Model save

README.md

@@ -1,17 +1,11 @@
 ---
 library_name: transformers
-license: llama3.1
-base_model: meta-llama/Meta-Llama-3.1-8B-Instruct
+license: llama3
+base_model: meta-llama/Meta-Llama-3-8B-Instruct
 tags:
-- alignment-handbook
 - trl
 - dpo
 - generated_from_trainer
-- trl
-- dpo
-- generated_from_trainer
-datasets:
-- HuggingFaceH4/ultrafeedback_binarized
 model-index:
 - name: llama-3-8b-dpo-full
   results: []
@@ -22,17 +16,17 @@ should probably proofread and complete it, then remove this comment. -->
 
 # llama-3-8b-dpo-full
 
-This model is a fine-tuned version of [meta-llama/Meta-Llama-3.1-8B-Instruct](https://huggingface.co/meta-llama/Meta-Llama-3.1-8B-Instruct) on the HuggingFaceH4/ultrafeedback_binarized dataset.
+This model is a fine-tuned version of [meta-llama/Meta-Llama-3-8B-Instruct](https://huggingface.co/meta-llama/Meta-Llama-3-8B-Instruct) on an unknown dataset.
 It achieves the following results on the evaluation set:
-- Loss: 0.5576
-- Rewards/chosen: -1.3507
-- Rewards/rejected: -2.0430
-- Rewards/accuracies: 0.7341
-- Rewards/margins: 0.6923
-- Logps/rejected: -599.9959
-- Logps/chosen: -539.5551
-- Logits/rejected: -2.7918
-- Logits/chosen: -2.6247
+- Loss: 0.6494
+- Rewards/chosen: -0.1752
+- Rewards/rejected: -0.2195
+- Rewards/accuracies: 0.5625
+- Rewards/margins: 0.0443
+- Logps/rejected: -418.5782
+- Logps/chosen: -335.3811
+- Logits/rejected: -1.1582
+- Logits/chosen: -1.0463
 
 ## Model description
 
@@ -56,10 +50,10 @@ The following hyperparameters were used during training:
 - eval_batch_size: 4
 - seed: 42
 - distributed_type: multi-GPU
-- num_devices: 8
+- num_devices: 32
 - gradient_accumulation_steps: 8
-- total_train_batch_size: 128
-- total_eval_batch_size: 32
+- total_train_batch_size: 512
+- total_eval_batch_size: 128
 - optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
 - lr_scheduler_type: cosine
 - lr_scheduler_warmup_ratio: 0.1
@@ -69,10 +63,7 @@ The following hyperparameters were used during training:
 
 | Training Loss | Epoch  | Step | Validation Loss | Rewards/chosen | Rewards/rejected | Rewards/accuracies | Rewards/margins | Logps/rejected | Logps/chosen | Logits/rejected | Logits/chosen |
 |:-------------:|:------:|:----:|:---------------:|:--------------:|:----------------:|:------------------:|:---------------:|:--------------:|:------------:|:---------------:|:-------------:|
-| 0.6341        | 0.2094 | 100  | 0.6223          | -0.6021        | -0.8779          | 0.7103             | 0.2758          | -483.4830      | -464.6949    | -2.6189         | -2.4353       |
-| 0.5887        | 0.4187 | 200  | 0.5796          | -1.0505        | -1.5993          | 0.7143             | 0.5488          | -555.6263      | -509.5346    | -2.6508         | -2.4854       |
-| 0.5667        | 0.6281 | 300  | 0.5653          | -1.0427        | -1.6191          | 0.7222             | 0.5764          | -557.6055      | -508.7539    | -2.6684         | -2.5120       |
-| 0.5803        | 0.8375 | 400  | 0.5582          | -1.3603        | -2.0529          | 0.7262             | 0.6926          | -600.9839      | -540.5128    | -2.7438         | -2.5853       |
+| 0.6411        | 0.8239 | 100  | 0.6494          | -0.1752        | -0.2195          | 0.5625             | 0.0443          | -418.5782      | -335.3811    | -1.1582         | -1.0463       |
 
 
 ### Framework versions

all_results.json

@@ -1,22 +1,9 @@
 {
-    "epoch": 0.998691442030882,
-    "eval_logits/chosen": -2.6246745586395264,
-    "eval_logits/rejected": -2.7917981147766113,
-    "eval_logps/chosen": -539.5550537109375,
-    "eval_logps/rejected": -599.9959106445312,
-    "eval_loss": 0.5575631260871887,
-    "eval_rewards/accuracies": 0.7341269850730896,
-    "eval_rewards/chosen": -1.3506927490234375,
-    "eval_rewards/margins": 0.6923374533653259,
-    "eval_rewards/rejected": -2.043030261993408,
-    "eval_runtime": 39.0144,
-    "eval_samples": 2000,
-    "eval_samples_per_second": 51.263,
-    "eval_steps_per_second": 1.615,
+    "epoch": 0.9969104016477858,
     "total_flos": 0.0,
-    "train_loss": 0.5939158873488068,
-    "train_runtime": 3918.7641,
-    "train_samples": 61135,
-    "train_samples_per_second": 15.601,
-    "train_steps_per_second": 0.122
+    "train_loss": 0.6619468080110786,
+    "train_runtime": 1288.9507,
+    "train_samples": 62135,
+    "train_samples_per_second": 48.206,
+    "train_steps_per_second": 0.094
 }

generation_config.json

@@ -3,9 +3,9 @@
   "do_sample": true,
   "eos_token_id": [
     128001,
-    128008,
     128009
   ],
+  "max_length": 4096,
   "temperature": 0.6,
   "top_p": 0.9,
   "transformers_version": "4.45.1"

model-00001-of-00004.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:40d744ea63ef2270f5100c63a17d0a0b26f5bec9d939002092eec4047c0eacba
+oid sha256:f9e81f67fea17a60a9e82e3d68141b38768c0ea9c8d6d0dfbb04ec1b23d53479
 size 4976698672

model-00002-of-00004.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:40bfb2d580322069e93bd0d346fbd1752fc73aa7e9c8a3325d90710b3e08fc9f
+oid sha256:5660b6a64d157b764893f6ed0e8a2b7722e931084e3063e01192aa19792edf83
 size 4999802720

model-00003-of-00004.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:79849487167122cfd3f80b6b27f124eedb28398dca7bc827dcca805f9d82d701
+oid sha256:53a7e3ac53c81aacb82a20f481e282419377879962700d300dc73b2aa9b6d268
 size 4915916176

model-00004-of-00004.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:0707d187c67088c3a2cde6247d5cae889702c93821f65ca4f60a7604eb5ca903
+oid sha256:4cf561ec0ca067efdb8e3e2f2480afb303e3238e4412882b48b75c45048494a5
 size 1168138808

runs/Oct08_11-25-31_a2ap-dgx018/events.out.tfevents.1728358134.a2ap-dgx018.1893204.0

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:d446165e2f87e2549b0cdfe298ae7c8e660ba0bea55e2d95d7edb478adfffd97
-size 14446
+oid sha256:0d3ac22f9e2d624c93b6acd5e547cb97ae0b0a45ab5a94b63283ea93475d95d8
+size 16146

train_results.json

@@ -1,9 +1,9 @@
 {
-    "epoch": 0.998691442030882,
+    "epoch": 0.9969104016477858,
     "total_flos": 0.0,
-    "train_loss": 0.5939158873488068,
-    "train_runtime": 3918.7641,
-    "train_samples": 61135,
-    "train_samples_per_second": 15.601,
-    "train_steps_per_second": 0.122
+    "train_loss": 0.6619468080110786,
+    "train_runtime": 1288.9507,
+    "train_samples": 62135,
+    "train_samples_per_second": 48.206,
+    "train_steps_per_second": 0.094
 }

trainer_state.json

@@ -1,21 +1,21 @@
 {
   "best_metric": null,
   "best_model_checkpoint": null,
-  "epoch": 0.998691442030882,
+  "epoch": 0.9969104016477858,
   "eval_steps": 100,
-  "global_step": 477,
+  "global_step": 121,
   "is_hyper_param_search": false,
   "is_local_process_zero": true,
   "is_world_process_zero": true,
   "log_history": [
     {
-      "epoch": 0.002093692750588851,
-      "grad_norm": 5.844382227449463,
-      "learning_rate": 6.25e-09,
-      "logits/chosen": -1.6512014865875244,
-      "logits/rejected": -1.4355601072311401,
-      "logps/chosen": -375.0550842285156,
-      "logps/rejected": -355.35711669921875,
+      "epoch": 0.008238928939237899,
+      "grad_norm": 3.6043033117037053,
+      "learning_rate": 2.3076923076923076e-08,
+      "logits/chosen": -0.971167802810669,
+      "logits/rejected": -1.2423464059829712,
+      "logps/chosen": -290.201171875,
+      "logps/rejected": -263.9928894042969,
       "loss": 0.6931,
       "rewards/accuracies": 0.0,
       "rewards/chosen": 0.0,
@@ -24,786 +24,213 @@
       "step": 1
     },
     {
-      "epoch": 0.02093692750588851,
-      "grad_norm": 5.235821845579244,
-      "learning_rate": 6.25e-08,
-      "logits/chosen": -1.7515054941177368,
-      "logits/rejected": -2.0137522220611572,
-      "logps/chosen": -417.4104919433594,
-      "logps/rejected": -399.6319274902344,
-      "loss": 0.693,
-      "rewards/accuracies": 0.4305555522441864,
-      "rewards/chosen": -0.0020523108541965485,
-      "rewards/margins": -0.00028108435799367726,
-      "rewards/rejected": -0.0017712265253067017,
+      "epoch": 0.082389289392379,
+      "grad_norm": 3.4484942444647335,
+      "learning_rate": 2.3076923076923078e-07,
+      "logits/chosen": -0.9607771635055542,
+      "logits/rejected": -0.9658682942390442,
+      "logps/chosen": -330.0989685058594,
+      "logps/rejected": -330.9446716308594,
+      "loss": 0.6931,
+      "rewards/accuracies": 0.4236111044883728,
+      "rewards/chosen": 0.0003925244091078639,
+      "rewards/margins": 4.308321513235569e-05,
+      "rewards/rejected": 0.0003494411939755082,
       "step": 10
     },
     {
-      "epoch": 0.04187385501177702,
-      "grad_norm": 4.969390451186454,
-      "learning_rate": 1.25e-07,
-      "logits/chosen": -1.6114051342010498,
-      "logits/rejected": -2.0293774604797363,
-      "logps/chosen": -404.6941833496094,
-      "logps/rejected": -367.32257080078125,
-      "loss": 0.6929,
-      "rewards/accuracies": 0.4937500059604645,
-      "rewards/chosen": -0.0030800788663327694,
-      "rewards/margins": -2.9044622351648286e-05,
-      "rewards/rejected": -0.00305103394202888,
+      "epoch": 0.164778578784758,
+      "grad_norm": 3.5480284480156086,
+      "learning_rate": 2.969010932648327e-07,
+      "logits/chosen": -1.0245492458343506,
+      "logits/rejected": -1.0337975025177002,
+      "logps/chosen": -347.72344970703125,
+      "logps/rejected": -345.3406066894531,
+      "loss": 0.6907,
+      "rewards/accuracies": 0.543749988079071,
+      "rewards/chosen": 0.007954178377985954,
+      "rewards/margins": 0.005248874891549349,
+      "rewards/rejected": 0.00270530441775918,
       "step": 20
     },
     {
-      "epoch": 0.06281078251766553,
-      "grad_norm": 5.046869414952022,
-      "learning_rate": 1.875e-07,
-      "logits/chosen": -1.7840598821640015,
-      "logits/rejected": -2.01045298576355,
-      "logps/chosen": -380.27178955078125,
-      "logps/rejected": -359.34735107421875,
-      "loss": 0.6913,
-      "rewards/accuracies": 0.550000011920929,
-      "rewards/chosen": -0.013285043649375439,
-      "rewards/margins": 0.0011252568801864982,
-      "rewards/rejected": -0.01441030204296112,
+      "epoch": 0.24716786817713698,
+      "grad_norm": 3.121263888821987,
+      "learning_rate": 2.8203020867701666e-07,
+      "logits/chosen": -0.8977586030960083,
+      "logits/rejected": -1.0179085731506348,
+      "logps/chosen": -304.6155700683594,
+      "logps/rejected": -275.14447021484375,
+      "loss": 0.6823,
+      "rewards/accuracies": 0.731249988079071,
+      "rewards/chosen": 0.025795791298151016,
+      "rewards/margins": 0.025993749499320984,
+      "rewards/rejected": -0.00019796025299001485,
       "step": 30
     },
     {
-      "epoch": 0.08374771002355404,
-      "grad_norm": 4.698944247342457,
-      "learning_rate": 2.5e-07,
-      "logits/chosen": -1.6258041858673096,
-      "logits/rejected": -1.7915970087051392,
-      "logps/chosen": -407.2330017089844,
-      "logps/rejected": -409.3331298828125,
-      "loss": 0.6861,
-      "rewards/accuracies": 0.668749988079071,
-      "rewards/chosen": -0.038303375244140625,
-      "rewards/margins": 0.012813824228942394,
-      "rewards/rejected": -0.05111720412969589,
+      "epoch": 0.329557157569516,
+      "grad_norm": 2.6779190942219393,
+      "learning_rate": 2.560660171779821e-07,
+      "logits/chosen": -0.9125121235847473,
+      "logits/rejected": -1.0349690914154053,
+      "logps/chosen": -340.72442626953125,
+      "logps/rejected": -313.9574890136719,
+      "loss": 0.675,
+      "rewards/accuracies": 0.65625,
+      "rewards/chosen": 0.02803516387939453,
+      "rewards/margins": 0.04235466197133064,
+      "rewards/rejected": -0.01431949995458126,
       "step": 40
     },
     {
-      "epoch": 0.10468463752944256,
-      "grad_norm": 5.305945732445058,
-      "learning_rate": 2.9998391212614154e-07,
-      "logits/chosen": -1.5322725772857666,
-      "logits/rejected": -1.7322235107421875,
-      "logps/chosen": -404.21575927734375,
-      "logps/rejected": -431.61004638671875,
-      "loss": 0.6751,
-      "rewards/accuracies": 0.625,
-      "rewards/chosen": -0.09563098847866058,
-      "rewards/margins": 0.032842181622982025,
-      "rewards/rejected": -0.1284731775522232,
+      "epoch": 0.411946446961895,
+      "grad_norm": 2.814818260153817,
+      "learning_rate": 2.2119005546214604e-07,
+      "logits/chosen": -0.9567445516586304,
+      "logits/rejected": -0.9532279968261719,
+      "logps/chosen": -272.302978515625,
+      "logps/rejected": -295.3742370605469,
+      "loss": 0.6649,
+      "rewards/accuracies": 0.643750011920929,
+      "rewards/chosen": 0.006898154504597187,
+      "rewards/margins": 0.06003887206315994,
+      "rewards/rejected": -0.05314071848988533,
       "step": 50
     },
     {
-      "epoch": 0.12562156503533106,
-      "grad_norm": 5.972525123376799,
-      "learning_rate": 2.9942119880575817e-07,
-      "logits/chosen": -1.8156770467758179,
-      "logits/rejected": -1.956964135169983,
-      "logps/chosen": -413.17218017578125,
-      "logps/rejected": -401.9190979003906,
-      "loss": 0.662,
-      "rewards/accuracies": 0.6187499761581421,
-      "rewards/chosen": -0.22033607959747314,
-      "rewards/margins": 0.05222884565591812,
-      "rewards/rejected": -0.27256491780281067,
+      "epoch": 0.49433573635427397,
+      "grad_norm": 2.653170547007469,
+      "learning_rate": 1.803326358498057e-07,
+      "logits/chosen": -0.8227835893630981,
+      "logits/rejected": -0.8951934576034546,
+      "logps/chosen": -364.8822021484375,
+      "logps/rejected": -332.07489013671875,
+      "loss": 0.6563,
+      "rewards/accuracies": 0.6812499761581421,
+      "rewards/chosen": 0.002518509980291128,
+      "rewards/margins": 0.14299212396144867,
+      "rewards/rejected": -0.1404736191034317,
       "step": 60
     },
     {
-      "epoch": 0.14655849254121958,
-      "grad_norm": 8.237654744616636,
-      "learning_rate": 2.980575393956869e-07,
-      "logits/chosen": -1.9767135381698608,
-      "logits/rejected": -2.0496904850006104,
-      "logps/chosen": -442.13055419921875,
-      "logps/rejected": -466.5682678222656,
-      "loss": 0.6501,
-      "rewards/accuracies": 0.606249988079071,
-      "rewards/chosen": -0.4638500213623047,
-      "rewards/margins": 0.14070633053779602,
-      "rewards/rejected": -0.6045563817024231,
+      "epoch": 0.576725025746653,
+      "grad_norm": 2.7392936630658764,
+      "learning_rate": 1.3692663858785124e-07,
+      "logits/chosen": -0.8262346386909485,
+      "logits/rejected": -0.9113919138908386,
+      "logps/chosen": -337.5333251953125,
+      "logps/rejected": -330.14459228515625,
+      "loss": 0.6558,
+      "rewards/accuracies": 0.59375,
+      "rewards/chosen": -0.06830023974180222,
+      "rewards/margins": 0.05756063386797905,
+      "rewards/rejected": -0.12586086988449097,
       "step": 70
     },
     {
-      "epoch": 0.16749542004710807,
-      "grad_norm": 7.104867946653516,
-      "learning_rate": 2.9590024355266256e-07,
-      "logits/chosen": -1.9303478002548218,
-      "logits/rejected": -2.1261794567108154,
-      "logps/chosen": -456.86334228515625,
-      "logps/rejected": -468.8231506347656,
-      "loss": 0.6315,
-      "rewards/accuracies": 0.6312500238418579,
-      "rewards/chosen": -0.610644519329071,
-      "rewards/margins": 0.14992806315422058,
-      "rewards/rejected": -0.7605725526809692,
+      "epoch": 0.659114315139032,
+      "grad_norm": 4.467409541426095,
+      "learning_rate": 9.461907790309735e-08,
+      "logits/chosen": -0.8138895034790039,
+      "logits/rejected": -0.9478031992912292,
+      "logps/chosen": -370.1873779296875,
+      "logps/rejected": -331.826904296875,
+      "loss": 0.645,
+      "rewards/accuracies": 0.6812499761581421,
+      "rewards/chosen": -0.046670813113451004,
+      "rewards/margins": 0.14053179323673248,
+      "rewards/rejected": -0.187202587723732,
       "step": 80
     },
     {
-      "epoch": 0.1884323475529966,
-      "grad_norm": 7.380030502862527,
-      "learning_rate": 2.929608750821129e-07,
-      "logits/chosen": -1.861717939376831,
-      "logits/rejected": -2.3124728202819824,
-      "logps/chosen": -473.9281311035156,
-      "logps/rejected": -462.75018310546875,
-      "loss": 0.6356,
-      "rewards/accuracies": 0.6187499761581421,
-      "rewards/chosen": -0.72138911485672,
-      "rewards/margins": 0.17633597552776337,
-      "rewards/rejected": -0.8977251052856445,
+      "epoch": 0.741503604531411,
+      "grad_norm": 2.8035199165767315,
+      "learning_rate": 5.6964676309761004e-08,
+      "logits/chosen": -0.8749955296516418,
+      "logits/rejected": -0.9215911030769348,
+      "logps/chosen": -359.80511474609375,
+      "logps/rejected": -361.1700134277344,
+      "loss": 0.647,
+      "rewards/accuracies": 0.675000011920929,
+      "rewards/chosen": -0.08991885185241699,
+      "rewards/margins": 0.14792749285697937,
+      "rewards/rejected": -0.23784634470939636,
       "step": 90
     },
     {
-      "epoch": 0.2093692750588851,
-      "grad_norm": 7.1928607065555585,
-      "learning_rate": 2.892551899524109e-07,
-      "logits/chosen": -1.8947665691375732,
-      "logits/rejected": -2.0735626220703125,
-      "logps/chosen": -474.9530334472656,
-      "logps/rejected": -498.10223388671875,
-      "loss": 0.6341,
-      "rewards/accuracies": 0.699999988079071,
-      "rewards/chosen": -0.6549180746078491,
-      "rewards/margins": 0.23183219134807587,
-      "rewards/rejected": -0.886750340461731,
+      "epoch": 0.82389289392379,
+      "grad_norm": 3.018969560415363,
+      "learning_rate": 2.7127193356651214e-08,
+      "logits/chosen": -0.838853657245636,
+      "logits/rejected": -0.981770396232605,
+      "logps/chosen": -380.34307861328125,
+      "logps/rejected": -369.41180419921875,
+      "loss": 0.6411,
+      "rewards/accuracies": 0.668749988079071,
+      "rewards/chosen": -0.10747754573822021,
+      "rewards/margins": 0.17505204677581787,
+      "rewards/rejected": -0.28252965211868286,
       "step": 100
     },
     {
-      "epoch": 0.2093692750588851,
-      "eval_logits/chosen": -2.435333490371704,
-      "eval_logits/rejected": -2.6189088821411133,
-      "eval_logps/chosen": -464.6949462890625,
-      "eval_logps/rejected": -483.48297119140625,
-      "eval_loss": 0.6222504377365112,
-      "eval_rewards/accuracies": 0.7103174328804016,
-      "eval_rewards/chosen": -0.6020920872688293,
-      "eval_rewards/margins": 0.2758086025714874,
-      "eval_rewards/rejected": -0.8779006600379944,
-      "eval_runtime": 40.059,
-      "eval_samples_per_second": 49.926,
-      "eval_steps_per_second": 1.573,
+      "epoch": 0.82389289392379,
+      "eval_logits/chosen": -1.0462558269500732,
+      "eval_logits/rejected": -1.1581745147705078,
+      "eval_logps/chosen": -335.3811340332031,
+      "eval_logps/rejected": -418.57818603515625,
+      "eval_loss": 0.6494396924972534,
+      "eval_rewards/accuracies": 0.5625,
+      "eval_rewards/chosen": -0.17523552477359772,
+      "eval_rewards/margins": 0.04425430670380592,
+      "eval_rewards/rejected": -0.21948984265327454,
+      "eval_runtime": 9.1337,
+      "eval_samples_per_second": 109.485,
+      "eval_steps_per_second": 0.876,
       "step": 100
     },
     {
-      "epoch": 0.23030620256477363,
-      "grad_norm": 7.415850428843577,
-      "learning_rate": 2.848030518377739e-07,
-      "logits/chosen": -1.8223705291748047,
-      "logits/rejected": -2.2238452434539795,
-      "logps/chosen": -484.059814453125,
-      "logps/rejected": -462.3560485839844,
-      "loss": 0.6067,
-      "rewards/accuracies": 0.6812499761581421,
-      "rewards/chosen": -0.7122976779937744,
-      "rewards/margins": 0.2089848816394806,
-      "rewards/rejected": -0.9212825894355774,
+      "epoch": 0.9062821833161689,
+      "grad_norm": 2.971778165880155,
+      "learning_rate": 7.613603464044916e-09,
+      "logits/chosen": -0.8196293115615845,
+      "logits/rejected": -0.9439749717712402,
+      "logps/chosen": -354.5921630859375,
+      "logps/rejected": -318.67962646484375,
+      "loss": 0.6449,
+      "rewards/accuracies": 0.612500011920929,
+      "rewards/chosen": -0.1414671391248703,
+      "rewards/margins": 0.08098333328962326,
+      "rewards/rejected": -0.22245049476623535,
       "step": 110
     },
     {
-      "epoch": 0.2512431300706621,
-      "grad_norm": 10.357041495108014,
-      "learning_rate": 2.7962832564252725e-07,
-      "logits/chosen": -2.0539345741271973,
-      "logits/rejected": -2.1838014125823975,
-      "logps/chosen": -490.2127990722656,
-      "logps/rejected": -524.0123901367188,
-      "loss": 0.6168,
-      "rewards/accuracies": 0.668749988079071,
-      "rewards/chosen": -0.9230595827102661,
-      "rewards/margins": 0.2751563489437103,
-      "rewards/rejected": -1.1982159614562988,
+      "epoch": 0.9886714727085479,
+      "grad_norm": 2.832084621380708,
+      "learning_rate": 6.34574876699101e-11,
+      "logits/chosen": -0.9086641073226929,
+      "logits/rejected": -0.9848729372024536,
+      "logps/chosen": -327.4648132324219,
+      "logps/rejected": -343.4488525390625,
+      "loss": 0.6437,
+      "rewards/accuracies": 0.6812499761581421,
+      "rewards/chosen": -0.10709067434072495,
+      "rewards/margins": 0.1814233958721161,
+      "rewards/rejected": -0.28851407766342163,
       "step": 120
     },
     {
-      "epoch": 0.2721800575765506,
-      "grad_norm": 8.727812082209544,
-      "learning_rate": 2.737587495774764e-07,
-      "logits/chosen": -2.0211193561553955,
-      "logits/rejected": -2.091749906539917,
-      "logps/chosen": -476.74884033203125,
-      "logps/rejected": -513.7708740234375,
-      "loss": 0.6219,
-      "rewards/accuracies": 0.6937500238418579,
-      "rewards/chosen": -0.8179013133049011,
-      "rewards/margins": 0.2984716296195984,
-      "rewards/rejected": -1.116373062133789,
-      "step": 130
-    },
-    {
-      "epoch": 0.29311698508243916,
-      "grad_norm": 10.936927213853266,
-      "learning_rate": 2.672257864741005e-07,
-      "logits/chosen": -2.028709888458252,
-      "logits/rejected": -2.1645443439483643,
-      "logps/chosen": -472.2229919433594,
-      "logps/rejected": -513.4671020507812,
-      "loss": 0.5967,
-      "rewards/accuracies": 0.7124999761581421,
-      "rewards/chosen": -0.7414780855178833,
-      "rewards/margins": 0.44093647599220276,
-      "rewards/rejected": -1.1824145317077637,
-      "step": 140
-    },
-    {
-      "epoch": 0.31405391258832765,
-      "grad_norm": 11.027705905016305,
-      "learning_rate": 2.6006445513357056e-07,
-      "logits/chosen": -1.9190442562103271,
-      "logits/rejected": -2.1971538066864014,
-      "logps/chosen": -533.9457397460938,
-      "logps/rejected": -540.30322265625,
-      "loss": 0.6027,
-      "rewards/accuracies": 0.6875,
-      "rewards/chosen": -1.019513726234436,
-      "rewards/margins": 0.4124508500099182,
-      "rewards/rejected": -1.431964635848999,
-      "step": 150
-    },
-    {
-      "epoch": 0.33499084009421615,
-      "grad_norm": 14.081707059974091,
-      "learning_rate": 2.523131426146173e-07,
-      "logits/chosen": -2.0433707237243652,
-      "logits/rejected": -2.262349843978882,
-      "logps/chosen": -504.19873046875,
-      "logps/rejected": -551.8690185546875,
-      "loss": 0.5571,
-      "rewards/accuracies": 0.7562500238418579,
-      "rewards/chosen": -1.0314991474151611,
-      "rewards/margins": 0.5809075832366943,
-      "rewards/rejected": -1.6124067306518555,
-      "step": 160
-    },
-    {
-      "epoch": 0.3559277676001047,
-      "grad_norm": 11.826987567285538,
-      "learning_rate": 2.440133984664454e-07,
-      "logits/chosen": -1.9292476177215576,
-      "logits/rejected": -2.1319830417633057,
-      "logps/chosen": -521.266845703125,
-      "logps/rejected": -494.2020568847656,
-      "loss": 0.5976,
-      "rewards/accuracies": 0.6625000238418579,
-      "rewards/chosen": -1.065916657447815,
-      "rewards/margins": 0.2420383244752884,
-      "rewards/rejected": -1.307955026626587,
-      "step": 170
-    },
-    {
-      "epoch": 0.3768646951059932,
-      "grad_norm": 9.632333751579734,
-      "learning_rate": 2.3520971200967334e-07,
-      "logits/chosen": -2.046644449234009,
-      "logits/rejected": -2.234806537628174,
-      "logps/chosen": -473.70770263671875,
-      "logps/rejected": -492.889404296875,
-      "loss": 0.5926,
-      "rewards/accuracies": 0.6875,
-      "rewards/chosen": -0.6946296095848083,
-      "rewards/margins": 0.37669837474823,
-      "rewards/rejected": -1.0713279247283936,
-      "step": 180
-    },
-    {
-      "epoch": 0.39780162261188173,
-      "grad_norm": 11.984832910221622,
-      "learning_rate": 2.2594927385914543e-07,
-      "logits/chosen": -1.9993702173233032,
-      "logits/rejected": -2.236987590789795,
-      "logps/chosen": -489.000244140625,
-      "logps/rejected": -519.3321533203125,
-      "loss": 0.5755,
-      "rewards/accuracies": 0.768750011920929,
-      "rewards/chosen": -1.0067808628082275,
-      "rewards/margins": 0.5224589705467224,
-      "rewards/rejected": -1.5292397737503052,
-      "step": 190
-    },
-    {
-      "epoch": 0.4187385501177702,
-      "grad_norm": 12.816716120801315,
-      "learning_rate": 2.162817229669295e-07,
-      "logits/chosen": -2.1309211254119873,
-      "logits/rejected": -2.3286731243133545,
-      "logps/chosen": -511.89959716796875,
-      "logps/rejected": -556.114990234375,
-      "loss": 0.5887,
-      "rewards/accuracies": 0.71875,
-      "rewards/chosen": -1.1402111053466797,
-      "rewards/margins": 0.55492103099823,
-      "rewards/rejected": -1.6951322555541992,
-      "step": 200
-    },
-    {
-      "epoch": 0.4187385501177702,
-      "eval_logits/chosen": -2.485398769378662,
-      "eval_logits/rejected": -2.6507983207702637,
-      "eval_logps/chosen": -509.5345764160156,
-      "eval_logps/rejected": -555.6263427734375,
-      "eval_loss": 0.5795689821243286,
-      "eval_rewards/accuracies": 0.7142857313156128,
-      "eval_rewards/chosen": -1.0504883527755737,
-      "eval_rewards/margins": 0.5488466620445251,
-      "eval_rewards/rejected": -1.599334955215454,
-      "eval_runtime": 38.586,
-      "eval_samples_per_second": 51.832,
-      "eval_steps_per_second": 1.633,
-      "step": 200
-    },
-    {
-      "epoch": 0.4396754776236587,
-      "grad_norm": 12.909762723332605,
-      "learning_rate": 2.0625888054143427e-07,
-      "logits/chosen": -2.1252217292785645,
-      "logits/rejected": -2.271873950958252,
-      "logps/chosen": -487.28045654296875,
-      "logps/rejected": -531.2611083984375,
-      "loss": 0.5759,
-      "rewards/accuracies": 0.7124999761581421,
-      "rewards/chosen": -1.0537116527557373,
-      "rewards/margins": 0.4694788455963135,
-      "rewards/rejected": -1.5231904983520508,
-      "step": 210
-    },
-    {
-      "epoch": 0.46061240512954726,
-      "grad_norm": 14.088630852147947,
-      "learning_rate": 1.9593447226892386e-07,
-      "logits/chosen": -2.2382607460021973,
-      "logits/rejected": -2.3763673305511475,
-      "logps/chosen": -526.3925170898438,
-      "logps/rejected": -573.842529296875,
-      "loss": 0.5762,
-      "rewards/accuracies": 0.75,
-      "rewards/chosen": -1.2542815208435059,
-      "rewards/margins": 0.5530741810798645,
-      "rewards/rejected": -1.807355523109436,
-      "step": 220
-    },
-    {
-      "epoch": 0.48154933263543576,
-      "grad_norm": 15.369265950204866,
-      "learning_rate": 1.853638403264141e-07,
-      "logits/chosen": -2.082986354827881,
-      "logits/rejected": -2.2771239280700684,
-      "logps/chosen": -481.3154296875,
-      "logps/rejected": -538.543212890625,
-      "loss": 0.5646,
-      "rewards/accuracies": 0.7749999761581421,
-      "rewards/chosen": -0.9790234565734863,
-      "rewards/margins": 0.5569753050804138,
-      "rewards/rejected": -1.535998821258545,
-      "step": 230
-    },
-    {
-      "epoch": 0.5024862601413242,
-      "grad_norm": 14.740154999219994,
-      "learning_rate": 1.7460364672965327e-07,
-      "logits/chosen": -2.0640745162963867,
-      "logits/rejected": -2.2337191104888916,
-      "logps/chosen": -490.3001403808594,
-      "logps/rejected": -569.6817626953125,
-      "loss": 0.5559,
-      "rewards/accuracies": 0.7250000238418579,
-      "rewards/chosen": -1.325878381729126,
-      "rewards/margins": 0.620599627494812,
-      "rewards/rejected": -1.9464781284332275,
-      "step": 240
-    },
-    {
-      "epoch": 0.5234231876472127,
-      "grad_norm": 11.893999213827813,
-      "learning_rate": 1.6371156960634017e-07,
-      "logits/chosen": -2.005420684814453,
-      "logits/rejected": -2.3055458068847656,
-      "logps/chosen": -528.0526733398438,
-      "logps/rejected": -554.3103637695312,
-      "loss": 0.5972,
-      "rewards/accuracies": 0.7562500238418579,
-      "rewards/chosen": -1.2345831394195557,
-      "rewards/margins": 0.4900982975959778,
-      "rewards/rejected": -1.7246814966201782,
-      "step": 250
-    },
-    {
-      "epoch": 0.5443601151531012,
-      "grad_norm": 12.09341713044732,
-      "learning_rate": 1.527459940226516e-07,
-      "logits/chosen": -2.0832793712615967,
-      "logits/rejected": -2.202723741531372,
-      "logps/chosen": -532.1043701171875,
-      "logps/rejected": -550.4468994140625,
-      "loss": 0.5794,
-      "rewards/accuracies": 0.699999988079071,
-      "rewards/chosen": -1.092650294303894,
-      "rewards/margins": 0.4655452370643616,
-      "rewards/rejected": -1.5581955909729004,
-      "step": 260
-    },
-    {
-      "epoch": 0.5652970426589898,
-      "grad_norm": 13.578869062104298,
-      "learning_rate": 1.4176569902035086e-07,
-      "logits/chosen": -2.2014851570129395,
-      "logits/rejected": -2.173820972442627,
-      "logps/chosen": -536.9981689453125,
-      "logps/rejected": -623.5565185546875,
-      "loss": 0.5642,
-      "rewards/accuracies": 0.7250000238418579,
-      "rewards/chosen": -1.3192862272262573,
-      "rewards/margins": 0.5946845412254333,
-      "rewards/rejected": -1.913970708847046,
-      "step": 270
-    },
-    {
-      "epoch": 0.5862339701648783,
-      "grad_norm": 11.79408971766642,
-      "learning_rate": 1.3082954254205928e-07,
-      "logits/chosen": -2.1410746574401855,
-      "logits/rejected": -2.3836779594421387,
-      "logps/chosen": -525.254638671875,
-      "logps/rejected": -562.7789306640625,
-      "loss": 0.5766,
-      "rewards/accuracies": 0.6625000238418579,
-      "rewards/chosen": -1.3711321353912354,
-      "rewards/margins": 0.5404887199401855,
-      "rewards/rejected": -1.9116207361221313,
-      "step": 280
-    },
-    {
-      "epoch": 0.6071708976707668,
-      "grad_norm": 11.223086706265748,
-      "learning_rate": 1.1999614593359334e-07,
-      "logits/chosen": -2.239353895187378,
-      "logits/rejected": -2.298745632171631,
-      "logps/chosen": -540.8309936523438,
-      "logps/rejected": -590.9267578125,
-      "loss": 0.5838,
-      "rewards/accuracies": 0.699999988079071,
-      "rewards/chosen": -1.269129991531372,
-      "rewards/margins": 0.5602521300315857,
-      "rewards/rejected": -1.829382300376892,
-      "step": 290
-    },
-    {
-      "epoch": 0.6281078251766553,
-      "grad_norm": 11.552990511047495,
-      "learning_rate": 1.0932357971453743e-07,
-      "logits/chosen": -2.2803478240966797,
-      "logits/rejected": -2.3654093742370605,
-      "logps/chosen": -488.808837890625,
-      "logps/rejected": -528.6336059570312,
-      "loss": 0.5667,
-      "rewards/accuracies": 0.7562500238418579,
-      "rewards/chosen": -1.0937414169311523,
-      "rewards/margins": 0.4998832643032074,
-      "rewards/rejected": -1.5936245918273926,
-      "step": 300
-    },
-    {
-      "epoch": 0.6281078251766553,
-      "eval_logits/chosen": -2.511992931365967,
-      "eval_logits/rejected": -2.6684465408325195,
-      "eval_logps/chosen": -508.75390625,
-      "eval_logps/rejected": -557.6055297851562,
-      "eval_loss": 0.5652720928192139,
-      "eval_rewards/accuracies": 0.7222222089767456,
-      "eval_rewards/chosen": -1.0426820516586304,
-      "eval_rewards/margins": 0.5764438509941101,
-      "eval_rewards/rejected": -1.6191262006759644,
-      "eval_runtime": 38.5252,
-      "eval_samples_per_second": 51.914,
-      "eval_steps_per_second": 1.635,
-      "step": 300
-    },
-    {
-      "epoch": 0.6490447526825438,
-      "grad_norm": 11.573976352456944,
-      "learning_rate": 9.886905230142431e-08,
-      "logits/chosen": -2.1670825481414795,
-      "logits/rejected": -2.3857502937316895,
-      "logps/chosen": -506.4759216308594,
-      "logps/rejected": -566.917724609375,
-      "loss": 0.5538,
-      "rewards/accuracies": 0.7749999761581421,
-      "rewards/chosen": -1.1441192626953125,
-      "rewards/margins": 0.5939659476280212,
-      "rewards/rejected": -1.738085389137268,
-      "step": 310
-    },
-    {
-      "epoch": 0.6699816801884323,
-      "grad_norm": 11.811853616414988,
-      "learning_rate": 8.868860335206678e-08,
-      "logits/chosen": -2.2445249557495117,
-      "logits/rejected": -2.408531665802002,
-      "logps/chosen": -516.3126220703125,
-      "logps/rejected": -578.1045532226562,
-      "loss": 0.5687,
-      "rewards/accuracies": 0.7124999761581421,
-      "rewards/chosen": -1.3060988187789917,
-      "rewards/margins": 0.6043392419815063,
-      "rewards/rejected": -1.9104379415512085,
-      "step": 320
-    },
-    {
-      "epoch": 0.6909186076943209,
-      "grad_norm": 15.067885639931706,
-      "learning_rate": 7.883680337481599e-08,
-      "logits/chosen": -2.2785134315490723,
-      "logits/rejected": -2.36972713470459,
-      "logps/chosen": -516.9320678710938,
-      "logps/rejected": -572.5419311523438,
-      "loss": 0.5766,
-      "rewards/accuracies": 0.75,
-      "rewards/chosen": -1.1142256259918213,
-      "rewards/margins": 0.5552430748939514,
-      "rewards/rejected": -1.669468641281128,
-      "step": 330
-    },
-    {
-      "epoch": 0.7118555352002094,
-      "grad_norm": 11.201673342190341,
-      "learning_rate": 6.936646121293653e-08,
-      "logits/chosen": -2.2320239543914795,
-      "logits/rejected": -2.4628634452819824,
-      "logps/chosen": -464.08502197265625,
-      "logps/rejected": -495.380859375,
-      "loss": 0.576,
-      "rewards/accuracies": 0.7124999761581421,
-      "rewards/chosen": -1.0917171239852905,
-      "rewards/margins": 0.5427553653717041,
-      "rewards/rejected": -1.6344726085662842,
-      "step": 340
-    },
-    {
-      "epoch": 0.7327924627060979,
-      "grad_norm": 11.655689256355245,
-      "learning_rate": 6.032834097207888e-08,
-      "logits/chosen": -2.152733325958252,
-      "logits/rejected": -2.451592445373535,
-      "logps/chosen": -545.7279663085938,
-      "logps/rejected": -572.2804565429688,
-      "loss": 0.569,
-      "rewards/accuracies": 0.6875,
-      "rewards/chosen": -1.1281747817993164,
-      "rewards/margins": 0.5794798731803894,
-      "rewards/rejected": -1.7076547145843506,
-      "step": 350
-    },
-    {
-      "epoch": 0.7537293902119864,
-      "grad_norm": 13.61262579360644,
-      "learning_rate": 5.177088990820725e-08,
-      "logits/chosen": -2.3029279708862305,
-      "logits/rejected": -2.4399266242980957,
-      "logps/chosen": -488.69921875,
-      "logps/rejected": -547.46142578125,
-      "loss": 0.572,
-      "rewards/accuracies": 0.75,
-      "rewards/chosen": -1.1252793073654175,
-      "rewards/margins": 0.5917247533798218,
-      "rewards/rejected": -1.7170041799545288,
-      "step": 360
-    },
-    {
-      "epoch": 0.7746663177178749,
-      "grad_norm": 12.249704332676693,
-      "learning_rate": 4.373997873459449e-08,
-      "logits/chosen": -2.2016916275024414,
-      "logits/rejected": -2.460952043533325,
-      "logps/chosen": -532.7511596679688,
-      "logps/rejected": -570.0086669921875,
-      "loss": 0.5523,
-      "rewards/accuracies": 0.7749999761581421,
-      "rewards/chosen": -1.1311514377593994,
-      "rewards/margins": 0.6326709985733032,
-      "rewards/rejected": -1.7638225555419922,
-      "step": 370
-    },
-    {
-      "epoch": 0.7956032452237635,
-      "grad_norm": 16.95867889347482,
-      "learning_rate": 3.627865573992087e-08,
-      "logits/chosen": -2.2143394947052,
-      "logits/rejected": -2.3504748344421387,
-      "logps/chosen": -514.0362548828125,
-      "logps/rejected": -542.30712890625,
-      "loss": 0.5731,
-      "rewards/accuracies": 0.6875,
-      "rewards/chosen": -1.1771132946014404,
-      "rewards/margins": 0.5136756300926208,
-      "rewards/rejected": -1.690788984298706,
-      "step": 380
-    },
-    {
-      "epoch": 0.816540172729652,
-      "grad_norm": 12.40383909871666,
-      "learning_rate": 2.942691603548416e-08,
-      "logits/chosen": -2.1582541465759277,
-      "logits/rejected": -2.3233611583709717,
-      "logps/chosen": -540.1260375976562,
-      "logps/rejected": -588.2706909179688,
-      "loss": 0.5499,
-      "rewards/accuracies": 0.699999988079071,
-      "rewards/chosen": -1.2853848934173584,
-      "rewards/margins": 0.5986013412475586,
-      "rewards/rejected": -1.883986473083496,
-      "step": 390
-    },
-    {
-      "epoch": 0.8374771002355405,
-      "grad_norm": 15.729235077567395,
-      "learning_rate": 2.322148716843081e-08,
-      "logits/chosen": -2.103240966796875,
-      "logits/rejected": -2.3556132316589355,
-      "logps/chosen": -578.9059448242188,
-      "logps/rejected": -590.7369384765625,
-      "loss": 0.5803,
-      "rewards/accuracies": 0.7749999761581421,
-      "rewards/chosen": -1.4074345827102661,
-      "rewards/margins": 0.5764046907424927,
-      "rewards/rejected": -1.9838392734527588,
-      "step": 400
-    },
-    {
-      "epoch": 0.8374771002355405,
-      "eval_logits/chosen": -2.5853030681610107,
-      "eval_logits/rejected": -2.7438056468963623,
-      "eval_logps/chosen": -540.5127563476562,
-      "eval_logps/rejected": -600.98388671875,
-      "eval_loss": 0.5581567287445068,
-      "eval_rewards/accuracies": 0.726190447807312,
-      "eval_rewards/chosen": -1.3602701425552368,
-      "eval_rewards/margins": 0.6926405429840088,
-      "eval_rewards/rejected": -2.052910804748535,
-      "eval_runtime": 38.5105,
-      "eval_samples_per_second": 51.934,
-      "eval_steps_per_second": 1.636,
-      "step": 400
-    },
-    {
-      "epoch": 0.8584140277414289,
-      "grad_norm": 13.568875613159559,
-      "learning_rate": 1.7695632250191e-08,
-      "logits/chosen": -2.2451727390289307,
-      "logits/rejected": -2.3778178691864014,
-      "logps/chosen": -545.51318359375,
-      "logps/rejected": -575.8936157226562,
-      "loss": 0.5625,
-      "rewards/accuracies": 0.6937500238418579,
-      "rewards/chosen": -1.4706048965454102,
-      "rewards/margins": 0.4073447287082672,
-      "rewards/rejected": -1.8779499530792236,
-      "step": 410
-    },
-    {
-      "epoch": 0.8793509552473174,
-      "grad_norm": 15.557207200171755,
-      "learning_rate": 1.2878971655412513e-08,
-      "logits/chosen": -2.2278409004211426,
-      "logits/rejected": -2.379141330718994,
-      "logps/chosen": -526.79248046875,
-      "logps/rejected": -568.4296875,
-      "loss": 0.5456,
-      "rewards/accuracies": 0.6875,
-      "rewards/chosen": -1.2908048629760742,
-      "rewards/margins": 0.5515668392181396,
-      "rewards/rejected": -1.8423715829849243,
-      "step": 420
-    },
-    {
-      "epoch": 0.9002878827532059,
-      "grad_norm": 13.028076908432435,
-      "learning_rate": 8.79732424714541e-09,
-      "logits/chosen": -2.252420663833618,
-      "logits/rejected": -2.4495749473571777,
-      "logps/chosen": -547.81787109375,
-      "logps/rejected": -600.2841186523438,
-      "loss": 0.5432,
-      "rewards/accuracies": 0.731249988079071,
-      "rewards/chosen": -1.333070993423462,
-      "rewards/margins": 0.7184783220291138,
-      "rewards/rejected": -2.051549196243286,
-      "step": 430
-    },
-    {
-      "epoch": 0.9212248102590945,
-      "grad_norm": 11.81802801299209,
-      "learning_rate": 5.472568979361852e-09,
-      "logits/chosen": -2.3049111366271973,
-      "logits/rejected": -2.2651658058166504,
-      "logps/chosen": -557.698486328125,
-      "logps/rejected": -628.7259521484375,
-      "loss": 0.5597,
-      "rewards/accuracies": 0.7250000238418579,
-      "rewards/chosen": -1.3383581638336182,
-      "rewards/margins": 0.5372647047042847,
-      "rewards/rejected": -1.8756227493286133,
-      "step": 440
-    },
-    {
-      "epoch": 0.942161737764983,
-      "grad_norm": 13.690789391511412,
-      "learning_rate": 2.922527618666465e-09,
-      "logits/chosen": -2.1817500591278076,
-      "logits/rejected": -2.4527294635772705,
-      "logps/chosen": -550.814697265625,
-      "logps/rejected": -563.7845458984375,
-      "loss": 0.5551,
-      "rewards/accuracies": 0.7124999761581421,
-      "rewards/chosen": -1.2616031169891357,
-      "rewards/margins": 0.5693789720535278,
-      "rewards/rejected": -1.8309822082519531,
-      "step": 450
-    },
-    {
-      "epoch": 0.9630986652708715,
-      "grad_norm": 16.036131164059057,
-      "learning_rate": 1.1608692138469378e-09,
-      "logits/chosen": -2.184504508972168,
-      "logits/rejected": -2.432163715362549,
-      "logps/chosen": -579.7196044921875,
-      "logps/rejected": -595.66162109375,
-      "loss": 0.5728,
-      "rewards/accuracies": 0.699999988079071,
-      "rewards/chosen": -1.4352037906646729,
-      "rewards/margins": 0.5179115533828735,
-      "rewards/rejected": -1.9531152248382568,
-      "step": 460
-    },
-    {
-      "epoch": 0.98403559277676,
-      "grad_norm": 10.779013615465365,
-      "learning_rate": 1.9703682533901976e-10,
-      "logits/chosen": -2.1344995498657227,
-      "logits/rejected": -2.3982386589050293,
-      "logps/chosen": -520.6542358398438,
-      "logps/rejected": -542.9959716796875,
-      "loss": 0.5528,
-      "rewards/accuracies": 0.7562500238418579,
-      "rewards/chosen": -1.2908748388290405,
-      "rewards/margins": 0.5777503252029419,
-      "rewards/rejected": -1.8686254024505615,
-      "step": 470
-    },
-    {
-      "epoch": 0.998691442030882,
-      "step": 477,
+      "epoch": 0.9969104016477858,
+      "step": 121,
       "total_flos": 0.0,
-      "train_loss": 0.5939158873488068,
-      "train_runtime": 3918.7641,
-      "train_samples_per_second": 15.601,
-      "train_steps_per_second": 0.122
+      "train_loss": 0.6619468080110786,
+      "train_runtime": 1288.9507,
+      "train_samples_per_second": 48.206,
+      "train_steps_per_second": 0.094
     }
   ],
   "logging_steps": 10,
-  "max_steps": 477,
+  "max_steps": 121,
   "num_input_tokens_seen": 0,
   "num_train_epochs": 1,
   "save_steps": 100,