Model save

README.md

@@ -1,15 +1,9 @@
 ---
 base_model: dmis-lab/selfbiorag_7b
 tags:
-- alignment-handbook
 - trl
 - dpo
 - generated_from_trainer
-- trl
-- dpo
-- generated_from_trainer
-datasets:
-- HuggingFaceH4/ultrafeedback_binarized
 model-index:
 - name: selfbiorag-7b-dpo-full-wo-live_qa-ep3
   results: []
@@ -20,17 +14,17 @@ should probably proofread and complete it, then remove this comment. -->
 
 # selfbiorag-7b-dpo-full-wo-live_qa-ep3
 
-This model is a fine-tuned version of [dmis-lab/selfbiorag_7b](https://huggingface.co/dmis-lab/selfbiorag_7b) on the HuggingFaceH4/ultrafeedback_binarized dataset.
+This model is a fine-tuned version of [dmis-lab/selfbiorag_7b](https://huggingface.co/dmis-lab/selfbiorag_7b) on an unknown dataset.
 It achieves the following results on the evaluation set:
-- Loss: 0.4246
-- Rewards/chosen: 0.5568
-- Rewards/rejected: -0.1398
-- Rewards/accuracies: 0.9891
-- Rewards/margins: 0.6965
-- Logps/rejected: -106.8022
-- Logps/chosen: -251.5511
-- Logits/rejected: -1.9696
-- Logits/chosen: -1.9998
+- Loss: 0.6502
+- Rewards/chosen: 0.1544
+- Rewards/rejected: 0.0500
+- Rewards/accuracies: 0.7273
+- Rewards/margins: 0.1043
+- Logps/rejected: -152.4127
+- Logps/chosen: -129.2778
+- Logits/rejected: -1.6898
+- Logits/chosen: -1.9159
 
 ## Model description
 
@@ -54,25 +48,28 @@ The following hyperparameters were used during training:
 - eval_batch_size: 8
 - seed: 42
 - distributed_type: multi-GPU
-- num_devices: 8
+- num_devices: 4
 - gradient_accumulation_steps: 2
-- total_train_batch_size: 128
-- total_eval_batch_size: 64
+- total_train_batch_size: 64
+- total_eval_batch_size: 32
 - optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
 - lr_scheduler_type: cosine
 - lr_scheduler_warmup_ratio: 0.1
-- num_epochs: 3
+- num_epochs: 1
 
 ### Training results
 
 | Training Loss | Epoch | Step | Validation Loss | Rewards/chosen | Rewards/rejected | Rewards/accuracies | Rewards/margins | Logps/rejected | Logps/chosen | Logits/rejected | Logits/chosen |
 |:-------------:|:-----:|:----:|:---------------:|:--------------:|:----------------:|:------------------:|:---------------:|:--------------:|:------------:|:---------------:|:-------------:|
-| 0.2874        | 2.63  | 100  | 0.4252          | 0.5557         | -0.1385          | 0.9891             | 0.6942          | -106.6720      | -251.6549    | -1.9692         | -2.0004       |
+| 0.6586        | 0.22  | 100  | 0.6694          | 0.1304         | 0.0785           | 0.6932             | 0.0520          | -149.5688      | -131.6699    | -1.5849         | -1.7960       |
+| 0.6342        | 0.44  | 200  | 0.6581          | 0.1743         | 0.0934           | 0.7273             | 0.0808          | -148.0715      | -127.2864    | -1.5703         | -1.7959       |
+| 0.5967        | 0.66  | 300  | 0.6527          | 0.1658         | 0.0685           | 0.7159             | 0.0973          | -150.5620      | -128.1308    | -1.6454         | -1.8741       |
+| 0.5979        | 0.88  | 400  | 0.6502          | 0.1544         | 0.0500           | 0.7273             | 0.1043          | -152.4127      | -129.2778    | -1.6898         | -1.9159       |
 
 
 ### Framework versions
 
 - Transformers 4.39.0.dev0
-- Pytorch 2.2.1+cu121
+- Pytorch 2.1.2
 - Datasets 2.14.6
 - Tokenizers 0.15.2

all_results.json

@@ -1,21 +1,8 @@
 {
-    "epoch": 3.0,
-    "eval_logits/chosen": -1.999770164489746,
-    "eval_logits/rejected": -1.969599723815918,
-    "eval_logps/chosen": -251.5510711669922,
-    "eval_logps/rejected": -106.80218505859375,
-    "eval_loss": 0.42463189363479614,
-    "eval_rewards/accuracies": 0.989130437374115,
-    "eval_rewards/chosen": 0.5567559599876404,
-    "eval_rewards/margins": 0.6965265870094299,
-    "eval_rewards/rejected": -0.13977064192295074,
-    "eval_runtime": 38.2309,
-    "eval_samples": 1445,
-    "eval_samples_per_second": 37.797,
-    "eval_steps_per_second": 0.602,
-    "train_loss": 0.41232719756009284,
-    "train_runtime": 892.3177,
-    "train_samples": 4859,
-    "train_samples_per_second": 16.336,
-    "train_steps_per_second": 0.128
+    "epoch": 1.0,
+    "train_loss": 0.6286900518214809,
+    "train_runtime": 6365.7753,
+    "train_samples": 28906,
+    "train_samples_per_second": 4.541,
+    "train_steps_per_second": 0.071
 }

model-00001-of-00003.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:ae57e1f0c3eb5798ae125339feed63aa2cb050eb43f282f7e60f49b5f79660e8
+oid sha256:691fd6b365b1e86bf603fe852a578748e4b3263fa6d7dffc08fc8e48ef93a92d
 size 4939116424

model-00002-of-00003.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:74190c046d6cf1d841bee53a3e1976be5d005f9f603fffdf60c6e8ab067d2b21
+oid sha256:86981eaebf61419ea432021d6b8b75510b432dabad944e9bf3157294ed6a981f
 size 4947390880

model-00003-of-00003.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:e915be3dc2e2ab26b9e2fe1b0b76a9f2fca315f07c49d20a3c4619254572dbbc
+oid sha256:43a1311e7a67e3b2dc129b8560e6bb104d3ed5f59e6ed141200f322198dd26b1
 size 3590619888

train_results.json

@@ -1,8 +1,8 @@
 {
-    "epoch": 3.0,
-    "train_loss": 0.41232719756009284,
-    "train_runtime": 892.3177,
-    "train_samples": 4859,
-    "train_samples_per_second": 16.336,
-    "train_steps_per_second": 0.128
+    "epoch": 1.0,
+    "train_loss": 0.6286900518214809,
+    "train_runtime": 6365.7753,
+    "train_samples": 28906,
+    "train_samples_per_second": 4.541,
+    "train_steps_per_second": 0.071
 }

trainer_state.json

@@ -1,21 +1,21 @@
 {
   "best_metric": null,
   "best_model_checkpoint": null,
-  "epoch": 3.0,
+  "epoch": 1.0,
   "eval_steps": 100,
-  "global_step": 114,
+  "global_step": 452,
   "is_hyper_param_search": false,
   "is_local_process_zero": true,
   "is_world_process_zero": true,
   "log_history": [
     {
-      "epoch": 0.03,
-      "grad_norm": 15.695575290564046,
-      "learning_rate": 4.166666666666666e-08,
-      "logits/chosen": -1.6109695434570312,
-      "logits/rejected": -1.8240289688110352,
-      "logps/chosen": -377.82647705078125,
-      "logps/rejected": -85.1103744506836,
+      "epoch": 0.0,
+      "grad_norm": 6.375993978538379,
+      "learning_rate": 1.0869565217391303e-08,
+      "logits/chosen": -1.8598690032958984,
+      "logits/rejected": -1.813749074935913,
+      "logps/chosen": -155.32705688476562,
+      "logps/rejected": -119.10728454589844,
       "loss": 0.6931,
       "rewards/accuracies": 0.0,
       "rewards/chosen": 0.0,
@@ -24,200 +24,758 @@
       "step": 1
     },
     {
-      "epoch": 0.26,
-      "grad_norm": 15.298800622377637,
-      "learning_rate": 4.1666666666666667e-07,
-      "logits/chosen": -1.7764019966125488,
-      "logits/rejected": -1.7756922245025635,
-      "logps/chosen": -359.9149475097656,
-      "logps/rejected": -89.94466400146484,
-      "loss": 0.6918,
-      "rewards/accuracies": 0.5763888955116272,
-      "rewards/chosen": 0.002338339574635029,
-      "rewards/margins": 0.0021099813748151064,
-      "rewards/rejected": 0.00022835780691821128,
+      "epoch": 0.02,
+      "grad_norm": 6.761130122357207,
+      "learning_rate": 1.0869565217391303e-07,
+      "logits/chosen": -1.7754908800125122,
+      "logits/rejected": -1.9477850198745728,
+      "logps/chosen": -164.1376495361328,
+      "logps/rejected": -177.90771484375,
+      "loss": 0.693,
+      "rewards/accuracies": 0.4583333432674408,
+      "rewards/chosen": -0.0001304509787587449,
+      "rewards/margins": -0.0002223679330199957,
+      "rewards/rejected": 9.191703429678455e-05,
       "step": 10
     },
     {
-      "epoch": 0.53,
-      "grad_norm": 13.520085274594496,
-      "learning_rate": 4.924492340087524e-07,
-      "logits/chosen": -1.7332627773284912,
-      "logits/rejected": -1.8106008768081665,
-      "logps/chosen": -382.57012939453125,
-      "logps/rejected": -91.563720703125,
-      "loss": 0.6519,
-      "rewards/accuracies": 0.96875,
-      "rewards/chosen": 0.09311604499816895,
-      "rewards/margins": 0.09122468531131744,
-      "rewards/rejected": 0.001891360036097467,
+      "epoch": 0.04,
+      "grad_norm": 6.476650286184237,
+      "learning_rate": 2.1739130434782607e-07,
+      "logits/chosen": -1.813336730003357,
+      "logits/rejected": -1.8786985874176025,
+      "logps/chosen": -154.4861297607422,
+      "logps/rejected": -178.77951049804688,
+      "loss": 0.6931,
+      "rewards/accuracies": 0.550000011920929,
+      "rewards/chosen": 0.000607121444772929,
+      "rewards/margins": 0.000769242993555963,
+      "rewards/rejected": -0.00016212157788686454,
       "step": 20
     },
     {
-      "epoch": 0.79,
-      "grad_norm": 10.836566523756993,
-      "learning_rate": 4.6255428393240354e-07,
-      "logits/chosen": -1.6527366638183594,
-      "logits/rejected": -1.8300998210906982,
-      "logps/chosen": -329.12725830078125,
-      "logps/rejected": -93.04560852050781,
-      "loss": 0.5493,
-      "rewards/accuracies": 0.981249988079071,
-      "rewards/chosen": 0.3091312050819397,
-      "rewards/margins": 0.30923140048980713,
-      "rewards/rejected": -0.00010019731416832656,
+      "epoch": 0.07,
+      "grad_norm": 6.2480047328303705,
+      "learning_rate": 3.260869565217391e-07,
+      "logits/chosen": -1.729405164718628,
+      "logits/rejected": -1.8028945922851562,
+      "logps/chosen": -159.96788024902344,
+      "logps/rejected": -153.2834930419922,
+      "loss": 0.6927,
+      "rewards/accuracies": 0.574999988079071,
+      "rewards/chosen": 0.003659260692074895,
+      "rewards/margins": 0.0018362473929300904,
+      "rewards/rejected": 0.001823012949898839,
       "step": 30
     },
     {
-      "epoch": 1.05,
-      "grad_norm": 8.223735591833103,
-      "learning_rate": 4.126545750510605e-07,
-      "logits/chosen": -1.6139713525772095,
-      "logits/rejected": -1.7977306842803955,
-      "logps/chosen": -339.6487731933594,
-      "logps/rejected": -91.19889831542969,
-      "loss": 0.4628,
-      "rewards/accuracies": 1.0,
-      "rewards/chosen": 0.541479229927063,
-      "rewards/margins": 0.5557602643966675,
-      "rewards/rejected": -0.014281067065894604,
+      "epoch": 0.09,
+      "grad_norm": 6.780490815116132,
+      "learning_rate": 4.3478260869565214e-07,
+      "logits/chosen": -1.715442419052124,
+      "logits/rejected": -1.8770908117294312,
+      "logps/chosen": -168.16839599609375,
+      "logps/rejected": -172.1099853515625,
+      "loss": 0.6899,
+      "rewards/accuracies": 0.65625,
+      "rewards/chosen": 0.010979695245623589,
+      "rewards/margins": 0.005665521137416363,
+      "rewards/rejected": 0.0053141750395298,
       "step": 40
     },
     {
-      "epoch": 1.32,
-      "grad_norm": 5.164509496454736,
-      "learning_rate": 3.474464683231698e-07,
-      "logits/chosen": -1.5413013696670532,
-      "logits/rejected": -1.6259466409683228,
-      "logps/chosen": -299.329345703125,
-      "logps/rejected": -98.39735412597656,
-      "loss": 0.3926,
-      "rewards/accuracies": 1.0,
-      "rewards/chosen": 0.7213589549064636,
-      "rewards/margins": 0.7860541343688965,
-      "rewards/rejected": -0.06469517201185226,
+      "epoch": 0.11,
+      "grad_norm": 6.269256344947319,
+      "learning_rate": 4.998802589665008e-07,
+      "logits/chosen": -1.6737487316131592,
+      "logits/rejected": -1.8300039768218994,
+      "logps/chosen": -157.9713134765625,
+      "logps/rejected": -184.8284912109375,
+      "loss": 0.6862,
+      "rewards/accuracies": 0.6625000238418579,
+      "rewards/chosen": 0.024453260004520416,
+      "rewards/margins": 0.0103492122143507,
+      "rewards/rejected": 0.014104047790169716,
       "step": 50
     },
     {
-      "epoch": 1.58,
-      "grad_norm": 3.8172387720508434,
-      "learning_rate": 2.730670898658255e-07,
-      "logits/chosen": -1.4262573719024658,
-      "logits/rejected": -1.8172870874404907,
-      "logps/chosen": -286.40216064453125,
-      "logps/rejected": -96.14507293701172,
-      "loss": 0.329,
-      "rewards/accuracies": 1.0,
-      "rewards/chosen": 0.9281619191169739,
-      "rewards/margins": 1.0235956907272339,
-      "rewards/rejected": -0.09543389827013016,
+      "epoch": 0.13,
+      "grad_norm": 5.9502015182077,
+      "learning_rate": 4.985344892885899e-07,
+      "logits/chosen": -1.7060184478759766,
+      "logits/rejected": -1.8548672199249268,
+      "logps/chosen": -182.36965942382812,
+      "logps/rejected": -164.8349151611328,
+      "loss": 0.6821,
+      "rewards/accuracies": 0.706250011920929,
+      "rewards/chosen": 0.052476245909929276,
+      "rewards/margins": 0.026238251477479935,
+      "rewards/rejected": 0.02623800002038479,
       "step": 60
     },
     {
-      "epoch": 1.84,
-      "grad_norm": 3.5476238893060894,
-      "learning_rate": 1.965167291983757e-07,
-      "logits/chosen": -1.45745849609375,
-      "logits/rejected": -1.7625938653945923,
-      "logps/chosen": -261.16015625,
-      "logps/rejected": -112.32124328613281,
-      "loss": 0.3178,
-      "rewards/accuracies": 1.0,
-      "rewards/chosen": 0.8942011594772339,
-      "rewards/margins": 1.0334198474884033,
-      "rewards/rejected": -0.13921861350536346,
+      "epoch": 0.15,
+      "grad_norm": 6.211310793736214,
+      "learning_rate": 4.957013543421161e-07,
+      "logits/chosen": -1.6781314611434937,
+      "logits/rejected": -1.7970473766326904,
+      "logps/chosen": -156.598388671875,
+      "logps/rejected": -159.87399291992188,
+      "loss": 0.6735,
+      "rewards/accuracies": 0.75,
+      "rewards/chosen": 0.06103619933128357,
+      "rewards/margins": 0.040171049535274506,
+      "rewards/rejected": 0.020865142345428467,
       "step": 70
     },
     {
-      "epoch": 2.11,
-      "grad_norm": 3.3551163964170834,
-      "learning_rate": 1.2500000000000005e-07,
-      "logits/chosen": -1.5400848388671875,
-      "logits/rejected": -1.910940408706665,
-      "logps/chosen": -271.2547607421875,
-      "logps/rejected": -104.63142395019531,
-      "loss": 0.3055,
-      "rewards/accuracies": 1.0,
-      "rewards/chosen": 0.9246896505355835,
-      "rewards/margins": 1.0737279653549194,
-      "rewards/rejected": -0.1490381807088852,
+      "epoch": 0.18,
+      "grad_norm": 5.997223838682246,
+      "learning_rate": 4.913978091441985e-07,
+      "logits/chosen": -1.591805338859558,
+      "logits/rejected": -1.6250261068344116,
+      "logps/chosen": -149.93402099609375,
+      "logps/rejected": -149.83056640625,
+      "loss": 0.6686,
+      "rewards/accuracies": 0.7124999761581421,
+      "rewards/chosen": 0.08923570066690445,
+      "rewards/margins": 0.0577419213950634,
+      "rewards/rejected": 0.03149377182126045,
       "step": 80
     },
     {
-      "epoch": 2.37,
-      "grad_norm": 3.219821221036962,
-      "learning_rate": 6.524777069483525e-08,
-      "logits/chosen": -1.4664478302001953,
-      "logits/rejected": -1.8648977279663086,
-      "logps/chosen": -264.0588073730469,
-      "logps/rejected": -113.32071685791016,
-      "loss": 0.2951,
-      "rewards/accuracies": 1.0,
-      "rewards/chosen": 0.932086169719696,
-      "rewards/margins": 1.1108168363571167,
-      "rewards/rejected": -0.17873072624206543,
+      "epoch": 0.2,
+      "grad_norm": 5.698461942985052,
+      "learning_rate": 4.856496084449218e-07,
+      "logits/chosen": -1.6099445819854736,
+      "logits/rejected": -1.7586275339126587,
+      "logps/chosen": -168.14712524414062,
+      "logps/rejected": -164.38485717773438,
+      "loss": 0.6648,
+      "rewards/accuracies": 0.762499988079071,
+      "rewards/chosen": 0.0959181934595108,
+      "rewards/margins": 0.06268787384033203,
+      "rewards/rejected": 0.03323032334446907,
       "step": 90
     },
     {
-      "epoch": 2.63,
-      "grad_norm": 3.1639794499179605,
-      "learning_rate": 2.2883682045119062e-08,
-      "logits/chosen": -1.5078153610229492,
-      "logits/rejected": -1.8548780679702759,
-      "logps/chosen": -261.926513671875,
-      "logps/rejected": -104.53482818603516,
-      "loss": 0.2874,
-      "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": 0.9581863284111023,
-      "rewards/margins": 1.1179349422454834,
-      "rewards/rejected": -0.15974867343902588,
+      "epoch": 0.22,
+      "grad_norm": 5.821198867398168,
+      "learning_rate": 4.784911525969344e-07,
+      "logits/chosen": -1.5159536600112915,
+      "logits/rejected": -1.6637051105499268,
+      "logps/chosen": -144.79641723632812,
+      "logps/rejected": -155.20135498046875,
+      "loss": 0.6586,
+      "rewards/accuracies": 0.706250011920929,
+      "rewards/chosen": 0.11348827183246613,
+      "rewards/margins": 0.057287197560071945,
+      "rewards/rejected": 0.056201063096523285,
       "step": 100
     },
     {
-      "epoch": 2.63,
-      "eval_logits/chosen": -2.0004403591156006,
-      "eval_logits/rejected": -1.9692000150680542,
-      "eval_logps/chosen": -251.65493774414062,
-      "eval_logps/rejected": -106.67204284667969,
-      "eval_loss": 0.42517274618148804,
-      "eval_rewards/accuracies": 0.989130437374115,
-      "eval_rewards/chosen": 0.555717408657074,
-      "eval_rewards/margins": 0.6941863894462585,
-      "eval_rewards/rejected": -0.1384691298007965,
-      "eval_runtime": 38.519,
-      "eval_samples_per_second": 37.514,
-      "eval_steps_per_second": 0.597,
+      "epoch": 0.22,
+      "eval_logits/chosen": -1.7960463762283325,
+      "eval_logits/rejected": -1.5849276781082153,
+      "eval_logps/chosen": -131.66993713378906,
+      "eval_logps/rejected": -149.56881713867188,
+      "eval_loss": 0.6694082617759705,
+      "eval_rewards/accuracies": 0.6931818127632141,
+      "eval_rewards/chosen": 0.130432590842247,
+      "eval_rewards/margins": 0.051959823817014694,
+      "eval_rewards/rejected": 0.07847274094820023,
+      "eval_runtime": 64.0941,
+      "eval_samples_per_second": 10.516,
+      "eval_steps_per_second": 0.343,
       "step": 100
     },
     {
-      "epoch": 2.89,
-      "grad_norm": 3.372775379055756,
-      "learning_rate": 1.8948725820160662e-09,
-      "logits/chosen": -1.5847268104553223,
-      "logits/rejected": -1.8966096639633179,
-      "logps/chosen": -271.20782470703125,
-      "logps/rejected": -107.42216491699219,
-      "loss": 0.2942,
-      "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": 0.9737249612808228,
-      "rewards/margins": 1.1161648035049438,
-      "rewards/rejected": -0.14243988692760468,
+      "epoch": 0.24,
+      "grad_norm": 5.737042856571591,
+      "learning_rate": 4.699652816850686e-07,
+      "logits/chosen": -1.5323385000228882,
+      "logits/rejected": -1.623490571975708,
+      "logps/chosen": -153.6015625,
+      "logps/rejected": -182.46792602539062,
+      "loss": 0.6609,
+      "rewards/accuracies": 0.7562500238418579,
+      "rewards/chosen": 0.11847398430109024,
+      "rewards/margins": 0.07614362984895706,
+      "rewards/rejected": 0.04233035817742348,
       "step": 110
     },
     {
-      "epoch": 3.0,
-      "step": 114,
+      "epoch": 0.27,
+      "grad_norm": 5.744846632920582,
+      "learning_rate": 4.6012301914802236e-07,
+      "logits/chosen": -1.4478352069854736,
+      "logits/rejected": -1.6309044361114502,
+      "logps/chosen": -145.10116577148438,
+      "logps/rejected": -180.8983154296875,
+      "loss": 0.6529,
+      "rewards/accuracies": 0.7749999761581421,
+      "rewards/chosen": 0.1110893040895462,
+      "rewards/margins": 0.07829709351062775,
+      "rewards/rejected": 0.032792218029499054,
+      "step": 120
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 5.866234048530386,
+      "learning_rate": 4.490232664264109e-07,
+      "logits/chosen": -1.5028413534164429,
+      "logits/rejected": -1.4950730800628662,
+      "logps/chosen": -157.7068634033203,
+      "logps/rejected": -163.91094970703125,
+      "loss": 0.6543,
+      "rewards/accuracies": 0.768750011920929,
+      "rewards/chosen": 0.13327065110206604,
+      "rewards/margins": 0.0991601049900055,
+      "rewards/rejected": 0.03411053121089935,
+      "step": 130
+    },
+    {
+      "epoch": 0.31,
+      "grad_norm": 5.2985790715652294,
+      "learning_rate": 4.3673245046457916e-07,
+      "logits/chosen": -1.439444899559021,
+      "logits/rejected": -1.4795516729354858,
+      "logps/chosen": -133.6232147216797,
+      "logps/rejected": -162.96536254882812,
+      "loss": 0.6428,
+      "rewards/accuracies": 0.793749988079071,
+      "rewards/chosen": 0.12999793887138367,
+      "rewards/margins": 0.10853303968906403,
+      "rewards/rejected": 0.021464908495545387,
+      "step": 140
+    },
+    {
+      "epoch": 0.33,
+      "grad_norm": 5.49213913037938,
+      "learning_rate": 4.2332412617571544e-07,
+      "logits/chosen": -1.470943570137024,
+      "logits/rejected": -1.632310152053833,
+      "logps/chosen": -152.8888702392578,
+      "logps/rejected": -153.59072875976562,
+      "loss": 0.6469,
+      "rewards/accuracies": 0.75,
+      "rewards/chosen": 0.13645324110984802,
+      "rewards/margins": 0.09464041888713837,
+      "rewards/rejected": 0.041812822222709656,
+      "step": 150
+    },
+    {
+      "epoch": 0.35,
+      "grad_norm": 5.401465863070874,
+      "learning_rate": 4.088785362493313e-07,
+      "logits/chosen": -1.5582928657531738,
+      "logits/rejected": -1.6405029296875,
+      "logps/chosen": -157.95230102539062,
+      "logps/rejected": -166.15577697753906,
+      "loss": 0.6374,
+      "rewards/accuracies": 0.8062499761581421,
+      "rewards/chosen": 0.1728488802909851,
+      "rewards/margins": 0.13387203216552734,
+      "rewards/rejected": 0.03897686302661896,
+      "step": 160
+    },
+    {
+      "epoch": 0.38,
+      "grad_norm": 5.73235647435814,
+      "learning_rate": 3.93482130935458e-07,
+      "logits/chosen": -1.4991676807403564,
+      "logits/rejected": -1.6196138858795166,
+      "logps/chosen": -147.5597381591797,
+      "logps/rejected": -162.0667266845703,
+      "loss": 0.6319,
+      "rewards/accuracies": 0.8062499761581421,
+      "rewards/chosen": 0.1789514720439911,
+      "rewards/margins": 0.13890545070171356,
+      "rewards/rejected": 0.040045998990535736,
+      "step": 170
+    },
+    {
+      "epoch": 0.4,
+      "grad_norm": 5.503184592670663,
+      "learning_rate": 3.772270506794322e-07,
+      "logits/chosen": -1.4963356256484985,
+      "logits/rejected": -1.6418964862823486,
+      "logps/chosen": -137.38124084472656,
+      "logps/rejected": -166.94546508789062,
+      "loss": 0.6335,
+      "rewards/accuracies": 0.768750011920929,
+      "rewards/chosen": 0.13972221314907074,
+      "rewards/margins": 0.11094947904348373,
+      "rewards/rejected": 0.028772741556167603,
+      "step": 180
+    },
+    {
+      "epoch": 0.42,
+      "grad_norm": 5.574317629060601,
+      "learning_rate": 3.6021057470346455e-07,
+      "logits/chosen": -1.4465644359588623,
+      "logits/rejected": -1.5275087356567383,
+      "logps/chosen": -142.6824951171875,
+      "logps/rejected": -147.18019104003906,
+      "loss": 0.6315,
+      "rewards/accuracies": 0.800000011920929,
+      "rewards/chosen": 0.1697317659854889,
+      "rewards/margins": 0.1489827036857605,
+      "rewards/rejected": 0.020749058574438095,
+      "step": 190
+    },
+    {
+      "epoch": 0.44,
+      "grad_norm": 5.601814789329134,
+      "learning_rate": 3.4253453883497864e-07,
+      "logits/chosen": -1.5157060623168945,
+      "logits/rejected": -1.5182088613510132,
+      "logps/chosen": -138.18411254882812,
+      "logps/rejected": -177.93359375,
+      "loss": 0.6342,
+      "rewards/accuracies": 0.7437499761581421,
+      "rewards/chosen": 0.15571925044059753,
+      "rewards/margins": 0.12424109876155853,
+      "rewards/rejected": 0.0314781591296196,
+      "step": 200
+    },
+    {
+      "epoch": 0.44,
+      "eval_logits/chosen": -1.7958874702453613,
+      "eval_logits/rejected": -1.5702953338623047,
+      "eval_logps/chosen": -127.2863540649414,
+      "eval_logps/rejected": -148.07151794433594,
+      "eval_loss": 0.6580806970596313,
+      "eval_rewards/accuracies": 0.7272727489471436,
+      "eval_rewards/chosen": 0.1742684245109558,
+      "eval_rewards/margins": 0.08082254230976105,
+      "eval_rewards/rejected": 0.09344588965177536,
+      "eval_runtime": 64.1886,
+      "eval_samples_per_second": 10.5,
+      "eval_steps_per_second": 0.343,
+      "step": 200
+    },
+    {
+      "epoch": 0.46,
+      "grad_norm": 6.199728402265358,
+      "learning_rate": 3.2430472606575104e-07,
+      "logits/chosen": -1.5152722597122192,
+      "logits/rejected": -1.5520877838134766,
+      "logps/chosen": -131.77743530273438,
+      "logps/rejected": -161.01565551757812,
+      "loss": 0.6259,
+      "rewards/accuracies": 0.8125,
+      "rewards/chosen": 0.17089861631393433,
+      "rewards/margins": 0.1506127417087555,
+      "rewards/rejected": 0.020285870879888535,
+      "step": 210
+    },
+    {
+      "epoch": 0.49,
+      "grad_norm": 5.744865266952806,
+      "learning_rate": 3.056302334890786e-07,
+      "logits/chosen": -1.5393413305282593,
+      "logits/rejected": -1.6224792003631592,
+      "logps/chosen": -139.15541076660156,
+      "logps/rejected": -153.6171417236328,
+      "loss": 0.6265,
+      "rewards/accuracies": 0.824999988079071,
+      "rewards/chosen": 0.181275874376297,
+      "rewards/margins": 0.16251231729984283,
+      "rewards/rejected": 0.018763558939099312,
+      "step": 220
+    },
+    {
+      "epoch": 0.51,
+      "grad_norm": 5.612480895027258,
+      "learning_rate": 2.866228194035623e-07,
+      "logits/chosen": -1.5460705757141113,
+      "logits/rejected": -1.6816442012786865,
+      "logps/chosen": -143.0059356689453,
+      "logps/rejected": -156.33035278320312,
+      "loss": 0.6183,
+      "rewards/accuracies": 0.793749988079071,
+      "rewards/chosen": 0.17381103336811066,
+      "rewards/margins": 0.15255849063396454,
+      "rewards/rejected": 0.021252544596791267,
+      "step": 230
+    },
+    {
+      "epoch": 0.53,
+      "grad_norm": 5.69741102915849,
+      "learning_rate": 2.673962344907953e-07,
+      "logits/chosen": -1.509167194366455,
+      "logits/rejected": -1.6316970586776733,
+      "logps/chosen": -162.8449249267578,
+      "logps/rejected": -151.0370635986328,
+      "loss": 0.6158,
+      "rewards/accuracies": 0.8500000238418579,
+      "rewards/chosen": 0.17691971361637115,
+      "rewards/margins": 0.16015170514583588,
+      "rewards/rejected": 0.016767997294664383,
+      "step": 240
+    },
+    {
+      "epoch": 0.55,
+      "grad_norm": 6.055880696709496,
+      "learning_rate": 2.4806554106954945e-07,
+      "logits/chosen": -1.558767557144165,
+      "logits/rejected": -1.66998291015625,
+      "logps/chosen": -154.49603271484375,
+      "logps/rejected": -168.3751220703125,
+      "loss": 0.6173,
+      "rewards/accuracies": 0.824999988079071,
+      "rewards/chosen": 0.17227458953857422,
+      "rewards/margins": 0.16756592690944672,
+      "rewards/rejected": 0.004708637483417988,
+      "step": 250
+    },
+    {
+      "epoch": 0.58,
+      "grad_norm": 5.955808609049162,
+      "learning_rate": 2.287464245004132e-07,
+      "logits/chosen": -1.4977203607559204,
+      "logits/rejected": -1.6133390665054321,
+      "logps/chosen": -134.21728515625,
+      "logps/rejected": -176.34817504882812,
+      "loss": 0.6141,
+      "rewards/accuracies": 0.800000011920929,
+      "rewards/chosen": 0.16645415127277374,
+      "rewards/margins": 0.17111308872699738,
+      "rewards/rejected": -0.004658935125917196,
+      "step": 260
+    },
+    {
+      "epoch": 0.6,
+      "grad_norm": 5.716956108430631,
+      "learning_rate": 2.0955450086180881e-07,
+      "logits/chosen": -1.5042165517807007,
+      "logits/rejected": -1.6705764532089233,
+      "logps/chosen": -128.129638671875,
+      "logps/rejected": -168.9424591064453,
+      "loss": 0.6087,
+      "rewards/accuracies": 0.856249988079071,
+      "rewards/chosen": 0.18247731029987335,
+      "rewards/margins": 0.19755597412586212,
+      "rewards/rejected": -0.015078653581440449,
+      "step": 270
+    },
+    {
+      "epoch": 0.62,
+      "grad_norm": 6.529628318945036,
+      "learning_rate": 1.9060462504063227e-07,
+      "logits/chosen": -1.6098215579986572,
+      "logits/rejected": -1.7113733291625977,
+      "logps/chosen": -152.48533630371094,
+      "logps/rejected": -169.33935546875,
+      "loss": 0.607,
+      "rewards/accuracies": 0.862500011920929,
+      "rewards/chosen": 0.1736232191324234,
+      "rewards/margins": 0.19309216737747192,
+      "rewards/rejected": -0.019468944519758224,
+      "step": 280
+    },
+    {
+      "epoch": 0.64,
+      "grad_norm": 5.964891862645329,
+      "learning_rate": 1.7201020337827556e-07,
+      "logits/chosen": -1.5367965698242188,
+      "logits/rejected": -1.6592748165130615,
+      "logps/chosen": -155.07492065429688,
+      "logps/rejected": -157.03305053710938,
+      "loss": 0.6077,
+      "rewards/accuracies": 0.824999988079071,
+      "rewards/chosen": 0.15863782167434692,
+      "rewards/margins": 0.16328461468219757,
+      "rewards/rejected": -0.004646780900657177,
+      "step": 290
+    },
+    {
+      "epoch": 0.66,
+      "grad_norm": 5.912017097517653,
+      "learning_rate": 1.5388251498553261e-07,
+      "logits/chosen": -1.5541818141937256,
+      "logits/rejected": -1.7061693668365479,
+      "logps/chosen": -151.648681640625,
+      "logps/rejected": -161.48257446289062,
+      "loss": 0.5967,
+      "rewards/accuracies": 0.875,
+      "rewards/chosen": 0.17861925065517426,
+      "rewards/margins": 0.2286207377910614,
+      "rewards/rejected": -0.050001464784145355,
+      "step": 300
+    },
+    {
+      "epoch": 0.66,
+      "eval_logits/chosen": -1.8740853071212769,
+      "eval_logits/rejected": -1.6454341411590576,
+      "eval_logps/chosen": -128.13084411621094,
+      "eval_logps/rejected": -150.56195068359375,
+      "eval_loss": 0.6526921391487122,
+      "eval_rewards/accuracies": 0.7159090638160706,
+      "eval_rewards/chosen": 0.16582368314266205,
+      "eval_rewards/margins": 0.09728217869997025,
+      "eval_rewards/rejected": 0.06854148954153061,
+      "eval_runtime": 64.2292,
+      "eval_samples_per_second": 10.494,
+      "eval_steps_per_second": 0.343,
+      "step": 300
+    },
+    {
+      "epoch": 0.69,
+      "grad_norm": 5.769946207769061,
+      "learning_rate": 1.3633004578800611e-07,
+      "logits/chosen": -1.6252295970916748,
+      "logits/rejected": -1.6870454549789429,
+      "logps/chosen": -145.02096557617188,
+      "logps/rejected": -175.39373779296875,
+      "loss": 0.6066,
+      "rewards/accuracies": 0.831250011920929,
+      "rewards/chosen": 0.1655764877796173,
+      "rewards/margins": 0.19951465725898743,
+      "rewards/rejected": -0.03393816202878952,
+      "step": 310
+    },
+    {
+      "epoch": 0.71,
+      "grad_norm": 6.460474591534718,
+      "learning_rate": 1.1945783928745184e-07,
+      "logits/chosen": -1.5879411697387695,
+      "logits/rejected": -1.6782830953598022,
+      "logps/chosen": -148.7205810546875,
+      "logps/rejected": -193.89002990722656,
+      "loss": 0.6004,
+      "rewards/accuracies": 0.893750011920929,
+      "rewards/chosen": 0.179766446352005,
+      "rewards/margins": 0.21573027968406677,
+      "rewards/rejected": -0.03596385195851326,
+      "step": 320
+    },
+    {
+      "epoch": 0.73,
+      "grad_norm": 5.905172245521576,
+      "learning_rate": 1.0336686792445423e-07,
+      "logits/chosen": -1.6096925735473633,
+      "logits/rejected": -1.7555859088897705,
+      "logps/chosen": -141.56198120117188,
+      "logps/rejected": -160.07009887695312,
+      "loss": 0.599,
+      "rewards/accuracies": 0.856249988079071,
+      "rewards/chosen": 0.1507815718650818,
+      "rewards/margins": 0.19255796074867249,
+      "rewards/rejected": -0.0417763814330101,
+      "step": 330
+    },
+    {
+      "epoch": 0.75,
+      "grad_norm": 5.690958514404008,
+      "learning_rate": 8.81534288045431e-08,
+      "logits/chosen": -1.6023919582366943,
+      "logits/rejected": -1.743801474571228,
+      "logps/chosen": -155.75018310546875,
+      "logps/rejected": -177.66357421875,
+      "loss": 0.5925,
+      "rewards/accuracies": 0.8187500238418579,
+      "rewards/chosen": 0.18638750910758972,
+      "rewards/margins": 0.23221752047538757,
+      "rewards/rejected": -0.045830003917217255,
+      "step": 340
+    },
+    {
+      "epoch": 0.77,
+      "grad_norm": 5.590062248471453,
+      "learning_rate": 7.390856740405091e-08,
+      "logits/chosen": -1.6454668045043945,
+      "logits/rejected": -1.7857444286346436,
+      "logps/chosen": -164.88294982910156,
+      "logps/rejected": -179.4400177001953,
+      "loss": 0.5915,
+      "rewards/accuracies": 0.8125,
+      "rewards/chosen": 0.16329999268054962,
+      "rewards/margins": 0.2226523905992508,
+      "rewards/rejected": -0.059352416545152664,
+      "step": 350
+    },
+    {
+      "epoch": 0.8,
+      "grad_norm": 5.731203070847599,
+      "learning_rate": 6.071753270457065e-08,
+      "logits/chosen": -1.5961381196975708,
+      "logits/rejected": -1.81313157081604,
+      "logps/chosen": -150.7357940673828,
+      "logps/rejected": -198.715087890625,
+      "loss": 0.5961,
+      "rewards/accuracies": 0.8125,
+      "rewards/chosen": 0.13613273203372955,
+      "rewards/margins": 0.21987445652484894,
+      "rewards/rejected": -0.08374173194169998,
+      "step": 360
+    },
+    {
+      "epoch": 0.82,
+      "grad_norm": 5.979073484319341,
+      "learning_rate": 4.865926701678352e-08,
+      "logits/chosen": -1.6725813150405884,
+      "logits/rejected": -1.7875378131866455,
+      "logps/chosen": -145.93125915527344,
+      "logps/rejected": -173.05287170410156,
+      "loss": 0.5926,
+      "rewards/accuracies": 0.893750011920929,
+      "rewards/chosen": 0.1760322004556656,
+      "rewards/margins": 0.2271731197834015,
+      "rewards/rejected": -0.0511409156024456,
+      "step": 370
+    },
+    {
+      "epoch": 0.84,
+      "grad_norm": 6.048552200178175,
+      "learning_rate": 3.780593354682826e-08,
+      "logits/chosen": -1.589115858078003,
+      "logits/rejected": -1.8489913940429688,
+      "logps/chosen": -151.71109008789062,
+      "logps/rejected": -172.7513885498047,
+      "loss": 0.6031,
+      "rewards/accuracies": 0.856249988079071,
+      "rewards/chosen": 0.15712784230709076,
+      "rewards/margins": 0.2085273712873459,
+      "rewards/rejected": -0.05139952152967453,
+      "step": 380
+    },
+    {
+      "epoch": 0.86,
+      "grad_norm": 6.206696499253566,
+      "learning_rate": 2.8222484532511166e-08,
+      "logits/chosen": -1.626539945602417,
+      "logits/rejected": -1.8017622232437134,
+      "logps/chosen": -152.2079620361328,
+      "logps/rejected": -174.45358276367188,
+      "loss": 0.5972,
+      "rewards/accuracies": 0.84375,
+      "rewards/chosen": 0.15285542607307434,
+      "rewards/margins": 0.1900016814470291,
+      "rewards/rejected": -0.037146251648664474,
+      "step": 390
+    },
+    {
+      "epoch": 0.88,
+      "grad_norm": 6.415272732792148,
+      "learning_rate": 1.9966272533864183e-08,
+      "logits/chosen": -1.6374372243881226,
+      "logits/rejected": -1.8020416498184204,
+      "logps/chosen": -154.29025268554688,
+      "logps/rejected": -180.31027221679688,
+      "loss": 0.5979,
+      "rewards/accuracies": 0.8374999761581421,
+      "rewards/chosen": 0.17826148867607117,
+      "rewards/margins": 0.23326411843299866,
+      "rewards/rejected": -0.055002618581056595,
+      "step": 400
+    },
+    {
+      "epoch": 0.88,
+      "eval_logits/chosen": -1.9158859252929688,
+      "eval_logits/rejected": -1.6898193359375,
+      "eval_logps/chosen": -129.27784729003906,
+      "eval_logps/rejected": -152.41265869140625,
+      "eval_loss": 0.6502383947372437,
+      "eval_rewards/accuracies": 0.7272727489471436,
+      "eval_rewards/chosen": 0.15435346961021423,
+      "eval_rewards/margins": 0.10431905090808868,
+      "eval_rewards/rejected": 0.05003440007567406,
+      "eval_runtime": 64.2154,
+      "eval_samples_per_second": 10.496,
+      "eval_steps_per_second": 0.343,
+      "step": 400
+    },
+    {
+      "epoch": 0.91,
+      "grad_norm": 5.862851313494248,
+      "learning_rate": 1.3086707204299413e-08,
+      "logits/chosen": -1.5973860025405884,
+      "logits/rejected": -1.8493759632110596,
+      "logps/chosen": -151.65713500976562,
+      "logps/rejected": -169.72344970703125,
+      "loss": 0.5886,
+      "rewards/accuracies": 0.8187500238418579,
+      "rewards/chosen": 0.13965830206871033,
+      "rewards/margins": 0.21700701117515564,
+      "rewards/rejected": -0.07734867930412292,
+      "step": 410
+    },
+    {
+      "epoch": 0.93,
+      "grad_norm": 6.0290500426097875,
+      "learning_rate": 7.624959596427145e-09,
+      "logits/chosen": -1.6760833263397217,
+      "logits/rejected": -1.7056286334991455,
+      "logps/chosen": -150.13731384277344,
+      "logps/rejected": -179.03042602539062,
+      "loss": 0.5987,
+      "rewards/accuracies": 0.84375,
+      "rewards/chosen": 0.15684738755226135,
+      "rewards/margins": 0.21405240893363953,
+      "rewards/rejected": -0.057205021381378174,
+      "step": 420
+    },
+    {
+      "epoch": 0.95,
+      "grad_norm": 5.566801847010683,
+      "learning_rate": 3.6137157721330967e-09,
+      "logits/chosen": -1.6355880498886108,
+      "logits/rejected": -1.7391598224639893,
+      "logps/chosen": -138.0072479248047,
+      "logps/rejected": -193.86627197265625,
+      "loss": 0.5925,
+      "rewards/accuracies": 0.90625,
+      "rewards/chosen": 0.18388640880584717,
+      "rewards/margins": 0.2870241403579712,
+      "rewards/rejected": -0.10313773155212402,
+      "step": 430
+    },
+    {
+      "epoch": 0.97,
+      "grad_norm": 5.3200698426502,
+      "learning_rate": 1.0769811914444204e-09,
+      "logits/chosen": -1.5580041408538818,
+      "logits/rejected": -1.690857172012329,
+      "logps/chosen": -135.1139373779297,
+      "logps/rejected": -164.4469757080078,
+      "loss": 0.588,
+      "rewards/accuracies": 0.856249988079071,
+      "rewards/chosen": 0.159897118806839,
+      "rewards/margins": 0.23895928263664246,
+      "rewards/rejected": -0.07906216382980347,
+      "step": 440
+    },
+    {
+      "epoch": 1.0,
+      "grad_norm": 5.753301995096959,
+      "learning_rate": 2.993705082879328e-11,
+      "logits/chosen": -1.6066157817840576,
+      "logits/rejected": -1.796430230140686,
+      "logps/chosen": -155.62608337402344,
+      "logps/rejected": -184.45956420898438,
+      "loss": 0.5864,
+      "rewards/accuracies": 0.8999999761581421,
+      "rewards/chosen": 0.15883687138557434,
+      "rewards/margins": 0.24033670127391815,
+      "rewards/rejected": -0.0814998596906662,
+      "step": 450
+    },
+    {
+      "epoch": 1.0,
+      "step": 452,
       "total_flos": 0.0,
-      "train_loss": 0.41232719756009284,
-      "train_runtime": 892.3177,
-      "train_samples_per_second": 16.336,
-      "train_steps_per_second": 0.128
+      "train_loss": 0.6286900518214809,
+      "train_runtime": 6365.7753,
+      "train_samples_per_second": 4.541,
+      "train_steps_per_second": 0.071
     }
   ],
   "logging_steps": 10,
-  "max_steps": 114,
+  "max_steps": 452,
   "num_input_tokens_seen": 0,
-  "num_train_epochs": 3,
+  "num_train_epochs": 1,
   "save_steps": 100,
   "total_flos": 0.0,
   "train_batch_size": 8,