracies": 0.550000011920929,
-      "rewards/chosen": 0.016154326498508453,
-      "rewards/margins": 0.036555372178554535,
-      "rewards/rejected": -0.020401041954755783,
+      "logits/chosen": -2.0473170280456543,
+      "logits/rejected": -2.0525729656219482,
+      "logps/chosen": -32.52764129638672,
+      "logps/rejected": -32.50192642211914,
+      "loss": 0.9695,
+      "rewards/accuracies": 0.574999988079071,
+      "rewards/chosen": 0.015429094433784485,
+      "rewards/margins": 0.030534446239471436,
+      "rewards/rejected": -0.015105349011719227,
       "step": 180
     },
     {
       "epoch": 0.49,
       "learning_rate": 2.9960716642946403e-07,
-      "logits/chosen": -2.048490524291992,
-      "logits/rejected": -2.0457024574279785,
-      "logps/chosen": -31.492746353149414,
-      "logps/rejected": -31.319293975830078,
-      "loss": 1.0022,
-      "rewards/accuracies": 0.5,
-      "rewards/chosen": -0.003033895045518875,
-      "rewards/margins": -0.002170243766158819,
-      "rewards/rejected": -0.0008636537822894752,
+      "logits/chosen": -2.048530340194702,
+      "logits/rejected": -2.045767307281494,
+      "logps/chosen": -31.485219955444336,
+      "logps/rejected": -31.34262466430664,
+      "loss": 0.9747,
+      "rewards/accuracies": 0.5874999761581421,
+      "rewards/chosen": 0.003360496135428548,
+      "rewards/margins": 0.02533094584941864,
+      "rewards/rejected": -0.02197045087814331,
       "step": 190
     },
     {
       "epoch": 0.52,
       "learning_rate": 2.771853789806683e-07,
-      "logits/chosen": -1.9185683727264404,
-      "logits/rejected": -1.9232347011566162,
-      "logps/chosen": -31.5926513671875,
-      "logps/rejected": -32.78697204589844,
-      "loss": 0.9923,
-      "rewards/accuracies": 0.5874999761581421,
-      "rewards/chosen": 0.01090637780725956,
-      "rewards/margins": 0.007740379776805639,
-      "rewards/rejected": 0.0031659987289458513,
+      "logits/chosen": -1.918745756149292,
+      "logits/rejected": -1.9234097003936768,
+      "logps/chosen": -31.576425552368164,
+      "logps/rejected": -32.79480743408203,
+      "loss": 0.9696,
+      "rewards/accuracies": 0.5375000238418579,
+      "rewards/chosen": 0.026870254427194595,
+      "rewards/margins": 0.03035963699221611,
+      "rewards/rejected": -0.0034893869888037443,
       "step": 200
     },
     {
       "epoch": 0.52,
-      "eval_logits/chosen": -2.243807315826416,
-      "eval_logits/rejected": -2.2389235496520996,
-      "eval_logps/chosen": -34.018714904785156,
-      "eval_logps/rejected": -37.517478942871094,
-      "eval_loss": 0.9864435195922852,
-      "eval_rewards/accuracies": 0.5328072905540466,
-      "eval_rewards/chosen": 0.012671924196183681,
-      "eval_rewards/margins": 0.013361074030399323,
-      "eval_rewards/rejected": -0.0006891504744999111,
-      "eval_runtime": 145.9847,
-      "eval_samples_per_second": 2.35,
+      "eval_logits/chosen": -2.243912696838379,
+      "eval_logits/rejected": -2.2390317916870117,
+      "eval_logps/chosen": -34.021690368652344,
+      "eval_logps/rejected": -37.502418518066406,
+      "eval_loss": 1.001371145248413,
+      "eval_rewards/accuracies": 0.4692690968513489,
+      "eval_rewards/chosen": 0.011576101183891296,
+      "eval_rewards/margins": -0.0012021720176562667,
+      "eval_rewards/rejected": 0.012778270989656448,
+      "eval_runtime": 145.8493,
+      "eval_samples_per_second": 2.352,
       "eval_steps_per_second": 0.295,
       "step": 200
     },
     {
       "epoch": 0.55,
       "learning_rate": 2.5453962426402e-07,
-      "logits/chosen": -2.0318503379821777,
-      "logits/rejected": -2.042539596557617,
-      "logps/chosen": -31.948400497436523,
-      "logps/rejected": -33.86983871459961,
-      "loss": 0.976,
-      "rewards/accuracies": 0.5,
-      "rewards/chosen": 0.018323851749300957,
-      "rewards/margins": 0.024035325273871422,
-      "rewards/rejected": -0.005711475852876902,
+      "logits/chosen": -2.0318515300750732,
+      "logits/rejected": -2.0425400733947754,
+      "logps/chosen": -31.96026039123535,
+      "logps/rejected": -33.8691291809082,
+      "loss": 0.9843,
+      "rewards/accuracies": 0.625,
+      "rewards/chosen": 0.009936656802892685,
+      "rewards/margins": 0.015721982344985008,
+      "rewards/rejected": -0.005785329267382622,
       "step": 210
     },
     {
       "epoch": 0.57,
       "learning_rate": 2.318564697655179e-07,
-      "logits/chosen": -1.9251388311386108,
-      "logits/rejected": -1.9399843215942383,
-      "logps/chosen": -30.099853515625,
-      "logps/rejected": -31.55409812927246,
-      "loss": 0.9939,
-      "rewards/accuracies": 0.5249999761581421,
-      "rewards/chosen": 0.009348717518150806,
-      "rewards/margins": 0.006087350193411112,
-      "rewards/rejected": 0.003261365694925189,
+      "logits/chosen": -1.9256188869476318,
+      "logits/rejected": -1.9404585361480713,
+      "logps/chosen": -30.080001831054688,
+      "logps/rejected": -31.53633689880371,
+      "loss": 0.9913,
+      "rewards/accuracies": 0.512499988079071,
+      "rewards/chosen": 0.028384273871779442,
+      "rewards/margins": 0.00873211957514286,
+      "rewards/rejected": 0.01965215802192688,
       "step": 220
     },
     {
       "epoch": 0.6,
       "learning_rate": 2.093227910899832e-07,
-      "logits/chosen": -1.9835479259490967,
-      "logits/rejected": -1.9875112771987915,
-      "logps/chosen": -33.387638092041016,
-      "logps/rejected": -31.554845809936523,
-      "loss": 0.9717,
-      "rewards/accuracies": 0.550000011920929,
-      "rewards/chosen": 0.023866940289735794,
-      "rewards/margins": 0.028299665078520775,
-      "rewards/rejected": -0.00443272665143013,
+      "logits/chosen": -1.9830671548843384,
+      "logits/rejected": -1.9870269298553467,
+      "logps/chosen": -33.386474609375,
+      "logps/rejected": -31.55719566345215,
+      "loss": 0.965,
+      "rewards/accuracies": 0.5249999761581421,
+      "rewards/chosen": 0.02789575420320034,
+      "rewards/margins": 0.03499894589185715,
+      "rewards/rejected": -0.007103185169398785,
       "step": 230
     },
     {
       "epoch": 0.62,
       "learning_rate": 1.8712423238279356e-07,
-      "logits/chosen": -1.9826898574829102,
-      "logits/rejected": -1.9607274532318115,
-      "logps/chosen": -34.158443450927734,
-      "logps/rejected": -34.963783264160156,
-      "loss": 0.9934,
-      "rewards/accuracies": 0.512499988079071,
-      "rewards/chosen": -0.0037571662105619907,
-      "rewards/margins": 0.0066327848471701145,
-      "rewards/rejected": -0.01038995012640953,
+      "logits/chosen": -1.9833042621612549,
+      "logits/rejected": -1.96132493019104,
+      "logps/chosen": -34.13536834716797,
+      "logps/rejected": -34.952049255371094,
+      "loss": 0.9823,
+      "rewards/accuracies": 0.5375000238418579,
+      "rewards/chosen": 0.016542484983801842,
+      "rewards/margins": 0.01767394319176674,
+      "rewards/rejected": -0.0011314575094729662,
       "step": 240
     },
     {
       "epoch": 0.65,
       "learning_rate": 1.654436768970182e-07,
-      "logits/chosen": -2.024381160736084,
-      "logits/rejected": -2.0210862159729004,
-      "logps/chosen": -32.9254035949707,
-      "logps/rejected": -36.251712799072266,
-      "loss": 0.9788,
-      "rewards/accuracies": 0.5375000238418579,
-      "rewards/chosen": -0.0002662956831045449,
-      "rewards/margins": 0.02116318605840206,
-      "rewards/rejected": -0.021429482847452164,
+      "logits/chosen": -2.0247886180877686,
+      "logits/rejected": -2.0214812755584717,
+      "logps/chosen": -32.913856506347656,
+      "logps/rejected": -36.23151397705078,
+      "loss": 0.984,
+      "rewards/accuracies": 0.5625,
+      "rewards/chosen": 0.010092089883983135,
+      "rewards/margins": 0.016022875905036926,
+      "rewards/rejected": -0.0059307836927473545,
       "step": 250
     },
     {
       "epoch": 0.68,
       "learning_rate": 1.444597403062196e-07,
-      "logits/chosen": -1.8911311626434326,
-      "logits/rejected": -1.8886839151382446,
-      "logps/chosen": -34.194557189941406,
-      "logps/rejected": -35.51445770263672,
-      "loss": 0.9971,
-      "rewards/accuracies": 0.48750001192092896,
-      "rewards/chosen": -0.004926486872136593,
-      "rewards/margins": 0.002939151134341955,
-      "rewards/rejected": -0.007865638472139835,
+      "logits/chosen": -1.8914467096328735,
+      "logits/rejected": -1.8890043497085571,
+      "logps/chosen": -34.18690872192383,
+      "logps/rejected": -35.524742126464844,
+      "loss": 0.9806,
+      "rewards/accuracies": 0.5375000238418579,
+      "rewards/chosen": 0.0013398710871115327,
+      "rewards/margins": 0.019445054233074188,
+      "rewards/rejected": -0.01810518093407154,
       "step": 260
     },
     {
       "epoch": 0.7,
       "learning_rate": 1.2434529917578887e-07,
-      "logits/chosen": -1.8759450912475586,
-      "logits/rejected": -1.8734045028686523,
-      "logps/chosen": -34.40558624267578,
-      "logps/rejected": -31.752349853515625,
-      "loss": 1.0216,
-      "rewards/accuracies": 0.4625000059604645,
-      "rewards/chosen": -0.0063878437504172325,
-      "rewards/margins": -0.02157408930361271,
-      "rewards/rejected": 0.015186244621872902,
+      "logits/chosen": -1.8761389255523682,
+      "logits/rejected": -1.8735923767089844,
+      "logps/chosen": -34.396785736083984,
+      "logps/rejected": -31.74139976501465,
+      "loss": 1.0262,
+      "rewards/accuracies": 0.4000000059604645,
+      "rewards/chosen": 0.000734941684640944,
+      "rewards/margins": -0.026202131062746048,
+      "rewards/rejected": 0.02693706750869751,
       "step": 270
     },
     {
       "epoch": 0.73,
       "learning_rate": 1.0526606671603521e-07,
-      "logits/chosen": -1.980015754699707,
-      "logits/rejected": -1.9693737030029297,
-      "logps/chosen": -35.33230209350586,
-      "logps/rejected": -31.845691680908203,
-      "loss": 0.9768,
-      "rewards/accuracies": 0.512499988079071,
-      "rewards/chosen": 0.019471382722258568,
-      "rewards/margins": 0.023230455815792084,
-      "rewards/rejected": -0.0037590693682432175,
+      "logits/chosen": -1.9801757335662842,
+      "logits/rejected": -1.9695546627044678,
+      "logps/chosen": -35.34246063232422,
+      "logps/rejected": -31.823230743408203,
+      "loss": 1.0032,
+      "rewards/accuracies": 0.4749999940395355,
+      "rewards/chosen": 0.012763315811753273,
+      "rewards/margins": -0.003222113475203514,
+      "rewards/rejected": 0.015985429286956787,
       "step": 280
     },
     {
       "epoch": 0.75,
       "learning_rate": 8.737922755071453e-08,
-      "logits/chosen": -2.0756278038024902,
-      "logits/rejected": -2.060606002807617,
-      "logps/chosen": -30.907390594482422,
-      "logps/rejected": -32.64055252075195,
-      "loss": 0.9956,
-      "rewards/accuracies": 0.550000011920929,
-      "rewards/chosen": 0.01338677667081356,
-      "rewards/margins": 0.0043600378558039665,
-      "rewards/rejected": 0.00902673788368702,
+      "logits/chosen": -2.075316905975342,
+      "logits/rejected": -2.0602917671203613,
+      "logps/chosen": -30.897375106811523,
+      "logps/rejected": -32.64093780517578,
+      "loss": 0.9857,
+      "rewards/accuracies": 0.5249999761581421,
+      "rewards/chosen": 0.024072837084531784,
+      "rewards/margins": 0.014264687895774841,
+      "rewards/rejected": 0.009808152914047241,
       "step": 290
     },
     {
       "epoch": 0.78,
       "learning_rate": 7.08321427484816e-08,
-      "logits/chosen": -1.946616768836975,
-      "logits/rejected": -1.9440828561782837,
-      "logps/chosen": -32.894561767578125,
-      "logps/rejected": -30.812387466430664,
-      "loss": 0.9775,
-      "rewards/accuracies": 0.5,
-      "rewards/chosen": 0.026041675359010696,
-      "rewards/margins": 0.022502180188894272,
-      "rewards/rejected": 0.003539492143318057,
+      "logits/chosen": -1.946758508682251,
+      "logits/rejected": -1.9442226886749268,
+      "logps/chosen": -32.89502716064453,
+      "logps/rejected": -30.831274032592773,
+      "loss": 0.9581,
+      "rewards/accuracies": 0.5625,
+      "rewards/chosen": 0.0288736280053854,
+      "rewards/margins": 0.04189068451523781,
+      "rewards/rejected": -0.013017055578529835,
       "step": 300
     },
     {
       "epoch": 0.78,
-      "eval_logits/chosen": -2.2434821128845215,
-      "eval_logits/rejected": -2.2386035919189453,
-      "eval_logps/chosen": -34.017669677734375,
-      "eval_logps/rejected": -37.50018310546875,
-      "eval_loss": 0.9997690916061401,
-      "eval_rewards/accuracies": 0.490448534488678,
-      "eval_rewards/chosen": 0.013507848605513573,
-      "eval_rewards/margins": 0.000357679498847574,
-      "eval_rewards/rejected": 0.0131501704454422,
-      "eval_runtime": 145.7344,
-      "eval_samples_per_second": 2.354,
+      "eval_logits/chosen": -2.2437047958374023,
+      "eval_logits/rejected": -2.2388339042663574,
+      "eval_logps/chosen": -34.03062438964844,
+      "eval_logps/rejected": -37.499412536621094,
+      "eval_loss": 1.0120128393173218,
+      "eval_rewards/accuracies": 0.472591370344162,
+      "eval_rewards/chosen": 0.003532935632392764,
+      "eval_rewards/margins": -0.011953731998801231,
+      "eval_rewards/rejected": 0.015486669726669788,
+      "eval_runtime": 145.7744,
+      "eval_samples_per_second": 2.353,
       "eval_steps_per_second": 0.295,
       "step": 300
     },
     {
       "epoch": 0.81,
       "learning_rate": 5.576113578589034e-08,
-      "logits/chosen": -1.9287067651748657,
-      "logits/rejected": -1.925451636314392,
-      "logps/chosen": -31.603496551513672,
-      "logps/rejected": -33.734046936035156,
-      "loss": 1.0066,
-      "rewards/accuracies": 0.4749999940395355,
-      "rewards/chosen": -0.00011934786743950099,
-      "rewards/margins": -0.006604696158319712,
-      "rewards/rejected": 0.006485348101705313,
+      "logits/chosen": -1.928391695022583,
+      "logits/rejected": -1.925132393836975,
+      "logps/chosen": -31.573719024658203,
+      "logps/rejected": -33.733333587646484,
+      "loss": 0.9813,
+      "rewards/accuracies": 0.574999988079071,
+      "rewards/chosen": 0.026669049635529518,
+      "rewards/margins": 0.01873381994664669,
+      "rewards/rejected": 0.007935228757560253,
       "step": 310
     },
     {
       "epoch": 0.83,
       "learning_rate": 4.229036944380912e-08,
-      "logits/chosen": -1.9802377223968506,
-      "logits/rejected": -1.9679291248321533,
-      "logps/chosen": -34.585323333740234,
-      "logps/rejected": -33.57084274291992,
-      "loss": 0.9658,
-      "rewards/accuracies": 0.5375000238418579,
-      "rewards/chosen": 0.00860034953802824,
-      "rewards/margins": 0.034203141927719116,
-      "rewards/rejected": -0.02560279332101345,
+      "logits/chosen": -1.9808565378189087,
+      "logits/rejected": -1.9685465097427368,
+      "logps/chosen": -34.569026947021484,
+      "logps/rejected": -33.5385856628418,
+      "loss": 0.9759,
+      "rewards/accuracies": 0.5874999761581421,
+      "rewards/chosen": 0.02433854714035988,
+      "rewards/margins": 0.024108227342367172,
+      "rewards/rejected": 0.00023032091849017888,
       "step": 320
     },
     {
       "epoch": 0.86,
       "learning_rate": 3.053082288996112e-08,
-      "logits/chosen": -2.015906810760498,
-      "logits/rejected": -2.014427661895752,
-      "logps/chosen": -33.49116516113281,
-      "logps/rejected": -32.47978973388672,
-      "loss": 1.0041,
-      "rewards/accuracies": 0.4375,
-      "rewards/chosen": -0.0042965407483279705,
-      "rewards/margins": -0.004129170440137386,
-      "rewards/rejected": -0.00016737030819058418,
+      "logits/chosen": -2.0160164833068848,
+      "logits/rejected": -2.0145697593688965,
+      "logps/chosen": -33.470794677734375,
+      "logps/rejected": -32.465301513671875,
+      "loss": 0.9994,
+      "rewards/accuracies": 0.5,
+      "rewards/chosen": 0.013503548689186573,
+      "rewards/margins": 0.0006495133275166154,
+      "rewards/rejected": 0.012854036875069141,
       "step": 330
     },
     {
       "epoch": 0.88,
       "learning_rate": 2.05793773749158e-08,
-      "logits/chosen": -2.1030631065368652,
-      "logits/rejected": -2.0872654914855957,
-      "logps/chosen": -34.18492889404297,
-      "logps/rejected": -33.08319854736328,
-      "loss": 1.0323,
-      "rewards/accuracies": 0.42500001192092896,
-      "rewards/chosen": -0.00442854966968298,
-      "rewards/margins": -0.03229869529604912,
-      "rewards/rejected": 0.027870142832398415,
+      "logits/chosen": -2.1030516624450684,
+      "logits/rejected": -2.087252140045166,
+      "logps/chosen": -34.154754638671875,
+      "logps/rejected": -33.087013244628906,
+      "loss": 1.0057,
+      "rewards/accuracies": 0.512499988079071,
+      "rewards/chosen": 0.02217467688024044,
+      "rewards/margins": -0.005742859095335007,
+      "rewards/rejected": 0.027917543426156044,
       "step": 340
     },
     {
       "epoch": 0.91,
       "learning_rate": 1.251801807404168e-08,
-      "logits/chosen": -1.9745471477508545,
-      "logits/rejected": -1.9736032485961914,
-      "logps/chosen": -33.23271942138672,
-      "logps/rejected": -32.4765510559082,
-      "loss": 0.9514,
-      "rewards/accuracies": 0.612500011920929,
-      "rewards/chosen": 0.0395023450255394,
-      "rewards/margins": 0.04862459748983383,
-      "rewards/rejected": -0.009122253395617008,
+      "logits/chosen": -1.9748103618621826,
+      "logits/rejected": -1.9738750457763672,
+      "logps/chosen": -33.23518371582031,
+      "logps/rejected": -32.46066665649414,
+      "loss": 0.9618,
+      "rewards/accuracies": 0.574999988079071,
+      "rewards/chosen": 0.04222174361348152,
+      "rewards/margins": 0.03818847984075546,
+      "rewards/rejected": 0.004033264704048634,
       "step": 350
     },
     {
       "epoch": 0.94,
       "learning_rate": 6.41315865106129e-09,
-      "logits/chosen": -1.9305438995361328,
-      "logits/rejected": -1.940913438796997,
-      "logps/chosen": -32.22040939331055,
-      "logps/rejected": -35.28728103637695,
-      "loss": 1.0138,
-      "rewards/accuracies": 0.48750001192092896,
-      "rewards/chosen": -0.012576865963637829,
-      "rewards/margins": -0.013758530840277672,
-      "rewards/rejected": 0.0011816651094704866,
+      "logits/chosen": -1.9303611516952515,
+      "logits/rejected": -1.940708875656128,
+      "logps/chosen": -32.200523376464844,
+      "logps/rejected": -35.27861785888672,
+      "loss": 1.0054,
+      "rewards/accuracies": 0.42500001192092896,
+      "rewards/chosen": 0.0037514485884457827,
+      "rewards/margins": -0.005373493768274784,
+      "rewards/rejected": 0.009124943986535072,
       "step": 360
     },
     {
       "epoch": 0.96,
       "learning_rate": 2.3150941078050324e-09,
-      "logits/chosen": -2.069648265838623,
-      "logits/rejected": -2.0630898475646973,
-      "logps/chosen": -33.63695526123047,
-      "logps/rejected": -29.226470947265625,
-      "loss": 0.988,
-      "rewards/accuracies": 0.5625,
-      "rewards/chosen": 0.012009462341666222,
-      "rewards/margins": 0.012046033516526222,
-      "rewards/rejected": -3.6573597753886133e-05,
+      "logits/chosen": -2.069342613220215,
+      "logits/rejected": -2.062788486480713,
+      "logps/chosen": -33.65803527832031,
+      "logps/rejected": -29.199758529663086,
+      "loss": 1.0295,
+      "rewards/accuracies": 0.5,
+      "rewards/chosen": -0.005459528882056475,
+      "rewards/margins": -0.029461050406098366,
+      "rewards/rejected": 0.02400151826441288,
       "step": 370
     },
     {
       "epoch": 0.99,
       "learning_rate": 2.575864278703266e-10,
-      "logits/chosen": -1.928865671157837,
-      "logits/rejected": -1.9310123920440674,
-      "logps/chosen": -34.243560791015625,
-      "logps/rejected": -30.892742156982422,
-      "loss": 0.9875,
-      "rewards/accuracies": 0.5249999761581421,
-      "rewards/chosen": -0.003606534795835614,
-      "rewards/margins": 0.012486499734222889,
-      "rewards/rejected": -0.01609303429722786,
+      "logits/chosen": -1.9290788173675537,
+      "logits/rejected": -1.931239366531372,
+      "logps/chosen": -34.229915618896484,
+      "logps/rejected": -30.90093994140625,
+      "loss": 0.9663,
+      "rewards/accuracies": 0.5625,
+      "rewards/chosen": 0.00822196900844574,
+      "rewards/margins": 0.03370029479265213,
+      "rewards/rejected": -0.02547832392156124,
       "step": 380
     },
     {
       "epoch": 1.0,
       "step": 385,
       "total_flos": 0.0,
-      "train_loss": 0.9929689927534624,
-      "train_runtime": 3255.1859,
+      "train_loss": 0.9887851405453373,
+      "train_runtime": 3254.6365,
       "train_samples_per_second": 0.946,
       "train_steps_per_second": 0.118
     }