Model save

README.md

@@ -1,10 +1,8 @@
 ---
-datasets: open-r1/OpenR1-Math-220k
 library_name: transformers
 model_name: Qwen2.5-0.5B-Open-R1-Distill
 tags:
 - generated_from_trainer
-- open-r1
 - trl
 - sft
 licence: license
@@ -12,7 +10,7 @@ licence: license
 
 # Model Card for Qwen2.5-0.5B-Open-R1-Distill
 
-This model is a fine-tuned version of [None](https://huggingface.co/None) on the [open-r1/OpenR1-Math-220k](https://huggingface.co/datasets/open-r1/OpenR1-Math-220k) dataset.
+This model is a fine-tuned version of [None](https://huggingface.co/None).
 It has been trained using [TRL](https://github.com/huggingface/trl).
 
 ## Quick start
@@ -28,7 +26,7 @@ print(output["generated_text"])
 
 ## Training procedure
 
-[<img src="https://raw.githubusercontent.com/wandb/assets/main/wandb-github-badge-28.svg" alt="Visualize in Weights & Biases" width="150" height="24"/>](https://wandb.ai/648848014/huggingface/runs/0iamwd3c) 
+[<img src="https://raw.githubusercontent.com/wandb/assets/main/wandb-github-badge-28.svg" alt="Visualize in Weights & Biases" width="150" height="24"/>](https://wandb.ai/648848014/huggingface/runs/56r8ofq1) 
 
 
 This model was trained with SFT.

all_results.json

@@ -1,8 +1,8 @@
 {
-    "total_flos": 1.3158373184765952e+17,
-    "train_loss": 0.0,
-    "train_runtime": 2.524,
+    "total_flos": 1.3211147408769024e+17,
+    "train_loss": 0.4334589367217206,
+    "train_runtime": 1817.3607,
     "train_samples": 3472,
-    "train_samples_per_second": 5950.808,
-    "train_steps_per_second": 372.421
+    "train_samples_per_second": 8.265,
+    "train_steps_per_second": 0.517
 }

train_results.json

@@ -1,8 +1,8 @@
 {
-    "total_flos": 1.3158373184765952e+17,
-    "train_loss": 0.0,
-    "train_runtime": 2.524,
+    "total_flos": 1.3211147408769024e+17,
+    "train_loss": 0.4334589367217206,
+    "train_runtime": 1817.3607,
     "train_samples": 3472,
-    "train_samples_per_second": 5950.808,
-    "train_steps_per_second": 372.421
+    "train_samples_per_second": 8.265,
+    "train_steps_per_second": 0.517
 }

trainer_state.json

@@ -10,1150 +10,1150 @@
   "log_history": [
     {
       "epoch": 0.10638297872340426,
-      "grad_norm": 7.65625,
+      "grad_norm": 7.71875,
       "learning_rate": 5.319148936170213e-06,
-      "loss": 1.8681,
+      "loss": 1.874,
       "step": 5
     },
     {
       "epoch": 0.2127659574468085,
-      "grad_norm": 4.03125,
+      "grad_norm": 3.96875,
       "learning_rate": 1.0638297872340426e-05,
-      "loss": 1.8104,
+      "loss": 1.8146,
       "step": 10
     },
     {
       "epoch": 0.3191489361702128,
-      "grad_norm": 3.078125,
+      "grad_norm": 3.09375,
       "learning_rate": 1.595744680851064e-05,
-      "loss": 1.6921,
+      "loss": 1.6658,
       "step": 15
     },
     {
       "epoch": 0.425531914893617,
-      "grad_norm": 2.796875,
+      "grad_norm": 2.5625,
       "learning_rate": 2.1276595744680852e-05,
-      "loss": 1.6016,
+      "loss": 1.5605,
       "step": 20
     },
     {
       "epoch": 0.5319148936170213,
-      "grad_norm": 2.34375,
+      "grad_norm": 2.296875,
       "learning_rate": 2.6595744680851064e-05,
-      "loss": 1.4323,
+      "loss": 1.4474,
       "step": 25
     },
     {
       "epoch": 0.6382978723404256,
-      "grad_norm": 2.03125,
+      "grad_norm": 1.953125,
       "learning_rate": 3.191489361702128e-05,
-      "loss": 1.3657,
+      "loss": 1.367,
       "step": 30
     },
     {
       "epoch": 0.7446808510638298,
-      "grad_norm": 2.015625,
+      "grad_norm": 2.09375,
       "learning_rate": 3.723404255319149e-05,
-      "loss": 1.3284,
+      "loss": 1.3428,
       "step": 35
     },
     {
       "epoch": 0.851063829787234,
-      "grad_norm": 1.953125,
+      "grad_norm": 2.0625,
       "learning_rate": 4.2553191489361704e-05,
-      "loss": 1.2982,
+      "loss": 1.2922,
       "step": 40
     },
     {
       "epoch": 0.9574468085106383,
-      "grad_norm": 2.140625,
+      "grad_norm": 2.4375,
       "learning_rate": 4.787234042553192e-05,
-      "loss": 1.229,
+      "loss": 1.2636,
       "step": 45
     },
     {
       "epoch": 1.0638297872340425,
-      "grad_norm": 2.171875,
+      "grad_norm": 2.140625,
       "learning_rate": 4.9998746893945805e-05,
-      "loss": 1.1899,
+      "loss": 1.2129,
       "step": 50
     },
     {
       "epoch": 1.1702127659574468,
-      "grad_norm": 2.171875,
+      "grad_norm": 1.984375,
       "learning_rate": 4.99910895290836e-05,
-      "loss": 1.1671,
+      "loss": 1.1482,
       "step": 55
     },
     {
       "epoch": 1.2765957446808511,
-      "grad_norm": 2.03125,
+      "grad_norm": 2.15625,
       "learning_rate": 4.997647333573498e-05,
-      "loss": 1.139,
+      "loss": 1.1343,
       "step": 60
     },
     {
       "epoch": 1.3829787234042552,
-      "grad_norm": 1.953125,
+      "grad_norm": 2.140625,
       "learning_rate": 4.995490283620569e-05,
-      "loss": 1.1244,
+      "loss": 1.1265,
       "step": 65
     },
     {
       "epoch": 1.4893617021276595,
-      "grad_norm": 2.265625,
+      "grad_norm": 1.9453125,
       "learning_rate": 4.992638470449016e-05,
-      "loss": 1.0981,
+      "loss": 1.1193,
       "step": 70
     },
     {
       "epoch": 1.5957446808510638,
-      "grad_norm": 1.9765625,
+      "grad_norm": 1.9921875,
       "learning_rate": 4.98909277642066e-05,
-      "loss": 1.0794,
+      "loss": 1.1042,
       "step": 75
     },
     {
       "epoch": 1.702127659574468,
-      "grad_norm": 1.9765625,
+      "grad_norm": 2.171875,
       "learning_rate": 4.984854298586695e-05,
-      "loss": 1.1011,
+      "loss": 1.0683,
       "step": 80
     },
     {
       "epoch": 1.8085106382978724,
-      "grad_norm": 1.8046875,
+      "grad_norm": 1.8984375,
       "learning_rate": 4.979924348348256e-05,
-      "loss": 1.0679,
+      "loss": 1.0495,
       "step": 85
     },
     {
       "epoch": 1.9148936170212765,
-      "grad_norm": 2.015625,
+      "grad_norm": 1.8125,
       "learning_rate": 4.9743044510506596e-05,
-      "loss": 1.0971,
+      "loss": 1.0677,
       "step": 90
     },
     {
       "epoch": 2.021276595744681,
-      "grad_norm": 2.09375,
+      "grad_norm": 2.109375,
       "learning_rate": 4.967996345511468e-05,
-      "loss": 1.0524,
+      "loss": 1.058,
       "step": 95
     },
     {
       "epoch": 2.127659574468085,
-      "grad_norm": 2.296875,
+      "grad_norm": 2.484375,
       "learning_rate": 4.9610019834824824e-05,
-      "loss": 0.9611,
+      "loss": 0.9445,
       "step": 100
     },
     {
       "epoch": 2.2340425531914896,
-      "grad_norm": 2.1875,
+      "grad_norm": 2.140625,
       "learning_rate": 4.953323529045868e-05,
-      "loss": 0.9699,
+      "loss": 0.9534,
       "step": 105
     },
     {
       "epoch": 2.3404255319148937,
-      "grad_norm": 2.15625,
+      "grad_norm": 1.9375,
       "learning_rate": 4.9449633579445785e-05,
-      "loss": 0.9598,
+      "loss": 0.9609,
       "step": 110
     },
     {
       "epoch": 2.4468085106382977,
-      "grad_norm": 2.25,
+      "grad_norm": 1.9765625,
       "learning_rate": 4.935924056847292e-05,
-      "loss": 0.9543,
+      "loss": 0.9451,
       "step": 115
     },
     {
       "epoch": 2.5531914893617023,
-      "grad_norm": 1.875,
+      "grad_norm": 1.9375,
       "learning_rate": 4.926208422548085e-05,
-      "loss": 0.9315,
+      "loss": 0.9381,
       "step": 120
     },
     {
       "epoch": 2.6595744680851063,
-      "grad_norm": 1.890625,
+      "grad_norm": 2.03125,
       "learning_rate": 4.915819461101094e-05,
-      "loss": 0.9201,
+      "loss": 0.9412,
       "step": 125
     },
     {
       "epoch": 2.7659574468085104,
-      "grad_norm": 2.109375,
+      "grad_norm": 2.0625,
       "learning_rate": 4.904760386890431e-05,
-      "loss": 0.9237,
+      "loss": 0.9306,
       "step": 130
     },
     {
       "epoch": 2.872340425531915,
-      "grad_norm": 2.015625,
+      "grad_norm": 2.046875,
       "learning_rate": 4.893034621635644e-05,
-      "loss": 0.9201,
+      "loss": 0.9287,
       "step": 135
     },
     {
       "epoch": 2.978723404255319,
-      "grad_norm": 2.09375,
+      "grad_norm": 2.0625,
       "learning_rate": 4.8806457933330185e-05,
-      "loss": 0.9136,
+      "loss": 0.9184,
       "step": 140
     },
     {
       "epoch": 3.0851063829787235,
       "grad_norm": 2.1875,
       "learning_rate": 4.8675977351330635e-05,
-      "loss": 0.8568,
+      "loss": 0.83,
       "step": 145
     },
     {
       "epoch": 3.1914893617021276,
-      "grad_norm": 2.140625,
+      "grad_norm": 2.015625,
       "learning_rate": 4.853894484154523e-05,
-      "loss": 0.8213,
+      "loss": 0.8281,
       "step": 150
     },
     {
       "epoch": 3.297872340425532,
-      "grad_norm": 2.3125,
+      "grad_norm": 2.234375,
       "learning_rate": 4.839540280235274e-05,
-      "loss": 0.7954,
+      "loss": 0.8054,
       "step": 155
     },
     {
       "epoch": 3.404255319148936,
-      "grad_norm": 2.0,
+      "grad_norm": 2.03125,
       "learning_rate": 4.8245395646205024e-05,
-      "loss": 0.809,
+      "loss": 0.8155,
       "step": 160
     },
     {
       "epoch": 3.5106382978723403,
-      "grad_norm": 2.125,
+      "grad_norm": 2.140625,
       "learning_rate": 4.808896978588571e-05,
-      "loss": 0.8244,
+      "loss": 0.799,
       "step": 165
     },
     {
       "epoch": 3.617021276595745,
-      "grad_norm": 2.0625,
+      "grad_norm": 2.171875,
       "learning_rate": 4.792617362014982e-05,
-      "loss": 0.7982,
+      "loss": 0.8008,
       "step": 170
     },
     {
       "epoch": 3.723404255319149,
-      "grad_norm": 2.0,
+      "grad_norm": 2.125,
       "learning_rate": 4.7757057518749085e-05,
-      "loss": 0.8034,
+      "loss": 0.7805,
       "step": 175
     },
     {
       "epoch": 3.829787234042553,
-      "grad_norm": 2.015625,
+      "grad_norm": 2.140625,
       "learning_rate": 4.758167380684733e-05,
-      "loss": 0.7917,
+      "loss": 0.8217,
       "step": 180
     },
     {
       "epoch": 3.9361702127659575,
-      "grad_norm": 2.25,
+      "grad_norm": 2.53125,
       "learning_rate": 4.7400076748830835e-05,
-      "loss": 0.8016,
+      "loss": 0.8064,
       "step": 185
     },
     {
       "epoch": 4.042553191489362,
-      "grad_norm": 2.125,
+      "grad_norm": 2.296875,
       "learning_rate": 4.721232253151879e-05,
-      "loss": 0.7626,
+      "loss": 0.7495,
       "step": 190
     },
     {
       "epoch": 4.148936170212766,
-      "grad_norm": 2.15625,
+      "grad_norm": 2.0625,
       "learning_rate": 4.701846924677881e-05,
-      "loss": 0.7081,
+      "loss": 0.7,
       "step": 195
     },
     {
       "epoch": 4.25531914893617,
-      "grad_norm": 1.984375,
+      "grad_norm": 2.109375,
       "learning_rate": 4.681857687355316e-05,
-      "loss": 0.6809,
+      "loss": 0.6738,
       "step": 200
     },
     {
       "epoch": 4.361702127659575,
-      "grad_norm": 1.9453125,
+      "grad_norm": 2.015625,
       "learning_rate": 4.661270725930099e-05,
-      "loss": 0.7027,
+      "loss": 0.6929,
       "step": 205
     },
     {
       "epoch": 4.468085106382979,
-      "grad_norm": 2.15625,
+      "grad_norm": 2.125,
       "learning_rate": 4.6400924100862526e-05,
-      "loss": 0.6892,
+      "loss": 0.6916,
       "step": 210
     },
     {
       "epoch": 4.574468085106383,
-      "grad_norm": 2.1875,
+      "grad_norm": 2.59375,
       "learning_rate": 4.6183292924751005e-05,
-      "loss": 0.6986,
+      "loss": 0.6948,
       "step": 215
     },
     {
       "epoch": 4.680851063829787,
-      "grad_norm": 2.328125,
+      "grad_norm": 2.53125,
       "learning_rate": 4.595988106687854e-05,
-      "loss": 0.692,
+      "loss": 0.6938,
       "step": 220
     },
     {
       "epoch": 4.787234042553192,
-      "grad_norm": 2.015625,
+      "grad_norm": 2.203125,
       "learning_rate": 4.573075765172214e-05,
-      "loss": 0.6969,
+      "loss": 0.6836,
       "step": 225
     },
     {
       "epoch": 4.8936170212765955,
-      "grad_norm": 2.25,
+      "grad_norm": 2.515625,
       "learning_rate": 4.549599357093629e-05,
-      "loss": 0.6812,
+      "loss": 0.6804,
       "step": 230
     },
     {
       "epoch": 5.0,
-      "grad_norm": 2.171875,
+      "grad_norm": 2.25,
       "learning_rate": 4.5255661461418854e-05,
-      "loss": 0.6853,
+      "loss": 0.6922,
       "step": 235
     },
     {
       "epoch": 5.1063829787234045,
-      "grad_norm": 2.46875,
+      "grad_norm": 2.15625,
       "learning_rate": 4.500983568283694e-05,
-      "loss": 0.5988,
+      "loss": 0.5917,
       "step": 240
     },
     {
       "epoch": 5.212765957446808,
       "grad_norm": 2.1875,
       "learning_rate": 4.4758592294619696e-05,
-      "loss": 0.5863,
+      "loss": 0.582,
       "step": 245
     },
     {
       "epoch": 5.319148936170213,
-      "grad_norm": 2.28125,
+      "grad_norm": 2.15625,
       "learning_rate": 4.450200903242526e-05,
-      "loss": 0.5902,
+      "loss": 0.5846,
       "step": 250
     },
     {
       "epoch": 5.425531914893617,
       "grad_norm": 2.171875,
       "learning_rate": 4.4240165284089065e-05,
-      "loss": 0.5869,
+      "loss": 0.5845,
       "step": 255
     },
     {
       "epoch": 5.531914893617021,
-      "grad_norm": 2.078125,
+      "grad_norm": 2.21875,
       "learning_rate": 4.397314206506092e-05,
-      "loss": 0.5789,
+      "loss": 0.5968,
       "step": 260
     },
     {
       "epoch": 5.638297872340425,
-      "grad_norm": 2.03125,
+      "grad_norm": 2.15625,
       "learning_rate": 4.3701021993338534e-05,
-      "loss": 0.579,
+      "loss": 0.5739,
       "step": 265
     },
     {
       "epoch": 5.74468085106383,
-      "grad_norm": 2.15625,
+      "grad_norm": 2.171875,
       "learning_rate": 4.342388926390517e-05,
-      "loss": 0.5854,
+      "loss": 0.5974,
       "step": 270
     },
     {
       "epoch": 5.851063829787234,
-      "grad_norm": 2.171875,
+      "grad_norm": 2.140625,
       "learning_rate": 4.314182962267946e-05,
-      "loss": 0.5823,
+      "loss": 0.5955,
       "step": 275
     },
     {
       "epoch": 5.957446808510638,
-      "grad_norm": 2.03125,
+      "grad_norm": 2.0625,
       "learning_rate": 4.285493033998519e-05,
-      "loss": 0.5968,
+      "loss": 0.5848,
       "step": 280
     },
     {
       "epoch": 6.0638297872340425,
-      "grad_norm": 2.28125,
+      "grad_norm": 2.375,
       "learning_rate": 4.256328018354964e-05,
-      "loss": 0.5438,
+      "loss": 0.5369,
       "step": 285
     },
     {
       "epoch": 6.170212765957447,
       "grad_norm": 2.21875,
       "learning_rate": 4.226696939103846e-05,
-      "loss": 0.4993,
+      "loss": 0.4998,
       "step": 290
     },
     {
       "epoch": 6.276595744680851,
-      "grad_norm": 2.078125,
+      "grad_norm": 2.21875,
       "learning_rate": 4.196608964213574e-05,
-      "loss": 0.4965,
+      "loss": 0.4911,
       "step": 295
     },
     {
       "epoch": 6.382978723404255,
-      "grad_norm": 2.25,
+      "grad_norm": 2.15625,
       "learning_rate": 4.16607340301781e-05,
-      "loss": 0.4886,
+      "loss": 0.4974,
       "step": 300
     },
     {
       "epoch": 6.48936170212766,
-      "grad_norm": 2.03125,
+      "grad_norm": 2.078125,
       "learning_rate": 4.1350997033351126e-05,
-      "loss": 0.5026,
+      "loss": 0.4991,
       "step": 305
     },
     {
       "epoch": 6.595744680851064,
-      "grad_norm": 2.0625,
+      "grad_norm": 2.125,
       "learning_rate": 4.103697448545753e-05,
-      "loss": 0.4985,
+      "loss": 0.499,
       "step": 310
     },
     {
       "epoch": 6.702127659574468,
-      "grad_norm": 2.109375,
+      "grad_norm": 2.140625,
       "learning_rate": 4.071876354626581e-05,
-      "loss": 0.4953,
+      "loss": 0.4991,
       "step": 315
     },
     {
       "epoch": 6.808510638297872,
-      "grad_norm": 2.03125,
+      "grad_norm": 2.09375,
       "learning_rate": 4.0396462671448616e-05,
-      "loss": 0.5016,
+      "loss": 0.5022,
       "step": 320
     },
     {
       "epoch": 6.914893617021277,
-      "grad_norm": 1.8828125,
+      "grad_norm": 2.03125,
       "learning_rate": 4.007017158212022e-05,
-      "loss": 0.4867,
+      "loss": 0.5008,
       "step": 325
     },
     {
       "epoch": 7.0212765957446805,
-      "grad_norm": 2.375,
+      "grad_norm": 2.625,
       "learning_rate": 3.9739991233982394e-05,
-      "loss": 0.4817,
+      "loss": 0.4899,
       "step": 330
     },
     {
       "epoch": 7.127659574468085,
-      "grad_norm": 2.328125,
+      "grad_norm": 2.203125,
       "learning_rate": 3.9406023786088356e-05,
-      "loss": 0.4092,
+      "loss": 0.4098,
       "step": 335
     },
     {
       "epoch": 7.23404255319149,
-      "grad_norm": 2.328125,
+      "grad_norm": 2.234375,
       "learning_rate": 3.906837256923439e-05,
-      "loss": 0.4203,
+      "loss": 0.4213,
       "step": 340
     },
     {
       "epoch": 7.340425531914893,
-      "grad_norm": 2.0625,
+      "grad_norm": 2.046875,
       "learning_rate": 3.872714205398886e-05,
-      "loss": 0.4135,
+      "loss": 0.4184,
       "step": 345
     },
     {
       "epoch": 7.446808510638298,
-      "grad_norm": 2.15625,
+      "grad_norm": 2.140625,
       "learning_rate": 3.838243781836864e-05,
-      "loss": 0.423,
+      "loss": 0.4223,
       "step": 350
     },
     {
       "epoch": 7.553191489361702,
-      "grad_norm": 2.109375,
+      "grad_norm": 2.15625,
       "learning_rate": 3.803436651517288e-05,
-      "loss": 0.3946,
+      "loss": 0.4052,
       "step": 355
     },
     {
       "epoch": 7.659574468085106,
-      "grad_norm": 2.1875,
+      "grad_norm": 2.25,
       "learning_rate": 3.7683035838984224e-05,
-      "loss": 0.4204,
+      "loss": 0.4189,
       "step": 360
     },
     {
       "epoch": 7.76595744680851,
-      "grad_norm": 2.203125,
+      "grad_norm": 2.3125,
       "learning_rate": 3.732855449284769e-05,
-      "loss": 0.43,
+      "loss": 0.4152,
       "step": 365
     },
     {
       "epoch": 7.872340425531915,
-      "grad_norm": 2.140625,
+      "grad_norm": 2.21875,
       "learning_rate": 3.697103215463754e-05,
-      "loss": 0.4231,
+      "loss": 0.4349,
       "step": 370
     },
     {
       "epoch": 7.9787234042553195,
-      "grad_norm": 2.265625,
+      "grad_norm": 2.171875,
       "learning_rate": 3.661057944312252e-05,
-      "loss": 0.4282,
+      "loss": 0.4235,
       "step": 375
     },
     {
       "epoch": 8.085106382978724,
-      "grad_norm": 2.5625,
+      "grad_norm": 2.75,
       "learning_rate": 3.624730788374003e-05,
-      "loss": 0.3744,
+      "loss": 0.3635,
       "step": 380
     },
     {
       "epoch": 8.191489361702128,
-      "grad_norm": 2.265625,
+      "grad_norm": 2.125,
       "learning_rate": 3.5881329874089664e-05,
-      "loss": 0.3512,
+      "loss": 0.355,
       "step": 385
     },
     {
       "epoch": 8.297872340425531,
-      "grad_norm": 2.265625,
+      "grad_norm": 2.15625,
       "learning_rate": 3.551275864915703e-05,
-      "loss": 0.335,
+      "loss": 0.3456,
       "step": 390
     },
     {
       "epoch": 8.404255319148936,
-      "grad_norm": 2.234375,
+      "grad_norm": 2.171875,
       "learning_rate": 3.5141708246278385e-05,
-      "loss": 0.3427,
+      "loss": 0.3448,
       "step": 395
     },
     {
       "epoch": 8.51063829787234,
-      "grad_norm": 2.171875,
+      "grad_norm": 2.125,
       "learning_rate": 3.476829346985704e-05,
-      "loss": 0.3489,
+      "loss": 0.3448,
       "step": 400
     },
     {
       "epoch": 8.617021276595745,
       "grad_norm": 2.125,
       "learning_rate": 3.4392629855842475e-05,
-      "loss": 0.3559,
+      "loss": 0.3551,
       "step": 405
     },
     {
       "epoch": 8.72340425531915,
-      "grad_norm": 2.109375,
+      "grad_norm": 2.234375,
       "learning_rate": 3.401483363598305e-05,
-      "loss": 0.3538,
+      "loss": 0.3574,
       "step": 410
     },
     {
       "epoch": 8.829787234042554,
-      "grad_norm": 2.140625,
+      "grad_norm": 2.0625,
       "learning_rate": 3.363502170186346e-05,
-      "loss": 0.3509,
+      "loss": 0.3502,
       "step": 415
     },
     {
       "epoch": 8.936170212765958,
-      "grad_norm": 2.21875,
+      "grad_norm": 2.1875,
       "learning_rate": 3.325331156873807e-05,
-      "loss": 0.3531,
+      "loss": 0.3579,
       "step": 420
     },
     {
       "epoch": 9.042553191489361,
-      "grad_norm": 2.1875,
+      "grad_norm": 2.046875,
       "learning_rate": 3.2869821339171195e-05,
-      "loss": 0.33,
+      "loss": 0.3326,
       "step": 425
     },
     {
       "epoch": 9.148936170212766,
-      "grad_norm": 2.328125,
+      "grad_norm": 2.28125,
       "learning_rate": 3.248466966649579e-05,
-      "loss": 0.2935,
+      "loss": 0.2921,
       "step": 430
     },
     {
       "epoch": 9.25531914893617,
-      "grad_norm": 2.046875,
+      "grad_norm": 2.15625,
       "learning_rate": 3.2097975718101544e-05,
-      "loss": 0.282,
+      "loss": 0.2834,
       "step": 435
     },
     {
       "epoch": 9.361702127659575,
-      "grad_norm": 2.15625,
+      "grad_norm": 2.375,
       "learning_rate": 3.170985913856418e-05,
-      "loss": 0.2882,
+      "loss": 0.2942,
       "step": 440
     },
     {
       "epoch": 9.46808510638298,
-      "grad_norm": 2.3125,
+      "grad_norm": 2.09375,
       "learning_rate": 3.1320440012626864e-05,
-      "loss": 0.3017,
+      "loss": 0.294,
       "step": 445
     },
     {
       "epoch": 9.574468085106384,
-      "grad_norm": 2.1875,
+      "grad_norm": 2.140625,
       "learning_rate": 3.0929838828045566e-05,
-      "loss": 0.2897,
+      "loss": 0.2844,
       "step": 450
     },
     {
       "epoch": 9.680851063829786,
-      "grad_norm": 2.15625,
+      "grad_norm": 2.125,
       "learning_rate": 3.053817643830969e-05,
-      "loss": 0.2889,
+      "loss": 0.295,
       "step": 455
     },
     {
       "epoch": 9.787234042553191,
-      "grad_norm": 2.15625,
+      "grad_norm": 2.078125,
       "learning_rate": 3.0145574025249544e-05,
-      "loss": 0.2911,
+      "loss": 0.2875,
       "step": 460
     },
     {
       "epoch": 9.893617021276595,
-      "grad_norm": 2.28125,
+      "grad_norm": 2.25,
       "learning_rate": 2.975215306154222e-05,
-      "loss": 0.2988,
+      "loss": 0.3021,
       "step": 465
     },
     {
       "epoch": 10.0,
-      "grad_norm": 2.421875,
+      "grad_norm": 2.375,
       "learning_rate": 2.9358035273127483e-05,
-      "loss": 0.2952,
+      "loss": 0.296,
       "step": 470
     },
     {
       "epoch": 10.106382978723405,
-      "grad_norm": 2.484375,
+      "grad_norm": 2.40625,
       "learning_rate": 2.896334260154532e-05,
-      "loss": 0.2421,
+      "loss": 0.2404,
       "step": 475
     },
     {
       "epoch": 10.212765957446809,
-      "grad_norm": 2.171875,
+      "grad_norm": 2.0,
       "learning_rate": 2.856819716620674e-05,
-      "loss": 0.2477,
+      "loss": 0.2408,
       "step": 480
     },
     {
       "epoch": 10.319148936170214,
-      "grad_norm": 2.109375,
+      "grad_norm": 2.1875,
       "learning_rate": 2.81727212266096e-05,
-      "loss": 0.2408,
+      "loss": 0.2386,
       "step": 485
     },
     {
       "epoch": 10.425531914893616,
       "grad_norm": 2.1875,
       "learning_rate": 2.7777037144510993e-05,
-      "loss": 0.2371,
+      "loss": 0.2381,
       "step": 490
     },
     {
       "epoch": 10.53191489361702,
-      "grad_norm": 2.015625,
+      "grad_norm": 1.90625,
       "learning_rate": 2.7381267346068112e-05,
-      "loss": 0.2335,
+      "loss": 0.2355,
       "step": 495
     },
     {
       "epoch": 10.638297872340425,
-      "grad_norm": 2.015625,
+      "grad_norm": 2.125,
       "learning_rate": 2.6985534283959123e-05,
-      "loss": 0.2432,
+      "loss": 0.2509,
       "step": 500
     },
     {
       "epoch": 10.74468085106383,
-      "grad_norm": 2.125,
+      "grad_norm": 2.1875,
       "learning_rate": 2.6589960399495834e-05,
-      "loss": 0.2472,
+      "loss": 0.2449,
       "step": 505
     },
     {
       "epoch": 10.851063829787234,
-      "grad_norm": 2.140625,
+      "grad_norm": 2.125,
       "learning_rate": 2.6194668084739833e-05,
-      "loss": 0.2459,
+      "loss": 0.2462,
       "step": 510
     },
     {
       "epoch": 10.957446808510639,
-      "grad_norm": 2.09375,
+      "grad_norm": 2.015625,
       "learning_rate": 2.579977964463403e-05,
-      "loss": 0.2453,
+      "loss": 0.2482,
       "step": 515
     },
     {
       "epoch": 11.063829787234043,
-      "grad_norm": 2.015625,
+      "grad_norm": 1.96875,
       "learning_rate": 2.5405417259160968e-05,
-      "loss": 0.2259,
+      "loss": 0.2267,
       "step": 520
     },
     {
       "epoch": 11.170212765957446,
-      "grad_norm": 2.21875,
+      "grad_norm": 2.046875,
       "learning_rate": 2.5011702945539882e-05,
-      "loss": 0.2042,
+      "loss": 0.1964,
       "step": 525
     },
     {
       "epoch": 11.27659574468085,
-      "grad_norm": 2.015625,
+      "grad_norm": 1.9765625,
       "learning_rate": 2.4618758520474116e-05,
-      "loss": 0.1996,
+      "loss": 0.1997,
       "step": 530
     },
     {
       "epoch": 11.382978723404255,
-      "grad_norm": 2.0625,
+      "grad_norm": 2.125,
       "learning_rate": 2.422670556246059e-05,
-      "loss": 0.2063,
+      "loss": 0.2016,
       "step": 535
     },
     {
       "epoch": 11.48936170212766,
-      "grad_norm": 1.9140625,
+      "grad_norm": 2.21875,
       "learning_rate": 2.38356653741729e-05,
-      "loss": 0.203,
+      "loss": 0.2074,
       "step": 540
     },
     {
       "epoch": 11.595744680851064,
-      "grad_norm": 1.9453125,
+      "grad_norm": 2.078125,
       "learning_rate": 2.3445758944929875e-05,
-      "loss": 0.1979,
+      "loss": 0.197,
       "step": 545
     },
     {
       "epoch": 11.702127659574469,
-      "grad_norm": 1.8046875,
+      "grad_norm": 1.8671875,
       "learning_rate": 2.3057106913260972e-05,
-      "loss": 0.1985,
+      "loss": 0.2065,
       "step": 550
     },
     {
       "epoch": 11.808510638297872,
-      "grad_norm": 2.09375,
+      "grad_norm": 2.046875,
       "learning_rate": 2.266982952958029e-05,
-      "loss": 0.2078,
+      "loss": 0.2105,
       "step": 555
     },
     {
       "epoch": 11.914893617021276,
-      "grad_norm": 2.03125,
+      "grad_norm": 1.9609375,
       "learning_rate": 2.2284046618980613e-05,
-      "loss": 0.2066,
+      "loss": 0.2061,
       "step": 560
     },
     {
       "epoch": 12.02127659574468,
-      "grad_norm": 1.9140625,
+      "grad_norm": 1.84375,
       "learning_rate": 2.1899877544159092e-05,
-      "loss": 0.2014,
+      "loss": 0.2009,
       "step": 565
     },
     {
       "epoch": 12.127659574468085,
       "grad_norm": 2.0625,
       "learning_rate": 2.1517441168485936e-05,
-      "loss": 0.1756,
+      "loss": 0.1751,
       "step": 570
     },
     {
       "epoch": 12.23404255319149,
-      "grad_norm": 2.03125,
+      "grad_norm": 1.84375,
       "learning_rate": 2.1136855819227686e-05,
-      "loss": 0.1743,
+      "loss": 0.1741,
       "step": 575
     },
     {
       "epoch": 12.340425531914894,
-      "grad_norm": 1.859375,
+      "grad_norm": 1.984375,
       "learning_rate": 2.075823925093627e-05,
-      "loss": 0.1699,
+      "loss": 0.1727,
       "step": 580
     },
     {
       "epoch": 12.446808510638299,
-      "grad_norm": 1.8046875,
+      "grad_norm": 1.734375,
       "learning_rate": 2.038170860901531e-05,
-      "loss": 0.1696,
+      "loss": 0.1685,
       "step": 585
     },
     {
       "epoch": 12.553191489361701,
-      "grad_norm": 1.9375,
+      "grad_norm": 1.8671875,
       "learning_rate": 2.00073803934749e-05,
-      "loss": 0.1765,
+      "loss": 0.1707,
       "step": 590
     },
     {
       "epoch": 12.659574468085106,
-      "grad_norm": 1.984375,
+      "grad_norm": 1.828125,
       "learning_rate": 1.9635370422886033e-05,
-      "loss": 0.1721,
+      "loss": 0.1762,
       "step": 595
     },
     {
       "epoch": 12.76595744680851,
-      "grad_norm": 1.96875,
+      "grad_norm": 1.796875,
       "learning_rate": 1.9265793798545934e-05,
-      "loss": 0.1799,
+      "loss": 0.1737,
       "step": 600
     },
     {
       "epoch": 12.872340425531915,
-      "grad_norm": 1.90625,
+      "grad_norm": 1.828125,
       "learning_rate": 1.8898764868865243e-05,
-      "loss": 0.1765,
+      "loss": 0.1753,
       "step": 605
     },
     {
       "epoch": 12.97872340425532,
-      "grad_norm": 1.8984375,
+      "grad_norm": 1.9296875,
       "learning_rate": 1.8534397193988202e-05,
-      "loss": 0.1729,
+      "loss": 0.1747,
       "step": 610
     },
     {
       "epoch": 13.085106382978724,
-      "grad_norm": 1.6796875,
+      "grad_norm": 1.8203125,
       "learning_rate": 1.8172803510656728e-05,
-      "loss": 0.1589,
+      "loss": 0.1539,
       "step": 615
     },
     {
       "epoch": 13.191489361702128,
-      "grad_norm": 1.8125,
+      "grad_norm": 1.75,
       "learning_rate": 1.781409569732922e-05,
-      "loss": 0.1509,
+      "loss": 0.1467,
       "step": 620
     },
     {
       "epoch": 13.297872340425531,
-      "grad_norm": 1.765625,
+      "grad_norm": 1.7421875,
       "learning_rate": 1.745838473956492e-05,
-      "loss": 0.1484,
+      "loss": 0.1477,
       "step": 625
     },
     {
       "epoch": 13.404255319148936,
-      "grad_norm": 1.7109375,
+      "grad_norm": 1.7265625,
       "learning_rate": 1.71057806956846e-05,
-      "loss": 0.1463,
+      "loss": 0.1545,
       "step": 630
     },
     {
       "epoch": 13.51063829787234,
-      "grad_norm": 1.8359375,
+      "grad_norm": 2.015625,
       "learning_rate": 1.6756392662718072e-05,
-      "loss": 0.1501,
+      "loss": 0.1461,
       "step": 635
     },
     {
       "epoch": 13.617021276595745,
-      "grad_norm": 1.6796875,
+      "grad_norm": 1.78125,
       "learning_rate": 1.6410328742649175e-05,
-      "loss": 0.151,
+      "loss": 0.1538,
       "step": 640
     },
     {
       "epoch": 13.72340425531915,
-      "grad_norm": 1.765625,
+      "grad_norm": 1.7265625,
       "learning_rate": 1.606769600896859e-05,
-      "loss": 0.155,
+      "loss": 0.1521,
       "step": 645
     },
     {
       "epoch": 13.829787234042554,
-      "grad_norm": 1.8359375,
+      "grad_norm": 1.71875,
       "learning_rate": 1.5728600473544896e-05,
-      "loss": 0.1534,
+      "loss": 0.1513,
       "step": 650
     },
     {
       "epoch": 13.936170212765958,
       "grad_norm": 1.796875,
       "learning_rate": 1.539314705382412e-05,
-      "loss": 0.1519,
+      "loss": 0.1511,
       "step": 655
     },
     {
       "epoch": 14.042553191489361,
-      "grad_norm": 1.671875,
+      "grad_norm": 1.4296875,
       "learning_rate": 1.506143954036784e-05,
-      "loss": 0.1479,
+      "loss": 0.1421,
       "step": 660
     },
     {
       "epoch": 14.148936170212766,
-      "grad_norm": 1.703125,
+      "grad_norm": 1.6875,
       "learning_rate": 1.4733580564740035e-05,
-      "loss": 0.1351,
+      "loss": 0.1355,
       "step": 665
     },
     {
       "epoch": 14.25531914893617,
-      "grad_norm": 1.5234375,
+      "grad_norm": 1.71875,
       "learning_rate": 1.4409671567752462e-05,
-      "loss": 0.1321,
+      "loss": 0.1358,
       "step": 670
     },
     {
       "epoch": 14.361702127659575,
-      "grad_norm": 1.6015625,
+      "grad_norm": 1.5625,
       "learning_rate": 1.4089812768078495e-05,
-      "loss": 0.1359,
+      "loss": 0.1325,
       "step": 675
     },
     {
       "epoch": 14.46808510638298,
-      "grad_norm": 1.6796875,
+      "grad_norm": 1.6015625,
       "learning_rate": 1.3774103131245064e-05,
-      "loss": 0.1351,
+      "loss": 0.1325,
       "step": 680
     },
     {
       "epoch": 14.574468085106384,
-      "grad_norm": 1.7109375,
+      "grad_norm": 1.7421875,
       "learning_rate": 1.3462640339012389e-05,
-      "loss": 0.1356,
+      "loss": 0.1331,
       "step": 685
     },
     {
       "epoch": 14.680851063829786,
-      "grad_norm": 1.546875,
+      "grad_norm": 1.640625,
       "learning_rate": 1.3155520759150825e-05,
-      "loss": 0.1351,
+      "loss": 0.1336,
       "step": 690
     },
     {
       "epoch": 14.787234042553191,
-      "grad_norm": 1.6953125,
+      "grad_norm": 1.5546875,
       "learning_rate": 1.2852839415624358e-05,
-      "loss": 0.1351,
+      "loss": 0.1379,
       "step": 695
     },
     {
       "epoch": 14.893617021276595,
-      "grad_norm": 1.6015625,
+      "grad_norm": 1.515625,
       "learning_rate": 1.2554689959189781e-05,
-      "loss": 0.1355,
+      "loss": 0.1349,
       "step": 700
     },
     {
       "epoch": 15.0,
-      "grad_norm": 1.9375,
+      "grad_norm": 1.6875,
       "learning_rate": 1.2261164638420832e-05,
-      "loss": 0.136,
+      "loss": 0.1374,
       "step": 705
     },
     {
       "epoch": 15.106382978723405,
-      "grad_norm": 1.5078125,
+      "grad_norm": 1.453125,
       "learning_rate": 1.1972354271166089e-05,
-      "loss": 0.127,
+      "loss": 0.1248,
       "step": 710
     },
     {
       "epoch": 15.212765957446809,
-      "grad_norm": 1.5625,
+      "grad_norm": 1.5546875,
       "learning_rate": 1.168834821644964e-05,
-      "loss": 0.1225,
+      "loss": 0.1268,
       "step": 715
     },
     {
       "epoch": 15.319148936170214,
-      "grad_norm": 1.5078125,
+      "grad_norm": 1.53125,
       "learning_rate": 1.1409234346823028e-05,
-      "loss": 0.123,
+      "loss": 0.1227,
       "step": 720
     },
     {
       "epoch": 15.425531914893616,
-      "grad_norm": 1.453125,
+      "grad_norm": 1.515625,
       "learning_rate": 1.1135099021177155e-05,
-      "loss": 0.1214,
+      "loss": 0.1247,
       "step": 725
     },
     {
       "epoch": 15.53191489361702,
-      "grad_norm": 1.4609375,
+      "grad_norm": 1.453125,
       "learning_rate": 1.0866027058022515e-05,
-      "loss": 0.1221,
+      "loss": 0.1207,
       "step": 730
     },
     {
       "epoch": 15.638297872340425,
-      "grad_norm": 1.5,
+      "grad_norm": 1.53125,
       "learning_rate": 1.0602101709246026e-05,
-      "loss": 0.1233,
+      "loss": 0.1237,
       "step": 735
     },
     {
       "epoch": 15.74468085106383,
-      "grad_norm": 1.5234375,
+      "grad_norm": 1.46875,
       "learning_rate": 1.0343404634352569e-05,
-      "loss": 0.1248,
+      "loss": 0.1238,
       "step": 740
     },
     {
       "epoch": 15.851063829787234,
       "grad_norm": 1.546875,
       "learning_rate": 1.0090015875199252e-05,
-      "loss": 0.1246,
+      "loss": 0.1251,
       "step": 745
     },
     {
       "epoch": 15.957446808510639,
-      "grad_norm": 1.5390625,
+      "grad_norm": 1.5078125,
       "learning_rate": 9.842013831230129e-06,
-      "loss": 0.1272,
+      "loss": 0.122,
       "step": 750
     },
     {
       "epoch": 16.06382978723404,
-      "grad_norm": 1.4609375,
+      "grad_norm": 1.328125,
       "learning_rate": 9.59947523521913e-06,
-      "loss": 0.1206,
+      "loss": 0.1187,
       "step": 755
     },
     {
       "epoch": 16.170212765957448,
-      "grad_norm": 1.4375,
+      "grad_norm": 1.46875,
       "learning_rate": 9.362475129528648e-06,
-      "loss": 0.1172,
+      "loss": 0.1125,
       "step": 760
     },
     {
       "epoch": 16.27659574468085,
-      "grad_norm": 1.4140625,
+      "grad_norm": 1.4765625,
       "learning_rate": 9.131086842891132e-06,
-      "loss": 0.1165,
+      "loss": 0.1155,
       "step": 765
     },
     {
       "epoch": 16.382978723404257,
-      "grad_norm": 1.3984375,
+      "grad_norm": 1.5,
       "learning_rate": 8.905381967720925e-06,
-      "loss": 0.1153,
+      "loss": 0.1161,
       "step": 770
     },
     {
       "epoch": 16.48936170212766,
-      "grad_norm": 1.390625,
+      "grad_norm": 1.4453125,
       "learning_rate": 8.685430337963278e-06,
-      "loss": 0.1168,
+      "loss": 0.1181,
       "step": 775
     },
     {
       "epoch": 16.595744680851062,
-      "grad_norm": 1.4140625,
+      "grad_norm": 1.4375,
       "learning_rate": 8.471300007487473e-06,
-      "loss": 0.1162,
+      "loss": 0.1173,
       "step": 780
     },
     {
       "epoch": 16.70212765957447,
-      "grad_norm": 1.3984375,
+      "grad_norm": 1.4375,
       "learning_rate": 8.263057229030688e-06,
-      "loss": 0.1157,
+      "loss": 0.1181,
       "step": 785
     },
     {
       "epoch": 16.80851063829787,
-      "grad_norm": 1.4609375,
+      "grad_norm": 1.4921875,
       "learning_rate": 8.060766433699175e-06,
-      "loss": 0.1188,
+      "loss": 0.1168,
       "step": 790
     },
     {
       "epoch": 16.914893617021278,
-      "grad_norm": 1.4453125,
+      "grad_norm": 1.3984375,
       "learning_rate": 7.864490211032975e-06,
-      "loss": 0.1189,
+      "loss": 0.1174,
       "step": 795
     },
     {
       "epoch": 17.02127659574468,
-      "grad_norm": 1.4296875,
+      "grad_norm": 1.328125,
       "learning_rate": 7.67428928964054e-06,
-      "loss": 0.1189,
+      "loss": 0.1153,
       "step": 800
     },
     {
       "epoch": 17.127659574468087,
-      "grad_norm": 1.375,
+      "grad_norm": 1.3984375,
       "learning_rate": 7.490222518409038e-06,
-      "loss": 0.11,
+      "loss": 0.111,
       "step": 805
     },
     {
       "epoch": 17.23404255319149,
-      "grad_norm": 1.40625,
+      "grad_norm": 1.4375,
       "learning_rate": 7.312346848296281e-06,
-      "loss": 0.1146,
+      "loss": 0.1105,
       "step": 810
     },
     {
       "epoch": 17.340425531914892,
-      "grad_norm": 1.359375,
+      "grad_norm": 1.5703125,
       "learning_rate": 7.140717314709908e-06,
       "loss": 0.1113,
       "step": 815
     },
     {
       "epoch": 17.4468085106383,
-      "grad_norm": 1.375,
+      "grad_norm": 1.4140625,
       "learning_rate": 6.975387020479173e-06,
-      "loss": 0.114,
+      "loss": 0.1119,
       "step": 820
     },
     {
@@ -1165,173 +1165,173 @@
     },
     {
       "epoch": 17.659574468085108,
-      "grad_norm": 1.390625,
+      "grad_norm": 1.3984375,
       "learning_rate": 6.663826800531703e-06,
-      "loss": 0.1134,
+      "loss": 0.1115,
       "step": 830
     },
     {
       "epoch": 17.76595744680851,
-      "grad_norm": 1.359375,
+      "grad_norm": 1.3359375,
       "learning_rate": 6.517693272729839e-06,
-      "loss": 0.1124,
+      "loss": 0.1126,
       "step": 835
     },
     {
       "epoch": 17.872340425531917,
-      "grad_norm": 1.390625,
+      "grad_norm": 1.3828125,
       "learning_rate": 6.3780517502874835e-06,
-      "loss": 0.1106,
+      "loss": 0.1157,
       "step": 840
     },
     {
       "epoch": 17.97872340425532,
-      "grad_norm": 1.484375,
+      "grad_norm": 1.5,
       "learning_rate": 6.24494543882183e-06,
       "loss": 0.1128,
       "step": 845
     },
     {
       "epoch": 18.085106382978722,
-      "grad_norm": 1.3046875,
+      "grad_norm": 1.3203125,
       "learning_rate": 6.118415521930966e-06,
-      "loss": 0.1098,
+      "loss": 0.11,
       "step": 850
     },
     {
       "epoch": 18.19148936170213,
-      "grad_norm": 1.25,
+      "grad_norm": 1.28125,
       "learning_rate": 5.998501148451544e-06,
-      "loss": 0.1072,
+      "loss": 0.1065,
       "step": 855
     },
     {
       "epoch": 18.29787234042553,
-      "grad_norm": 1.390625,
+      "grad_norm": 1.3125,
       "learning_rate": 5.885239420345969e-06,
-      "loss": 0.1085,
+      "loss": 0.106,
       "step": 860
     },
     {
       "epoch": 18.404255319148938,
       "grad_norm": 1.3359375,
       "learning_rate": 5.778665381222906e-06,
-      "loss": 0.1115,
+      "loss": 0.1106,
       "step": 865
     },
     {
       "epoch": 18.51063829787234,
-      "grad_norm": 1.390625,
+      "grad_norm": 1.40625,
       "learning_rate": 5.678812005494661e-06,
-      "loss": 0.1101,
+      "loss": 0.1109,
       "step": 870
     },
     {
       "epoch": 18.617021276595743,
-      "grad_norm": 1.3671875,
+      "grad_norm": 1.40625,
       "learning_rate": 5.585710188174777e-06,
-      "loss": 0.111,
+      "loss": 0.1093,
       "step": 875
     },
     {
       "epoch": 18.72340425531915,
-      "grad_norm": 1.328125,
+      "grad_norm": 1.3203125,
       "learning_rate": 5.499388735318982e-06,
-      "loss": 0.1077,
+      "loss": 0.1081,
       "step": 880
     },
     {
       "epoch": 18.829787234042552,
-      "grad_norm": 1.359375,
+      "grad_norm": 1.46875,
       "learning_rate": 5.419874355112507e-06,
-      "loss": 0.1107,
+      "loss": 0.1123,
       "step": 885
     },
     {
       "epoch": 18.93617021276596,
-      "grad_norm": 1.4453125,
+      "grad_norm": 1.6796875,
       "learning_rate": 5.34719164960646e-06,
       "loss": 0.1107,
       "step": 890
     },
     {
       "epoch": 19.04255319148936,
-      "grad_norm": 1.34375,
+      "grad_norm": 1.359375,
       "learning_rate": 5.281363107105859e-06,
-      "loss": 0.1113,
+      "loss": 0.1104,
       "step": 895
     },
     {
       "epoch": 19.148936170212767,
-      "grad_norm": 1.3359375,
+      "grad_norm": 1.3515625,
       "learning_rate": 5.222409095211667e-06,
-      "loss": 0.1069,
+      "loss": 0.1087,
       "step": 900
     },
     {
       "epoch": 19.25531914893617,
-      "grad_norm": 1.3046875,
+      "grad_norm": 1.359375,
       "learning_rate": 5.170347854518983e-06,
-      "loss": 0.1071,
+      "loss": 0.1057,
       "step": 905
     },
     {
       "epoch": 19.361702127659573,
-      "grad_norm": 1.3984375,
+      "grad_norm": 1.3359375,
       "learning_rate": 5.1251954929733194e-06,
-      "loss": 0.1095,
+      "loss": 0.1097,
       "step": 910
     },
     {
       "epoch": 19.46808510638298,
-      "grad_norm": 1.3203125,
+      "grad_norm": 1.28125,
       "learning_rate": 5.086965980886762e-06,
-      "loss": 0.1074,
+      "loss": 0.1084,
       "step": 915
     },
     {
       "epoch": 19.574468085106382,
-      "grad_norm": 1.3515625,
+      "grad_norm": 1.3046875,
       "learning_rate": 5.0556711466154845e-06,
-      "loss": 0.1077,
+      "loss": 0.1062,
       "step": 920
     },
     {
       "epoch": 19.680851063829788,
-      "grad_norm": 1.3515625,
+      "grad_norm": 1.4453125,
       "learning_rate": 5.0313206729000174e-06,
-      "loss": 0.1082,
+      "loss": 0.1055,
       "step": 925
     },
     {
       "epoch": 19.78723404255319,
-      "grad_norm": 1.328125,
+      "grad_norm": 1.4140625,
       "learning_rate": 5.013922093869376e-06,
-      "loss": 0.1085,
+      "loss": 0.107,
       "step": 930
     },
     {
       "epoch": 19.893617021276597,
-      "grad_norm": 1.4375,
+      "grad_norm": 1.4296875,
       "learning_rate": 5.003480792709965e-06,
-      "loss": 0.109,
+      "loss": 0.1074,
       "step": 935
     },
     {
       "epoch": 20.0,
-      "grad_norm": 1.5546875,
+      "grad_norm": 1.46875,
       "learning_rate": 5e-06,
-      "loss": 0.1072,
+      "loss": 0.1086,
       "step": 940
     },
     {
       "epoch": 20.0,
       "step": 940,
-      "total_flos": 1.3158373184765952e+17,
-      "train_loss": 0.0,
-      "train_runtime": 2.524,
-      "train_samples_per_second": 5950.808,
-      "train_steps_per_second": 372.421
+      "total_flos": 1.3211147408769024e+17,
+      "train_loss": 0.4334589367217206,
+      "train_runtime": 1817.3607,
+      "train_samples_per_second": 8.265,
+      "train_steps_per_second": 0.517
     }
   ],
   "logging_steps": 5,
@@ -1351,7 +1351,7 @@
       "attributes": {}
     }
   },
-  "total_flos": 1.3158373184765952e+17,
+  "total_flos": 1.3211147408769024e+17,
   "train_batch_size": 16,
   "trial_name": null,
   "trial_params": null