Model save

README.md

@@ -1,4 +1,5 @@
 ---
+base_model: princeton-nlp/Mistral-7B-Base-SFT-DPO
 tags:
 - trl
 - dpo
@@ -13,7 +14,7 @@ should probably proofread and complete it, then remove this comment. -->
 
 # zephyr-7b-dpo-full
 
-This model was trained from scratch on the None dataset.
+This model is a fine-tuned version of [princeton-nlp/Mistral-7B-Base-SFT-DPO](https://huggingface.co/princeton-nlp/Mistral-7B-Base-SFT-DPO) on the None dataset.
 
 ## Model description
 
@@ -32,7 +33,7 @@ More information needed
 ### Training hyperparameters
 
 The following hyperparameters were used during training:
-- learning_rate: 1e-08
+- learning_rate: 5e-07
 - train_batch_size: 8
 - eval_batch_size: 8
 - seed: 42
@@ -44,7 +45,7 @@ The following hyperparameters were used during training:
 - optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
 - lr_scheduler_type: cosine
 - lr_scheduler_warmup_ratio: 0.1
-- num_epochs: 2
+- num_epochs: 1
 
 ### Training results
 

all_results.json

@@ -1,8 +1,8 @@
 {
-    "epoch": 2.0,
-    "train_loss": 0.26489045179408527,
-    "train_runtime": 4520.7899,
-    "train_samples": 47095,
-    "train_samples_per_second": 20.835,
-    "train_steps_per_second": 0.081
+    "epoch": 1.0,
+    "train_loss": 0.33391942269093283,
+    "train_runtime": 5319.9853,
+    "train_samples": 47302,
+    "train_samples_per_second": 8.891,
+    "train_steps_per_second": 0.035
 }

config.json

@@ -20,7 +20,7 @@
   "sliding_window": 4096,
   "tie_word_embeddings": false,
   "torch_dtype": "bfloat16",
-  "transformers_version": "4.41.1",
+  "transformers_version": "4.39.3",
   "use_cache": false,
   "vocab_size": 32000
 }

model-00001-of-00003.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:89021f925795319ec6cd79c57490ff100b4d209128279330c40d869a1257ea07
+oid sha256:082f207ea835a592e1e8eb2cd383fe7080164e426d45e909c8d42aca68d2441d
 size 4943162336

model-00002-of-00003.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:cc9bb6d1ff6e5e594487f490ec146b7666201b9e396d8c36acadcf28b8944e2e
+oid sha256:1f1ab93164931c699c31d8f9a6662aa4d21ae5163cf4ca656ec82228d11a555b
 size 4999819336

model-00003-of-00003.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:d64a4a2eb8fd994c6c2e00218546b04c07ec10f9918e36094d7621932e29d265
+oid sha256:241e2b2864b464b12cb35da89659de3b3cce018b44fd5b62ae428621c0077ca8
 size 4540516344

runs/Jun21_00-12-13_n136-100-194/events.out.tfevents.1718899955.n136-100-194.494165.0

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:c97c12876d1c4d23198b5d9f91d061f458ebd51aac69372e75f566ce403f78b3
-size 12321
+oid sha256:30db6263136339554ee965a3c30d4aa91ded54ecf93706de542e54b6c0c8595d
+size 18155

tokenizer.json

@@ -134,7 +134,6 @@
     "end_of_word_suffix": null,
     "fuse_unk": true,
     "byte_fallback": true,
-    "ignore_merges": false,
     "vocab": {
       "<unk>": 0,
       "<s>": 1,

train_results.json

@@ -1,8 +1,8 @@
 {
-    "epoch": 2.0,
-    "train_loss": 0.26489045179408527,
-    "train_runtime": 4520.7899,
-    "train_samples": 47095,
-    "train_samples_per_second": 20.835,
-    "train_steps_per_second": 0.081
+    "epoch": 1.0,
+    "train_loss": 0.33391942269093283,
+    "train_runtime": 5319.9853,
+    "train_samples": 47302,
+    "train_samples_per_second": 8.891,
+    "train_steps_per_second": 0.035
 }

trainer_state.json

@@ -1,22 +1,22 @@
 {
   "best_metric": null,
   "best_model_checkpoint": null,
-  "epoch": 2.0,
+  "epoch": 1.0,
   "eval_steps": 500,
-  "global_step": 368,
+  "global_step": 185,
   "is_hyper_param_search": false,
   "is_local_process_zero": true,
   "is_world_process_zero": true,
   "log_history": [
     {
       "epoch": 0.01,
-      "grad_norm": 2733.5875966596914,
-      "learning_rate": 2.702702702702703e-10,
-      "logits/chosen": -1.3332719802856445,
-      "logits/rejected": -1.246394395828247,
-      "logps/chosen": -286.9539794921875,
-      "logps/rejected": -263.3782958984375,
-      "loss": 0.7283,
+      "grad_norm": 171.98892218238854,
+      "learning_rate": 2.6315789473684208e-08,
+      "logits/chosen": -0.1266070306301117,
+      "logits/rejected": 0.7204304933547974,
+      "logps/chosen": -319.01666259765625,
+      "logps/rejected": -252.47039794921875,
+      "loss": 0.6916,
       "rewards/accuracies": 0.0,
       "rewards/chosen": 0.0,
       "rewards/margins": 0.0,
@@ -25,558 +25,288 @@
     },
     {
       "epoch": 0.05,
-      "grad_norm": 3426.28374639058,
-      "learning_rate": 2.702702702702703e-09,
-      "logits/chosen": -1.617490530014038,
-      "logits/rejected": -1.3964743614196777,
-      "logps/chosen": -342.53607177734375,
-      "logps/rejected": -294.5452575683594,
-      "loss": 0.9019,
-      "rewards/accuracies": 0.4375,
-      "rewards/chosen": -0.0025859144516289234,
-      "rewards/margins": 0.014665775932371616,
-      "rewards/rejected": -0.017251690849661827,
+      "grad_norm": 158.2614639136714,
+      "learning_rate": 2.631578947368421e-07,
+      "logits/chosen": -0.3861861824989319,
+      "logits/rejected": 0.33749374747276306,
+      "logps/chosen": -266.4891052246094,
+      "logps/rejected": -224.11000061035156,
+      "loss": 0.6758,
+      "rewards/accuracies": 0.5520833134651184,
+      "rewards/chosen": -0.03102089650928974,
+      "rewards/margins": 0.034922875463962555,
+      "rewards/rejected": -0.06594377011060715,
       "step": 10
     },
     {
       "epoch": 0.11,
-      "grad_norm": 3068.517544136213,
-      "learning_rate": 5.405405405405406e-09,
-      "logits/chosen": -1.4905732870101929,
-      "logits/rejected": -1.3132953643798828,
-      "logps/chosen": -314.7499084472656,
-      "logps/rejected": -279.27752685546875,
-      "loss": 0.9225,
-      "rewards/accuracies": 0.4937500059604645,
-      "rewards/chosen": 0.03222837299108505,
-      "rewards/margins": -0.024351513013243675,
-      "rewards/rejected": 0.05657988786697388,
+      "grad_norm": 104.41587535161224,
+      "learning_rate": 4.999552306674344e-07,
+      "logits/chosen": -0.24374540150165558,
+      "logits/rejected": 0.8117060661315918,
+      "logps/chosen": -289.02911376953125,
+      "logps/rejected": -250.653564453125,
+      "loss": 0.478,
+      "rewards/accuracies": 0.793749988079071,
+      "rewards/chosen": -0.4349571764469147,
+      "rewards/margins": 1.263426661491394,
+      "rewards/rejected": -1.6983836889266968,
       "step": 20
     },
     {
       "epoch": 0.16,
-      "grad_norm": 3019.197796031677,
-      "learning_rate": 8.108108108108109e-09,
-      "logits/chosen": -1.5479624271392822,
-      "logits/rejected": -1.3802028894424438,
-      "logps/chosen": -324.89044189453125,
-      "logps/rejected": -286.2395324707031,
-      "loss": 0.9562,
-      "rewards/accuracies": 0.5218750238418579,
-      "rewards/chosen": 0.031223665922880173,
-      "rewards/margins": -0.0476018562912941,
-      "rewards/rejected": 0.07882551848888397,
+      "grad_norm": 98.86884631406178,
+      "learning_rate": 4.946022852363932e-07,
+      "logits/chosen": -0.2871348261833191,
+      "logits/rejected": 0.6740838289260864,
+      "logps/chosen": -281.1429748535156,
+      "logps/rejected": -271.7496032714844,
+      "loss": 0.4067,
+      "rewards/accuracies": 0.800000011920929,
+      "rewards/chosen": -1.6509501934051514,
+      "rewards/margins": 2.906687021255493,
+      "rewards/rejected": -4.5576372146606445,
       "step": 30
     },
     {
       "epoch": 0.22,
-      "grad_norm": 2861.837005400921,
-      "learning_rate": 9.997973265157192e-09,
-      "logits/chosen": -1.5354044437408447,
-      "logits/rejected": -1.3576419353485107,
-      "logps/chosen": -325.43408203125,
-      "logps/rejected": -285.6204528808594,
-      "loss": 0.9309,
-      "rewards/accuracies": 0.484375,
-      "rewards/chosen": -0.08140890300273895,
-      "rewards/margins": -0.13951030373573303,
-      "rewards/rejected": 0.05810140445828438,
+      "grad_norm": 95.68866413164287,
+      "learning_rate": 4.805146507594034e-07,
+      "logits/chosen": -0.5090769529342651,
+      "logits/rejected": 0.5341213345527649,
+      "logps/chosen": -283.4405517578125,
+      "logps/rejected": -268.97686767578125,
+      "loss": 0.371,
+      "rewards/accuracies": 0.8500000238418579,
+      "rewards/chosen": -2.4618725776672363,
+      "rewards/margins": 3.340365171432495,
+      "rewards/rejected": -5.802238464355469,
       "step": 40
     },
     {
       "epoch": 0.27,
-      "grad_norm": 3285.7135249709568,
-      "learning_rate": 9.961988113473708e-09,
-      "logits/chosen": -1.534355640411377,
-      "logits/rejected": -1.3875898122787476,
-      "logps/chosen": -337.02044677734375,
-      "logps/rejected": -297.35101318359375,
-      "loss": 0.88,
-      "rewards/accuracies": 0.546875,
-      "rewards/chosen": 0.0009558796882629395,
-      "rewards/margins": 0.08057532459497452,
-      "rewards/rejected": -0.07961944490671158,
+      "grad_norm": 103.23233983894589,
+      "learning_rate": 4.581953932909403e-07,
+      "logits/chosen": -0.4626421332359314,
+      "logits/rejected": 0.5320831537246704,
+      "logps/chosen": -313.1284484863281,
+      "logps/rejected": -299.7115173339844,
+      "loss": 0.335,
+      "rewards/accuracies": 0.862500011920929,
+      "rewards/chosen": -2.947516441345215,
+      "rewards/margins": 3.4062907695770264,
+      "rewards/rejected": -6.353806495666504,
       "step": 50
     },
     {
-      "epoch": 0.33,
-      "grad_norm": 2547.1665545835035,
-      "learning_rate": 9.881337335184878e-09,
-      "logits/chosen": -1.5822935104370117,
-      "logits/rejected": -1.4333903789520264,
-      "logps/chosen": -319.79644775390625,
-      "logps/rejected": -285.0381164550781,
-      "loss": 0.8105,
-      "rewards/accuracies": 0.6156250238418579,
-      "rewards/chosen": 0.09285839647054672,
-      "rewards/margins": 0.40408092737197876,
-      "rewards/rejected": -0.31122252345085144,
+      "epoch": 0.32,
+      "grad_norm": 91.31338474042249,
+      "learning_rate": 4.284415281717847e-07,
+      "logits/chosen": -0.3830726444721222,
+      "logits/rejected": 0.7034914493560791,
+      "logps/chosen": -302.44549560546875,
+      "logps/rejected": -295.2908020019531,
+      "loss": 0.2941,
+      "rewards/accuracies": 0.890625,
+      "rewards/chosen": -2.427272081375122,
+      "rewards/margins": 3.3501389026641846,
+      "rewards/rejected": -5.777410507202148,
       "step": 60
     },
     {
       "epoch": 0.38,
-      "grad_norm": 3178.8313195657583,
-      "learning_rate": 9.756746912994832e-09,
-      "logits/chosen": -1.5119212865829468,
-      "logits/rejected": -1.350838541984558,
-      "logps/chosen": -312.1349182128906,
-      "logps/rejected": -275.08660888671875,
-      "loss": 0.7993,
-      "rewards/accuracies": 0.5874999761581421,
-      "rewards/chosen": -0.05482473224401474,
-      "rewards/margins": 0.165395587682724,
-      "rewards/rejected": -0.22022032737731934,
+      "grad_norm": 108.21771645007362,
+      "learning_rate": 3.923155588020165e-07,
+      "logits/chosen": -0.050761766731739044,
+      "logits/rejected": 1.1738256216049194,
+      "logps/chosen": -279.0822448730469,
+      "logps/rejected": -271.3674011230469,
+      "loss": 0.3118,
+      "rewards/accuracies": 0.859375,
+      "rewards/chosen": -2.9542946815490723,
+      "rewards/margins": 3.31215238571167,
+      "rewards/rejected": -6.266446590423584,
       "step": 70
     },
     {
       "epoch": 0.43,
-      "grad_norm": 2596.199281277795,
-      "learning_rate": 9.589338354885628e-09,
-      "logits/chosen": -1.5992329120635986,
-      "logits/rejected": -1.4463211297988892,
-      "logps/chosen": -323.2821960449219,
-      "logps/rejected": -288.0993347167969,
-      "loss": 0.7772,
-      "rewards/accuracies": 0.574999988079071,
-      "rewards/chosen": 0.20231203734874725,
-      "rewards/margins": 0.4638887345790863,
-      "rewards/rejected": -0.26157671213150024,
+      "grad_norm": 110.48344039822193,
+      "learning_rate": 3.511075348989692e-07,
+      "logits/chosen": -0.02379416488111019,
+      "logits/rejected": 0.9985305666923523,
+      "logps/chosen": -291.3994140625,
+      "logps/rejected": -278.81207275390625,
+      "loss": 0.3145,
+      "rewards/accuracies": 0.862500011920929,
+      "rewards/chosen": -1.9188703298568726,
+      "rewards/margins": 3.239673614501953,
+      "rewards/rejected": -5.158544063568115,
       "step": 80
     },
     {
       "epoch": 0.49,
-      "grad_norm": 2404.162696635418,
-      "learning_rate": 9.380618598797472e-09,
-      "logits/chosen": -1.6108148097991943,
-      "logits/rejected": -1.4147026538848877,
-      "logps/chosen": -319.95526123046875,
-      "logps/rejected": -281.7666015625,
-      "loss": 0.7649,
-      "rewards/accuracies": 0.640625,
-      "rewards/chosen": 0.2500740885734558,
-      "rewards/margins": 0.5574880838394165,
-      "rewards/rejected": -0.3074139356613159,
+      "grad_norm": 94.44668679211257,
+      "learning_rate": 3.062889851306735e-07,
+      "logits/chosen": 0.15241345763206482,
+      "logits/rejected": 1.204730749130249,
+      "logps/chosen": -285.8970947265625,
+      "logps/rejected": -274.2763671875,
+      "loss": 0.3256,
+      "rewards/accuracies": 0.859375,
+      "rewards/chosen": -2.3829667568206787,
+      "rewards/margins": 3.3176727294921875,
+      "rewards/rejected": -5.700639724731445,
       "step": 90
     },
     {
       "epoch": 0.54,
-      "grad_norm": 2500.271614922207,
-      "learning_rate": 9.132466447838596e-09,
-      "logits/chosen": -1.542976975440979,
-      "logits/rejected": -1.3676128387451172,
-      "logps/chosen": -321.9007263183594,
-      "logps/rejected": -282.65899658203125,
-      "loss": 0.7305,
-      "rewards/accuracies": 0.659375011920929,
-      "rewards/chosen": 0.4254188537597656,
-      "rewards/margins": 0.763160228729248,
-      "rewards/rejected": -0.3377414047718048,
+      "grad_norm": 72.25131318661623,
+      "learning_rate": 2.594603691794176e-07,
+      "logits/chosen": 0.017316246405243874,
+      "logits/rejected": 1.112657070159912,
+      "logps/chosen": -291.90631103515625,
+      "logps/rejected": -277.5140686035156,
+      "loss": 0.2907,
+      "rewards/accuracies": 0.846875011920929,
+      "rewards/chosen": -2.7051641941070557,
+      "rewards/margins": 3.0991756916046143,
+      "rewards/rejected": -5.804339408874512,
       "step": 100
     },
     {
-      "epoch": 0.6,
-      "grad_norm": 2525.572173445101,
-      "learning_rate": 8.847115658129039e-09,
-      "logits/chosen": -1.512939453125,
-      "logits/rejected": -1.3849382400512695,
-      "logps/chosen": -318.14813232421875,
-      "logps/rejected": -287.1947937011719,
-      "loss": 0.7164,
-      "rewards/accuracies": 0.628125011920929,
-      "rewards/chosen": 0.3916184604167938,
-      "rewards/margins": 0.657262921333313,
-      "rewards/rejected": -0.265644371509552,
+      "epoch": 0.59,
+      "grad_norm": 106.17479973453032,
+      "learning_rate": 2.1229392570965654e-07,
+      "logits/chosen": 0.5481065511703491,
+      "logits/rejected": 1.4057379961013794,
+      "logps/chosen": -290.0019226074219,
+      "logps/rejected": -288.4178161621094,
+      "loss": 0.2795,
+      "rewards/accuracies": 0.859375,
+      "rewards/chosen": -3.1080322265625,
+      "rewards/margins": 3.114968776702881,
+      "rewards/rejected": -6.223001003265381,
       "step": 110
     },
     {
       "epoch": 0.65,
-      "grad_norm": 2498.220455730566,
-      "learning_rate": 8.527134831514116e-09,
-      "logits/chosen": -1.5739517211914062,
-      "logits/rejected": -1.41860032081604,
-      "logps/chosen": -331.3175354003906,
-      "logps/rejected": -297.8718566894531,
-      "loss": 0.7018,
-      "rewards/accuracies": 0.637499988079071,
-      "rewards/chosen": 0.6175383925437927,
-      "rewards/margins": 0.6906081438064575,
-      "rewards/rejected": -0.07306969165802002,
+      "grad_norm": 88.25415485320248,
+      "learning_rate": 1.6647395712565254e-07,
+      "logits/chosen": 0.10530638694763184,
+      "logits/rejected": 1.3136330842971802,
+      "logps/chosen": -303.7025451660156,
+      "logps/rejected": -291.4312438964844,
+      "loss": 0.3024,
+      "rewards/accuracies": 0.8812500238418579,
+      "rewards/chosen": -2.87852144241333,
+      "rewards/margins": 3.439791440963745,
+      "rewards/rejected": -6.318312644958496,
       "step": 120
     },
     {
-      "epoch": 0.71,
-      "grad_norm": 2170.854176631694,
-      "learning_rate": 8.175404294144481e-09,
-      "logits/chosen": -1.616276502609253,
-      "logits/rejected": -1.429518699645996,
-      "logps/chosen": -317.1609802246094,
-      "logps/rejected": -271.5557861328125,
-      "loss": 0.6719,
-      "rewards/accuracies": 0.6625000238418579,
-      "rewards/chosen": 0.6955646872520447,
-      "rewards/margins": 0.8560658693313599,
-      "rewards/rejected": -0.160501167178154,
+      "epoch": 0.7,
+      "grad_norm": 81.41509800140894,
+      "learning_rate": 1.2363668353585485e-07,
+      "logits/chosen": 0.025721266865730286,
+      "logits/rejected": 1.1706856489181519,
+      "logps/chosen": -291.2774963378906,
+      "logps/rejected": -280.7757873535156,
+      "loss": 0.2712,
+      "rewards/accuracies": 0.903124988079071,
+      "rewards/chosen": -2.8241302967071533,
+      "rewards/margins": 3.6137948036193848,
+      "rewards/rejected": -6.437924385070801,
       "step": 130
     },
     {
       "epoch": 0.76,
-      "grad_norm": 2147.543587634845,
-      "learning_rate": 7.79509016905158e-09,
-      "logits/chosen": -1.5726101398468018,
-      "logits/rejected": -1.4245671033859253,
-      "logps/chosen": -331.12109375,
-      "logps/rejected": -294.2488098144531,
-      "loss": 0.6686,
-      "rewards/accuracies": 0.6968749761581421,
-      "rewards/chosen": 0.9418653249740601,
-      "rewards/margins": 0.9832620620727539,
-      "rewards/rejected": -0.04139674827456474,
+      "grad_norm": 76.56961564493653,
+      "learning_rate": 8.53118137245516e-08,
+      "logits/chosen": 0.24798288941383362,
+      "logits/rejected": 1.3128881454467773,
+      "logps/chosen": -298.71783447265625,
+      "logps/rejected": -297.16790771484375,
+      "loss": 0.2607,
+      "rewards/accuracies": 0.8656250238418579,
+      "rewards/chosen": -2.95615553855896,
+      "rewards/margins": 3.7294158935546875,
+      "rewards/rejected": -6.685571193695068,
       "step": 140
     },
     {
-      "epoch": 0.82,
-      "grad_norm": 2298.0130679506387,
-      "learning_rate": 7.389615876105773e-09,
-      "logits/chosen": -1.5536715984344482,
-      "logits/rejected": -1.4254592657089233,
-      "logps/chosen": -314.55267333984375,
-      "logps/rejected": -291.81536865234375,
-      "loss": 0.6793,
-      "rewards/accuracies": 0.675000011920929,
-      "rewards/chosen": 1.0258944034576416,
-      "rewards/margins": 0.981005072593689,
-      "rewards/rejected": 0.04488936811685562,
+      "epoch": 0.81,
+      "grad_norm": 101.60579173655283,
+      "learning_rate": 5.2867919617408553e-08,
+      "logits/chosen": 0.16610342264175415,
+      "logits/rejected": 1.297738790512085,
+      "logps/chosen": -296.17230224609375,
+      "logps/rejected": -285.56707763671875,
+      "loss": 0.2777,
+      "rewards/accuracies": 0.903124988079071,
+      "rewards/chosen": -2.7571194171905518,
+      "rewards/margins": 3.536668062210083,
+      "rewards/rejected": -6.293786525726318,
       "step": 150
     },
     {
-      "epoch": 0.87,
-      "grad_norm": 2220.286727308493,
-      "learning_rate": 6.962631315901861e-09,
-      "logits/chosen": -1.5181314945220947,
-      "logits/rejected": -1.4019381999969482,
-      "logps/chosen": -318.02752685546875,
-      "logps/rejected": -291.03936767578125,
-      "loss": 0.6742,
-      "rewards/accuracies": 0.628125011920929,
-      "rewards/chosen": 0.9784830808639526,
-      "rewards/margins": 0.819505512714386,
-      "rewards/rejected": 0.15897764265537262,
+      "epoch": 0.86,
+      "grad_norm": 92.310593955402,
+      "learning_rate": 2.7463564905650853e-08,
+      "logits/chosen": 0.06046704202890396,
+      "logits/rejected": 1.0854153633117676,
+      "logps/chosen": -297.1445007324219,
+      "logps/rejected": -291.33868408203125,
+      "loss": 0.2684,
+      "rewards/accuracies": 0.893750011920929,
+      "rewards/chosen": -2.6816515922546387,
+      "rewards/margins": 3.552661418914795,
+      "rewards/rejected": -6.234313011169434,
       "step": 160
     },
     {
       "epoch": 0.92,
-      "grad_norm": 2125.4983562302123,
-      "learning_rate": 6.517980014965139e-09,
-      "logits/chosen": -1.5958881378173828,
-      "logits/rejected": -1.4071909189224243,
-      "logps/chosen": -331.4378356933594,
-      "logps/rejected": -289.5236511230469,
-      "loss": 0.6456,
-      "rewards/accuracies": 0.731249988079071,
-      "rewards/chosen": 1.1029136180877686,
-      "rewards/margins": 1.160766363143921,
-      "rewards/rejected": -0.05785265564918518,
+      "grad_norm": 83.71834684366553,
+      "learning_rate": 1.0005933014019307e-08,
+      "logits/chosen": 0.15604642033576965,
+      "logits/rejected": 1.338841199874878,
+      "logps/chosen": -298.0588684082031,
+      "logps/rejected": -293.54638671875,
+      "loss": 0.2745,
+      "rewards/accuracies": 0.903124988079071,
+      "rewards/chosen": -3.0692405700683594,
+      "rewards/margins": 3.527927875518799,
+      "rewards/rejected": -6.59716796875,
       "step": 170
     },
     {
-      "epoch": 0.98,
-      "grad_norm": 2116.9259500110184,
-      "learning_rate": 6.059664528022266e-09,
-      "logits/chosen": -1.5962104797363281,
-      "logits/rejected": -1.445967197418213,
-      "logps/chosen": -315.10467529296875,
-      "logps/rejected": -276.73443603515625,
-      "loss": 0.6191,
-      "rewards/accuracies": 0.6812499761581421,
-      "rewards/chosen": 1.123002052307129,
-      "rewards/margins": 1.1910655498504639,
-      "rewards/rejected": -0.0680634081363678,
+      "epoch": 0.97,
+      "grad_norm": 78.86616344216218,
+      "learning_rate": 1.1184317978602808e-09,
+      "logits/chosen": -0.07575028389692307,
+      "logits/rejected": 1.0216057300567627,
+      "logps/chosen": -288.5888366699219,
+      "logps/rejected": -287.2474670410156,
+      "loss": 0.3031,
+      "rewards/accuracies": 0.90625,
+      "rewards/chosen": -2.831172466278076,
+      "rewards/margins": 3.852785587310791,
+      "rewards/rejected": -6.683958530426025,
       "step": 180
     },
     {
-      "epoch": 1.03,
-      "grad_norm": 2067.8979230397717,
-      "learning_rate": 5.591810408770492e-09,
-      "logits/chosen": -1.55275559425354,
-      "logits/rejected": -1.3787180185317993,
-      "logps/chosen": -315.572509765625,
-      "logps/rejected": -278.71087646484375,
-      "loss": 0.6052,
-      "rewards/accuracies": 0.71875,
-      "rewards/chosen": 1.2037546634674072,
-      "rewards/margins": 1.2858160734176636,
-      "rewards/rejected": -0.0820615291595459,
-      "step": 190
-    },
-    {
-      "epoch": 1.09,
-      "grad_norm": 2203.6189854484824,
-      "learning_rate": 5.118629073464423e-09,
-      "logits/chosen": -1.5673738718032837,
-      "logits/rejected": -1.3565856218338013,
-      "logps/chosen": -325.91680908203125,
-      "logps/rejected": -282.65869140625,
-      "loss": 0.6024,
-      "rewards/accuracies": 0.746874988079071,
-      "rewards/chosen": 1.4020355939865112,
-      "rewards/margins": 1.1880966424942017,
-      "rewards/rejected": 0.21393892168998718,
-      "step": 200
-    },
-    {
-      "epoch": 1.14,
-      "grad_norm": 2263.504332795979,
-      "learning_rate": 4.644379891605983e-09,
-      "logits/chosen": -1.611310601234436,
-      "logits/rejected": -1.4343440532684326,
-      "logps/chosen": -324.752197265625,
-      "logps/rejected": -291.36102294921875,
-      "loss": 0.5985,
-      "rewards/accuracies": 0.699999988079071,
-      "rewards/chosen": 1.2598803043365479,
-      "rewards/margins": 1.2701908349990845,
-      "rewards/rejected": -0.010310685262084007,
-      "step": 210
-    },
-    {
-      "epoch": 1.2,
-      "grad_norm": 2324.309417872748,
-      "learning_rate": 4.173331844980362e-09,
-      "logits/chosen": -1.5291264057159424,
-      "logits/rejected": -1.4033840894699097,
-      "logps/chosen": -323.9982604980469,
-      "logps/rejected": -293.4136047363281,
-      "loss": 0.5948,
-      "rewards/accuracies": 0.7281249761581421,
-      "rewards/chosen": 1.1985571384429932,
-      "rewards/margins": 1.105764627456665,
-      "rewards/rejected": 0.09279236942529678,
-      "step": 220
-    },
-    {
-      "epoch": 1.25,
-      "grad_norm": 2275.313721629071,
-      "learning_rate": 3.7097251001664824e-09,
-      "logits/chosen": -1.5342741012573242,
-      "logits/rejected": -1.3754017353057861,
-      "logps/chosen": -323.9897766113281,
-      "logps/rejected": -287.0173645019531,
-      "loss": 0.577,
-      "rewards/accuracies": 0.71875,
-      "rewards/chosen": 1.2511075735092163,
-      "rewards/margins": 1.3087403774261475,
-      "rewards/rejected": -0.05763290077447891,
-      "step": 230
-    },
-    {
-      "epoch": 1.3,
-      "grad_norm": 2261.889907648677,
-      "learning_rate": 3.2577328404292057e-09,
-      "logits/chosen": -1.5497777462005615,
-      "logits/rejected": -1.4208284616470337,
-      "logps/chosen": -312.53802490234375,
-      "logps/rejected": -285.97076416015625,
-      "loss": 0.5741,
-      "rewards/accuracies": 0.699999988079071,
-      "rewards/chosen": 1.4049217700958252,
-      "rewards/margins": 1.3057136535644531,
-      "rewards/rejected": 0.09920807182788849,
-      "step": 240
-    },
-    {
-      "epoch": 1.36,
-      "grad_norm": 2520.220972155196,
-      "learning_rate": 2.821423700565763e-09,
-      "logits/chosen": -1.5996572971343994,
-      "logits/rejected": -1.4216984510421753,
-      "logps/chosen": -350.76129150390625,
-      "logps/rejected": -306.58831787109375,
-      "loss": 0.5681,
-      "rewards/accuracies": 0.762499988079071,
-      "rewards/chosen": 1.537647008895874,
-      "rewards/margins": 1.4706511497497559,
-      "rewards/rejected": 0.06699595600366592,
-      "step": 250
-    },
-    {
-      "epoch": 1.41,
-      "grad_norm": 2185.2135747623915,
-      "learning_rate": 2.4047251428513483e-09,
-      "logits/chosen": -1.61586594581604,
-      "logits/rejected": -1.4618706703186035,
-      "logps/chosen": -325.3050537109375,
-      "logps/rejected": -291.10345458984375,
-      "loss": 0.5977,
-      "rewards/accuracies": 0.731249988079071,
-      "rewards/chosen": 1.5201152563095093,
-      "rewards/margins": 1.4326369762420654,
-      "rewards/rejected": 0.08747831732034683,
-      "step": 260
-    },
-    {
-      "epoch": 1.47,
-      "grad_norm": 1735.3140228188304,
-      "learning_rate": 2.011388103757442e-09,
-      "logits/chosen": -1.5243465900421143,
-      "logits/rejected": -1.3802506923675537,
-      "logps/chosen": -316.4330139160156,
-      "logps/rejected": -285.81353759765625,
-      "loss": 0.5429,
-      "rewards/accuracies": 0.7406250238418579,
-      "rewards/chosen": 1.550986647605896,
-      "rewards/margins": 1.45218026638031,
-      "rewards/rejected": 0.09880634397268295,
-      "step": 270
-    },
-    {
-      "epoch": 1.52,
-      "grad_norm": 2121.5830866823144,
-      "learning_rate": 1.644953229677474e-09,
-      "logits/chosen": -1.6015859842300415,
-      "logits/rejected": -1.4193016290664673,
-      "logps/chosen": -326.1202087402344,
-      "logps/rejected": -284.7384033203125,
-      "loss": 0.5887,
-      "rewards/accuracies": 0.746874988079071,
-      "rewards/chosen": 1.580999732017517,
-      "rewards/margins": 1.3716144561767578,
-      "rewards/rejected": 0.20938535034656525,
-      "step": 280
-    },
-    {
-      "epoch": 1.58,
-      "grad_norm": 2103.020456118981,
-      "learning_rate": 1.308719005590957e-09,
-      "logits/chosen": -1.509340524673462,
-      "logits/rejected": -1.3944005966186523,
-      "logps/chosen": -318.451416015625,
-      "logps/rejected": -282.4563293457031,
-      "loss": 0.5721,
-      "rewards/accuracies": 0.731249988079071,
-      "rewards/chosen": 1.603921890258789,
-      "rewards/margins": 1.4211392402648926,
-      "rewards/rejected": 0.182782843708992,
-      "step": 290
-    },
-    {
-      "epoch": 1.63,
-      "grad_norm": 2010.5734100649909,
-      "learning_rate": 1.005712063557776e-09,
-      "logits/chosen": -1.6272541284561157,
-      "logits/rejected": -1.4480578899383545,
-      "logps/chosen": -324.20068359375,
-      "logps/rejected": -290.54803466796875,
-      "loss": 0.5898,
-      "rewards/accuracies": 0.6968749761581421,
-      "rewards/chosen": 1.3664501905441284,
-      "rewards/margins": 1.1471518278121948,
-      "rewards/rejected": 0.21929831802845,
-      "step": 300
-    },
-    {
-      "epoch": 1.68,
-      "grad_norm": 1862.455491029429,
-      "learning_rate": 7.386599383124321e-10,
-      "logits/chosen": -1.563561201095581,
-      "logits/rejected": -1.3803369998931885,
-      "logps/chosen": -321.889404296875,
-      "logps/rejected": -285.8083190917969,
-      "loss": 0.5879,
-      "rewards/accuracies": 0.7124999761581421,
-      "rewards/chosen": 1.4487113952636719,
-      "rewards/margins": 1.3910987377166748,
-      "rewards/rejected": 0.057612527161836624,
-      "step": 310
-    },
-    {
-      "epoch": 1.74,
-      "grad_norm": 1875.0391267528262,
-      "learning_rate": 5.099665152003929e-10,
-      "logits/chosen": -1.5980346202850342,
-      "logits/rejected": -1.3878109455108643,
-      "logps/chosen": -333.843994140625,
-      "logps/rejected": -289.8874816894531,
-      "loss": 0.5634,
-      "rewards/accuracies": 0.7593749761581421,
-      "rewards/chosen": 1.5412708520889282,
-      "rewards/margins": 1.3988701105117798,
-      "rewards/rejected": 0.14240065217018127,
-      "step": 320
-    },
-    {
-      "epoch": 1.79,
-      "grad_norm": 2232.545936783362,
-      "learning_rate": 3.216903914633745e-10,
-      "logits/chosen": -1.5563807487487793,
-      "logits/rejected": -1.4342092275619507,
-      "logps/chosen": -325.34674072265625,
-      "logps/rejected": -296.1554870605469,
-      "loss": 0.5762,
-      "rewards/accuracies": 0.706250011920929,
-      "rewards/chosen": 1.3614461421966553,
-      "rewards/margins": 1.3264009952545166,
-      "rewards/rejected": 0.03504505008459091,
-      "step": 330
-    },
-    {
-      "epoch": 1.85,
-      "grad_norm": 1895.2431252651277,
-      "learning_rate": 1.7552634565570324e-10,
-      "logits/chosen": -1.5503065586090088,
-      "logits/rejected": -1.381874680519104,
-      "logps/chosen": -329.953857421875,
-      "logps/rejected": -292.89447021484375,
-      "loss": 0.5681,
-      "rewards/accuracies": 0.7718750238418579,
-      "rewards/chosen": 1.7590471506118774,
-      "rewards/margins": 1.6621748208999634,
-      "rewards/rejected": 0.09687252342700958,
-      "step": 340
-    },
-    {
-      "epoch": 1.9,
-      "grad_norm": 2070.257116146559,
-      "learning_rate": 7.279008199590543e-11,
-      "logits/chosen": -1.541084885597229,
-      "logits/rejected": -1.378144383430481,
-      "logps/chosen": -326.51507568359375,
-      "logps/rejected": -291.95123291015625,
-      "loss": 0.5797,
-      "rewards/accuracies": 0.7281249761581421,
-      "rewards/chosen": 1.5987285375595093,
-      "rewards/margins": 1.4538962841033936,
-      "rewards/rejected": 0.1448323279619217,
-      "step": 350
-    },
-    {
-      "epoch": 1.96,
-      "grad_norm": 2166.9724743285583,
-      "learning_rate": 1.4406386978128017e-11,
-      "logits/chosen": -1.6209802627563477,
-      "logits/rejected": -1.42485511302948,
-      "logps/chosen": -331.2142639160156,
-      "logps/rejected": -291.69842529296875,
-      "loss": 0.5582,
-      "rewards/accuracies": 0.778124988079071,
-      "rewards/chosen": 1.724250078201294,
-      "rewards/margins": 1.5519315004348755,
-      "rewards/rejected": 0.1723184883594513,
-      "step": 360
-    },
-    {
-      "epoch": 2.0,
-      "step": 368,
+      "epoch": 1.0,
+      "step": 185,
       "total_flos": 0.0,
-      "train_loss": 0.26489045179408527,
-      "train_runtime": 4520.7899,
-      "train_samples_per_second": 20.835,
-      "train_steps_per_second": 0.081
+      "train_loss": 0.33391942269093283,
+      "train_runtime": 5319.9853,
+      "train_samples_per_second": 8.891,
+      "train_steps_per_second": 0.035
     }
   ],
   "logging_steps": 10,
-  "max_steps": 368,
+  "max_steps": 185,
   "num_input_tokens_seen": 0,
-  "num_train_epochs": 2,
+  "num_train_epochs": 1,
   "save_steps": 100,
   "total_flos": 0.0,
   "train_batch_size": 8,

training_args.bin

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:7ac969fef84df010a18a3f4d0a9620668cc46f72adfaf3b07d1d383dfb0d5734
-size 6520
+oid sha256:684c529dee2a83fa137d0e6a638952a379fc861d4c597982d59871fc123d5d7b
+size 6264