Model save

README.md

@@ -17,18 +17,18 @@ should probably proofread and complete it, then remove this comment. -->
 
 This model is a fine-tuned version of [HuggingFaceH4/mistral-7b-sft-beta](https://huggingface.co/HuggingFaceH4/mistral-7b-sft-beta) on the None dataset.
 It achieves the following results on the evaluation set:
-- Loss: 0.1314
-- Rewards/chosen: -1.5200
-- Rewards/rejected: -2.4344
-- Rewards/accuracies: 0.75
-- Rewards/margins: 0.9144
-- Logps/rejected: -500.7934
-- Logps/chosen: -409.0388
-- Logits/rejected: -2.1508
-- Logits/chosen: -2.1830
-- Debug/policy Weights: 0.2589
-- Debug/losses: 0.1297
-- Debug/raw Losses: 0.4817
+- Loss: 0.1422
+- Rewards/chosen: -1.3154
+- Rewards/rejected: -2.2768
+- Rewards/accuracies: 0.7617
+- Rewards/margins: 0.9613
+- Logps/rejected: -483.9327
+- Logps/chosen: -386.7366
+- Logits/rejected: -2.1695
+- Logits/chosen: -2.2036
+- Debug/policy Weights: 0.2815
+- Debug/losses: 0.1397
+- Debug/raw Losses: 0.4727
 
 ## Model description
 
@@ -65,15 +65,15 @@ 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 | Debug/policy Weights | Debug/losses | Debug/raw Losses |
 |:-------------:|:-----:|:----:|:---------------:|:--------------:|:----------------:|:------------------:|:---------------:|:--------------:|:------------:|:---------------:|:-------------:|:--------------------:|:------------:|:----------------:|
-| 0.2168        | 0.21  | 100  | 0.2150          | -0.5440        | -1.0580          | 0.7383             | 0.5141          | -363.1571      | -311.4377    | -2.6827         | -2.6979       | 0.3735               | 0.2082       | 0.5529           |
-| 0.1396        | 0.42  | 200  | 0.1416          | -1.3480        | -2.1286          | 0.7656             | 0.7807          | -470.2158      | -391.8350    | -2.2733         | -2.2968       | 0.2687               | 0.1390       | 0.5030           |
-| 0.1294        | 0.63  | 300  | 0.1309          | -1.6003        | -2.4486          | 0.7383             | 0.8483          | -502.2112      | -417.0714    | -2.1589         | -2.1885       | 0.2545               | 0.1284       | 0.4935           |
-| 0.1329        | 0.84  | 400  | 0.1314          | -1.5200        | -2.4344          | 0.75               | 0.9144          | -500.7934      | -409.0388    | -2.1508         | -2.1830       | 0.2589               | 0.1297       | 0.4817           |
+| 0.1781        | 0.21  | 100  | 0.2007          | -0.6478        | -1.1693          | 0.7344             | 0.5214          | -373.1867      | -319.9806    | -2.6910         | -2.7080       | 0.3512               | 0.1953       | 0.5590           |
+| 0.1616        | 0.42  | 200  | 0.1669          | -0.8830        | -1.6003          | 0.7109             | 0.7173          | -416.2844      | -343.4914    | -2.4277         | -2.4499       | 0.3174               | 0.1671       | 0.5079           |
+| 0.1343        | 0.63  | 300  | 0.1368          | -1.5021        | -2.3715          | 0.7578             | 0.8695          | -493.4114      | -405.4042    | -2.2283         | -2.2618       | 0.2666               | 0.1365       | 0.4953           |
+| 0.1398        | 0.84  | 400  | 0.1422          | -1.3154        | -2.2768          | 0.7617             | 0.9613          | -483.9327      | -386.7366    | -2.1695         | -2.2036       | 0.2815               | 0.1397       | 0.4727           |
 
 
 ### Framework versions
 
-- Transformers 4.35.2
+- Transformers 4.39.3
 - Pytorch 2.1.2+cu121
 - Datasets 2.14.6
-- Tokenizers 0.14.1
+- Tokenizers 0.15.2

all_results.json

@@ -1,8 +1,8 @@
 {
     "epoch": 1.0,
-    "train_loss": 0.17621936496331603,
-    "train_runtime": 4510.4366,
+    "train_loss": 0.17903520272865456,
+    "train_runtime": 4631.245,
     "train_samples": 61134,
-    "train_samples_per_second": 13.554,
-    "train_steps_per_second": 0.106
+    "train_samples_per_second": 13.2,
+    "train_steps_per_second": 0.103
 }

config.json

@@ -3,6 +3,7 @@
   "architectures": [
     "MistralForCausalLM"
   ],
+  "attention_dropout": 0.0,
   "bos_token_id": 1,
   "eos_token_id": 2,
   "hidden_act": "silu",
@@ -19,7 +20,7 @@
   "sliding_window": 4096,
   "tie_word_embeddings": false,
   "torch_dtype": "bfloat16",
-  "transformers_version": "4.35.2",
+  "transformers_version": "4.39.3",
   "use_cache": false,
   "vocab_size": 32000
 }

generation_config.json

@@ -2,5 +2,5 @@
   "_from_model_config": true,
   "bos_token_id": 1,
   "eos_token_id": 2,
-  "transformers_version": "4.35.2"
+  "transformers_version": "4.39.3"
 }

model-00001-of-00003.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:09cc7ca4a2419236436e6c007340b593bba2a32c8ae2632430bedae57caa0e7d
+oid sha256:18d00ba6a8e8636533d5f9f026b67bade132d932eb7e25cf20a5471d14cf4938
 size 4943162336

model-00002-of-00003.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:4a18ed3e61ca7e185538cfad27e95e49182c44b8e7cf83e0a3e9e75dfa4402b4
+oid sha256:a726c7d9b0e02053f892980f475aa8bdf053ba38518c5ec8f29a48cd43f1e728
 size 4999819336

model-00003-of-00003.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:fe431dd8a696ee319a6385010068d99a7950c45d95d27f41443264941966e681
+oid sha256:abb64bdbf0bdca4349e17eb7d2424fa102b61d7e91947dde62627f2995cbf4b8
 size 4540516344

tokenizer_config.json

@@ -1,4 +1,6 @@
 {
+  "add_bos_token": true,
+  "add_eos_token": false,
   "added_tokens_decoder": {
     "0": {
       "content": "<unk>",

train_results.json

@@ -1,8 +1,8 @@
 {
     "epoch": 1.0,
-    "train_loss": 0.17621936496331603,
-    "train_runtime": 4510.4366,
+    "train_loss": 0.17903520272865456,
+    "train_runtime": 4631.245,
     "train_samples": 61134,
-    "train_samples_per_second": 13.554,
-    "train_steps_per_second": 0.106
+    "train_samples_per_second": 13.2,
+    "train_steps_per_second": 0.103
 }

trainer_state.json

@@ -9,16 +9,17 @@
   "is_world_process_zero": true,
   "log_history": [
     {
-      "debug/losses": 0.34217238426208496,
-      "debug/policy_weights": 0.4936503767967224,
+      "debug/losses": 0.36003828048706055,
+      "debug/policy_weights": 0.5194254517555237,
       "debug/raw_losses": 0.6931471824645996,
       "epoch": 0.0,
+      "grad_norm": 5.3582401111914875,
       "learning_rate": 1.0416666666666666e-08,
-      "logits/chosen": -2.8099329471588135,
-      "logits/rejected": -2.7572641372680664,
-      "logps/chosen": -241.48843383789062,
-      "logps/rejected": -197.4517822265625,
-      "loss": 0.3561,
+      "logits/chosen": -2.888590097427368,
+      "logits/rejected": -2.871957540512085,
+      "logps/chosen": -261.88616943359375,
+      "logps/rejected": -355.31793212890625,
+      "loss": 0.3648,
       "rewards/accuracies": 0.0,
       "rewards/chosen": 0.0,
       "rewards/margins": 0.0,
@@ -26,895 +27,944 @@
       "step": 1
     },
     {
-      "debug/losses": 0.3613118529319763,
-      "debug/policy_weights": 0.5213115215301514,
-      "debug/raw_losses": 0.6931909918785095,
+      "debug/losses": 0.3548344671726227,
+      "debug/policy_weights": 0.5118452906608582,
+      "debug/raw_losses": 0.6932389736175537,
       "epoch": 0.02,
+      "grad_norm": 5.014402110225361,
       "learning_rate": 1.0416666666666667e-07,
-      "logits/chosen": -2.8320045471191406,
-      "logits/rejected": -2.8085670471191406,
-      "logps/chosen": -292.685546875,
-      "logps/rejected": -278.5729064941406,
-      "loss": 0.3674,
+      "logits/chosen": -2.7435038089752197,
+      "logits/rejected": -2.7258925437927246,
+      "logps/chosen": -256.5353088378906,
+      "logps/rejected": -244.49546813964844,
+      "loss": 0.3619,
       "rewards/accuracies": 0.4236111044883728,
-      "rewards/chosen": 5.248460729490034e-05,
-      "rewards/margins": -7.99686458776705e-05,
-      "rewards/rejected": 0.00013245324953459203,
+      "rewards/chosen": -8.211319800466299e-05,
+      "rewards/margins": -0.00017834355821833014,
+      "rewards/rejected": 9.623030200600624e-05,
       "step": 10
     },
     {
-      "debug/losses": 0.3490375578403473,
-      "debug/policy_weights": 0.5044432878494263,
-      "debug/raw_losses": 0.6918557286262512,
+      "debug/losses": 0.3572663962841034,
+      "debug/policy_weights": 0.5156484842300415,
+      "debug/raw_losses": 0.6928108930587769,
       "epoch": 0.04,
+      "grad_norm": 4.715098712970233,
       "learning_rate": 2.0833333333333333e-07,
-      "logits/chosen": -2.811972141265869,
-      "logits/rejected": -2.78340482711792,
-      "logps/chosen": -290.2806396484375,
-      "logps/rejected": -290.8512268066406,
-      "loss": 0.3549,
-      "rewards/accuracies": 0.612500011920929,
-      "rewards/chosen": 0.0011896035866811872,
-      "rewards/margins": 0.0026031401939690113,
-      "rewards/rejected": -0.0014135364908725023,
+      "logits/chosen": -2.8007731437683105,
+      "logits/rejected": -2.7729618549346924,
+      "logps/chosen": -271.37860107421875,
+      "logps/rejected": -288.34271240234375,
+      "loss": 0.3584,
+      "rewards/accuracies": 0.543749988079071,
+      "rewards/chosen": 0.0006374852382577956,
+      "rewards/margins": 0.0006833201623521745,
+      "rewards/rejected": -4.583501868182793e-05,
       "step": 20
     },
     {
-      "debug/losses": 0.3571945130825043,
-      "debug/policy_weights": 0.518287181854248,
-      "debug/raw_losses": 0.6891354322433472,
+      "debug/losses": 0.35951218008995056,
+      "debug/policy_weights": 0.521370530128479,
+      "debug/raw_losses": 0.6895479559898376,
       "epoch": 0.06,
+      "grad_norm": 4.7982052271779825,
       "learning_rate": 3.1249999999999997e-07,
-      "logits/chosen": -2.759937047958374,
-      "logits/rejected": -2.7286112308502197,
-      "logps/chosen": -246.35159301757812,
-      "logps/rejected": -227.08651733398438,
-      "loss": 0.3602,
-      "rewards/accuracies": 0.637499988079071,
-      "rewards/chosen": 0.00010563675459707156,
-      "rewards/margins": 0.008171903900802135,
-      "rewards/rejected": -0.008066266775131226,
+      "logits/chosen": -2.810387134552002,
+      "logits/rejected": -2.773030996322632,
+      "logps/chosen": -291.4071350097656,
+      "logps/rejected": -258.0267639160156,
+      "loss": 0.3579,
+      "rewards/accuracies": 0.6312500238418579,
+      "rewards/chosen": 0.004485988523811102,
+      "rewards/margins": 0.00730940094217658,
+      "rewards/rejected": -0.002823411952704191,
       "step": 30
     },
     {
-      "debug/losses": 0.3552504777908325,
-      "debug/policy_weights": 0.52684086561203,
-      "debug/raw_losses": 0.6755487322807312,
+      "debug/losses": 0.35061347484588623,
+      "debug/policy_weights": 0.5189496278762817,
+      "debug/raw_losses": 0.6759483218193054,
       "epoch": 0.08,
+      "grad_norm": 4.841334528975048,
       "learning_rate": 4.1666666666666667e-07,
-      "logits/chosen": -2.7985873222351074,
-      "logits/rejected": -2.7698562145233154,
-      "logps/chosen": -298.9928283691406,
-      "logps/rejected": -264.44781494140625,
-      "loss": 0.3544,
-      "rewards/accuracies": 0.7124999761581421,
-      "rewards/chosen": 0.005288169719278812,
-      "rewards/margins": 0.03713225945830345,
-      "rewards/rejected": -0.031844086945056915,
+      "logits/chosen": -2.7438368797302246,
+      "logits/rejected": -2.7376155853271484,
+      "logps/chosen": -252.50436401367188,
+      "logps/rejected": -240.8105010986328,
+      "loss": 0.3573,
+      "rewards/accuracies": 0.762499988079071,
+      "rewards/chosen": 0.015405787155032158,
+      "rewards/margins": 0.03552238270640373,
+      "rewards/rejected": -0.020116599276661873,
       "step": 40
     },
     {
-      "debug/losses": 0.3372410833835602,
-      "debug/policy_weights": 0.50788414478302,
-      "debug/raw_losses": 0.6635575890541077,
+      "debug/losses": 0.34173911809921265,
+      "debug/policy_weights": 0.5125550627708435,
+      "debug/raw_losses": 0.666517436504364,
       "epoch": 0.1,
+      "grad_norm": 5.135483442753442,
       "learning_rate": 4.999733114418725e-07,
-      "logits/chosen": -2.719874143600464,
-      "logits/rejected": -2.698538064956665,
-      "logps/chosen": -279.17694091796875,
-      "logps/rejected": -272.10687255859375,
-      "loss": 0.3418,
-      "rewards/accuracies": 0.65625,
-      "rewards/chosen": -0.017464958131313324,
-      "rewards/margins": 0.06663360446691513,
-      "rewards/rejected": -0.08409856259822845,
+      "logits/chosen": -2.6830456256866455,
+      "logits/rejected": -2.6514101028442383,
+      "logps/chosen": -269.3827819824219,
+      "logps/rejected": -287.2772216796875,
+      "loss": 0.3437,
+      "rewards/accuracies": 0.6812499761581421,
+      "rewards/chosen": 0.0019381232559680939,
+      "rewards/margins": 0.05897800251841545,
+      "rewards/rejected": -0.05703987926244736,
       "step": 50
     },
     {
-      "debug/losses": 0.32032984495162964,
-      "debug/policy_weights": 0.49015456438064575,
-      "debug/raw_losses": 0.6468743085861206,
+      "debug/losses": 0.3377828598022461,
+      "debug/policy_weights": 0.5224286317825317,
+      "debug/raw_losses": 0.6448591947555542,
       "epoch": 0.13,
+      "grad_norm": 5.5606740216520905,
       "learning_rate": 4.990398100856366e-07,
-      "logits/chosen": -2.741781234741211,
-      "logits/rejected": -2.718924045562744,
-      "logps/chosen": -264.25994873046875,
-      "logps/rejected": -249.49978637695312,
-      "loss": 0.3195,
-      "rewards/accuracies": 0.6499999761581421,
-      "rewards/chosen": -0.05183681100606918,
-      "rewards/margins": 0.11701079457998276,
-      "rewards/rejected": -0.16884759068489075,
+      "logits/chosen": -2.745028018951416,
+      "logits/rejected": -2.713804244995117,
+      "logps/chosen": -266.05279541015625,
+      "logps/rejected": -305.28271484375,
+      "loss": 0.3305,
+      "rewards/accuracies": 0.6937500238418579,
+      "rewards/chosen": -0.02193317376077175,
+      "rewards/margins": 0.1177050918340683,
+      "rewards/rejected": -0.1396382749080658,
       "step": 60
     },
     {
-      "debug/losses": 0.2948753237724304,
-      "debug/policy_weights": 0.4617387652397156,
-      "debug/raw_losses": 0.6441112756729126,
+      "debug/losses": 0.28029459714889526,
+      "debug/policy_weights": 0.47101029753685,
+      "debug/raw_losses": 0.6030661463737488,
       "epoch": 0.15,
+      "grad_norm": 6.284287160526739,
       "learning_rate": 4.967775735898179e-07,
-      "logits/chosen": -2.7274184226989746,
-      "logits/rejected": -2.697922706604004,
-      "logps/chosen": -283.56719970703125,
-      "logps/rejected": -265.3944091796875,
-      "loss": 0.2944,
-      "rewards/accuracies": 0.6499999761581421,
-      "rewards/chosen": -0.15144629776477814,
-      "rewards/margins": 0.15366096794605255,
-      "rewards/rejected": -0.3051072657108307,
+      "logits/chosen": -2.7072081565856934,
+      "logits/rejected": -2.7113521099090576,
+      "logps/chosen": -272.530029296875,
+      "logps/rejected": -289.87713623046875,
+      "loss": 0.2826,
+      "rewards/accuracies": 0.7875000238418579,
+      "rewards/chosen": -0.1557193100452423,
+      "rewards/margins": 0.24685576558113098,
+      "rewards/rejected": -0.4025750756263733,
       "step": 70
     },
     {
-      "debug/losses": 0.24030272662639618,
-      "debug/policy_weights": 0.39300116896629333,
-      "debug/raw_losses": 0.61235511302948,
+      "debug/losses": 0.26233458518981934,
+      "debug/policy_weights": 0.42553257942199707,
+      "debug/raw_losses": 0.6218484044075012,
       "epoch": 0.17,
+      "grad_norm": 6.8893773770621305,
       "learning_rate": 4.931986719649298e-07,
-      "logits/chosen": -2.6943717002868652,
-      "logits/rejected": -2.6917672157287598,
-      "logps/chosen": -306.8209533691406,
-      "logps/rejected": -331.4168701171875,
-      "loss": 0.2366,
+      "logits/chosen": -2.800093412399292,
+      "logits/rejected": -2.7768216133117676,
+      "logps/chosen": -360.3621520996094,
+      "logps/rejected": -329.7616882324219,
+      "loss": 0.2516,
       "rewards/accuracies": 0.6875,
-      "rewards/chosen": -0.36089134216308594,
-      "rewards/margins": 0.27803000807762146,
-      "rewards/rejected": -0.638921320438385,
+      "rewards/chosen": -0.45188745856285095,
+      "rewards/margins": 0.26955321431159973,
+      "rewards/rejected": -0.7214406728744507,
       "step": 80
     },
     {
-      "debug/losses": 0.21284589171409607,
-      "debug/policy_weights": 0.36208364367485046,
-      "debug/raw_losses": 0.6084356307983398,
+      "debug/losses": 0.18998686969280243,
+      "debug/policy_weights": 0.34327536821365356,
+      "debug/raw_losses": 0.5527450442314148,
       "epoch": 0.19,
+      "grad_norm": 6.5870040213670435,
       "learning_rate": 4.883222001996351e-07,
-      "logits/chosen": -2.742755889892578,
-      "logits/rejected": -2.7089426517486572,
-      "logps/chosen": -306.11700439453125,
-      "logps/rejected": -313.5643005371094,
-      "loss": 0.2153,
-      "rewards/accuracies": 0.675000011920929,
-      "rewards/chosen": -0.5115253925323486,
-      "rewards/margins": 0.27717387676239014,
-      "rewards/rejected": -0.7886992692947388,
+      "logits/chosen": -2.724050998687744,
+      "logits/rejected": -2.690390110015869,
+      "logps/chosen": -303.0259704589844,
+      "logps/rejected": -344.49505615234375,
+      "loss": 0.2198,
+      "rewards/accuracies": 0.768750011920929,
+      "rewards/chosen": -0.5087189078330994,
+      "rewards/margins": 0.4743797779083252,
+      "rewards/rejected": -0.9830986857414246,
       "step": 90
     },
     {
-      "debug/losses": 0.23982000350952148,
-      "debug/policy_weights": 0.39243510365486145,
-      "debug/raw_losses": 0.625305712223053,
+      "debug/losses": 0.16577747464179993,
+      "debug/policy_weights": 0.29914408922195435,
+      "debug/raw_losses": 0.5552432537078857,
       "epoch": 0.21,
+      "grad_norm": 7.249323730073561,
       "learning_rate": 4.821741763807186e-07,
-      "logits/chosen": -2.754852533340454,
-      "logits/rejected": -2.7216262817382812,
-      "logps/chosen": -346.60498046875,
-      "logps/rejected": -359.0435485839844,
-      "loss": 0.2168,
-      "rewards/accuracies": 0.637499988079071,
-      "rewards/chosen": -0.5716595649719238,
-      "rewards/margins": 0.3401753306388855,
-      "rewards/rejected": -0.9118350148200989,
+      "logits/chosen": -2.6653432846069336,
+      "logits/rejected": -2.658141613006592,
+      "logps/chosen": -325.9325866699219,
+      "logps/rejected": -375.12127685546875,
+      "loss": 0.1781,
+      "rewards/accuracies": 0.706250011920929,
+      "rewards/chosen": -0.7297804951667786,
+      "rewards/margins": 0.511761486530304,
+      "rewards/rejected": -1.2415419816970825,
       "step": 100
     },
     {
       "epoch": 0.21,
-      "eval_debug/losses": 0.2082262486219406,
-      "eval_debug/policy_weights": 0.37346428632736206,
-      "eval_debug/raw_losses": 0.5528886318206787,
-      "eval_logits/chosen": -2.697880268096924,
-      "eval_logits/rejected": -2.6826982498168945,
-      "eval_logps/chosen": -311.4377136230469,
-      "eval_logps/rejected": -363.1571350097656,
-      "eval_loss": 0.21503373980522156,
-      "eval_rewards/accuracies": 0.73828125,
-      "eval_rewards/chosen": -0.5439806580543518,
-      "eval_rewards/margins": 0.5140582323074341,
-      "eval_rewards/rejected": -1.0580389499664307,
-      "eval_runtime": 53.0291,
-      "eval_samples_per_second": 37.715,
-      "eval_steps_per_second": 0.603,
+      "eval_debug/losses": 0.19525206089019775,
+      "eval_debug/policy_weights": 0.3511899709701538,
+      "eval_debug/raw_losses": 0.5590351819992065,
+      "eval_logits/chosen": -2.70804500579834,
+      "eval_logits/rejected": -2.6910133361816406,
+      "eval_logps/chosen": -319.9805603027344,
+      "eval_logps/rejected": -373.18670654296875,
+      "eval_loss": 0.20066915452480316,
+      "eval_rewards/accuracies": 0.734375,
+      "eval_rewards/chosen": -0.6478448510169983,
+      "eval_rewards/margins": 0.5214487910270691,
+      "eval_rewards/rejected": -1.1692936420440674,
+      "eval_runtime": 43.2006,
+      "eval_samples_per_second": 46.296,
+      "eval_steps_per_second": 0.741,
       "step": 100
     },
     {
-      "debug/losses": 0.1748097836971283,
-      "debug/policy_weights": 0.3250483572483063,
-      "debug/raw_losses": 0.5473231077194214,
+      "debug/losses": 0.1987285315990448,
+      "debug/policy_weights": 0.34678488969802856,
+      "debug/raw_losses": 0.5721240043640137,
       "epoch": 0.23,
+      "grad_norm": 6.9123559213202554,
       "learning_rate": 4.747874028753375e-07,
-      "logits/chosen": -2.5498530864715576,
-      "logits/rejected": -2.5195746421813965,
-      "logps/chosen": -338.24639892578125,
-      "logps/rejected": -351.0300598144531,
-      "loss": 0.2041,
-      "rewards/accuracies": 0.6875,
-      "rewards/chosen": -0.6814893484115601,
-      "rewards/margins": 0.6107165813446045,
-      "rewards/rejected": -1.292205810546875,
+      "logits/chosen": -2.7045469284057617,
+      "logits/rejected": -2.672999858856201,
+      "logps/chosen": -369.9187927246094,
+      "logps/rejected": -373.6562194824219,
+      "loss": 0.1986,
+      "rewards/accuracies": 0.706250011920929,
+      "rewards/chosen": -0.7313551902770996,
+      "rewards/margins": 0.47859472036361694,
+      "rewards/rejected": -1.2099498510360718,
       "step": 110
     },
     {
-      "debug/losses": 0.1997809112071991,
-      "debug/policy_weights": 0.35038530826568604,
-      "debug/raw_losses": 0.5690494775772095,
+      "debug/losses": 0.18334314227104187,
+      "debug/policy_weights": 0.3257436156272888,
+      "debug/raw_losses": 0.5537685751914978,
       "epoch": 0.25,
+      "grad_norm": 7.316158832496888,
       "learning_rate": 4.662012913161997e-07,
-      "logits/chosen": -2.6230885982513428,
-      "logits/rejected": -2.5906565189361572,
-      "logps/chosen": -337.9828186035156,
-      "logps/rejected": -383.8708801269531,
-      "loss": 0.1885,
-      "rewards/accuracies": 0.7250000238418579,
-      "rewards/chosen": -0.6913173794746399,
-      "rewards/margins": 0.5188819169998169,
-      "rewards/rejected": -1.2101994752883911,
+      "logits/chosen": -2.6026859283447266,
+      "logits/rejected": -2.600971221923828,
+      "logps/chosen": -346.50555419921875,
+      "logps/rejected": -379.53887939453125,
+      "loss": 0.1846,
+      "rewards/accuracies": 0.731249988079071,
+      "rewards/chosen": -0.7285534739494324,
+      "rewards/margins": 0.5237767100334167,
+      "rewards/rejected": -1.2523301839828491,
       "step": 120
     },
     {
-      "debug/losses": 0.17902129888534546,
-      "debug/policy_weights": 0.3178775906562805,
-      "debug/raw_losses": 0.5685083866119385,
+      "debug/losses": 0.17840833961963654,
+      "debug/policy_weights": 0.3271704614162445,
+      "debug/raw_losses": 0.5529276132583618,
       "epoch": 0.27,
+      "grad_norm": 6.11630399118371,
       "learning_rate": 4.5646165232345103e-07,
-      "logits/chosen": -2.591015100479126,
-      "logits/rejected": -2.576317548751831,
-      "logps/chosen": -350.18768310546875,
-      "logps/rejected": -370.1181640625,
-      "loss": 0.1792,
-      "rewards/accuracies": 0.6937500238418579,
-      "rewards/chosen": -0.7569286227226257,
-      "rewards/margins": 0.4377259314060211,
-      "rewards/rejected": -1.1946544647216797,
+      "logits/chosen": -2.5919995307922363,
+      "logits/rejected": -2.598027467727661,
+      "logps/chosen": -346.44610595703125,
+      "logps/rejected": -391.03778076171875,
+      "loss": 0.1758,
+      "rewards/accuracies": 0.731249988079071,
+      "rewards/chosen": -0.7799400687217712,
+      "rewards/margins": 0.5067718029022217,
+      "rewards/rejected": -1.2867119312286377,
       "step": 130
     },
     {
-      "debug/losses": 0.18843333423137665,
-      "debug/policy_weights": 0.3299049437046051,
-      "debug/raw_losses": 0.550617516040802,
+      "debug/losses": 0.18315494060516357,
+      "debug/policy_weights": 0.3452090620994568,
+      "debug/raw_losses": 0.56131511926651,
       "epoch": 0.29,
+      "grad_norm": 6.845717287191781,
       "learning_rate": 4.456204510851956e-07,
-      "logits/chosen": -2.570383071899414,
-      "logits/rejected": -2.5405478477478027,
-      "logps/chosen": -350.4234924316406,
-      "logps/rejected": -398.32403564453125,
-      "loss": 0.1848,
+      "logits/chosen": -2.572934627532959,
+      "logits/rejected": -2.5713508129119873,
+      "logps/chosen": -390.03253173828125,
+      "logps/rejected": -440.16461181640625,
+      "loss": 0.1668,
       "rewards/accuracies": 0.7250000238418579,
-      "rewards/chosen": -0.69789057970047,
-      "rewards/margins": 0.577714741230011,
-      "rewards/rejected": -1.275605320930481,
+      "rewards/chosen": -0.8578819036483765,
+      "rewards/margins": 0.5912094712257385,
+      "rewards/rejected": -1.4490914344787598,
       "step": 140
     },
     {
-      "debug/losses": 0.18008050322532654,
-      "debug/policy_weights": 0.34716594219207764,
-      "debug/raw_losses": 0.5200980305671692,
+      "debug/losses": 0.19362398982048035,
+      "debug/policy_weights": 0.33687275648117065,
+      "debug/raw_losses": 0.559166669845581,
       "epoch": 0.31,
+      "grad_norm": 7.029648911332738,
       "learning_rate": 4.337355301007335e-07,
-      "logits/chosen": -2.520066738128662,
-      "logits/rejected": -2.5489494800567627,
-      "logps/chosen": -280.36163330078125,
-      "logps/rejected": -348.47869873046875,
-      "loss": 0.1859,
-      "rewards/accuracies": 0.793749988079071,
-      "rewards/chosen": -0.5368272066116333,
-      "rewards/margins": 0.585098147392273,
-      "rewards/rejected": -1.1219253540039062,
+      "logits/chosen": -2.512932300567627,
+      "logits/rejected": -2.4797143936157227,
+      "logps/chosen": -357.82464599609375,
+      "logps/rejected": -402.2618713378906,
+      "loss": 0.1771,
+      "rewards/accuracies": 0.7124999761581421,
+      "rewards/chosen": -0.8432974815368652,
+      "rewards/margins": 0.48099932074546814,
+      "rewards/rejected": -1.3242968320846558,
       "step": 150
     },
     {
-      "debug/losses": 0.17896616458892822,
-      "debug/policy_weights": 0.3156106472015381,
-      "debug/raw_losses": 0.599802553653717,
+      "debug/losses": 0.14652669429779053,
+      "debug/policy_weights": 0.28093475103378296,
+      "debug/raw_losses": 0.5318498611450195,
       "epoch": 0.33,
+      "grad_norm": 6.358610241024355,
       "learning_rate": 4.2087030056579986e-07,
-      "logits/chosen": -2.5032947063446045,
-      "logits/rejected": -2.476680278778076,
-      "logps/chosen": -341.430908203125,
-      "logps/rejected": -376.94244384765625,
-      "loss": 0.1759,
-      "rewards/accuracies": 0.675000011920929,
-      "rewards/chosen": -0.8545902967453003,
-      "rewards/margins": 0.4560604691505432,
-      "rewards/rejected": -1.3106508255004883,
+      "logits/chosen": -2.464831590652466,
+      "logits/rejected": -2.4293084144592285,
+      "logps/chosen": -370.14971923828125,
+      "logps/rejected": -408.4079284667969,
+      "loss": 0.1583,
+      "rewards/accuracies": 0.7124999761581421,
+      "rewards/chosen": -1.0606319904327393,
+      "rewards/margins": 0.596233606338501,
+      "rewards/rejected": -1.6568657159805298,
       "step": 160
     },
     {
-      "debug/losses": 0.12902560830116272,
-      "debug/policy_weights": 0.25470516085624695,
-      "debug/raw_losses": 0.5344475507736206,
+      "debug/losses": 0.1372167319059372,
+      "debug/policy_weights": 0.2504947781562805,
+      "debug/raw_losses": 0.6000548005104065,
       "epoch": 0.36,
+      "grad_norm": 7.75801911972225,
       "learning_rate": 4.070934040463998e-07,
-      "logits/chosen": -2.3815720081329346,
-      "logits/rejected": -2.365286350250244,
-      "logps/chosen": -394.91192626953125,
-      "logps/rejected": -446.85321044921875,
-      "loss": 0.1496,
-      "rewards/accuracies": 0.7437499761581421,
-      "rewards/chosen": -1.1274508237838745,
-      "rewards/margins": 0.7393988966941833,
-      "rewards/rejected": -1.866849660873413,
+      "logits/chosen": -2.361786365509033,
+      "logits/rejected": -2.332592010498047,
+      "logps/chosen": -389.86767578125,
+      "logps/rejected": -423.97698974609375,
+      "loss": 0.1417,
+      "rewards/accuracies": 0.675000011920929,
+      "rewards/chosen": -1.4737131595611572,
+      "rewards/margins": 0.511743426322937,
+      "rewards/rejected": -1.9854564666748047,
       "step": 170
     },
     {
-      "debug/losses": 0.12488824129104614,
-      "debug/policy_weights": 0.24605941772460938,
-      "debug/raw_losses": 0.5098173022270203,
+      "debug/losses": 0.11748126894235611,
+      "debug/policy_weights": 0.2296670377254486,
+      "debug/raw_losses": 0.5410301089286804,
       "epoch": 0.38,
+      "grad_norm": 6.142786182484198,
       "learning_rate": 3.9247834624635404e-07,
-      "logits/chosen": -2.3833303451538086,
-      "logits/rejected": -2.389314889907837,
-      "logps/chosen": -389.9188232421875,
-      "logps/rejected": -450.25140380859375,
-      "loss": 0.1436,
-      "rewards/accuracies": 0.768750011920929,
-      "rewards/chosen": -1.2074089050292969,
-      "rewards/margins": 0.7412688136100769,
-      "rewards/rejected": -1.9486777782440186,
+      "logits/chosen": -2.3254575729370117,
+      "logits/rejected": -2.3115153312683105,
+      "logps/chosen": -391.5261535644531,
+      "logps/rejected": -429.21875,
+      "loss": 0.1327,
+      "rewards/accuracies": 0.706250011920929,
+      "rewards/chosen": -1.446401834487915,
+      "rewards/margins": 0.6310542821884155,
+      "rewards/rejected": -2.07745623588562,
       "step": 180
     },
     {
-      "debug/losses": 0.1292407363653183,
-      "debug/policy_weights": 0.26770901679992676,
-      "debug/raw_losses": 0.48357778787612915,
+      "debug/losses": 0.16298121213912964,
+      "debug/policy_weights": 0.2831325829029083,
+      "debug/raw_losses": 0.5709558725357056,
       "epoch": 0.4,
+      "grad_norm": 6.941055961208667,
       "learning_rate": 3.7710310482256523e-07,
-      "logits/chosen": -2.3599636554718018,
-      "logits/rejected": -2.3223559856414795,
-      "logps/chosen": -414.10699462890625,
-      "logps/rejected": -469.5603942871094,
-      "loss": 0.1421,
-      "rewards/accuracies": 0.793749988079071,
-      "rewards/chosen": -1.227112054824829,
-      "rewards/margins": 0.8277386426925659,
-      "rewards/rejected": -2.0548505783081055,
+      "logits/chosen": -2.378763198852539,
+      "logits/rejected": -2.3435521125793457,
+      "logps/chosen": -381.296630859375,
+      "logps/rejected": -435.13494873046875,
+      "loss": 0.1447,
+      "rewards/accuracies": 0.675000011920929,
+      "rewards/chosen": -1.178696870803833,
+      "rewards/margins": 0.6175613403320312,
+      "rewards/rejected": -1.7962583303451538,
       "step": 190
     },
     {
-      "debug/losses": 0.15382704138755798,
-      "debug/policy_weights": 0.2786335349082947,
-      "debug/raw_losses": 0.5533261299133301,
+      "debug/losses": 0.1531984657049179,
+      "debug/policy_weights": 0.27346447110176086,
+      "debug/raw_losses": 0.5578541159629822,
       "epoch": 0.42,
+      "grad_norm": 9.54639374222291,
       "learning_rate": 3.610497133404795e-07,
-      "logits/chosen": -2.2979869842529297,
-      "logits/rejected": -2.312802791595459,
-      "logps/chosen": -419.5997009277344,
-      "logps/rejected": -456.7002868652344,
-      "loss": 0.1396,
+      "logits/chosen": -2.3584351539611816,
+      "logits/rejected": -2.362753391265869,
+      "logps/chosen": -359.6307067871094,
+      "logps/rejected": -422.2249450683594,
+      "loss": 0.1616,
       "rewards/accuracies": 0.731249988079071,
-      "rewards/chosen": -1.2343220710754395,
-      "rewards/margins": 0.6895232200622559,
-      "rewards/rejected": -1.9238452911376953,
+      "rewards/chosen": -1.0809690952301025,
+      "rewards/margins": 0.6879221796989441,
+      "rewards/rejected": -1.7688913345336914,
       "step": 200
     },
     {
       "epoch": 0.42,
-      "eval_debug/losses": 0.13899114727973938,
-      "eval_debug/policy_weights": 0.2686985731124878,
-      "eval_debug/raw_losses": 0.5029721260070801,
-      "eval_logits/chosen": -2.2968499660491943,
-      "eval_logits/rejected": -2.273340940475464,
-      "eval_logps/chosen": -391.8349609375,
-      "eval_logps/rejected": -470.2158203125,
-      "eval_loss": 0.14160528779029846,
-      "eval_rewards/accuracies": 0.765625,
-      "eval_rewards/chosen": -1.3479530811309814,
-      "eval_rewards/margins": 0.7806724309921265,
-      "eval_rewards/rejected": -2.1286253929138184,
-      "eval_runtime": 52.9895,
-      "eval_samples_per_second": 37.743,
-      "eval_steps_per_second": 0.604,
+      "eval_debug/losses": 0.16714033484458923,
+      "eval_debug/policy_weights": 0.3173619508743286,
+      "eval_debug/raw_losses": 0.5079437494277954,
+      "eval_logits/chosen": -2.44987154006958,
+      "eval_logits/rejected": -2.4276504516601562,
+      "eval_logps/chosen": -343.49139404296875,
+      "eval_logps/rejected": -416.28436279296875,
+      "eval_loss": 0.16694626212120056,
+      "eval_rewards/accuracies": 0.7109375,
+      "eval_rewards/chosen": -0.8829529881477356,
+      "eval_rewards/margins": 0.7173169851303101,
+      "eval_rewards/rejected": -1.6002700328826904,
+      "eval_runtime": 43.2126,
+      "eval_samples_per_second": 46.283,
+      "eval_steps_per_second": 0.741,
       "step": 200
     },
     {
-      "debug/losses": 0.12412895262241364,
-      "debug/policy_weights": 0.26891231536865234,
-      "debug/raw_losses": 0.47236162424087524,
+      "debug/losses": 0.14734311401844025,
+      "debug/policy_weights": 0.2870820164680481,
+      "debug/raw_losses": 0.5041711330413818,
       "epoch": 0.44,
+      "grad_norm": 7.104709810051969,
       "learning_rate": 3.4440382358952115e-07,
-      "logits/chosen": -2.2376811504364014,
-      "logits/rejected": -2.2135262489318848,
-      "logps/chosen": -395.38421630859375,
-      "logps/rejected": -459.84210205078125,
-      "loss": 0.1418,
-      "rewards/accuracies": 0.762499988079071,
-      "rewards/chosen": -1.2235102653503418,
-      "rewards/margins": 0.8555533289909363,
-      "rewards/rejected": -2.0790634155273438,
+      "logits/chosen": -2.3574612140655518,
+      "logits/rejected": -2.326927900314331,
+      "logps/chosen": -381.2717590332031,
+      "logps/rejected": -413.890869140625,
+      "loss": 0.1636,
+      "rewards/accuracies": 0.7437499761581421,
+      "rewards/chosen": -0.9982665777206421,
+      "rewards/margins": 0.6929588317871094,
+      "rewards/rejected": -1.6912254095077515,
       "step": 210
     },
     {
-      "debug/losses": 0.13483984768390656,
-      "debug/policy_weights": 0.2588108479976654,
-      "debug/raw_losses": 0.5093666911125183,
+      "debug/losses": 0.13881739974021912,
+      "debug/policy_weights": 0.27268368005752563,
+      "debug/raw_losses": 0.5210752487182617,
       "epoch": 0.46,
+      "grad_norm": 8.340698654196919,
       "learning_rate": 3.272542485937368e-07,
-      "logits/chosen": -2.2282791137695312,
-      "logits/rejected": -2.1731905937194824,
-      "logps/chosen": -391.88104248046875,
-      "logps/rejected": -457.2334899902344,
-      "loss": 0.1418,
-      "rewards/accuracies": 0.7562500238418579,
-      "rewards/chosen": -1.1691535711288452,
-      "rewards/margins": 0.7801142930984497,
-      "rewards/rejected": -1.9492677450180054,
+      "logits/chosen": -2.3073763847351074,
+      "logits/rejected": -2.288245677947998,
+      "logps/chosen": -370.5925598144531,
+      "logps/rejected": -423.53021240234375,
+      "loss": 0.1549,
+      "rewards/accuracies": 0.75,
+      "rewards/chosen": -1.1713535785675049,
+      "rewards/margins": 0.7361623644828796,
+      "rewards/rejected": -1.9075158834457397,
       "step": 220
     },
     {
-      "debug/losses": 0.14488555490970612,
-      "debug/policy_weights": 0.26247432827949524,
-      "debug/raw_losses": 0.5395208597183228,
+      "debug/losses": 0.15816743671894073,
+      "debug/policy_weights": 0.2902621626853943,
+      "debug/raw_losses": 0.509421706199646,
       "epoch": 0.48,
+      "grad_norm": 7.2612463171873225,
       "learning_rate": 3.096924887558854e-07,
-      "logits/chosen": -2.2395882606506348,
-      "logits/rejected": -2.1690821647644043,
-      "logps/chosen": -417.9415588378906,
-      "logps/rejected": -489.2323303222656,
-      "loss": 0.1451,
-      "rewards/accuracies": 0.699999988079071,
-      "rewards/chosen": -1.433936595916748,
-      "rewards/margins": 0.758230984210968,
-      "rewards/rejected": -2.1921677589416504,
+      "logits/chosen": -2.339087724685669,
+      "logits/rejected": -2.31463360786438,
+      "logps/chosen": -342.2999572753906,
+      "logps/rejected": -425.86181640625,
+      "loss": 0.169,
+      "rewards/accuracies": 0.731249988079071,
+      "rewards/chosen": -1.0157617330551147,
+      "rewards/margins": 0.8210276365280151,
+      "rewards/rejected": -1.8367893695831299,
       "step": 230
     },
     {
-      "debug/losses": 0.1316806524991989,
-      "debug/policy_weights": 0.2545274794101715,
-      "debug/raw_losses": 0.5126517415046692,
+      "debug/losses": 0.15576457977294922,
+      "debug/policy_weights": 0.29614073038101196,
+      "debug/raw_losses": 0.4828997552394867,
       "epoch": 0.5,
+      "grad_norm": 7.345094511114747,
       "learning_rate": 2.9181224366319943e-07,
-      "logits/chosen": -2.171607732772827,
-      "logits/rejected": -2.1234748363494873,
-      "logps/chosen": -426.31719970703125,
-      "logps/rejected": -475.4518127441406,
-      "loss": 0.1325,
-      "rewards/accuracies": 0.75,
-      "rewards/chosen": -1.4014716148376465,
-      "rewards/margins": 0.8200809359550476,
-      "rewards/rejected": -2.2215523719787598,
+      "logits/chosen": -2.3611202239990234,
+      "logits/rejected": -2.317119836807251,
+      "logps/chosen": -368.0779113769531,
+      "logps/rejected": -434.31549072265625,
+      "loss": 0.1514,
+      "rewards/accuracies": 0.768750011920929,
+      "rewards/chosen": -1.1054710149765015,
+      "rewards/margins": 0.8346185684204102,
+      "rewards/rejected": -1.9400895833969116,
       "step": 240
     },
     {
-      "debug/losses": 0.14006611704826355,
-      "debug/policy_weights": 0.2591710388660431,
-      "debug/raw_losses": 0.5314095616340637,
+      "debug/losses": 0.12131881713867188,
+      "debug/policy_weights": 0.2461313009262085,
+      "debug/raw_losses": 0.46923789381980896,
       "epoch": 0.52,
+      "grad_norm": 6.902697741869489,
       "learning_rate": 2.7370891215954565e-07,
-      "logits/chosen": -2.1962928771972656,
-      "logits/rejected": -2.1328587532043457,
-      "logps/chosen": -403.8672790527344,
-      "logps/rejected": -458.2118225097656,
-      "loss": 0.1446,
-      "rewards/accuracies": 0.762499988079071,
-      "rewards/chosen": -1.3229501247406006,
-      "rewards/margins": 0.7149368524551392,
-      "rewards/rejected": -2.03788685798645,
+      "logits/chosen": -2.2654645442962646,
+      "logits/rejected": -2.2106964588165283,
+      "logps/chosen": -434.3672790527344,
+      "logps/rejected": -491.00750732421875,
+      "loss": 0.1285,
+      "rewards/accuracies": 0.78125,
+      "rewards/chosen": -1.4400012493133545,
+      "rewards/margins": 0.9503369331359863,
+      "rewards/rejected": -2.390338182449341,
       "step": 250
     },
     {
-      "debug/losses": 0.14360225200653076,
-      "debug/policy_weights": 0.26045817136764526,
-      "debug/raw_losses": 0.5265286564826965,
+      "debug/losses": 0.1332876831293106,
+      "debug/policy_weights": 0.27705293893814087,
+      "debug/raw_losses": 0.5086624622344971,
       "epoch": 0.54,
+      "grad_norm": 7.179715362550834,
       "learning_rate": 2.55479083351317e-07,
-      "logits/chosen": -2.1760964393615723,
-      "logits/rejected": -2.150712013244629,
-      "logps/chosen": -401.54754638671875,
-      "logps/rejected": -479.643310546875,
-      "loss": 0.1352,
-      "rewards/accuracies": 0.7124999761581421,
-      "rewards/chosen": -1.4240009784698486,
-      "rewards/margins": 0.7899759411811829,
-      "rewards/rejected": -2.2139768600463867,
+      "logits/chosen": -2.316641330718994,
+      "logits/rejected": -2.3014323711395264,
+      "logps/chosen": -410.114990234375,
+      "logps/rejected": -454.8689880371094,
+      "loss": 0.1406,
+      "rewards/accuracies": 0.762499988079071,
+      "rewards/chosen": -1.2414360046386719,
+      "rewards/margins": 0.8403455018997192,
+      "rewards/rejected": -2.0817813873291016,
       "step": 260
     },
     {
-      "debug/losses": 0.11131677776575089,
-      "debug/policy_weights": 0.24807500839233398,
-      "debug/raw_losses": 0.47175589203834534,
+      "debug/losses": 0.168529212474823,
+      "debug/policy_weights": 0.2988266348838806,
+      "debug/raw_losses": 0.5682334899902344,
       "epoch": 0.56,
+      "grad_norm": 7.216808591858915,
       "learning_rate": 2.3722002126275822e-07,
-      "logits/chosen": -2.186707019805908,
-      "logits/rejected": -2.1590161323547363,
-      "logps/chosen": -427.5020446777344,
-      "logps/rejected": -495.8419494628906,
-      "loss": 0.1267,
-      "rewards/accuracies": 0.75,
-      "rewards/chosen": -1.3904914855957031,
-      "rewards/margins": 0.9172846078872681,
-      "rewards/rejected": -2.3077759742736816,
+      "logits/chosen": -2.3377718925476074,
+      "logits/rejected": -2.3152408599853516,
+      "logps/chosen": -379.34661865234375,
+      "logps/rejected": -430.7207946777344,
+      "loss": 0.1519,
+      "rewards/accuracies": 0.675000011920929,
+      "rewards/chosen": -1.0926826000213623,
+      "rewards/margins": 0.6951464414596558,
+      "rewards/rejected": -1.7878293991088867,
       "step": 270
     },
     {
-      "debug/losses": 0.11321704089641571,
-      "debug/policy_weights": 0.24243195354938507,
-      "debug/raw_losses": 0.4841908812522888,
+      "debug/losses": 0.15156733989715576,
+      "debug/policy_weights": 0.2692407965660095,
+      "debug/raw_losses": 0.5413981080055237,
       "epoch": 0.59,
+      "grad_norm": 9.464551455783996,
       "learning_rate": 2.19029145890313e-07,
-      "logits/chosen": -2.1000428199768066,
-      "logits/rejected": -2.089947462081909,
-      "logps/chosen": -469.45867919921875,
-      "logps/rejected": -520.9786987304688,
-      "loss": 0.1249,
-      "rewards/accuracies": 0.8187500238418579,
-      "rewards/chosen": -1.617157220840454,
-      "rewards/margins": 0.9369556307792664,
-      "rewards/rejected": -2.554112672805786,
+      "logits/chosen": -2.295574426651001,
+      "logits/rejected": -2.2587995529174805,
+      "logps/chosen": -383.4193420410156,
+      "logps/rejected": -444.91680908203125,
+      "loss": 0.1498,
+      "rewards/accuracies": 0.731249988079071,
+      "rewards/chosen": -1.302856683731079,
+      "rewards/margins": 0.7735105156898499,
+      "rewards/rejected": -2.076367139816284,
       "step": 280
     },
     {
-      "debug/losses": 0.11086218059062958,
-      "debug/policy_weights": 0.22184400260448456,
-      "debug/raw_losses": 0.5123878717422485,
+      "debug/losses": 0.14380133152008057,
+      "debug/policy_weights": 0.26454827189445496,
+      "debug/raw_losses": 0.5405210256576538,
       "epoch": 0.61,
+      "grad_norm": 7.542791313020671,
       "learning_rate": 2.0100351342479216e-07,
-      "logits/chosen": -2.1559338569641113,
-      "logits/rejected": -2.1183762550354004,
-      "logps/chosen": -447.17138671875,
-      "logps/rejected": -499.66143798828125,
-      "loss": 0.1214,
-      "rewards/accuracies": 0.71875,
-      "rewards/chosen": -1.6940768957138062,
-      "rewards/margins": 0.7695088386535645,
-      "rewards/rejected": -2.46358585357666,
+      "logits/chosen": -2.3062846660614014,
+      "logits/rejected": -2.2651028633117676,
+      "logps/chosen": -386.890625,
+      "logps/rejected": -450.9478454589844,
+      "loss": 0.1352,
+      "rewards/accuracies": 0.737500011920929,
+      "rewards/chosen": -1.4489481449127197,
+      "rewards/margins": 0.7288294434547424,
+      "rewards/rejected": -2.1777775287628174,
       "step": 290
     },
     {
-      "debug/losses": 0.13476888835430145,
-      "debug/policy_weights": 0.23854057490825653,
-      "debug/raw_losses": 0.5516811013221741,
+      "debug/losses": 0.14622345566749573,
+      "debug/policy_weights": 0.2636463940143585,
+      "debug/raw_losses": 0.536266028881073,
       "epoch": 0.63,
+      "grad_norm": 7.687631612230176,
       "learning_rate": 1.8323929841460178e-07,
-      "logits/chosen": -2.199371099472046,
-      "logits/rejected": -2.178723096847534,
-      "logps/chosen": -422.80450439453125,
-      "logps/rejected": -485.37890625,
-      "loss": 0.1294,
-      "rewards/accuracies": 0.7250000238418579,
-      "rewards/chosen": -1.7119791507720947,
-      "rewards/margins": 0.7184675335884094,
-      "rewards/rejected": -2.4304463863372803,
+      "logits/chosen": -2.2653095722198486,
+      "logits/rejected": -2.2190821170806885,
+      "logps/chosen": -444.52960205078125,
+      "logps/rejected": -489.06817626953125,
+      "loss": 0.1343,
+      "rewards/accuracies": 0.699999988079071,
+      "rewards/chosen": -1.5472103357315063,
+      "rewards/margins": 0.8187928199768066,
+      "rewards/rejected": -2.3660032749176025,
       "step": 300
     },
     {
       "epoch": 0.63,
-      "eval_debug/losses": 0.12840886414051056,
-      "eval_debug/policy_weights": 0.25453710556030273,
-      "eval_debug/raw_losses": 0.4935261309146881,
-      "eval_logits/chosen": -2.1884968280792236,
-      "eval_logits/rejected": -2.158949851989746,
-      "eval_logps/chosen": -417.07135009765625,
-      "eval_logps/rejected": -502.21124267578125,
-      "eval_loss": 0.13086578249931335,
-      "eval_rewards/accuracies": 0.73828125,
-      "eval_rewards/chosen": -1.6003175973892212,
-      "eval_rewards/margins": 0.8482623100280762,
-      "eval_rewards/rejected": -2.448579788208008,
-      "eval_runtime": 53.0489,
-      "eval_samples_per_second": 37.701,
-      "eval_steps_per_second": 0.603,
+      "eval_debug/losses": 0.13651318848133087,
+      "eval_debug/policy_weights": 0.26656585931777954,
+      "eval_debug/raw_losses": 0.49531808495521545,
+      "eval_logits/chosen": -2.2618136405944824,
+      "eval_logits/rejected": -2.228300094604492,
+      "eval_logps/chosen": -405.40423583984375,
+      "eval_logps/rejected": -493.411376953125,
+      "eval_loss": 0.13676287233829498,
+      "eval_rewards/accuracies": 0.7578125,
+      "eval_rewards/chosen": -1.5020815134048462,
+      "eval_rewards/margins": 0.8694581985473633,
+      "eval_rewards/rejected": -2.371539831161499,
+      "eval_runtime": 43.5081,
+      "eval_samples_per_second": 45.968,
+      "eval_steps_per_second": 0.735,
       "step": 300
     },
     {
-      "debug/losses": 0.12721626460552216,
-      "debug/policy_weights": 0.25444597005844116,
-      "debug/raw_losses": 0.49146708846092224,
+      "debug/losses": 0.1321336030960083,
+      "debug/policy_weights": 0.2647426724433899,
+      "debug/raw_losses": 0.4956802725791931,
       "epoch": 0.65,
+      "grad_norm": 7.197171827503445,
       "learning_rate": 1.6583128063291573e-07,
-      "logits/chosen": -2.169189453125,
-      "logits/rejected": -2.12001895904541,
-      "logps/chosen": -451.0543518066406,
-      "logps/rejected": -501.00830078125,
-      "loss": 0.1339,
-      "rewards/accuracies": 0.7749999761581421,
-      "rewards/chosen": -1.6334816217422485,
-      "rewards/margins": 0.8188700675964355,
-      "rewards/rejected": -2.4523518085479736,
+      "logits/chosen": -2.186096429824829,
+      "logits/rejected": -2.185868501663208,
+      "logps/chosen": -443.67803955078125,
+      "logps/rejected": -486.25091552734375,
+      "loss": 0.1268,
+      "rewards/accuracies": 0.768750011920929,
+      "rewards/chosen": -1.5781536102294922,
+      "rewards/margins": 0.7728461027145386,
+      "rewards/rejected": -2.3510000705718994,
       "step": 310
     },
     {
-      "debug/losses": 0.1101643294095993,
-      "debug/policy_weights": 0.24654574692249298,
-      "debug/raw_losses": 0.46620503067970276,
+      "debug/losses": 0.13546636700630188,
+      "debug/policy_weights": 0.2598869204521179,
+      "debug/raw_losses": 0.5187257528305054,
       "epoch": 0.67,
+      "grad_norm": 7.989288502873238,
       "learning_rate": 1.488723393865766e-07,
-      "logits/chosen": -2.1372463703155518,
-      "logits/rejected": -2.0980920791625977,
-      "logps/chosen": -443.265625,
-      "logps/rejected": -513.0931396484375,
-      "loss": 0.1228,
-      "rewards/accuracies": 0.78125,
-      "rewards/chosen": -1.5740420818328857,
-      "rewards/margins": 0.9036803245544434,
-      "rewards/rejected": -2.477722644805908,
+      "logits/chosen": -2.2410237789154053,
+      "logits/rejected": -2.2062346935272217,
+      "logps/chosen": -433.8523864746094,
+      "logps/rejected": -467.3929138183594,
+      "loss": 0.1308,
+      "rewards/accuracies": 0.7124999761581421,
+      "rewards/chosen": -1.493381142616272,
+      "rewards/margins": 0.7878578901290894,
+      "rewards/rejected": -2.2812390327453613,
       "step": 320
     },
     {
-      "debug/losses": 0.12689927220344543,
-      "debug/policy_weights": 0.25536665320396423,
-      "debug/raw_losses": 0.49713826179504395,
+      "debug/losses": 0.14906349778175354,
+      "debug/policy_weights": 0.2727554440498352,
+      "debug/raw_losses": 0.5552006959915161,
       "epoch": 0.69,
+      "grad_norm": 6.170487059058624,
       "learning_rate": 1.3245295796480788e-07,
-      "logits/chosen": -2.1500933170318604,
-      "logits/rejected": -2.1095988750457764,
-      "logps/chosen": -424.4521484375,
-      "logps/rejected": -503.61163330078125,
-      "loss": 0.1324,
-      "rewards/accuracies": 0.793749988079071,
-      "rewards/chosen": -1.5042273998260498,
-      "rewards/margins": 0.8895484805107117,
-      "rewards/rejected": -2.3937759399414062,
+      "logits/chosen": -2.236531972885132,
+      "logits/rejected": -2.1864519119262695,
+      "logps/chosen": -390.4073791503906,
+      "logps/rejected": -459.62347412109375,
+      "loss": 0.1411,
+      "rewards/accuracies": 0.7124999761581421,
+      "rewards/chosen": -1.373223066329956,
+      "rewards/margins": 0.6836794018745422,
+      "rewards/rejected": -2.0569024085998535,
       "step": 330
     },
     {
-      "debug/losses": 0.11528172343969345,
-      "debug/policy_weights": 0.24339346587657928,
-      "debug/raw_losses": 0.48817843198776245,
+      "debug/losses": 0.1494503617286682,
+      "debug/policy_weights": 0.28068724274635315,
+      "debug/raw_losses": 0.4996885359287262,
       "epoch": 0.71,
+      "grad_norm": 9.416700000227674,
       "learning_rate": 1.1666074087171627e-07,
-      "logits/chosen": -2.1724162101745605,
-      "logits/rejected": -2.1479129791259766,
-      "logps/chosen": -403.76806640625,
-      "logps/rejected": -496.89703369140625,
-      "loss": 0.1239,
-      "rewards/accuracies": 0.75,
-      "rewards/chosen": -1.4917190074920654,
-      "rewards/margins": 0.9479940533638,
-      "rewards/rejected": -2.4397130012512207,
+      "logits/chosen": -2.2197887897491455,
+      "logits/rejected": -2.1604628562927246,
+      "logps/chosen": -418.9659118652344,
+      "logps/rejected": -511.0225524902344,
+      "loss": 0.1446,
+      "rewards/accuracies": 0.800000011920929,
+      "rewards/chosen": -1.3679134845733643,
+      "rewards/margins": 1.0408259630203247,
+      "rewards/rejected": -2.4087395668029785,
       "step": 340
     },
     {
-      "debug/losses": 0.11798451095819473,
-      "debug/policy_weights": 0.23595662415027618,
-      "debug/raw_losses": 0.5302962064743042,
+      "debug/losses": 0.12328938394784927,
+      "debug/policy_weights": 0.24812527000904083,
+      "debug/raw_losses": 0.48738375306129456,
       "epoch": 0.73,
+      "grad_norm": 8.40666805347326,
       "learning_rate": 1.0157994641835734e-07,
-      "logits/chosen": -2.148632526397705,
-      "logits/rejected": -2.1165192127227783,
-      "logps/chosen": -442.53057861328125,
-      "logps/rejected": -513.9363403320312,
-      "loss": 0.1374,
+      "logits/chosen": -2.172837018966675,
+      "logits/rejected": -2.1325600147247314,
+      "logps/chosen": -398.5860900878906,
+      "logps/rejected": -471.38397216796875,
+      "loss": 0.1286,
       "rewards/accuracies": 0.7124999761581421,
-      "rewards/chosen": -1.6607694625854492,
-      "rewards/margins": 0.8479129672050476,
-      "rewards/rejected": -2.5086822509765625,
+      "rewards/chosen": -1.4896047115325928,
+      "rewards/margins": 0.9131780862808228,
+      "rewards/rejected": -2.402782917022705,
       "step": 350
     },
     {
-      "debug/losses": 0.10292885452508926,
-      "debug/policy_weights": 0.22015142440795898,
-      "debug/raw_losses": 0.4573485255241394,
+      "debug/losses": 0.1280422806739807,
+      "debug/policy_weights": 0.2567124366760254,
+      "debug/raw_losses": 0.5079829692840576,
       "epoch": 0.75,
+      "grad_norm": 7.5652554505895955,
       "learning_rate": 8.729103716819111e-08,
-      "logits/chosen": -2.1121697425842285,
-      "logits/rejected": -2.0813915729522705,
-      "logps/chosen": -401.754150390625,
-      "logps/rejected": -494.1514587402344,
-      "loss": 0.1311,
-      "rewards/accuracies": 0.762499988079071,
-      "rewards/chosen": -1.4336955547332764,
-      "rewards/margins": 1.074181079864502,
-      "rewards/rejected": -2.5078768730163574,
+      "logits/chosen": -2.230677366256714,
+      "logits/rejected": -2.138909101486206,
+      "logps/chosen": -448.53179931640625,
+      "logps/rejected": -494.7422790527344,
+      "loss": 0.1317,
+      "rewards/accuracies": 0.7437499761581421,
+      "rewards/chosen": -1.5306211709976196,
+      "rewards/margins": 0.8633092641830444,
+      "rewards/rejected": -2.393930435180664,
       "step": 360
     },
     {
-      "debug/losses": 0.13746492564678192,
-      "debug/policy_weights": 0.25476521253585815,
-      "debug/raw_losses": 0.5358820557594299,
+      "debug/losses": 0.13079939782619476,
+      "debug/policy_weights": 0.23427358269691467,
+      "debug/raw_losses": 0.5890456438064575,
       "epoch": 0.77,
+      "grad_norm": 9.87793971826204,
       "learning_rate": 7.387025063449081e-08,
-      "logits/chosen": -2.175912380218506,
-      "logits/rejected": -2.1576006412506104,
-      "logps/chosen": -416.6788024902344,
-      "logps/rejected": -477.13153076171875,
-      "loss": 0.1253,
-      "rewards/accuracies": 0.7124999761581421,
-      "rewards/chosen": -1.3861879110336304,
-      "rewards/margins": 0.7486017346382141,
-      "rewards/rejected": -2.1347897052764893,
+      "logits/chosen": -2.167722225189209,
+      "logits/rejected": -2.1336138248443604,
+      "logps/chosen": -432.41314697265625,
+      "logps/rejected": -470.364501953125,
+      "loss": 0.1313,
+      "rewards/accuracies": 0.71875,
+      "rewards/chosen": -1.6386140584945679,
+      "rewards/margins": 0.7472943067550659,
+      "rewards/rejected": -2.385908365249634,
       "step": 370
     },
     {
-      "debug/losses": 0.14446747303009033,
-      "debug/policy_weights": 0.2367408275604248,
-      "debug/raw_losses": 0.5707719326019287,
+      "debug/losses": 0.1057448610663414,
+      "debug/policy_weights": 0.2194373905658722,
+      "debug/raw_losses": 0.5027541518211365,
       "epoch": 0.79,
+      "grad_norm": 7.002597773553437,
       "learning_rate": 6.138919252022435e-08,
-      "logits/chosen": -2.084808111190796,
-      "logits/rejected": -2.070844888687134,
-      "logps/chosen": -394.0305480957031,
-      "logps/rejected": -481.216552734375,
-      "loss": 0.1285,
-      "rewards/accuracies": 0.699999988079071,
-      "rewards/chosen": -1.7118937969207764,
-      "rewards/margins": 0.6706421375274658,
-      "rewards/rejected": -2.3825364112854004,
+      "logits/chosen": -2.093229055404663,
+      "logits/rejected": -2.0845484733581543,
+      "logps/chosen": -409.2686462402344,
+      "logps/rejected": -525.614013671875,
+      "loss": 0.1212,
+      "rewards/accuracies": 0.75,
+      "rewards/chosen": -1.7782268524169922,
+      "rewards/margins": 0.96544349193573,
+      "rewards/rejected": -2.743670701980591,
       "step": 380
     },
     {
-      "debug/losses": 0.12546047568321228,
-      "debug/policy_weights": 0.24170584976673126,
-      "debug/raw_losses": 0.4937317967414856,
+      "debug/losses": 0.14649717509746552,
+      "debug/policy_weights": 0.27705007791519165,
+      "debug/raw_losses": 0.5246635675430298,
       "epoch": 0.82,
+      "grad_norm": 7.870379900484857,
       "learning_rate": 4.991445467064689e-08,
-      "logits/chosen": -2.1555511951446533,
-      "logits/rejected": -2.1209189891815186,
-      "logps/chosen": -406.2367248535156,
-      "logps/rejected": -479.50567626953125,
-      "loss": 0.1302,
-      "rewards/accuracies": 0.7875000238418579,
-      "rewards/chosen": -1.5467679500579834,
-      "rewards/margins": 0.8932439684867859,
-      "rewards/rejected": -2.440011501312256,
+      "logits/chosen": -2.1749279499053955,
+      "logits/rejected": -2.138232946395874,
+      "logps/chosen": -439.6005859375,
+      "logps/rejected": -510.23260498046875,
+      "loss": 0.1384,
+      "rewards/accuracies": 0.71875,
+      "rewards/chosen": -1.4645540714263916,
+      "rewards/margins": 0.8761545419692993,
+      "rewards/rejected": -2.3407082557678223,
       "step": 390
     },
     {
-      "debug/losses": 0.13418573141098022,
-      "debug/policy_weights": 0.24948246777057648,
-      "debug/raw_losses": 0.5159034132957458,
+      "debug/losses": 0.1481957882642746,
+      "debug/policy_weights": 0.25738272070884705,
+      "debug/raw_losses": 0.5208276510238647,
       "epoch": 0.84,
+      "grad_norm": 7.317947546801369,
       "learning_rate": 3.9507259776993954e-08,
-      "logits/chosen": -2.1299073696136475,
-      "logits/rejected": -2.0702521800994873,
-      "logps/chosen": -448.74298095703125,
-      "logps/rejected": -487.4234313964844,
-      "loss": 0.1329,
-      "rewards/accuracies": 0.7562500238418579,
-      "rewards/chosen": -1.6303924322128296,
-      "rewards/margins": 0.7685388326644897,
-      "rewards/rejected": -2.3989315032958984,
+      "logits/chosen": -2.1648905277252197,
+      "logits/rejected": -2.1050162315368652,
+      "logps/chosen": -416.4757385253906,
+      "logps/rejected": -496.00927734375,
+      "loss": 0.1398,
+      "rewards/accuracies": 0.75,
+      "rewards/chosen": -1.536012887954712,
+      "rewards/margins": 0.8311750292778015,
+      "rewards/rejected": -2.367187976837158,
       "step": 400
     },
     {
       "epoch": 0.84,
-      "eval_debug/losses": 0.12970629334449768,
-      "eval_debug/policy_weights": 0.25886857509613037,
-      "eval_debug/raw_losses": 0.48170554637908936,
-      "eval_logits/chosen": -2.18296217918396,
-      "eval_logits/rejected": -2.1507883071899414,
-      "eval_logps/chosen": -409.0387878417969,
-      "eval_logps/rejected": -500.7933654785156,
-      "eval_loss": 0.1314304769039154,
-      "eval_rewards/accuracies": 0.75,
-      "eval_rewards/chosen": -1.519991159439087,
-      "eval_rewards/margins": 0.9144098162651062,
-      "eval_rewards/rejected": -2.434401035308838,
-      "eval_runtime": 53.0316,
-      "eval_samples_per_second": 37.713,
-      "eval_steps_per_second": 0.603,
+      "eval_debug/losses": 0.13974608480930328,
+      "eval_debug/policy_weights": 0.28152135014533997,
+      "eval_debug/raw_losses": 0.47273018956184387,
+      "eval_logits/chosen": -2.2036194801330566,
+      "eval_logits/rejected": -2.1694607734680176,
+      "eval_logps/chosen": -386.73663330078125,
+      "eval_logps/rejected": -483.9327392578125,
+      "eval_loss": 0.14222779870033264,
+      "eval_rewards/accuracies": 0.76171875,
+      "eval_rewards/chosen": -1.3154058456420898,
+      "eval_rewards/margins": 0.9613481163978577,
+      "eval_rewards/rejected": -2.2767536640167236,
+      "eval_runtime": 98.9464,
+      "eval_samples_per_second": 20.213,
+      "eval_steps_per_second": 0.323,
       "step": 400
     },
     {
-      "debug/losses": 0.13898980617523193,
-      "debug/policy_weights": 0.24470162391662598,
-      "debug/raw_losses": 0.5698193907737732,
+      "debug/losses": 0.13129359483718872,
+      "debug/policy_weights": 0.25963759422302246,
+      "debug/raw_losses": 0.5042656660079956,
       "epoch": 0.86,
+      "grad_norm": 7.093812921964164,
       "learning_rate": 3.022313472693447e-08,
-      "logits/chosen": -2.1896605491638184,
-      "logits/rejected": -2.1111254692077637,
-      "logps/chosen": -431.5157165527344,
-      "logps/rejected": -519.735595703125,
-      "loss": 0.1335,
-      "rewards/accuracies": 0.71875,
-      "rewards/chosen": -1.67926025390625,
-      "rewards/margins": 0.7034608125686646,
-      "rewards/rejected": -2.382721185684204,
+      "logits/chosen": -2.2121942043304443,
+      "logits/rejected": -2.1647019386291504,
+      "logps/chosen": -432.4383850097656,
+      "logps/rejected": -489.4990234375,
+      "loss": 0.1404,
+      "rewards/accuracies": 0.75,
+      "rewards/chosen": -1.4152858257293701,
+      "rewards/margins": 0.8453054428100586,
+      "rewards/rejected": -2.260591506958008,
       "step": 410
     },
     {
-      "debug/losses": 0.13722026348114014,
-      "debug/policy_weights": 0.25542253255844116,
-      "debug/raw_losses": 0.553850531578064,
+      "debug/losses": 0.15901803970336914,
+      "debug/policy_weights": 0.2908507287502289,
+      "debug/raw_losses": 0.5441660284996033,
       "epoch": 0.88,
+      "grad_norm": 7.194889894590537,
       "learning_rate": 2.2111614344599684e-08,
-      "logits/chosen": -2.1816725730895996,
-      "logits/rejected": -2.1436891555786133,
-      "logps/chosen": -450.2588806152344,
-      "logps/rejected": -532.4340209960938,
-      "loss": 0.132,
-      "rewards/accuracies": 0.668749988079071,
-      "rewards/chosen": -1.6344585418701172,
-      "rewards/margins": 0.765791654586792,
-      "rewards/rejected": -2.40024995803833,
+      "logits/chosen": -2.1668639183044434,
+      "logits/rejected": -2.1467199325561523,
+      "logps/chosen": -427.835693359375,
+      "logps/rejected": -492.3506774902344,
+      "loss": 0.1412,
+      "rewards/accuracies": 0.6937500238418579,
+      "rewards/chosen": -1.4115890264511108,
+      "rewards/margins": 0.840238094329834,
+      "rewards/rejected": -2.2518272399902344,
       "step": 420
     },
     {
-      "debug/losses": 0.11242847144603729,
-      "debug/policy_weights": 0.23566405475139618,
-      "debug/raw_losses": 0.474797785282135,
+      "debug/losses": 0.1401083916425705,
+      "debug/policy_weights": 0.2716709077358246,
+      "debug/raw_losses": 0.49931907653808594,
       "epoch": 0.9,
+      "grad_norm": 6.971935536096578,
       "learning_rate": 1.521597710086439e-08,
-      "logits/chosen": -2.1367993354797363,
-      "logits/rejected": -2.101963520050049,
-      "logps/chosen": -447.2886657714844,
-      "logps/rejected": -528.4554443359375,
-      "loss": 0.135,
-      "rewards/accuracies": 0.768750011920929,
-      "rewards/chosen": -1.57035493850708,
-      "rewards/margins": 0.9703021049499512,
-      "rewards/rejected": -2.5406570434570312,
+      "logits/chosen": -2.106959342956543,
+      "logits/rejected": -2.051405668258667,
+      "logps/chosen": -412.853759765625,
+      "logps/rejected": -481.4043884277344,
+      "loss": 0.1291,
+      "rewards/accuracies": 0.7875000238418579,
+      "rewards/chosen": -1.4160857200622559,
+      "rewards/margins": 0.8900654911994934,
+      "rewards/rejected": -2.3061509132385254,
       "step": 430
     },
     {
-      "debug/losses": 0.14394986629486084,
-      "debug/policy_weights": 0.2594669461250305,
-      "debug/raw_losses": 0.551374614238739,
+      "debug/losses": 0.14032089710235596,
+      "debug/policy_weights": 0.28002962470054626,
+      "debug/raw_losses": 0.4943667948246002,
       "epoch": 0.92,
+      "grad_norm": 7.598363791526828,
       "learning_rate": 9.57301420397924e-09,
-      "logits/chosen": -2.1795363426208496,
-      "logits/rejected": -2.1459131240844727,
-      "logps/chosen": -479.01873779296875,
-      "logps/rejected": -490.289794921875,
-      "loss": 0.128,
-      "rewards/accuracies": 0.7250000238418579,
-      "rewards/chosen": -1.6396774053573608,
-      "rewards/margins": 0.6072880625724792,
-      "rewards/rejected": -2.2469656467437744,
+      "logits/chosen": -2.1854939460754395,
+      "logits/rejected": -2.1297545433044434,
+      "logps/chosen": -409.72625732421875,
+      "logps/rejected": -484.60546875,
+      "loss": 0.1401,
+      "rewards/accuracies": 0.737500011920929,
+      "rewards/chosen": -1.338160753250122,
+      "rewards/margins": 0.8834611773490906,
+      "rewards/rejected": -2.2216219902038574,
       "step": 440
     },
     {
-      "debug/losses": 0.12136085331439972,
-      "debug/policy_weights": 0.25487110018730164,
-      "debug/raw_losses": 0.5141120553016663,
+      "debug/losses": 0.1532571017742157,
+      "debug/policy_weights": 0.28182727098464966,
+      "debug/raw_losses": 0.5403295159339905,
       "epoch": 0.94,
+      "grad_norm": 7.420969377583743,
       "learning_rate": 5.212833302556258e-09,
-      "logits/chosen": -2.131383180618286,
-      "logits/rejected": -2.101245880126953,
-      "logps/chosen": -428.8694763183594,
-      "logps/rejected": -482.65380859375,
-      "loss": 0.1317,
-      "rewards/accuracies": 0.762499988079071,
-      "rewards/chosen": -1.4801188707351685,
-      "rewards/margins": 0.8445402383804321,
-      "rewards/rejected": -2.3246593475341797,
+      "logits/chosen": -2.108179807662964,
+      "logits/rejected": -2.0896925926208496,
+      "logps/chosen": -421.1512756347656,
+      "logps/rejected": -522.5540771484375,
+      "loss": 0.1398,
+      "rewards/accuracies": 0.699999988079071,
+      "rewards/chosen": -1.3998409509658813,
+      "rewards/margins": 0.8198366165161133,
+      "rewards/rejected": -2.219677448272705,
       "step": 450
     },
     {
-      "debug/losses": 0.13340520858764648,
-      "debug/policy_weights": 0.26456892490386963,
-      "debug/raw_losses": 0.5068139433860779,
+      "debug/losses": 0.14014151692390442,
+      "debug/policy_weights": 0.27090734243392944,
+      "debug/raw_losses": 0.5257419347763062,
       "epoch": 0.96,
+      "grad_norm": 6.950959627699968,
       "learning_rate": 2.158697848236607e-09,
-      "logits/chosen": -2.1625778675079346,
-      "logits/rejected": -2.118445873260498,
-      "logps/chosen": -446.69012451171875,
-      "logps/rejected": -497.6316833496094,
-      "loss": 0.1415,
-      "rewards/accuracies": 0.7749999761581421,
-      "rewards/chosen": -1.46303391456604,
-      "rewards/margins": 0.8069203495979309,
-      "rewards/rejected": -2.2699544429779053,
+      "logits/chosen": -2.1658027172088623,
+      "logits/rejected": -2.1314563751220703,
+      "logps/chosen": -402.38995361328125,
+      "logps/rejected": -448.8475646972656,
+      "loss": 0.1362,
+      "rewards/accuracies": 0.71875,
+      "rewards/chosen": -1.3720080852508545,
+      "rewards/margins": 0.7829836010932922,
+      "rewards/rejected": -2.154991626739502,
       "step": 460
     },
     {
-      "debug/losses": 0.13318563997745514,
-      "debug/policy_weights": 0.2579984962940216,
-      "debug/raw_losses": 0.5174868702888489,
+      "debug/losses": 0.14437520503997803,
+      "debug/policy_weights": 0.27037832140922546,
+      "debug/raw_losses": 0.5343233942985535,
       "epoch": 0.98,
+      "grad_norm": 8.44671954335882,
       "learning_rate": 4.269029751107489e-10,
-      "logits/chosen": -2.1812849044799805,
-      "logits/rejected": -2.100320816040039,
-      "logps/chosen": -435.08428955078125,
-      "logps/rejected": -452.8404235839844,
-      "loss": 0.1302,
-      "rewards/accuracies": 0.699999988079071,
-      "rewards/chosen": -1.3687784671783447,
-      "rewards/margins": 0.8292155265808105,
-      "rewards/rejected": -2.1979942321777344,
+      "logits/chosen": -2.151285171508789,
+      "logits/rejected": -2.119642972946167,
+      "logps/chosen": -403.640625,
+      "logps/rejected": -483.1455078125,
+      "loss": 0.1371,
+      "rewards/accuracies": 0.7124999761581421,
+      "rewards/chosen": -1.3858269453048706,
+      "rewards/margins": 0.757634699344635,
+      "rewards/rejected": -2.1434617042541504,
       "step": 470
     },
     {
       "epoch": 1.0,
       "step": 478,
       "total_flos": 0.0,
-      "train_loss": 0.17621936496331603,
-      "train_runtime": 4510.4366,
-      "train_samples_per_second": 13.554,
-      "train_steps_per_second": 0.106
+      "train_loss": 0.17903520272865456,
+      "train_runtime": 4631.245,
+      "train_samples_per_second": 13.2,
+      "train_steps_per_second": 0.103
     }
   ],
   "logging_steps": 10,
   "max_steps": 478,
+  "num_input_tokens_seen": 0,
   "num_train_epochs": 1,
   "save_steps": 100,
   "total_flos": 0.0,
+  "train_batch_size": 8,
   "trial_name": null,
   "trial_params": null
 }

training_args.bin

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:9719e3f0efcf7076e0c24a28a8de3b015157defd25e092d527d5719a985b314c
-size 5944
+oid sha256:399bbe17d43609818d6de82c511649d9abdfbf26f93b2196d4e24cacea1c61ee
+size 6328