Model save

README.md

@@ -1,13 +1,11 @@
 ---
+license: apache-2.0
 library_name: peft
 tags:
-- alignment-handbook
 - trl
 - dpo
 - generated_from_trainer
 base_model: norallm/normistral-7b-warm
-datasets:
-- hugodk-sch/aftonposten_title_prefs
 model-index:
 - name: ap-normistral-7b-align-scan
   results: []
@@ -18,17 +16,17 @@ should probably proofread and complete it, then remove this comment. -->
 
 # ap-normistral-7b-align-scan
 
-This model is a fine-tuned version of [data/ap-normistral-7b-sft-qlora](https://huggingface.co/data/ap-normistral-7b-sft-qlora) on the hugodk-sch/aftonposten_title_prefs dataset.
+This model is a fine-tuned version of [norallm/normistral-7b-warm](https://huggingface.co/norallm/normistral-7b-warm) on the None dataset.
 It achieves the following results on the evaluation set:
-- Loss: 0.6803
-- Rewards/chosen: -0.0838
-- Rewards/rejected: -0.1384
-- Rewards/accuracies: 0.5187
-- Rewards/margins: 0.0546
-- Logps/rejected: -36.6583
-- Logps/chosen: -32.8620
-- Logits/rejected: 98.2235
-- Logits/chosen: 98.2555
+- Loss: 0.4980
+- Rewards/chosen: 0.1165
+- Rewards/rejected: 0.1232
+- Rewards/accuracies: 0.5017
+- Rewards/margins: -0.0067
+- Logps/rejected: -35.7905
+- Logps/chosen: -32.2767
+- Logits/rejected: 98.3630
+- Logits/chosen: 98.3713
 
 ## Model description
 
@@ -63,9 +61,9 @@ The following hyperparameters were used during training:
 
 | Training Loss | Epoch | Step | Validation Loss | Rewards/chosen | Rewards/rejected | Rewards/accuracies | Rewards/margins | Logps/rejected | Logps/chosen | Logits/rejected | Logits/chosen |
 |:-------------:|:-----:|:----:|:---------------:|:--------------:|:----------------:|:------------------:|:---------------:|:--------------:|:------------:|:---------------:|:-------------:|
-| 0.6551        | 0.26  | 100  | 0.6957          | -0.0035        | -0.0114          | 0.5224             | 0.0079          | -36.0235       | -32.4608     | 98.7375         | 98.7478       |
-| 0.5948        | 0.52  | 200  | 0.6720          | -0.0629        | -0.1290          | 0.5515             | 0.0661          | -36.6115       | -32.7579     | 98.3088         | 98.3381       |
-| 0.5903        | 0.78  | 300  | 0.6774          | -0.0728        | -0.1310          | 0.5341             | 0.0583          | -36.6218       | -32.8071     | 98.2471         | 98.2788       |
+| 0.461         | 0.26  | 100  | 0.5000          | 0.0677         | 0.0660           | 0.5158             | 0.0017          | -35.8723       | -32.3464     | 98.6883         | 98.7043       |
+| 0.3172        | 0.52  | 200  | 0.4956          | 0.0869         | 0.0578           | 0.5133             | 0.0291          | -35.8840       | -32.3190     | 98.4409         | 98.4463       |
+| 0.3303        | 0.78  | 300  | 0.4980          | 0.1165         | 0.1232           | 0.5017             | -0.0067         | -35.7905       | -32.2767     | 98.3630         | 98.3713       |
 
 
 ### Framework versions

adapter_model.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:81ea432485ae5594f553e6c220b246a98ea4e7ece55c6a8f62d04d488ab7f717
+oid sha256:c469839f0357fe0aae4b2cbe9eb6f372e9c962908b8e6777a086f551553efcf2
 size 671150064

all_results.json

@@ -1,20 +1,7 @@
 {
     "epoch": 1.0,
-    "eval_logits/chosen": 98.25546264648438,
-    "eval_logits/rejected": 98.2234878540039,
-    "eval_logps/chosen": -32.8620491027832,
-    "eval_logps/rejected": -36.65834426879883,
-    "eval_loss": 0.6803030967712402,
-    "eval_rewards/accuracies": 0.5186877250671387,
-    "eval_rewards/chosen": -0.0837748572230339,
-    "eval_rewards/margins": 0.05458855256438255,
-    "eval_rewards/rejected": -0.13836342096328735,
-    "eval_runtime": 103.6862,
-    "eval_samples": 343,
-    "eval_samples_per_second": 3.308,
-    "eval_steps_per_second": 0.415,
-    "train_loss": 0.6168378699909557,
-    "train_runtime": 2560.1818,
+    "train_loss": 0.38857739250381274,
+    "train_runtime": 2559.074,
     "train_samples": 3079,
     "train_samples_per_second": 1.203,
     "train_steps_per_second": 0.15

train_results.json

@@ -1,7 +1,7 @@
 {
     "epoch": 1.0,
-    "train_loss": 0.6168378699909557,
-    "train_runtime": 2560.1818,
+    "train_loss": 0.38857739250381274,
+    "train_runtime": 2559.074,
     "train_samples": 3079,
     "train_samples_per_second": 1.203,
     "train_steps_per_second": 0.15

trainer_state.json

@@ -10,13 +10,13 @@
   "log_history": [
     {
       "epoch": 0.0,
-      "grad_norm": 13.0625,
+      "grad_norm": 22.875,
       "learning_rate": 1.282051282051282e-07,
       "logits/chosen": 88.18099975585938,
       "logits/rejected": 88.25153350830078,
       "logps/chosen": -29.073104858398438,
       "logps/rejected": -26.25731658935547,
-      "loss": 0.6931,
+      "loss": 0.5,
       "rewards/accuracies": 0.0,
       "rewards/chosen": 0.0,
       "rewards/margins": 0.0,
@@ -25,628 +25,628 @@
     },
     {
       "epoch": 0.03,
-      "grad_norm": 10.6875,
+      "grad_norm": 8.75,
       "learning_rate": 1.282051282051282e-06,
-      "logits/chosen": 81.09551239013672,
-      "logits/rejected": 80.80068969726562,
-      "logps/chosen": -34.28557586669922,
-      "logps/rejected": -33.081260681152344,
-      "loss": 0.6879,
-      "rewards/accuracies": 0.5277777910232544,
-      "rewards/chosen": -0.008770154789090157,
-      "rewards/margins": 0.01580061763525009,
-      "rewards/rejected": -0.0245707705616951,
+      "logits/chosen": 81.07839965820312,
+      "logits/rejected": 80.78140258789062,
+      "logps/chosen": -34.22772216796875,
+      "logps/rejected": -33.004886627197266,
+      "loss": 0.493,
+      "rewards/accuracies": 0.4583333432674408,
+      "rewards/chosen": 0.009805028326809406,
+      "rewards/margins": 0.0423436164855957,
+      "rewards/rejected": -0.03253858909010887,
       "step": 10
     },
     {
       "epoch": 0.05,
-      "grad_norm": 11.8125,
+      "grad_norm": 15.5,
       "learning_rate": 2.564102564102564e-06,
-      "logits/chosen": 80.68476867675781,
-      "logits/rejected": 80.57356262207031,
-      "logps/chosen": -33.56308364868164,
-      "logps/rejected": -30.81569480895996,
-      "loss": 0.6814,
-      "rewards/accuracies": 0.574999988079071,
-      "rewards/chosen": 0.02536492981016636,
-      "rewards/margins": 0.03300638496875763,
-      "rewards/rejected": -0.007641455624252558,
+      "logits/chosen": 80.6578369140625,
+      "logits/rejected": 80.54452514648438,
+      "logps/chosen": -33.567832946777344,
+      "logps/rejected": -30.769500732421875,
+      "loss": 0.4824,
+      "rewards/accuracies": 0.4749999940395355,
+      "rewards/chosen": 0.0854543000459671,
+      "rewards/margins": 0.07986308634281158,
+      "rewards/rejected": 0.005591208580881357,
       "step": 20
     },
     {
       "epoch": 0.08,
-      "grad_norm": 12.4375,
+      "grad_norm": 17.875,
       "learning_rate": 3.846153846153847e-06,
-      "logits/chosen": 82.52334594726562,
-      "logits/rejected": 82.55731964111328,
-      "logps/chosen": -33.843528747558594,
-      "logps/rejected": -31.111019134521484,
-      "loss": 0.7142,
+      "logits/chosen": 82.4737777709961,
+      "logits/rejected": 82.50477600097656,
+      "logps/chosen": -33.87335205078125,
+      "logps/rejected": -31.195053100585938,
+      "loss": 0.5111,
       "rewards/accuracies": 0.4124999940395355,
-      "rewards/chosen": 0.03385842964053154,
-      "rewards/margins": -0.028703877702355385,
-      "rewards/rejected": 0.06256230920553207,
+      "rewards/chosen": 0.09762730449438095,
+      "rewards/margins": -0.06251715123653412,
+      "rewards/rejected": 0.16014449298381805,
       "step": 30
     },
     {
       "epoch": 0.1,
-      "grad_norm": 11.0,
+      "grad_norm": 15.625,
       "learning_rate": 4.999896948438434e-06,
-      "logits/chosen": 81.0788803100586,
-      "logits/rejected": 81.07527923583984,
-      "logps/chosen": -32.68939971923828,
-      "logps/rejected": -33.106483459472656,
-      "loss": 0.6804,
-      "rewards/accuracies": 0.5375000238418579,
-      "rewards/chosen": 0.0969332829117775,
-      "rewards/margins": 0.044524990022182465,
-      "rewards/rejected": 0.05240829661488533,
+      "logits/chosen": 81.05291748046875,
+      "logits/rejected": 81.04745483398438,
+      "logps/chosen": -32.597801208496094,
+      "logps/rejected": -33.1552734375,
+      "loss": 0.4525,
+      "rewards/accuracies": 0.550000011920929,
+      "rewards/chosen": 0.40338650345802307,
+      "rewards/margins": 0.25411099195480347,
+      "rewards/rejected": 0.14927545189857483,
       "step": 40
     },
     {
       "epoch": 0.13,
-      "grad_norm": 9.25,
+      "grad_norm": 14.5625,
       "learning_rate": 4.987541037542187e-06,
-      "logits/chosen": 78.8182373046875,
-      "logits/rejected": 78.82775115966797,
-      "logps/chosen": -30.46297836303711,
-      "logps/rejected": -30.7069091796875,
-      "loss": 0.6746,
-      "rewards/accuracies": 0.5625,
-      "rewards/chosen": 0.14845404028892517,
-      "rewards/margins": 0.07517841458320618,
-      "rewards/rejected": 0.073275625705719,
+      "logits/chosen": 78.7928466796875,
+      "logits/rejected": 78.8023910522461,
+      "logps/chosen": -30.487594604492188,
+      "logps/rejected": -30.689586639404297,
+      "loss": 0.4635,
+      "rewards/accuracies": 0.6000000238418579,
+      "rewards/chosen": 0.5023558735847473,
+      "rewards/margins": 0.23376783728599548,
+      "rewards/rejected": 0.26858800649642944,
       "step": 50
     },
     {
       "epoch": 0.16,
-      "grad_norm": 10.0,
+      "grad_norm": 13.375,
       "learning_rate": 4.954691471941119e-06,
-      "logits/chosen": 83.39994812011719,
-      "logits/rejected": 83.45416259765625,
-      "logps/chosen": -30.7622013092041,
-      "logps/rejected": -29.376235961914062,
-      "loss": 0.6858,
-      "rewards/accuracies": 0.5,
-      "rewards/chosen": 0.08259173482656479,
-      "rewards/margins": 0.03973449766635895,
-      "rewards/rejected": 0.04285724088549614,
+      "logits/chosen": 83.43833923339844,
+      "logits/rejected": 83.5004653930664,
+      "logps/chosen": -30.851482391357422,
+      "logps/rejected": -29.280162811279297,
+      "loss": 0.501,
+      "rewards/accuracies": 0.42500001192092896,
+      "rewards/chosen": 0.22657600045204163,
+      "rewards/margins": 0.009325137361884117,
+      "rewards/rejected": 0.21725085377693176,
       "step": 60
     },
     {
       "epoch": 0.18,
-      "grad_norm": 14.3125,
+      "grad_norm": 13.5,
       "learning_rate": 4.901618883413549e-06,
-      "logits/chosen": 84.03162384033203,
-      "logits/rejected": 84.06212615966797,
-      "logps/chosen": -30.389270782470703,
-      "logps/rejected": -32.93138885498047,
-      "loss": 0.6894,
-      "rewards/accuracies": 0.4625000059604645,
-      "rewards/chosen": 0.04795076698064804,
-      "rewards/margins": 0.02668655477464199,
-      "rewards/rejected": 0.02126421593129635,
+      "logits/chosen": 84.01557159423828,
+      "logits/rejected": 84.04491424560547,
+      "logps/chosen": -30.311126708984375,
+      "logps/rejected": -32.73175811767578,
+      "loss": 0.4997,
+      "rewards/accuracies": 0.42500001192092896,
+      "rewards/chosen": 0.22253043949604034,
+      "rewards/margins": 0.008363036438822746,
+      "rewards/rejected": 0.21416740119457245,
       "step": 70
     },
     {
       "epoch": 0.21,
-      "grad_norm": 12.0,
+      "grad_norm": 16.375,
       "learning_rate": 4.828760511501322e-06,
-      "logits/chosen": 81.65167236328125,
-      "logits/rejected": 81.63506317138672,
-      "logps/chosen": -31.229572296142578,
-      "logps/rejected": -30.818950653076172,
-      "loss": 0.6715,
+      "logits/chosen": 81.79966735839844,
+      "logits/rejected": 81.77557373046875,
+      "logps/chosen": -31.009265899658203,
+      "logps/rejected": -30.690744400024414,
+      "loss": 0.4434,
       "rewards/accuracies": 0.574999988079071,
-      "rewards/chosen": 0.05507441610097885,
-      "rewards/margins": 0.06733567267656326,
-      "rewards/rejected": -0.012261250987648964,
+      "rewards/chosen": 0.34697332978248596,
+      "rewards/margins": 0.3001402020454407,
+      "rewards/rejected": 0.04683312773704529,
       "step": 80
     },
     {
       "epoch": 0.23,
-      "grad_norm": 13.875,
+      "grad_norm": 20.75,
       "learning_rate": 4.7367166013034295e-06,
-      "logits/chosen": 78.45601654052734,
-      "logits/rejected": 78.42174530029297,
-      "logps/chosen": -32.425289154052734,
-      "logps/rejected": -31.132177352905273,
-      "loss": 0.6671,
-      "rewards/accuracies": 0.550000011920929,
-      "rewards/chosen": 0.04257836565375328,
-      "rewards/margins": 0.07871954143047333,
-      "rewards/rejected": -0.03614116832613945,
+      "logits/chosen": 78.57239532470703,
+      "logits/rejected": 78.55184173583984,
+      "logps/chosen": -32.28063201904297,
+      "logps/rejected": -30.936914443969727,
+      "loss": 0.4612,
+      "rewards/accuracies": 0.5249999761581421,
+      "rewards/chosen": 0.2502840459346771,
+      "rewards/margins": 0.2400922328233719,
+      "rewards/rejected": 0.0101918401196599,
       "step": 90
     },
     {
       "epoch": 0.26,
-      "grad_norm": 11.4375,
+      "grad_norm": 15.5625,
       "learning_rate": 4.626245458345211e-06,
-      "logits/chosen": 83.59445190429688,
-      "logits/rejected": 83.61743927001953,
-      "logps/chosen": -33.955238342285156,
-      "logps/rejected": -31.89117431640625,
-      "loss": 0.6551,
-      "rewards/accuracies": 0.5874999761581421,
-      "rewards/chosen": 0.06995107233524323,
-      "rewards/margins": 0.09807688742876053,
-      "rewards/rejected": -0.028125811368227005,
+      "logits/chosen": 83.69164276123047,
+      "logits/rejected": 83.69895935058594,
+      "logps/chosen": -33.99507141113281,
+      "logps/rejected": -31.751794815063477,
+      "loss": 0.461,
+      "rewards/accuracies": 0.512499988079071,
+      "rewards/chosen": 0.21694469451904297,
+      "rewards/margins": 0.21781782805919647,
+      "rewards/rejected": -0.0008731067064218223,
       "step": 100
     },
     {
       "epoch": 0.26,
-      "eval_logits/chosen": 98.74779510498047,
-      "eval_logits/rejected": 98.73747253417969,
-      "eval_logps/chosen": -32.46078872680664,
-      "eval_logps/rejected": -36.02348327636719,
-      "eval_loss": 0.6956869959831238,
-      "eval_rewards/accuracies": 0.5224252939224243,
-      "eval_rewards/chosen": -0.003522348590195179,
-      "eval_rewards/margins": 0.00786901917308569,
-      "eval_rewards/rejected": -0.011391367763280869,
-      "eval_runtime": 104.3728,
-      "eval_samples_per_second": 3.286,
-      "eval_steps_per_second": 0.412,
+      "eval_logits/chosen": 98.70431518554688,
+      "eval_logits/rejected": 98.68829345703125,
+      "eval_logps/chosen": -32.346439361572266,
+      "eval_logps/rejected": -35.87228775024414,
+      "eval_loss": 0.5000348687171936,
+      "eval_rewards/accuracies": 0.5157807469367981,
+      "eval_rewards/chosen": 0.06771623343229294,
+      "eval_rewards/margins": 0.0017476709326729178,
+      "eval_rewards/rejected": 0.06596855819225311,
+      "eval_runtime": 104.2308,
+      "eval_samples_per_second": 3.291,
+      "eval_steps_per_second": 0.413,
       "step": 100
     },
     {
       "epoch": 0.29,
-      "grad_norm": 14.0,
+      "grad_norm": 17.625,
       "learning_rate": 4.498257201263691e-06,
-      "logits/chosen": 83.79670715332031,
-      "logits/rejected": 83.66975402832031,
-      "logps/chosen": -32.42261505126953,
-      "logps/rejected": -32.766204833984375,
-      "loss": 0.6376,
-      "rewards/accuracies": 0.675000011920929,
-      "rewards/chosen": 0.12003795802593231,
-      "rewards/margins": 0.1421007364988327,
-      "rewards/rejected": -0.02206278033554554,
+      "logits/chosen": 83.92507934570312,
+      "logits/rejected": 83.79533386230469,
+      "logps/chosen": -32.187767028808594,
+      "logps/rejected": -32.65043258666992,
+      "loss": 0.3988,
+      "rewards/accuracies": 0.637499988079071,
+      "rewards/chosen": 0.5845246911048889,
+      "rewards/margins": 0.5807043313980103,
+      "rewards/rejected": 0.0038203417789191008,
       "step": 110
     },
     {
       "epoch": 0.31,
-      "grad_norm": 13.4375,
+      "grad_norm": 19.375,
       "learning_rate": 4.353806263777678e-06,
-      "logits/chosen": 83.85148620605469,
-      "logits/rejected": 83.95706939697266,
-      "logps/chosen": -28.113927841186523,
-      "logps/rejected": -35.665374755859375,
-      "loss": 0.6109,
-      "rewards/accuracies": 0.7124999761581421,
-      "rewards/chosen": 0.1631302535533905,
-      "rewards/margins": 0.20437581837177277,
-      "rewards/rejected": -0.04124556854367256,
+      "logits/chosen": 83.96737670898438,
+      "logits/rejected": 84.08332061767578,
+      "logps/chosen": -28.09195327758789,
+      "logps/rejected": -35.17109680175781,
+      "loss": 0.4351,
+      "rewards/accuracies": 0.550000011920929,
+      "rewards/chosen": 0.5863426327705383,
+      "rewards/margins": 0.3847105801105499,
+      "rewards/rejected": 0.2016320526599884,
       "step": 120
     },
     {
       "epoch": 0.34,
-      "grad_norm": 9.5,
+      "grad_norm": 13.25,
       "learning_rate": 4.1940827077152755e-06,
-      "logits/chosen": 80.99261474609375,
-      "logits/rejected": 81.01200866699219,
-      "logps/chosen": -30.389972686767578,
-      "logps/rejected": -32.014381408691406,
-      "loss": 0.6511,
-      "rewards/accuracies": 0.625,
-      "rewards/chosen": 0.10347676277160645,
-      "rewards/margins": 0.120028056204319,
-      "rewards/rejected": -0.0165513064712286,
+      "logits/chosen": 81.23406219482422,
+      "logits/rejected": 81.2734603881836,
+      "logps/chosen": -30.209375381469727,
+      "logps/rejected": -31.70619773864746,
+      "loss": 0.4414,
+      "rewards/accuracies": 0.6000000238418579,
+      "rewards/chosen": 0.48858681321144104,
+      "rewards/margins": 0.330787718296051,
+      "rewards/rejected": 0.1577991098165512,
       "step": 130
     },
     {
       "epoch": 0.36,
-      "grad_norm": 9.3125,
+      "grad_norm": 15.3125,
       "learning_rate": 4.0204024186666215e-06,
-      "logits/chosen": 82.00664520263672,
-      "logits/rejected": 82.01942443847656,
-      "logps/chosen": -27.06869125366211,
-      "logps/rejected": -33.10210418701172,
-      "loss": 0.611,
-      "rewards/accuracies": 0.6875,
-      "rewards/chosen": 0.07572650164365768,
-      "rewards/margins": 0.21147167682647705,
-      "rewards/rejected": -0.13574519753456116,
+      "logits/chosen": 82.31855773925781,
+      "logits/rejected": 82.33930206298828,
+      "logps/chosen": -26.636510848999023,
+      "logps/rejected": -32.51647186279297,
+      "loss": 0.3966,
+      "rewards/accuracies": 0.612500011920929,
+      "rewards/chosen": 0.5675702691078186,
+      "rewards/margins": 0.6327372789382935,
+      "rewards/rejected": -0.06516699492931366,
       "step": 140
     },
     {
       "epoch": 0.39,
-      "grad_norm": 10.0625,
+      "grad_norm": 14.5625,
       "learning_rate": 3.834196265035119e-06,
-      "logits/chosen": 80.3797378540039,
-      "logits/rejected": 80.35098266601562,
-      "logps/chosen": -28.888397216796875,
-      "logps/rejected": -33.30742645263672,
-      "loss": 0.596,
-      "rewards/accuracies": 0.7250000238418579,
-      "rewards/chosen": 0.10014281421899796,
-      "rewards/margins": 0.24832835793495178,
-      "rewards/rejected": -0.14818556606769562,
+      "logits/chosen": 80.7647476196289,
+      "logits/rejected": 80.73397064208984,
+      "logps/chosen": -28.527297973632812,
+      "logps/rejected": -32.617584228515625,
+      "loss": 0.3943,
+      "rewards/accuracies": 0.7124999761581421,
+      "rewards/chosen": 0.603269100189209,
+      "rewards/margins": 0.6390293836593628,
+      "rewards/rejected": -0.035760145634412766,
       "step": 150
     },
     {
       "epoch": 0.42,
-      "grad_norm": 13.9375,
+      "grad_norm": 16.0,
       "learning_rate": 3.636998309800573e-06,
-      "logits/chosen": 82.10334014892578,
-      "logits/rejected": 82.10926818847656,
-      "logps/chosen": -33.88823699951172,
-      "logps/rejected": -30.74798011779785,
-      "loss": 0.6209,
-      "rewards/accuracies": 0.75,
-      "rewards/chosen": 0.04970277473330498,
-      "rewards/margins": 0.2087908238172531,
-      "rewards/rejected": -0.15908804535865784,
+      "logits/chosen": 82.51421356201172,
+      "logits/rejected": 82.53550720214844,
+      "logps/chosen": -33.11312484741211,
+      "logps/rejected": -30.041656494140625,
+      "loss": 0.3649,
+      "rewards/accuracies": 0.675000011920929,
+      "rewards/chosen": 0.7165369987487793,
+      "rewards/margins": 0.7789196372032166,
+      "rewards/rejected": -0.06238250061869621,
       "step": 160
     },
     {
       "epoch": 0.44,
-      "grad_norm": 12.4375,
+      "grad_norm": 11.75,
       "learning_rate": 3.4304331721118078e-06,
-      "logits/chosen": 82.82877349853516,
-      "logits/rejected": 82.77672576904297,
-      "logps/chosen": -30.899311065673828,
-      "logps/rejected": -32.958866119384766,
-      "loss": 0.598,
-      "rewards/accuracies": 0.699999988079071,
-      "rewards/chosen": 0.0719953253865242,
-      "rewards/margins": 0.2597564160823822,
-      "rewards/rejected": -0.1877611130475998,
+      "logits/chosen": 83.22119903564453,
+      "logits/rejected": 83.18238067626953,
+      "logps/chosen": -30.43136978149414,
+      "logps/rejected": -32.384212493896484,
+      "loss": 0.3752,
+      "rewards/accuracies": 0.6499999761581421,
+      "rewards/chosen": 0.5795407295227051,
+      "rewards/margins": 0.8344497680664062,
+      "rewards/rejected": -0.25490903854370117,
       "step": 170
     },
     {
       "epoch": 0.47,
-      "grad_norm": 8.75,
+      "grad_norm": 7.625,
       "learning_rate": 3.2162026428305436e-06,
-      "logits/chosen": 80.19332885742188,
-      "logits/rejected": 80.17335510253906,
-      "logps/chosen": -30.60544204711914,
-      "logps/rejected": -31.953319549560547,
-      "loss": 0.5849,
-      "rewards/accuracies": 0.699999988079071,
-      "rewards/chosen": 0.1183418408036232,
-      "rewards/margins": 0.2775646448135376,
-      "rewards/rejected": -0.159222811460495,
+      "logits/chosen": 80.77767181396484,
+      "logits/rejected": 80.74992370605469,
+      "logps/chosen": -30.34661293029785,
+      "logps/rejected": -31.299999237060547,
+      "loss": 0.3834,
+      "rewards/accuracies": 0.6625000238418579,
+      "rewards/chosen": 0.5953787565231323,
+      "rewards/margins": 0.6953333616256714,
+      "rewards/rejected": -0.09995462000370026,
       "step": 180
     },
     {
       "epoch": 0.49,
-      "grad_norm": 6.59375,
+      "grad_norm": 7.5,
       "learning_rate": 2.996071664294641e-06,
-      "logits/chosen": 81.75331115722656,
-      "logits/rejected": 81.74398040771484,
-      "logps/chosen": -30.514236450195312,
-      "logps/rejected": -30.920120239257812,
-      "loss": 0.6564,
-      "rewards/accuracies": 0.625,
-      "rewards/chosen": 0.050435226410627365,
-      "rewards/margins": 0.14085446298122406,
-      "rewards/rejected": -0.090419240295887,
+      "logits/chosen": 82.40937805175781,
+      "logits/rejected": 82.38029479980469,
+      "logps/chosen": -29.8316593170166,
+      "logps/rejected": -30.288339614868164,
+      "loss": 0.4151,
+      "rewards/accuracies": 0.612500011920929,
+      "rewards/chosen": 0.6543310284614563,
+      "rewards/margins": 0.5285537838935852,
+      "rewards/rejected": 0.1257772147655487,
       "step": 190
     },
     {
       "epoch": 0.52,
-      "grad_norm": 8.9375,
+      "grad_norm": 8.5625,
       "learning_rate": 2.7718537898066833e-06,
-      "logits/chosen": 77.08003997802734,
-      "logits/rejected": 77.01986694335938,
-      "logps/chosen": -33.93625259399414,
-      "logps/rejected": -33.12162780761719,
-      "loss": 0.5948,
-      "rewards/accuracies": 0.6875,
-      "rewards/chosen": 0.16329774260520935,
-      "rewards/margins": 0.28894227743148804,
-      "rewards/rejected": -0.1256445348262787,
+      "logits/chosen": 77.76971435546875,
+      "logits/rejected": 77.71305847167969,
+      "logps/chosen": -33.00738525390625,
+      "logps/rejected": -32.31935119628906,
+      "loss": 0.3172,
+      "rewards/accuracies": 0.699999988079071,
+      "rewards/chosen": 1.2217466831207275,
+      "rewards/margins": 1.0999095439910889,
+      "rewards/rejected": 0.12183725833892822,
       "step": 200
     },
     {
       "epoch": 0.52,
-      "eval_logits/chosen": 98.33805847167969,
-      "eval_logits/rejected": 98.30875396728516,
-      "eval_logps/chosen": -32.75787353515625,
-      "eval_logps/rejected": -36.611534118652344,
-      "eval_loss": 0.672012209892273,
-      "eval_rewards/accuracies": 0.5514950156211853,
-      "eval_rewards/chosen": -0.06293957680463791,
-      "eval_rewards/margins": 0.06606128066778183,
-      "eval_rewards/rejected": -0.12900085747241974,
-      "eval_runtime": 104.1536,
-      "eval_samples_per_second": 3.293,
+      "eval_logits/chosen": 98.44633483886719,
+      "eval_logits/rejected": 98.44091033935547,
+      "eval_logps/chosen": -32.319034576416016,
+      "eval_logps/rejected": -35.88397979736328,
+      "eval_loss": 0.49560824036598206,
+      "eval_rewards/accuracies": 0.5132890343666077,
+      "eval_rewards/chosen": 0.08689937740564346,
+      "eval_rewards/margins": 0.029115768149495125,
+      "eval_rewards/rejected": 0.057783618569374084,
+      "eval_runtime": 104.0206,
+      "eval_samples_per_second": 3.297,
       "eval_steps_per_second": 0.413,
       "step": 200
     },
     {
       "epoch": 0.55,
-      "grad_norm": 15.0625,
+      "grad_norm": 18.875,
       "learning_rate": 2.5453962426402006e-06,
-      "logits/chosen": 79.71280670166016,
-      "logits/rejected": 79.62548065185547,
-      "logps/chosen": -33.420021057128906,
-      "logps/rejected": -35.76679229736328,
-      "loss": 0.6026,
-      "rewards/accuracies": 0.699999988079071,
-      "rewards/chosen": 0.10514093935489655,
-      "rewards/margins": 0.2612799406051636,
-      "rewards/rejected": -0.15613897144794464,
+      "logits/chosen": 80.35867309570312,
+      "logits/rejected": 80.27238464355469,
+      "logps/chosen": -32.687442779541016,
+      "logps/rejected": -34.85881423950195,
+      "loss": 0.3718,
+      "rewards/accuracies": 0.6625000238418579,
+      "rewards/chosen": 0.8807934522628784,
+      "rewards/margins": 0.7916989922523499,
+      "rewards/rejected": 0.08909468352794647,
       "step": 210
     },
     {
       "epoch": 0.57,
-      "grad_norm": 10.375,
+      "grad_norm": 15.625,
       "learning_rate": 2.3185646976551794e-06,
-      "logits/chosen": 81.73656463623047,
-      "logits/rejected": 81.82989501953125,
-      "logps/chosen": -31.172740936279297,
-      "logps/rejected": -31.53964614868164,
-      "loss": 0.5726,
-      "rewards/accuracies": 0.800000011920929,
-      "rewards/chosen": 0.15524254739284515,
-      "rewards/margins": 0.3173803985118866,
-      "rewards/rejected": -0.16213785111904144,
+      "logits/chosen": 82.43341064453125,
+      "logits/rejected": 82.50484466552734,
+      "logps/chosen": -30.421199798583984,
+      "logps/rejected": -30.724105834960938,
+      "loss": 0.3248,
+      "rewards/accuracies": 0.7250000238418579,
+      "rewards/chosen": 1.0694276094436646,
+      "rewards/margins": 1.0660343170166016,
+      "rewards/rejected": 0.0033930898644030094,
       "step": 220
     },
     {
       "epoch": 0.6,
-      "grad_norm": 11.5625,
+      "grad_norm": 11.6875,
       "learning_rate": 2.0932279108998323e-06,
-      "logits/chosen": 78.81214904785156,
-      "logits/rejected": 78.8668212890625,
-      "logps/chosen": -32.4487190246582,
-      "logps/rejected": -34.767303466796875,
-      "loss": 0.6172,
-      "rewards/accuracies": 0.637499988079071,
-      "rewards/chosen": 0.07140493392944336,
-      "rewards/margins": 0.22243526577949524,
-      "rewards/rejected": -0.15103033185005188,
+      "logits/chosen": 79.52705383300781,
+      "logits/rejected": 79.57157897949219,
+      "logps/chosen": -31.819992065429688,
+      "logps/rejected": -33.83860397338867,
+      "loss": 0.4096,
+      "rewards/accuracies": 0.5874999761581421,
+      "rewards/chosen": 0.6900272369384766,
+      "rewards/margins": 0.5685457587242126,
+      "rewards/rejected": 0.1214815154671669,
       "step": 230
     },
     {
       "epoch": 0.62,
-      "grad_norm": 14.25,
+      "grad_norm": 17.25,
       "learning_rate": 1.8712423238279358e-06,
-      "logits/chosen": 81.07857513427734,
-      "logits/rejected": 81.38665771484375,
-      "logps/chosen": -31.030254364013672,
-      "logps/rejected": -32.145328521728516,
-      "loss": 0.5977,
-      "rewards/accuracies": 0.737500011920929,
-      "rewards/chosen": 0.13156220316886902,
-      "rewards/margins": 0.254219114780426,
-      "rewards/rejected": -0.12265688180923462,
+      "logits/chosen": 81.8853988647461,
+      "logits/rejected": 82.16355895996094,
+      "logps/chosen": -30.063451766967773,
+      "logps/rejected": -31.44558334350586,
+      "loss": 0.3113,
+      "rewards/accuracies": 0.762499988079071,
+      "rewards/chosen": 1.137229561805725,
+      "rewards/margins": 1.0767086744308472,
+      "rewards/rejected": 0.06052086502313614,
       "step": 240
     },
     {
       "epoch": 0.65,
-      "grad_norm": 12.75,
+      "grad_norm": 17.0,
       "learning_rate": 1.6544367689701824e-06,
-      "logits/chosen": 79.71233367919922,
-      "logits/rejected": 79.77317810058594,
-      "logps/chosen": -27.3262939453125,
-      "logps/rejected": -30.594249725341797,
-      "loss": 0.6333,
-      "rewards/accuracies": 0.5625,
-      "rewards/chosen": 0.053920842707157135,
-      "rewards/margins": 0.19262339174747467,
-      "rewards/rejected": -0.13870255649089813,
+      "logits/chosen": 80.61282348632812,
+      "logits/rejected": 80.67658996582031,
+      "logps/chosen": -26.506671905517578,
+      "logps/rejected": -29.721996307373047,
+      "loss": 0.4011,
+      "rewards/accuracies": 0.6000000238418579,
+      "rewards/chosen": 0.7624577283859253,
+      "rewards/margins": 0.6373406648635864,
+      "rewards/rejected": 0.12511704862117767,
       "step": 250
     },
     {
       "epoch": 0.68,
-      "grad_norm": 10.5,
+      "grad_norm": 12.25,
       "learning_rate": 1.4445974030621963e-06,
-      "logits/chosen": 76.8447036743164,
-      "logits/rejected": 76.98011779785156,
-      "logps/chosen": -30.586124420166016,
-      "logps/rejected": -37.12653350830078,
-      "loss": 0.5519,
-      "rewards/accuracies": 0.7124999761581421,
-      "rewards/chosen": 0.19497378170490265,
-      "rewards/margins": 0.3999541699886322,
-      "rewards/rejected": -0.20498037338256836,
+      "logits/chosen": 77.79762268066406,
+      "logits/rejected": 77.97430419921875,
+      "logps/chosen": -29.87813949584961,
+      "logps/rejected": -35.95857238769531,
+      "loss": 0.3311,
+      "rewards/accuracies": 0.675000011920929,
+      "rewards/chosen": 1.1779998540878296,
+      "rewards/margins": 1.0778592824935913,
+      "rewards/rejected": 0.10014041513204575,
       "step": 260
     },
     {
       "epoch": 0.7,
-      "grad_norm": 9.3125,
+      "grad_norm": 8.8125,
       "learning_rate": 1.243452991757889e-06,
-      "logits/chosen": 76.1262435913086,
-      "logits/rejected": 76.15739440917969,
-      "logps/chosen": -31.1810245513916,
-      "logps/rejected": -32.34056854248047,
-      "loss": 0.5874,
-      "rewards/accuracies": 0.75,
-      "rewards/chosen": 0.11957981437444687,
-      "rewards/margins": 0.28660956025123596,
-      "rewards/rejected": -0.1670297235250473,
+      "logits/chosen": 77.17610168457031,
+      "logits/rejected": 77.21318817138672,
+      "logps/chosen": -30.205867767333984,
+      "logps/rejected": -31.4431095123291,
+      "loss": 0.3276,
+      "rewards/accuracies": 0.737500011920929,
+      "rewards/chosen": 1.1011395454406738,
+      "rewards/margins": 1.0575218200683594,
+      "rewards/rejected": 0.04361782968044281,
       "step": 270
     },
     {
       "epoch": 0.73,
-      "grad_norm": 15.75,
+      "grad_norm": 17.375,
       "learning_rate": 1.0526606671603523e-06,
-      "logits/chosen": 78.95848846435547,
-      "logits/rejected": 78.7259750366211,
-      "logps/chosen": -31.400470733642578,
-      "logps/rejected": -30.376489639282227,
-      "loss": 0.6251,
-      "rewards/accuracies": 0.7749999761581421,
-      "rewards/chosen": 0.08188416808843613,
-      "rewards/margins": 0.22378277778625488,
-      "rewards/rejected": -0.14189860224723816,
+      "logits/chosen": 79.84697723388672,
+      "logits/rejected": 79.62041473388672,
+      "logps/chosen": -30.549224853515625,
+      "logps/rejected": -29.2475643157959,
+      "loss": 0.3847,
+      "rewards/accuracies": 0.699999988079071,
+      "rewards/chosen": 0.8824693560600281,
+      "rewards/margins": 0.588868260383606,
+      "rewards/rejected": 0.2936010956764221,
       "step": 280
     },
     {
       "epoch": 0.75,
-      "grad_norm": 9.3125,
+      "grad_norm": 13.125,
       "learning_rate": 8.737922755071455e-07,
-      "logits/chosen": 79.04127502441406,
-      "logits/rejected": 78.95845031738281,
-      "logps/chosen": -33.34519577026367,
-      "logps/rejected": -33.35126876831055,
-      "loss": 0.5388,
-      "rewards/accuracies": 0.800000011920929,
-      "rewards/chosen": 0.15181000530719757,
-      "rewards/margins": 0.4186907410621643,
-      "rewards/rejected": -0.26688069105148315,
+      "logits/chosen": 80.05425262451172,
+      "logits/rejected": 79.96458435058594,
+      "logps/chosen": -32.43302536010742,
+      "logps/rejected": -32.11684036254883,
+      "loss": 0.3019,
+      "rewards/accuracies": 0.7875000238418579,
+      "rewards/chosen": 1.169856309890747,
+      "rewards/margins": 1.2398390769958496,
+      "rewards/rejected": -0.06998284161090851,
       "step": 290
     },
     {
       "epoch": 0.78,
-      "grad_norm": 12.8125,
+      "grad_norm": 9.5625,
       "learning_rate": 7.08321427484816e-07,
-      "logits/chosen": 74.71063995361328,
-      "logits/rejected": 74.8211441040039,
-      "logps/chosen": -32.499046325683594,
-      "logps/rejected": -29.784500122070312,
-      "loss": 0.5903,
-      "rewards/accuracies": 0.699999988079071,
-      "rewards/chosen": 0.18133236467838287,
-      "rewards/margins": 0.3178277611732483,
-      "rewards/rejected": -0.13649536669254303,
+      "logits/chosen": 75.59251403808594,
+      "logits/rejected": 75.66484069824219,
+      "logps/chosen": -31.76091957092285,
+      "logps/rejected": -28.981002807617188,
+      "loss": 0.3303,
+      "rewards/accuracies": 0.7124999761581421,
+      "rewards/chosen": 1.151352882385254,
+      "rewards/margins": 1.0666378736495972,
+      "rewards/rejected": 0.08471502363681793,
       "step": 300
     },
     {
       "epoch": 0.78,
-      "eval_logits/chosen": 98.2787857055664,
-      "eval_logits/rejected": 98.24710083007812,
-      "eval_logps/chosen": -32.80708312988281,
-      "eval_logps/rejected": -36.62175750732422,
-      "eval_loss": 0.6774110198020935,
-      "eval_rewards/accuracies": 0.5340532064437866,
-      "eval_rewards/chosen": -0.07278140634298325,
-      "eval_rewards/margins": 0.05826449766755104,
-      "eval_rewards/rejected": -0.13104592263698578,
-      "eval_runtime": 104.1558,
-      "eval_samples_per_second": 3.293,
+      "eval_logits/chosen": 98.37130737304688,
+      "eval_logits/rejected": 98.36297607421875,
+      "eval_logps/chosen": -32.2767219543457,
+      "eval_logps/rejected": -35.79054260253906,
+      "eval_loss": 0.49796876311302185,
+      "eval_rewards/accuracies": 0.5016611218452454,
+      "eval_rewards/chosen": 0.11651887744665146,
+      "eval_rewards/margins": -0.0066675543785095215,
+      "eval_rewards/rejected": 0.12318644672632217,
+      "eval_runtime": 104.1032,
+      "eval_samples_per_second": 3.295,
       "eval_steps_per_second": 0.413,
       "step": 300
     },
     {
       "epoch": 0.81,
-      "grad_norm": 10.0,
+      "grad_norm": 10.625,
       "learning_rate": 5.576113578589035e-07,
-      "logits/chosen": 81.97877502441406,
-      "logits/rejected": 82.00166320800781,
-      "logps/chosen": -30.284442901611328,
-      "logps/rejected": -33.162872314453125,
-      "loss": 0.579,
+      "logits/chosen": 82.78644561767578,
+      "logits/rejected": 82.81319427490234,
+      "logps/chosen": -29.468921661376953,
+      "logps/rejected": -32.31016159057617,
+      "loss": 0.3414,
       "rewards/accuracies": 0.7124999761581421,
-      "rewards/chosen": 0.11581814289093018,
-      "rewards/margins": 0.30844008922576904,
-      "rewards/rejected": -0.19262196123600006,
+      "rewards/chosen": 0.9762304425239563,
+      "rewards/margins": 1.0535131692886353,
+      "rewards/rejected": -0.07728256285190582,
       "step": 310
     },
     {
       "epoch": 0.83,
-      "grad_norm": 9.8125,
+      "grad_norm": 13.0625,
       "learning_rate": 4.229036944380913e-07,
-      "logits/chosen": 79.28789520263672,
-      "logits/rejected": 79.29541778564453,
-      "logps/chosen": -30.684707641601562,
-      "logps/rejected": -29.662694931030273,
-      "loss": 0.5734,
-      "rewards/accuracies": 0.7124999761581421,
-      "rewards/chosen": 0.1827705204486847,
-      "rewards/margins": 0.3314109444618225,
-      "rewards/rejected": -0.14864039421081543,
+      "logits/chosen": 80.16923522949219,
+      "logits/rejected": 80.17256164550781,
+      "logps/chosen": -29.88885498046875,
+      "logps/rejected": -28.86279296875,
+      "loss": 0.3116,
+      "rewards/accuracies": 0.699999988079071,
+      "rewards/chosen": 1.1967929601669312,
+      "rewards/margins": 1.157100796699524,
+      "rewards/rejected": 0.03969240561127663,
       "step": 320
     },
     {
       "epoch": 0.86,
-      "grad_norm": 9.4375,
+      "grad_norm": 11.6875,
       "learning_rate": 3.053082288996112e-07,
-      "logits/chosen": 76.40345764160156,
-      "logits/rejected": 76.45513916015625,
-      "logps/chosen": -29.180185317993164,
-      "logps/rejected": -33.6518440246582,
-      "loss": 0.5234,
+      "logits/chosen": 77.39164733886719,
+      "logits/rejected": 77.41889953613281,
+      "logps/chosen": -28.371896743774414,
+      "logps/rejected": -32.67755889892578,
+      "loss": 0.2678,
       "rewards/accuracies": 0.75,
-      "rewards/chosen": 0.2344847470521927,
-      "rewards/margins": 0.43788036704063416,
-      "rewards/rejected": -0.20339563488960266,
+      "rewards/chosen": 1.3864963054656982,
+      "rewards/margins": 1.4163827896118164,
+      "rewards/rejected": -0.02988656982779503,
       "step": 330
     },
     {
       "epoch": 0.88,
-      "grad_norm": 16.875,
+      "grad_norm": 15.3125,
       "learning_rate": 2.0579377374915805e-07,
-      "logits/chosen": 80.8781509399414,
-      "logits/rejected": 80.92503356933594,
-      "logps/chosen": -32.53740310668945,
-      "logps/rejected": -34.52736282348633,
-      "loss": 0.5705,
-      "rewards/accuracies": 0.7124999761581421,
-      "rewards/chosen": 0.13494810461997986,
-      "rewards/margins": 0.3666823208332062,
-      "rewards/rejected": -0.23173420131206512,
+      "logits/chosen": 81.65691375732422,
+      "logits/rejected": 81.6773681640625,
+      "logps/chosen": -31.6644229888916,
+      "logps/rejected": -33.2541389465332,
+      "loss": 0.3491,
+      "rewards/accuracies": 0.737500011920929,
+      "rewards/chosen": 1.0834088325500488,
+      "rewards/margins": 1.003218412399292,
+      "rewards/rejected": 0.08019042015075684,
       "step": 340
     },
     {
       "epoch": 0.91,
-      "grad_norm": 9.375,
+      "grad_norm": 9.5625,
       "learning_rate": 1.2518018074041684e-07,
-      "logits/chosen": 79.81520080566406,
-      "logits/rejected": 79.83287048339844,
-      "logps/chosen": -32.53202819824219,
-      "logps/rejected": -33.971153259277344,
-      "loss": 0.561,
-      "rewards/accuracies": 0.737500011920929,
-      "rewards/chosen": 0.22234222292900085,
-      "rewards/margins": 0.39821282029151917,
-      "rewards/rejected": -0.1758706122636795,
+      "logits/chosen": 80.73246765136719,
+      "logits/rejected": 80.75402069091797,
+      "logps/chosen": -31.80560874938965,
+      "logps/rejected": -32.868934631347656,
+      "loss": 0.3373,
+      "rewards/accuracies": 0.7250000238418579,
+      "rewards/chosen": 1.2866908311843872,
+      "rewards/margins": 1.1306837797164917,
+      "rewards/rejected": 0.15600721538066864,
       "step": 350
     },
     {
       "epoch": 0.94,
-      "grad_norm": 9.25,
+      "grad_norm": 10.5,
       "learning_rate": 6.41315865106129e-08,
-      "logits/chosen": 81.41835021972656,
-      "logits/rejected": 81.45225524902344,
-      "logps/chosen": -28.548419952392578,
-      "logps/rejected": -32.2237434387207,
-      "loss": 0.5819,
-      "rewards/accuracies": 0.675000011920929,
-      "rewards/chosen": 0.20075836777687073,
-      "rewards/margins": 0.3155379891395569,
-      "rewards/rejected": -0.11477959156036377,
+      "logits/chosen": 82.23421478271484,
+      "logits/rejected": 82.29096221923828,
+      "logps/chosen": -27.619741439819336,
+      "logps/rejected": -31.5566349029541,
+      "loss": 0.2941,
+      "rewards/accuracies": 0.7749999761581421,
+      "rewards/chosen": 1.3527272939682007,
+      "rewards/margins": 1.2874782085418701,
+      "rewards/rejected": 0.06524892896413803,
       "step": 360
     },
     {
       "epoch": 0.96,
-      "grad_norm": 12.4375,
+      "grad_norm": 11.5,
       "learning_rate": 2.3150941078050325e-08,
-      "logits/chosen": 80.87324523925781,
-      "logits/rejected": 80.88945007324219,
-      "logps/chosen": -32.29994583129883,
-      "logps/rejected": -36.0130615234375,
-      "loss": 0.6003,
-      "rewards/accuracies": 0.6625000238418579,
-      "rewards/chosen": 0.1152433529496193,
-      "rewards/margins": 0.295980840921402,
-      "rewards/rejected": -0.1807374656200409,
+      "logits/chosen": 81.71080017089844,
+      "logits/rejected": 81.74567413330078,
+      "logps/chosen": -31.354503631591797,
+      "logps/rejected": -34.771522521972656,
+      "loss": 0.3671,
+      "rewards/accuracies": 0.612500011920929,
+      "rewards/chosen": 1.0651617050170898,
+      "rewards/margins": 0.8286684155464172,
+      "rewards/rejected": 0.23649325966835022,
       "step": 370
     },
     {
       "epoch": 0.99,
-      "grad_norm": 11.1875,
+      "grad_norm": 13.8125,
       "learning_rate": 2.575864278703266e-09,
-      "logits/chosen": 74.64203643798828,
-      "logits/rejected": 74.50636291503906,
-      "logps/chosen": -29.965755462646484,
-      "logps/rejected": -28.94148826599121,
-      "loss": 0.5896,
-      "rewards/accuracies": 0.737500011920929,
-      "rewards/chosen": 0.12207402288913727,
-      "rewards/margins": 0.2782566249370575,
-      "rewards/rejected": -0.15618260204792023,
+      "logits/chosen": 75.57669067382812,
+      "logits/rejected": 75.45292663574219,
+      "logps/chosen": -29.130199432373047,
+      "logps/rejected": -27.9501953125,
+      "loss": 0.3558,
+      "rewards/accuracies": 0.7250000238418579,
+      "rewards/chosen": 1.0121484994888306,
+      "rewards/margins": 0.8648824691772461,
+      "rewards/rejected": 0.14726600050926208,
       "step": 380
     },
     {
       "epoch": 1.0,
       "step": 385,
       "total_flos": 0.0,
-      "train_loss": 0.6168378699909557,
-      "train_runtime": 2560.1818,
+      "train_loss": 0.38857739250381274,
+      "train_runtime": 2559.074,
       "train_samples_per_second": 1.203,
       "train_steps_per_second": 0.15
     }