Upload with huggingface_hub

added_tokens.json

@@ -1,3 +1,66 @@
 {
-  "<pad>": 32000
+  "<pad>": 32000,
+  "{'pad_token_0': '<pad0>'}": 32001,
+  "{'pad_token_1': '<pad1>'}": 32002,
+  "{'pad_token_10': '<pad10>'}": 32011,
+  "{'pad_token_11': '<pad11>'}": 32012,
+  "{'pad_token_12': '<pad12>'}": 32013,
+  "{'pad_token_13': '<pad13>'}": 32014,
+  "{'pad_token_14': '<pad14>'}": 32015,
+  "{'pad_token_15': '<pad15>'}": 32016,
+  "{'pad_token_16': '<pad16>'}": 32017,
+  "{'pad_token_17': '<pad17>'}": 32018,
+  "{'pad_token_18': '<pad18>'}": 32019,
+  "{'pad_token_19': '<pad19>'}": 32020,
+  "{'pad_token_2': '<pad2>'}": 32003,
+  "{'pad_token_20': '<pad20>'}": 32021,
+  "{'pad_token_21': '<pad21>'}": 32022,
+  "{'pad_token_22': '<pad22>'}": 32023,
+  "{'pad_token_23': '<pad23>'}": 32024,
+  "{'pad_token_24': '<pad24>'}": 32025,
+  "{'pad_token_25': '<pad25>'}": 32026,
+  "{'pad_token_26': '<pad26>'}": 32027,
+  "{'pad_token_27': '<pad27>'}": 32028,
+  "{'pad_token_28': '<pad28>'}": 32029,
+  "{'pad_token_29': '<pad29>'}": 32030,
+  "{'pad_token_3': '<pad3>'}": 32004,
+  "{'pad_token_30': '<pad30>'}": 32031,
+  "{'pad_token_31': '<pad31>'}": 32032,
+  "{'pad_token_32': '<pad32>'}": 32033,
+  "{'pad_token_33': '<pad33>'}": 32034,
+  "{'pad_token_34': '<pad34>'}": 32035,
+  "{'pad_token_35': '<pad35>'}": 32036,
+  "{'pad_token_36': '<pad36>'}": 32037,
+  "{'pad_token_37': '<pad37>'}": 32038,
+  "{'pad_token_38': '<pad38>'}": 32039,
+  "{'pad_token_39': '<pad39>'}": 32040,
+  "{'pad_token_4': '<pad4>'}": 32005,
+  "{'pad_token_40': '<pad40>'}": 32041,
+  "{'pad_token_41': '<pad41>'}": 32042,
+  "{'pad_token_42': '<pad42>'}": 32043,
+  "{'pad_token_43': '<pad43>'}": 32044,
+  "{'pad_token_44': '<pad44>'}": 32045,
+  "{'pad_token_45': '<pad45>'}": 32046,
+  "{'pad_token_46': '<pad46>'}": 32047,
+  "{'pad_token_47': '<pad47>'}": 32048,
+  "{'pad_token_48': '<pad48>'}": 32049,
+  "{'pad_token_49': '<pad49>'}": 32050,
+  "{'pad_token_5': '<pad5>'}": 32006,
+  "{'pad_token_50': '<pad50>'}": 32051,
+  "{'pad_token_51': '<pad51>'}": 32052,
+  "{'pad_token_52': '<pad52>'}": 32053,
+  "{'pad_token_53': '<pad53>'}": 32054,
+  "{'pad_token_54': '<pad54>'}": 32055,
+  "{'pad_token_55': '<pad55>'}": 32056,
+  "{'pad_token_56': '<pad56>'}": 32057,
+  "{'pad_token_57': '<pad57>'}": 32058,
+  "{'pad_token_58': '<pad58>'}": 32059,
+  "{'pad_token_59': '<pad59>'}": 32060,
+  "{'pad_token_6': '<pad6>'}": 32007,
+  "{'pad_token_60': '<pad60>'}": 32061,
+  "{'pad_token_61': '<pad61>'}": 32062,
+  "{'pad_token_62': '<pad62>'}": 32063,
+  "{'pad_token_7': '<pad7>'}": 32008,
+  "{'pad_token_8': '<pad8>'}": 32009,
+  "{'pad_token_9': '<pad9>'}": 32010
 }

config.json

@@ -18,5 +18,5 @@
   "torch_dtype": "float32",
   "transformers_version": "4.28.0.dev0",
   "use_cache": true,
-  "vocab_size": 32001
+  "vocab_size": 32064
 }

optimizer.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ec9ecf988d8d4a60332ec5360c3da5ab811aeac7ac05a28a5baf1d1b1df09620
+size 13016554941

pytorch_model-00001-of-00006.bin

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:c79e83b4b6e934facb51f76de7425701fd53520793fc3a62ee6f6e986d1dfd38
-size 9956564363
+oid sha256:e1c27d16b04485852f0236368e69a1247be3004657e6d4c94252a82a23b479ed
+size 9957853579

pytorch_model-00002-of-00006.bin

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:54376ff86e52e85da8064236d2420e489ddd80ee5852c3c671d7211e0a3d9629
-size 9940856385
+oid sha256:939a1b7ce53ec8d7b0d0caa7baeeb6bf65acf47c4c7fab9002bdca32ccb1ab62
+size 9940855489

pytorch_model-00003-of-00006.bin

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:a4a3294d4ae14e52260acbd6fba427c21e4729fd6b4948f0748b535924d51209
-size 9940856943
+oid sha256:f1fb78394a052d6150e75760eac998957465dd95945624a46529de702b6ebcf5
+size 9940855919

pytorch_model-00004-of-00006.bin

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:cf196ac8e5dfa0e9b478f71b54bfbd050931242a20f37bcf6f550fc0e5c9ff0b
-size 9867415289
+oid sha256:6685e10066982987100c4253d630868ed04a4911cdced5914ebb7300c0faedf5
+size 9867414329

pytorch_model-00005-of-00006.bin

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:0b7113f68e1a8cd3b1ce01926f3e6b219300ac3ed226c25eb7ab4069af8131c1
-size 9867456961
+oid sha256:a5a01ac2d21dbafdc859cd45788dcf63f1c42847d916b6939f9964af1c597b38
+size 9867455937

pytorch_model-00006-of-00006.bin

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:a7860cdc7e83a7380f546a53c6382eca2d41291b0f1ace9bd392c2d5b39d2d9c
-size 2490496687
+oid sha256:a4e449ebab139dc5b3a447072fad6d10887d36e9962f856b6be2a314d5a7e4a4
+size 2491786863

pytorch_model.bin.index.json

@@ -1,6 +1,6 @@
 {
   "metadata": {
-    "total_size": 52063508480
+    "total_size": 52066083840
   },
   "weight_map": {
     "lm_head.weight": "pytorch_model-00006-of-00006.bin",

rng_state_0.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:3a791362b7c290af6a599b520026a6a983b08a42f3aac3106f8425f52a00a146
+size 14583

rng_state_1.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d8c985887f5fce914e7c179ceb8237cc83fc2d1a1dc23fd23d1b7612c9749ad0
+size 14583

rng_state_2.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:168536fb391ac24abcc5e2ae9ef056f2eec661440e3ab6a6227b3978b2ac537d
+size 14583

rng_state_3.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7430ee71f201154f1aeb42a2b9f2c07987b46b39ebbb4db65c5b58d70a619610
+size 14583

rng_state_4.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ec8ef8fcccffe36623208f05195bcd2a1739acd1ec1e673c1af2a7cf9af2a778
+size 14583

rng_state_5.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:49989e535b8fe57f1d0481e19c5514686f062472da148f61fb73cae25244812b
+size 14583

rng_state_6.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c1d9287e7a213727fc3131dbdd120289a010181f1f1f823a0bb78f6f39222bce
+size 14583

rng_state_7.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:4514c6199ebef7957cac7461e5366c5c5458c4330e446bfe8894ee7c8686d3af
+size 14583

scaler.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:513d5f5ddcc46a2601fe0530fcd3472e57eb4200c1291537c5007d601cfda25e
+size 557

scheduler.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ab312fa421bc1fba2e5bf93e48a06c324c864277ef9d03feb770fe34a89549d4
+size 627

tokenizer_config.json

@@ -1,9 +1,10 @@
 {
-  "bos_token": "<s>",
-  "eos_token": "</s>",
+  "bos_token": "",
+  "eos_token": "",
   "model_max_length": 512,
   "padding_side": "right",
-  "special_tokens_map_file": "/home/paperspace/.cache/huggingface/hub/models--swype--deepshard-13B-raw/snapshots/956dd7d2c805d149e9c3d4bef1c16c2df6802408/special_tokens_map.json",
+  "special_tokens_map_file": "/root/.cache/huggingface/hub/models--swype--deepshard-13B-raw/snapshots/598d3a1c7219145b3e16aef7987cc57e75fc0e08/special_tokens_map.json",
   "tokenizer_class": "LlamaTokenizer",
-  "unk_token": "<unk>"
+  "unk_token": "",
+  "use_fast": false
 }

trainer_state.json

@@ -2,42 +2,2936 @@
   "best_metric": null,
   "best_model_checkpoint": null,
   "epoch": 3.0,
-  "global_step": 3,
+  "global_step": 4800,
   "is_hyper_param_search": false,
   "is_local_process_zero": true,
   "is_world_process_zero": true,
   "log_history": [
+    {
+      "epoch": 0.01,
+      "learning_rate": 6.944444444444446e-07,
+      "loss": 1.4352,
+      "step": 10
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 2.0833333333333334e-06,
+      "loss": 1.4072,
+      "step": 20
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 3.4722222222222224e-06,
+      "loss": 1.0503,
+      "step": 30
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 4.861111111111111e-06,
+      "loss": 0.8994,
+      "step": 40
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 6.25e-06,
+      "loss": 0.7676,
+      "step": 50
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 7.638888888888888e-06,
+      "loss": 0.7975,
+      "step": 60
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 9.027777777777779e-06,
+      "loss": 0.7504,
+      "step": 70
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 1.0416666666666668e-05,
+      "loss": 0.7608,
+      "step": 80
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 1.1805555555555557e-05,
+      "loss": 0.7681,
+      "step": 90
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 1.3194444444444446e-05,
+      "loss": 0.7687,
+      "step": 100
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 1.4583333333333333e-05,
+      "loss": 0.7127,
+      "step": 110
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 1.5972222222222224e-05,
+      "loss": 0.7564,
+      "step": 120
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 1.7361111111111114e-05,
+      "loss": 0.7366,
+      "step": 130
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 1.8750000000000002e-05,
+      "loss": 0.7586,
+      "step": 140
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 1.9999997723625092e-05,
+      "loss": 0.7792,
+      "step": 150
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 1.999972455989006e-05,
+      "loss": 0.7505,
+      "step": 160
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 1.9998996135423355e-05,
+      "loss": 0.7458,
+      "step": 170
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 1.9997812483388194e-05,
+      "loss": 0.7705,
+      "step": 180
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 1.9996173657673094e-05,
+      "loss": 0.7879,
+      "step": 190
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 1.9994079732889404e-05,
+      "loss": 0.7624,
+      "step": 200
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 1.999153080436793e-05,
+      "loss": 0.7882,
+      "step": 210
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 1.998852698815457e-05,
+      "loss": 0.744,
+      "step": 220
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 1.998506842100505e-05,
+      "loss": 0.7527,
+      "step": 230
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 1.9981155260378685e-05,
+      "loss": 0.7637,
+      "step": 240
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 1.9976787684431223e-05,
+      "loss": 0.7317,
+      "step": 250
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 1.997196589200672e-05,
+      "loss": 0.7602,
+      "step": 260
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 1.9966690102628496e-05,
+      "loss": 0.7491,
+      "step": 270
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 1.9960960556489145e-05,
+      "loss": 0.7414,
+      "step": 280
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 1.995477751443959e-05,
+      "loss": 0.7513,
+      "step": 290
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 1.9948141257977198e-05,
+      "loss": 0.7192,
+      "step": 300
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 1.9941052089233e-05,
+      "loss": 0.7267,
+      "step": 310
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 1.9933510330957896e-05,
+      "loss": 0.7066,
+      "step": 320
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 1.9925516326507984e-05,
+      "loss": 0.6888,
+      "step": 330
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 1.9917070439828916e-05,
+      "loss": 0.7601,
+      "step": 340
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 1.9908173055439343e-05,
+      "loss": 0.7598,
+      "step": 350
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 1.989882457841339e-05,
+      "loss": 0.7245,
+      "step": 360
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 1.9889025434362236e-05,
+      "loss": 0.6829,
+      "step": 370
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 1.9878776069414714e-05,
+      "loss": 0.7353,
+      "step": 380
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 1.986807695019701e-05,
+      "loss": 0.7357,
+      "step": 390
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 1.9856928563811438e-05,
+      "loss": 0.7185,
+      "step": 400
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 1.9845331417814223e-05,
+      "loss": 0.7281,
+      "step": 410
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 1.983328604019243e-05,
+      "loss": 0.6729,
+      "step": 420
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 1.9820792979339897e-05,
+      "loss": 0.7019,
+      "step": 430
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 1.9807852804032306e-05,
+      "loss": 0.6777,
+      "step": 440
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 1.9794466103401244e-05,
+      "loss": 0.661,
+      "step": 450
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 1.978063348690741e-05,
+      "loss": 0.679,
+      "step": 460
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 1.9766355584312866e-05,
+      "loss": 0.7216,
+      "step": 470
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 1.975163304565235e-05,
+      "loss": 0.6873,
+      "step": 480
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 1.97364665412037e-05,
+      "loss": 0.7134,
+      "step": 490
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 1.9720856761457326e-05,
+      "loss": 0.6959,
+      "step": 500
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 1.9704804417084772e-05,
+      "loss": 0.7185,
+      "step": 510
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 1.9688310238906378e-05,
+      "loss": 0.7001,
+      "step": 520
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 1.9671374977857987e-05,
+      "loss": 0.6652,
+      "step": 530
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 1.9653999404956775e-05,
+      "loss": 0.6838,
+      "step": 540
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 1.9636184311266126e-05,
+      "loss": 0.7071,
+      "step": 550
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 1.9617930507859643e-05,
+      "loss": 0.6807,
+      "step": 560
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 1.9599238825784212e-05,
+      "loss": 0.6743,
+      "step": 570
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 1.9580110116022145e-05,
+      "loss": 0.6475,
+      "step": 580
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 1.9560545249452477e-05,
+      "loss": 0.7439,
+      "step": 590
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 1.9540545116811295e-05,
+      "loss": 0.6432,
+      "step": 600
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 1.9520110628651174e-05,
+      "loss": 0.6582,
+      "step": 610
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 1.9499242715299743e-05,
+      "loss": 0.6583,
+      "step": 620
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 1.9477942326817322e-05,
+      "loss": 0.6494,
+      "step": 630
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 1.9456210432953663e-05,
+      "loss": 0.6734,
+      "step": 640
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 1.943404802310381e-05,
+      "loss": 0.6404,
+      "step": 650
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 1.9411456106263053e-05,
+      "loss": 0.7011,
+      "step": 660
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 1.9388435710980975e-05,
+      "loss": 0.6624,
+      "step": 670
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 1.9364987885314645e-05,
+      "loss": 0.6611,
+      "step": 680
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 1.9341113696780892e-05,
+      "loss": 0.6348,
+      "step": 690
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 1.931681423230771e-05,
+      "loss": 0.6237,
+      "step": 700
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 1.9292090598184768e-05,
+      "loss": 0.6512,
+      "step": 710
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 1.926694392001305e-05,
+      "loss": 0.6436,
+      "step": 720
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 1.924137534265359e-05,
+      "loss": 0.6544,
+      "step": 730
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 1.9215386030175383e-05,
+      "loss": 0.6266,
+      "step": 740
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 1.9188977165802362e-05,
+      "loss": 0.6249,
+      "step": 750
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 1.9162149951859527e-05,
+      "loss": 0.6258,
+      "step": 760
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 1.9134905609718235e-05,
+      "loss": 0.6292,
+      "step": 770
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 1.910724537974056e-05,
+      "loss": 0.6194,
+      "step": 780
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 1.9079170521222835e-05,
+      "loss": 0.6591,
+      "step": 790
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 1.905068231233834e-05,
+      "loss": 0.6126,
+      "step": 800
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 1.9021782050079074e-05,
+      "loss": 0.6783,
+      "step": 810
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 1.899247105019674e-05,
+      "loss": 0.6676,
+      "step": 820
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 1.8962750647142808e-05,
+      "loss": 0.6596,
+      "step": 830
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 1.89326221940078e-05,
+      "loss": 0.6616,
+      "step": 840
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 1.8902087062459658e-05,
+      "loss": 0.6219,
+      "step": 850
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 1.8871146642681304e-05,
+      "loss": 0.6721,
+      "step": 860
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 1.8839802343307354e-05,
+      "loss": 0.6224,
+      "step": 870
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 1.880805559135998e-05,
+      "loss": 0.6115,
+      "step": 880
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 1.8775907832183945e-05,
+      "loss": 0.6151,
+      "step": 890
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 1.8743360529380794e-05,
+      "loss": 0.6023,
+      "step": 900
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 1.871041516474224e-05,
+      "loss": 0.5917,
+      "step": 910
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 1.8677073238182667e-05,
+      "loss": 0.6032,
+      "step": 920
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 1.8643336267670885e-05,
+      "loss": 0.5802,
+      "step": 930
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 1.860920578916098e-05,
+      "loss": 0.5687,
+      "step": 940
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 1.8574683356522416e-05,
+      "loss": 0.5965,
+      "step": 950
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 1.853977054146928e-05,
+      "loss": 0.6706,
+      "step": 960
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 1.850446893348872e-05,
+      "loss": 0.5983,
+      "step": 970
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 1.8468780139768602e-05,
+      "loss": 0.5642,
+      "step": 980
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 1.8432705785124306e-05,
+      "loss": 0.5775,
+      "step": 990
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 1.8396247511924782e-05,
+      "loss": 0.6504,
+      "step": 1000
+    },
+    {
+      "epoch": 0.62,
+      "eval_loss": NaN,
+      "eval_runtime": 16.558,
+      "eval_samples_per_second": 61.843,
+      "eval_steps_per_second": 3.865,
+      "step": 1000
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 1.8359406980017763e-05,
+      "loss": 0.5675,
+      "step": 1010
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 1.8322185866654192e-05,
+      "loss": 0.5895,
+      "step": 1020
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 1.8284585866411882e-05,
+      "loss": 0.5816,
+      "step": 1030
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 1.8246608691118343e-05,
+      "loss": 0.5847,
+      "step": 1040
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 1.820825606977286e-05,
+      "loss": 0.6139,
+      "step": 1050
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 1.816952974846777e-05,
+      "loss": 0.6137,
+      "step": 1060
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 1.813043149030898e-05,
+      "loss": 0.5607,
+      "step": 1070
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 1.809096307533567e-05,
+      "loss": 0.6062,
+      "step": 1080
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 1.8051126300439295e-05,
+      "loss": 0.5783,
+      "step": 1090
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 1.801092297928173e-05,
+      "loss": 0.5812,
+      "step": 1100
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 1.7970354942212736e-05,
+      "loss": 0.5843,
+      "step": 1110
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 1.7929424036186603e-05,
+      "loss": 0.5849,
+      "step": 1120
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 1.788813212467809e-05,
+      "loss": 0.6044,
+      "step": 1130
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 1.7846481087597553e-05,
+      "loss": 0.5326,
+      "step": 1140
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 1.7804472821205387e-05,
+      "loss": 0.5767,
+      "step": 1150
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 1.7762109238025682e-05,
+      "loss": 0.5525,
+      "step": 1160
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 1.771939226675915e-05,
+      "loss": 0.5936,
+      "step": 1170
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 1.7676323852195313e-05,
+      "loss": 0.5523,
+      "step": 1180
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 1.763290595512398e-05,
+      "loss": 0.5267,
+      "step": 1190
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 1.7589140552245946e-05,
+      "loss": 0.5738,
+      "step": 1200
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 1.7545029636083036e-05,
+      "loss": 0.5236,
+      "step": 1210
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 1.7500575214887354e-05,
+      "loss": 0.5457,
+      "step": 1220
+    },
+    {
+      "epoch": 0.77,
+      "learning_rate": 1.7455779312549885e-05,
+      "loss": 0.5873,
+      "step": 1230
+    },
+    {
+      "epoch": 0.78,
+      "learning_rate": 1.7410643968508324e-05,
+      "loss": 0.5426,
+      "step": 1240
+    },
+    {
+      "epoch": 0.78,
+      "learning_rate": 1.736517123765425e-05,
+      "loss": 0.5286,
+      "step": 1250
+    },
+    {
+      "epoch": 0.79,
+      "learning_rate": 1.731936319023956e-05,
+      "loss": 0.5222,
+      "step": 1260
+    },
+    {
+      "epoch": 0.79,
+      "learning_rate": 1.727322191178221e-05,
+      "loss": 0.5428,
+      "step": 1270
+    },
+    {
+      "epoch": 0.8,
+      "learning_rate": 1.7226749502971288e-05,
+      "loss": 0.528,
+      "step": 1280
+    },
+    {
+      "epoch": 0.81,
+      "learning_rate": 1.717994807957135e-05,
+      "loss": 0.5526,
+      "step": 1290
+    },
+    {
+      "epoch": 0.81,
+      "learning_rate": 1.7132819772326107e-05,
+      "loss": 0.5461,
+      "step": 1300
+    },
+    {
+      "epoch": 0.82,
+      "learning_rate": 1.708536672686143e-05,
+      "loss": 0.5559,
+      "step": 1310
+    },
+    {
+      "epoch": 0.82,
+      "learning_rate": 1.7037591103587643e-05,
+      "loss": 0.5404,
+      "step": 1320
+    },
+    {
+      "epoch": 0.83,
+      "learning_rate": 1.6989495077601174e-05,
+      "loss": 0.5629,
+      "step": 1330
+    },
+    {
+      "epoch": 0.84,
+      "learning_rate": 1.6941080838585537e-05,
+      "loss": 0.529,
+      "step": 1340
+    },
+    {
+      "epoch": 0.84,
+      "learning_rate": 1.6892350590711634e-05,
+      "loss": 0.5623,
+      "step": 1350
+    },
+    {
+      "epoch": 0.85,
+      "learning_rate": 1.6843306552537394e-05,
+      "loss": 0.5111,
+      "step": 1360
+    },
+    {
+      "epoch": 0.86,
+      "learning_rate": 1.67939509569068e-05,
+      "loss": 0.5376,
+      "step": 1370
+    },
+    {
+      "epoch": 0.86,
+      "learning_rate": 1.6744286050848194e-05,
+      "loss": 0.5547,
+      "step": 1380
+    },
+    {
+      "epoch": 0.87,
+      "learning_rate": 1.6694314095472006e-05,
+      "loss": 0.5552,
+      "step": 1390
+    },
+    {
+      "epoch": 0.88,
+      "learning_rate": 1.66440373658678e-05,
+      "loss": 0.5061,
+      "step": 1400
+    },
+    {
+      "epoch": 0.88,
+      "learning_rate": 1.659345815100069e-05,
+      "loss": 0.5242,
+      "step": 1410
+    },
+    {
+      "epoch": 0.89,
+      "learning_rate": 1.654257875360714e-05,
+      "loss": 0.5066,
+      "step": 1420
+    },
+    {
+      "epoch": 0.89,
+      "learning_rate": 1.6491401490090125e-05,
+      "loss": 0.5358,
+      "step": 1430
+    },
+    {
+      "epoch": 0.9,
+      "learning_rate": 1.6439928690413663e-05,
+      "loss": 0.5006,
+      "step": 1440
+    },
+    {
+      "epoch": 0.91,
+      "learning_rate": 1.638816269799674e-05,
+      "loss": 0.5012,
+      "step": 1450
+    },
+    {
+      "epoch": 0.91,
+      "learning_rate": 1.633610586960664e-05,
+      "loss": 0.5365,
+      "step": 1460
+    },
+    {
+      "epoch": 0.92,
+      "learning_rate": 1.6283760575251613e-05,
+      "loss": 0.513,
+      "step": 1470
+    },
+    {
+      "epoch": 0.93,
+      "learning_rate": 1.6231129198073e-05,
+      "loss": 0.5552,
+      "step": 1480
+    },
+    {
+      "epoch": 0.93,
+      "learning_rate": 1.6178214134236733e-05,
+      "loss": 0.5281,
+      "step": 1490
+    },
+    {
+      "epoch": 0.94,
+      "learning_rate": 1.6125017792824237e-05,
+      "loss": 0.5118,
+      "step": 1500
+    },
+    {
+      "epoch": 0.94,
+      "learning_rate": 1.607154259572275e-05,
+      "loss": 0.4905,
+      "step": 1510
+    },
+    {
+      "epoch": 0.95,
+      "learning_rate": 1.6017790977515063e-05,
+      "loss": 0.5183,
+      "step": 1520
+    },
+    {
+      "epoch": 0.96,
+      "learning_rate": 1.5963765385368687e-05,
+      "loss": 0.5085,
+      "step": 1530
+    },
+    {
+      "epoch": 0.96,
+      "learning_rate": 1.5909468278924433e-05,
+      "loss": 0.5221,
+      "step": 1540
+    },
+    {
+      "epoch": 0.97,
+      "learning_rate": 1.5854902130184426e-05,
+      "loss": 0.5477,
+      "step": 1550
+    },
+    {
+      "epoch": 0.97,
+      "learning_rate": 1.5800069423399576e-05,
+      "loss": 0.5533,
+      "step": 1560
+    },
+    {
+      "epoch": 0.98,
+      "learning_rate": 1.5744972654956466e-05,
+      "loss": 0.5319,
+      "step": 1570
+    },
+    {
+      "epoch": 0.99,
+      "learning_rate": 1.568961433326369e-05,
+      "loss": 0.5048,
+      "step": 1580
+    },
+    {
+      "epoch": 0.99,
+      "learning_rate": 1.5633996978637685e-05,
+      "loss": 0.4891,
+      "step": 1590
+    },
     {
       "epoch": 1.0,
-      "learning_rate": 2e-05,
-      "loss": 1.0501,
-      "step": 1
+      "learning_rate": 1.5578123123187944e-05,
+      "loss": 0.531,
+      "step": 1600
+    },
+    {
+      "epoch": 1.01,
+      "learning_rate": 1.5521995310701762e-05,
+      "loss": 0.4531,
+      "step": 1610
+    },
+    {
+      "epoch": 1.01,
+      "learning_rate": 1.5465616096528427e-05,
+      "loss": 0.4815,
+      "step": 1620
+    },
+    {
+      "epoch": 1.02,
+      "learning_rate": 1.5408988047462866e-05,
+      "loss": 0.5315,
+      "step": 1630
+    },
+    {
+      "epoch": 1.02,
+      "learning_rate": 1.5352113741628795e-05,
+      "loss": 0.4574,
+      "step": 1640
+    },
+    {
+      "epoch": 1.03,
+      "learning_rate": 1.5294995768361342e-05,
+      "loss": 0.477,
+      "step": 1650
+    },
+    {
+      "epoch": 1.04,
+      "learning_rate": 1.5237636728089169e-05,
+      "loss": 0.4898,
+      "step": 1660
+    },
+    {
+      "epoch": 1.04,
+      "learning_rate": 1.5180039232216062e-05,
+      "loss": 0.4988,
+      "step": 1670
+    },
+    {
+      "epoch": 1.05,
+      "learning_rate": 1.5122205903002068e-05,
+      "loss": 0.4846,
+      "step": 1680
+    },
+    {
+      "epoch": 1.06,
+      "learning_rate": 1.5064139373444077e-05,
+      "loss": 0.4404,
+      "step": 1690
+    },
+    {
+      "epoch": 1.06,
+      "learning_rate": 1.500584228715599e-05,
+      "loss": 0.4469,
+      "step": 1700
+    },
+    {
+      "epoch": 1.07,
+      "learning_rate": 1.494731729824832e-05,
+      "loss": 0.4866,
+      "step": 1710
+    },
+    {
+      "epoch": 1.07,
+      "learning_rate": 1.4888567071207383e-05,
+      "loss": 0.4554,
+      "step": 1720
+    },
+    {
+      "epoch": 1.08,
+      "learning_rate": 1.4829594280773993e-05,
+      "loss": 0.4532,
+      "step": 1730
+    },
+    {
+      "epoch": 1.09,
+      "learning_rate": 1.4770401611821672e-05,
+      "loss": 0.4777,
+      "step": 1740
+    },
+    {
+      "epoch": 1.09,
+      "learning_rate": 1.4710991759234425e-05,
+      "loss": 0.4504,
+      "step": 1750
+    },
+    {
+      "epoch": 1.1,
+      "learning_rate": 1.4651367427784049e-05,
+      "loss": 0.4489,
+      "step": 1760
+    },
+    {
+      "epoch": 1.11,
+      "learning_rate": 1.4591531332006987e-05,
+      "loss": 0.496,
+      "step": 1770
+    },
+    {
+      "epoch": 1.11,
+      "learning_rate": 1.4531486196080753e-05,
+      "loss": 0.4507,
+      "step": 1780
+    },
+    {
+      "epoch": 1.12,
+      "learning_rate": 1.4471234753699887e-05,
+      "loss": 0.4795,
+      "step": 1790
+    },
+    {
+      "epoch": 1.12,
+      "learning_rate": 1.4410779747951526e-05,
+      "loss": 0.4479,
+      "step": 1800
+    },
+    {
+      "epoch": 1.13,
+      "learning_rate": 1.4350123931190498e-05,
+      "loss": 0.3965,
+      "step": 1810
+    },
+    {
+      "epoch": 1.14,
+      "learning_rate": 1.4289270064914012e-05,
+      "loss": 0.5061,
+      "step": 1820
+    },
+    {
+      "epoch": 1.14,
+      "learning_rate": 1.4228220919635946e-05,
+      "loss": 0.3941,
+      "step": 1830
+    },
+    {
+      "epoch": 1.15,
+      "learning_rate": 1.4166979274760718e-05,
+      "loss": 0.4335,
+      "step": 1840
+    },
+    {
+      "epoch": 1.16,
+      "learning_rate": 1.4105547918456726e-05,
+      "loss": 0.4662,
+      "step": 1850
+    },
+    {
+      "epoch": 1.16,
+      "learning_rate": 1.4043929647529424e-05,
+      "loss": 0.4777,
+      "step": 1860
+    },
+    {
+      "epoch": 1.17,
+      "learning_rate": 1.3982127267294e-05,
+      "loss": 0.4501,
+      "step": 1870
+    },
+    {
+      "epoch": 1.18,
+      "learning_rate": 1.3920143591447635e-05,
+      "loss": 0.4332,
+      "step": 1880
+    },
+    {
+      "epoch": 1.18,
+      "learning_rate": 1.3857981441941419e-05,
+      "loss": 0.4321,
+      "step": 1890
+    },
+    {
+      "epoch": 1.19,
+      "learning_rate": 1.3795643648851869e-05,
+      "loss": 0.4029,
+      "step": 1900
+    },
+    {
+      "epoch": 1.19,
+      "learning_rate": 1.3733133050252087e-05,
+      "loss": 0.4371,
+      "step": 1910
+    },
+    {
+      "epoch": 1.2,
+      "learning_rate": 1.3670452492082552e-05,
+      "loss": 0.4139,
+      "step": 1920
+    },
+    {
+      "epoch": 1.21,
+      "learning_rate": 1.3607604828021534e-05,
+      "loss": 0.4569,
+      "step": 1930
+    },
+    {
+      "epoch": 1.21,
+      "learning_rate": 1.3544592919355203e-05,
+      "loss": 0.4088,
+      "step": 1940
+    },
+    {
+      "epoch": 1.22,
+      "learning_rate": 1.348141963484734e-05,
+      "loss": 0.4386,
+      "step": 1950
+    },
+    {
+      "epoch": 1.23,
+      "learning_rate": 1.3418087850608735e-05,
+      "loss": 0.4337,
+      "step": 1960
+    },
+    {
+      "epoch": 1.23,
+      "learning_rate": 1.335460044996625e-05,
+      "loss": 0.4365,
+      "step": 1970
+    },
+    {
+      "epoch": 1.24,
+      "learning_rate": 1.329096032333155e-05,
+      "loss": 0.4581,
+      "step": 1980
+    },
+    {
+      "epoch": 1.24,
+      "learning_rate": 1.3227170368069496e-05,
+      "loss": 0.4316,
+      "step": 1990
+    },
+    {
+      "epoch": 1.25,
+      "learning_rate": 1.3163233488366254e-05,
+      "loss": 0.3938,
+      "step": 2000
+    },
+    {
+      "epoch": 1.25,
+      "eval_loss": NaN,
+      "eval_runtime": 17.5701,
+      "eval_samples_per_second": 58.281,
+      "eval_steps_per_second": 3.643,
+      "step": 2000
+    },
+    {
+      "epoch": 1.26,
+      "learning_rate": 1.309915259509706e-05,
+      "loss": 0.4294,
+      "step": 2010
+    },
+    {
+      "epoch": 1.26,
+      "learning_rate": 1.3034930605693716e-05,
+      "loss": 0.4351,
+      "step": 2020
+    },
+    {
+      "epoch": 1.27,
+      "learning_rate": 1.2970570444011739e-05,
+      "loss": 0.4263,
+      "step": 2030
+    },
+    {
+      "epoch": 1.27,
+      "learning_rate": 1.290607504019727e-05,
+      "loss": 0.412,
+      "step": 2040
+    },
+    {
+      "epoch": 1.28,
+      "learning_rate": 1.2841447330553658e-05,
+      "loss": 0.4085,
+      "step": 2050
+    },
+    {
+      "epoch": 1.29,
+      "learning_rate": 1.2776690257407782e-05,
+      "loss": 0.4281,
+      "step": 2060
+    },
+    {
+      "epoch": 1.29,
+      "learning_rate": 1.2711806768976102e-05,
+      "loss": 0.4265,
+      "step": 2070
+    },
+    {
+      "epoch": 1.3,
+      "learning_rate": 1.264679981923042e-05,
+      "loss": 0.4109,
+      "step": 2080
+    },
+    {
+      "epoch": 1.31,
+      "learning_rate": 1.2581672367763408e-05,
+      "loss": 0.4149,
+      "step": 2090
+    },
+    {
+      "epoch": 1.31,
+      "learning_rate": 1.251642737965386e-05,
+      "loss": 0.4164,
+      "step": 2100
+    },
+    {
+      "epoch": 1.32,
+      "learning_rate": 1.245106782533169e-05,
+      "loss": 0.4153,
+      "step": 2110
+    },
+    {
+      "epoch": 1.32,
+      "learning_rate": 1.2385596680442715e-05,
+      "loss": 0.4099,
+      "step": 2120
+    },
+    {
+      "epoch": 1.33,
+      "learning_rate": 1.2320016925713168e-05,
+      "loss": 0.4614,
+      "step": 2130
+    },
+    {
+      "epoch": 1.34,
+      "learning_rate": 1.2254331546813995e-05,
+      "loss": 0.421,
+      "step": 2140
+    },
+    {
+      "epoch": 1.34,
+      "learning_rate": 1.218854353422494e-05,
+      "loss": 0.4066,
+      "step": 2150
+    },
+    {
+      "epoch": 1.35,
+      "learning_rate": 1.2122655883098369e-05,
+      "loss": 0.3837,
+      "step": 2160
+    },
+    {
+      "epoch": 1.36,
+      "learning_rate": 1.205667159312293e-05,
+      "loss": 0.3882,
+      "step": 2170
+    },
+    {
+      "epoch": 1.36,
+      "learning_rate": 1.199059366838699e-05,
+      "loss": 0.3884,
+      "step": 2180
+    },
+    {
+      "epoch": 1.37,
+      "learning_rate": 1.1924425117241837e-05,
+      "loss": 0.4101,
+      "step": 2190
+    },
+    {
+      "epoch": 1.38,
+      "learning_rate": 1.1858168952164757e-05,
+      "loss": 0.3678,
+      "step": 2200
+    },
+    {
+      "epoch": 1.38,
+      "learning_rate": 1.1791828189621848e-05,
+      "loss": 0.4123,
+      "step": 2210
+    },
+    {
+      "epoch": 1.39,
+      "learning_rate": 1.1725405849930713e-05,
+      "loss": 0.4034,
+      "step": 2220
+    },
+    {
+      "epoch": 1.39,
+      "learning_rate": 1.1658904957122941e-05,
+      "loss": 0.3858,
+      "step": 2230
+    },
+    {
+      "epoch": 1.4,
+      "learning_rate": 1.1592328538806439e-05,
+      "loss": 0.4078,
+      "step": 2240
+    },
+    {
+      "epoch": 1.41,
+      "learning_rate": 1.1525679626027585e-05,
+      "loss": 0.404,
+      "step": 2250
+    },
+    {
+      "epoch": 1.41,
+      "learning_rate": 1.1458961253133238e-05,
+      "loss": 0.3735,
+      "step": 2260
+    },
+    {
+      "epoch": 1.42,
+      "learning_rate": 1.1392176457632586e-05,
+      "loss": 0.3782,
+      "step": 2270
+    },
+    {
+      "epoch": 1.43,
+      "learning_rate": 1.1325328280058867e-05,
+      "loss": 0.3717,
+      "step": 2280
+    },
+    {
+      "epoch": 1.43,
+      "learning_rate": 1.125841976383093e-05,
+      "loss": 0.3843,
+      "step": 2290
+    },
+    {
+      "epoch": 1.44,
+      "learning_rate": 1.1191453955114681e-05,
+      "loss": 0.368,
+      "step": 2300
+    },
+    {
+      "epoch": 1.44,
+      "learning_rate": 1.1124433902684411e-05,
+      "loss": 0.3863,
+      "step": 2310
+    },
+    {
+      "epoch": 1.45,
+      "learning_rate": 1.1057362657783967e-05,
+      "loss": 0.3432,
+      "step": 2320
+    },
+    {
+      "epoch": 1.46,
+      "learning_rate": 1.0990243273987863e-05,
+      "loss": 0.3875,
+      "step": 2330
+    },
+    {
+      "epoch": 1.46,
+      "learning_rate": 1.0923078807062245e-05,
+      "loss": 0.3518,
+      "step": 2340
+    },
+    {
+      "epoch": 1.47,
+      "learning_rate": 1.0855872314825772e-05,
+      "loss": 0.3623,
+      "step": 2350
+    },
+    {
+      "epoch": 1.48,
+      "learning_rate": 1.0788626857010404e-05,
+      "loss": 0.3577,
+      "step": 2360
+    },
+    {
+      "epoch": 1.48,
+      "learning_rate": 1.0721345495122098e-05,
+      "loss": 0.3666,
+      "step": 2370
+    },
+    {
+      "epoch": 1.49,
+      "learning_rate": 1.0654031292301432e-05,
+      "loss": 0.3714,
+      "step": 2380
+    },
+    {
+      "epoch": 1.49,
+      "learning_rate": 1.0586687313184141e-05,
+      "loss": 0.3783,
+      "step": 2390
+    },
+    {
+      "epoch": 1.5,
+      "learning_rate": 1.0519316623761593e-05,
+      "loss": 0.3453,
+      "step": 2400
+    },
+    {
+      "epoch": 1.51,
+      "learning_rate": 1.045192229124121e-05,
+      "loss": 0.385,
+      "step": 2410
+    },
+    {
+      "epoch": 1.51,
+      "learning_rate": 1.0384507383906819e-05,
+      "loss": 0.3752,
+      "step": 2420
+    },
+    {
+      "epoch": 1.52,
+      "learning_rate": 1.0317074970978957e-05,
+      "loss": 0.3714,
+      "step": 2430
+    },
+    {
+      "epoch": 1.52,
+      "learning_rate": 1.0249628122475154e-05,
+      "loss": 0.3871,
+      "step": 2440
+    },
+    {
+      "epoch": 1.53,
+      "learning_rate": 1.0188916154305646e-05,
+      "loss": 0.3366,
+      "step": 2450
+    },
+    {
+      "epoch": 1.54,
+      "learning_rate": 1.012145033834699e-05,
+      "loss": 0.3724,
+      "step": 2460
+    },
+    {
+      "epoch": 1.54,
+      "learning_rate": 1.0053978993079046e-05,
+      "loss": 0.3878,
+      "step": 2470
+    },
+    {
+      "epoch": 1.55,
+      "learning_rate": 9.986505190291822e-06,
+      "loss": 0.351,
+      "step": 2480
+    },
+    {
+      "epoch": 1.56,
+      "learning_rate": 9.919032001887215e-06,
+      "loss": 0.3289,
+      "step": 2490
+    },
+    {
+      "epoch": 1.56,
+      "learning_rate": 9.85156249973915e-06,
+      "loss": 0.351,
+      "step": 2500
+    },
+    {
+      "epoch": 1.57,
+      "learning_rate": 9.784099755553723e-06,
+      "loss": 0.3428,
+      "step": 2510
+    },
+    {
+      "epoch": 1.57,
+      "learning_rate": 9.71664684072936e-06,
+      "loss": 0.3405,
+      "step": 2520
+    },
+    {
+      "epoch": 1.58,
+      "learning_rate": 9.649206826216988e-06,
+      "loss": 0.3786,
+      "step": 2530
+    },
+    {
+      "epoch": 1.59,
+      "learning_rate": 9.581782782380208e-06,
+      "loss": 0.3498,
+      "step": 2540
+    },
+    {
+      "epoch": 1.59,
+      "learning_rate": 9.514377778855521e-06,
+      "loss": 0.3247,
+      "step": 2550
+    },
+    {
+      "epoch": 1.6,
+      "learning_rate": 9.446994884412575e-06,
+      "loss": 0.3601,
+      "step": 2560
+    },
+    {
+      "epoch": 1.61,
+      "learning_rate": 9.37963716681446e-06,
+      "loss": 0.3569,
+      "step": 2570
+    },
+    {
+      "epoch": 1.61,
+      "learning_rate": 9.312307692678016e-06,
+      "loss": 0.3534,
+      "step": 2580
+    },
+    {
+      "epoch": 1.62,
+      "learning_rate": 9.245009527334243e-06,
+      "loss": 0.3435,
+      "step": 2590
+    },
+    {
+      "epoch": 1.62,
+      "learning_rate": 9.17774573468873e-06,
+      "loss": 0.3236,
+      "step": 2600
+    },
+    {
+      "epoch": 1.63,
+      "learning_rate": 9.110519377082174e-06,
+      "loss": 0.3543,
+      "step": 2610
+    },
+    {
+      "epoch": 1.64,
+      "learning_rate": 9.04333351515095e-06,
+      "loss": 0.3193,
+      "step": 2620
+    },
+    {
+      "epoch": 1.64,
+      "learning_rate": 8.976191207687775e-06,
+      "loss": 0.3259,
+      "step": 2630
+    },
+    {
+      "epoch": 1.65,
+      "learning_rate": 8.909095511502452e-06,
+      "loss": 0.3576,
+      "step": 2640
+    },
+    {
+      "epoch": 1.66,
+      "learning_rate": 8.842049481282691e-06,
+      "loss": 0.3422,
+      "step": 2650
+    },
+    {
+      "epoch": 1.66,
+      "learning_rate": 8.775056169455045e-06,
+      "loss": 0.3463,
+      "step": 2660
+    },
+    {
+      "epoch": 1.67,
+      "learning_rate": 8.708118626045939e-06,
+      "loss": 0.3219,
+      "step": 2670
+    },
+    {
+      "epoch": 1.68,
+      "learning_rate": 8.641239898542814e-06,
+      "loss": 0.3258,
+      "step": 2680
+    },
+    {
+      "epoch": 1.68,
+      "learning_rate": 8.574423031755377e-06,
+      "loss": 0.3325,
+      "step": 2690
+    },
+    {
+      "epoch": 1.69,
+      "learning_rate": 8.50767106767698e-06,
+      "loss": 0.3271,
+      "step": 2700
+    },
+    {
+      "epoch": 1.69,
+      "learning_rate": 8.440987045346135e-06,
+      "loss": 0.3171,
+      "step": 2710
+    },
+    {
+      "epoch": 1.7,
+      "learning_rate": 8.374374000708146e-06,
+      "loss": 0.3464,
+      "step": 2720
+    },
+    {
+      "epoch": 1.71,
+      "learning_rate": 8.307834966476885e-06,
+      "loss": 0.3445,
+      "step": 2730
+    },
+    {
+      "epoch": 1.71,
+      "learning_rate": 8.241372971996735e-06,
+      "loss": 0.3334,
+      "step": 2740
+    },
+    {
+      "epoch": 1.72,
+      "learning_rate": 8.174991043104662e-06,
+      "loss": 0.3,
+      "step": 2750
+    },
+    {
+      "epoch": 1.73,
+      "learning_rate": 8.108692201992466e-06,
+      "loss": 0.3452,
+      "step": 2760
+    },
+    {
+      "epoch": 1.73,
+      "learning_rate": 8.04247946706917e-06,
+      "loss": 0.3307,
+      "step": 2770
+    },
+    {
+      "epoch": 1.74,
+      "learning_rate": 7.976355852823628e-06,
+      "loss": 0.3035,
+      "step": 2780
+    },
+    {
+      "epoch": 1.74,
+      "learning_rate": 7.91032436968725e-06,
+      "loss": 0.3533,
+      "step": 2790
+    },
+    {
+      "epoch": 1.75,
+      "learning_rate": 7.844388023896981e-06,
+      "loss": 0.3318,
+      "step": 2800
+    },
+    {
+      "epoch": 1.76,
+      "learning_rate": 7.778549817358404e-06,
+      "loss": 0.3277,
+      "step": 2810
+    },
+    {
+      "epoch": 1.76,
+      "learning_rate": 7.712812747509091e-06,
+      "loss": 0.3024,
+      "step": 2820
+    },
+    {
+      "epoch": 1.77,
+      "learning_rate": 7.647179807182125e-06,
+      "loss": 0.3337,
+      "step": 2830
+    },
+    {
+      "epoch": 1.77,
+      "learning_rate": 7.5816539844698565e-06,
+      "loss": 0.3133,
+      "step": 2840
+    },
+    {
+      "epoch": 1.78,
+      "learning_rate": 7.516238262587851e-06,
+      "loss": 0.3381,
+      "step": 2850
+    },
+    {
+      "epoch": 1.79,
+      "learning_rate": 7.450935619739087e-06,
+      "loss": 0.3025,
+      "step": 2860
+    },
+    {
+      "epoch": 1.79,
+      "learning_rate": 7.385749028978347e-06,
+      "loss": 0.3152,
+      "step": 2870
+    },
+    {
+      "epoch": 1.8,
+      "learning_rate": 7.320681458076871e-06,
+      "loss": 0.3325,
+      "step": 2880
+    },
+    {
+      "epoch": 1.81,
+      "learning_rate": 7.255735869387257e-06,
+      "loss": 0.2968,
+      "step": 2890
+    },
+    {
+      "epoch": 1.81,
+      "learning_rate": 7.190915219708564e-06,
+      "loss": 0.3332,
+      "step": 2900
+    },
+    {
+      "epoch": 1.82,
+      "learning_rate": 7.126222460151719e-06,
+      "loss": 0.3049,
+      "step": 2910
+    },
+    {
+      "epoch": 1.82,
+      "learning_rate": 7.061660536005151e-06,
+      "loss": 0.3051,
+      "step": 2920
+    },
+    {
+      "epoch": 1.83,
+      "learning_rate": 6.997232386600706e-06,
+      "loss": 0.3294,
+      "step": 2930
+    },
+    {
+      "epoch": 1.84,
+      "learning_rate": 6.932940945179818e-06,
+      "loss": 0.3166,
+      "step": 2940
+    },
+    {
+      "epoch": 1.84,
+      "learning_rate": 6.868789138759977e-06,
+      "loss": 0.3045,
+      "step": 2950
+    },
+    {
+      "epoch": 1.85,
+      "learning_rate": 6.804779888001461e-06,
+      "loss": 0.2994,
+      "step": 2960
+    },
+    {
+      "epoch": 1.86,
+      "learning_rate": 6.740916107074372e-06,
+      "loss": 0.3216,
+      "step": 2970
+    },
+    {
+      "epoch": 1.86,
+      "learning_rate": 6.677200703525959e-06,
+      "loss": 0.2938,
+      "step": 2980
+    },
+    {
+      "epoch": 1.87,
+      "learning_rate": 6.613636578148242e-06,
+      "loss": 0.3407,
+      "step": 2990
+    },
+    {
+      "epoch": 1.88,
+      "learning_rate": 6.550226624845961e-06,
+      "loss": 0.343,
+      "step": 3000
+    },
+    {
+      "epoch": 1.88,
+      "eval_loss": NaN,
+      "eval_runtime": 16.6263,
+      "eval_samples_per_second": 61.589,
+      "eval_steps_per_second": 3.849,
+      "step": 3000
+    },
+    {
+      "epoch": 1.88,
+      "learning_rate": 6.4869737305047996e-06,
+      "loss": 0.3083,
+      "step": 3010
+    },
+    {
+      "epoch": 1.89,
+      "learning_rate": 6.423880774859978e-06,
+      "loss": 0.2817,
+      "step": 3020
+    },
+    {
+      "epoch": 1.89,
+      "learning_rate": 6.360950630365126e-06,
+      "loss": 0.2781,
+      "step": 3030
+    },
+    {
+      "epoch": 1.9,
+      "learning_rate": 6.298186162061521e-06,
+      "loss": 0.2762,
+      "step": 3040
+    },
+    {
+      "epoch": 1.91,
+      "learning_rate": 6.241842155612241e-06,
+      "loss": 0.2512,
+      "step": 3050
+    },
+    {
+      "epoch": 1.91,
+      "learning_rate": 6.179400338141371e-06,
+      "loss": 0.2593,
+      "step": 3060
+    },
+    {
+      "epoch": 1.92,
+      "learning_rate": 6.1171324623613016e-06,
+      "loss": 0.2904,
+      "step": 3070
+    },
+    {
+      "epoch": 1.93,
+      "learning_rate": 6.055041363161986e-06,
+      "loss": 0.2917,
+      "step": 3080
+    },
+    {
+      "epoch": 1.93,
+      "learning_rate": 5.99312986738521e-06,
+      "loss": 0.3004,
+      "step": 3090
+    },
+    {
+      "epoch": 1.94,
+      "learning_rate": 5.9314007936959006e-06,
+      "loss": 0.2892,
+      "step": 3100
+    },
+    {
+      "epoch": 1.94,
+      "learning_rate": 5.869856952453792e-06,
+      "loss": 0.2986,
+      "step": 3110
+    },
+    {
+      "epoch": 1.95,
+      "learning_rate": 5.8085011455854766e-06,
+      "loss": 0.2674,
+      "step": 3120
+    },
+    {
+      "epoch": 1.96,
+      "learning_rate": 5.747336166456849e-06,
+      "loss": 0.2898,
+      "step": 3130
+    },
+    {
+      "epoch": 1.96,
+      "learning_rate": 5.686364799745923e-06,
+      "loss": 0.2871,
+      "step": 3140
+    },
+    {
+      "epoch": 1.97,
+      "learning_rate": 5.625589821316065e-06,
+      "loss": 0.2682,
+      "step": 3150
+    },
+    {
+      "epoch": 1.98,
+      "learning_rate": 5.5650139980895985e-06,
+      "loss": 0.3133,
+      "step": 3160
+    },
+    {
+      "epoch": 1.98,
+      "learning_rate": 5.504640087921847e-06,
+      "loss": 0.2823,
+      "step": 3170
+    },
+    {
+      "epoch": 1.99,
+      "learning_rate": 5.444470839475571e-06,
+      "loss": 0.3103,
+      "step": 3180
+    },
+    {
+      "epoch": 1.99,
+      "learning_rate": 5.38450899209583e-06,
+      "loss": 0.2896,
+      "step": 3190
+    },
+    {
+      "epoch": 2.0,
+      "learning_rate": 5.324757275685269e-06,
+      "loss": 0.2672,
+      "step": 3200
+    },
+    {
+      "epoch": 2.01,
+      "learning_rate": 5.265218410579827e-06,
+      "loss": 0.2438,
+      "step": 3210
+    },
+    {
+      "epoch": 2.01,
+      "learning_rate": 5.2058951074248985e-06,
+      "loss": 0.2523,
+      "step": 3220
+    },
+    {
+      "epoch": 2.02,
+      "learning_rate": 5.146790067051912e-06,
+      "loss": 0.233,
+      "step": 3230
+    },
+    {
+      "epoch": 2.02,
+      "learning_rate": 5.087905980355381e-06,
+      "loss": 0.2697,
+      "step": 3240
+    },
+    {
+      "epoch": 2.03,
+      "learning_rate": 5.029245528170383e-06,
+      "loss": 0.2586,
+      "step": 3250
+    },
+    {
+      "epoch": 2.04,
+      "learning_rate": 4.97081138115052e-06,
+      "loss": 0.2368,
+      "step": 3260
+    },
+    {
+      "epoch": 2.04,
+      "learning_rate": 4.9126061996463195e-06,
+      "loss": 0.2319,
+      "step": 3270
+    },
+    {
+      "epoch": 2.05,
+      "learning_rate": 4.854632633584118e-06,
+      "loss": 0.2774,
+      "step": 3280
+    },
+    {
+      "epoch": 2.06,
+      "learning_rate": 4.796893322345435e-06,
+      "loss": 0.2738,
+      "step": 3290
+    },
+    {
+      "epoch": 2.06,
+      "learning_rate": 4.739390894646779e-06,
+      "loss": 0.242,
+      "step": 3300
+    },
+    {
+      "epoch": 2.07,
+      "learning_rate": 4.68212796841999e-06,
+      "loss": 0.2434,
+      "step": 3310
+    },
+    {
+      "epoch": 2.08,
+      "learning_rate": 4.625107150693042e-06,
+      "loss": 0.284,
+      "step": 3320
+    },
+    {
+      "epoch": 2.08,
+      "learning_rate": 4.568331037471364e-06,
+      "loss": 0.2368,
+      "step": 3330
+    },
+    {
+      "epoch": 2.09,
+      "learning_rate": 4.511802213619635e-06,
+      "loss": 0.2386,
+      "step": 3340
+    },
+    {
+      "epoch": 2.09,
+      "learning_rate": 4.455523252744114e-06,
+      "loss": 0.2971,
+      "step": 3350
+    },
+    {
+      "epoch": 2.1,
+      "learning_rate": 4.399496717075465e-06,
+      "loss": 0.2351,
+      "step": 3360
+    },
+    {
+      "epoch": 2.11,
+      "learning_rate": 4.34372515735211e-06,
+      "loss": 0.2279,
+      "step": 3370
+    },
+    {
+      "epoch": 2.11,
+      "learning_rate": 4.288211112704092e-06,
+      "loss": 0.2699,
+      "step": 3380
+    },
+    {
+      "epoch": 2.12,
+      "learning_rate": 4.232957110537488e-06,
+      "loss": 0.2333,
+      "step": 3390
+    },
+    {
+      "epoch": 2.12,
+      "learning_rate": 4.17796566641933e-06,
+      "loss": 0.2472,
+      "step": 3400
+    },
+    {
+      "epoch": 2.13,
+      "learning_rate": 4.123239283963087e-06,
+      "loss": 0.248,
+      "step": 3410
+    },
+    {
+      "epoch": 2.14,
+      "learning_rate": 4.068780454714677e-06,
+      "loss": 0.2294,
+      "step": 3420
+    },
+    {
+      "epoch": 2.14,
+      "learning_rate": 4.0145916580390335e-06,
+      "loss": 0.2326,
+      "step": 3430
+    },
+    {
+      "epoch": 2.15,
+      "learning_rate": 3.96067536100724e-06,
+      "loss": 0.2489,
+      "step": 3440
+    },
+    {
+      "epoch": 2.16,
+      "learning_rate": 3.907034018284186e-06,
+      "loss": 0.2375,
+      "step": 3450
+    },
+    {
+      "epoch": 2.16,
+      "learning_rate": 3.853670072016833e-06,
+      "loss": 0.255,
+      "step": 3460
+    },
+    {
+      "epoch": 2.17,
+      "learning_rate": 3.8005859517230214e-06,
+      "loss": 0.258,
+      "step": 3470
+    },
+    {
+      "epoch": 2.17,
+      "learning_rate": 3.747784074180865e-06,
+      "loss": 0.2595,
+      "step": 3480
+    },
+    {
+      "epoch": 2.18,
+      "learning_rate": 3.6952668433187145e-06,
+      "loss": 0.2647,
+      "step": 3490
+    },
+    {
+      "epoch": 2.19,
+      "learning_rate": 3.643036650105725e-06,
+      "loss": 0.2052,
+      "step": 3500
+    },
+    {
+      "epoch": 2.19,
+      "learning_rate": 3.591095872442989e-06,
+      "loss": 0.26,
+      "step": 3510
+    },
+    {
+      "epoch": 2.2,
+      "learning_rate": 3.539446875055287e-06,
+      "loss": 0.2436,
+      "step": 3520
     },
     {
-      "epoch": 2.0,
-      "learning_rate": 1e-05,
-      "loss": 1.0767,
-      "step": 2
+      "epoch": 2.21,
+      "learning_rate": 3.488092009383418e-06,
+      "loss": 0.2306,
+      "step": 3530
+    },
+    {
+      "epoch": 2.21,
+      "learning_rate": 3.4370336134771567e-06,
+      "loss": 0.1932,
+      "step": 3540
+    },
+    {
+      "epoch": 2.22,
+      "learning_rate": 3.386274011888796e-06,
+      "loss": 0.1953,
+      "step": 3550
+    },
+    {
+      "epoch": 2.23,
+      "learning_rate": 3.3358155155673257e-06,
+      "loss": 0.2648,
+      "step": 3560
+    },
+    {
+      "epoch": 2.23,
+      "learning_rate": 3.285660421753216e-06,
+      "loss": 0.2133,
+      "step": 3570
+    },
+    {
+      "epoch": 2.24,
+      "learning_rate": 3.2358110138738297e-06,
+      "loss": 0.2209,
+      "step": 3580
+    },
+    {
+      "epoch": 2.24,
+      "learning_rate": 3.1862695614394745e-06,
+      "loss": 0.2401,
+      "step": 3590
+    },
+    {
+      "epoch": 2.25,
+      "learning_rate": 3.1370383199400613e-06,
+      "loss": 0.214,
+      "step": 3600
+    },
+    {
+      "epoch": 2.26,
+      "learning_rate": 3.0881195307424282e-06,
+      "loss": 0.23,
+      "step": 3610
+    },
+    {
+      "epoch": 2.26,
+      "learning_rate": 3.039515420988297e-06,
+      "loss": 0.2272,
+      "step": 3620
+    },
+    {
+      "epoch": 2.27,
+      "learning_rate": 2.991228203492873e-06,
+      "loss": 0.2326,
+      "step": 3630
+    },
+    {
+      "epoch": 2.27,
+      "learning_rate": 2.9432600766441066e-06,
+      "loss": 0.2299,
+      "step": 3640
+    },
+    {
+      "epoch": 2.28,
+      "learning_rate": 2.8956132243026025e-06,
+      "loss": 0.2481,
+      "step": 3650
+    },
+    {
+      "epoch": 2.29,
+      "learning_rate": 2.8482898157021945e-06,
+      "loss": 0.2266,
+      "step": 3660
+    },
+    {
+      "epoch": 2.29,
+      "learning_rate": 2.8012920053511916e-06,
+      "loss": 0.2223,
+      "step": 3670
+    },
+    {
+      "epoch": 2.3,
+      "learning_rate": 2.7546219329342792e-06,
+      "loss": 0.2181,
+      "step": 3680
+    },
+    {
+      "epoch": 2.31,
+      "learning_rate": 2.7082817232151128e-06,
+      "loss": 0.2287,
+      "step": 3690
+    },
+    {
+      "epoch": 2.31,
+      "learning_rate": 2.662273485939586e-06,
+      "loss": 0.2314,
+      "step": 3700
+    },
+    {
+      "epoch": 2.32,
+      "learning_rate": 2.616599315739766e-06,
+      "loss": 0.2275,
+      "step": 3710
+    },
+    {
+      "epoch": 2.33,
+      "learning_rate": 2.571261292038546e-06,
+      "loss": 0.213,
+      "step": 3720
+    },
+    {
+      "epoch": 2.33,
+      "learning_rate": 2.5262614789549624e-06,
+      "loss": 0.2008,
+      "step": 3730
+    },
+    {
+      "epoch": 2.34,
+      "learning_rate": 2.4816019252102274e-06,
+      "loss": 0.2189,
+      "step": 3740
+    },
+    {
+      "epoch": 2.34,
+      "learning_rate": 2.437284664034457e-06,
+      "loss": 0.2167,
+      "step": 3750
+    },
+    {
+      "epoch": 2.35,
+      "learning_rate": 2.3933117130741e-06,
+      "loss": 0.2059,
+      "step": 3760
+    },
+    {
+      "epoch": 2.36,
+      "learning_rate": 2.3496850743000775e-06,
+      "loss": 0.2265,
+      "step": 3770
+    },
+    {
+      "epoch": 2.36,
+      "learning_rate": 2.306406733916654e-06,
+      "loss": 0.1929,
+      "step": 3780
+    },
+    {
+      "epoch": 2.37,
+      "learning_rate": 2.263478662270987e-06,
+      "loss": 0.2038,
+      "step": 3790
+    },
+    {
+      "epoch": 2.38,
+      "learning_rate": 2.2209028137634402e-06,
+      "loss": 0.1998,
+      "step": 3800
+    },
+    {
+      "epoch": 2.38,
+      "learning_rate": 2.1786811267585984e-06,
+      "loss": 0.2111,
+      "step": 3810
+    },
+    {
+      "epoch": 2.39,
+      "learning_rate": 2.136815523497019e-06,
+      "loss": 0.1965,
+      "step": 3820
+    },
+    {
+      "epoch": 2.39,
+      "learning_rate": 2.09530791000772e-06,
+      "loss": 0.1969,
+      "step": 3830
+    },
+    {
+      "epoch": 2.4,
+      "learning_rate": 2.054160176021399e-06,
+      "loss": 0.2224,
+      "step": 3840
+    },
+    {
+      "epoch": 2.41,
+      "learning_rate": 2.0133741948844056e-06,
+      "loss": 0.2081,
+      "step": 3850
+    },
+    {
+      "epoch": 2.41,
+      "learning_rate": 1.9729518234734455e-06,
+      "loss": 0.2033,
+      "step": 3860
+    },
+    {
+      "epoch": 2.42,
+      "learning_rate": 1.9328949021110453e-06,
+      "loss": 0.1862,
+      "step": 3870
+    },
+    {
+      "epoch": 2.42,
+      "learning_rate": 1.8932052544817747e-06,
+      "loss": 0.2318,
+      "step": 3880
+    },
+    {
+      "epoch": 2.43,
+      "learning_rate": 1.8538846875492034e-06,
+      "loss": 0.2283,
+      "step": 3890
+    },
+    {
+      "epoch": 2.44,
+      "learning_rate": 1.814934991473647e-06,
+      "loss": 0.1989,
+      "step": 3900
+    },
+    {
+      "epoch": 2.44,
+      "learning_rate": 1.776357939530663e-06,
+      "loss": 0.1953,
+      "step": 3910
+    },
+    {
+      "epoch": 2.45,
+      "learning_rate": 1.7381552880303155e-06,
+      "loss": 0.2307,
+      "step": 3920
+    },
+    {
+      "epoch": 2.46,
+      "learning_rate": 1.7003287762372178e-06,
+      "loss": 0.2179,
+      "step": 3930
+    },
+    {
+      "epoch": 2.46,
+      "learning_rate": 1.6628801262913485e-06,
+      "loss": 0.2061,
+      "step": 3940
+    },
+    {
+      "epoch": 2.47,
+      "learning_rate": 1.625811043129647e-06,
+      "loss": 0.1814,
+      "step": 3950
+    },
+    {
+      "epoch": 2.48,
+      "learning_rate": 1.589123214408388e-06,
+      "loss": 0.2217,
+      "step": 3960
+    },
+    {
+      "epoch": 2.48,
+      "learning_rate": 1.552818310426356e-06,
+      "loss": 0.2369,
+      "step": 3970
+    },
+    {
+      "epoch": 2.49,
+      "learning_rate": 1.5168979840487897e-06,
+      "loss": 0.2065,
+      "step": 3980
+    },
+    {
+      "epoch": 2.49,
+      "learning_rate": 1.481363870632142e-06,
+      "loss": 0.2191,
+      "step": 3990
+    },
+    {
+      "epoch": 2.5,
+      "learning_rate": 1.4462175879496198e-06,
+      "loss": 0.2158,
+      "step": 4000
+    },
+    {
+      "epoch": 2.5,
+      "eval_loss": NaN,
+      "eval_runtime": 16.1092,
+      "eval_samples_per_second": 63.566,
+      "eval_steps_per_second": 3.973,
+      "step": 4000
+    },
+    {
+      "epoch": 2.51,
+      "learning_rate": 1.4114607361175314e-06,
+      "loss": 0.2249,
+      "step": 4010
+    },
+    {
+      "epoch": 2.51,
+      "learning_rate": 1.3770948975224462e-06,
+      "loss": 0.2279,
+      "step": 4020
+    },
+    {
+      "epoch": 2.52,
+      "learning_rate": 1.3431216367491384e-06,
+      "loss": 0.1812,
+      "step": 4030
+    },
+    {
+      "epoch": 2.52,
+      "learning_rate": 1.3095425005093676e-06,
+      "loss": 0.2139,
+      "step": 4040
+    },
+    {
+      "epoch": 2.53,
+      "learning_rate": 1.2763590175714547e-06,
+      "loss": 0.2395,
+      "step": 4050
+    },
+    {
+      "epoch": 2.54,
+      "learning_rate": 1.243572698690685e-06,
+      "loss": 0.1955,
+      "step": 4060
+    },
+    {
+      "epoch": 2.54,
+      "learning_rate": 1.2111850365405242e-06,
+      "loss": 0.2058,
+      "step": 4070
+    },
+    {
+      "epoch": 2.55,
+      "learning_rate": 1.1791975056446626e-06,
+      "loss": 0.2055,
+      "step": 4080
+    },
+    {
+      "epoch": 2.56,
+      "learning_rate": 1.147611562309887e-06,
+      "loss": 0.202,
+      "step": 4090
+    },
+    {
+      "epoch": 2.56,
+      "learning_rate": 1.1164286445597716e-06,
+      "loss": 0.2077,
+      "step": 4100
+    },
+    {
+      "epoch": 2.57,
+      "learning_rate": 1.0856501720692159e-06,
+      "loss": 0.1978,
+      "step": 4110
+    },
+    {
+      "epoch": 2.58,
+      "learning_rate": 1.0552775460998067e-06,
+      "loss": 0.1776,
+      "step": 4120
+    },
+    {
+      "epoch": 2.58,
+      "learning_rate": 1.025312149436024e-06,
+      "loss": 0.1917,
+      "step": 4130
+    },
+    {
+      "epoch": 2.59,
+      "learning_rate": 9.957553463222847e-07,
+      "loss": 0.2137,
+      "step": 4140
+    },
+    {
+      "epoch": 2.59,
+      "learning_rate": 9.666084824008349e-07,
+      "loss": 0.1914,
+      "step": 4150
+    },
+    {
+      "epoch": 2.6,
+      "learning_rate": 9.378728846504825e-07,
+      "loss": 0.2026,
+      "step": 4160
+    },
+    {
+      "epoch": 2.61,
+      "learning_rate": 9.095498613261911e-07,
+      "loss": 0.2462,
+      "step": 4170
+    },
+    {
+      "epoch": 2.61,
+      "learning_rate": 8.816407018995088e-07,
+      "loss": 0.1953,
+      "step": 4180
+    },
+    {
+      "epoch": 2.62,
+      "learning_rate": 8.541466769998685e-07,
+      "loss": 0.2095,
+      "step": 4190
+    },
+    {
+      "epoch": 2.62,
+      "learning_rate": 8.270690383567359e-07,
+      "loss": 0.1931,
+      "step": 4200
+    },
+    {
+      "epoch": 2.63,
+      "learning_rate": 8.004090187426238e-07,
+      "loss": 0.2122,
+      "step": 4210
+    },
+    {
+      "epoch": 2.64,
+      "learning_rate": 7.741678319169698e-07,
+      "loss": 0.2336,
+      "step": 4220
+    },
+    {
+      "epoch": 2.64,
+      "learning_rate": 7.48346672570871e-07,
+      "loss": 0.1691,
+      "step": 4230
+    },
+    {
+      "epoch": 2.65,
+      "learning_rate": 7.229467162726966e-07,
+      "loss": 0.1789,
+      "step": 4240
+    },
+    {
+      "epoch": 2.66,
+      "learning_rate": 6.979691194145677e-07,
+      "loss": 0.1936,
+      "step": 4250
+    },
+    {
+      "epoch": 2.66,
+      "learning_rate": 6.734150191597078e-07,
+      "loss": 0.1923,
+      "step": 4260
+    },
+    {
+      "epoch": 2.67,
+      "learning_rate": 6.492855333906733e-07,
+      "loss": 0.2141,
+      "step": 4270
+    },
+    {
+      "epoch": 2.67,
+      "learning_rate": 6.255817606584558e-07,
+      "loss": 0.2156,
+      "step": 4280
+    },
+    {
+      "epoch": 2.68,
+      "learning_rate": 6.023047801324744e-07,
+      "loss": 0.1892,
+      "step": 4290
+    },
+    {
+      "epoch": 2.69,
+      "learning_rate": 5.794556515514327e-07,
+      "loss": 0.2173,
+      "step": 4300
+    },
+    {
+      "epoch": 2.69,
+      "learning_rate": 5.570354151750878e-07,
+      "loss": 0.1992,
+      "step": 4310
+    },
+    {
+      "epoch": 2.7,
+      "learning_rate": 5.350450917368755e-07,
+      "loss": 0.1994,
+      "step": 4320
+    },
+    {
+      "epoch": 2.71,
+      "learning_rate": 5.134856823974444e-07,
+      "loss": 0.2123,
+      "step": 4330
+    },
+    {
+      "epoch": 2.71,
+      "learning_rate": 4.923581686990775e-07,
+      "loss": 0.1867,
+      "step": 4340
+    },
+    {
+      "epoch": 2.72,
+      "learning_rate": 4.716635125210034e-07,
+      "loss": 0.1895,
+      "step": 4350
+    },
+    {
+      "epoch": 2.73,
+      "learning_rate": 4.514026560356044e-07,
+      "loss": 0.2056,
+      "step": 4360
+    },
+    {
+      "epoch": 2.73,
+      "learning_rate": 4.315765216655232e-07,
+      "loss": 0.1885,
+      "step": 4370
+    },
+    {
+      "epoch": 2.74,
+      "learning_rate": 4.121860120416654e-07,
+      "loss": 0.1791,
+      "step": 4380
+    },
+    {
+      "epoch": 2.74,
+      "learning_rate": 3.9323200996210673e-07,
+      "loss": 0.201,
+      "step": 4390
+    },
+    {
+      "epoch": 2.75,
+      "learning_rate": 3.7471537835190087e-07,
+      "loss": 0.2065,
+      "step": 4400
+    },
+    {
+      "epoch": 2.76,
+      "learning_rate": 3.566369602237929e-07,
+      "loss": 0.2152,
+      "step": 4410
+    },
+    {
+      "epoch": 2.76,
+      "learning_rate": 3.389975786398403e-07,
+      "loss": 0.1918,
+      "step": 4420
+    },
+    {
+      "epoch": 2.77,
+      "learning_rate": 3.217980366739404e-07,
+      "loss": 0.2076,
+      "step": 4430
+    },
+    {
+      "epoch": 2.77,
+      "learning_rate": 3.0503911737526536e-07,
+      "loss": 0.2107,
+      "step": 4440
+    },
+    {
+      "epoch": 2.78,
+      "learning_rate": 2.8872158373261847e-07,
+      "loss": 0.1817,
+      "step": 4450
+    },
+    {
+      "epoch": 2.79,
+      "learning_rate": 2.7284617863969296e-07,
+      "loss": 0.1849,
+      "step": 4460
+    },
+    {
+      "epoch": 2.79,
+      "learning_rate": 2.5741362486125023e-07,
+      "loss": 0.2002,
+      "step": 4470
+    },
+    {
+      "epoch": 2.8,
+      "learning_rate": 2.424246250002138e-07,
+      "loss": 0.2012,
+      "step": 4480
+    },
+    {
+      "epoch": 2.81,
+      "learning_rate": 2.278798614656863e-07,
+      "loss": 0.1792,
+      "step": 4490
+    },
+    {
+      "epoch": 2.81,
+      "learning_rate": 2.1377999644187518e-07,
+      "loss": 0.1965,
+      "step": 4500
+    },
+    {
+      "epoch": 2.82,
+      "learning_rate": 2.0012567185794808e-07,
+      "loss": 0.2079,
+      "step": 4510
+    },
+    {
+      "epoch": 2.83,
+      "learning_rate": 1.8691750935880826e-07,
+      "loss": 0.1776,
+      "step": 4520
+    },
+    {
+      "epoch": 2.83,
+      "learning_rate": 1.741561102767908e-07,
+      "loss": 0.1856,
+      "step": 4530
+    },
+    {
+      "epoch": 2.84,
+      "learning_rate": 1.6184205560428655e-07,
+      "loss": 0.1927,
+      "step": 4540
+    },
+    {
+      "epoch": 2.84,
+      "learning_rate": 1.4997590596729227e-07,
+      "loss": 0.1844,
+      "step": 4550
+    },
+    {
+      "epoch": 2.85,
+      "learning_rate": 1.385582015998832e-07,
+      "loss": 0.2104,
+      "step": 4560
+    },
+    {
+      "epoch": 2.86,
+      "learning_rate": 1.2758946231962389e-07,
+      "loss": 0.192,
+      "step": 4570
+    },
+    {
+      "epoch": 2.86,
+      "learning_rate": 1.1707018750389376e-07,
+      "loss": 0.2106,
+      "step": 4580
+    },
+    {
+      "epoch": 2.87,
+      "learning_rate": 1.0700085606716092e-07,
+      "loss": 0.1914,
+      "step": 4590
+    },
+    {
+      "epoch": 2.88,
+      "learning_rate": 9.738192643917066e-08,
+      "loss": 0.1844,
+      "step": 4600
+    },
+    {
+      "epoch": 2.88,
+      "learning_rate": 8.8213836544081e-08,
+      "loss": 0.1783,
+      "step": 4610
+    },
+    {
+      "epoch": 2.89,
+      "learning_rate": 7.949700378051872e-08,
+      "loss": 0.1934,
+      "step": 4620
+    },
+    {
+      "epoch": 2.89,
+      "learning_rate": 7.123182500258119e-08,
+      "loss": 0.1946,
+      "step": 4630
+    },
+    {
+      "epoch": 2.9,
+      "learning_rate": 6.341867650176637e-08,
+      "loss": 0.2097,
+      "step": 4640
+    },
+    {
+      "epoch": 2.91,
+      "learning_rate": 5.605791398984317e-08,
+      "loss": 0.1757,
+      "step": 4650
+    },
+    {
+      "epoch": 2.91,
+      "learning_rate": 4.914987258265558e-08,
+      "loss": 0.1656,
+      "step": 4660
+    },
+    {
+      "epoch": 2.92,
+      "learning_rate": 4.2694866784862565e-08,
+      "loss": 0.1923,
+      "step": 4670
+    },
+    {
+      "epoch": 2.92,
+      "learning_rate": 3.669319047562514e-08,
+      "loss": 0.2078,
+      "step": 4680
+    },
+    {
+      "epoch": 2.93,
+      "learning_rate": 3.114511689522592e-08,
+      "loss": 0.2223,
+      "step": 4690
+    },
+    {
+      "epoch": 2.94,
+      "learning_rate": 2.6050898632625777e-08,
+      "loss": 0.2062,
+      "step": 4700
+    },
+    {
+      "epoch": 2.94,
+      "learning_rate": 2.1410767613965212e-08,
+      "loss": 0.1708,
+      "step": 4710
+    },
+    {
+      "epoch": 2.95,
+      "learning_rate": 1.722493509200729e-08,
+      "loss": 0.2104,
+      "step": 4720
+    },
+    {
+      "epoch": 2.96,
+      "learning_rate": 1.3493591636519753e-08,
+      "loss": 0.1751,
+      "step": 4730
+    },
+    {
+      "epoch": 2.96,
+      "learning_rate": 1.021690712559642e-08,
+      "loss": 0.2018,
+      "step": 4740
+    },
+    {
+      "epoch": 2.97,
+      "learning_rate": 7.395030737924469e-09,
+      "loss": 0.1806,
+      "step": 4750
+    },
+    {
+      "epoch": 2.98,
+      "learning_rate": 5.028090945993214e-09,
+      "loss": 0.1807,
+      "step": 4760
+    },
+    {
+      "epoch": 2.98,
+      "learning_rate": 3.1161955102465558e-09,
+      "loss": 0.19,
+      "step": 4770
+    },
+    {
+      "epoch": 2.99,
+      "learning_rate": 1.6594314741724682e-09,
+      "loss": 0.1996,
+      "step": 4780
+    },
+    {
+      "epoch": 2.99,
+      "learning_rate": 6.578651603439401e-10,
+      "loss": 0.2078,
+      "step": 4790
     },
     {
       "epoch": 3.0,
-      "learning_rate": 0.0,
-      "loss": 0.7536,
-      "step": 3
+      "learning_rate": 1.115421673980599e-10,
+      "loss": 0.2101,
+      "step": 4800
     },
     {
       "epoch": 3.0,
-      "step": 3,
-      "total_flos": 2327628803473408.0,
-      "train_loss": 0.9601547519365946,
-      "train_runtime": 1446.9004,
-      "train_samples_per_second": 0.531,
-      "train_steps_per_second": 0.002
+      "step": 4800,
+      "total_flos": 9.142542195721503e+17,
+      "train_loss": 0.4117240861803293,
+      "train_runtime": 65160.1252,
+      "train_samples_per_second": 4.715,
+      "train_steps_per_second": 0.074
     }
   ],
-  "max_steps": 3,
+  "max_steps": 4800,
   "num_train_epochs": 3,
-  "total_flos": 2327628803473408.0,
+  "total_flos": 9.142542195721503e+17,
   "trial_name": null,
   "trial_params": null
 }

training_args.bin

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:a37c01339dfc23b560f46c214c764dcc232eff864973e5037cb88d8c3da23be3
-size 3835
+oid sha256:9df58c911ce5d65244c39e5ebc34e9406d4523b4e5c2e73cf24a0da4967b85b2
+size 3771