Model save

README.md

@@ -4,6 +4,7 @@ base_model: alignment-handbook/zephyr-7b-sft-full
 tags:
 - trl
 - dpo
+- alignment-handbook
 - generated_from_trainer
 model-index:
 - name: zephyr-7b-dpo-full-prometheus_consistent-3
@@ -17,15 +18,15 @@ should probably proofread and complete it, then remove this comment. -->
 
 This model is a fine-tuned version of [alignment-handbook/zephyr-7b-sft-full](https://huggingface.co/alignment-handbook/zephyr-7b-sft-full) on the None dataset.
 It achieves the following results on the evaluation set:
-- Loss: 0.5175
-- Rewards/chosen: -0.8697
-- Rewards/rejected: -1.8604
-- Rewards/accuracies: 0.7716
-- Rewards/margins: 0.9907
-- Logps/rejected: -405.1198
-- Logps/chosen: -362.5746
-- Logits/rejected: 1.7998
-- Logits/chosen: 0.2407
+- Loss: 0.5201
+- Rewards/chosen: -0.8222
+- Rewards/rejected: -1.7957
+- Rewards/accuracies: 0.7759
+- Rewards/margins: 0.9734
+- Logps/rejected: -398.6412
+- Logps/chosen: -357.8258
+- Logits/rejected: 1.5724
+- Logits/chosen: -0.0130
 
 ## Model description
 
@@ -62,10 +63,10 @@ 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 |
 |:-------------:|:------:|:----:|:---------------:|:--------------:|:----------------:|:------------------:|:---------------:|:--------------:|:------------:|:---------------:|:-------------:|
-| 0.5982        | 0.2286 | 100  | 0.5954          | -0.1704        | -0.5853          | 0.7155             | 0.4149          | -277.6054      | -292.6412    | -2.2902         | -2.4282       |
-| 0.5439        | 0.4571 | 200  | 0.5353          | -0.8124        | -1.6747          | 0.7586             | 0.8623          | -386.5413      | -356.8392    | -0.0415         | -1.0380       |
-| 0.5109        | 0.6857 | 300  | 0.5190          | -0.6985        | -1.6462          | 0.7716             | 0.9478          | -383.7008      | -345.4529    | 1.1016          | -0.3942       |
-| 0.5457        | 0.9143 | 400  | 0.5175          | -0.8697        | -1.8604          | 0.7716             | 0.9907          | -405.1198      | -362.5746    | 1.7998          | 0.2407        |
+| 0.5995        | 0.2286 | 100  | 0.5960          | -0.3095        | -0.7496          | 0.7155             | 0.4400          | -294.0323      | -306.5585    | -2.3479         | -2.4655       |
+| 0.5431        | 0.4571 | 200  | 0.5326          | -0.6273        | -1.5233          | 0.7629             | 0.8960          | -371.4070      | -338.3311    | -0.0959         | -1.0624       |
+| 0.5138        | 0.6857 | 300  | 0.5242          | -0.5729        | -1.4989          | 0.7629             | 0.9260          | -368.9617      | -332.8911    | 0.6740          | -0.7677       |
+| 0.5493        | 0.9143 | 400  | 0.5201          | -0.8222        | -1.7957          | 0.7759             | 0.9734          | -398.6412      | -357.8258    | 1.5724          | -0.0130       |
 
 
 ### Framework versions

all_results.json

@@ -1,9 +1,22 @@
 {
     "epoch": 0.9988571428571429,
+    "eval_logits/chosen": 0.20247821509838104,
+    "eval_logits/rejected": 1.7624026536941528,
+    "eval_logps/chosen": -361.47540283203125,
+    "eval_logps/rejected": -403.6280212402344,
+    "eval_loss": 0.5172589421272278,
+    "eval_rewards/accuracies": 0.7715517282485962,
+    "eval_rewards/chosen": -0.8587150573730469,
+    "eval_rewards/margins": 0.9868037104606628,
+    "eval_rewards/rejected": -1.8455188274383545,
+    "eval_runtime": 92.8884,
+    "eval_samples": 1831,
+    "eval_samples_per_second": 19.712,
+    "eval_steps_per_second": 0.312,
     "total_flos": 0.0,
-    "train_loss": 0.5586249282758351,
-    "train_runtime": 11006.8714,
+    "train_loss": 0.5608050315822016,
+    "train_runtime": 10950.5403,
     "train_samples": 55999,
-    "train_samples_per_second": 5.088,
+    "train_samples_per_second": 5.114,
     "train_steps_per_second": 0.04
 }

eval_results.json

@@ -0,0 +1,16 @@
+{
+    "epoch": 0.9988571428571429,
+    "eval_logits/chosen": 0.20247821509838104,
+    "eval_logits/rejected": 1.7624026536941528,
+    "eval_logps/chosen": -361.47540283203125,
+    "eval_logps/rejected": -403.6280212402344,
+    "eval_loss": 0.5172589421272278,
+    "eval_rewards/accuracies": 0.7715517282485962,
+    "eval_rewards/chosen": -0.8587150573730469,
+    "eval_rewards/margins": 0.9868037104606628,
+    "eval_rewards/rejected": -1.8455188274383545,
+    "eval_runtime": 92.8884,
+    "eval_samples": 1831,
+    "eval_samples_per_second": 19.712,
+    "eval_steps_per_second": 0.312
+}

model-00001-of-00003.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:ed7e82c33753ebe7317e10065846399acd6ae49db5448e2543b94b38f044271c
+oid sha256:7f66e2ff3080532a30015144afa05dcc1dd216d3cc4828ddf602e91af33d0e56
 size 4943162336

model-00002-of-00003.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:dea90300dc04b256ea0e99b1394e2649e384be3bb13ca1813045d4bd6fe43707
+oid sha256:e404da7afef74b5e46fb84fc567bdbdf32efb0c5d0c3cbdf4d92d3af8e30f5dd
 size 4999819336

model-00003-of-00003.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:74f820255c3b771e3d26ddbcf78fee50ba0d85e82dbf36b349abb19e185bf7e8
+oid sha256:0629d4551cf26be95bc9b0b2a4dfe11d38bcf348d6b510e16cda8a406e7ba608
 size 4540516344

train_results.json

@@ -1,9 +1,9 @@
 {
     "epoch": 0.9988571428571429,
     "total_flos": 0.0,
-    "train_loss": 0.5586249282758351,
-    "train_runtime": 11006.8714,
+    "train_loss": 0.5608050315822016,
+    "train_runtime": 10950.5403,
     "train_samples": 55999,
-    "train_samples_per_second": 5.088,
+    "train_samples_per_second": 5.114,
     "train_steps_per_second": 0.04
 }

trainer_state.json

@@ -10,720 +10,720 @@
   "log_history": [
     {
       "epoch": 0.022857142857142857,
-      "grad_norm": 9.822878263417993,
+      "grad_norm": 9.858836405653433,
       "learning_rate": 1.1363636363636363e-07,
-      "logits/chosen": -2.700991630554199,
-      "logits/rejected": -2.625051975250244,
-      "logps/chosen": -301.26373291015625,
-      "logps/rejected": -281.7487487792969,
+      "logits/chosen": -2.7008285522460938,
+      "logits/rejected": -2.6250243186950684,
+      "logps/chosen": -301.27081298828125,
+      "logps/rejected": -281.75146484375,
       "loss": 0.693,
-      "rewards/accuracies": 0.40625,
-      "rewards/chosen": 0.00022422037727665156,
-      "rewards/margins": 0.0002149858046323061,
-      "rewards/rejected": 9.234552635462023e-06,
+      "rewards/accuracies": 0.41874998807907104,
+      "rewards/chosen": 0.00015341350808739662,
+      "rewards/margins": 0.0001716136175673455,
+      "rewards/rejected": -1.8200071281171404e-05,
       "step": 10
     },
     {
       "epoch": 0.045714285714285714,
-      "grad_norm": 7.790697115653487,
+      "grad_norm": 7.79798162706573,
       "learning_rate": 2.2727272727272726e-07,
-      "logits/chosen": -2.6405527591705322,
-      "logits/rejected": -2.6053428649902344,
-      "logps/chosen": -278.93023681640625,
-      "logps/rejected": -254.70883178710938,
+      "logits/chosen": -2.6407124996185303,
+      "logits/rejected": -2.6055800914764404,
+      "logps/chosen": -278.97711181640625,
+      "logps/rejected": -254.7215576171875,
       "loss": 0.6923,
       "rewards/accuracies": 0.5874999761581421,
-      "rewards/chosen": 0.002409698674455285,
-      "rewards/margins": 0.0018069592770189047,
-      "rewards/rejected": 0.0006027388735674322,
+      "rewards/chosen": 0.0019410619279369712,
+      "rewards/margins": 0.0014655741397291422,
+      "rewards/rejected": 0.0004754880501423031,
       "step": 20
     },
     {
       "epoch": 0.06857142857142857,
-      "grad_norm": 8.37723659427994,
+      "grad_norm": 8.391842432337471,
       "learning_rate": 3.4090909090909085e-07,
-      "logits/chosen": -2.637849807739258,
-      "logits/rejected": -2.616870403289795,
-      "logps/chosen": -263.4436950683594,
-      "logps/rejected": -263.60205078125,
+      "logits/chosen": -2.63759183883667,
+      "logits/rejected": -2.6166491508483887,
+      "logps/chosen": -263.44866943359375,
+      "logps/rejected": -263.5602111816406,
       "loss": 0.6871,
-      "rewards/accuracies": 0.612500011920929,
-      "rewards/chosen": 0.012393913231790066,
-      "rewards/margins": 0.00848361849784851,
-      "rewards/rejected": 0.003910294268280268,
+      "rewards/accuracies": 0.6000000238418579,
+      "rewards/chosen": 0.012344349175691605,
+      "rewards/margins": 0.008015613071620464,
+      "rewards/rejected": 0.004328734241425991,
       "step": 30
     },
     {
       "epoch": 0.09142857142857143,
-      "grad_norm": 9.39434892949384,
+      "grad_norm": 9.39930524490582,
       "learning_rate": 4.545454545454545e-07,
-      "logits/chosen": -2.64766263961792,
-      "logits/rejected": -2.585386276245117,
-      "logps/chosen": -290.52587890625,
-      "logps/rejected": -268.34405517578125,
-      "loss": 0.6756,
-      "rewards/accuracies": 0.5874999761581421,
-      "rewards/chosen": 0.03471367806196213,
-      "rewards/margins": 0.041840970516204834,
-      "rewards/rejected": -0.007127293851226568,
+      "logits/chosen": -2.6476945877075195,
+      "logits/rejected": -2.5853049755096436,
+      "logps/chosen": -290.5145568847656,
+      "logps/rejected": -268.3503723144531,
+      "loss": 0.6755,
+      "rewards/accuracies": 0.606249988079071,
+      "rewards/chosen": 0.034826718270778656,
+      "rewards/margins": 0.042017363011837006,
+      "rewards/rejected": -0.007190642412751913,
       "step": 40
     },
     {
       "epoch": 0.11428571428571428,
-      "grad_norm": 13.922334228887747,
+      "grad_norm": 13.99279327188838,
       "learning_rate": 4.997124959943201e-07,
-      "logits/chosen": -2.676335573196411,
-      "logits/rejected": -2.5971837043762207,
-      "logps/chosen": -294.0897216796875,
-      "logps/rejected": -253.9338836669922,
-      "loss": 0.6628,
+      "logits/chosen": -2.6765246391296387,
+      "logits/rejected": -2.5974183082580566,
+      "logps/chosen": -294.23516845703125,
+      "logps/rejected": -254.03042602539062,
+      "loss": 0.6629,
       "rewards/accuracies": 0.6625000238418579,
-      "rewards/chosen": 0.022205885499715805,
-      "rewards/margins": 0.09290142357349396,
-      "rewards/rejected": -0.07069554179906845,
+      "rewards/chosen": 0.02075173892080784,
+      "rewards/margins": 0.0924127846956253,
+      "rewards/rejected": -0.07166104018688202,
       "step": 50
     },
     {
       "epoch": 0.13714285714285715,
-      "grad_norm": 11.361649612146103,
+      "grad_norm": 10.917441498337828,
       "learning_rate": 4.979579212164186e-07,
-      "logits/chosen": -2.5759811401367188,
-      "logits/rejected": -2.472777843475342,
-      "logps/chosen": -290.1505432128906,
-      "logps/rejected": -270.30242919921875,
-      "loss": 0.6374,
-      "rewards/accuracies": 0.65625,
-      "rewards/chosen": -0.09644584357738495,
-      "rewards/margins": 0.1210569366812706,
-      "rewards/rejected": -0.21750275790691376,
+      "logits/chosen": -2.5758731365203857,
+      "logits/rejected": -2.472479820251465,
+      "logps/chosen": -290.3058166503906,
+      "logps/rejected": -270.32891845703125,
+      "loss": 0.6375,
+      "rewards/accuracies": 0.668749988079071,
+      "rewards/chosen": -0.09799840301275253,
+      "rewards/margins": 0.11976947635412216,
+      "rewards/rejected": -0.21776790916919708,
       "step": 60
     },
     {
       "epoch": 0.16,
-      "grad_norm": 14.922787112175543,
+      "grad_norm": 14.438870715842985,
       "learning_rate": 4.946196886175515e-07,
-      "logits/chosen": -2.5715270042419434,
-      "logits/rejected": -2.5131115913391113,
-      "logps/chosen": -284.8800964355469,
-      "logps/rejected": -291.2018737792969,
-      "loss": 0.6102,
-      "rewards/accuracies": 0.6812499761581421,
-      "rewards/chosen": -0.09101851284503937,
-      "rewards/margins": 0.22473247349262238,
-      "rewards/rejected": -0.31575098633766174,
+      "logits/chosen": -2.568722724914551,
+      "logits/rejected": -2.5102906227111816,
+      "logps/chosen": -284.9170227050781,
+      "logps/rejected": -291.45648193359375,
+      "loss": 0.6099,
+      "rewards/accuracies": 0.675000011920929,
+      "rewards/chosen": -0.09138783067464828,
+      "rewards/margins": 0.22690913081169128,
+      "rewards/rejected": -0.31829696893692017,
       "step": 70
     },
     {
       "epoch": 0.18285714285714286,
-      "grad_norm": 11.602482470889782,
+      "grad_norm": 12.571362458092375,
       "learning_rate": 4.897191188239667e-07,
-      "logits/chosen": -2.5607805252075195,
-      "logits/rejected": -2.50270938873291,
-      "logps/chosen": -291.697998046875,
-      "logps/rejected": -307.26983642578125,
-      "loss": 0.6078,
+      "logits/chosen": -2.5521557331085205,
+      "logits/rejected": -2.4944987297058105,
+      "logps/chosen": -291.53851318359375,
+      "logps/rejected": -307.15631103515625,
+      "loss": 0.6075,
       "rewards/accuracies": 0.706250011920929,
-      "rewards/chosen": -0.2466896027326584,
-      "rewards/margins": 0.27291998267173767,
-      "rewards/rejected": -0.5196096301078796,
+      "rewards/chosen": -0.24509508907794952,
+      "rewards/margins": 0.27337878942489624,
+      "rewards/rejected": -0.5184738636016846,
       "step": 80
     },
     {
       "epoch": 0.2057142857142857,
-      "grad_norm": 17.533063770924706,
+      "grad_norm": 13.355734952743322,
       "learning_rate": 4.832875107981763e-07,
-      "logits/chosen": -2.6349377632141113,
-      "logits/rejected": -2.5697710514068604,
-      "logps/chosen": -295.2223815917969,
-      "logps/rejected": -310.2930603027344,
-      "loss": 0.6076,
-      "rewards/accuracies": 0.6937500238418579,
-      "rewards/chosen": -0.18937139213085175,
-      "rewards/margins": 0.3580796718597412,
-      "rewards/rejected": -0.5474510192871094,
+      "logits/chosen": -2.6172804832458496,
+      "logits/rejected": -2.551274538040161,
+      "logps/chosen": -292.2721252441406,
+      "logps/rejected": -307.9034423828125,
+      "loss": 0.6079,
+      "rewards/accuracies": 0.7124999761581421,
+      "rewards/chosen": -0.15986846387386322,
+      "rewards/margins": 0.3636865019798279,
+      "rewards/rejected": -0.5235549807548523,
       "step": 90
     },
     {
       "epoch": 0.22857142857142856,
-      "grad_norm": 16.194420237420413,
+      "grad_norm": 14.60913947420442,
       "learning_rate": 4.753659419387223e-07,
-      "logits/chosen": -2.6082775592803955,
-      "logits/rejected": -2.501986026763916,
-      "logps/chosen": -317.41748046875,
-      "logps/rejected": -301.41156005859375,
-      "loss": 0.5982,
-      "rewards/accuracies": 0.7437499761581421,
-      "rewards/chosen": -0.31031134724617004,
-      "rewards/margins": 0.3692900538444519,
-      "rewards/rejected": -0.6796014904975891,
+      "logits/chosen": -2.629087448120117,
+      "logits/rejected": -2.5364227294921875,
+      "logps/chosen": -312.4242858886719,
+      "logps/rejected": -296.9600524902344,
+      "loss": 0.5995,
+      "rewards/accuracies": 0.737500011920929,
+      "rewards/chosen": -0.2603791356086731,
+      "rewards/margins": 0.3747071921825409,
+      "rewards/rejected": -0.6350862979888916,
       "step": 100
     },
     {
       "epoch": 0.22857142857142856,
-      "eval_logits/chosen": -2.428199529647827,
-      "eval_logits/rejected": -2.290175437927246,
-      "eval_logps/chosen": -292.6412048339844,
-      "eval_logps/rejected": -277.6054382324219,
-      "eval_loss": 0.5954251885414124,
+      "eval_logits/chosen": -2.465451717376709,
+      "eval_logits/rejected": -2.34794545173645,
+      "eval_logps/chosen": -306.55853271484375,
+      "eval_logps/rejected": -294.0323486328125,
+      "eval_loss": 0.5960295796394348,
       "eval_rewards/accuracies": 0.7155172228813171,
-      "eval_rewards/chosen": -0.17037244141101837,
-      "eval_rewards/margins": 0.41492074728012085,
-      "eval_rewards/rejected": -0.5852932333946228,
-      "eval_runtime": 92.2211,
-      "eval_samples_per_second": 19.854,
-      "eval_steps_per_second": 0.314,
+      "eval_rewards/chosen": -0.30954551696777344,
+      "eval_rewards/margins": 0.4400167167186737,
+      "eval_rewards/rejected": -0.7495622038841248,
+      "eval_runtime": 90.9114,
+      "eval_samples_per_second": 20.14,
+      "eval_steps_per_second": 0.319,
       "step": 100
     },
     {
       "epoch": 0.25142857142857145,
-      "grad_norm": 17.063109893518277,
+      "grad_norm": 16.802955839707842,
       "learning_rate": 4.660050057270191e-07,
-      "logits/chosen": -2.2657978534698486,
-      "logits/rejected": -2.190037965774536,
-      "logps/chosen": -358.4273376464844,
-      "logps/rejected": -371.21563720703125,
-      "loss": 0.591,
-      "rewards/accuracies": 0.6187499761581421,
-      "rewards/chosen": -0.4212372303009033,
-      "rewards/margins": 0.31182247400283813,
-      "rewards/rejected": -0.7330597639083862,
+      "logits/chosen": -2.4196040630340576,
+      "logits/rejected": -2.3442416191101074,
+      "logps/chosen": -358.9793395996094,
+      "logps/rejected": -372.57965087890625,
+      "loss": 0.5847,
+      "rewards/accuracies": 0.6312500238418579,
+      "rewards/chosen": -0.42675742506980896,
+      "rewards/margins": 0.3199427127838135,
+      "rewards/rejected": -0.7467001676559448,
       "step": 110
     },
     {
       "epoch": 0.2742857142857143,
-      "grad_norm": 18.104917147072054,
+      "grad_norm": 18.751366323915487,
       "learning_rate": 4.5526448859687144e-07,
-      "logits/chosen": -1.961683988571167,
-      "logits/rejected": -1.7426531314849854,
-      "logps/chosen": -373.987060546875,
-      "logps/rejected": -336.79095458984375,
-      "loss": 0.5711,
-      "rewards/accuracies": 0.6937500238418579,
-      "rewards/chosen": -0.6308258175849915,
-      "rewards/margins": 0.4199417233467102,
-      "rewards/rejected": -1.0507675409317017,
+      "logits/chosen": -2.141848087310791,
+      "logits/rejected": -1.955249547958374,
+      "logps/chosen": -351.53961181640625,
+      "logps/rejected": -315.23406982421875,
+      "loss": 0.571,
+      "rewards/accuracies": 0.6875,
+      "rewards/chosen": -0.40635138750076294,
+      "rewards/margins": 0.42884722352027893,
+      "rewards/rejected": -0.8351985812187195,
       "step": 120
     },
     {
       "epoch": 0.29714285714285715,
-      "grad_norm": 23.635801722109512,
+      "grad_norm": 24.80835479779821,
       "learning_rate": 4.432129880904388e-07,
-      "logits/chosen": -1.0558098554611206,
-      "logits/rejected": -0.7218537330627441,
-      "logps/chosen": -395.3338317871094,
-      "logps/rejected": -383.4067077636719,
-      "loss": 0.551,
-      "rewards/accuracies": 0.6312500238418579,
-      "rewards/chosen": -0.901779055595398,
-      "rewards/margins": 0.42758530378341675,
-      "rewards/rejected": -1.3293644189834595,
+      "logits/chosen": -1.6259305477142334,
+      "logits/rejected": -1.344472885131836,
+      "logps/chosen": -370.61798095703125,
+      "logps/rejected": -360.02374267578125,
+      "loss": 0.5487,
+      "rewards/accuracies": 0.6625000238418579,
+      "rewards/chosen": -0.6546204686164856,
+      "rewards/margins": 0.4409145414829254,
+      "rewards/rejected": -1.0955349206924438,
       "step": 130
     },
     {
       "epoch": 0.32,
-      "grad_norm": 23.352536101734486,
+      "grad_norm": 22.937175363847114,
       "learning_rate": 4.299274747394055e-07,
-      "logits/chosen": -0.8035767674446106,
-      "logits/rejected": -0.4614217281341553,
-      "logps/chosen": -391.3183288574219,
-      "logps/rejected": -408.2334289550781,
-      "loss": 0.5542,
-      "rewards/accuracies": 0.731249988079071,
-      "rewards/chosen": -0.7587070465087891,
-      "rewards/margins": 0.6045126914978027,
-      "rewards/rejected": -1.3632197380065918,
+      "logits/chosen": -1.5916811227798462,
+      "logits/rejected": -1.4237914085388184,
+      "logps/chosen": -361.98211669921875,
+      "logps/rejected": -371.1536560058594,
+      "loss": 0.5577,
+      "rewards/accuracies": 0.7250000238418579,
+      "rewards/chosen": -0.4653448462486267,
+      "rewards/margins": 0.5270770788192749,
+      "rewards/rejected": -0.9924219250679016,
       "step": 140
     },
     {
       "epoch": 0.34285714285714286,
-      "grad_norm": 20.925582545841166,
+      "grad_norm": 22.019311232158735,
       "learning_rate": 4.1549280046953653e-07,
-      "logits/chosen": -0.7857555150985718,
-      "logits/rejected": -0.3248134255409241,
-      "logps/chosen": -371.37091064453125,
-      "logps/rejected": -420.19720458984375,
-      "loss": 0.5294,
-      "rewards/accuracies": 0.706250011920929,
-      "rewards/chosen": -0.8438502550125122,
-      "rewards/margins": 0.5951322317123413,
-      "rewards/rejected": -1.438982367515564,
+      "logits/chosen": -0.7861512303352356,
+      "logits/rejected": -0.3463224768638611,
+      "logps/chosen": -362.31756591796875,
+      "logps/rejected": -414.545166015625,
+      "loss": 0.5335,
+      "rewards/accuracies": 0.71875,
+      "rewards/chosen": -0.7533167004585266,
+      "rewards/margins": 0.6291457414627075,
+      "rewards/rejected": -1.3824622631072998,
       "step": 150
     },
     {
       "epoch": 0.3657142857142857,
-      "grad_norm": 24.471284191884987,
+      "grad_norm": 28.0394510543571,
       "learning_rate": 4.000011566683401e-07,
-      "logits/chosen": -0.6328128576278687,
-      "logits/rejected": -0.07759309560060501,
-      "logps/chosen": -406.115966796875,
-      "logps/rejected": -437.8179626464844,
-      "loss": 0.5424,
-      "rewards/accuracies": 0.737500011920929,
-      "rewards/chosen": -1.0677564144134521,
-      "rewards/margins": 0.7539954781532288,
-      "rewards/rejected": -1.8217518329620361,
+      "logits/chosen": -0.27471694350242615,
+      "logits/rejected": 0.3287709653377533,
+      "logps/chosen": -392.02667236328125,
+      "logps/rejected": -424.70556640625,
+      "loss": 0.5445,
+      "rewards/accuracies": 0.7562500238418579,
+      "rewards/chosen": -0.9268640279769897,
+      "rewards/margins": 0.7637636065483093,
+      "rewards/rejected": -1.6906276941299438,
       "step": 160
     },
     {
       "epoch": 0.38857142857142857,
-      "grad_norm": 24.225836916738423,
+      "grad_norm": 26.83796617150145,
       "learning_rate": 3.8355148537705047e-07,
-      "logits/chosen": -1.5064557790756226,
-      "logits/rejected": -1.0156046152114868,
-      "logps/chosen": -380.225830078125,
-      "logps/rejected": -391.9797668457031,
-      "loss": 0.5359,
-      "rewards/accuracies": 0.7124999761581421,
-      "rewards/chosen": -0.7612096667289734,
-      "rewards/margins": 0.5012774467468262,
-      "rewards/rejected": -1.2624871730804443,
+      "logits/chosen": -0.9808514714241028,
+      "logits/rejected": -0.5751891732215881,
+      "logps/chosen": -374.8761291503906,
+      "logps/rejected": -386.53851318359375,
+      "loss": 0.5452,
+      "rewards/accuracies": 0.7250000238418579,
+      "rewards/chosen": -0.7077123522758484,
+      "rewards/margins": 0.5003622770309448,
+      "rewards/rejected": -1.2080745697021484,
       "step": 170
     },
     {
       "epoch": 0.4114285714285714,
-      "grad_norm": 21.58523060279369,
+      "grad_norm": 24.04944603295859,
       "learning_rate": 3.662488473675315e-07,
-      "logits/chosen": -1.3468199968338013,
-      "logits/rejected": -0.5501200556755066,
-      "logps/chosen": -415.23846435546875,
-      "logps/rejected": -432.3148498535156,
-      "loss": 0.5511,
+      "logits/chosen": -1.283348798751831,
+      "logits/rejected": -0.556102991104126,
+      "logps/chosen": -405.58306884765625,
+      "logps/rejected": -420.27606201171875,
+      "loss": 0.5456,
       "rewards/accuracies": 0.7875000238418579,
-      "rewards/chosen": -0.8505870699882507,
-      "rewards/margins": 0.873055636882782,
-      "rewards/rejected": -1.7236427068710327,
+      "rewards/chosen": -0.7540328502655029,
+      "rewards/margins": 0.8492224812507629,
+      "rewards/rejected": -1.603255271911621,
       "step": 180
     },
     {
       "epoch": 0.4342857142857143,
-      "grad_norm": 22.105812847980097,
+      "grad_norm": 24.200865372247023,
       "learning_rate": 3.48203751140067e-07,
-      "logits/chosen": -1.2279399633407593,
-      "logits/rejected": -0.5397592782974243,
-      "logps/chosen": -367.1853942871094,
-      "logps/rejected": -383.40606689453125,
-      "loss": 0.5454,
-      "rewards/accuracies": 0.7124999761581421,
-      "rewards/chosen": -0.9092782735824585,
-      "rewards/margins": 0.5798004269599915,
-      "rewards/rejected": -1.4890786409378052,
+      "logits/chosen": -0.9612107276916504,
+      "logits/rejected": -0.4138285517692566,
+      "logps/chosen": -374.7398376464844,
+      "logps/rejected": -386.4855041503906,
+      "loss": 0.5501,
+      "rewards/accuracies": 0.6937500238418579,
+      "rewards/chosen": -0.9848228693008423,
+      "rewards/margins": 0.5350502133369446,
+      "rewards/rejected": -1.5198729038238525,
       "step": 190
     },
     {
       "epoch": 0.45714285714285713,
-      "grad_norm": 18.556629832547383,
+      "grad_norm": 21.201969213616838,
       "learning_rate": 3.2953144712759537e-07,
-      "logits/chosen": -1.407790184020996,
-      "logits/rejected": -0.6917704343795776,
-      "logps/chosen": -349.32354736328125,
-      "logps/rejected": -378.60113525390625,
-      "loss": 0.5439,
+      "logits/chosen": -1.3318579196929932,
+      "logits/rejected": -0.688397228717804,
+      "logps/chosen": -338.3636169433594,
+      "logps/rejected": -369.132080078125,
+      "loss": 0.5431,
       "rewards/accuracies": 0.71875,
-      "rewards/chosen": -0.8098081350326538,
-      "rewards/margins": 0.687397301197052,
-      "rewards/rejected": -1.4972054958343506,
+      "rewards/chosen": -0.7002092599868774,
+      "rewards/margins": 0.7023059725761414,
+      "rewards/rejected": -1.402515172958374,
       "step": 200
     },
     {
       "epoch": 0.45714285714285713,
-      "eval_logits/chosen": -1.037967562675476,
-      "eval_logits/rejected": -0.04150526225566864,
-      "eval_logps/chosen": -356.8392333984375,
-      "eval_logps/rejected": -386.5412902832031,
-      "eval_loss": 0.5353143215179443,
-      "eval_rewards/accuracies": 0.7586206793785095,
-      "eval_rewards/chosen": -0.8123530149459839,
-      "eval_rewards/margins": 0.8622987270355225,
-      "eval_rewards/rejected": -1.674651861190796,
-      "eval_runtime": 90.9527,
-      "eval_samples_per_second": 20.131,
+      "eval_logits/chosen": -1.0623676776885986,
+      "eval_logits/rejected": -0.09592445194721222,
+      "eval_logps/chosen": -338.33111572265625,
+      "eval_logps/rejected": -371.40704345703125,
+      "eval_loss": 0.5325908660888672,
+      "eval_rewards/accuracies": 0.7629310488700867,
+      "eval_rewards/chosen": -0.627271831035614,
+      "eval_rewards/margins": 0.8960375785827637,
+      "eval_rewards/rejected": -1.523309350013733,
+      "eval_runtime": 90.9273,
+      "eval_samples_per_second": 20.137,
       "eval_steps_per_second": 0.319,
       "step": 200
     },
     {
       "epoch": 0.48,
-      "grad_norm": 20.355837183026274,
+      "grad_norm": 26.3766221546479,
       "learning_rate": 3.103511916141658e-07,
-      "logits/chosen": -0.7288470268249512,
-      "logits/rejected": -0.05566452816128731,
-      "logps/chosen": -357.4866943359375,
-      "logps/rejected": -403.9605407714844,
-      "loss": 0.5321,
-      "rewards/accuracies": 0.706250011920929,
-      "rewards/chosen": -0.9403315782546997,
-      "rewards/margins": 0.6464741230010986,
-      "rewards/rejected": -1.5868055820465088,
+      "logits/chosen": -0.7819164991378784,
+      "logits/rejected": -0.08627365529537201,
+      "logps/chosen": -337.33160400390625,
+      "logps/rejected": -386.0670471191406,
+      "loss": 0.5355,
+      "rewards/accuracies": 0.699999988079071,
+      "rewards/chosen": -0.7387806177139282,
+      "rewards/margins": 0.6690900921821594,
+      "rewards/rejected": -1.4078707695007324,
       "step": 210
     },
     {
       "epoch": 0.5028571428571429,
-      "grad_norm": 22.193503671674048,
+      "grad_norm": 25.508902398231484,
       "learning_rate": 2.9078548506882117e-07,
-      "logits/chosen": -0.47817331552505493,
-      "logits/rejected": 0.3535853326320648,
-      "logps/chosen": -388.7688293457031,
-      "logps/rejected": -414.6498107910156,
-      "loss": 0.5487,
-      "rewards/accuracies": 0.7437499761581421,
-      "rewards/chosen": -1.0370280742645264,
-      "rewards/margins": 0.6476354002952576,
-      "rewards/rejected": -1.6846634149551392,
+      "logits/chosen": -0.4892755448818207,
+      "logits/rejected": 0.36961695551872253,
+      "logps/chosen": -375.30145263671875,
+      "logps/rejected": -400.32037353515625,
+      "loss": 0.5589,
+      "rewards/accuracies": 0.7124999761581421,
+      "rewards/chosen": -0.90235435962677,
+      "rewards/margins": 0.6390146613121033,
+      "rewards/rejected": -1.5413691997528076,
       "step": 220
     },
     {
       "epoch": 0.5257142857142857,
-      "grad_norm": 20.666846900072,
+      "grad_norm": 20.815219094385377,
       "learning_rate": 2.709592897595191e-07,
-      "logits/chosen": -0.1568877398967743,
-      "logits/rejected": 0.7903480529785156,
-      "logps/chosen": -379.3858642578125,
-      "logps/rejected": -402.08740234375,
-      "loss": 0.531,
-      "rewards/accuracies": 0.731249988079071,
-      "rewards/chosen": -0.983267605304718,
-      "rewards/margins": 0.6447398662567139,
-      "rewards/rejected": -1.6280075311660767,
+      "logits/chosen": -0.5455812215805054,
+      "logits/rejected": 0.40110301971435547,
+      "logps/chosen": -356.85198974609375,
+      "logps/rejected": -377.463623046875,
+      "loss": 0.5282,
+      "rewards/accuracies": 0.7250000238418579,
+      "rewards/chosen": -0.7579285502433777,
+      "rewards/margins": 0.6238406896591187,
+      "rewards/rejected": -1.3817692995071411,
       "step": 230
     },
     {
       "epoch": 0.5485714285714286,
-      "grad_norm": 26.65066022502421,
+      "grad_norm": 29.165147607391557,
       "learning_rate": 2.509992316440332e-07,
-      "logits/chosen": -0.4197590947151184,
-      "logits/rejected": 0.6013392210006714,
-      "logps/chosen": -385.48638916015625,
-      "logps/rejected": -448.73175048828125,
-      "loss": 0.5245,
-      "rewards/accuracies": 0.78125,
-      "rewards/chosen": -0.8677653074264526,
-      "rewards/margins": 0.9053336977958679,
-      "rewards/rejected": -1.7730991840362549,
+      "logits/chosen": -0.44089436531066895,
+      "logits/rejected": 0.5840796828269958,
+      "logps/chosen": -384.84747314453125,
+      "logps/rejected": -446.173828125,
+      "loss": 0.5281,
+      "rewards/accuracies": 0.7749999761581421,
+      "rewards/chosen": -0.8613764047622681,
+      "rewards/margins": 0.8861438632011414,
+      "rewards/rejected": -1.7475202083587646,
       "step": 240
     },
     {
       "epoch": 0.5714285714285714,
-      "grad_norm": 21.858480788312963,
+      "grad_norm": 21.600370107234998,
       "learning_rate": 2.3103279163519918e-07,
-      "logits/chosen": -0.17110395431518555,
-      "logits/rejected": 0.49592718482017517,
-      "logps/chosen": -368.26263427734375,
-      "logps/rejected": -430.48468017578125,
-      "loss": 0.5351,
-      "rewards/accuracies": 0.7250000238418579,
-      "rewards/chosen": -0.8763637542724609,
-      "rewards/margins": 0.7264786958694458,
-      "rewards/rejected": -1.6028425693511963,
+      "logits/chosen": -0.7112084031105042,
+      "logits/rejected": -0.16417662799358368,
+      "logps/chosen": -350.6177978515625,
+      "logps/rejected": -410.4039001464844,
+      "loss": 0.5421,
+      "rewards/accuracies": 0.737500011920929,
+      "rewards/chosen": -0.69991534948349,
+      "rewards/margins": 0.7021188735961914,
+      "rewards/rejected": -1.4020342826843262,
       "step": 250
     },
     {
       "epoch": 0.5942857142857143,
-      "grad_norm": 23.021393487481784,
+      "grad_norm": 25.092625042887974,
       "learning_rate": 2.1118749140573358e-07,
-      "logits/chosen": -0.3194964528083801,
-      "logits/rejected": 0.2826068699359894,
-      "logps/chosen": -373.82830810546875,
-      "logps/rejected": -423.619384765625,
-      "loss": 0.5387,
+      "logits/chosen": -0.9435871243476868,
+      "logits/rejected": -0.3378845751285553,
+      "logps/chosen": -350.99188232421875,
+      "logps/rejected": -403.04901123046875,
+      "loss": 0.5464,
       "rewards/accuracies": 0.706250011920929,
-      "rewards/chosen": -0.9404159784317017,
-      "rewards/margins": 0.5623483657836914,
-      "rewards/rejected": -1.502764344215393,
+      "rewards/chosen": -0.7120517492294312,
+      "rewards/margins": 0.5850083231925964,
+      "rewards/rejected": -1.297060251235962,
       "step": 260
     },
     {
       "epoch": 0.6171428571428571,
-      "grad_norm": 28.56553316601309,
+      "grad_norm": 27.282048688910123,
       "learning_rate": 1.9159007893272703e-07,
-      "logits/chosen": 0.10294933617115021,
-      "logits/rejected": 1.3178448677062988,
-      "logps/chosen": -370.7629089355469,
-      "logps/rejected": -405.086181640625,
-      "loss": 0.5159,
-      "rewards/accuracies": 0.7250000238418579,
-      "rewards/chosen": -0.9823705554008484,
-      "rewards/margins": 0.7634164690971375,
-      "rewards/rejected": -1.7457869052886963,
+      "logits/chosen": -0.04894972965121269,
+      "logits/rejected": 1.124455213546753,
+      "logps/chosen": -359.29815673828125,
+      "logps/rejected": -393.60260009765625,
+      "loss": 0.5192,
+      "rewards/accuracies": 0.71875,
+      "rewards/chosen": -0.8677228689193726,
+      "rewards/margins": 0.7632287740707397,
+      "rewards/rejected": -1.6309516429901123,
       "step": 270
     },
     {
       "epoch": 0.64,
-      "grad_norm": 24.30460588422491,
+      "grad_norm": 28.952729734552342,
       "learning_rate": 1.7236571898357766e-07,
-      "logits/chosen": 0.1169511079788208,
-      "logits/rejected": 1.0327664613723755,
-      "logps/chosen": -367.6756591796875,
-      "logps/rejected": -442.3148498535156,
-      "loss": 0.5251,
-      "rewards/accuracies": 0.7250000238418579,
-      "rewards/chosen": -0.9463755488395691,
-      "rewards/margins": 0.8457048535346985,
-      "rewards/rejected": -1.792080283164978,
+      "logits/chosen": 0.5159433484077454,
+      "logits/rejected": 1.3371174335479736,
+      "logps/chosen": -371.30694580078125,
+      "logps/rejected": -441.8828125,
+      "loss": 0.5321,
+      "rewards/accuracies": 0.7124999761581421,
+      "rewards/chosen": -0.9826878309249878,
+      "rewards/margins": 0.8050721287727356,
+      "rewards/rejected": -1.7877601385116577,
       "step": 280
     },
     {
       "epoch": 0.6628571428571428,
-      "grad_norm": 28.120774422152078,
+      "grad_norm": 26.553591107110048,
       "learning_rate": 1.5363719371356882e-07,
-      "logits/chosen": 0.10812599956989288,
-      "logits/rejected": 1.0026741027832031,
-      "logps/chosen": -394.53759765625,
-      "logps/rejected": -435.6455078125,
-      "loss": 0.5215,
-      "rewards/accuracies": 0.7562500238418579,
-      "rewards/chosen": -0.9363173246383667,
-      "rewards/margins": 0.764653205871582,
-      "rewards/rejected": -1.7009706497192383,
+      "logits/chosen": 0.31891578435897827,
+      "logits/rejected": 1.1744709014892578,
+      "logps/chosen": -396.26849365234375,
+      "logps/rejected": -436.003173828125,
+      "loss": 0.527,
+      "rewards/accuracies": 0.768750011920929,
+      "rewards/chosen": -0.9536263346672058,
+      "rewards/margins": 0.7509206533432007,
+      "rewards/rejected": -1.7045469284057617,
       "step": 290
     },
     {
       "epoch": 0.6857142857142857,
-      "grad_norm": 20.243539038486684,
+      "grad_norm": 21.68374961944068,
       "learning_rate": 1.3552411848071565e-07,
-      "logits/chosen": -0.2318529188632965,
-      "logits/rejected": 1.2198255062103271,
-      "logps/chosen": -389.1872863769531,
-      "logps/rejected": -424.034912109375,
-      "loss": 0.5109,
+      "logits/chosen": -0.4674099385738373,
+      "logits/rejected": 0.9225466847419739,
+      "logps/chosen": -378.4803771972656,
+      "logps/rejected": -411.07000732421875,
+      "loss": 0.5138,
       "rewards/accuracies": 0.762499988079071,
-      "rewards/chosen": -0.8727809190750122,
-      "rewards/margins": 0.8411470651626587,
-      "rewards/rejected": -1.713927984237671,
+      "rewards/chosen": -0.7657118439674377,
+      "rewards/margins": 0.8185675740242004,
+      "rewards/rejected": -1.5842792987823486,
       "step": 300
     },
     {
       "epoch": 0.6857142857142857,
-      "eval_logits/chosen": -0.3942234218120575,
-      "eval_logits/rejected": 1.1016443967819214,
-      "eval_logps/chosen": -345.452880859375,
-      "eval_logps/rejected": -383.7008361816406,
-      "eval_loss": 0.5189641118049622,
-      "eval_rewards/accuracies": 0.7715517282485962,
-      "eval_rewards/chosen": -0.6984895467758179,
-      "eval_rewards/margins": 0.9477577805519104,
-      "eval_rewards/rejected": -1.6462475061416626,
-      "eval_runtime": 91.163,
-      "eval_samples_per_second": 20.085,
-      "eval_steps_per_second": 0.318,
+      "eval_logits/chosen": -0.7676966190338135,
+      "eval_logits/rejected": 0.6739733219146729,
+      "eval_logps/chosen": -332.8910827636719,
+      "eval_logps/rejected": -368.96173095703125,
+      "eval_loss": 0.524158775806427,
+      "eval_rewards/accuracies": 0.7629310488700867,
+      "eval_rewards/chosen": -0.5728713274002075,
+      "eval_rewards/margins": 0.9259848594665527,
+      "eval_rewards/rejected": -1.4988560676574707,
+      "eval_runtime": 90.1222,
+      "eval_samples_per_second": 20.317,
+      "eval_steps_per_second": 0.322,
       "step": 300
     },
     {
       "epoch": 0.7085714285714285,
-      "grad_norm": 21.712761933059525,
+      "grad_norm": 21.32729150729434,
       "learning_rate": 1.1814217788631473e-07,
-      "logits/chosen": -0.18514932692050934,
-      "logits/rejected": 0.6901147961616516,
-      "logps/chosen": -342.0721740722656,
-      "logps/rejected": -390.0445861816406,
-      "loss": 0.5324,
-      "rewards/accuracies": 0.6937500238418579,
-      "rewards/chosen": -0.8576802015304565,
-      "rewards/margins": 0.6708122491836548,
-      "rewards/rejected": -1.5284925699234009,
+      "logits/chosen": -0.6603255271911621,
+      "logits/rejected": 0.27117711305618286,
+      "logps/chosen": -326.17193603515625,
+      "logps/rejected": -374.136474609375,
+      "loss": 0.5335,
+      "rewards/accuracies": 0.6875,
+      "rewards/chosen": -0.6986777782440186,
+      "rewards/margins": 0.6707334518432617,
+      "rewards/rejected": -1.3694112300872803,
       "step": 310
     },
     {
       "epoch": 0.7314285714285714,
-      "grad_norm": 22.89579586434415,
+      "grad_norm": 26.8635604286969,
       "learning_rate": 1.0160238692045331e-07,
-      "logits/chosen": 0.14465376734733582,
-      "logits/rejected": 0.9812980890274048,
-      "logps/chosen": -342.1743469238281,
-      "logps/rejected": -399.4386901855469,
-      "loss": 0.5259,
-      "rewards/accuracies": 0.675000011920929,
-      "rewards/chosen": -0.9789196848869324,
-      "rewards/margins": 0.6055852174758911,
-      "rewards/rejected": -1.5845048427581787,
+      "logits/chosen": -0.23095369338989258,
+      "logits/rejected": 0.5602467656135559,
+      "logps/chosen": -329.86492919921875,
+      "logps/rejected": -387.0187683105469,
+      "loss": 0.5276,
+      "rewards/accuracies": 0.6625000238418579,
+      "rewards/chosen": -0.8558257222175598,
+      "rewards/margins": 0.6044800281524658,
+      "rewards/rejected": -1.4603056907653809,
       "step": 320
     },
     {
       "epoch": 0.7542857142857143,
-      "grad_norm": 21.156935890608,
+      "grad_norm": 24.235988068898312,
       "learning_rate": 8.601038193139438e-08,
-      "logits/chosen": -0.23074492812156677,
-      "logits/rejected": 0.9273271560668945,
-      "logps/chosen": -383.02691650390625,
-      "logps/rejected": -410.26519775390625,
-      "loss": 0.5203,
+      "logits/chosen": -0.3879459798336029,
+      "logits/rejected": 0.737913966178894,
+      "logps/chosen": -381.5906677246094,
+      "logps/rejected": -407.6598815917969,
+      "loss": 0.5205,
       "rewards/accuracies": 0.78125,
-      "rewards/chosen": -0.8380874395370483,
-      "rewards/margins": 0.7866854071617126,
-      "rewards/rejected": -1.6247727870941162,
+      "rewards/chosen": -0.823724627494812,
+      "rewards/margins": 0.7749950289726257,
+      "rewards/rejected": -1.5987197160720825,
       "step": 330
     },
     {
       "epoch": 0.7771428571428571,
-      "grad_norm": 21.402992880624122,
+      "grad_norm": 19.828650597594493,
       "learning_rate": 7.146574594727572e-08,
-      "logits/chosen": 0.4683713912963867,
-      "logits/rejected": 1.2558484077453613,
-      "logps/chosen": -365.50213623046875,
-      "logps/rejected": -427.18682861328125,
-      "loss": 0.5174,
-      "rewards/accuracies": 0.7562500238418579,
-      "rewards/chosen": -0.9955114126205444,
-      "rewards/margins": 0.8561191558837891,
-      "rewards/rejected": -1.8516305685043335,
+      "logits/chosen": 0.20474159717559814,
+      "logits/rejected": 1.0162971019744873,
+      "logps/chosen": -362.6044921875,
+      "logps/rejected": -421.4503479003906,
+      "loss": 0.521,
+      "rewards/accuracies": 0.768750011920929,
+      "rewards/chosen": -0.9665349721908569,
+      "rewards/margins": 0.8277307748794556,
+      "rewards/rejected": -1.7942657470703125,
       "step": 340
     },
     {
       "epoch": 0.8,
-      "grad_norm": 23.419375804789023,
+      "grad_norm": 22.77041280299393,
       "learning_rate": 5.8061372659157306e-08,
-      "logits/chosen": -0.03433036804199219,
-      "logits/rejected": 1.043505072593689,
-      "logps/chosen": -382.8368225097656,
-      "logps/rejected": -416.30145263671875,
-      "loss": 0.5252,
-      "rewards/accuracies": 0.6937500238418579,
-      "rewards/chosen": -0.9312959909439087,
-      "rewards/margins": 0.6887425780296326,
-      "rewards/rejected": -1.620038628578186,
+      "logits/chosen": -0.23095539212226868,
+      "logits/rejected": 0.8589683771133423,
+      "logps/chosen": -380.7283630371094,
+      "logps/rejected": -412.0213317871094,
+      "loss": 0.5294,
+      "rewards/accuracies": 0.699999988079071,
+      "rewards/chosen": -0.9102112054824829,
+      "rewards/margins": 0.667026162147522,
+      "rewards/rejected": -1.5772373676300049,
       "step": 350
     },
     {
       "epoch": 0.8228571428571428,
-      "grad_norm": 21.8208411590094,
+      "grad_norm": 24.963575760417292,
       "learning_rate": 4.5882873127531614e-08,
-      "logits/chosen": -0.20763058960437775,
-      "logits/rejected": 1.1433674097061157,
-      "logps/chosen": -370.57281494140625,
-      "logps/rejected": -423.702392578125,
-      "loss": 0.5099,
-      "rewards/accuracies": 0.793749988079071,
-      "rewards/chosen": -0.8516775369644165,
-      "rewards/margins": 0.8744813799858093,
-      "rewards/rejected": -1.726158857345581,
+      "logits/chosen": -0.4319024682044983,
+      "logits/rejected": 0.861344039440155,
+      "logps/chosen": -366.2247314453125,
+      "logps/rejected": -415.48028564453125,
+      "loss": 0.5177,
+      "rewards/accuracies": 0.7875000238418579,
+      "rewards/chosen": -0.8081969022750854,
+      "rewards/margins": 0.8357402086257935,
+      "rewards/rejected": -1.643937110900879,
       "step": 360
     },
     {
       "epoch": 0.8457142857142858,
-      "grad_norm": 21.479692671582267,
+      "grad_norm": 22.865584640185784,
       "learning_rate": 3.500802900154412e-08,
-      "logits/chosen": -0.02174173854291439,
-      "logits/rejected": 1.3574512004852295,
-      "logps/chosen": -351.3383483886719,
-      "logps/rejected": -407.7884826660156,
-      "loss": 0.515,
+      "logits/chosen": -0.290349543094635,
+      "logits/rejected": 1.0458359718322754,
+      "logps/chosen": -349.37005615234375,
+      "logps/rejected": -402.3358459472656,
+      "loss": 0.5229,
       "rewards/accuracies": 0.762499988079071,
-      "rewards/chosen": -0.8027693033218384,
-      "rewards/margins": 0.8909598588943481,
-      "rewards/rejected": -1.6937291622161865,
+      "rewards/chosen": -0.7830866575241089,
+      "rewards/margins": 0.856115996837616,
+      "rewards/rejected": -1.6392027139663696,
       "step": 370
     },
     {
       "epoch": 0.8685714285714285,
-      "grad_norm": 23.836354879990434,
+      "grad_norm": 25.875193424699745,
       "learning_rate": 2.550629574310309e-08,
-      "logits/chosen": -0.2868313491344452,
-      "logits/rejected": 1.0226128101348877,
-      "logps/chosen": -414.9786682128906,
-      "logps/rejected": -421.4689025878906,
-      "loss": 0.5176,
-      "rewards/accuracies": 0.71875,
-      "rewards/chosen": -0.969223141670227,
-      "rewards/margins": 0.6714719533920288,
-      "rewards/rejected": -1.6406952142715454,
+      "logits/chosen": -0.4227335453033447,
+      "logits/rejected": 0.8518049120903015,
+      "logps/chosen": -414.580322265625,
+      "logps/rejected": -417.60418701171875,
+      "loss": 0.5205,
+      "rewards/accuracies": 0.737500011920929,
+      "rewards/chosen": -0.9652398824691772,
+      "rewards/margins": 0.6368075609207153,
+      "rewards/rejected": -1.602047324180603,
       "step": 380
     },
     {
       "epoch": 0.8914285714285715,
-      "grad_norm": 24.654106172641406,
+      "grad_norm": 23.202834331091832,
       "learning_rate": 1.7438359028687983e-08,
-      "logits/chosen": 0.04682580381631851,
-      "logits/rejected": 0.8462721109390259,
-      "logps/chosen": -395.02130126953125,
-      "logps/rejected": -450.59375,
-      "loss": 0.525,
-      "rewards/accuracies": 0.71875,
-      "rewards/chosen": -0.836483359336853,
-      "rewards/margins": 0.7217278480529785,
-      "rewards/rejected": -1.5582112073898315,
+      "logits/chosen": -0.143943652510643,
+      "logits/rejected": 0.5985423922538757,
+      "logps/chosen": -392.14068603515625,
+      "logps/rejected": -445.93560791015625,
+      "loss": 0.526,
+      "rewards/accuracies": 0.7124999761581421,
+      "rewards/chosen": -0.8076774477958679,
+      "rewards/margins": 0.7039517760276794,
+      "rewards/rejected": -1.5116291046142578,
       "step": 390
     },
     {
       "epoch": 0.9142857142857143,
-      "grad_norm": 35.84442600986327,
+      "grad_norm": 31.25928182468869,
       "learning_rate": 1.0855747162029361e-08,
-      "logits/chosen": 0.2935855984687805,
-      "logits/rejected": 0.8280227780342102,
-      "logps/chosen": -379.0578918457031,
-      "logps/rejected": -430.62261962890625,
-      "loss": 0.5457,
-      "rewards/accuracies": 0.731249988079071,
-      "rewards/chosen": -1.0032463073730469,
-      "rewards/margins": 0.6372436881065369,
-      "rewards/rejected": -1.640489935874939,
+      "logits/chosen": 0.005290505476295948,
+      "logits/rejected": 0.5700523257255554,
+      "logps/chosen": -371.91815185546875,
+      "logps/rejected": -422.913818359375,
+      "loss": 0.5493,
+      "rewards/accuracies": 0.7250000238418579,
+      "rewards/chosen": -0.9318493604660034,
+      "rewards/margins": 0.6315523982048035,
+      "rewards/rejected": -1.563401699066162,
       "step": 400
     },
     {
       "epoch": 0.9142857142857143,
-      "eval_logits/chosen": 0.24073126912117004,
-      "eval_logits/rejected": 1.7997794151306152,
-      "eval_logps/chosen": -362.5745849609375,
-      "eval_logps/rejected": -405.11981201171875,
-      "eval_loss": 0.5175118446350098,
-      "eval_rewards/accuracies": 0.7715517282485962,
-      "eval_rewards/chosen": -0.8697066307067871,
-      "eval_rewards/margins": 0.9907302856445312,
-      "eval_rewards/rejected": -1.8604369163513184,
-      "eval_runtime": 90.8335,
-      "eval_samples_per_second": 20.158,
+      "eval_logits/chosen": -0.012980658560991287,
+      "eval_logits/rejected": 1.572424292564392,
+      "eval_logps/chosen": -357.8257751464844,
+      "eval_logps/rejected": -398.6412353515625,
+      "eval_loss": 0.5201366543769836,
+      "eval_rewards/accuracies": 0.7758620977401733,
+      "eval_rewards/chosen": -0.8222182989120483,
+      "eval_rewards/margins": 0.9734326004981995,
+      "eval_rewards/rejected": -1.795650839805603,
+      "eval_runtime": 90.9384,
+      "eval_samples_per_second": 20.135,
       "eval_steps_per_second": 0.319,
       "step": 400
     },
     {
       "epoch": 0.9371428571428572,
-      "grad_norm": 27.47931239793458,
+      "grad_norm": 24.160283510679385,
       "learning_rate": 5.8005019731033615e-09,
-      "logits/chosen": 0.10173013061285019,
-      "logits/rejected": 1.089665174484253,
-      "logps/chosen": -389.411865234375,
-      "logps/rejected": -432.0257873535156,
-      "loss": 0.5148,
-      "rewards/accuracies": 0.731249988079071,
-      "rewards/chosen": -1.0123193264007568,
-      "rewards/margins": 0.710638165473938,
-      "rewards/rejected": -1.7229573726654053,
+      "logits/chosen": -0.10076072067022324,
+      "logits/rejected": 0.8628407716751099,
+      "logps/chosen": -385.8863525390625,
+      "logps/rejected": -425.5079040527344,
+      "loss": 0.5157,
+      "rewards/accuracies": 0.7250000238418579,
+      "rewards/chosen": -0.977064311504364,
+      "rewards/margins": 0.6807142496109009,
+      "rewards/rejected": -1.6577785015106201,
       "step": 410
     },
     {
       "epoch": 0.96,
-      "grad_norm": 21.72575201877371,
+      "grad_norm": 21.949347011437847,
       "learning_rate": 2.3049103053431886e-09,
-      "logits/chosen": -0.029790764674544334,
-      "logits/rejected": 1.4268020391464233,
-      "logps/chosen": -357.87603759765625,
-      "logps/rejected": -413.14227294921875,
-      "loss": 0.5143,
-      "rewards/accuracies": 0.7749999761581421,
-      "rewards/chosen": -0.7512796521186829,
-      "rewards/margins": 1.0402743816375732,
-      "rewards/rejected": -1.7915542125701904,
+      "logits/chosen": -0.21599116921424866,
+      "logits/rejected": 1.2091736793518066,
+      "logps/chosen": -353.96514892578125,
+      "logps/rejected": -403.6047058105469,
+      "loss": 0.5167,
+      "rewards/accuracies": 0.78125,
+      "rewards/chosen": -0.7121706604957581,
+      "rewards/margins": 0.9840081930160522,
+      "rewards/rejected": -1.696178674697876,
       "step": 420
     },
     {
       "epoch": 0.9828571428571429,
-      "grad_norm": 22.231222051131574,
+      "grad_norm": 23.473520497590073,
       "learning_rate": 3.9129780600541397e-10,
-      "logits/chosen": 0.29859107732772827,
-      "logits/rejected": 1.3065626621246338,
-      "logps/chosen": -373.48834228515625,
-      "logps/rejected": -435.72515869140625,
-      "loss": 0.5136,
+      "logits/chosen": 0.03587682545185089,
+      "logits/rejected": 1.0841922760009766,
+      "logps/chosen": -372.0699768066406,
+      "logps/rejected": -431.086181640625,
+      "loss": 0.5158,
       "rewards/accuracies": 0.7562500238418579,
-      "rewards/chosen": -0.9125651121139526,
-      "rewards/margins": 0.8079904317855835,
-      "rewards/rejected": -1.7205555438995361,
+      "rewards/chosen": -0.8983810544013977,
+      "rewards/margins": 0.775784432888031,
+      "rewards/rejected": -1.6741654872894287,
       "step": 430
     },
     {
       "epoch": 0.9988571428571429,
       "step": 437,
       "total_flos": 0.0,
-      "train_loss": 0.5586249282758351,
-      "train_runtime": 11006.8714,
-      "train_samples_per_second": 5.088,
+      "train_loss": 0.5608050315822016,
+      "train_runtime": 10950.5403,
+      "train_samples_per_second": 5.114,
       "train_steps_per_second": 0.04
     }
   ],

training_args.bin

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:7aa0c3ecfea1b56864b7fbfe5f813be8913ccf567da5242b94111688b9a04360
+oid sha256:7b039acd68c2072172fff79eb06019ffcc90aef6b5f626cda75fb3fb82e4df03
 size 7608