Hugging Face's logo

saim1212
/
qwen2_2b_instruct_modified_loras

PEFT
TensorBoard
Safetensors
llama-factory
lora
Generated from Trainer
Model card Files
xet
Metrics Community
saim1212 commited on
Commit
0eb33ac
·
verified ·
1 Parent(s): 6e993a2

second model upload

Browse files
Files changed (50) hide show
  1. README.md +5 -5
  2. adapter_config.json +1 -1
  3. adapter_model.safetensors +2 -2
  4. all_results.json +6 -6
  5. checkpoint-1000/adapter_config.json +1 -1
  6. checkpoint-1000/adapter_model.safetensors +2 -2
  7. checkpoint-1000/optimizer.pt +2 -2
  8. checkpoint-1000/rng_state_0.pth +1 -1
  9. checkpoint-1000/rng_state_1.pth +1 -1
  10. checkpoint-1000/scheduler.pt +1 -1
  11. checkpoint-1000/trainer_state.json +405 -405
  12. checkpoint-1000/training_args.bin +1 -1
  13. checkpoint-2000/adapter_config.json +1 -1
  14. checkpoint-2000/adapter_model.safetensors +2 -2
  15. checkpoint-2000/optimizer.pt +2 -2
  16. checkpoint-2000/rng_state_0.pth +1 -1
  17. checkpoint-2000/rng_state_1.pth +1 -1
  18. checkpoint-2000/scheduler.pt +1 -1
  19. checkpoint-2000/trainer_state.json +805 -805
  20. checkpoint-2000/training_args.bin +1 -1
  21. checkpoint-3000/adapter_config.json +1 -1
  22. checkpoint-3000/adapter_model.safetensors +2 -2
  23. checkpoint-3000/optimizer.pt +2 -2
  24. checkpoint-3000/rng_state_0.pth +1 -1
  25. checkpoint-3000/rng_state_1.pth +1 -1
  26. checkpoint-3000/scheduler.pt +1 -1
  27. checkpoint-3000/trainer_state.json +1198 -1198
  28. checkpoint-3000/training_args.bin +1 -1
  29. checkpoint-3125/README.md +202 -0
  30. checkpoint-3125/adapter_config.json +26 -0
  31. checkpoint-3125/adapter_model.safetensors +3 -0
  32. checkpoint-3125/added_tokens.json +16 -0
  33. checkpoint-3125/chat_template.json +3 -0
  34. checkpoint-3125/merges.txt +0 -0
  35. checkpoint-3125/optimizer.pt +3 -0
  36. checkpoint-3125/preprocessor_config.json +29 -0
  37. checkpoint-3125/rng_state_0.pth +3 -0
  38. checkpoint-3125/rng_state_1.pth +3 -0
  39. checkpoint-3125/scheduler.pt +3 -0
  40. checkpoint-3125/special_tokens_map.json +31 -0
  41. checkpoint-3125/tokenizer.json +3 -0
  42. checkpoint-3125/tokenizer_config.json +145 -0
  43. checkpoint-3125/trainer_state.json +2217 -0
  44. checkpoint-3125/training_args.bin +3 -0
  45. checkpoint-3125/vocab.json +0 -0
  46. runs/Feb03_13-32-36_120d765cf84c/events.out.tfevents.1738589701.120d765cf84c.147.0 +3 -0
  47. train_results.json +6 -6
  48. trainer_log.jsonl +0 -0
  49. trainer_state.json +0 -0
  50. training_args.bin +1 -1
README.md CHANGED
@@ -36,17 +36,17 @@ More information needed
36
 
37
  The following hyperparameters were used during training:
38
  - learning_rate: 2e-05
39
- - train_batch_size: 3
40
  - eval_batch_size: 8
41
  - seed: 42
42
  - distributed_type: multi-GPU
43
  - num_devices: 2
44
- - total_train_batch_size: 6
45
  - total_eval_batch_size: 16
46
  - optimizer: Use adamw_torch with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
47
  - lr_scheduler_type: cosine
48
  - lr_scheduler_warmup_ratio: 0.1
49
- - num_epochs: 40.0
50
  - mixed_precision_training: Native AMP
51
 
52
  ### Training results
@@ -57,6 +57,6 @@ The following hyperparameters were used during training:
57
 
58
  - PEFT 0.12.0
59
  - Transformers 4.48.2
60
- - Pytorch 2.4.0
61
- - Datasets 3.1.0
62
  - Tokenizers 0.21.0
 
36
 
37
  The following hyperparameters were used during training:
38
  - learning_rate: 2e-05
39
+ - train_batch_size: 2
40
  - eval_batch_size: 8
41
  - seed: 42
42
  - distributed_type: multi-GPU
43
  - num_devices: 2
44
+ - total_train_batch_size: 4
45
  - total_eval_batch_size: 16
46
  - optimizer: Use adamw_torch with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
47
  - lr_scheduler_type: cosine
48
  - lr_scheduler_warmup_ratio: 0.1
49
+ - num_epochs: 25.0
50
  - mixed_precision_training: Native AMP
51
 
52
  ### Training results
 
57
 
58
  - PEFT 0.12.0
59
  - Transformers 4.48.2
60
+ - Pytorch 2.4.1+cu121
61
+ - Datasets 3.2.0
62
  - Tokenizers 0.21.0
adapter_config.json CHANGED
@@ -19,7 +19,7 @@
19
  "r": 16,
20
  "rank_pattern": {},
21
  "revision": null,
22
- "target_modules": "^(?!.*visual.patch_embed|visual.blocks).*(?:o_proj|v_proj|down_proj|k_proj|q_proj|up_proj|gate_proj).*",
23
  "task_type": "CAUSAL_LM",
24
  "use_dora": false,
25
  "use_rslora": false
 
19
  "r": 16,
20
  "rank_pattern": {},
21
  "revision": null,
22
+ "target_modules": "^(?!.*patch_embed).*(?:o_proj|fc2|v_proj|down_proj|q_proj|up_proj|proj|qkv|gate_proj|k_proj|fc1).*",
23
  "task_type": "CAUSAL_LM",
24
  "use_dora": false,
25
  "use_rslora": false
adapter_model.safetensors CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:788ec8b5da707abf6948e70f786bef2295da243d71c6e22f81ad1a15d216c6d4
3
- size 73911112
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:335a6122bd58fa538a3cd935a28d1549f4413046e91785944e52bf8e4db9e78c
3
+ size 115886968
all_results.json CHANGED
@@ -1,8 +1,8 @@
1
  {
2
- "epoch": 40.0,
3
- "total_flos": 2.004550530755461e+17,
4
- "train_loss": 0.09495920253196216,
5
- "train_runtime": 28025.0795,
6
- "train_samples_per_second": 0.714,
7
- "train_steps_per_second": 0.12
8
  }
 
1
  {
2
+ "epoch": 25.0,
3
+ "total_flos": 1.246578685771776e+17,
4
+ "train_loss": 0.15220735648881645,
5
+ "train_runtime": 31927.3271,
6
+ "train_samples_per_second": 0.392,
7
+ "train_steps_per_second": 0.098
8
  }
checkpoint-1000/adapter_config.json CHANGED
@@ -19,7 +19,7 @@
19
  "r": 16,
20
  "rank_pattern": {},
21
  "revision": null,
22
- "target_modules": "^(?!.*visual.patch_embed|visual.blocks).*(?:o_proj|v_proj|down_proj|k_proj|q_proj|up_proj|gate_proj).*",
23
  "task_type": "CAUSAL_LM",
24
  "use_dora": false,
25
  "use_rslora": false
 
19
  "r": 16,
20
  "rank_pattern": {},
21
  "revision": null,
22
+ "target_modules": "^(?!.*patch_embed).*(?:o_proj|fc2|v_proj|down_proj|q_proj|up_proj|proj|qkv|gate_proj|k_proj|fc1).*",
23
  "task_type": "CAUSAL_LM",
24
  "use_dora": false,
25
  "use_rslora": false
checkpoint-1000/adapter_model.safetensors CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:e3deb2c723ca194b1a5a0be565dc2099d36a5628ab4101146cefaf66983f7615
3
- size 73911112
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:e9534a957d067e480096493495461e8157fe6067e955859f7c64b335a32c0860
3
+ size 115886968
checkpoint-1000/optimizer.pt CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:deb228cd06b4526571fbb9a3a19a1bc7cb1dca378bc66a296104de67254345af
3
- size 148047978
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:62893254c570f37d24aebbc71654a729eaa947be50edcb95370ec06c1b863b8a
3
+ size 232150634
checkpoint-1000/rng_state_0.pth CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:6f3803bff3f596c03b55881de967a825b5734e4a581739164f9cb9e7fd1aee89
3
  size 14512
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:3dcb161b22b2558dbf7e3f8c871050cec383d11a40423fab11f18d5e630639bf
3
  size 14512
checkpoint-1000/rng_state_1.pth CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:d768a04b798e2ca42effbe096b8e4481f32a402a9125a2ced390586dab8eb29e
3
  size 14512
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:d50af6aef769414a6f28fa1b1bc51ce707dc8ecd15474e03f99a2f10fde086be
3
  size 14512
checkpoint-1000/scheduler.pt CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:f08df4df4cc87960e1f920262b834a1d27f67a7cbdd32431fe0bca63620d0ec0
3
  size 1064
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:16a6fcf6da79db5e52d37f34a4a3ac0e2edb5e36532b8138b70ac0fb7b89a42b
3
  size 1064
checkpoint-1000/trainer_state.json CHANGED
@@ -1,7 +1,7 @@
1
  {
2
  "best_metric": null,
3
  "best_model_checkpoint": null,
4
- "epoch": 11.904761904761905,
5
  "eval_steps": 500,
6
  "global_step": 1000,
7
  "is_hyper_param_search": false,
@@ -9,710 +9,710 @@
9
  "is_world_process_zero": true,
10
  "log_history": [
11
  {
12
- "epoch": 0.11904761904761904,
13
- "grad_norm": 5.44253396987915,
14
- "learning_rate": 5.952380952380953e-07,
15
- "loss": 1.5321,
16
  "step": 10
17
  },
18
  {
19
- "epoch": 0.23809523809523808,
20
- "grad_norm": 2.2140841484069824,
21
- "learning_rate": 1.1904761904761906e-06,
22
- "loss": 1.1934,
23
  "step": 20
24
  },
25
  {
26
- "epoch": 0.35714285714285715,
27
- "grad_norm": 1.413270115852356,
28
- "learning_rate": 1.7857142857142859e-06,
29
- "loss": 0.9128,
30
  "step": 30
31
  },
32
  {
33
- "epoch": 0.47619047619047616,
34
- "grad_norm": 1.0657840967178345,
35
- "learning_rate": 2.380952380952381e-06,
36
- "loss": 0.7772,
37
  "step": 40
38
  },
39
  {
40
- "epoch": 0.5952380952380952,
41
- "grad_norm": 0.9593137502670288,
42
- "learning_rate": 2.9761904761904763e-06,
43
- "loss": 0.7849,
44
  "step": 50
45
  },
46
  {
47
- "epoch": 0.7142857142857143,
48
- "grad_norm": 1.134382724761963,
49
- "learning_rate": 3.5714285714285718e-06,
50
- "loss": 0.716,
51
  "step": 60
52
  },
53
  {
54
- "epoch": 0.8333333333333334,
55
- "grad_norm": 0.618364691734314,
56
- "learning_rate": 4.166666666666667e-06,
57
- "loss": 0.6678,
58
  "step": 70
59
  },
60
  {
61
- "epoch": 0.9523809523809523,
62
- "grad_norm": 0.7465847134590149,
63
- "learning_rate": 4.761904761904762e-06,
64
- "loss": 0.671,
65
  "step": 80
66
  },
67
  {
68
- "epoch": 1.0714285714285714,
69
- "grad_norm": 0.6726512908935547,
70
- "learning_rate": 5.357142857142857e-06,
71
- "loss": 0.6186,
72
  "step": 90
73
  },
74
  {
75
- "epoch": 1.1904761904761905,
76
- "grad_norm": 0.6353351473808289,
77
- "learning_rate": 5.9523809523809525e-06,
78
- "loss": 0.5818,
79
  "step": 100
80
  },
81
  {
82
- "epoch": 1.3095238095238095,
83
- "grad_norm": 0.7020516991615295,
84
- "learning_rate": 6.547619047619048e-06,
85
- "loss": 0.6056,
86
  "step": 110
87
  },
88
  {
89
- "epoch": 1.4285714285714286,
90
- "grad_norm": 1.1165708303451538,
91
- "learning_rate": 7.1428571428571436e-06,
92
- "loss": 0.6225,
93
  "step": 120
94
  },
95
  {
96
- "epoch": 1.5476190476190477,
97
- "grad_norm": 1.7502918243408203,
98
- "learning_rate": 7.738095238095238e-06,
99
- "loss": 0.6321,
100
  "step": 130
101
  },
102
  {
103
- "epoch": 1.6666666666666665,
104
- "grad_norm": 0.6962683796882629,
105
- "learning_rate": 8.333333333333334e-06,
106
- "loss": 0.5751,
107
  "step": 140
108
  },
109
  {
110
- "epoch": 1.7857142857142856,
111
- "grad_norm": 0.876686155796051,
112
- "learning_rate": 8.92857142857143e-06,
113
- "loss": 0.5974,
114
  "step": 150
115
  },
116
  {
117
- "epoch": 1.9047619047619047,
118
- "grad_norm": 0.5832173228263855,
119
- "learning_rate": 9.523809523809525e-06,
120
- "loss": 0.6085,
121
  "step": 160
122
  },
123
  {
124
- "epoch": 2.0238095238095237,
125
- "grad_norm": 0.44337984919548035,
126
- "learning_rate": 1.011904761904762e-05,
127
- "loss": 0.5713,
128
  "step": 170
129
  },
130
  {
131
- "epoch": 2.142857142857143,
132
- "grad_norm": 1.2403652667999268,
133
- "learning_rate": 1.0714285714285714e-05,
134
- "loss": 0.5165,
135
  "step": 180
136
  },
137
  {
138
- "epoch": 2.261904761904762,
139
- "grad_norm": 0.7495181560516357,
140
- "learning_rate": 1.130952380952381e-05,
141
- "loss": 0.5448,
142
  "step": 190
143
  },
144
  {
145
- "epoch": 2.380952380952381,
146
- "grad_norm": 0.8501800298690796,
147
- "learning_rate": 1.1904761904761905e-05,
148
- "loss": 0.5122,
149
  "step": 200
150
  },
151
  {
152
- "epoch": 2.5,
153
- "grad_norm": 0.8133195638656616,
154
- "learning_rate": 1.25e-05,
155
- "loss": 0.5337,
156
  "step": 210
157
  },
158
  {
159
- "epoch": 2.619047619047619,
160
- "grad_norm": 1.3526533842086792,
161
- "learning_rate": 1.3095238095238096e-05,
162
- "loss": 0.4943,
163
  "step": 220
164
  },
165
  {
166
- "epoch": 2.738095238095238,
167
- "grad_norm": 1.0535343885421753,
168
- "learning_rate": 1.3690476190476192e-05,
169
- "loss": 0.521,
170
  "step": 230
171
  },
172
  {
173
- "epoch": 2.857142857142857,
174
- "grad_norm": 1.2215440273284912,
175
- "learning_rate": 1.4285714285714287e-05,
176
- "loss": 0.5461,
177
  "step": 240
178
  },
179
  {
180
- "epoch": 2.9761904761904763,
181
- "grad_norm": 1.1781115531921387,
182
- "learning_rate": 1.4880952380952383e-05,
183
- "loss": 0.5346,
184
  "step": 250
185
  },
186
  {
187
- "epoch": 3.0952380952380953,
188
- "grad_norm": 0.762434184551239,
189
- "learning_rate": 1.5476190476190476e-05,
190
- "loss": 0.4855,
191
  "step": 260
192
  },
193
  {
194
- "epoch": 3.2142857142857144,
195
- "grad_norm": 1.287593126296997,
196
- "learning_rate": 1.6071428571428572e-05,
197
- "loss": 0.473,
198
  "step": 270
199
  },
200
  {
201
- "epoch": 3.3333333333333335,
202
- "grad_norm": 0.9310547113418579,
203
- "learning_rate": 1.6666666666666667e-05,
204
- "loss": 0.4201,
205
  "step": 280
206
  },
207
  {
208
- "epoch": 3.4523809523809526,
209
- "grad_norm": 1.1341677904129028,
210
- "learning_rate": 1.7261904761904763e-05,
211
- "loss": 0.3884,
212
  "step": 290
213
  },
214
  {
215
- "epoch": 3.571428571428571,
216
- "grad_norm": 1.483049750328064,
217
- "learning_rate": 1.785714285714286e-05,
218
- "loss": 0.4662,
219
  "step": 300
220
  },
221
  {
222
- "epoch": 3.6904761904761907,
223
- "grad_norm": 0.9312372803688049,
224
- "learning_rate": 1.8452380952380954e-05,
225
- "loss": 0.4688,
226
  "step": 310
227
  },
228
  {
229
- "epoch": 3.8095238095238093,
230
- "grad_norm": 1.0440223217010498,
231
- "learning_rate": 1.904761904761905e-05,
232
- "loss": 0.4727,
233
  "step": 320
234
  },
235
  {
236
- "epoch": 3.928571428571429,
237
- "grad_norm": 1.20750892162323,
238
- "learning_rate": 1.9642857142857145e-05,
239
- "loss": 0.4062,
240
  "step": 330
241
  },
242
  {
243
- "epoch": 4.0476190476190474,
244
- "grad_norm": 1.5511012077331543,
245
- "learning_rate": 1.999991365731819e-05,
246
- "loss": 0.4459,
247
  "step": 340
248
  },
249
  {
250
- "epoch": 4.166666666666667,
251
- "grad_norm": 1.3020656108856201,
252
- "learning_rate": 1.9998942319271076e-05,
253
- "loss": 0.3667,
254
  "step": 350
255
  },
256
  {
257
- "epoch": 4.285714285714286,
258
- "grad_norm": 0.9357143640518188,
259
- "learning_rate": 1.9996891820008165e-05,
260
- "loss": 0.3703,
261
  "step": 360
262
  },
263
  {
264
- "epoch": 4.404761904761905,
265
- "grad_norm": 1.120471477508545,
266
- "learning_rate": 1.9993762380834786e-05,
267
- "loss": 0.3797,
268
  "step": 370
269
  },
270
  {
271
- "epoch": 4.523809523809524,
272
- "grad_norm": 1.4476983547210693,
273
- "learning_rate": 1.9989554339503612e-05,
274
- "loss": 0.366,
275
  "step": 380
276
  },
277
  {
278
- "epoch": 4.642857142857143,
279
- "grad_norm": 0.8721790313720703,
280
- "learning_rate": 1.998426815017817e-05,
281
- "loss": 0.3914,
282
  "step": 390
283
  },
284
  {
285
- "epoch": 4.761904761904762,
286
- "grad_norm": 1.4711418151855469,
287
- "learning_rate": 1.997790438338385e-05,
288
- "loss": 0.3665,
289
  "step": 400
290
  },
291
  {
292
- "epoch": 4.880952380952381,
293
- "grad_norm": 1.216174840927124,
294
- "learning_rate": 1.9970463725946335e-05,
295
- "loss": 0.345,
296
  "step": 410
297
  },
298
  {
299
- "epoch": 5.0,
300
- "grad_norm": 0.8730125427246094,
301
- "learning_rate": 1.9962847054074817e-05,
302
- "loss": 0.3732,
303
  "step": 420
304
  },
305
  {
306
- "epoch": 5.119047619047619,
307
- "grad_norm": 1.7755807638168335,
308
- "learning_rate": 1.9953362612814294e-05,
309
- "loss": 0.2578,
310
  "step": 430
311
  },
312
  {
313
- "epoch": 5.238095238095238,
314
- "grad_norm": 1.7037990093231201,
315
- "learning_rate": 1.9942803929643423e-05,
316
- "loss": 0.2561,
317
  "step": 440
318
  },
319
  {
320
- "epoch": 5.357142857142857,
321
- "grad_norm": 1.140688180923462,
322
- "learning_rate": 1.9931172144134874e-05,
323
- "loss": 0.2525,
324
  "step": 450
325
  },
326
  {
327
- "epoch": 5.476190476190476,
328
- "grad_norm": 1.492291808128357,
329
- "learning_rate": 1.99184685116786e-05,
330
- "loss": 0.2727,
331
  "step": 460
332
  },
333
  {
334
- "epoch": 5.595238095238095,
335
- "grad_norm": 1.1798468828201294,
336
- "learning_rate": 1.9904694403346362e-05,
337
- "loss": 0.291,
338
  "step": 470
339
  },
340
  {
341
- "epoch": 5.714285714285714,
342
- "grad_norm": 1.1415112018585205,
343
- "learning_rate": 1.9889851305743742e-05,
344
- "loss": 0.2845,
345
  "step": 480
346
  },
347
  {
348
- "epoch": 5.833333333333333,
349
- "grad_norm": 1.8219681978225708,
350
- "learning_rate": 1.9873940820849714e-05,
351
- "loss": 0.2746,
352
  "step": 490
353
  },
354
  {
355
- "epoch": 5.9523809523809526,
356
- "grad_norm": 1.3858968019485474,
357
- "learning_rate": 1.9856964665843717e-05,
358
- "loss": 0.2861,
359
  "step": 500
360
  },
361
  {
362
- "epoch": 6.071428571428571,
363
- "grad_norm": 1.2878497838974,
364
- "learning_rate": 1.9838924672920344e-05,
365
- "loss": 0.239,
366
  "step": 510
367
  },
368
  {
369
- "epoch": 6.190476190476191,
370
- "grad_norm": 2.5550217628479004,
371
- "learning_rate": 1.9819822789091597e-05,
372
- "loss": 0.1605,
373
  "step": 520
374
  },
375
  {
376
- "epoch": 6.309523809523809,
377
- "grad_norm": 2.016493082046509,
378
- "learning_rate": 1.9799661075976754e-05,
379
- "loss": 0.2069,
380
  "step": 530
381
  },
382
  {
383
- "epoch": 6.428571428571429,
384
- "grad_norm": 1.8905669450759888,
385
- "learning_rate": 1.977844170957984e-05,
386
- "loss": 0.1953,
387
  "step": 540
388
  },
389
  {
390
- "epoch": 6.5476190476190474,
391
- "grad_norm": 1.6992967128753662,
392
- "learning_rate": 1.9756166980054812e-05,
393
- "loss": 0.1837,
394
  "step": 550
395
  },
396
  {
397
- "epoch": 6.666666666666667,
398
- "grad_norm": 1.893633246421814,
399
- "learning_rate": 1.973283929145836e-05,
400
- "loss": 0.1947,
401
  "step": 560
402
  },
403
  {
404
- "epoch": 6.785714285714286,
405
- "grad_norm": 1.6204354763031006,
406
- "learning_rate": 1.9708461161490456e-05,
407
- "loss": 0.1836,
408
  "step": 570
409
  },
410
  {
411
- "epoch": 6.904761904761905,
412
- "grad_norm": 2.0902771949768066,
413
- "learning_rate": 1.9683035221222617e-05,
414
- "loss": 0.1971,
415
  "step": 580
416
  },
417
  {
418
- "epoch": 7.023809523809524,
419
- "grad_norm": 2.2528581619262695,
420
- "learning_rate": 1.9656564214813948e-05,
421
- "loss": 0.1864,
422
  "step": 590
423
  },
424
  {
425
- "epoch": 7.142857142857143,
426
- "grad_norm": 0.9720670580863953,
427
- "learning_rate": 1.9629050999214977e-05,
428
- "loss": 0.114,
429
  "step": 600
430
  },
431
  {
432
- "epoch": 7.261904761904762,
433
- "grad_norm": 1.1338551044464111,
434
- "learning_rate": 1.960049854385929e-05,
435
- "loss": 0.1331,
436
  "step": 610
437
  },
438
  {
439
- "epoch": 7.380952380952381,
440
- "grad_norm": 1.8178822994232178,
441
- "learning_rate": 1.957090993034306e-05,
442
- "loss": 0.1144,
443
  "step": 620
444
  },
445
  {
446
- "epoch": 7.5,
447
- "grad_norm": 2.2525339126586914,
448
- "learning_rate": 1.9540288352092474e-05,
449
- "loss": 0.1298,
450
  "step": 630
451
  },
452
  {
453
- "epoch": 7.619047619047619,
454
- "grad_norm": 2.6778416633605957,
455
- "learning_rate": 1.9508637114019037e-05,
456
- "loss": 0.131,
457
  "step": 640
458
  },
459
  {
460
- "epoch": 7.738095238095238,
461
- "grad_norm": 1.2893160581588745,
462
- "learning_rate": 1.947595963216292e-05,
463
- "loss": 0.1332,
464
  "step": 650
465
  },
466
  {
467
- "epoch": 7.857142857142857,
468
- "grad_norm": 2.0039525032043457,
469
- "learning_rate": 1.9442259433324247e-05,
470
- "loss": 0.1385,
471
  "step": 660
472
  },
473
  {
474
- "epoch": 7.976190476190476,
475
- "grad_norm": 1.5008896589279175,
476
- "learning_rate": 1.9407540154682473e-05,
477
- "loss": 0.1564,
478
  "step": 670
479
  },
480
  {
481
- "epoch": 8.095238095238095,
482
- "grad_norm": 1.5358928442001343,
483
- "learning_rate": 1.9371805543403826e-05,
484
- "loss": 0.0916,
485
  "step": 680
486
  },
487
  {
488
- "epoch": 8.214285714285714,
489
- "grad_norm": 1.3121554851531982,
490
- "learning_rate": 1.9335059456236877e-05,
491
- "loss": 0.0928,
492
  "step": 690
493
  },
494
  {
495
- "epoch": 8.333333333333334,
496
- "grad_norm": 1.5265170335769653,
497
- "learning_rate": 1.9297305859096305e-05,
498
- "loss": 0.0858,
499
  "step": 700
500
  },
501
  {
502
- "epoch": 8.452380952380953,
503
- "grad_norm": 2.1085143089294434,
504
- "learning_rate": 1.925854882663486e-05,
505
- "loss": 0.0821,
506
  "step": 710
507
  },
508
  {
509
- "epoch": 8.571428571428571,
510
- "grad_norm": 1.8447340726852417,
511
- "learning_rate": 1.9218792541803582e-05,
512
- "loss": 0.1064,
513
  "step": 720
514
  },
515
  {
516
- "epoch": 8.69047619047619,
517
- "grad_norm": 1.8958710432052612,
518
- "learning_rate": 1.9178041295400383e-05,
519
- "loss": 0.082,
520
  "step": 730
521
  },
522
  {
523
- "epoch": 8.80952380952381,
524
- "grad_norm": 1.9607478380203247,
525
- "learning_rate": 1.9136299485606898e-05,
526
- "loss": 0.0901,
527
  "step": 740
528
  },
529
  {
530
- "epoch": 8.928571428571429,
531
- "grad_norm": 2.034801959991455,
532
- "learning_rate": 1.9093571617513853e-05,
533
- "loss": 0.1067,
534
  "step": 750
535
  },
536
  {
537
- "epoch": 9.047619047619047,
538
- "grad_norm": 1.1250438690185547,
539
- "learning_rate": 1.90498623026348e-05,
540
- "loss": 0.0558,
541
  "step": 760
542
  },
543
  {
544
- "epoch": 9.166666666666666,
545
- "grad_norm": 3.2768161296844482,
546
- "learning_rate": 1.9005176258408434e-05,
547
- "loss": 0.0464,
548
  "step": 770
549
  },
550
  {
551
- "epoch": 9.285714285714286,
552
- "grad_norm": 1.0632288455963135,
553
- "learning_rate": 1.8959518307689444e-05,
554
- "loss": 0.0499,
555
  "step": 780
556
  },
557
  {
558
- "epoch": 9.404761904761905,
559
- "grad_norm": 1.1634855270385742,
560
- "learning_rate": 1.8912893378227984e-05,
561
- "loss": 0.0665,
562
  "step": 790
563
  },
564
  {
565
- "epoch": 9.523809523809524,
566
- "grad_norm": 2.089411497116089,
567
- "learning_rate": 1.8865306502137852e-05,
568
- "loss": 0.0715,
569
  "step": 800
570
  },
571
  {
572
- "epoch": 9.642857142857142,
573
- "grad_norm": 0.46574532985687256,
574
- "learning_rate": 1.8816762815353368e-05,
575
- "loss": 0.0453,
576
  "step": 810
577
  },
578
  {
579
- "epoch": 9.761904761904763,
580
- "grad_norm": 2.3916738033294678,
581
- "learning_rate": 1.876726755707508e-05,
582
- "loss": 0.1086,
583
  "step": 820
584
  },
585
  {
586
- "epoch": 9.880952380952381,
587
- "grad_norm": 1.7373374700546265,
588
- "learning_rate": 1.8716826069204297e-05,
589
- "loss": 0.1003,
590
  "step": 830
591
  },
592
  {
593
- "epoch": 10.0,
594
- "grad_norm": 1.2700127363204956,
595
- "learning_rate": 1.8665443795766566e-05,
596
- "loss": 0.054,
597
  "step": 840
598
  },
599
  {
600
- "epoch": 10.119047619047619,
601
- "grad_norm": 0.9935422539710999,
602
- "learning_rate": 1.8613126282324092e-05,
603
- "loss": 0.0403,
604
  "step": 850
605
  },
606
  {
607
- "epoch": 10.238095238095237,
608
- "grad_norm": 1.3984347581863403,
609
- "learning_rate": 1.855987917537724e-05,
610
- "loss": 0.0584,
611
  "step": 860
612
  },
613
  {
614
- "epoch": 10.357142857142858,
615
- "grad_norm": 2.4390032291412354,
616
- "learning_rate": 1.8505708221755106e-05,
617
- "loss": 0.0672,
618
  "step": 870
619
  },
620
  {
621
- "epoch": 10.476190476190476,
622
- "grad_norm": 0.9531667828559875,
623
- "learning_rate": 1.8450619267995283e-05,
624
- "loss": 0.0425,
625
  "step": 880
626
  },
627
  {
628
- "epoch": 10.595238095238095,
629
- "grad_norm": 0.42088013887405396,
630
- "learning_rate": 1.8394618259712866e-05,
631
- "loss": 0.0399,
632
  "step": 890
633
  },
634
  {
635
- "epoch": 10.714285714285714,
636
- "grad_norm": 0.8044432997703552,
637
- "learning_rate": 1.8337711240958734e-05,
638
- "loss": 0.0516,
639
  "step": 900
640
  },
641
  {
642
- "epoch": 10.833333333333334,
643
- "grad_norm": 1.7414997816085815,
644
- "learning_rate": 1.827990435356725e-05,
645
- "loss": 0.043,
646
  "step": 910
647
  },
648
  {
649
- "epoch": 10.952380952380953,
650
- "grad_norm": 0.6912918090820312,
651
- "learning_rate": 1.822120383649339e-05,
652
- "loss": 0.0444,
653
  "step": 920
654
  },
655
  {
656
- "epoch": 11.071428571428571,
657
- "grad_norm": 1.3013064861297607,
658
- "learning_rate": 1.8161616025139375e-05,
659
- "loss": 0.0436,
660
  "step": 930
661
  },
662
  {
663
- "epoch": 11.19047619047619,
664
- "grad_norm": 1.2811461687088013,
665
- "learning_rate": 1.8101147350670905e-05,
666
- "loss": 0.039,
667
  "step": 940
668
  },
669
  {
670
- "epoch": 11.30952380952381,
671
- "grad_norm": 6.565953254699707,
672
- "learning_rate": 1.8039804339323083e-05,
673
- "loss": 0.0411,
674
  "step": 950
675
  },
676
  {
677
- "epoch": 11.428571428571429,
678
- "grad_norm": 1.2721556425094604,
679
- "learning_rate": 1.7977593611696017e-05,
680
- "loss": 0.0242,
681
  "step": 960
682
  },
683
  {
684
- "epoch": 11.547619047619047,
685
- "grad_norm": 1.6954939365386963,
686
- "learning_rate": 1.791452188204031e-05,
687
- "loss": 0.0421,
688
  "step": 970
689
  },
690
  {
691
- "epoch": 11.666666666666666,
692
- "grad_norm": 1.3118443489074707,
693
- "learning_rate": 1.7850595957532394e-05,
694
- "loss": 0.0524,
695
  "step": 980
696
  },
697
  {
698
- "epoch": 11.785714285714286,
699
- "grad_norm": 0.5428506731987,
700
- "learning_rate": 1.778582273753984e-05,
701
- "loss": 0.0484,
702
  "step": 990
703
  },
704
  {
705
- "epoch": 11.904761904761905,
706
- "grad_norm": 0.4753515124320984,
707
- "learning_rate": 1.772020921287674e-05,
708
- "loss": 0.0347,
709
  "step": 1000
710
  }
711
  ],
712
  "logging_steps": 10,
713
- "max_steps": 3360,
714
  "num_input_tokens_seen": 0,
715
- "num_train_epochs": 40,
716
  "save_steps": 1000,
717
  "stateful_callbacks": {
718
  "TrainerControl": {
@@ -726,8 +726,8 @@
726
  "attributes": {}
727
  }
728
  },
729
- "total_flos": 5.965853086449664e+16,
730
- "train_batch_size": 3,
731
  "trial_name": null,
732
  "trial_params": null
733
  }
 
1
  {
2
  "best_metric": null,
3
  "best_model_checkpoint": null,
4
+ "epoch": 8.0,
5
  "eval_steps": 500,
6
  "global_step": 1000,
7
  "is_hyper_param_search": false,
 
9
  "is_world_process_zero": true,
10
  "log_history": [
11
  {
12
+ "epoch": 0.08,
13
+ "grad_norm": 6.874109268188477,
14
+ "learning_rate": 2.5559105431309904e-07,
15
+ "loss": 1.5574,
16
  "step": 10
17
  },
18
  {
19
+ "epoch": 0.16,
20
+ "grad_norm": 6.183030128479004,
21
+ "learning_rate": 8.306709265175719e-07,
22
+ "loss": 1.4724,
23
  "step": 20
24
  },
25
  {
26
+ "epoch": 0.24,
27
+ "grad_norm": 2.6565823554992676,
28
+ "learning_rate": 1.4696485623003196e-06,
29
+ "loss": 1.054,
30
  "step": 30
31
  },
32
  {
33
+ "epoch": 0.32,
34
+ "grad_norm": 3.227349042892456,
35
+ "learning_rate": 2.1086261980830672e-06,
36
+ "loss": 0.8732,
37
  "step": 40
38
  },
39
  {
40
+ "epoch": 0.4,
41
+ "grad_norm": 1.1049450635910034,
42
+ "learning_rate": 2.747603833865815e-06,
43
+ "loss": 0.7617,
44
  "step": 50
45
  },
46
  {
47
+ "epoch": 0.48,
48
+ "grad_norm": 1.8624663352966309,
49
+ "learning_rate": 3.386581469648563e-06,
50
+ "loss": 0.7144,
51
  "step": 60
52
  },
53
  {
54
+ "epoch": 0.56,
55
+ "grad_norm": 2.1665029525756836,
56
+ "learning_rate": 4.02555910543131e-06,
57
+ "loss": 0.7612,
58
  "step": 70
59
  },
60
  {
61
+ "epoch": 0.64,
62
+ "grad_norm": 1.0787376165390015,
63
+ "learning_rate": 4.664536741214058e-06,
64
+ "loss": 0.6864,
65
  "step": 80
66
  },
67
  {
68
+ "epoch": 0.72,
69
+ "grad_norm": 1.4465893507003784,
70
+ "learning_rate": 5.303514376996806e-06,
71
+ "loss": 0.6558,
72
  "step": 90
73
  },
74
  {
75
+ "epoch": 0.8,
76
+ "grad_norm": 1.1737278699874878,
77
+ "learning_rate": 5.8785942492012785e-06,
78
+ "loss": 0.6295,
79
  "step": 100
80
  },
81
  {
82
+ "epoch": 0.88,
83
+ "grad_norm": 1.7027487754821777,
84
+ "learning_rate": 6.517571884984026e-06,
85
+ "loss": 0.638,
86
  "step": 110
87
  },
88
  {
89
+ "epoch": 0.96,
90
+ "grad_norm": 1.5483449697494507,
91
+ "learning_rate": 7.156549520766773e-06,
92
+ "loss": 0.6173,
93
  "step": 120
94
  },
95
  {
96
+ "epoch": 1.04,
97
+ "grad_norm": 1.9829121828079224,
98
+ "learning_rate": 7.795527156549521e-06,
99
+ "loss": 0.5511,
100
  "step": 130
101
  },
102
  {
103
+ "epoch": 1.12,
104
+ "grad_norm": 1.466599464416504,
105
+ "learning_rate": 8.434504792332269e-06,
106
+ "loss": 0.5366,
107
  "step": 140
108
  },
109
  {
110
+ "epoch": 1.2,
111
+ "grad_norm": 1.5559449195861816,
112
+ "learning_rate": 9.073482428115017e-06,
113
+ "loss": 0.5662,
114
  "step": 150
115
  },
116
  {
117
+ "epoch": 1.28,
118
+ "grad_norm": 1.33512282371521,
119
+ "learning_rate": 9.712460063897765e-06,
120
+ "loss": 0.5567,
121
  "step": 160
122
  },
123
  {
124
+ "epoch": 1.3599999999999999,
125
+ "grad_norm": 0.943450927734375,
126
+ "learning_rate": 1.035143769968051e-05,
127
+ "loss": 0.5997,
128
  "step": 170
129
  },
130
  {
131
+ "epoch": 1.44,
132
+ "grad_norm": 1.2756850719451904,
133
+ "learning_rate": 1.099041533546326e-05,
134
+ "loss": 0.5587,
135
  "step": 180
136
  },
137
  {
138
+ "epoch": 1.52,
139
+ "grad_norm": 1.0768604278564453,
140
+ "learning_rate": 1.1629392971246008e-05,
141
+ "loss": 0.6288,
142
  "step": 190
143
  },
144
  {
145
+ "epoch": 1.6,
146
+ "grad_norm": 1.082158088684082,
147
+ "learning_rate": 1.2268370607028754e-05,
148
+ "loss": 0.5475,
149
  "step": 200
150
  },
151
  {
152
+ "epoch": 1.6800000000000002,
153
+ "grad_norm": 1.4872550964355469,
154
+ "learning_rate": 1.2907348242811502e-05,
155
+ "loss": 0.6002,
156
  "step": 210
157
  },
158
  {
159
+ "epoch": 1.76,
160
+ "grad_norm": 2.0925095081329346,
161
+ "learning_rate": 1.3546325878594251e-05,
162
+ "loss": 0.5887,
163
  "step": 220
164
  },
165
  {
166
+ "epoch": 1.8399999999999999,
167
+ "grad_norm": 1.4571633338928223,
168
+ "learning_rate": 1.4185303514376998e-05,
169
+ "loss": 0.6107,
170
  "step": 230
171
  },
172
  {
173
+ "epoch": 1.92,
174
+ "grad_norm": 2.4871673583984375,
175
+ "learning_rate": 1.4824281150159745e-05,
176
+ "loss": 0.5582,
177
  "step": 240
178
  },
179
  {
180
+ "epoch": 2.0,
181
+ "grad_norm": 1.3184454441070557,
182
+ "learning_rate": 1.5463258785942495e-05,
183
+ "loss": 0.582,
184
  "step": 250
185
  },
186
  {
187
+ "epoch": 2.08,
188
+ "grad_norm": 1.9747495651245117,
189
+ "learning_rate": 1.610223642172524e-05,
190
+ "loss": 0.4687,
191
  "step": 260
192
  },
193
  {
194
+ "epoch": 2.16,
195
+ "grad_norm": 1.7691447734832764,
196
+ "learning_rate": 1.6741214057507987e-05,
197
+ "loss": 0.5013,
198
  "step": 270
199
  },
200
  {
201
+ "epoch": 2.24,
202
+ "grad_norm": 1.8902829885482788,
203
+ "learning_rate": 1.7380191693290737e-05,
204
+ "loss": 0.5008,
205
  "step": 280
206
  },
207
  {
208
+ "epoch": 2.32,
209
+ "grad_norm": 1.3197141885757446,
210
+ "learning_rate": 1.8019169329073486e-05,
211
+ "loss": 0.5229,
212
  "step": 290
213
  },
214
  {
215
+ "epoch": 2.4,
216
+ "grad_norm": 2.325404167175293,
217
+ "learning_rate": 1.8658146964856232e-05,
218
+ "loss": 0.4975,
219
  "step": 300
220
  },
221
  {
222
+ "epoch": 2.48,
223
+ "grad_norm": 1.3323551416397095,
224
+ "learning_rate": 1.929712460063898e-05,
225
+ "loss": 0.5254,
226
  "step": 310
227
  },
228
  {
229
+ "epoch": 2.56,
230
+ "grad_norm": 0.7770269513130188,
231
+ "learning_rate": 1.9936102236421725e-05,
232
+ "loss": 0.5304,
233
  "step": 320
234
  },
235
  {
236
+ "epoch": 2.64,
237
+ "grad_norm": 1.887294054031372,
238
+ "learning_rate": 1.999949450079496e-05,
239
+ "loss": 0.4755,
240
  "step": 330
241
  },
242
  {
243
+ "epoch": 2.7199999999999998,
244
+ "grad_norm": 1.2255381345748901,
245
+ "learning_rate": 1.9997747161747696e-05,
246
+ "loss": 0.4982,
247
  "step": 340
248
  },
249
  {
250
+ "epoch": 2.8,
251
+ "grad_norm": 1.6865767240524292,
252
+ "learning_rate": 1.9994751960168383e-05,
253
+ "loss": 0.4857,
254
  "step": 350
255
  },
256
  {
257
+ "epoch": 2.88,
258
+ "grad_norm": 1.2997519969940186,
259
+ "learning_rate": 1.999050926990122e-05,
260
+ "loss": 0.4987,
261
  "step": 360
262
  },
263
  {
264
+ "epoch": 2.96,
265
+ "grad_norm": 1.4636356830596924,
266
+ "learning_rate": 1.9985019620494935e-05,
267
+ "loss": 0.4947,
268
  "step": 370
269
  },
270
  {
271
+ "epoch": 3.04,
272
+ "grad_norm": 1.3720135688781738,
273
+ "learning_rate": 1.9978283697136662e-05,
274
+ "loss": 0.5135,
275
  "step": 380
276
  },
277
  {
278
+ "epoch": 3.12,
279
+ "grad_norm": 1.6108952760696411,
280
+ "learning_rate": 1.997030234056645e-05,
281
+ "loss": 0.4405,
282
  "step": 390
283
  },
284
  {
285
+ "epoch": 3.2,
286
+ "grad_norm": 2.1790759563446045,
287
+ "learning_rate": 1.9961076546972304e-05,
288
+ "loss": 0.4468,
289
  "step": 400
290
  },
291
  {
292
+ "epoch": 3.2800000000000002,
293
+ "grad_norm": 4.406925678253174,
294
+ "learning_rate": 1.9950607467865856e-05,
295
+ "loss": 0.4675,
296
  "step": 410
297
  },
298
  {
299
+ "epoch": 3.36,
300
+ "grad_norm": 1.7346851825714111,
301
+ "learning_rate": 1.993889640993864e-05,
302
+ "loss": 0.4243,
303
  "step": 420
304
  },
305
  {
306
+ "epoch": 3.44,
307
+ "grad_norm": 1.2865455150604248,
308
+ "learning_rate": 1.9925944834898992e-05,
309
+ "loss": 0.3787,
310
  "step": 430
311
  },
312
  {
313
+ "epoch": 3.52,
314
+ "grad_norm": 1.321824550628662,
315
+ "learning_rate": 1.991175435928962e-05,
316
+ "loss": 0.3982,
317
  "step": 440
318
  },
319
  {
320
+ "epoch": 3.6,
321
+ "grad_norm": 2.45711350440979,
322
+ "learning_rate": 1.989632675428582e-05,
323
+ "loss": 0.4811,
324
  "step": 450
325
  },
326
  {
327
+ "epoch": 3.68,
328
+ "grad_norm": 2.5552310943603516,
329
+ "learning_rate": 1.9879663945474416e-05,
330
+ "loss": 0.4255,
331
  "step": 460
332
  },
333
  {
334
+ "epoch": 3.76,
335
+ "grad_norm": 1.763244867324829,
336
+ "learning_rate": 1.9861768012613435e-05,
337
+ "loss": 0.4579,
338
  "step": 470
339
  },
340
  {
341
+ "epoch": 3.84,
342
+ "grad_norm": 1.0951169729232788,
343
+ "learning_rate": 1.984264118937249e-05,
344
+ "loss": 0.398,
345
  "step": 480
346
  },
347
  {
348
+ "epoch": 3.92,
349
+ "grad_norm": 8.235908508300781,
350
+ "learning_rate": 1.9822285863054e-05,
351
+ "loss": 0.4101,
352
  "step": 490
353
  },
354
  {
355
+ "epoch": 4.0,
356
+ "grad_norm": 19.39371681213379,
357
+ "learning_rate": 1.9800704574295246e-05,
358
+ "loss": 0.4645,
359
  "step": 500
360
  },
361
  {
362
+ "epoch": 4.08,
363
+ "grad_norm": 1.2594205141067505,
364
+ "learning_rate": 1.9777900016751224e-05,
365
+ "loss": 0.3632,
366
  "step": 510
367
  },
368
  {
369
+ "epoch": 4.16,
370
+ "grad_norm": 1.7697833776474,
371
+ "learning_rate": 1.9753875036758464e-05,
372
+ "loss": 0.3356,
373
  "step": 520
374
  },
375
  {
376
+ "epoch": 4.24,
377
+ "grad_norm": 1.66884446144104,
378
+ "learning_rate": 1.9728632632979746e-05,
379
+ "loss": 0.382,
380
  "step": 530
381
  },
382
  {
383
+ "epoch": 4.32,
384
+ "grad_norm": 1.7648805379867554,
385
+ "learning_rate": 1.970217595602985e-05,
386
+ "loss": 0.3997,
387
  "step": 540
388
  },
389
  {
390
+ "epoch": 4.4,
391
+ "grad_norm": 1.6132514476776123,
392
+ "learning_rate": 1.967450830808228e-05,
393
+ "loss": 0.3476,
394
  "step": 550
395
  },
396
  {
397
+ "epoch": 4.48,
398
+ "grad_norm": 1.3217155933380127,
399
+ "learning_rate": 1.9645633142457143e-05,
400
+ "loss": 0.3362,
401
  "step": 560
402
  },
403
  {
404
+ "epoch": 4.5600000000000005,
405
+ "grad_norm": 1.5668702125549316,
406
+ "learning_rate": 1.9615554063190098e-05,
407
+ "loss": 0.3996,
408
  "step": 570
409
  },
410
  {
411
+ "epoch": 4.64,
412
+ "grad_norm": 2.051454544067383,
413
+ "learning_rate": 1.958427482458253e-05,
414
+ "loss": 0.4104,
415
  "step": 580
416
  },
417
  {
418
+ "epoch": 4.72,
419
+ "grad_norm": 1.84868586063385,
420
+ "learning_rate": 1.9551799330732954e-05,
421
+ "loss": 0.374,
422
  "step": 590
423
  },
424
  {
425
+ "epoch": 4.8,
426
+ "grad_norm": 1.3498121500015259,
427
+ "learning_rate": 1.9518131635049745e-05,
428
+ "loss": 0.3444,
429
  "step": 600
430
  },
431
  {
432
+ "epoch": 4.88,
433
+ "grad_norm": 1.962553858757019,
434
+ "learning_rate": 1.9483275939745184e-05,
435
+ "loss": 0.3153,
436
  "step": 610
437
  },
438
  {
439
+ "epoch": 4.96,
440
+ "grad_norm": 1.7132457494735718,
441
+ "learning_rate": 1.944723659531099e-05,
442
+ "loss": 0.383,
443
  "step": 620
444
  },
445
  {
446
+ "epoch": 5.04,
447
+ "grad_norm": 1.8860396146774292,
448
+ "learning_rate": 1.9410018099975297e-05,
449
+ "loss": 0.327,
450
  "step": 630
451
  },
452
  {
453
+ "epoch": 5.12,
454
+ "grad_norm": 1.9154820442199707,
455
+ "learning_rate": 1.9371625099141223e-05,
456
+ "loss": 0.2704,
457
  "step": 640
458
  },
459
  {
460
+ "epoch": 5.2,
461
+ "grad_norm": 1.4077069759368896,
462
+ "learning_rate": 1.9332062384807058e-05,
463
+ "loss": 0.2659,
464
  "step": 650
465
  },
466
  {
467
+ "epoch": 5.28,
468
+ "grad_norm": 1.8489642143249512,
469
+ "learning_rate": 1.9291334894968133e-05,
470
+ "loss": 0.2523,
471
  "step": 660
472
  },
473
  {
474
+ "epoch": 5.36,
475
+ "grad_norm": 2.3917829990386963,
476
+ "learning_rate": 1.9249447713000515e-05,
477
+ "loss": 0.2885,
478
  "step": 670
479
  },
480
  {
481
+ "epoch": 5.44,
482
+ "grad_norm": 2.5429599285125732,
483
+ "learning_rate": 1.9206406067026506e-05,
484
+ "loss": 0.3097,
485
  "step": 680
486
  },
487
  {
488
+ "epoch": 5.52,
489
+ "grad_norm": 3.4684863090515137,
490
+ "learning_rate": 1.9162215329262115e-05,
491
+ "loss": 0.2868,
492
  "step": 690
493
  },
494
  {
495
+ "epoch": 5.6,
496
+ "grad_norm": 2.662013292312622,
497
+ "learning_rate": 1.9116881015346517e-05,
498
+ "loss": 0.3281,
499
  "step": 700
500
  },
501
  {
502
+ "epoch": 5.68,
503
+ "grad_norm": 2.223334312438965,
504
+ "learning_rate": 1.9070408783653627e-05,
505
+ "loss": 0.2799,
506
  "step": 710
507
  },
508
  {
509
+ "epoch": 5.76,
510
+ "grad_norm": 2.992532730102539,
511
+ "learning_rate": 1.9022804434585854e-05,
512
+ "loss": 0.3252,
513
  "step": 720
514
  },
515
  {
516
+ "epoch": 5.84,
517
+ "grad_norm": 1.6079481840133667,
518
+ "learning_rate": 1.8974073909850125e-05,
519
+ "loss": 0.2791,
520
  "step": 730
521
  },
522
  {
523
+ "epoch": 5.92,
524
+ "grad_norm": 2.97104549407959,
525
+ "learning_rate": 1.8924223291716274e-05,
526
+ "loss": 0.3029,
527
  "step": 740
528
  },
529
  {
530
+ "epoch": 6.0,
531
+ "grad_norm": 2.1889212131500244,
532
+ "learning_rate": 1.887325880225789e-05,
533
+ "loss": 0.293,
534
  "step": 750
535
  },
536
  {
537
+ "epoch": 6.08,
538
+ "grad_norm": 2.123610019683838,
539
+ "learning_rate": 1.882118680257572e-05,
540
+ "loss": 0.2308,
541
  "step": 760
542
  },
543
  {
544
+ "epoch": 6.16,
545
+ "grad_norm": 3.71563720703125,
546
+ "learning_rate": 1.8768013792003683e-05,
547
+ "loss": 0.216,
548
  "step": 770
549
  },
550
  {
551
+ "epoch": 6.24,
552
+ "grad_norm": 2.459091901779175,
553
+ "learning_rate": 1.8713746407297703e-05,
554
+ "loss": 0.1972,
555
  "step": 780
556
  },
557
  {
558
+ "epoch": 6.32,
559
+ "grad_norm": 2.669945240020752,
560
+ "learning_rate": 1.8658391421807313e-05,
561
+ "loss": 0.2478,
562
  "step": 790
563
  },
564
  {
565
+ "epoch": 6.4,
566
+ "grad_norm": 2.1581931114196777,
567
+ "learning_rate": 1.8601955744630255e-05,
568
+ "loss": 0.2295,
569
  "step": 800
570
  },
571
  {
572
+ "epoch": 6.48,
573
+ "grad_norm": 2.567908763885498,
574
+ "learning_rate": 1.8544446419750125e-05,
575
+ "loss": 0.2239,
576
  "step": 810
577
  },
578
  {
579
+ "epoch": 6.5600000000000005,
580
+ "grad_norm": 2.0205740928649902,
581
+ "learning_rate": 1.8485870625157186e-05,
582
+ "loss": 0.216,
583
  "step": 820
584
  },
585
  {
586
+ "epoch": 6.64,
587
+ "grad_norm": 2.2898030281066895,
588
+ "learning_rate": 1.8426235671952452e-05,
589
+ "loss": 0.2343,
590
  "step": 830
591
  },
592
  {
593
+ "epoch": 6.72,
594
+ "grad_norm": 2.2003941535949707,
595
+ "learning_rate": 1.836554900343514e-05,
596
+ "loss": 0.2076,
597
  "step": 840
598
  },
599
  {
600
+ "epoch": 6.8,
601
+ "grad_norm": 1.5117710828781128,
602
+ "learning_rate": 1.8303818194173665e-05,
603
+ "loss": 0.1904,
604
  "step": 850
605
  },
606
  {
607
+ "epoch": 6.88,
608
+ "grad_norm": 1.2647337913513184,
609
+ "learning_rate": 1.824105094906021e-05,
610
+ "loss": 0.2341,
611
  "step": 860
612
  },
613
  {
614
+ "epoch": 6.96,
615
+ "grad_norm": 2.00254225730896,
616
+ "learning_rate": 1.8177255102349047e-05,
617
+ "loss": 0.2288,
618
  "step": 870
619
  },
620
  {
621
+ "epoch": 7.04,
622
+ "grad_norm": 2.516129493713379,
623
+ "learning_rate": 1.8112438616678712e-05,
624
+ "loss": 0.1685,
625
  "step": 880
626
  },
627
  {
628
+ "epoch": 7.12,
629
+ "grad_norm": 3.6129150390625,
630
+ "learning_rate": 1.8046609582078147e-05,
631
+ "loss": 0.1625,
632
  "step": 890
633
  },
634
  {
635
+ "epoch": 7.2,
636
+ "grad_norm": 3.1554768085479736,
637
+ "learning_rate": 1.797977621495696e-05,
638
+ "loss": 0.2011,
639
  "step": 900
640
  },
641
  {
642
+ "epoch": 7.28,
643
+ "grad_norm": 3.8677282333374023,
644
+ "learning_rate": 1.7911946857079886e-05,
645
+ "loss": 0.1579,
646
  "step": 910
647
  },
648
  {
649
+ "epoch": 7.36,
650
+ "grad_norm": 3.062568426132202,
651
+ "learning_rate": 1.784312997452562e-05,
652
+ "loss": 0.1705,
653
  "step": 920
654
  },
655
  {
656
+ "epoch": 7.44,
657
+ "grad_norm": 1.8281478881835938,
658
+ "learning_rate": 1.777333415663014e-05,
659
+ "loss": 0.1708,
660
  "step": 930
661
  },
662
  {
663
+ "epoch": 7.52,
664
+ "grad_norm": 1.1312201023101807,
665
+ "learning_rate": 1.7702568114914607e-05,
666
+ "loss": 0.1644,
667
  "step": 940
668
  },
669
  {
670
+ "epoch": 7.6,
671
+ "grad_norm": 1.1482549905776978,
672
+ "learning_rate": 1.7630840681998068e-05,
673
+ "loss": 0.1297,
674
  "step": 950
675
  },
676
  {
677
+ "epoch": 7.68,
678
+ "grad_norm": 3.109949827194214,
679
+ "learning_rate": 1.755816081049501e-05,
680
+ "loss": 0.1812,
681
  "step": 960
682
  },
683
  {
684
+ "epoch": 7.76,
685
+ "grad_norm": 1.2507814168930054,
686
+ "learning_rate": 1.7484537571897943e-05,
687
+ "loss": 0.1331,
688
  "step": 970
689
  },
690
  {
691
+ "epoch": 7.84,
692
+ "grad_norm": 2.898353338241577,
693
+ "learning_rate": 1.740998015544514e-05,
694
+ "loss": 0.1625,
695
  "step": 980
696
  },
697
  {
698
+ "epoch": 7.92,
699
+ "grad_norm": 2.2018096446990967,
700
+ "learning_rate": 1.7334497866973716e-05,
701
+ "loss": 0.1441,
702
  "step": 990
703
  },
704
  {
705
+ "epoch": 8.0,
706
+ "grad_norm": 2.1120376586914062,
707
+ "learning_rate": 1.725810012775808e-05,
708
+ "loss": 0.1804,
709
  "step": 1000
710
  }
711
  ],
712
  "logging_steps": 10,
713
+ "max_steps": 3125,
714
  "num_input_tokens_seen": 0,
715
+ "num_train_epochs": 25,
716
  "save_steps": 1000,
717
  "stateful_callbacks": {
718
  "TrainerControl": {
 
726
  "attributes": {}
727
  }
728
  },
729
+ "total_flos": 3.989316595980698e+16,
730
+ "train_batch_size": 2,
731
  "trial_name": null,
732
  "trial_params": null
733
  }
checkpoint-1000/training_args.bin CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:fe16ad85cf67a1d616e8a664749a043a7e42845bd6dfe8881e3e3c947d23497a
3
  size 5624
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:27ff69d62d45c0f05227ab8bfdfb70f5de878e29a775a5568fe676d60d677a54
3
  size 5624
checkpoint-2000/adapter_config.json CHANGED
@@ -19,7 +19,7 @@
19
  "r": 16,
20
  "rank_pattern": {},
21
  "revision": null,
22
- "target_modules": "^(?!.*visual.patch_embed|visual.blocks).*(?:o_proj|v_proj|down_proj|k_proj|q_proj|up_proj|gate_proj).*",
23
  "task_type": "CAUSAL_LM",
24
  "use_dora": false,
25
  "use_rslora": false
 
19
  "r": 16,
20
  "rank_pattern": {},
21
  "revision": null,
22
+ "target_modules": "^(?!.*patch_embed).*(?:o_proj|fc2|v_proj|down_proj|q_proj|up_proj|proj|qkv|gate_proj|k_proj|fc1).*",
23
  "task_type": "CAUSAL_LM",
24
  "use_dora": false,
25
  "use_rslora": false
checkpoint-2000/adapter_model.safetensors CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:03928b0ecd6b0549b45181cc9efb4fbf4040792fdf1f257edf0b7c21fbc2ee40
3
- size 73911112
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:d0ccf2b7b362ab3db16adf2548b738315e048bc9865fdd613c3b8dbf8e81e3eb
3
+ size 115886968
checkpoint-2000/optimizer.pt CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:95a36ccd6117677e5f04e807f62ecb626c9dca12a1beaf724fa9f43cc788efac
3
- size 148047978
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:fa97a1aea3dde7b3cf80d8d086b0ac5f69b309b85418d33438ec8ebc63f3c26f
3
+ size 232150634
checkpoint-2000/rng_state_0.pth CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:23c5c1a139c10fe0906308288b9990e59b0830f85f1180b142012af4957b1ba2
3
  size 14512
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:9281be04f2316562db36bd289b226e07237cabe0e64c821e5b8cfedcb5f17669
3
  size 14512
checkpoint-2000/rng_state_1.pth CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:e02bd5e21ab993cf26deb90360061cfc27b8440d255a29134ff03a42f4f9abff
3
  size 14512
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:658f78ca761388a441a3eaf439ede5c9b4dc1f731762e906f81abe29d7755ff1
3
  size 14512
checkpoint-2000/scheduler.pt CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:849e569d25a585ac5f4565034971d52ce26f78183bf5984dd969f75ddb21fb9b
3
  size 1064
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:05f5842cbd1f9aa3089485ac6c5b4ff67fddae1f9025135ab08c01df6214b761
3
  size 1064
checkpoint-2000/trainer_state.json CHANGED
@@ -1,7 +1,7 @@
1
  {
2
  "best_metric": null,
3
  "best_model_checkpoint": null,
4
- "epoch": 23.80952380952381,
5
  "eval_steps": 500,
6
  "global_step": 2000,
7
  "is_hyper_param_search": false,
@@ -9,1410 +9,1410 @@
9
  "is_world_process_zero": true,
10
  "log_history": [
11
  {
12
- "epoch": 0.11904761904761904,
13
- "grad_norm": 5.44253396987915,
14
- "learning_rate": 5.952380952380953e-07,
15
- "loss": 1.5321,
16
  "step": 10
17
  },
18
  {
19
- "epoch": 0.23809523809523808,
20
- "grad_norm": 2.2140841484069824,
21
- "learning_rate": 1.1904761904761906e-06,
22
- "loss": 1.1934,
23
  "step": 20
24
  },
25
  {
26
- "epoch": 0.35714285714285715,
27
- "grad_norm": 1.413270115852356,
28
- "learning_rate": 1.7857142857142859e-06,
29
- "loss": 0.9128,
30
  "step": 30
31
  },
32
  {
33
- "epoch": 0.47619047619047616,
34
- "grad_norm": 1.0657840967178345,
35
- "learning_rate": 2.380952380952381e-06,
36
- "loss": 0.7772,
37
  "step": 40
38
  },
39
  {
40
- "epoch": 0.5952380952380952,
41
- "grad_norm": 0.9593137502670288,
42
- "learning_rate": 2.9761904761904763e-06,
43
- "loss": 0.7849,
44
  "step": 50
45
  },
46
  {
47
- "epoch": 0.7142857142857143,
48
- "grad_norm": 1.134382724761963,
49
- "learning_rate": 3.5714285714285718e-06,
50
- "loss": 0.716,
51
  "step": 60
52
  },
53
  {
54
- "epoch": 0.8333333333333334,
55
- "grad_norm": 0.618364691734314,
56
- "learning_rate": 4.166666666666667e-06,
57
- "loss": 0.6678,
58
  "step": 70
59
  },
60
  {
61
- "epoch": 0.9523809523809523,
62
- "grad_norm": 0.7465847134590149,
63
- "learning_rate": 4.761904761904762e-06,
64
- "loss": 0.671,
65
  "step": 80
66
  },
67
  {
68
- "epoch": 1.0714285714285714,
69
- "grad_norm": 0.6726512908935547,
70
- "learning_rate": 5.357142857142857e-06,
71
- "loss": 0.6186,
72
  "step": 90
73
  },
74
  {
75
- "epoch": 1.1904761904761905,
76
- "grad_norm": 0.6353351473808289,
77
- "learning_rate": 5.9523809523809525e-06,
78
- "loss": 0.5818,
79
  "step": 100
80
  },
81
  {
82
- "epoch": 1.3095238095238095,
83
- "grad_norm": 0.7020516991615295,
84
- "learning_rate": 6.547619047619048e-06,
85
- "loss": 0.6056,
86
  "step": 110
87
  },
88
  {
89
- "epoch": 1.4285714285714286,
90
- "grad_norm": 1.1165708303451538,
91
- "learning_rate": 7.1428571428571436e-06,
92
- "loss": 0.6225,
93
  "step": 120
94
  },
95
  {
96
- "epoch": 1.5476190476190477,
97
- "grad_norm": 1.7502918243408203,
98
- "learning_rate": 7.738095238095238e-06,
99
- "loss": 0.6321,
100
  "step": 130
101
  },
102
  {
103
- "epoch": 1.6666666666666665,
104
- "grad_norm": 0.6962683796882629,
105
- "learning_rate": 8.333333333333334e-06,
106
- "loss": 0.5751,
107
  "step": 140
108
  },
109
  {
110
- "epoch": 1.7857142857142856,
111
- "grad_norm": 0.876686155796051,
112
- "learning_rate": 8.92857142857143e-06,
113
- "loss": 0.5974,
114
  "step": 150
115
  },
116
  {
117
- "epoch": 1.9047619047619047,
118
- "grad_norm": 0.5832173228263855,
119
- "learning_rate": 9.523809523809525e-06,
120
- "loss": 0.6085,
121
  "step": 160
122
  },
123
  {
124
- "epoch": 2.0238095238095237,
125
- "grad_norm": 0.44337984919548035,
126
- "learning_rate": 1.011904761904762e-05,
127
- "loss": 0.5713,
128
  "step": 170
129
  },
130
  {
131
- "epoch": 2.142857142857143,
132
- "grad_norm": 1.2403652667999268,
133
- "learning_rate": 1.0714285714285714e-05,
134
- "loss": 0.5165,
135
  "step": 180
136
  },
137
  {
138
- "epoch": 2.261904761904762,
139
- "grad_norm": 0.7495181560516357,
140
- "learning_rate": 1.130952380952381e-05,
141
- "loss": 0.5448,
142
  "step": 190
143
  },
144
  {
145
- "epoch": 2.380952380952381,
146
- "grad_norm": 0.8501800298690796,
147
- "learning_rate": 1.1904761904761905e-05,
148
- "loss": 0.5122,
149
  "step": 200
150
  },
151
  {
152
- "epoch": 2.5,
153
- "grad_norm": 0.8133195638656616,
154
- "learning_rate": 1.25e-05,
155
- "loss": 0.5337,
156
  "step": 210
157
  },
158
  {
159
- "epoch": 2.619047619047619,
160
- "grad_norm": 1.3526533842086792,
161
- "learning_rate": 1.3095238095238096e-05,
162
- "loss": 0.4943,
163
  "step": 220
164
  },
165
  {
166
- "epoch": 2.738095238095238,
167
- "grad_norm": 1.0535343885421753,
168
- "learning_rate": 1.3690476190476192e-05,
169
- "loss": 0.521,
170
  "step": 230
171
  },
172
  {
173
- "epoch": 2.857142857142857,
174
- "grad_norm": 1.2215440273284912,
175
- "learning_rate": 1.4285714285714287e-05,
176
- "loss": 0.5461,
177
  "step": 240
178
  },
179
  {
180
- "epoch": 2.9761904761904763,
181
- "grad_norm": 1.1781115531921387,
182
- "learning_rate": 1.4880952380952383e-05,
183
- "loss": 0.5346,
184
  "step": 250
185
  },
186
  {
187
- "epoch": 3.0952380952380953,
188
- "grad_norm": 0.762434184551239,
189
- "learning_rate": 1.5476190476190476e-05,
190
- "loss": 0.4855,
191
  "step": 260
192
  },
193
  {
194
- "epoch": 3.2142857142857144,
195
- "grad_norm": 1.287593126296997,
196
- "learning_rate": 1.6071428571428572e-05,
197
- "loss": 0.473,
198
  "step": 270
199
  },
200
  {
201
- "epoch": 3.3333333333333335,
202
- "grad_norm": 0.9310547113418579,
203
- "learning_rate": 1.6666666666666667e-05,
204
- "loss": 0.4201,
205
  "step": 280
206
  },
207
  {
208
- "epoch": 3.4523809523809526,
209
- "grad_norm": 1.1341677904129028,
210
- "learning_rate": 1.7261904761904763e-05,
211
- "loss": 0.3884,
212
  "step": 290
213
  },
214
  {
215
- "epoch": 3.571428571428571,
216
- "grad_norm": 1.483049750328064,
217
- "learning_rate": 1.785714285714286e-05,
218
- "loss": 0.4662,
219
  "step": 300
220
  },
221
  {
222
- "epoch": 3.6904761904761907,
223
- "grad_norm": 0.9312372803688049,
224
- "learning_rate": 1.8452380952380954e-05,
225
- "loss": 0.4688,
226
  "step": 310
227
  },
228
  {
229
- "epoch": 3.8095238095238093,
230
- "grad_norm": 1.0440223217010498,
231
- "learning_rate": 1.904761904761905e-05,
232
- "loss": 0.4727,
233
  "step": 320
234
  },
235
  {
236
- "epoch": 3.928571428571429,
237
- "grad_norm": 1.20750892162323,
238
- "learning_rate": 1.9642857142857145e-05,
239
- "loss": 0.4062,
240
  "step": 330
241
  },
242
  {
243
- "epoch": 4.0476190476190474,
244
- "grad_norm": 1.5511012077331543,
245
- "learning_rate": 1.999991365731819e-05,
246
- "loss": 0.4459,
247
  "step": 340
248
  },
249
  {
250
- "epoch": 4.166666666666667,
251
- "grad_norm": 1.3020656108856201,
252
- "learning_rate": 1.9998942319271076e-05,
253
- "loss": 0.3667,
254
  "step": 350
255
  },
256
  {
257
- "epoch": 4.285714285714286,
258
- "grad_norm": 0.9357143640518188,
259
- "learning_rate": 1.9996891820008165e-05,
260
- "loss": 0.3703,
261
  "step": 360
262
  },
263
  {
264
- "epoch": 4.404761904761905,
265
- "grad_norm": 1.120471477508545,
266
- "learning_rate": 1.9993762380834786e-05,
267
- "loss": 0.3797,
268
  "step": 370
269
  },
270
  {
271
- "epoch": 4.523809523809524,
272
- "grad_norm": 1.4476983547210693,
273
- "learning_rate": 1.9989554339503612e-05,
274
- "loss": 0.366,
275
  "step": 380
276
  },
277
  {
278
- "epoch": 4.642857142857143,
279
- "grad_norm": 0.8721790313720703,
280
- "learning_rate": 1.998426815017817e-05,
281
- "loss": 0.3914,
282
  "step": 390
283
  },
284
  {
285
- "epoch": 4.761904761904762,
286
- "grad_norm": 1.4711418151855469,
287
- "learning_rate": 1.997790438338385e-05,
288
- "loss": 0.3665,
289
  "step": 400
290
  },
291
  {
292
- "epoch": 4.880952380952381,
293
- "grad_norm": 1.216174840927124,
294
- "learning_rate": 1.9970463725946335e-05,
295
- "loss": 0.345,
296
  "step": 410
297
  },
298
  {
299
- "epoch": 5.0,
300
- "grad_norm": 0.8730125427246094,
301
- "learning_rate": 1.9962847054074817e-05,
302
- "loss": 0.3732,
303
  "step": 420
304
  },
305
  {
306
- "epoch": 5.119047619047619,
307
- "grad_norm": 1.7755807638168335,
308
- "learning_rate": 1.9953362612814294e-05,
309
- "loss": 0.2578,
310
  "step": 430
311
  },
312
  {
313
- "epoch": 5.238095238095238,
314
- "grad_norm": 1.7037990093231201,
315
- "learning_rate": 1.9942803929643423e-05,
316
- "loss": 0.2561,
317
  "step": 440
318
  },
319
  {
320
- "epoch": 5.357142857142857,
321
- "grad_norm": 1.140688180923462,
322
- "learning_rate": 1.9931172144134874e-05,
323
- "loss": 0.2525,
324
  "step": 450
325
  },
326
  {
327
- "epoch": 5.476190476190476,
328
- "grad_norm": 1.492291808128357,
329
- "learning_rate": 1.99184685116786e-05,
330
- "loss": 0.2727,
331
  "step": 460
332
  },
333
  {
334
- "epoch": 5.595238095238095,
335
- "grad_norm": 1.1798468828201294,
336
- "learning_rate": 1.9904694403346362e-05,
337
- "loss": 0.291,
338
  "step": 470
339
  },
340
  {
341
- "epoch": 5.714285714285714,
342
- "grad_norm": 1.1415112018585205,
343
- "learning_rate": 1.9889851305743742e-05,
344
- "loss": 0.2845,
345
  "step": 480
346
  },
347
  {
348
- "epoch": 5.833333333333333,
349
- "grad_norm": 1.8219681978225708,
350
- "learning_rate": 1.9873940820849714e-05,
351
- "loss": 0.2746,
352
  "step": 490
353
  },
354
  {
355
- "epoch": 5.9523809523809526,
356
- "grad_norm": 1.3858968019485474,
357
- "learning_rate": 1.9856964665843717e-05,
358
- "loss": 0.2861,
359
  "step": 500
360
  },
361
  {
362
- "epoch": 6.071428571428571,
363
- "grad_norm": 1.2878497838974,
364
- "learning_rate": 1.9838924672920344e-05,
365
- "loss": 0.239,
366
  "step": 510
367
  },
368
  {
369
- "epoch": 6.190476190476191,
370
- "grad_norm": 2.5550217628479004,
371
- "learning_rate": 1.9819822789091597e-05,
372
- "loss": 0.1605,
373
  "step": 520
374
  },
375
  {
376
- "epoch": 6.309523809523809,
377
- "grad_norm": 2.016493082046509,
378
- "learning_rate": 1.9799661075976754e-05,
379
- "loss": 0.2069,
380
  "step": 530
381
  },
382
  {
383
- "epoch": 6.428571428571429,
384
- "grad_norm": 1.8905669450759888,
385
- "learning_rate": 1.977844170957984e-05,
386
- "loss": 0.1953,
387
  "step": 540
388
  },
389
  {
390
- "epoch": 6.5476190476190474,
391
- "grad_norm": 1.6992967128753662,
392
- "learning_rate": 1.9756166980054812e-05,
393
- "loss": 0.1837,
394
  "step": 550
395
  },
396
  {
397
- "epoch": 6.666666666666667,
398
- "grad_norm": 1.893633246421814,
399
- "learning_rate": 1.973283929145836e-05,
400
- "loss": 0.1947,
401
  "step": 560
402
  },
403
  {
404
- "epoch": 6.785714285714286,
405
- "grad_norm": 1.6204354763031006,
406
- "learning_rate": 1.9708461161490456e-05,
407
- "loss": 0.1836,
408
  "step": 570
409
  },
410
  {
411
- "epoch": 6.904761904761905,
412
- "grad_norm": 2.0902771949768066,
413
- "learning_rate": 1.9683035221222617e-05,
414
- "loss": 0.1971,
415
  "step": 580
416
  },
417
  {
418
- "epoch": 7.023809523809524,
419
- "grad_norm": 2.2528581619262695,
420
- "learning_rate": 1.9656564214813948e-05,
421
- "loss": 0.1864,
422
  "step": 590
423
  },
424
  {
425
- "epoch": 7.142857142857143,
426
- "grad_norm": 0.9720670580863953,
427
- "learning_rate": 1.9629050999214977e-05,
428
- "loss": 0.114,
429
  "step": 600
430
  },
431
  {
432
- "epoch": 7.261904761904762,
433
- "grad_norm": 1.1338551044464111,
434
- "learning_rate": 1.960049854385929e-05,
435
- "loss": 0.1331,
436
  "step": 610
437
  },
438
  {
439
- "epoch": 7.380952380952381,
440
- "grad_norm": 1.8178822994232178,
441
- "learning_rate": 1.957090993034306e-05,
442
- "loss": 0.1144,
443
  "step": 620
444
  },
445
  {
446
- "epoch": 7.5,
447
- "grad_norm": 2.2525339126586914,
448
- "learning_rate": 1.9540288352092474e-05,
449
- "loss": 0.1298,
450
  "step": 630
451
  },
452
  {
453
- "epoch": 7.619047619047619,
454
- "grad_norm": 2.6778416633605957,
455
- "learning_rate": 1.9508637114019037e-05,
456
- "loss": 0.131,
457
  "step": 640
458
  },
459
  {
460
- "epoch": 7.738095238095238,
461
- "grad_norm": 1.2893160581588745,
462
- "learning_rate": 1.947595963216292e-05,
463
- "loss": 0.1332,
464
  "step": 650
465
  },
466
  {
467
- "epoch": 7.857142857142857,
468
- "grad_norm": 2.0039525032043457,
469
- "learning_rate": 1.9442259433324247e-05,
470
- "loss": 0.1385,
471
  "step": 660
472
  },
473
  {
474
- "epoch": 7.976190476190476,
475
- "grad_norm": 1.5008896589279175,
476
- "learning_rate": 1.9407540154682473e-05,
477
- "loss": 0.1564,
478
  "step": 670
479
  },
480
  {
481
- "epoch": 8.095238095238095,
482
- "grad_norm": 1.5358928442001343,
483
- "learning_rate": 1.9371805543403826e-05,
484
- "loss": 0.0916,
485
  "step": 680
486
  },
487
  {
488
- "epoch": 8.214285714285714,
489
- "grad_norm": 1.3121554851531982,
490
- "learning_rate": 1.9335059456236877e-05,
491
- "loss": 0.0928,
492
  "step": 690
493
  },
494
  {
495
- "epoch": 8.333333333333334,
496
- "grad_norm": 1.5265170335769653,
497
- "learning_rate": 1.9297305859096305e-05,
498
- "loss": 0.0858,
499
  "step": 700
500
  },
501
  {
502
- "epoch": 8.452380952380953,
503
- "grad_norm": 2.1085143089294434,
504
- "learning_rate": 1.925854882663486e-05,
505
- "loss": 0.0821,
506
  "step": 710
507
  },
508
  {
509
- "epoch": 8.571428571428571,
510
- "grad_norm": 1.8447340726852417,
511
- "learning_rate": 1.9218792541803582e-05,
512
- "loss": 0.1064,
513
  "step": 720
514
  },
515
  {
516
- "epoch": 8.69047619047619,
517
- "grad_norm": 1.8958710432052612,
518
- "learning_rate": 1.9178041295400383e-05,
519
- "loss": 0.082,
520
  "step": 730
521
  },
522
  {
523
- "epoch": 8.80952380952381,
524
- "grad_norm": 1.9607478380203247,
525
- "learning_rate": 1.9136299485606898e-05,
526
- "loss": 0.0901,
527
  "step": 740
528
  },
529
  {
530
- "epoch": 8.928571428571429,
531
- "grad_norm": 2.034801959991455,
532
- "learning_rate": 1.9093571617513853e-05,
533
- "loss": 0.1067,
534
  "step": 750
535
  },
536
  {
537
- "epoch": 9.047619047619047,
538
- "grad_norm": 1.1250438690185547,
539
- "learning_rate": 1.90498623026348e-05,
540
- "loss": 0.0558,
541
  "step": 760
542
  },
543
  {
544
- "epoch": 9.166666666666666,
545
- "grad_norm": 3.2768161296844482,
546
- "learning_rate": 1.9005176258408434e-05,
547
- "loss": 0.0464,
548
  "step": 770
549
  },
550
  {
551
- "epoch": 9.285714285714286,
552
- "grad_norm": 1.0632288455963135,
553
- "learning_rate": 1.8959518307689444e-05,
554
- "loss": 0.0499,
555
  "step": 780
556
  },
557
  {
558
- "epoch": 9.404761904761905,
559
- "grad_norm": 1.1634855270385742,
560
- "learning_rate": 1.8912893378227984e-05,
561
- "loss": 0.0665,
562
  "step": 790
563
  },
564
  {
565
- "epoch": 9.523809523809524,
566
- "grad_norm": 2.089411497116089,
567
- "learning_rate": 1.8865306502137852e-05,
568
- "loss": 0.0715,
569
  "step": 800
570
  },
571
  {
572
- "epoch": 9.642857142857142,
573
- "grad_norm": 0.46574532985687256,
574
- "learning_rate": 1.8816762815353368e-05,
575
- "loss": 0.0453,
576
  "step": 810
577
  },
578
  {
579
- "epoch": 9.761904761904763,
580
- "grad_norm": 2.3916738033294678,
581
- "learning_rate": 1.876726755707508e-05,
582
- "loss": 0.1086,
583
  "step": 820
584
  },
585
  {
586
- "epoch": 9.880952380952381,
587
- "grad_norm": 1.7373374700546265,
588
- "learning_rate": 1.8716826069204297e-05,
589
- "loss": 0.1003,
590
  "step": 830
591
  },
592
  {
593
- "epoch": 10.0,
594
- "grad_norm": 1.2700127363204956,
595
- "learning_rate": 1.8665443795766566e-05,
596
- "loss": 0.054,
597
  "step": 840
598
  },
599
  {
600
- "epoch": 10.119047619047619,
601
- "grad_norm": 0.9935422539710999,
602
- "learning_rate": 1.8613126282324092e-05,
603
- "loss": 0.0403,
604
  "step": 850
605
  },
606
  {
607
- "epoch": 10.238095238095237,
608
- "grad_norm": 1.3984347581863403,
609
- "learning_rate": 1.855987917537724e-05,
610
- "loss": 0.0584,
611
  "step": 860
612
  },
613
  {
614
- "epoch": 10.357142857142858,
615
- "grad_norm": 2.4390032291412354,
616
- "learning_rate": 1.8505708221755106e-05,
617
- "loss": 0.0672,
618
  "step": 870
619
  },
620
  {
621
- "epoch": 10.476190476190476,
622
- "grad_norm": 0.9531667828559875,
623
- "learning_rate": 1.8450619267995283e-05,
624
- "loss": 0.0425,
625
  "step": 880
626
  },
627
  {
628
- "epoch": 10.595238095238095,
629
- "grad_norm": 0.42088013887405396,
630
- "learning_rate": 1.8394618259712866e-05,
631
- "loss": 0.0399,
632
  "step": 890
633
  },
634
  {
635
- "epoch": 10.714285714285714,
636
- "grad_norm": 0.8044432997703552,
637
- "learning_rate": 1.8337711240958734e-05,
638
- "loss": 0.0516,
639
  "step": 900
640
  },
641
  {
642
- "epoch": 10.833333333333334,
643
- "grad_norm": 1.7414997816085815,
644
- "learning_rate": 1.827990435356725e-05,
645
- "loss": 0.043,
646
  "step": 910
647
  },
648
  {
649
- "epoch": 10.952380952380953,
650
- "grad_norm": 0.6912918090820312,
651
- "learning_rate": 1.822120383649339e-05,
652
- "loss": 0.0444,
653
  "step": 920
654
  },
655
  {
656
- "epoch": 11.071428571428571,
657
- "grad_norm": 1.3013064861297607,
658
- "learning_rate": 1.8161616025139375e-05,
659
- "loss": 0.0436,
660
  "step": 930
661
  },
662
  {
663
- "epoch": 11.19047619047619,
664
- "grad_norm": 1.2811461687088013,
665
- "learning_rate": 1.8101147350670905e-05,
666
- "loss": 0.039,
667
  "step": 940
668
  },
669
  {
670
- "epoch": 11.30952380952381,
671
- "grad_norm": 6.565953254699707,
672
- "learning_rate": 1.8039804339323083e-05,
673
- "loss": 0.0411,
674
  "step": 950
675
  },
676
  {
677
- "epoch": 11.428571428571429,
678
- "grad_norm": 1.2721556425094604,
679
- "learning_rate": 1.7977593611696017e-05,
680
- "loss": 0.0242,
681
  "step": 960
682
  },
683
  {
684
- "epoch": 11.547619047619047,
685
- "grad_norm": 1.6954939365386963,
686
- "learning_rate": 1.791452188204031e-05,
687
- "loss": 0.0421,
688
  "step": 970
689
  },
690
  {
691
- "epoch": 11.666666666666666,
692
- "grad_norm": 1.3118443489074707,
693
- "learning_rate": 1.7850595957532394e-05,
694
- "loss": 0.0524,
695
  "step": 980
696
  },
697
  {
698
- "epoch": 11.785714285714286,
699
- "grad_norm": 0.5428506731987,
700
- "learning_rate": 1.778582273753984e-05,
701
- "loss": 0.0484,
702
  "step": 990
703
  },
704
  {
705
- "epoch": 11.904761904761905,
706
- "grad_norm": 0.4753515124320984,
707
- "learning_rate": 1.772020921287674e-05,
708
- "loss": 0.0347,
709
  "step": 1000
710
  },
711
  {
712
- "epoch": 12.023809523809524,
713
- "grad_norm": 1.2880959510803223,
714
- "learning_rate": 1.7653762465049185e-05,
715
- "loss": 0.0408,
716
  "step": 1010
717
  },
718
  {
719
- "epoch": 12.142857142857142,
720
- "grad_norm": 2.5490682125091553,
721
- "learning_rate": 1.7586489665491013e-05,
722
- "loss": 0.0461,
723
  "step": 1020
724
  },
725
  {
726
- "epoch": 12.261904761904763,
727
- "grad_norm": 0.1983003169298172,
728
- "learning_rate": 1.7518398074789776e-05,
729
- "loss": 0.0391,
730
  "step": 1030
731
  },
732
  {
733
- "epoch": 12.380952380952381,
734
- "grad_norm": 1.2978509664535522,
735
- "learning_rate": 1.7449495041903136e-05,
736
- "loss": 0.0272,
737
  "step": 1040
738
  },
739
  {
740
- "epoch": 12.5,
741
- "grad_norm": 0.3622434437274933,
742
- "learning_rate": 1.737978800336572e-05,
743
- "loss": 0.0431,
744
  "step": 1050
745
  },
746
  {
747
- "epoch": 12.619047619047619,
748
- "grad_norm": 2.1278223991394043,
749
- "learning_rate": 1.7309284482486494e-05,
750
- "loss": 0.0241,
751
  "step": 1060
752
  },
753
  {
754
- "epoch": 12.738095238095237,
755
- "grad_norm": 2.409932851791382,
756
- "learning_rate": 1.723799208853681e-05,
757
- "loss": 0.0247,
758
  "step": 1070
759
  },
760
  {
761
- "epoch": 12.857142857142858,
762
- "grad_norm": 0.8543738126754761,
763
- "learning_rate": 1.7165918515929147e-05,
764
- "loss": 0.0385,
765
  "step": 1080
766
  },
767
  {
768
- "epoch": 12.976190476190476,
769
- "grad_norm": 0.301162451505661,
770
- "learning_rate": 1.7093071543386667e-05,
771
- "loss": 0.0362,
772
  "step": 1090
773
  },
774
  {
775
- "epoch": 13.095238095238095,
776
- "grad_norm": 0.9338281750679016,
777
- "learning_rate": 1.7019459033103684e-05,
778
- "loss": 0.0297,
779
  "step": 1100
780
  },
781
  {
782
- "epoch": 13.214285714285714,
783
- "grad_norm": 1.0773793458938599,
784
- "learning_rate": 1.6945088929897115e-05,
785
- "loss": 0.0312,
786
  "step": 1110
787
  },
788
  {
789
- "epoch": 13.333333333333334,
790
- "grad_norm": 0.34426188468933105,
791
- "learning_rate": 1.686996926034902e-05,
792
- "loss": 0.0146,
793
  "step": 1120
794
  },
795
  {
796
- "epoch": 13.452380952380953,
797
- "grad_norm": 0.22328321635723114,
798
- "learning_rate": 1.67941081319403e-05,
799
- "loss": 0.0195,
800
  "step": 1130
801
  },
802
  {
803
- "epoch": 13.571428571428571,
804
- "grad_norm": 0.5331404805183411,
805
- "learning_rate": 1.6717513732175686e-05,
806
- "loss": 0.025,
807
  "step": 1140
808
  },
809
  {
810
- "epoch": 13.69047619047619,
811
- "grad_norm": 2.170349359512329,
812
- "learning_rate": 1.6640194327700087e-05,
813
- "loss": 0.0132,
814
  "step": 1150
815
  },
816
  {
817
- "epoch": 13.80952380952381,
818
- "grad_norm": 1.2327094078063965,
819
- "learning_rate": 1.6562158263406376e-05,
820
- "loss": 0.0175,
821
  "step": 1160
822
  },
823
  {
824
- "epoch": 13.928571428571429,
825
- "grad_norm": 0.2306894212961197,
826
- "learning_rate": 1.6483413961534764e-05,
827
- "loss": 0.0195,
828
  "step": 1170
829
  },
830
  {
831
- "epoch": 14.047619047619047,
832
- "grad_norm": 0.8987044095993042,
833
- "learning_rate": 1.640396992076379e-05,
834
- "loss": 0.0186,
835
  "step": 1180
836
  },
837
  {
838
- "epoch": 14.166666666666666,
839
- "grad_norm": 1.013019323348999,
840
- "learning_rate": 1.6323834715293094e-05,
841
- "loss": 0.0259,
842
  "step": 1190
843
  },
844
  {
845
- "epoch": 14.285714285714286,
846
- "grad_norm": 0.14262178540229797,
847
- "learning_rate": 1.6243016993918022e-05,
848
- "loss": 0.0158,
849
  "step": 1200
850
  },
851
  {
852
- "epoch": 14.404761904761905,
853
- "grad_norm": 0.7596869468688965,
854
- "learning_rate": 1.616152547909618e-05,
855
- "loss": 0.0265,
856
  "step": 1210
857
  },
858
  {
859
- "epoch": 14.523809523809524,
860
- "grad_norm": 0.2511692941188812,
861
- "learning_rate": 1.6079368966006022e-05,
862
- "loss": 0.022,
863
  "step": 1220
864
  },
865
  {
866
- "epoch": 14.642857142857142,
867
- "grad_norm": 0.16154764592647552,
868
- "learning_rate": 1.5996556321597663e-05,
869
- "loss": 0.0222,
870
  "step": 1230
871
  },
872
  {
873
- "epoch": 14.761904761904763,
874
- "grad_norm": 0.6887322664260864,
875
- "learning_rate": 1.5913096483635827e-05,
876
- "loss": 0.008,
877
  "step": 1240
878
  },
879
  {
880
- "epoch": 14.880952380952381,
881
- "grad_norm": 0.07254794239997864,
882
- "learning_rate": 1.582899845973525e-05,
883
- "loss": 0.0034,
884
  "step": 1250
885
  },
886
  {
887
- "epoch": 15.0,
888
- "grad_norm": 0.4768582284450531,
889
- "learning_rate": 1.574427132638851e-05,
890
- "loss": 0.0083,
891
  "step": 1260
892
  },
893
  {
894
- "epoch": 15.119047619047619,
895
- "grad_norm": 0.04978374391794205,
896
- "learning_rate": 1.5658924227986415e-05,
897
- "loss": 0.0102,
898
  "step": 1270
899
  },
900
  {
901
- "epoch": 15.238095238095237,
902
- "grad_norm": 1.5553603172302246,
903
- "learning_rate": 1.5572966375831072e-05,
904
- "loss": 0.0082,
905
  "step": 1280
906
  },
907
  {
908
- "epoch": 15.357142857142858,
909
- "grad_norm": 0.1932092010974884,
910
- "learning_rate": 1.5486407047141745e-05,
911
- "loss": 0.0191,
912
  "step": 1290
913
  },
914
  {
915
- "epoch": 15.476190476190476,
916
- "grad_norm": 0.12488748133182526,
917
- "learning_rate": 1.5399255584053568e-05,
918
- "loss": 0.0068,
919
  "step": 1300
920
  },
921
  {
922
- "epoch": 15.595238095238095,
923
- "grad_norm": 0.04466537758708,
924
- "learning_rate": 1.5311521392609283e-05,
925
- "loss": 0.0067,
926
  "step": 1310
927
  },
928
  {
929
- "epoch": 15.714285714285714,
930
- "grad_norm": 0.1408109813928604,
931
- "learning_rate": 1.5223213941744078e-05,
932
- "loss": 0.0076,
933
  "step": 1320
934
  },
935
  {
936
- "epoch": 15.833333333333334,
937
- "grad_norm": 0.4874057471752167,
938
- "learning_rate": 1.5134342762263606e-05,
939
- "loss": 0.0109,
940
  "step": 1330
941
  },
942
  {
943
- "epoch": 15.952380952380953,
944
- "grad_norm": 0.15269064903259277,
945
- "learning_rate": 1.504491744581537e-05,
946
- "loss": 0.0073,
947
  "step": 1340
948
  },
949
  {
950
- "epoch": 16.071428571428573,
951
- "grad_norm": 0.192997545003891,
952
- "learning_rate": 1.4954947643853508e-05,
953
- "loss": 0.0223,
954
  "step": 1350
955
  },
956
  {
957
- "epoch": 16.19047619047619,
958
- "grad_norm": 0.9681908488273621,
959
- "learning_rate": 1.486444306659714e-05,
960
- "loss": 0.0037,
961
  "step": 1360
962
  },
963
  {
964
- "epoch": 16.30952380952381,
965
- "grad_norm": 1.2680203914642334,
966
- "learning_rate": 1.4773413481982359e-05,
967
- "loss": 0.008,
968
  "step": 1370
969
  },
970
  {
971
- "epoch": 16.428571428571427,
972
- "grad_norm": 0.07117388397455215,
973
- "learning_rate": 1.4681868714608021e-05,
974
- "loss": 0.0046,
975
  "step": 1380
976
  },
977
  {
978
- "epoch": 16.547619047619047,
979
- "grad_norm": 0.031909532845020294,
980
- "learning_rate": 1.4589818644675378e-05,
981
- "loss": 0.0101,
982
  "step": 1390
983
  },
984
  {
985
- "epoch": 16.666666666666668,
986
- "grad_norm": 1.234426498413086,
987
- "learning_rate": 1.4497273206921747e-05,
988
- "loss": 0.0051,
989
  "step": 1400
990
  },
991
  {
992
- "epoch": 16.785714285714285,
993
- "grad_norm": 0.43718039989471436,
994
- "learning_rate": 1.4404242389548274e-05,
995
- "loss": 0.0063,
996
  "step": 1410
997
  },
998
  {
999
- "epoch": 16.904761904761905,
1000
- "grad_norm": 3.5411911010742188,
1001
- "learning_rate": 1.4310736233141926e-05,
1002
- "loss": 0.0177,
1003
  "step": 1420
1004
  },
1005
  {
1006
- "epoch": 17.023809523809526,
1007
- "grad_norm": 0.09840346872806549,
1008
- "learning_rate": 1.4216764829591853e-05,
1009
- "loss": 0.0071,
1010
  "step": 1430
1011
  },
1012
  {
1013
- "epoch": 17.142857142857142,
1014
- "grad_norm": 0.09478545933961868,
1015
- "learning_rate": 1.4122338321000175e-05,
1016
- "loss": 0.0032,
1017
  "step": 1440
1018
  },
1019
  {
1020
- "epoch": 17.261904761904763,
1021
- "grad_norm": 0.040692731738090515,
1022
- "learning_rate": 1.4027466898587375e-05,
1023
- "loss": 0.0036,
1024
  "step": 1450
1025
  },
1026
  {
1027
- "epoch": 17.38095238095238,
1028
- "grad_norm": 0.6164568066596985,
1029
- "learning_rate": 1.3932160801592395e-05,
1030
- "loss": 0.0066,
1031
  "step": 1460
1032
  },
1033
  {
1034
- "epoch": 17.5,
1035
- "grad_norm": 0.2039034515619278,
1036
- "learning_rate": 1.3836430316167533e-05,
1037
- "loss": 0.008,
1038
  "step": 1470
1039
  },
1040
  {
1041
- "epoch": 17.61904761904762,
1042
- "grad_norm": 0.06619343906641006,
1043
- "learning_rate": 1.3740285774268282e-05,
1044
- "loss": 0.0068,
1045
  "step": 1480
1046
  },
1047
  {
1048
- "epoch": 17.738095238095237,
1049
- "grad_norm": 0.050607554614543915,
1050
- "learning_rate": 1.3643737552538237e-05,
1051
- "loss": 0.0058,
1052
  "step": 1490
1053
  },
1054
  {
1055
- "epoch": 17.857142857142858,
1056
- "grad_norm": 0.6153551340103149,
1057
- "learning_rate": 1.3546796071189153e-05,
1058
- "loss": 0.0089,
1059
  "step": 1500
1060
  },
1061
  {
1062
- "epoch": 17.976190476190474,
1063
- "grad_norm": 0.07599499821662903,
1064
- "learning_rate": 1.3449471792876333e-05,
1065
- "loss": 0.0063,
1066
  "step": 1510
1067
  },
1068
  {
1069
- "epoch": 18.095238095238095,
1070
- "grad_norm": 0.0883430764079094,
1071
- "learning_rate": 1.3351775221569416e-05,
1072
- "loss": 0.0054,
1073
  "step": 1520
1074
  },
1075
  {
1076
- "epoch": 18.214285714285715,
1077
- "grad_norm": 0.02305910363793373,
1078
- "learning_rate": 1.3253716901418702e-05,
1079
- "loss": 0.0028,
1080
  "step": 1530
1081
  },
1082
  {
1083
- "epoch": 18.333333333333332,
1084
- "grad_norm": 0.6192808151245117,
1085
- "learning_rate": 1.3155307415617156e-05,
1086
- "loss": 0.002,
1087
  "step": 1540
1088
  },
1089
  {
1090
- "epoch": 18.452380952380953,
1091
- "grad_norm": 0.23332145810127258,
1092
- "learning_rate": 1.3056557385258181e-05,
1093
- "loss": 0.0066,
1094
  "step": 1550
1095
  },
1096
  {
1097
- "epoch": 18.571428571428573,
1098
- "grad_norm": 0.045441482216119766,
1099
- "learning_rate": 1.2957477468189321e-05,
1100
- "loss": 0.0015,
1101
  "step": 1560
1102
  },
1103
  {
1104
- "epoch": 18.69047619047619,
1105
- "grad_norm": 0.009192162193357944,
1106
- "learning_rate": 1.2858078357861979e-05,
1107
- "loss": 0.0128,
1108
  "step": 1570
1109
  },
1110
  {
1111
- "epoch": 18.80952380952381,
1112
- "grad_norm": 0.05836764723062515,
1113
- "learning_rate": 1.2758370782177296e-05,
1114
- "loss": 0.0048,
1115
  "step": 1580
1116
  },
1117
  {
1118
- "epoch": 18.928571428571427,
1119
- "grad_norm": 1.0975781679153442,
1120
- "learning_rate": 1.2658365502328329e-05,
1121
- "loss": 0.007,
1122
  "step": 1590
1123
  },
1124
  {
1125
- "epoch": 19.047619047619047,
1126
- "grad_norm": 1.1575758457183838,
1127
- "learning_rate": 1.2558073311638604e-05,
1128
- "loss": 0.018,
1129
  "step": 1600
1130
  },
1131
  {
1132
- "epoch": 19.166666666666668,
1133
- "grad_norm": 0.1114954873919487,
1134
- "learning_rate": 1.2457505034397223e-05,
1135
- "loss": 0.0023,
1136
  "step": 1610
1137
  },
1138
  {
1139
- "epoch": 19.285714285714285,
1140
- "grad_norm": 0.07574819773435593,
1141
- "learning_rate": 1.2356671524690645e-05,
1142
- "loss": 0.0019,
1143
  "step": 1620
1144
  },
1145
  {
1146
- "epoch": 19.404761904761905,
1147
- "grad_norm": 0.009654861874878407,
1148
- "learning_rate": 1.2255583665231196e-05,
1149
- "loss": 0.0026,
1150
  "step": 1630
1151
  },
1152
  {
1153
- "epoch": 19.523809523809526,
1154
- "grad_norm": 0.02393646165728569,
1155
- "learning_rate": 1.2154252366182564e-05,
1156
- "loss": 0.0014,
1157
  "step": 1640
1158
  },
1159
  {
1160
- "epoch": 19.642857142857142,
1161
- "grad_norm": 0.020082678645849228,
1162
- "learning_rate": 1.2052688563982257e-05,
1163
- "loss": 0.0012,
1164
  "step": 1650
1165
  },
1166
  {
1167
- "epoch": 19.761904761904763,
1168
- "grad_norm": 0.9789303541183472,
1169
- "learning_rate": 1.1950903220161286e-05,
1170
- "loss": 0.0063,
1171
  "step": 1660
1172
  },
1173
  {
1174
- "epoch": 19.88095238095238,
1175
- "grad_norm": 0.2839144766330719,
1176
- "learning_rate": 1.1848907320161092e-05,
1177
- "loss": 0.0034,
1178
  "step": 1670
1179
  },
1180
  {
1181
- "epoch": 20.0,
1182
- "grad_norm": 0.010598956607282162,
1183
- "learning_rate": 1.174671187214794e-05,
1184
- "loss": 0.0054,
1185
  "step": 1680
1186
  },
1187
  {
1188
- "epoch": 20.11904761904762,
1189
- "grad_norm": 0.07026033848524094,
1190
- "learning_rate": 1.1644327905824808e-05,
1191
- "loss": 0.0084,
1192
  "step": 1690
1193
  },
1194
  {
1195
- "epoch": 20.238095238095237,
1196
- "grad_norm": 0.2679937779903412,
1197
- "learning_rate": 1.1541766471241e-05,
1198
- "loss": 0.0022,
1199
  "step": 1700
1200
  },
1201
  {
1202
- "epoch": 20.357142857142858,
1203
- "grad_norm": 0.10295789688825607,
1204
- "learning_rate": 1.1439038637599529e-05,
1205
- "loss": 0.0008,
1206
  "step": 1710
1207
  },
1208
  {
1209
- "epoch": 20.476190476190474,
1210
- "grad_norm": 0.4343288242816925,
1211
- "learning_rate": 1.1336155492062439e-05,
1212
- "loss": 0.0021,
1213
  "step": 1720
1214
  },
1215
  {
1216
- "epoch": 20.595238095238095,
1217
- "grad_norm": 0.018273649737238884,
1218
- "learning_rate": 1.1233128138554222e-05,
1219
- "loss": 0.0006,
1220
  "step": 1730
1221
  },
1222
  {
1223
- "epoch": 20.714285714285715,
1224
- "grad_norm": 0.008875339291989803,
1225
- "learning_rate": 1.112996769656337e-05,
1226
- "loss": 0.0003,
1227
  "step": 1740
1228
  },
1229
  {
1230
- "epoch": 20.833333333333332,
1231
- "grad_norm": 0.21868984401226044,
1232
- "learning_rate": 1.1026685299942286e-05,
1233
- "loss": 0.0022,
1234
  "step": 1750
1235
  },
1236
  {
1237
- "epoch": 20.952380952380953,
1238
- "grad_norm": 0.024242576211690903,
1239
- "learning_rate": 1.0923292095705642e-05,
1240
- "loss": 0.003,
1241
  "step": 1760
1242
  },
1243
  {
1244
- "epoch": 21.071428571428573,
1245
- "grad_norm": 0.008948619477450848,
1246
- "learning_rate": 1.0819799242827299e-05,
1247
- "loss": 0.0046,
1248
  "step": 1770
1249
  },
1250
  {
1251
- "epoch": 21.19047619047619,
1252
- "grad_norm": 0.01921582594513893,
1253
- "learning_rate": 1.0716217911035952e-05,
1254
- "loss": 0.002,
1255
  "step": 1780
1256
  },
1257
  {
1258
- "epoch": 21.30952380952381,
1259
- "grad_norm": 0.015329894609749317,
1260
- "learning_rate": 1.0612559279609606e-05,
1261
- "loss": 0.0004,
1262
  "step": 1790
1263
  },
1264
  {
1265
- "epoch": 21.428571428571427,
1266
- "grad_norm": 0.03253422677516937,
1267
- "learning_rate": 1.0508834536169028e-05,
1268
- "loss": 0.0032,
1269
  "step": 1800
1270
  },
1271
  {
1272
- "epoch": 21.547619047619047,
1273
- "grad_norm": 0.007767899427562952,
1274
- "learning_rate": 1.0405054875470287e-05,
1275
- "loss": 0.0003,
1276
  "step": 1810
1277
  },
1278
  {
1279
- "epoch": 21.666666666666668,
1280
- "grad_norm": 0.014929098077118397,
1281
- "learning_rate": 1.030123149819654e-05,
1282
- "loss": 0.0052,
1283
  "step": 1820
1284
  },
1285
  {
1286
- "epoch": 21.785714285714285,
1287
- "grad_norm": 0.020312979817390442,
1288
- "learning_rate": 1.0197375609749164e-05,
1289
- "loss": 0.0006,
1290
  "step": 1830
1291
  },
1292
  {
1293
- "epoch": 21.904761904761905,
1294
- "grad_norm": 0.04301303252577782,
1295
- "learning_rate": 1.0093498419038394e-05,
1296
- "loss": 0.002,
1297
  "step": 1840
1298
  },
1299
  {
1300
- "epoch": 22.023809523809526,
1301
- "grad_norm": 0.011388319544494152,
1302
- "learning_rate": 9.989611137273556e-06,
1303
- "loss": 0.004,
1304
  "step": 1850
1305
  },
1306
  {
1307
- "epoch": 22.142857142857142,
1308
- "grad_norm": 0.02648976631462574,
1309
- "learning_rate": 9.885724976753078e-06,
1310
- "loss": 0.0005,
1311
  "step": 1860
1312
  },
1313
  {
1314
- "epoch": 22.261904761904763,
1315
- "grad_norm": 0.016934091225266457,
1316
- "learning_rate": 9.78185114965439e-06,
1317
- "loss": 0.0007,
1318
  "step": 1870
1319
  },
1320
  {
1321
- "epoch": 22.38095238095238,
1322
- "grad_norm": 0.008141864091157913,
1323
- "learning_rate": 9.678000866823789e-06,
1324
- "loss": 0.0031,
1325
  "step": 1880
1326
  },
1327
  {
1328
- "epoch": 22.5,
1329
- "grad_norm": 0.015341422520577908,
1330
- "learning_rate": 9.574185336566503e-06,
1331
- "loss": 0.0007,
1332
  "step": 1890
1333
  },
1334
  {
1335
- "epoch": 22.61904761904762,
1336
- "grad_norm": 0.013799485750496387,
1337
- "learning_rate": 9.470415763437003e-06,
1338
- "loss": 0.0033,
1339
  "step": 1900
1340
  },
1341
  {
1342
- "epoch": 22.738095238095237,
1343
- "grad_norm": 0.032908592373132706,
1344
- "learning_rate": 9.366703347029725e-06,
1345
- "loss": 0.001,
1346
  "step": 1910
1347
  },
1348
  {
1349
- "epoch": 22.857142857142858,
1350
- "grad_norm": 0.06811460107564926,
1351
- "learning_rate": 9.263059280770303e-06,
1352
- "loss": 0.0024,
1353
  "step": 1920
1354
  },
1355
  {
1356
- "epoch": 22.976190476190474,
1357
- "grad_norm": 0.045220863074064255,
1358
- "learning_rate": 9.159494750707527e-06,
1359
- "loss": 0.0004,
1360
  "step": 1930
1361
  },
1362
  {
1363
- "epoch": 23.095238095238095,
1364
- "grad_norm": 0.003873666515573859,
1365
- "learning_rate": 9.05602093430603e-06,
1366
- "loss": 0.0003,
1367
  "step": 1940
1368
  },
1369
  {
1370
- "epoch": 23.214285714285715,
1371
- "grad_norm": 0.013574629090726376,
1372
- "learning_rate": 8.952648999239948e-06,
1373
- "loss": 0.0035,
1374
  "step": 1950
1375
  },
1376
  {
1377
- "epoch": 23.333333333333332,
1378
- "grad_norm": 0.009078383445739746,
1379
- "learning_rate": 8.849390102187615e-06,
1380
- "loss": 0.0002,
1381
  "step": 1960
1382
  },
1383
  {
1384
- "epoch": 23.452380952380953,
1385
- "grad_norm": 0.0057939100079238415,
1386
- "learning_rate": 8.746255387627444e-06,
1387
- "loss": 0.0004,
1388
  "step": 1970
1389
  },
1390
  {
1391
- "epoch": 23.571428571428573,
1392
- "grad_norm": 0.019538553431630135,
1393
- "learning_rate": 8.643255986635152e-06,
1394
- "loss": 0.0002,
1395
  "step": 1980
1396
  },
1397
  {
1398
- "epoch": 23.69047619047619,
1399
- "grad_norm": 0.007959152571856976,
1400
- "learning_rate": 8.540403015682382e-06,
1401
- "loss": 0.0009,
1402
  "step": 1990
1403
  },
1404
  {
1405
- "epoch": 23.80952380952381,
1406
- "grad_norm": 0.009719247929751873,
1407
- "learning_rate": 8.437707575436943e-06,
1408
- "loss": 0.0078,
1409
  "step": 2000
1410
  }
1411
  ],
1412
  "logging_steps": 10,
1413
- "max_steps": 3360,
1414
  "num_input_tokens_seen": 0,
1415
- "num_train_epochs": 40,
1416
  "save_steps": 1000,
1417
  "stateful_callbacks": {
1418
  "TrainerControl": {
@@ -1426,8 +1426,8 @@
1426
  "attributes": {}
1427
  }
1428
  },
1429
- "total_flos": 1.1931390331741798e+17,
1430
- "train_batch_size": 3,
1431
  "trial_name": null,
1432
  "trial_params": null
1433
  }
 
1
  {
2
  "best_metric": null,
3
  "best_model_checkpoint": null,
4
+ "epoch": 16.0,
5
  "eval_steps": 500,
6
  "global_step": 2000,
7
  "is_hyper_param_search": false,
 
9
  "is_world_process_zero": true,
10
  "log_history": [
11
  {
12
+ "epoch": 0.08,
13
+ "grad_norm": 6.874109268188477,
14
+ "learning_rate": 2.5559105431309904e-07,
15
+ "loss": 1.5574,
16
  "step": 10
17
  },
18
  {
19
+ "epoch": 0.16,
20
+ "grad_norm": 6.183030128479004,
21
+ "learning_rate": 8.306709265175719e-07,
22
+ "loss": 1.4724,
23
  "step": 20
24
  },
25
  {
26
+ "epoch": 0.24,
27
+ "grad_norm": 2.6565823554992676,
28
+ "learning_rate": 1.4696485623003196e-06,
29
+ "loss": 1.054,
30
  "step": 30
31
  },
32
  {
33
+ "epoch": 0.32,
34
+ "grad_norm": 3.227349042892456,
35
+ "learning_rate": 2.1086261980830672e-06,
36
+ "loss": 0.8732,
37
  "step": 40
38
  },
39
  {
40
+ "epoch": 0.4,
41
+ "grad_norm": 1.1049450635910034,
42
+ "learning_rate": 2.747603833865815e-06,
43
+ "loss": 0.7617,
44
  "step": 50
45
  },
46
  {
47
+ "epoch": 0.48,
48
+ "grad_norm": 1.8624663352966309,
49
+ "learning_rate": 3.386581469648563e-06,
50
+ "loss": 0.7144,
51
  "step": 60
52
  },
53
  {
54
+ "epoch": 0.56,
55
+ "grad_norm": 2.1665029525756836,
56
+ "learning_rate": 4.02555910543131e-06,
57
+ "loss": 0.7612,
58
  "step": 70
59
  },
60
  {
61
+ "epoch": 0.64,
62
+ "grad_norm": 1.0787376165390015,
63
+ "learning_rate": 4.664536741214058e-06,
64
+ "loss": 0.6864,
65
  "step": 80
66
  },
67
  {
68
+ "epoch": 0.72,
69
+ "grad_norm": 1.4465893507003784,
70
+ "learning_rate": 5.303514376996806e-06,
71
+ "loss": 0.6558,
72
  "step": 90
73
  },
74
  {
75
+ "epoch": 0.8,
76
+ "grad_norm": 1.1737278699874878,
77
+ "learning_rate": 5.8785942492012785e-06,
78
+ "loss": 0.6295,
79
  "step": 100
80
  },
81
  {
82
+ "epoch": 0.88,
83
+ "grad_norm": 1.7027487754821777,
84
+ "learning_rate": 6.517571884984026e-06,
85
+ "loss": 0.638,
86
  "step": 110
87
  },
88
  {
89
+ "epoch": 0.96,
90
+ "grad_norm": 1.5483449697494507,
91
+ "learning_rate": 7.156549520766773e-06,
92
+ "loss": 0.6173,
93
  "step": 120
94
  },
95
  {
96
+ "epoch": 1.04,
97
+ "grad_norm": 1.9829121828079224,
98
+ "learning_rate": 7.795527156549521e-06,
99
+ "loss": 0.5511,
100
  "step": 130
101
  },
102
  {
103
+ "epoch": 1.12,
104
+ "grad_norm": 1.466599464416504,
105
+ "learning_rate": 8.434504792332269e-06,
106
+ "loss": 0.5366,
107
  "step": 140
108
  },
109
  {
110
+ "epoch": 1.2,
111
+ "grad_norm": 1.5559449195861816,
112
+ "learning_rate": 9.073482428115017e-06,
113
+ "loss": 0.5662,
114
  "step": 150
115
  },
116
  {
117
+ "epoch": 1.28,
118
+ "grad_norm": 1.33512282371521,
119
+ "learning_rate": 9.712460063897765e-06,
120
+ "loss": 0.5567,
121
  "step": 160
122
  },
123
  {
124
+ "epoch": 1.3599999999999999,
125
+ "grad_norm": 0.943450927734375,
126
+ "learning_rate": 1.035143769968051e-05,
127
+ "loss": 0.5997,
128
  "step": 170
129
  },
130
  {
131
+ "epoch": 1.44,
132
+ "grad_norm": 1.2756850719451904,
133
+ "learning_rate": 1.099041533546326e-05,
134
+ "loss": 0.5587,
135
  "step": 180
136
  },
137
  {
138
+ "epoch": 1.52,
139
+ "grad_norm": 1.0768604278564453,
140
+ "learning_rate": 1.1629392971246008e-05,
141
+ "loss": 0.6288,
142
  "step": 190
143
  },
144
  {
145
+ "epoch": 1.6,
146
+ "grad_norm": 1.082158088684082,
147
+ "learning_rate": 1.2268370607028754e-05,
148
+ "loss": 0.5475,
149
  "step": 200
150
  },
151
  {
152
+ "epoch": 1.6800000000000002,
153
+ "grad_norm": 1.4872550964355469,
154
+ "learning_rate": 1.2907348242811502e-05,
155
+ "loss": 0.6002,
156
  "step": 210
157
  },
158
  {
159
+ "epoch": 1.76,
160
+ "grad_norm": 2.0925095081329346,
161
+ "learning_rate": 1.3546325878594251e-05,
162
+ "loss": 0.5887,
163
  "step": 220
164
  },
165
  {
166
+ "epoch": 1.8399999999999999,
167
+ "grad_norm": 1.4571633338928223,
168
+ "learning_rate": 1.4185303514376998e-05,
169
+ "loss": 0.6107,
170
  "step": 230
171
  },
172
  {
173
+ "epoch": 1.92,
174
+ "grad_norm": 2.4871673583984375,
175
+ "learning_rate": 1.4824281150159745e-05,
176
+ "loss": 0.5582,
177
  "step": 240
178
  },
179
  {
180
+ "epoch": 2.0,
181
+ "grad_norm": 1.3184454441070557,
182
+ "learning_rate": 1.5463258785942495e-05,
183
+ "loss": 0.582,
184
  "step": 250
185
  },
186
  {
187
+ "epoch": 2.08,
188
+ "grad_norm": 1.9747495651245117,
189
+ "learning_rate": 1.610223642172524e-05,
190
+ "loss": 0.4687,
191
  "step": 260
192
  },
193
  {
194
+ "epoch": 2.16,
195
+ "grad_norm": 1.7691447734832764,
196
+ "learning_rate": 1.6741214057507987e-05,
197
+ "loss": 0.5013,
198
  "step": 270
199
  },
200
  {
201
+ "epoch": 2.24,
202
+ "grad_norm": 1.8902829885482788,
203
+ "learning_rate": 1.7380191693290737e-05,
204
+ "loss": 0.5008,
205
  "step": 280
206
  },
207
  {
208
+ "epoch": 2.32,
209
+ "grad_norm": 1.3197141885757446,
210
+ "learning_rate": 1.8019169329073486e-05,
211
+ "loss": 0.5229,
212
  "step": 290
213
  },
214
  {
215
+ "epoch": 2.4,
216
+ "grad_norm": 2.325404167175293,
217
+ "learning_rate": 1.8658146964856232e-05,
218
+ "loss": 0.4975,
219
  "step": 300
220
  },
221
  {
222
+ "epoch": 2.48,
223
+ "grad_norm": 1.3323551416397095,
224
+ "learning_rate": 1.929712460063898e-05,
225
+ "loss": 0.5254,
226
  "step": 310
227
  },
228
  {
229
+ "epoch": 2.56,
230
+ "grad_norm": 0.7770269513130188,
231
+ "learning_rate": 1.9936102236421725e-05,
232
+ "loss": 0.5304,
233
  "step": 320
234
  },
235
  {
236
+ "epoch": 2.64,
237
+ "grad_norm": 1.887294054031372,
238
+ "learning_rate": 1.999949450079496e-05,
239
+ "loss": 0.4755,
240
  "step": 330
241
  },
242
  {
243
+ "epoch": 2.7199999999999998,
244
+ "grad_norm": 1.2255381345748901,
245
+ "learning_rate": 1.9997747161747696e-05,
246
+ "loss": 0.4982,
247
  "step": 340
248
  },
249
  {
250
+ "epoch": 2.8,
251
+ "grad_norm": 1.6865767240524292,
252
+ "learning_rate": 1.9994751960168383e-05,
253
+ "loss": 0.4857,
254
  "step": 350
255
  },
256
  {
257
+ "epoch": 2.88,
258
+ "grad_norm": 1.2997519969940186,
259
+ "learning_rate": 1.999050926990122e-05,
260
+ "loss": 0.4987,
261
  "step": 360
262
  },
263
  {
264
+ "epoch": 2.96,
265
+ "grad_norm": 1.4636356830596924,
266
+ "learning_rate": 1.9985019620494935e-05,
267
+ "loss": 0.4947,
268
  "step": 370
269
  },
270
  {
271
+ "epoch": 3.04,
272
+ "grad_norm": 1.3720135688781738,
273
+ "learning_rate": 1.9978283697136662e-05,
274
+ "loss": 0.5135,
275
  "step": 380
276
  },
277
  {
278
+ "epoch": 3.12,
279
+ "grad_norm": 1.6108952760696411,
280
+ "learning_rate": 1.997030234056645e-05,
281
+ "loss": 0.4405,
282
  "step": 390
283
  },
284
  {
285
+ "epoch": 3.2,
286
+ "grad_norm": 2.1790759563446045,
287
+ "learning_rate": 1.9961076546972304e-05,
288
+ "loss": 0.4468,
289
  "step": 400
290
  },
291
  {
292
+ "epoch": 3.2800000000000002,
293
+ "grad_norm": 4.406925678253174,
294
+ "learning_rate": 1.9950607467865856e-05,
295
+ "loss": 0.4675,
296
  "step": 410
297
  },
298
  {
299
+ "epoch": 3.36,
300
+ "grad_norm": 1.7346851825714111,
301
+ "learning_rate": 1.993889640993864e-05,
302
+ "loss": 0.4243,
303
  "step": 420
304
  },
305
  {
306
+ "epoch": 3.44,
307
+ "grad_norm": 1.2865455150604248,
308
+ "learning_rate": 1.9925944834898992e-05,
309
+ "loss": 0.3787,
310
  "step": 430
311
  },
312
  {
313
+ "epoch": 3.52,
314
+ "grad_norm": 1.321824550628662,
315
+ "learning_rate": 1.991175435928962e-05,
316
+ "loss": 0.3982,
317
  "step": 440
318
  },
319
  {
320
+ "epoch": 3.6,
321
+ "grad_norm": 2.45711350440979,
322
+ "learning_rate": 1.989632675428582e-05,
323
+ "loss": 0.4811,
324
  "step": 450
325
  },
326
  {
327
+ "epoch": 3.68,
328
+ "grad_norm": 2.5552310943603516,
329
+ "learning_rate": 1.9879663945474416e-05,
330
+ "loss": 0.4255,
331
  "step": 460
332
  },
333
  {
334
+ "epoch": 3.76,
335
+ "grad_norm": 1.763244867324829,
336
+ "learning_rate": 1.9861768012613435e-05,
337
+ "loss": 0.4579,
338
  "step": 470
339
  },
340
  {
341
+ "epoch": 3.84,
342
+ "grad_norm": 1.0951169729232788,
343
+ "learning_rate": 1.984264118937249e-05,
344
+ "loss": 0.398,
345
  "step": 480
346
  },
347
  {
348
+ "epoch": 3.92,
349
+ "grad_norm": 8.235908508300781,
350
+ "learning_rate": 1.9822285863054e-05,
351
+ "loss": 0.4101,
352
  "step": 490
353
  },
354
  {
355
+ "epoch": 4.0,
356
+ "grad_norm": 19.39371681213379,
357
+ "learning_rate": 1.9800704574295246e-05,
358
+ "loss": 0.4645,
359
  "step": 500
360
  },
361
  {
362
+ "epoch": 4.08,
363
+ "grad_norm": 1.2594205141067505,
364
+ "learning_rate": 1.9777900016751224e-05,
365
+ "loss": 0.3632,
366
  "step": 510
367
  },
368
  {
369
+ "epoch": 4.16,
370
+ "grad_norm": 1.7697833776474,
371
+ "learning_rate": 1.9753875036758464e-05,
372
+ "loss": 0.3356,
373
  "step": 520
374
  },
375
  {
376
+ "epoch": 4.24,
377
+ "grad_norm": 1.66884446144104,
378
+ "learning_rate": 1.9728632632979746e-05,
379
+ "loss": 0.382,
380
  "step": 530
381
  },
382
  {
383
+ "epoch": 4.32,
384
+ "grad_norm": 1.7648805379867554,
385
+ "learning_rate": 1.970217595602985e-05,
386
+ "loss": 0.3997,
387
  "step": 540
388
  },
389
  {
390
+ "epoch": 4.4,
391
+ "grad_norm": 1.6132514476776123,
392
+ "learning_rate": 1.967450830808228e-05,
393
+ "loss": 0.3476,
394
  "step": 550
395
  },
396
  {
397
+ "epoch": 4.48,
398
+ "grad_norm": 1.3217155933380127,
399
+ "learning_rate": 1.9645633142457143e-05,
400
+ "loss": 0.3362,
401
  "step": 560
402
  },
403
  {
404
+ "epoch": 4.5600000000000005,
405
+ "grad_norm": 1.5668702125549316,
406
+ "learning_rate": 1.9615554063190098e-05,
407
+ "loss": 0.3996,
408
  "step": 570
409
  },
410
  {
411
+ "epoch": 4.64,
412
+ "grad_norm": 2.051454544067383,
413
+ "learning_rate": 1.958427482458253e-05,
414
+ "loss": 0.4104,
415
  "step": 580
416
  },
417
  {
418
+ "epoch": 4.72,
419
+ "grad_norm": 1.84868586063385,
420
+ "learning_rate": 1.9551799330732954e-05,
421
+ "loss": 0.374,
422
  "step": 590
423
  },
424
  {
425
+ "epoch": 4.8,
426
+ "grad_norm": 1.3498121500015259,
427
+ "learning_rate": 1.9518131635049745e-05,
428
+ "loss": 0.3444,
429
  "step": 600
430
  },
431
  {
432
+ "epoch": 4.88,
433
+ "grad_norm": 1.962553858757019,
434
+ "learning_rate": 1.9483275939745184e-05,
435
+ "loss": 0.3153,
436
  "step": 610
437
  },
438
  {
439
+ "epoch": 4.96,
440
+ "grad_norm": 1.7132457494735718,
441
+ "learning_rate": 1.944723659531099e-05,
442
+ "loss": 0.383,
443
  "step": 620
444
  },
445
  {
446
+ "epoch": 5.04,
447
+ "grad_norm": 1.8860396146774292,
448
+ "learning_rate": 1.9410018099975297e-05,
449
+ "loss": 0.327,
450
  "step": 630
451
  },
452
  {
453
+ "epoch": 5.12,
454
+ "grad_norm": 1.9154820442199707,
455
+ "learning_rate": 1.9371625099141223e-05,
456
+ "loss": 0.2704,
457
  "step": 640
458
  },
459
  {
460
+ "epoch": 5.2,
461
+ "grad_norm": 1.4077069759368896,
462
+ "learning_rate": 1.9332062384807058e-05,
463
+ "loss": 0.2659,
464
  "step": 650
465
  },
466
  {
467
+ "epoch": 5.28,
468
+ "grad_norm": 1.8489642143249512,
469
+ "learning_rate": 1.9291334894968133e-05,
470
+ "loss": 0.2523,
471
  "step": 660
472
  },
473
  {
474
+ "epoch": 5.36,
475
+ "grad_norm": 2.3917829990386963,
476
+ "learning_rate": 1.9249447713000515e-05,
477
+ "loss": 0.2885,
478
  "step": 670
479
  },
480
  {
481
+ "epoch": 5.44,
482
+ "grad_norm": 2.5429599285125732,
483
+ "learning_rate": 1.9206406067026506e-05,
484
+ "loss": 0.3097,
485
  "step": 680
486
  },
487
  {
488
+ "epoch": 5.52,
489
+ "grad_norm": 3.4684863090515137,
490
+ "learning_rate": 1.9162215329262115e-05,
491
+ "loss": 0.2868,
492
  "step": 690
493
  },
494
  {
495
+ "epoch": 5.6,
496
+ "grad_norm": 2.662013292312622,
497
+ "learning_rate": 1.9116881015346517e-05,
498
+ "loss": 0.3281,
499
  "step": 700
500
  },
501
  {
502
+ "epoch": 5.68,
503
+ "grad_norm": 2.223334312438965,
504
+ "learning_rate": 1.9070408783653627e-05,
505
+ "loss": 0.2799,
506
  "step": 710
507
  },
508
  {
509
+ "epoch": 5.76,
510
+ "grad_norm": 2.992532730102539,
511
+ "learning_rate": 1.9022804434585854e-05,
512
+ "loss": 0.3252,
513
  "step": 720
514
  },
515
  {
516
+ "epoch": 5.84,
517
+ "grad_norm": 1.6079481840133667,
518
+ "learning_rate": 1.8974073909850125e-05,
519
+ "loss": 0.2791,
520
  "step": 730
521
  },
522
  {
523
+ "epoch": 5.92,
524
+ "grad_norm": 2.97104549407959,
525
+ "learning_rate": 1.8924223291716274e-05,
526
+ "loss": 0.3029,
527
  "step": 740
528
  },
529
  {
530
+ "epoch": 6.0,
531
+ "grad_norm": 2.1889212131500244,
532
+ "learning_rate": 1.887325880225789e-05,
533
+ "loss": 0.293,
534
  "step": 750
535
  },
536
  {
537
+ "epoch": 6.08,
538
+ "grad_norm": 2.123610019683838,
539
+ "learning_rate": 1.882118680257572e-05,
540
+ "loss": 0.2308,
541
  "step": 760
542
  },
543
  {
544
+ "epoch": 6.16,
545
+ "grad_norm": 3.71563720703125,
546
+ "learning_rate": 1.8768013792003683e-05,
547
+ "loss": 0.216,
548
  "step": 770
549
  },
550
  {
551
+ "epoch": 6.24,
552
+ "grad_norm": 2.459091901779175,
553
+ "learning_rate": 1.8713746407297703e-05,
554
+ "loss": 0.1972,
555
  "step": 780
556
  },
557
  {
558
+ "epoch": 6.32,
559
+ "grad_norm": 2.669945240020752,
560
+ "learning_rate": 1.8658391421807313e-05,
561
+ "loss": 0.2478,
562
  "step": 790
563
  },
564
  {
565
+ "epoch": 6.4,
566
+ "grad_norm": 2.1581931114196777,
567
+ "learning_rate": 1.8601955744630255e-05,
568
+ "loss": 0.2295,
569
  "step": 800
570
  },
571
  {
572
+ "epoch": 6.48,
573
+ "grad_norm": 2.567908763885498,
574
+ "learning_rate": 1.8544446419750125e-05,
575
+ "loss": 0.2239,
576
  "step": 810
577
  },
578
  {
579
+ "epoch": 6.5600000000000005,
580
+ "grad_norm": 2.0205740928649902,
581
+ "learning_rate": 1.8485870625157186e-05,
582
+ "loss": 0.216,
583
  "step": 820
584
  },
585
  {
586
+ "epoch": 6.64,
587
+ "grad_norm": 2.2898030281066895,
588
+ "learning_rate": 1.8426235671952452e-05,
589
+ "loss": 0.2343,
590
  "step": 830
591
  },
592
  {
593
+ "epoch": 6.72,
594
+ "grad_norm": 2.2003941535949707,
595
+ "learning_rate": 1.836554900343514e-05,
596
+ "loss": 0.2076,
597
  "step": 840
598
  },
599
  {
600
+ "epoch": 6.8,
601
+ "grad_norm": 1.5117710828781128,
602
+ "learning_rate": 1.8303818194173665e-05,
603
+ "loss": 0.1904,
604
  "step": 850
605
  },
606
  {
607
+ "epoch": 6.88,
608
+ "grad_norm": 1.2647337913513184,
609
+ "learning_rate": 1.824105094906021e-05,
610
+ "loss": 0.2341,
611
  "step": 860
612
  },
613
  {
614
+ "epoch": 6.96,
615
+ "grad_norm": 2.00254225730896,
616
+ "learning_rate": 1.8177255102349047e-05,
617
+ "loss": 0.2288,
618
  "step": 870
619
  },
620
  {
621
+ "epoch": 7.04,
622
+ "grad_norm": 2.516129493713379,
623
+ "learning_rate": 1.8112438616678712e-05,
624
+ "loss": 0.1685,
625
  "step": 880
626
  },
627
  {
628
+ "epoch": 7.12,
629
+ "grad_norm": 3.6129150390625,
630
+ "learning_rate": 1.8046609582078147e-05,
631
+ "loss": 0.1625,
632
  "step": 890
633
  },
634
  {
635
+ "epoch": 7.2,
636
+ "grad_norm": 3.1554768085479736,
637
+ "learning_rate": 1.797977621495696e-05,
638
+ "loss": 0.2011,
639
  "step": 900
640
  },
641
  {
642
+ "epoch": 7.28,
643
+ "grad_norm": 3.8677282333374023,
644
+ "learning_rate": 1.7911946857079886e-05,
645
+ "loss": 0.1579,
646
  "step": 910
647
  },
648
  {
649
+ "epoch": 7.36,
650
+ "grad_norm": 3.062568426132202,
651
+ "learning_rate": 1.784312997452562e-05,
652
+ "loss": 0.1705,
653
  "step": 920
654
  },
655
  {
656
+ "epoch": 7.44,
657
+ "grad_norm": 1.8281478881835938,
658
+ "learning_rate": 1.777333415663014e-05,
659
+ "loss": 0.1708,
660
  "step": 930
661
  },
662
  {
663
+ "epoch": 7.52,
664
+ "grad_norm": 1.1312201023101807,
665
+ "learning_rate": 1.7702568114914607e-05,
666
+ "loss": 0.1644,
667
  "step": 940
668
  },
669
  {
670
+ "epoch": 7.6,
671
+ "grad_norm": 1.1482549905776978,
672
+ "learning_rate": 1.7630840681998068e-05,
673
+ "loss": 0.1297,
674
  "step": 950
675
  },
676
  {
677
+ "epoch": 7.68,
678
+ "grad_norm": 3.109949827194214,
679
+ "learning_rate": 1.755816081049501e-05,
680
+ "loss": 0.1812,
681
  "step": 960
682
  },
683
  {
684
+ "epoch": 7.76,
685
+ "grad_norm": 1.2507814168930054,
686
+ "learning_rate": 1.7484537571897943e-05,
687
+ "loss": 0.1331,
688
  "step": 970
689
  },
690
  {
691
+ "epoch": 7.84,
692
+ "grad_norm": 2.898353338241577,
693
+ "learning_rate": 1.740998015544514e-05,
694
+ "loss": 0.1625,
695
  "step": 980
696
  },
697
  {
698
+ "epoch": 7.92,
699
+ "grad_norm": 2.2018096446990967,
700
+ "learning_rate": 1.7334497866973716e-05,
701
+ "loss": 0.1441,
702
  "step": 990
703
  },
704
  {
705
+ "epoch": 8.0,
706
+ "grad_norm": 2.1120376586914062,
707
+ "learning_rate": 1.725810012775808e-05,
708
+ "loss": 0.1804,
709
  "step": 1000
710
  },
711
  {
712
+ "epoch": 8.08,
713
+ "grad_norm": 1.815804362297058,
714
+ "learning_rate": 1.7180796473334075e-05,
715
+ "loss": 0.1145,
716
  "step": 1010
717
  },
718
  {
719
+ "epoch": 8.16,
720
+ "grad_norm": 2.5671639442443848,
721
+ "learning_rate": 1.7102596552308765e-05,
722
+ "loss": 0.1094,
723
  "step": 1020
724
  },
725
  {
726
+ "epoch": 8.24,
727
+ "grad_norm": 0.9062528610229492,
728
+ "learning_rate": 1.7023510125156173e-05,
729
+ "loss": 0.0924,
730
  "step": 1030
731
  },
732
  {
733
+ "epoch": 8.32,
734
+ "grad_norm": 3.5460164546966553,
735
+ "learning_rate": 1.6943547062999027e-05,
736
+ "loss": 0.1323,
737
  "step": 1040
738
  },
739
  {
740
+ "epoch": 8.4,
741
+ "grad_norm": 2.856273889541626,
742
+ "learning_rate": 1.6862717346376706e-05,
743
+ "loss": 0.0938,
744
  "step": 1050
745
  },
746
  {
747
+ "epoch": 8.48,
748
+ "grad_norm": 3.177924633026123,
749
+ "learning_rate": 1.6781031063999515e-05,
750
+ "loss": 0.1204,
751
  "step": 1060
752
  },
753
  {
754
+ "epoch": 8.56,
755
+ "grad_norm": 1.3371851444244385,
756
+ "learning_rate": 1.6698498411489477e-05,
757
+ "loss": 0.1342,
758
  "step": 1070
759
  },
760
  {
761
+ "epoch": 8.64,
762
+ "grad_norm": 2.0833916664123535,
763
+ "learning_rate": 1.6615129690107773e-05,
764
+ "loss": 0.1337,
765
  "step": 1080
766
  },
767
  {
768
+ "epoch": 8.72,
769
+ "grad_norm": 2.4857378005981445,
770
+ "learning_rate": 1.6530935305469e-05,
771
+ "loss": 0.1233,
772
  "step": 1090
773
  },
774
  {
775
+ "epoch": 8.8,
776
+ "grad_norm": 1.7040643692016602,
777
+ "learning_rate": 1.6445925766242392e-05,
778
+ "loss": 0.121,
779
  "step": 1100
780
  },
781
  {
782
+ "epoch": 8.88,
783
+ "grad_norm": 1.2290680408477783,
784
+ "learning_rate": 1.6360111682840184e-05,
785
+ "loss": 0.1037,
786
  "step": 1110
787
  },
788
  {
789
+ "epoch": 8.96,
790
+ "grad_norm": 3.9770867824554443,
791
+ "learning_rate": 1.62735037660933e-05,
792
+ "loss": 0.1133,
793
  "step": 1120
794
  },
795
  {
796
+ "epoch": 9.04,
797
+ "grad_norm": 1.0533943176269531,
798
+ "learning_rate": 1.618611282591446e-05,
799
+ "loss": 0.0943,
800
  "step": 1130
801
  },
802
  {
803
+ "epoch": 9.12,
804
+ "grad_norm": 1.9520882368087769,
805
+ "learning_rate": 1.609794976994897e-05,
806
+ "loss": 0.0725,
807
  "step": 1140
808
  },
809
  {
810
+ "epoch": 9.2,
811
+ "grad_norm": 3.610280990600586,
812
+ "learning_rate": 1.600902560221329e-05,
813
+ "loss": 0.1015,
814
  "step": 1150
815
  },
816
  {
817
+ "epoch": 9.28,
818
+ "grad_norm": 1.3682546615600586,
819
+ "learning_rate": 1.5919351421721548e-05,
820
+ "loss": 0.0945,
821
  "step": 1160
822
  },
823
  {
824
+ "epoch": 9.36,
825
+ "grad_norm": 2.125917673110962,
826
+ "learning_rate": 1.5828938421100266e-05,
827
+ "loss": 0.0733,
828
  "step": 1170
829
  },
830
  {
831
+ "epoch": 9.44,
832
+ "grad_norm": 2.5206658840179443,
833
+ "learning_rate": 1.5737797885191316e-05,
834
+ "loss": 0.0646,
835
  "step": 1180
836
  },
837
  {
838
+ "epoch": 9.52,
839
+ "grad_norm": 2.5717523097991943,
840
+ "learning_rate": 1.5645941189643444e-05,
841
+ "loss": 0.0836,
842
  "step": 1190
843
  },
844
  {
845
+ "epoch": 9.6,
846
+ "grad_norm": 2.1724965572357178,
847
+ "learning_rate": 1.55533797994924e-05,
848
+ "loss": 0.0933,
849
  "step": 1200
850
  },
851
  {
852
+ "epoch": 9.68,
853
+ "grad_norm": 3.0643951892852783,
854
+ "learning_rate": 1.546012526772996e-05,
855
+ "loss": 0.083,
856
  "step": 1210
857
  },
858
  {
859
+ "epoch": 9.76,
860
+ "grad_norm": 1.3585115671157837,
861
+ "learning_rate": 1.5366189233861933e-05,
862
+ "loss": 0.0804,
863
  "step": 1220
864
  },
865
  {
866
+ "epoch": 9.84,
867
+ "grad_norm": 2.106403112411499,
868
+ "learning_rate": 1.5271583422455373e-05,
869
+ "loss": 0.0941,
870
  "step": 1230
871
  },
872
  {
873
+ "epoch": 9.92,
874
+ "grad_norm": 1.7128456830978394,
875
+ "learning_rate": 1.5176319641675213e-05,
876
+ "loss": 0.0668,
877
  "step": 1240
878
  },
879
  {
880
+ "epoch": 10.0,
881
+ "grad_norm": 1.7037241458892822,
882
+ "learning_rate": 1.5080409781810406e-05,
883
+ "loss": 0.1043,
884
  "step": 1250
885
  },
886
  {
887
+ "epoch": 10.08,
888
+ "grad_norm": 0.8989495038986206,
889
+ "learning_rate": 1.4983865813789869e-05,
890
+ "loss": 0.0521,
891
  "step": 1260
892
  },
893
  {
894
+ "epoch": 10.16,
895
+ "grad_norm": 1.3013043403625488,
896
+ "learning_rate": 1.488669978768833e-05,
897
+ "loss": 0.0647,
898
  "step": 1270
899
  },
900
  {
901
+ "epoch": 10.24,
902
+ "grad_norm": 0.6377755999565125,
903
+ "learning_rate": 1.47889238312223e-05,
904
+ "loss": 0.0367,
905
  "step": 1280
906
  },
907
  {
908
+ "epoch": 10.32,
909
+ "grad_norm": 2.2679171562194824,
910
+ "learning_rate": 1.4690550148236371e-05,
911
+ "loss": 0.0769,
912
  "step": 1290
913
  },
914
  {
915
+ "epoch": 10.4,
916
+ "grad_norm": 2.3292899131774902,
917
+ "learning_rate": 1.4591591017179993e-05,
918
+ "loss": 0.0908,
919
  "step": 1300
920
  },
921
  {
922
+ "epoch": 10.48,
923
+ "grad_norm": 1.8322817087173462,
924
+ "learning_rate": 1.4492058789574948e-05,
925
+ "loss": 0.0551,
926
  "step": 1310
927
  },
928
  {
929
+ "epoch": 10.56,
930
+ "grad_norm": 5.285430431365967,
931
+ "learning_rate": 1.4391965888473705e-05,
932
+ "loss": 0.0584,
933
  "step": 1320
934
  },
935
  {
936
+ "epoch": 10.64,
937
+ "grad_norm": 0.768731415271759,
938
+ "learning_rate": 1.4291324806908846e-05,
939
+ "loss": 0.0833,
940
  "step": 1330
941
  },
942
  {
943
+ "epoch": 10.72,
944
+ "grad_norm": 1.1357346773147583,
945
+ "learning_rate": 1.419014810633374e-05,
946
+ "loss": 0.0596,
947
  "step": 1340
948
  },
949
  {
950
+ "epoch": 10.8,
951
+ "grad_norm": 0.9390424489974976,
952
+ "learning_rate": 1.408844841505473e-05,
953
+ "loss": 0.0436,
954
  "step": 1350
955
  },
956
  {
957
+ "epoch": 10.88,
958
+ "grad_norm": 0.775092363357544,
959
+ "learning_rate": 1.3986238426654894e-05,
960
+ "loss": 0.0663,
961
  "step": 1360
962
  },
963
  {
964
+ "epoch": 10.96,
965
+ "grad_norm": 1.6158931255340576,
966
+ "learning_rate": 1.3883530898409736e-05,
967
+ "loss": 0.0438,
968
  "step": 1370
969
  },
970
  {
971
+ "epoch": 11.04,
972
+ "grad_norm": 1.8279812335968018,
973
+ "learning_rate": 1.3780338649694874e-05,
974
+ "loss": 0.0944,
975
  "step": 1380
976
  },
977
  {
978
+ "epoch": 11.12,
979
+ "grad_norm": 1.8661407232284546,
980
+ "learning_rate": 1.3676674560386018e-05,
981
+ "loss": 0.0398,
982
  "step": 1390
983
  },
984
  {
985
+ "epoch": 11.2,
986
+ "grad_norm": 1.8592077493667603,
987
+ "learning_rate": 1.357255156925136e-05,
988
+ "loss": 0.0561,
989
  "step": 1400
990
  },
991
  {
992
+ "epoch": 11.28,
993
+ "grad_norm": 3.479532241821289,
994
+ "learning_rate": 1.3467982672336633e-05,
995
+ "loss": 0.0408,
996
  "step": 1410
997
  },
998
  {
999
+ "epoch": 11.36,
1000
+ "grad_norm": 2.4556691646575928,
1001
+ "learning_rate": 1.336298092134302e-05,
1002
+ "loss": 0.0603,
1003
  "step": 1420
1004
  },
1005
  {
1006
+ "epoch": 11.44,
1007
+ "grad_norm": 2.028334379196167,
1008
+ "learning_rate": 1.325755942199812e-05,
1009
+ "loss": 0.0423,
1010
  "step": 1430
1011
  },
1012
  {
1013
+ "epoch": 11.52,
1014
+ "grad_norm": 0.6319141387939453,
1015
+ "learning_rate": 1.3151731332420152e-05,
1016
+ "loss": 0.0717,
1017
  "step": 1440
1018
  },
1019
  {
1020
+ "epoch": 11.6,
1021
+ "grad_norm": 1.3395280838012695,
1022
+ "learning_rate": 1.3045509861475645e-05,
1023
+ "loss": 0.0505,
1024
  "step": 1450
1025
  },
1026
  {
1027
+ "epoch": 11.68,
1028
+ "grad_norm": 2.619269847869873,
1029
+ "learning_rate": 1.293890826713077e-05,
1030
+ "loss": 0.0477,
1031
  "step": 1460
1032
  },
1033
  {
1034
+ "epoch": 11.76,
1035
+ "grad_norm": 2.6248464584350586,
1036
+ "learning_rate": 1.2831939854796567e-05,
1037
+ "loss": 0.0426,
1038
  "step": 1470
1039
  },
1040
  {
1041
+ "epoch": 11.84,
1042
+ "grad_norm": 1.672989845275879,
1043
+ "learning_rate": 1.2724617975668229e-05,
1044
+ "loss": 0.0533,
1045
  "step": 1480
1046
  },
1047
  {
1048
+ "epoch": 11.92,
1049
+ "grad_norm": 5.669078350067139,
1050
+ "learning_rate": 1.2616956025058688e-05,
1051
+ "loss": 0.0413,
1052
  "step": 1490
1053
  },
1054
  {
1055
+ "epoch": 12.0,
1056
+ "grad_norm": 2.987652063369751,
1057
+ "learning_rate": 1.2508967440726689e-05,
1058
+ "loss": 0.0442,
1059
  "step": 1500
1060
  },
1061
  {
1062
+ "epoch": 12.08,
1063
+ "grad_norm": 1.2909622192382812,
1064
+ "learning_rate": 1.2400665701199541e-05,
1065
+ "loss": 0.0565,
1066
  "step": 1510
1067
  },
1068
  {
1069
+ "epoch": 12.16,
1070
+ "grad_norm": 0.264878511428833,
1071
+ "learning_rate": 1.2292064324090842e-05,
1072
+ "loss": 0.0243,
1073
  "step": 1520
1074
  },
1075
  {
1076
+ "epoch": 12.24,
1077
+ "grad_norm": 2.0698373317718506,
1078
+ "learning_rate": 1.2183176864413262e-05,
1079
+ "loss": 0.0241,
1080
  "step": 1530
1081
  },
1082
  {
1083
+ "epoch": 12.32,
1084
+ "grad_norm": 0.20062156021595,
1085
+ "learning_rate": 1.2074016912886683e-05,
1086
+ "loss": 0.0204,
1087
  "step": 1540
1088
  },
1089
  {
1090
+ "epoch": 12.4,
1091
+ "grad_norm": 0.5076002478599548,
1092
+ "learning_rate": 1.1964598094241884e-05,
1093
+ "loss": 0.0462,
1094
  "step": 1550
1095
  },
1096
  {
1097
+ "epoch": 12.48,
1098
+ "grad_norm": 1.1509472131729126,
1099
+ "learning_rate": 1.1854934065519986e-05,
1100
+ "loss": 0.0302,
1101
  "step": 1560
1102
  },
1103
  {
1104
+ "epoch": 12.56,
1105
+ "grad_norm": 1.603930950164795,
1106
+ "learning_rate": 1.1745038514367853e-05,
1107
+ "loss": 0.0282,
1108
  "step": 1570
1109
  },
1110
  {
1111
+ "epoch": 12.64,
1112
+ "grad_norm": 0.9757488965988159,
1113
+ "learning_rate": 1.1634925157329668e-05,
1114
+ "loss": 0.0387,
1115
  "step": 1580
1116
  },
1117
  {
1118
+ "epoch": 12.72,
1119
+ "grad_norm": 1.058209776878357,
1120
+ "learning_rate": 1.1524607738134928e-05,
1121
+ "loss": 0.0445,
1122
  "step": 1590
1123
  },
1124
  {
1125
+ "epoch": 12.8,
1126
+ "grad_norm": 2.3291449546813965,
1127
+ "learning_rate": 1.141410002598301e-05,
1128
+ "loss": 0.0572,
1129
  "step": 1600
1130
  },
1131
  {
1132
+ "epoch": 12.88,
1133
+ "grad_norm": 0.8108656406402588,
1134
+ "learning_rate": 1.1303415813824599e-05,
1135
+ "loss": 0.0337,
1136
  "step": 1610
1137
  },
1138
  {
1139
+ "epoch": 12.96,
1140
+ "grad_norm": 6.551490783691406,
1141
+ "learning_rate": 1.1192568916640107e-05,
1142
+ "loss": 0.0337,
1143
  "step": 1620
1144
  },
1145
  {
1146
+ "epoch": 13.04,
1147
+ "grad_norm": 0.2524939477443695,
1148
+ "learning_rate": 1.1081573169715379e-05,
1149
+ "loss": 0.0218,
1150
  "step": 1630
1151
  },
1152
  {
1153
+ "epoch": 13.12,
1154
+ "grad_norm": 0.17596031725406647,
1155
+ "learning_rate": 1.0970442426914847e-05,
1156
+ "loss": 0.0132,
1157
  "step": 1640
1158
  },
1159
  {
1160
+ "epoch": 13.2,
1161
+ "grad_norm": 0.7680679559707642,
1162
+ "learning_rate": 1.085919055895237e-05,
1163
+ "loss": 0.0207,
1164
  "step": 1650
1165
  },
1166
  {
1167
+ "epoch": 13.28,
1168
+ "grad_norm": 0.24066166579723358,
1169
+ "learning_rate": 1.0747831451659967e-05,
1170
+ "loss": 0.0142,
1171
  "step": 1660
1172
  },
1173
  {
1174
+ "epoch": 13.36,
1175
+ "grad_norm": 3.3667285442352295,
1176
+ "learning_rate": 1.0636379004254665e-05,
1177
+ "loss": 0.0183,
1178
  "step": 1670
1179
  },
1180
  {
1181
+ "epoch": 13.44,
1182
+ "grad_norm": 1.8831534385681152,
1183
+ "learning_rate": 1.0524847127603677e-05,
1184
+ "loss": 0.0253,
1185
  "step": 1680
1186
  },
1187
  {
1188
+ "epoch": 13.52,
1189
+ "grad_norm": 1.329085111618042,
1190
+ "learning_rate": 1.0413249742488132e-05,
1191
+ "loss": 0.0118,
1192
  "step": 1690
1193
  },
1194
  {
1195
+ "epoch": 13.6,
1196
+ "grad_norm": 2.0776255130767822,
1197
+ "learning_rate": 1.030160077786556e-05,
1198
+ "loss": 0.0128,
1199
  "step": 1700
1200
  },
1201
  {
1202
+ "epoch": 13.68,
1203
+ "grad_norm": 1.02082097530365,
1204
+ "learning_rate": 1.0189914169131341e-05,
1205
+ "loss": 0.0406,
1206
  "step": 1710
1207
  },
1208
  {
1209
+ "epoch": 13.76,
1210
+ "grad_norm": 0.7947062253952026,
1211
+ "learning_rate": 1.0078203856379394e-05,
1212
+ "loss": 0.0183,
1213
  "step": 1720
1214
  },
1215
  {
1216
+ "epoch": 13.84,
1217
+ "grad_norm": 0.3098564147949219,
1218
+ "learning_rate": 9.966483782662233e-06,
1219
+ "loss": 0.0251,
1220
  "step": 1730
1221
  },
1222
  {
1223
+ "epoch": 13.92,
1224
+ "grad_norm": 0.4250534772872925,
1225
+ "learning_rate": 9.854767892250692e-06,
1226
+ "loss": 0.0394,
1227
  "step": 1740
1228
  },
1229
  {
1230
+ "epoch": 14.0,
1231
+ "grad_norm": 1.0927317142486572,
1232
+ "learning_rate": 9.743070128893452e-06,
1233
+ "loss": 0.0187,
1234
  "step": 1750
1235
  },
1236
  {
1237
+ "epoch": 14.08,
1238
+ "grad_norm": 0.5906019806861877,
1239
+ "learning_rate": 9.631404434076687e-06,
1240
+ "loss": 0.0042,
1241
  "step": 1760
1242
  },
1243
  {
1244
+ "epoch": 14.16,
1245
+ "grad_norm": 0.4497601389884949,
1246
+ "learning_rate": 9.519784745283956e-06,
1247
+ "loss": 0.0402,
1248
  "step": 1770
1249
  },
1250
  {
1251
+ "epoch": 14.24,
1252
+ "grad_norm": 0.2257859855890274,
1253
+ "learning_rate": 9.408224994256603e-06,
1254
+ "loss": 0.0039,
1255
  "step": 1780
1256
  },
1257
  {
1258
+ "epoch": 14.32,
1259
+ "grad_norm": 0.9509989023208618,
1260
+ "learning_rate": 9.296739105254869e-06,
1261
+ "loss": 0.0156,
1262
  "step": 1790
1263
  },
1264
  {
1265
+ "epoch": 14.4,
1266
+ "grad_norm": 1.7652183771133423,
1267
+ "learning_rate": 9.185340993319977e-06,
1268
+ "loss": 0.0121,
1269
  "step": 1800
1270
  },
1271
  {
1272
+ "epoch": 14.48,
1273
+ "grad_norm": 8.208316802978516,
1274
+ "learning_rate": 9.074044562537284e-06,
1275
+ "loss": 0.0389,
1276
  "step": 1810
1277
  },
1278
  {
1279
+ "epoch": 14.56,
1280
+ "grad_norm": 0.13252846896648407,
1281
+ "learning_rate": 8.962863704300893e-06,
1282
+ "loss": 0.011,
1283
  "step": 1820
1284
  },
1285
  {
1286
+ "epoch": 14.64,
1287
+ "grad_norm": 0.23998035490512848,
1288
+ "learning_rate": 8.851812295579789e-06,
1289
+ "loss": 0.0248,
1290
  "step": 1830
1291
  },
1292
  {
1293
+ "epoch": 14.72,
1294
+ "grad_norm": 0.10990134626626968,
1295
+ "learning_rate": 8.740904197185794e-06,
1296
+ "loss": 0.0204,
1297
  "step": 1840
1298
  },
1299
  {
1300
+ "epoch": 14.8,
1301
+ "grad_norm": 3.542644500732422,
1302
+ "learning_rate": 8.630153252043543e-06,
1303
+ "loss": 0.0163,
1304
  "step": 1850
1305
  },
1306
  {
1307
+ "epoch": 14.88,
1308
+ "grad_norm": 0.7273716926574707,
1309
+ "learning_rate": 8.519573283462688e-06,
1310
+ "loss": 0.0068,
1311
  "step": 1860
1312
  },
1313
  {
1314
+ "epoch": 14.96,
1315
+ "grad_norm": 0.036577966064214706,
1316
+ "learning_rate": 8.409178093412549e-06,
1317
+ "loss": 0.0084,
1318
  "step": 1870
1319
  },
1320
  {
1321
+ "epoch": 15.04,
1322
+ "grad_norm": 0.1642957627773285,
1323
+ "learning_rate": 8.298981460799426e-06,
1324
+ "loss": 0.0093,
1325
  "step": 1880
1326
  },
1327
  {
1328
+ "epoch": 15.12,
1329
+ "grad_norm": 0.11017989367246628,
1330
+ "learning_rate": 8.188997139746807e-06,
1331
+ "loss": 0.0076,
1332
  "step": 1890
1333
  },
1334
  {
1335
+ "epoch": 15.2,
1336
+ "grad_norm": 0.04475900158286095,
1337
+ "learning_rate": 8.079238857878631e-06,
1338
+ "loss": 0.0059,
1339
  "step": 1900
1340
  },
1341
  {
1342
+ "epoch": 15.28,
1343
+ "grad_norm": 1.0872098207473755,
1344
+ "learning_rate": 7.969720314605915e-06,
1345
+ "loss": 0.0051,
1346
  "step": 1910
1347
  },
1348
  {
1349
+ "epoch": 15.36,
1350
+ "grad_norm": 0.32729968428611755,
1351
+ "learning_rate": 7.860455179416837e-06,
1352
+ "loss": 0.0016,
1353
  "step": 1920
1354
  },
1355
  {
1356
+ "epoch": 15.44,
1357
+ "grad_norm": 1.3896279335021973,
1358
+ "learning_rate": 7.751457090170616e-06,
1359
+ "loss": 0.0066,
1360
  "step": 1930
1361
  },
1362
  {
1363
+ "epoch": 15.52,
1364
+ "grad_norm": 0.24193909764289856,
1365
+ "learning_rate": 7.642739651395295e-06,
1366
+ "loss": 0.0049,
1367
  "step": 1940
1368
  },
1369
  {
1370
+ "epoch": 15.6,
1371
+ "grad_norm": 0.0498543456196785,
1372
+ "learning_rate": 7.534316432589706e-06,
1373
+ "loss": 0.0065,
1374
  "step": 1950
1375
  },
1376
  {
1377
+ "epoch": 15.68,
1378
+ "grad_norm": 0.053042277693748474,
1379
+ "learning_rate": 7.426200966529795e-06,
1380
+ "loss": 0.0096,
1381
  "step": 1960
1382
  },
1383
  {
1384
+ "epoch": 15.76,
1385
+ "grad_norm": 0.6754854917526245,
1386
+ "learning_rate": 7.318406747579556e-06,
1387
+ "loss": 0.0074,
1388
  "step": 1970
1389
  },
1390
  {
1391
+ "epoch": 15.84,
1392
+ "grad_norm": 0.05438900366425514,
1393
+ "learning_rate": 7.210947230006713e-06,
1394
+ "loss": 0.0025,
1395
  "step": 1980
1396
  },
1397
  {
1398
+ "epoch": 15.92,
1399
+ "grad_norm": 0.0401092991232872,
1400
+ "learning_rate": 7.103835826303451e-06,
1401
+ "loss": 0.0022,
1402
  "step": 1990
1403
  },
1404
  {
1405
+ "epoch": 16.0,
1406
+ "grad_norm": 0.05556059256196022,
1407
+ "learning_rate": 6.997085905512346e-06,
1408
+ "loss": 0.0049,
1409
  "step": 2000
1410
  }
1411
  ],
1412
  "logging_steps": 10,
1413
+ "max_steps": 3125,
1414
  "num_input_tokens_seen": 0,
1415
+ "num_train_epochs": 25,
1416
  "save_steps": 1000,
1417
  "stateful_callbacks": {
1418
  "TrainerControl": {
 
1426
  "attributes": {}
1427
  }
1428
  },
1429
+ "total_flos": 7.978267535894118e+16,
1430
+ "train_batch_size": 2,
1431
  "trial_name": null,
1432
  "trial_params": null
1433
  }
checkpoint-2000/training_args.bin CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:fe16ad85cf67a1d616e8a664749a043a7e42845bd6dfe8881e3e3c947d23497a
3
  size 5624
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:27ff69d62d45c0f05227ab8bfdfb70f5de878e29a775a5568fe676d60d677a54
3
  size 5624
checkpoint-3000/adapter_config.json CHANGED
@@ -19,7 +19,7 @@
19
  "r": 16,
20
  "rank_pattern": {},
21
  "revision": null,
22
- "target_modules": "^(?!.*visual.patch_embed|visual.blocks).*(?:o_proj|v_proj|down_proj|k_proj|q_proj|up_proj|gate_proj).*",
23
  "task_type": "CAUSAL_LM",
24
  "use_dora": false,
25
  "use_rslora": false
 
19
  "r": 16,
20
  "rank_pattern": {},
21
  "revision": null,
22
+ "target_modules": "^(?!.*patch_embed).*(?:o_proj|fc2|v_proj|down_proj|q_proj|up_proj|proj|qkv|gate_proj|k_proj|fc1).*",
23
  "task_type": "CAUSAL_LM",
24
  "use_dora": false,
25
  "use_rslora": false
checkpoint-3000/adapter_model.safetensors CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:9125f43477e44442eeeb0a5636962102d5fa86da13422a8be4b72a35cab5fd60
3
- size 73911112
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:95b37ec047318c37688632e44d0f926bde368be8bff784e13de9df5db55667b9
3
+ size 115886968
checkpoint-3000/optimizer.pt CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:5f287e8823b2411b8fb6a85cc5745d8b1b02be79332d47db6aa61959714c480e
3
- size 148047978
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:ba5668dde14f2cf2e520e246443545e149420e503334da71adf03bbe21529ec6
3
+ size 232150634
checkpoint-3000/rng_state_0.pth CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:6f55565ad4212d927304dc332f26c49b3f14337c859de7f43c26f8335e8cce96
3
  size 14512
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:23c5c1a139c10fe0906308288b9990e59b0830f85f1180b142012af4957b1ba2
3
  size 14512
checkpoint-3000/rng_state_1.pth CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:913a52c1867f7a99cedc95cd6ea7ae8a4c6c72231defe91735403a490336b526
3
  size 14512
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:e02bd5e21ab993cf26deb90360061cfc27b8440d255a29134ff03a42f4f9abff
3
  size 14512
checkpoint-3000/scheduler.pt CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:b1a53609d86d3b4ca9a96f73eb6c1136af30910ebd5b99c12f8da4ada9da8fd9
3
  size 1064
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:784c355c48bea6b46510dc8392c7ddb37d2c79801bfb385e330fc037ec63a968
3
  size 1064
checkpoint-3000/trainer_state.json CHANGED
@@ -1,7 +1,7 @@
1
  {
2
  "best_metric": null,
3
  "best_model_checkpoint": null,
4
- "epoch": 35.714285714285715,
5
  "eval_steps": 500,
6
  "global_step": 3000,
7
  "is_hyper_param_search": false,
@@ -9,2110 +9,2110 @@
9
  "is_world_process_zero": true,
10
  "log_history": [
11
  {
12
- "epoch": 0.11904761904761904,
13
- "grad_norm": 5.44253396987915,
14
- "learning_rate": 5.952380952380953e-07,
15
- "loss": 1.5321,
16
  "step": 10
17
  },
18
  {
19
- "epoch": 0.23809523809523808,
20
- "grad_norm": 2.2140841484069824,
21
- "learning_rate": 1.1904761904761906e-06,
22
- "loss": 1.1934,
23
  "step": 20
24
  },
25
  {
26
- "epoch": 0.35714285714285715,
27
- "grad_norm": 1.413270115852356,
28
- "learning_rate": 1.7857142857142859e-06,
29
- "loss": 0.9128,
30
  "step": 30
31
  },
32
  {
33
- "epoch": 0.47619047619047616,
34
- "grad_norm": 1.0657840967178345,
35
- "learning_rate": 2.380952380952381e-06,
36
- "loss": 0.7772,
37
  "step": 40
38
  },
39
  {
40
- "epoch": 0.5952380952380952,
41
- "grad_norm": 0.9593137502670288,
42
- "learning_rate": 2.9761904761904763e-06,
43
- "loss": 0.7849,
44
  "step": 50
45
  },
46
  {
47
- "epoch": 0.7142857142857143,
48
- "grad_norm": 1.134382724761963,
49
- "learning_rate": 3.5714285714285718e-06,
50
- "loss": 0.716,
51
  "step": 60
52
  },
53
  {
54
- "epoch": 0.8333333333333334,
55
- "grad_norm": 0.618364691734314,
56
- "learning_rate": 4.166666666666667e-06,
57
- "loss": 0.6678,
58
  "step": 70
59
  },
60
  {
61
- "epoch": 0.9523809523809523,
62
- "grad_norm": 0.7465847134590149,
63
- "learning_rate": 4.761904761904762e-06,
64
- "loss": 0.671,
65
  "step": 80
66
  },
67
  {
68
- "epoch": 1.0714285714285714,
69
- "grad_norm": 0.6726512908935547,
70
- "learning_rate": 5.357142857142857e-06,
71
- "loss": 0.6186,
72
  "step": 90
73
  },
74
  {
75
- "epoch": 1.1904761904761905,
76
- "grad_norm": 0.6353351473808289,
77
- "learning_rate": 5.9523809523809525e-06,
78
- "loss": 0.5818,
79
  "step": 100
80
  },
81
  {
82
- "epoch": 1.3095238095238095,
83
- "grad_norm": 0.7020516991615295,
84
- "learning_rate": 6.547619047619048e-06,
85
- "loss": 0.6056,
86
  "step": 110
87
  },
88
  {
89
- "epoch": 1.4285714285714286,
90
- "grad_norm": 1.1165708303451538,
91
- "learning_rate": 7.1428571428571436e-06,
92
- "loss": 0.6225,
93
  "step": 120
94
  },
95
  {
96
- "epoch": 1.5476190476190477,
97
- "grad_norm": 1.7502918243408203,
98
- "learning_rate": 7.738095238095238e-06,
99
- "loss": 0.6321,
100
  "step": 130
101
  },
102
  {
103
- "epoch": 1.6666666666666665,
104
- "grad_norm": 0.6962683796882629,
105
- "learning_rate": 8.333333333333334e-06,
106
- "loss": 0.5751,
107
  "step": 140
108
  },
109
  {
110
- "epoch": 1.7857142857142856,
111
- "grad_norm": 0.876686155796051,
112
- "learning_rate": 8.92857142857143e-06,
113
- "loss": 0.5974,
114
  "step": 150
115
  },
116
  {
117
- "epoch": 1.9047619047619047,
118
- "grad_norm": 0.5832173228263855,
119
- "learning_rate": 9.523809523809525e-06,
120
- "loss": 0.6085,
121
  "step": 160
122
  },
123
  {
124
- "epoch": 2.0238095238095237,
125
- "grad_norm": 0.44337984919548035,
126
- "learning_rate": 1.011904761904762e-05,
127
- "loss": 0.5713,
128
  "step": 170
129
  },
130
  {
131
- "epoch": 2.142857142857143,
132
- "grad_norm": 1.2403652667999268,
133
- "learning_rate": 1.0714285714285714e-05,
134
- "loss": 0.5165,
135
  "step": 180
136
  },
137
  {
138
- "epoch": 2.261904761904762,
139
- "grad_norm": 0.7495181560516357,
140
- "learning_rate": 1.130952380952381e-05,
141
- "loss": 0.5448,
142
  "step": 190
143
  },
144
  {
145
- "epoch": 2.380952380952381,
146
- "grad_norm": 0.8501800298690796,
147
- "learning_rate": 1.1904761904761905e-05,
148
- "loss": 0.5122,
149
  "step": 200
150
  },
151
  {
152
- "epoch": 2.5,
153
- "grad_norm": 0.8133195638656616,
154
- "learning_rate": 1.25e-05,
155
- "loss": 0.5337,
156
  "step": 210
157
  },
158
  {
159
- "epoch": 2.619047619047619,
160
- "grad_norm": 1.3526533842086792,
161
- "learning_rate": 1.3095238095238096e-05,
162
- "loss": 0.4943,
163
  "step": 220
164
  },
165
  {
166
- "epoch": 2.738095238095238,
167
- "grad_norm": 1.0535343885421753,
168
- "learning_rate": 1.3690476190476192e-05,
169
- "loss": 0.521,
170
  "step": 230
171
  },
172
  {
173
- "epoch": 2.857142857142857,
174
- "grad_norm": 1.2215440273284912,
175
- "learning_rate": 1.4285714285714287e-05,
176
- "loss": 0.5461,
177
  "step": 240
178
  },
179
  {
180
- "epoch": 2.9761904761904763,
181
- "grad_norm": 1.1781115531921387,
182
- "learning_rate": 1.4880952380952383e-05,
183
- "loss": 0.5346,
184
  "step": 250
185
  },
186
  {
187
- "epoch": 3.0952380952380953,
188
- "grad_norm": 0.762434184551239,
189
- "learning_rate": 1.5476190476190476e-05,
190
- "loss": 0.4855,
191
  "step": 260
192
  },
193
  {
194
- "epoch": 3.2142857142857144,
195
- "grad_norm": 1.287593126296997,
196
- "learning_rate": 1.6071428571428572e-05,
197
- "loss": 0.473,
198
  "step": 270
199
  },
200
  {
201
- "epoch": 3.3333333333333335,
202
- "grad_norm": 0.9310547113418579,
203
- "learning_rate": 1.6666666666666667e-05,
204
- "loss": 0.4201,
205
  "step": 280
206
  },
207
  {
208
- "epoch": 3.4523809523809526,
209
- "grad_norm": 1.1341677904129028,
210
- "learning_rate": 1.7261904761904763e-05,
211
- "loss": 0.3884,
212
  "step": 290
213
  },
214
  {
215
- "epoch": 3.571428571428571,
216
- "grad_norm": 1.483049750328064,
217
- "learning_rate": 1.785714285714286e-05,
218
- "loss": 0.4662,
219
  "step": 300
220
  },
221
  {
222
- "epoch": 3.6904761904761907,
223
- "grad_norm": 0.9312372803688049,
224
- "learning_rate": 1.8452380952380954e-05,
225
- "loss": 0.4688,
226
  "step": 310
227
  },
228
  {
229
- "epoch": 3.8095238095238093,
230
- "grad_norm": 1.0440223217010498,
231
- "learning_rate": 1.904761904761905e-05,
232
- "loss": 0.4727,
233
  "step": 320
234
  },
235
  {
236
- "epoch": 3.928571428571429,
237
- "grad_norm": 1.20750892162323,
238
- "learning_rate": 1.9642857142857145e-05,
239
- "loss": 0.4062,
240
  "step": 330
241
  },
242
  {
243
- "epoch": 4.0476190476190474,
244
- "grad_norm": 1.5511012077331543,
245
- "learning_rate": 1.999991365731819e-05,
246
- "loss": 0.4459,
247
  "step": 340
248
  },
249
  {
250
- "epoch": 4.166666666666667,
251
- "grad_norm": 1.3020656108856201,
252
- "learning_rate": 1.9998942319271076e-05,
253
- "loss": 0.3667,
254
  "step": 350
255
  },
256
  {
257
- "epoch": 4.285714285714286,
258
- "grad_norm": 0.9357143640518188,
259
- "learning_rate": 1.9996891820008165e-05,
260
- "loss": 0.3703,
261
  "step": 360
262
  },
263
  {
264
- "epoch": 4.404761904761905,
265
- "grad_norm": 1.120471477508545,
266
- "learning_rate": 1.9993762380834786e-05,
267
- "loss": 0.3797,
268
  "step": 370
269
  },
270
  {
271
- "epoch": 4.523809523809524,
272
- "grad_norm": 1.4476983547210693,
273
- "learning_rate": 1.9989554339503612e-05,
274
- "loss": 0.366,
275
  "step": 380
276
  },
277
  {
278
- "epoch": 4.642857142857143,
279
- "grad_norm": 0.8721790313720703,
280
- "learning_rate": 1.998426815017817e-05,
281
- "loss": 0.3914,
282
  "step": 390
283
  },
284
  {
285
- "epoch": 4.761904761904762,
286
- "grad_norm": 1.4711418151855469,
287
- "learning_rate": 1.997790438338385e-05,
288
- "loss": 0.3665,
289
  "step": 400
290
  },
291
  {
292
- "epoch": 4.880952380952381,
293
- "grad_norm": 1.216174840927124,
294
- "learning_rate": 1.9970463725946335e-05,
295
- "loss": 0.345,
296
  "step": 410
297
  },
298
  {
299
- "epoch": 5.0,
300
- "grad_norm": 0.8730125427246094,
301
- "learning_rate": 1.9962847054074817e-05,
302
- "loss": 0.3732,
303
  "step": 420
304
  },
305
  {
306
- "epoch": 5.119047619047619,
307
- "grad_norm": 1.7755807638168335,
308
- "learning_rate": 1.9953362612814294e-05,
309
- "loss": 0.2578,
310
  "step": 430
311
  },
312
  {
313
- "epoch": 5.238095238095238,
314
- "grad_norm": 1.7037990093231201,
315
- "learning_rate": 1.9942803929643423e-05,
316
- "loss": 0.2561,
317
  "step": 440
318
  },
319
  {
320
- "epoch": 5.357142857142857,
321
- "grad_norm": 1.140688180923462,
322
- "learning_rate": 1.9931172144134874e-05,
323
- "loss": 0.2525,
324
  "step": 450
325
  },
326
  {
327
- "epoch": 5.476190476190476,
328
- "grad_norm": 1.492291808128357,
329
- "learning_rate": 1.99184685116786e-05,
330
- "loss": 0.2727,
331
  "step": 460
332
  },
333
  {
334
- "epoch": 5.595238095238095,
335
- "grad_norm": 1.1798468828201294,
336
- "learning_rate": 1.9904694403346362e-05,
337
- "loss": 0.291,
338
  "step": 470
339
  },
340
  {
341
- "epoch": 5.714285714285714,
342
- "grad_norm": 1.1415112018585205,
343
- "learning_rate": 1.9889851305743742e-05,
344
- "loss": 0.2845,
345
  "step": 480
346
  },
347
  {
348
- "epoch": 5.833333333333333,
349
- "grad_norm": 1.8219681978225708,
350
- "learning_rate": 1.9873940820849714e-05,
351
- "loss": 0.2746,
352
  "step": 490
353
  },
354
  {
355
- "epoch": 5.9523809523809526,
356
- "grad_norm": 1.3858968019485474,
357
- "learning_rate": 1.9856964665843717e-05,
358
- "loss": 0.2861,
359
  "step": 500
360
  },
361
  {
362
- "epoch": 6.071428571428571,
363
- "grad_norm": 1.2878497838974,
364
- "learning_rate": 1.9838924672920344e-05,
365
- "loss": 0.239,
366
  "step": 510
367
  },
368
  {
369
- "epoch": 6.190476190476191,
370
- "grad_norm": 2.5550217628479004,
371
- "learning_rate": 1.9819822789091597e-05,
372
- "loss": 0.1605,
373
  "step": 520
374
  },
375
  {
376
- "epoch": 6.309523809523809,
377
- "grad_norm": 2.016493082046509,
378
- "learning_rate": 1.9799661075976754e-05,
379
- "loss": 0.2069,
380
  "step": 530
381
  },
382
  {
383
- "epoch": 6.428571428571429,
384
- "grad_norm": 1.8905669450759888,
385
- "learning_rate": 1.977844170957984e-05,
386
- "loss": 0.1953,
387
  "step": 540
388
  },
389
  {
390
- "epoch": 6.5476190476190474,
391
- "grad_norm": 1.6992967128753662,
392
- "learning_rate": 1.9756166980054812e-05,
393
- "loss": 0.1837,
394
  "step": 550
395
  },
396
  {
397
- "epoch": 6.666666666666667,
398
- "grad_norm": 1.893633246421814,
399
- "learning_rate": 1.973283929145836e-05,
400
- "loss": 0.1947,
401
  "step": 560
402
  },
403
  {
404
- "epoch": 6.785714285714286,
405
- "grad_norm": 1.6204354763031006,
406
- "learning_rate": 1.9708461161490456e-05,
407
- "loss": 0.1836,
408
  "step": 570
409
  },
410
  {
411
- "epoch": 6.904761904761905,
412
- "grad_norm": 2.0902771949768066,
413
- "learning_rate": 1.9683035221222617e-05,
414
- "loss": 0.1971,
415
  "step": 580
416
  },
417
  {
418
- "epoch": 7.023809523809524,
419
- "grad_norm": 2.2528581619262695,
420
- "learning_rate": 1.9656564214813948e-05,
421
- "loss": 0.1864,
422
  "step": 590
423
  },
424
  {
425
- "epoch": 7.142857142857143,
426
- "grad_norm": 0.9720670580863953,
427
- "learning_rate": 1.9629050999214977e-05,
428
- "loss": 0.114,
429
  "step": 600
430
  },
431
  {
432
- "epoch": 7.261904761904762,
433
- "grad_norm": 1.1338551044464111,
434
- "learning_rate": 1.960049854385929e-05,
435
- "loss": 0.1331,
436
  "step": 610
437
  },
438
  {
439
- "epoch": 7.380952380952381,
440
- "grad_norm": 1.8178822994232178,
441
- "learning_rate": 1.957090993034306e-05,
442
- "loss": 0.1144,
443
  "step": 620
444
  },
445
  {
446
- "epoch": 7.5,
447
- "grad_norm": 2.2525339126586914,
448
- "learning_rate": 1.9540288352092474e-05,
449
- "loss": 0.1298,
450
  "step": 630
451
  },
452
  {
453
- "epoch": 7.619047619047619,
454
- "grad_norm": 2.6778416633605957,
455
- "learning_rate": 1.9508637114019037e-05,
456
- "loss": 0.131,
457
  "step": 640
458
  },
459
  {
460
- "epoch": 7.738095238095238,
461
- "grad_norm": 1.2893160581588745,
462
- "learning_rate": 1.947595963216292e-05,
463
- "loss": 0.1332,
464
  "step": 650
465
  },
466
  {
467
- "epoch": 7.857142857142857,
468
- "grad_norm": 2.0039525032043457,
469
- "learning_rate": 1.9442259433324247e-05,
470
- "loss": 0.1385,
471
  "step": 660
472
  },
473
  {
474
- "epoch": 7.976190476190476,
475
- "grad_norm": 1.5008896589279175,
476
- "learning_rate": 1.9407540154682473e-05,
477
- "loss": 0.1564,
478
  "step": 670
479
  },
480
  {
481
- "epoch": 8.095238095238095,
482
- "grad_norm": 1.5358928442001343,
483
- "learning_rate": 1.9371805543403826e-05,
484
- "loss": 0.0916,
485
  "step": 680
486
  },
487
  {
488
- "epoch": 8.214285714285714,
489
- "grad_norm": 1.3121554851531982,
490
- "learning_rate": 1.9335059456236877e-05,
491
- "loss": 0.0928,
492
  "step": 690
493
  },
494
  {
495
- "epoch": 8.333333333333334,
496
- "grad_norm": 1.5265170335769653,
497
- "learning_rate": 1.9297305859096305e-05,
498
- "loss": 0.0858,
499
  "step": 700
500
  },
501
  {
502
- "epoch": 8.452380952380953,
503
- "grad_norm": 2.1085143089294434,
504
- "learning_rate": 1.925854882663486e-05,
505
- "loss": 0.0821,
506
  "step": 710
507
  },
508
  {
509
- "epoch": 8.571428571428571,
510
- "grad_norm": 1.8447340726852417,
511
- "learning_rate": 1.9218792541803582e-05,
512
- "loss": 0.1064,
513
  "step": 720
514
  },
515
  {
516
- "epoch": 8.69047619047619,
517
- "grad_norm": 1.8958710432052612,
518
- "learning_rate": 1.9178041295400383e-05,
519
- "loss": 0.082,
520
  "step": 730
521
  },
522
  {
523
- "epoch": 8.80952380952381,
524
- "grad_norm": 1.9607478380203247,
525
- "learning_rate": 1.9136299485606898e-05,
526
- "loss": 0.0901,
527
  "step": 740
528
  },
529
  {
530
- "epoch": 8.928571428571429,
531
- "grad_norm": 2.034801959991455,
532
- "learning_rate": 1.9093571617513853e-05,
533
- "loss": 0.1067,
534
  "step": 750
535
  },
536
  {
537
- "epoch": 9.047619047619047,
538
- "grad_norm": 1.1250438690185547,
539
- "learning_rate": 1.90498623026348e-05,
540
- "loss": 0.0558,
541
  "step": 760
542
  },
543
  {
544
- "epoch": 9.166666666666666,
545
- "grad_norm": 3.2768161296844482,
546
- "learning_rate": 1.9005176258408434e-05,
547
- "loss": 0.0464,
548
  "step": 770
549
  },
550
  {
551
- "epoch": 9.285714285714286,
552
- "grad_norm": 1.0632288455963135,
553
- "learning_rate": 1.8959518307689444e-05,
554
- "loss": 0.0499,
555
  "step": 780
556
  },
557
  {
558
- "epoch": 9.404761904761905,
559
- "grad_norm": 1.1634855270385742,
560
- "learning_rate": 1.8912893378227984e-05,
561
- "loss": 0.0665,
562
  "step": 790
563
  },
564
  {
565
- "epoch": 9.523809523809524,
566
- "grad_norm": 2.089411497116089,
567
- "learning_rate": 1.8865306502137852e-05,
568
- "loss": 0.0715,
569
  "step": 800
570
  },
571
  {
572
- "epoch": 9.642857142857142,
573
- "grad_norm": 0.46574532985687256,
574
- "learning_rate": 1.8816762815353368e-05,
575
- "loss": 0.0453,
576
  "step": 810
577
  },
578
  {
579
- "epoch": 9.761904761904763,
580
- "grad_norm": 2.3916738033294678,
581
- "learning_rate": 1.876726755707508e-05,
582
- "loss": 0.1086,
583
  "step": 820
584
  },
585
  {
586
- "epoch": 9.880952380952381,
587
- "grad_norm": 1.7373374700546265,
588
- "learning_rate": 1.8716826069204297e-05,
589
- "loss": 0.1003,
590
  "step": 830
591
  },
592
  {
593
- "epoch": 10.0,
594
- "grad_norm": 1.2700127363204956,
595
- "learning_rate": 1.8665443795766566e-05,
596
- "loss": 0.054,
597
  "step": 840
598
  },
599
  {
600
- "epoch": 10.119047619047619,
601
- "grad_norm": 0.9935422539710999,
602
- "learning_rate": 1.8613126282324092e-05,
603
- "loss": 0.0403,
604
  "step": 850
605
  },
606
  {
607
- "epoch": 10.238095238095237,
608
- "grad_norm": 1.3984347581863403,
609
- "learning_rate": 1.855987917537724e-05,
610
- "loss": 0.0584,
611
  "step": 860
612
  },
613
  {
614
- "epoch": 10.357142857142858,
615
- "grad_norm": 2.4390032291412354,
616
- "learning_rate": 1.8505708221755106e-05,
617
- "loss": 0.0672,
618
  "step": 870
619
  },
620
  {
621
- "epoch": 10.476190476190476,
622
- "grad_norm": 0.9531667828559875,
623
- "learning_rate": 1.8450619267995283e-05,
624
- "loss": 0.0425,
625
  "step": 880
626
  },
627
  {
628
- "epoch": 10.595238095238095,
629
- "grad_norm": 0.42088013887405396,
630
- "learning_rate": 1.8394618259712866e-05,
631
- "loss": 0.0399,
632
  "step": 890
633
  },
634
  {
635
- "epoch": 10.714285714285714,
636
- "grad_norm": 0.8044432997703552,
637
- "learning_rate": 1.8337711240958734e-05,
638
- "loss": 0.0516,
639
  "step": 900
640
  },
641
  {
642
- "epoch": 10.833333333333334,
643
- "grad_norm": 1.7414997816085815,
644
- "learning_rate": 1.827990435356725e-05,
645
- "loss": 0.043,
646
  "step": 910
647
  },
648
  {
649
- "epoch": 10.952380952380953,
650
- "grad_norm": 0.6912918090820312,
651
- "learning_rate": 1.822120383649339e-05,
652
- "loss": 0.0444,
653
  "step": 920
654
  },
655
  {
656
- "epoch": 11.071428571428571,
657
- "grad_norm": 1.3013064861297607,
658
- "learning_rate": 1.8161616025139375e-05,
659
- "loss": 0.0436,
660
  "step": 930
661
  },
662
  {
663
- "epoch": 11.19047619047619,
664
- "grad_norm": 1.2811461687088013,
665
- "learning_rate": 1.8101147350670905e-05,
666
- "loss": 0.039,
667
  "step": 940
668
  },
669
  {
670
- "epoch": 11.30952380952381,
671
- "grad_norm": 6.565953254699707,
672
- "learning_rate": 1.8039804339323083e-05,
673
- "loss": 0.0411,
674
  "step": 950
675
  },
676
  {
677
- "epoch": 11.428571428571429,
678
- "grad_norm": 1.2721556425094604,
679
- "learning_rate": 1.7977593611696017e-05,
680
- "loss": 0.0242,
681
  "step": 960
682
  },
683
  {
684
- "epoch": 11.547619047619047,
685
- "grad_norm": 1.6954939365386963,
686
- "learning_rate": 1.791452188204031e-05,
687
- "loss": 0.0421,
688
  "step": 970
689
  },
690
  {
691
- "epoch": 11.666666666666666,
692
- "grad_norm": 1.3118443489074707,
693
- "learning_rate": 1.7850595957532394e-05,
694
- "loss": 0.0524,
695
  "step": 980
696
  },
697
  {
698
- "epoch": 11.785714285714286,
699
- "grad_norm": 0.5428506731987,
700
- "learning_rate": 1.778582273753984e-05,
701
- "loss": 0.0484,
702
  "step": 990
703
  },
704
  {
705
- "epoch": 11.904761904761905,
706
- "grad_norm": 0.4753515124320984,
707
- "learning_rate": 1.772020921287674e-05,
708
- "loss": 0.0347,
709
  "step": 1000
710
  },
711
  {
712
- "epoch": 12.023809523809524,
713
- "grad_norm": 1.2880959510803223,
714
- "learning_rate": 1.7653762465049185e-05,
715
- "loss": 0.0408,
716
  "step": 1010
717
  },
718
  {
719
- "epoch": 12.142857142857142,
720
- "grad_norm": 2.5490682125091553,
721
- "learning_rate": 1.7586489665491013e-05,
722
- "loss": 0.0461,
723
  "step": 1020
724
  },
725
  {
726
- "epoch": 12.261904761904763,
727
- "grad_norm": 0.1983003169298172,
728
- "learning_rate": 1.7518398074789776e-05,
729
- "loss": 0.0391,
730
  "step": 1030
731
  },
732
  {
733
- "epoch": 12.380952380952381,
734
- "grad_norm": 1.2978509664535522,
735
- "learning_rate": 1.7449495041903136e-05,
736
- "loss": 0.0272,
737
  "step": 1040
738
  },
739
  {
740
- "epoch": 12.5,
741
- "grad_norm": 0.3622434437274933,
742
- "learning_rate": 1.737978800336572e-05,
743
- "loss": 0.0431,
744
  "step": 1050
745
  },
746
  {
747
- "epoch": 12.619047619047619,
748
- "grad_norm": 2.1278223991394043,
749
- "learning_rate": 1.7309284482486494e-05,
750
- "loss": 0.0241,
751
  "step": 1060
752
  },
753
  {
754
- "epoch": 12.738095238095237,
755
- "grad_norm": 2.409932851791382,
756
- "learning_rate": 1.723799208853681e-05,
757
- "loss": 0.0247,
758
  "step": 1070
759
  },
760
  {
761
- "epoch": 12.857142857142858,
762
- "grad_norm": 0.8543738126754761,
763
- "learning_rate": 1.7165918515929147e-05,
764
- "loss": 0.0385,
765
  "step": 1080
766
  },
767
  {
768
- "epoch": 12.976190476190476,
769
- "grad_norm": 0.301162451505661,
770
- "learning_rate": 1.7093071543386667e-05,
771
- "loss": 0.0362,
772
  "step": 1090
773
  },
774
  {
775
- "epoch": 13.095238095238095,
776
- "grad_norm": 0.9338281750679016,
777
- "learning_rate": 1.7019459033103684e-05,
778
- "loss": 0.0297,
779
  "step": 1100
780
  },
781
  {
782
- "epoch": 13.214285714285714,
783
- "grad_norm": 1.0773793458938599,
784
- "learning_rate": 1.6945088929897115e-05,
785
- "loss": 0.0312,
786
  "step": 1110
787
  },
788
  {
789
- "epoch": 13.333333333333334,
790
- "grad_norm": 0.34426188468933105,
791
- "learning_rate": 1.686996926034902e-05,
792
- "loss": 0.0146,
793
  "step": 1120
794
  },
795
  {
796
- "epoch": 13.452380952380953,
797
- "grad_norm": 0.22328321635723114,
798
- "learning_rate": 1.67941081319403e-05,
799
- "loss": 0.0195,
800
  "step": 1130
801
  },
802
  {
803
- "epoch": 13.571428571428571,
804
- "grad_norm": 0.5331404805183411,
805
- "learning_rate": 1.6717513732175686e-05,
806
- "loss": 0.025,
807
  "step": 1140
808
  },
809
  {
810
- "epoch": 13.69047619047619,
811
- "grad_norm": 2.170349359512329,
812
- "learning_rate": 1.6640194327700087e-05,
813
- "loss": 0.0132,
814
  "step": 1150
815
  },
816
  {
817
- "epoch": 13.80952380952381,
818
- "grad_norm": 1.2327094078063965,
819
- "learning_rate": 1.6562158263406376e-05,
820
- "loss": 0.0175,
821
  "step": 1160
822
  },
823
  {
824
- "epoch": 13.928571428571429,
825
- "grad_norm": 0.2306894212961197,
826
- "learning_rate": 1.6483413961534764e-05,
827
- "loss": 0.0195,
828
  "step": 1170
829
  },
830
  {
831
- "epoch": 14.047619047619047,
832
- "grad_norm": 0.8987044095993042,
833
- "learning_rate": 1.640396992076379e-05,
834
- "loss": 0.0186,
835
  "step": 1180
836
  },
837
  {
838
- "epoch": 14.166666666666666,
839
- "grad_norm": 1.013019323348999,
840
- "learning_rate": 1.6323834715293094e-05,
841
- "loss": 0.0259,
842
  "step": 1190
843
  },
844
  {
845
- "epoch": 14.285714285714286,
846
- "grad_norm": 0.14262178540229797,
847
- "learning_rate": 1.6243016993918022e-05,
848
- "loss": 0.0158,
849
  "step": 1200
850
  },
851
  {
852
- "epoch": 14.404761904761905,
853
- "grad_norm": 0.7596869468688965,
854
- "learning_rate": 1.616152547909618e-05,
855
- "loss": 0.0265,
856
  "step": 1210
857
  },
858
  {
859
- "epoch": 14.523809523809524,
860
- "grad_norm": 0.2511692941188812,
861
- "learning_rate": 1.6079368966006022e-05,
862
- "loss": 0.022,
863
  "step": 1220
864
  },
865
  {
866
- "epoch": 14.642857142857142,
867
- "grad_norm": 0.16154764592647552,
868
- "learning_rate": 1.5996556321597663e-05,
869
- "loss": 0.0222,
870
  "step": 1230
871
  },
872
  {
873
- "epoch": 14.761904761904763,
874
- "grad_norm": 0.6887322664260864,
875
- "learning_rate": 1.5913096483635827e-05,
876
- "loss": 0.008,
877
  "step": 1240
878
  },
879
  {
880
- "epoch": 14.880952380952381,
881
- "grad_norm": 0.07254794239997864,
882
- "learning_rate": 1.582899845973525e-05,
883
- "loss": 0.0034,
884
  "step": 1250
885
  },
886
  {
887
- "epoch": 15.0,
888
- "grad_norm": 0.4768582284450531,
889
- "learning_rate": 1.574427132638851e-05,
890
- "loss": 0.0083,
891
  "step": 1260
892
  },
893
  {
894
- "epoch": 15.119047619047619,
895
- "grad_norm": 0.04978374391794205,
896
- "learning_rate": 1.5658924227986415e-05,
897
- "loss": 0.0102,
898
  "step": 1270
899
  },
900
  {
901
- "epoch": 15.238095238095237,
902
- "grad_norm": 1.5553603172302246,
903
- "learning_rate": 1.5572966375831072e-05,
904
- "loss": 0.0082,
905
  "step": 1280
906
  },
907
  {
908
- "epoch": 15.357142857142858,
909
- "grad_norm": 0.1932092010974884,
910
- "learning_rate": 1.5486407047141745e-05,
911
- "loss": 0.0191,
912
  "step": 1290
913
  },
914
  {
915
- "epoch": 15.476190476190476,
916
- "grad_norm": 0.12488748133182526,
917
- "learning_rate": 1.5399255584053568e-05,
918
- "loss": 0.0068,
919
  "step": 1300
920
  },
921
  {
922
- "epoch": 15.595238095238095,
923
- "grad_norm": 0.04466537758708,
924
- "learning_rate": 1.5311521392609283e-05,
925
- "loss": 0.0067,
926
  "step": 1310
927
  },
928
  {
929
- "epoch": 15.714285714285714,
930
- "grad_norm": 0.1408109813928604,
931
- "learning_rate": 1.5223213941744078e-05,
932
- "loss": 0.0076,
933
  "step": 1320
934
  },
935
  {
936
- "epoch": 15.833333333333334,
937
- "grad_norm": 0.4874057471752167,
938
- "learning_rate": 1.5134342762263606e-05,
939
- "loss": 0.0109,
940
  "step": 1330
941
  },
942
  {
943
- "epoch": 15.952380952380953,
944
- "grad_norm": 0.15269064903259277,
945
- "learning_rate": 1.504491744581537e-05,
946
- "loss": 0.0073,
947
  "step": 1340
948
  },
949
  {
950
- "epoch": 16.071428571428573,
951
- "grad_norm": 0.192997545003891,
952
- "learning_rate": 1.4954947643853508e-05,
953
- "loss": 0.0223,
954
  "step": 1350
955
  },
956
  {
957
- "epoch": 16.19047619047619,
958
- "grad_norm": 0.9681908488273621,
959
- "learning_rate": 1.486444306659714e-05,
960
- "loss": 0.0037,
961
  "step": 1360
962
  },
963
  {
964
- "epoch": 16.30952380952381,
965
- "grad_norm": 1.2680203914642334,
966
- "learning_rate": 1.4773413481982359e-05,
967
- "loss": 0.008,
968
  "step": 1370
969
  },
970
  {
971
- "epoch": 16.428571428571427,
972
- "grad_norm": 0.07117388397455215,
973
- "learning_rate": 1.4681868714608021e-05,
974
- "loss": 0.0046,
975
  "step": 1380
976
  },
977
  {
978
- "epoch": 16.547619047619047,
979
- "grad_norm": 0.031909532845020294,
980
- "learning_rate": 1.4589818644675378e-05,
981
- "loss": 0.0101,
982
  "step": 1390
983
  },
984
  {
985
- "epoch": 16.666666666666668,
986
- "grad_norm": 1.234426498413086,
987
- "learning_rate": 1.4497273206921747e-05,
988
- "loss": 0.0051,
989
  "step": 1400
990
  },
991
  {
992
- "epoch": 16.785714285714285,
993
- "grad_norm": 0.43718039989471436,
994
- "learning_rate": 1.4404242389548274e-05,
995
- "loss": 0.0063,
996
  "step": 1410
997
  },
998
  {
999
- "epoch": 16.904761904761905,
1000
- "grad_norm": 3.5411911010742188,
1001
- "learning_rate": 1.4310736233141926e-05,
1002
- "loss": 0.0177,
1003
  "step": 1420
1004
  },
1005
  {
1006
- "epoch": 17.023809523809526,
1007
- "grad_norm": 0.09840346872806549,
1008
- "learning_rate": 1.4216764829591853e-05,
1009
- "loss": 0.0071,
1010
  "step": 1430
1011
  },
1012
  {
1013
- "epoch": 17.142857142857142,
1014
- "grad_norm": 0.09478545933961868,
1015
- "learning_rate": 1.4122338321000175e-05,
1016
- "loss": 0.0032,
1017
  "step": 1440
1018
  },
1019
  {
1020
- "epoch": 17.261904761904763,
1021
- "grad_norm": 0.040692731738090515,
1022
- "learning_rate": 1.4027466898587375e-05,
1023
- "loss": 0.0036,
1024
  "step": 1450
1025
  },
1026
  {
1027
- "epoch": 17.38095238095238,
1028
- "grad_norm": 0.6164568066596985,
1029
- "learning_rate": 1.3932160801592395e-05,
1030
- "loss": 0.0066,
1031
  "step": 1460
1032
  },
1033
  {
1034
- "epoch": 17.5,
1035
- "grad_norm": 0.2039034515619278,
1036
- "learning_rate": 1.3836430316167533e-05,
1037
- "loss": 0.008,
1038
  "step": 1470
1039
  },
1040
  {
1041
- "epoch": 17.61904761904762,
1042
- "grad_norm": 0.06619343906641006,
1043
- "learning_rate": 1.3740285774268282e-05,
1044
- "loss": 0.0068,
1045
  "step": 1480
1046
  },
1047
  {
1048
- "epoch": 17.738095238095237,
1049
- "grad_norm": 0.050607554614543915,
1050
- "learning_rate": 1.3643737552538237e-05,
1051
- "loss": 0.0058,
1052
  "step": 1490
1053
  },
1054
  {
1055
- "epoch": 17.857142857142858,
1056
- "grad_norm": 0.6153551340103149,
1057
- "learning_rate": 1.3546796071189153e-05,
1058
- "loss": 0.0089,
1059
  "step": 1500
1060
  },
1061
  {
1062
- "epoch": 17.976190476190474,
1063
- "grad_norm": 0.07599499821662903,
1064
- "learning_rate": 1.3449471792876333e-05,
1065
- "loss": 0.0063,
1066
  "step": 1510
1067
  },
1068
  {
1069
- "epoch": 18.095238095238095,
1070
- "grad_norm": 0.0883430764079094,
1071
- "learning_rate": 1.3351775221569416e-05,
1072
- "loss": 0.0054,
1073
  "step": 1520
1074
  },
1075
  {
1076
- "epoch": 18.214285714285715,
1077
- "grad_norm": 0.02305910363793373,
1078
- "learning_rate": 1.3253716901418702e-05,
1079
- "loss": 0.0028,
1080
  "step": 1530
1081
  },
1082
  {
1083
- "epoch": 18.333333333333332,
1084
- "grad_norm": 0.6192808151245117,
1085
- "learning_rate": 1.3155307415617156e-05,
1086
- "loss": 0.002,
1087
  "step": 1540
1088
  },
1089
  {
1090
- "epoch": 18.452380952380953,
1091
- "grad_norm": 0.23332145810127258,
1092
- "learning_rate": 1.3056557385258181e-05,
1093
- "loss": 0.0066,
1094
  "step": 1550
1095
  },
1096
  {
1097
- "epoch": 18.571428571428573,
1098
- "grad_norm": 0.045441482216119766,
1099
- "learning_rate": 1.2957477468189321e-05,
1100
- "loss": 0.0015,
1101
  "step": 1560
1102
  },
1103
  {
1104
- "epoch": 18.69047619047619,
1105
- "grad_norm": 0.009192162193357944,
1106
- "learning_rate": 1.2858078357861979e-05,
1107
- "loss": 0.0128,
1108
  "step": 1570
1109
  },
1110
  {
1111
- "epoch": 18.80952380952381,
1112
- "grad_norm": 0.05836764723062515,
1113
- "learning_rate": 1.2758370782177296e-05,
1114
- "loss": 0.0048,
1115
  "step": 1580
1116
  },
1117
  {
1118
- "epoch": 18.928571428571427,
1119
- "grad_norm": 1.0975781679153442,
1120
- "learning_rate": 1.2658365502328329e-05,
1121
- "loss": 0.007,
1122
  "step": 1590
1123
  },
1124
  {
1125
- "epoch": 19.047619047619047,
1126
- "grad_norm": 1.1575758457183838,
1127
- "learning_rate": 1.2558073311638604e-05,
1128
- "loss": 0.018,
1129
  "step": 1600
1130
  },
1131
  {
1132
- "epoch": 19.166666666666668,
1133
- "grad_norm": 0.1114954873919487,
1134
- "learning_rate": 1.2457505034397223e-05,
1135
- "loss": 0.0023,
1136
  "step": 1610
1137
  },
1138
  {
1139
- "epoch": 19.285714285714285,
1140
- "grad_norm": 0.07574819773435593,
1141
- "learning_rate": 1.2356671524690645e-05,
1142
- "loss": 0.0019,
1143
  "step": 1620
1144
  },
1145
  {
1146
- "epoch": 19.404761904761905,
1147
- "grad_norm": 0.009654861874878407,
1148
- "learning_rate": 1.2255583665231196e-05,
1149
- "loss": 0.0026,
1150
  "step": 1630
1151
  },
1152
  {
1153
- "epoch": 19.523809523809526,
1154
- "grad_norm": 0.02393646165728569,
1155
- "learning_rate": 1.2154252366182564e-05,
1156
- "loss": 0.0014,
1157
  "step": 1640
1158
  },
1159
  {
1160
- "epoch": 19.642857142857142,
1161
- "grad_norm": 0.020082678645849228,
1162
- "learning_rate": 1.2052688563982257e-05,
1163
- "loss": 0.0012,
1164
  "step": 1650
1165
  },
1166
  {
1167
- "epoch": 19.761904761904763,
1168
- "grad_norm": 0.9789303541183472,
1169
- "learning_rate": 1.1950903220161286e-05,
1170
- "loss": 0.0063,
1171
  "step": 1660
1172
  },
1173
  {
1174
- "epoch": 19.88095238095238,
1175
- "grad_norm": 0.2839144766330719,
1176
- "learning_rate": 1.1848907320161092e-05,
1177
- "loss": 0.0034,
1178
  "step": 1670
1179
  },
1180
  {
1181
- "epoch": 20.0,
1182
- "grad_norm": 0.010598956607282162,
1183
- "learning_rate": 1.174671187214794e-05,
1184
- "loss": 0.0054,
1185
  "step": 1680
1186
  },
1187
  {
1188
- "epoch": 20.11904761904762,
1189
- "grad_norm": 0.07026033848524094,
1190
- "learning_rate": 1.1644327905824808e-05,
1191
- "loss": 0.0084,
1192
  "step": 1690
1193
  },
1194
  {
1195
- "epoch": 20.238095238095237,
1196
- "grad_norm": 0.2679937779903412,
1197
- "learning_rate": 1.1541766471241e-05,
1198
- "loss": 0.0022,
1199
  "step": 1700
1200
  },
1201
  {
1202
- "epoch": 20.357142857142858,
1203
- "grad_norm": 0.10295789688825607,
1204
- "learning_rate": 1.1439038637599529e-05,
1205
- "loss": 0.0008,
1206
  "step": 1710
1207
  },
1208
  {
1209
- "epoch": 20.476190476190474,
1210
- "grad_norm": 0.4343288242816925,
1211
- "learning_rate": 1.1336155492062439e-05,
1212
- "loss": 0.0021,
1213
  "step": 1720
1214
  },
1215
  {
1216
- "epoch": 20.595238095238095,
1217
- "grad_norm": 0.018273649737238884,
1218
- "learning_rate": 1.1233128138554222e-05,
1219
- "loss": 0.0006,
1220
  "step": 1730
1221
  },
1222
  {
1223
- "epoch": 20.714285714285715,
1224
- "grad_norm": 0.008875339291989803,
1225
- "learning_rate": 1.112996769656337e-05,
1226
- "loss": 0.0003,
1227
  "step": 1740
1228
  },
1229
  {
1230
- "epoch": 20.833333333333332,
1231
- "grad_norm": 0.21868984401226044,
1232
- "learning_rate": 1.1026685299942286e-05,
1233
- "loss": 0.0022,
1234
  "step": 1750
1235
  },
1236
  {
1237
- "epoch": 20.952380952380953,
1238
- "grad_norm": 0.024242576211690903,
1239
- "learning_rate": 1.0923292095705642e-05,
1240
- "loss": 0.003,
1241
  "step": 1760
1242
  },
1243
  {
1244
- "epoch": 21.071428571428573,
1245
- "grad_norm": 0.008948619477450848,
1246
- "learning_rate": 1.0819799242827299e-05,
1247
- "loss": 0.0046,
1248
  "step": 1770
1249
  },
1250
  {
1251
- "epoch": 21.19047619047619,
1252
- "grad_norm": 0.01921582594513893,
1253
- "learning_rate": 1.0716217911035952e-05,
1254
- "loss": 0.002,
1255
  "step": 1780
1256
  },
1257
  {
1258
- "epoch": 21.30952380952381,
1259
- "grad_norm": 0.015329894609749317,
1260
- "learning_rate": 1.0612559279609606e-05,
1261
- "loss": 0.0004,
1262
  "step": 1790
1263
  },
1264
  {
1265
- "epoch": 21.428571428571427,
1266
- "grad_norm": 0.03253422677516937,
1267
- "learning_rate": 1.0508834536169028e-05,
1268
- "loss": 0.0032,
1269
  "step": 1800
1270
  },
1271
  {
1272
- "epoch": 21.547619047619047,
1273
- "grad_norm": 0.007767899427562952,
1274
- "learning_rate": 1.0405054875470287e-05,
1275
- "loss": 0.0003,
1276
  "step": 1810
1277
  },
1278
  {
1279
- "epoch": 21.666666666666668,
1280
- "grad_norm": 0.014929098077118397,
1281
- "learning_rate": 1.030123149819654e-05,
1282
- "loss": 0.0052,
1283
  "step": 1820
1284
  },
1285
  {
1286
- "epoch": 21.785714285714285,
1287
- "grad_norm": 0.020312979817390442,
1288
- "learning_rate": 1.0197375609749164e-05,
1289
- "loss": 0.0006,
1290
  "step": 1830
1291
  },
1292
  {
1293
- "epoch": 21.904761904761905,
1294
- "grad_norm": 0.04301303252577782,
1295
- "learning_rate": 1.0093498419038394e-05,
1296
- "loss": 0.002,
1297
  "step": 1840
1298
  },
1299
  {
1300
- "epoch": 22.023809523809526,
1301
- "grad_norm": 0.011388319544494152,
1302
- "learning_rate": 9.989611137273556e-06,
1303
- "loss": 0.004,
1304
  "step": 1850
1305
  },
1306
  {
1307
- "epoch": 22.142857142857142,
1308
- "grad_norm": 0.02648976631462574,
1309
- "learning_rate": 9.885724976753078e-06,
1310
- "loss": 0.0005,
1311
  "step": 1860
1312
  },
1313
  {
1314
- "epoch": 22.261904761904763,
1315
- "grad_norm": 0.016934091225266457,
1316
- "learning_rate": 9.78185114965439e-06,
1317
- "loss": 0.0007,
1318
  "step": 1870
1319
  },
1320
  {
1321
- "epoch": 22.38095238095238,
1322
- "grad_norm": 0.008141864091157913,
1323
- "learning_rate": 9.678000866823789e-06,
1324
- "loss": 0.0031,
1325
  "step": 1880
1326
  },
1327
  {
1328
- "epoch": 22.5,
1329
- "grad_norm": 0.015341422520577908,
1330
- "learning_rate": 9.574185336566503e-06,
1331
- "loss": 0.0007,
1332
  "step": 1890
1333
  },
1334
  {
1335
- "epoch": 22.61904761904762,
1336
- "grad_norm": 0.013799485750496387,
1337
- "learning_rate": 9.470415763437003e-06,
1338
- "loss": 0.0033,
1339
  "step": 1900
1340
  },
1341
  {
1342
- "epoch": 22.738095238095237,
1343
- "grad_norm": 0.032908592373132706,
1344
- "learning_rate": 9.366703347029725e-06,
1345
- "loss": 0.001,
1346
  "step": 1910
1347
  },
1348
  {
1349
- "epoch": 22.857142857142858,
1350
- "grad_norm": 0.06811460107564926,
1351
- "learning_rate": 9.263059280770303e-06,
1352
- "loss": 0.0024,
1353
  "step": 1920
1354
  },
1355
  {
1356
- "epoch": 22.976190476190474,
1357
- "grad_norm": 0.045220863074064255,
1358
- "learning_rate": 9.159494750707527e-06,
1359
- "loss": 0.0004,
1360
  "step": 1930
1361
  },
1362
  {
1363
- "epoch": 23.095238095238095,
1364
- "grad_norm": 0.003873666515573859,
1365
- "learning_rate": 9.05602093430603e-06,
1366
- "loss": 0.0003,
1367
  "step": 1940
1368
  },
1369
  {
1370
- "epoch": 23.214285714285715,
1371
- "grad_norm": 0.013574629090726376,
1372
- "learning_rate": 8.952648999239948e-06,
1373
- "loss": 0.0035,
1374
  "step": 1950
1375
  },
1376
  {
1377
- "epoch": 23.333333333333332,
1378
- "grad_norm": 0.009078383445739746,
1379
- "learning_rate": 8.849390102187615e-06,
1380
- "loss": 0.0002,
1381
  "step": 1960
1382
  },
1383
  {
1384
- "epoch": 23.452380952380953,
1385
- "grad_norm": 0.0057939100079238415,
1386
- "learning_rate": 8.746255387627444e-06,
1387
- "loss": 0.0004,
1388
  "step": 1970
1389
  },
1390
  {
1391
- "epoch": 23.571428571428573,
1392
- "grad_norm": 0.019538553431630135,
1393
- "learning_rate": 8.643255986635152e-06,
1394
- "loss": 0.0002,
1395
  "step": 1980
1396
  },
1397
  {
1398
- "epoch": 23.69047619047619,
1399
- "grad_norm": 0.007959152571856976,
1400
- "learning_rate": 8.540403015682382e-06,
1401
- "loss": 0.0009,
1402
  "step": 1990
1403
  },
1404
  {
1405
- "epoch": 23.80952380952381,
1406
- "grad_norm": 0.009719247929751873,
1407
- "learning_rate": 8.437707575436943e-06,
1408
- "loss": 0.0078,
1409
  "step": 2000
1410
  },
1411
  {
1412
- "epoch": 23.928571428571427,
1413
- "grad_norm": 0.10970814526081085,
1414
- "learning_rate": 8.335180749564759e-06,
1415
- "loss": 0.0007,
1416
  "step": 2010
1417
  },
1418
  {
1419
- "epoch": 24.047619047619047,
1420
- "grad_norm": 0.007809420116245747,
1421
- "learning_rate": 8.232833603533601e-06,
1422
- "loss": 0.0003,
1423
  "step": 2020
1424
  },
1425
  {
1426
- "epoch": 24.166666666666668,
1427
- "grad_norm": 0.037389811128377914,
1428
- "learning_rate": 8.130677183418851e-06,
1429
- "loss": 0.0002,
1430
  "step": 2030
1431
  },
1432
  {
1433
- "epoch": 24.285714285714285,
1434
- "grad_norm": 0.019924212247133255,
1435
- "learning_rate": 8.028722514711304e-06,
1436
- "loss": 0.0003,
1437
  "step": 2040
1438
  },
1439
  {
1440
- "epoch": 24.404761904761905,
1441
- "grad_norm": 0.005945433862507343,
1442
- "learning_rate": 7.926980601127225e-06,
1443
- "loss": 0.0014,
1444
  "step": 2050
1445
  },
1446
  {
1447
- "epoch": 24.523809523809526,
1448
- "grad_norm": 0.022461403161287308,
1449
- "learning_rate": 7.825462423420746e-06,
1450
- "loss": 0.0003,
1451
  "step": 2060
1452
  },
1453
  {
1454
- "epoch": 24.642857142857142,
1455
- "grad_norm": 0.013543925248086452,
1456
- "learning_rate": 7.724178938198734e-06,
1457
- "loss": 0.0002,
1458
  "step": 2070
1459
  },
1460
  {
1461
- "epoch": 24.761904761904763,
1462
- "grad_norm": 0.00889624934643507,
1463
- "learning_rate": 7.623141076738271e-06,
1464
- "loss": 0.0002,
1465
  "step": 2080
1466
  },
1467
  {
1468
- "epoch": 24.88095238095238,
1469
- "grad_norm": 0.006303866393864155,
1470
- "learning_rate": 7.522359743806877e-06,
1471
- "loss": 0.0004,
1472
  "step": 2090
1473
  },
1474
  {
1475
- "epoch": 25.0,
1476
- "grad_norm": 0.03090692311525345,
1477
- "learning_rate": 7.42184581648558e-06,
1478
- "loss": 0.0003,
1479
  "step": 2100
1480
  },
1481
  {
1482
- "epoch": 25.11904761904762,
1483
- "grad_norm": 0.006824655923992395,
1484
- "learning_rate": 7.321610142994971e-06,
1485
- "loss": 0.0002,
1486
  "step": 2110
1487
  },
1488
  {
1489
- "epoch": 25.238095238095237,
1490
- "grad_norm": 0.00634979922324419,
1491
- "learning_rate": 7.221663541524402e-06,
1492
- "loss": 0.0001,
1493
  "step": 2120
1494
  },
1495
  {
1496
- "epoch": 25.357142857142858,
1497
- "grad_norm": 0.004561006557196379,
1498
- "learning_rate": 7.122016799064381e-06,
1499
- "loss": 0.0003,
1500
  "step": 2130
1501
  },
1502
  {
1503
- "epoch": 25.476190476190474,
1504
- "grad_norm": 0.003320546355098486,
1505
- "learning_rate": 7.022680670242387e-06,
1506
- "loss": 0.0001,
1507
  "step": 2140
1508
  },
1509
  {
1510
- "epoch": 25.595238095238095,
1511
- "grad_norm": 0.009416792541742325,
1512
- "learning_rate": 6.923665876162118e-06,
1513
- "loss": 0.0003,
1514
  "step": 2150
1515
  },
1516
  {
1517
- "epoch": 25.714285714285715,
1518
- "grad_norm": 0.004399705212563276,
1519
- "learning_rate": 6.824983103246419e-06,
1520
- "loss": 0.0002,
1521
  "step": 2160
1522
  },
1523
  {
1524
- "epoch": 25.833333333333332,
1525
- "grad_norm": 0.003428778378292918,
1526
- "learning_rate": 6.7266430020839e-06,
1527
- "loss": 0.0001,
1528
  "step": 2170
1529
  },
1530
  {
1531
- "epoch": 25.952380952380953,
1532
- "grad_norm": 0.011793705634772778,
1533
- "learning_rate": 6.6286561862794496e-06,
1534
- "loss": 0.0037,
1535
  "step": 2180
1536
  },
1537
  {
1538
- "epoch": 26.071428571428573,
1539
- "grad_norm": 0.011102055199444294,
1540
- "learning_rate": 6.531033231308752e-06,
1541
- "loss": 0.0002,
1542
  "step": 2190
1543
  },
1544
  {
1545
- "epoch": 26.19047619047619,
1546
- "grad_norm": 0.005515962839126587,
1547
- "learning_rate": 6.43378467337687e-06,
1548
- "loss": 0.0001,
1549
  "step": 2200
1550
  },
1551
  {
1552
- "epoch": 26.30952380952381,
1553
- "grad_norm": 0.005885295569896698,
1554
- "learning_rate": 6.336921008281131e-06,
1555
- "loss": 0.0002,
1556
  "step": 2210
1557
  },
1558
  {
1559
- "epoch": 26.428571428571427,
1560
- "grad_norm": 0.0031049679964780807,
1561
- "learning_rate": 6.2404526902783205e-06,
1562
- "loss": 0.0002,
1563
  "step": 2220
1564
  },
1565
  {
1566
- "epoch": 26.547619047619047,
1567
- "grad_norm": 0.0042422860860824585,
1568
- "learning_rate": 6.144390130956384e-06,
1569
- "loss": 0.0001,
1570
  "step": 2230
1571
  },
1572
  {
1573
- "epoch": 26.666666666666668,
1574
- "grad_norm": 0.007420944049954414,
1575
- "learning_rate": 6.048743698110735e-06,
1576
- "loss": 0.0002,
1577
  "step": 2240
1578
  },
1579
  {
1580
- "epoch": 26.785714285714285,
1581
- "grad_norm": 0.00533581618219614,
1582
- "learning_rate": 5.953523714625279e-06,
1583
- "loss": 0.0002,
1584
  "step": 2250
1585
  },
1586
  {
1587
- "epoch": 26.904761904761905,
1588
- "grad_norm": 0.003453291719779372,
1589
- "learning_rate": 5.858740457358298e-06,
1590
- "loss": 0.0003,
1591
  "step": 2260
1592
  },
1593
  {
1594
- "epoch": 27.023809523809526,
1595
- "grad_norm": 0.004742017947137356,
1596
- "learning_rate": 5.764404156033279e-06,
1597
- "loss": 0.0002,
1598
  "step": 2270
1599
  },
1600
  {
1601
- "epoch": 27.142857142857142,
1602
- "grad_norm": 0.005143118556588888,
1603
- "learning_rate": 5.670524992134869e-06,
1604
- "loss": 0.0001,
1605
  "step": 2280
1606
  },
1607
  {
1608
- "epoch": 27.261904761904763,
1609
- "grad_norm": 0.005694309249520302,
1610
- "learning_rate": 5.5771130978099896e-06,
1611
- "loss": 0.0002,
1612
  "step": 2290
1613
  },
1614
  {
1615
- "epoch": 27.38095238095238,
1616
- "grad_norm": 0.004051010590046644,
1617
- "learning_rate": 5.484178554774304e-06,
1618
- "loss": 0.0001,
1619
  "step": 2300
1620
  },
1621
  {
1622
- "epoch": 27.5,
1623
- "grad_norm": 0.005464476533234119,
1624
- "learning_rate": 5.391731393224132e-06,
1625
- "loss": 0.0001,
1626
  "step": 2310
1627
  },
1628
  {
1629
- "epoch": 27.61904761904762,
1630
- "grad_norm": 0.0028825020417571068,
1631
- "learning_rate": 5.299781590753916e-06,
1632
  "loss": 0.0002,
1633
  "step": 2320
1634
  },
1635
  {
1636
- "epoch": 27.738095238095237,
1637
- "grad_norm": 0.0022702522110193968,
1638
- "learning_rate": 5.20833907127935e-06,
1639
- "loss": 0.0001,
1640
  "step": 2330
1641
  },
1642
  {
1643
- "epoch": 27.857142857142858,
1644
- "grad_norm": 0.004218791611492634,
1645
- "learning_rate": 5.11741370396632e-06,
1646
- "loss": 0.0001,
1647
  "step": 2340
1648
  },
1649
  {
1650
- "epoch": 27.976190476190474,
1651
- "grad_norm": 0.005840124096721411,
1652
- "learning_rate": 5.027015302165768e-06,
1653
- "loss": 0.0002,
1654
  "step": 2350
1655
  },
1656
  {
1657
- "epoch": 28.095238095238095,
1658
- "grad_norm": 0.0037928225938230753,
1659
- "learning_rate": 4.9371536223545295e-06,
1660
- "loss": 0.0001,
1661
  "step": 2360
1662
  },
1663
  {
1664
- "epoch": 28.214285714285715,
1665
- "grad_norm": 0.03005683235824108,
1666
- "learning_rate": 4.84783836308237e-06,
1667
- "loss": 0.0002,
1668
  "step": 2370
1669
  },
1670
  {
1671
- "epoch": 28.333333333333332,
1672
- "grad_norm": 0.005503533408045769,
1673
- "learning_rate": 4.759079163925223e-06,
1674
- "loss": 0.0001,
1675
  "step": 2380
1676
  },
1677
  {
1678
- "epoch": 28.452380952380953,
1679
- "grad_norm": 0.006272814702242613,
1680
- "learning_rate": 4.670885604444835e-06,
1681
- "loss": 0.0001,
1682
  "step": 2390
1683
  },
1684
  {
1685
- "epoch": 28.571428571428573,
1686
- "grad_norm": 0.004106464330106974,
1687
- "learning_rate": 4.583267203154844e-06,
1688
- "loss": 0.0001,
1689
  "step": 2400
1690
  },
1691
  {
1692
- "epoch": 28.69047619047619,
1693
- "grad_norm": 0.002469676546752453,
1694
- "learning_rate": 4.496233416493481e-06,
1695
- "loss": 0.0001,
1696
  "step": 2410
1697
  },
1698
  {
1699
- "epoch": 28.80952380952381,
1700
- "grad_norm": 0.0034991255961358547,
1701
- "learning_rate": 4.409793637802973e-06,
1702
- "loss": 0.0001,
1703
  "step": 2420
1704
  },
1705
  {
1706
- "epoch": 28.928571428571427,
1707
- "grad_norm": 0.002798357279971242,
1708
- "learning_rate": 4.323957196315714e-06,
1709
- "loss": 0.0001,
1710
  "step": 2430
1711
  },
1712
  {
1713
- "epoch": 29.047619047619047,
1714
- "grad_norm": 0.00471846479922533,
1715
- "learning_rate": 4.238733356147414e-06,
1716
- "loss": 0.0001,
1717
  "step": 2440
1718
  },
1719
  {
1720
- "epoch": 29.166666666666668,
1721
- "grad_norm": 0.004853650461882353,
1722
- "learning_rate": 4.154131315297219e-06,
1723
  "loss": 0.0002,
1724
  "step": 2450
1725
  },
1726
  {
1727
- "epoch": 29.285714285714285,
1728
- "grad_norm": 0.002823170507326722,
1729
- "learning_rate": 4.070160204655017e-06,
1730
  "loss": 0.0002,
1731
  "step": 2460
1732
  },
1733
  {
1734
- "epoch": 29.404761904761905,
1735
- "grad_norm": 0.003337468020617962,
1736
- "learning_rate": 3.986829087015941e-06,
1737
- "loss": 0.0001,
1738
  "step": 2470
1739
  },
1740
  {
1741
- "epoch": 29.523809523809526,
1742
- "grad_norm": 0.010672134347259998,
1743
- "learning_rate": 3.904146956102271e-06,
1744
- "loss": 0.0001,
1745
  "step": 2480
1746
  },
1747
  {
1748
- "epoch": 29.642857142857142,
1749
- "grad_norm": 0.005289705935865641,
1750
- "learning_rate": 3.8221227355927335e-06,
1751
- "loss": 0.0001,
1752
  "step": 2490
1753
  },
1754
  {
1755
- "epoch": 29.761904761904763,
1756
- "grad_norm": 0.0027344687841832638,
1757
- "learning_rate": 3.7407652781594094e-06,
1758
- "loss": 0.0001,
1759
  "step": 2500
1760
  },
1761
  {
1762
- "epoch": 29.88095238095238,
1763
- "grad_norm": 0.003742131870239973,
1764
- "learning_rate": 3.6600833645122924e-06,
1765
- "loss": 0.0001,
1766
  "step": 2510
1767
  },
1768
  {
1769
- "epoch": 30.0,
1770
- "grad_norm": 0.003494767937809229,
1771
- "learning_rate": 3.5800857024515844e-06,
1772
- "loss": 0.0001,
1773
  "step": 2520
1774
  },
1775
  {
1776
- "epoch": 30.11904761904762,
1777
- "grad_norm": 0.0017953565111383796,
1778
- "learning_rate": 3.5007809259279146e-06,
1779
- "loss": 0.0001,
1780
  "step": 2530
1781
  },
1782
  {
1783
- "epoch": 30.238095238095237,
1784
- "grad_norm": 0.006446167826652527,
1785
- "learning_rate": 3.4221775941104686e-06,
1786
- "loss": 0.0001,
1787
  "step": 2540
1788
  },
1789
  {
1790
- "epoch": 30.357142857142858,
1791
- "grad_norm": 0.004150099586695433,
1792
- "learning_rate": 3.344284190463244e-06,
1793
  "loss": 0.0002,
1794
  "step": 2550
1795
  },
1796
  {
1797
- "epoch": 30.476190476190474,
1798
- "grad_norm": 0.004592072684317827,
1799
- "learning_rate": 3.267109121829428e-06,
1800
- "loss": 0.0001,
1801
  "step": 2560
1802
  },
1803
  {
1804
- "epoch": 30.595238095238095,
1805
- "grad_norm": 0.010771949775516987,
1806
- "learning_rate": 3.1906607175240943e-06,
1807
- "loss": 0.0001,
1808
  "step": 2570
1809
  },
1810
  {
1811
- "epoch": 30.714285714285715,
1812
- "grad_norm": 0.004764531273394823,
1813
- "learning_rate": 3.1149472284352134e-06,
1814
- "loss": 0.0001,
1815
  "step": 2580
1816
  },
1817
  {
1818
- "epoch": 30.833333333333332,
1819
- "grad_norm": 0.0046601309441030025,
1820
- "learning_rate": 3.0399768261331664e-06,
1821
- "loss": 0.0001,
1822
  "step": 2590
1823
  },
1824
  {
1825
- "epoch": 30.952380952380953,
1826
- "grad_norm": 0.005445255897939205,
1827
- "learning_rate": 2.965757601988817e-06,
1828
- "loss": 0.0001,
1829
  "step": 2600
1830
  },
1831
  {
1832
- "epoch": 31.071428571428573,
1833
- "grad_norm": 0.00476863095536828,
1834
- "learning_rate": 2.892297566300214e-06,
1835
- "loss": 0.0001,
1836
  "step": 2610
1837
  },
1838
  {
1839
- "epoch": 31.19047619047619,
1840
- "grad_norm": 0.003519458929076791,
1841
- "learning_rate": 2.819604647428067e-06,
1842
- "loss": 0.0001,
1843
  "step": 2620
1844
  },
1845
  {
1846
- "epoch": 31.30952380952381,
1847
- "grad_norm": 0.002418002812191844,
1848
- "learning_rate": 2.7476866909400534e-06,
1849
- "loss": 0.0001,
1850
  "step": 2630
1851
  },
1852
  {
1853
- "epoch": 31.428571428571427,
1854
- "grad_norm": 0.0026580304838716984,
1855
- "learning_rate": 2.6765514587640815e-06,
1856
- "loss": 0.0001,
1857
  "step": 2640
1858
  },
1859
  {
1860
- "epoch": 31.547619047619047,
1861
- "grad_norm": 0.0021312797907739878,
1862
- "learning_rate": 2.6062066283505404e-06,
1863
- "loss": 0.0001,
1864
  "step": 2650
1865
  },
1866
  {
1867
- "epoch": 31.666666666666668,
1868
- "grad_norm": 0.00403893506154418,
1869
- "learning_rate": 2.536659791843705e-06,
1870
- "loss": 0.0001,
1871
  "step": 2660
1872
  },
1873
  {
1874
- "epoch": 31.785714285714285,
1875
- "grad_norm": 0.0029257936403155327,
1876
- "learning_rate": 2.4679184552623425e-06,
1877
- "loss": 0.0001,
1878
  "step": 2670
1879
  },
1880
  {
1881
- "epoch": 31.904761904761905,
1882
- "grad_norm": 0.0024930243380367756,
1883
- "learning_rate": 2.3999900376895844e-06,
1884
- "loss": 0.0001,
1885
  "step": 2680
1886
  },
1887
  {
1888
- "epoch": 32.023809523809526,
1889
- "grad_norm": 0.004459215793758631,
1890
- "learning_rate": 2.33288187047223e-06,
1891
- "loss": 0.0001,
1892
  "step": 2690
1893
  },
1894
  {
1895
- "epoch": 32.142857142857146,
1896
- "grad_norm": 0.002566899638622999,
1897
- "learning_rate": 2.2666011964294645e-06,
1898
- "loss": 0.0001,
1899
  "step": 2700
1900
  },
1901
  {
1902
- "epoch": 32.26190476190476,
1903
- "grad_norm": 0.003721205284819007,
1904
- "learning_rate": 2.201155169071184e-06,
1905
- "loss": 0.0001,
1906
  "step": 2710
1907
  },
1908
  {
1909
- "epoch": 32.38095238095238,
1910
- "grad_norm": 0.0032318762969225645,
1911
- "learning_rate": 2.1365508518259136e-06,
1912
- "loss": 0.0001,
1913
  "step": 2720
1914
  },
1915
  {
1916
- "epoch": 32.5,
1917
- "grad_norm": 0.002136578783392906,
1918
- "learning_rate": 2.0727952172784915e-06,
1919
- "loss": 0.0001,
1920
  "step": 2730
1921
  },
1922
  {
1923
- "epoch": 32.61904761904762,
1924
- "grad_norm": 0.0036290776915848255,
1925
- "learning_rate": 2.009895146417512e-06,
1926
- "loss": 0.0001,
1927
  "step": 2740
1928
  },
1929
  {
1930
- "epoch": 32.73809523809524,
1931
- "grad_norm": 0.010773614048957825,
1932
- "learning_rate": 1.9478574278926887e-06,
1933
- "loss": 0.0001,
1934
  "step": 2750
1935
  },
1936
  {
1937
- "epoch": 32.857142857142854,
1938
- "grad_norm": 0.0032484810799360275,
1939
- "learning_rate": 1.8866887572821813e-06,
1940
- "loss": 0.0001,
1941
  "step": 2760
1942
  },
1943
  {
1944
- "epoch": 32.976190476190474,
1945
- "grad_norm": 0.006387073080986738,
1946
- "learning_rate": 1.826395736369937e-06,
1947
  "loss": 0.0001,
1948
  "step": 2770
1949
  },
1950
  {
1951
- "epoch": 33.095238095238095,
1952
- "grad_norm": 0.005526574328541756,
1953
- "learning_rate": 1.7669848724331984e-06,
1954
- "loss": 0.0001,
1955
  "step": 2780
1956
  },
1957
  {
1958
- "epoch": 33.214285714285715,
1959
- "grad_norm": 0.002433312125504017,
1960
- "learning_rate": 1.7084625775401664e-06,
1961
- "loss": 0.0001,
1962
  "step": 2790
1963
  },
1964
  {
1965
- "epoch": 33.333333333333336,
1966
- "grad_norm": 0.002393395872786641,
1967
- "learning_rate": 1.6508351678579882e-06,
1968
- "loss": 0.0001,
1969
  "step": 2800
1970
  },
1971
  {
1972
- "epoch": 33.45238095238095,
1973
- "grad_norm": 0.003153387922793627,
1974
- "learning_rate": 1.59410886297104e-06,
1975
- "loss": 0.0001,
1976
  "step": 2810
1977
  },
1978
  {
1979
- "epoch": 33.57142857142857,
1980
- "grad_norm": 0.003180896397680044,
1981
- "learning_rate": 1.538289785209689e-06,
1982
- "loss": 0.0001,
1983
  "step": 2820
1984
  },
1985
  {
1986
- "epoch": 33.69047619047619,
1987
- "grad_norm": 0.009424399584531784,
1988
- "learning_rate": 1.4833839589895072e-06,
1989
- "loss": 0.0001,
1990
  "step": 2830
1991
  },
1992
  {
1993
- "epoch": 33.80952380952381,
1994
- "grad_norm": 0.0031415896955877542,
1995
- "learning_rate": 1.4293973101610748e-06,
1996
- "loss": 0.0001,
1997
  "step": 2840
1998
  },
1999
  {
2000
- "epoch": 33.92857142857143,
2001
- "grad_norm": 0.00263274391181767,
2002
- "learning_rate": 1.3763356653704274e-06,
2003
- "loss": 0.0001,
2004
  "step": 2850
2005
  },
2006
  {
2007
- "epoch": 34.04761904761905,
2008
- "grad_norm": 0.0029935326892882586,
2009
- "learning_rate": 1.324204751430186e-06,
2010
  "loss": 0.0001,
2011
  "step": 2860
2012
  },
2013
  {
2014
- "epoch": 34.166666666666664,
2015
- "grad_norm": 0.0044165910221636295,
2016
- "learning_rate": 1.2730101947014918e-06,
2017
- "loss": 0.0001,
2018
  "step": 2870
2019
  },
2020
  {
2021
- "epoch": 34.285714285714285,
2022
- "grad_norm": 0.003201395506039262,
2023
- "learning_rate": 1.2227575204867481e-06,
2024
- "loss": 0.0001,
2025
  "step": 2880
2026
  },
2027
  {
2028
- "epoch": 34.404761904761905,
2029
- "grad_norm": 0.003189435927197337,
2030
- "learning_rate": 1.1734521524333087e-06,
2031
- "loss": 0.0001,
2032
  "step": 2890
2033
  },
2034
  {
2035
- "epoch": 34.523809523809526,
2036
- "grad_norm": 0.0018071344820782542,
2037
- "learning_rate": 1.1250994119480985e-06,
2038
- "loss": 0.0001,
2039
  "step": 2900
2040
  },
2041
  {
2042
- "epoch": 34.642857142857146,
2043
- "grad_norm": 0.002372781280428171,
2044
- "learning_rate": 1.0777045176232992e-06,
2045
- "loss": 0.0001,
2046
  "step": 2910
2047
  },
2048
  {
2049
- "epoch": 34.76190476190476,
2050
- "grad_norm": 0.010425383225083351,
2051
- "learning_rate": 1.0312725846731174e-06,
2052
  "loss": 0.0001,
2053
  "step": 2920
2054
  },
2055
  {
2056
- "epoch": 34.88095238095238,
2057
- "grad_norm": 0.003908902872353792,
2058
- "learning_rate": 9.858086243817066e-07,
2059
- "loss": 0.0001,
2060
  "step": 2930
2061
  },
2062
  {
2063
- "epoch": 35.0,
2064
- "grad_norm": 0.002348334761336446,
2065
- "learning_rate": 9.41317543562319e-07,
2066
- "loss": 0.0001,
2067
  "step": 2940
2068
  },
2069
  {
2070
- "epoch": 35.11904761904762,
2071
- "grad_norm": 0.0023442634847015142,
2072
- "learning_rate": 8.978041440277163e-07,
2073
- "loss": 0.0001,
2074
  "step": 2950
2075
  },
2076
  {
2077
- "epoch": 35.23809523809524,
2078
- "grad_norm": 0.0019229685422033072,
2079
- "learning_rate": 8.552731220719357e-07,
2080
- "loss": 0.0001,
2081
  "step": 2960
2082
  },
2083
  {
2084
- "epoch": 35.357142857142854,
2085
- "grad_norm": 0.0020270678214728832,
2086
- "learning_rate": 8.137290679634124e-07,
2087
- "loss": 0.0001,
2088
  "step": 2970
2089
  },
2090
  {
2091
- "epoch": 35.476190476190474,
2092
- "grad_norm": 0.004971123766154051,
2093
- "learning_rate": 7.731764654495832e-07,
2094
- "loss": 0.0001,
2095
  "step": 2980
2096
  },
2097
  {
2098
- "epoch": 35.595238095238095,
2099
- "grad_norm": 0.0020201329607516527,
2100
- "learning_rate": 7.336196912729488e-07,
2101
- "loss": 0.0001,
2102
  "step": 2990
2103
  },
2104
  {
2105
- "epoch": 35.714285714285715,
2106
- "grad_norm": 0.00401673186570406,
2107
- "learning_rate": 6.950630146987103e-07,
2108
- "loss": 0.0001,
2109
  "step": 3000
2110
  }
2111
  ],
2112
  "logging_steps": 10,
2113
- "max_steps": 3360,
2114
  "num_input_tokens_seen": 0,
2115
- "num_train_epochs": 40,
2116
  "save_steps": 1000,
2117
  "stateful_callbacks": {
2118
  "TrainerControl": {
@@ -2126,8 +2126,8 @@
2126
  "attributes": {}
2127
  }
2128
  },
2129
- "total_flos": 1.7897473131832934e+17,
2130
- "train_batch_size": 3,
2131
  "trial_name": null,
2132
  "trial_params": null
2133
  }
 
1
  {
2
  "best_metric": null,
3
  "best_model_checkpoint": null,
4
+ "epoch": 24.0,
5
  "eval_steps": 500,
6
  "global_step": 3000,
7
  "is_hyper_param_search": false,
 
9
  "is_world_process_zero": true,
10
  "log_history": [
11
  {
12
+ "epoch": 0.08,
13
+ "grad_norm": 6.874109268188477,
14
+ "learning_rate": 2.5559105431309904e-07,
15
+ "loss": 1.5574,
16
  "step": 10
17
  },
18
  {
19
+ "epoch": 0.16,
20
+ "grad_norm": 6.183030128479004,
21
+ "learning_rate": 8.306709265175719e-07,
22
+ "loss": 1.4724,
23
  "step": 20
24
  },
25
  {
26
+ "epoch": 0.24,
27
+ "grad_norm": 2.6565823554992676,
28
+ "learning_rate": 1.4696485623003196e-06,
29
+ "loss": 1.054,
30
  "step": 30
31
  },
32
  {
33
+ "epoch": 0.32,
34
+ "grad_norm": 3.227349042892456,
35
+ "learning_rate": 2.1086261980830672e-06,
36
+ "loss": 0.8732,
37
  "step": 40
38
  },
39
  {
40
+ "epoch": 0.4,
41
+ "grad_norm": 1.1049450635910034,
42
+ "learning_rate": 2.747603833865815e-06,
43
+ "loss": 0.7617,
44
  "step": 50
45
  },
46
  {
47
+ "epoch": 0.48,
48
+ "grad_norm": 1.8624663352966309,
49
+ "learning_rate": 3.386581469648563e-06,
50
+ "loss": 0.7144,
51
  "step": 60
52
  },
53
  {
54
+ "epoch": 0.56,
55
+ "grad_norm": 2.1665029525756836,
56
+ "learning_rate": 4.02555910543131e-06,
57
+ "loss": 0.7612,
58
  "step": 70
59
  },
60
  {
61
+ "epoch": 0.64,
62
+ "grad_norm": 1.0787376165390015,
63
+ "learning_rate": 4.664536741214058e-06,
64
+ "loss": 0.6864,
65
  "step": 80
66
  },
67
  {
68
+ "epoch": 0.72,
69
+ "grad_norm": 1.4465893507003784,
70
+ "learning_rate": 5.303514376996806e-06,
71
+ "loss": 0.6558,
72
  "step": 90
73
  },
74
  {
75
+ "epoch": 0.8,
76
+ "grad_norm": 1.1737278699874878,
77
+ "learning_rate": 5.8785942492012785e-06,
78
+ "loss": 0.6295,
79
  "step": 100
80
  },
81
  {
82
+ "epoch": 0.88,
83
+ "grad_norm": 1.7027487754821777,
84
+ "learning_rate": 6.517571884984026e-06,
85
+ "loss": 0.638,
86
  "step": 110
87
  },
88
  {
89
+ "epoch": 0.96,
90
+ "grad_norm": 1.5483449697494507,
91
+ "learning_rate": 7.156549520766773e-06,
92
+ "loss": 0.6173,
93
  "step": 120
94
  },
95
  {
96
+ "epoch": 1.04,
97
+ "grad_norm": 1.9829121828079224,
98
+ "learning_rate": 7.795527156549521e-06,
99
+ "loss": 0.5511,
100
  "step": 130
101
  },
102
  {
103
+ "epoch": 1.12,
104
+ "grad_norm": 1.466599464416504,
105
+ "learning_rate": 8.434504792332269e-06,
106
+ "loss": 0.5366,
107
  "step": 140
108
  },
109
  {
110
+ "epoch": 1.2,
111
+ "grad_norm": 1.5559449195861816,
112
+ "learning_rate": 9.073482428115017e-06,
113
+ "loss": 0.5662,
114
  "step": 150
115
  },
116
  {
117
+ "epoch": 1.28,
118
+ "grad_norm": 1.33512282371521,
119
+ "learning_rate": 9.712460063897765e-06,
120
+ "loss": 0.5567,
121
  "step": 160
122
  },
123
  {
124
+ "epoch": 1.3599999999999999,
125
+ "grad_norm": 0.943450927734375,
126
+ "learning_rate": 1.035143769968051e-05,
127
+ "loss": 0.5997,
128
  "step": 170
129
  },
130
  {
131
+ "epoch": 1.44,
132
+ "grad_norm": 1.2756850719451904,
133
+ "learning_rate": 1.099041533546326e-05,
134
+ "loss": 0.5587,
135
  "step": 180
136
  },
137
  {
138
+ "epoch": 1.52,
139
+ "grad_norm": 1.0768604278564453,
140
+ "learning_rate": 1.1629392971246008e-05,
141
+ "loss": 0.6288,
142
  "step": 190
143
  },
144
  {
145
+ "epoch": 1.6,
146
+ "grad_norm": 1.082158088684082,
147
+ "learning_rate": 1.2268370607028754e-05,
148
+ "loss": 0.5475,
149
  "step": 200
150
  },
151
  {
152
+ "epoch": 1.6800000000000002,
153
+ "grad_norm": 1.4872550964355469,
154
+ "learning_rate": 1.2907348242811502e-05,
155
+ "loss": 0.6002,
156
  "step": 210
157
  },
158
  {
159
+ "epoch": 1.76,
160
+ "grad_norm": 2.0925095081329346,
161
+ "learning_rate": 1.3546325878594251e-05,
162
+ "loss": 0.5887,
163
  "step": 220
164
  },
165
  {
166
+ "epoch": 1.8399999999999999,
167
+ "grad_norm": 1.4571633338928223,
168
+ "learning_rate": 1.4185303514376998e-05,
169
+ "loss": 0.6107,
170
  "step": 230
171
  },
172
  {
173
+ "epoch": 1.92,
174
+ "grad_norm": 2.4871673583984375,
175
+ "learning_rate": 1.4824281150159745e-05,
176
+ "loss": 0.5582,
177
  "step": 240
178
  },
179
  {
180
+ "epoch": 2.0,
181
+ "grad_norm": 1.3184454441070557,
182
+ "learning_rate": 1.5463258785942495e-05,
183
+ "loss": 0.582,
184
  "step": 250
185
  },
186
  {
187
+ "epoch": 2.08,
188
+ "grad_norm": 1.9747495651245117,
189
+ "learning_rate": 1.610223642172524e-05,
190
+ "loss": 0.4687,
191
  "step": 260
192
  },
193
  {
194
+ "epoch": 2.16,
195
+ "grad_norm": 1.7691447734832764,
196
+ "learning_rate": 1.6741214057507987e-05,
197
+ "loss": 0.5013,
198
  "step": 270
199
  },
200
  {
201
+ "epoch": 2.24,
202
+ "grad_norm": 1.8902829885482788,
203
+ "learning_rate": 1.7380191693290737e-05,
204
+ "loss": 0.5008,
205
  "step": 280
206
  },
207
  {
208
+ "epoch": 2.32,
209
+ "grad_norm": 1.3197141885757446,
210
+ "learning_rate": 1.8019169329073486e-05,
211
+ "loss": 0.5229,
212
  "step": 290
213
  },
214
  {
215
+ "epoch": 2.4,
216
+ "grad_norm": 2.325404167175293,
217
+ "learning_rate": 1.8658146964856232e-05,
218
+ "loss": 0.4975,
219
  "step": 300
220
  },
221
  {
222
+ "epoch": 2.48,
223
+ "grad_norm": 1.3323551416397095,
224
+ "learning_rate": 1.929712460063898e-05,
225
+ "loss": 0.5254,
226
  "step": 310
227
  },
228
  {
229
+ "epoch": 2.56,
230
+ "grad_norm": 0.7770269513130188,
231
+ "learning_rate": 1.9936102236421725e-05,
232
+ "loss": 0.5304,
233
  "step": 320
234
  },
235
  {
236
+ "epoch": 2.64,
237
+ "grad_norm": 1.887294054031372,
238
+ "learning_rate": 1.999949450079496e-05,
239
+ "loss": 0.4755,
240
  "step": 330
241
  },
242
  {
243
+ "epoch": 2.7199999999999998,
244
+ "grad_norm": 1.2255381345748901,
245
+ "learning_rate": 1.9997747161747696e-05,
246
+ "loss": 0.4982,
247
  "step": 340
248
  },
249
  {
250
+ "epoch": 2.8,
251
+ "grad_norm": 1.6865767240524292,
252
+ "learning_rate": 1.9994751960168383e-05,
253
+ "loss": 0.4857,
254
  "step": 350
255
  },
256
  {
257
+ "epoch": 2.88,
258
+ "grad_norm": 1.2997519969940186,
259
+ "learning_rate": 1.999050926990122e-05,
260
+ "loss": 0.4987,
261
  "step": 360
262
  },
263
  {
264
+ "epoch": 2.96,
265
+ "grad_norm": 1.4636356830596924,
266
+ "learning_rate": 1.9985019620494935e-05,
267
+ "loss": 0.4947,
268
  "step": 370
269
  },
270
  {
271
+ "epoch": 3.04,
272
+ "grad_norm": 1.3720135688781738,
273
+ "learning_rate": 1.9978283697136662e-05,
274
+ "loss": 0.5135,
275
  "step": 380
276
  },
277
  {
278
+ "epoch": 3.12,
279
+ "grad_norm": 1.6108952760696411,
280
+ "learning_rate": 1.997030234056645e-05,
281
+ "loss": 0.4405,
282
  "step": 390
283
  },
284
  {
285
+ "epoch": 3.2,
286
+ "grad_norm": 2.1790759563446045,
287
+ "learning_rate": 1.9961076546972304e-05,
288
+ "loss": 0.4468,
289
  "step": 400
290
  },
291
  {
292
+ "epoch": 3.2800000000000002,
293
+ "grad_norm": 4.406925678253174,
294
+ "learning_rate": 1.9950607467865856e-05,
295
+ "loss": 0.4675,
296
  "step": 410
297
  },
298
  {
299
+ "epoch": 3.36,
300
+ "grad_norm": 1.7346851825714111,
301
+ "learning_rate": 1.993889640993864e-05,
302
+ "loss": 0.4243,
303
  "step": 420
304
  },
305
  {
306
+ "epoch": 3.44,
307
+ "grad_norm": 1.2865455150604248,
308
+ "learning_rate": 1.9925944834898992e-05,
309
+ "loss": 0.3787,
310
  "step": 430
311
  },
312
  {
313
+ "epoch": 3.52,
314
+ "grad_norm": 1.321824550628662,
315
+ "learning_rate": 1.991175435928962e-05,
316
+ "loss": 0.3982,
317
  "step": 440
318
  },
319
  {
320
+ "epoch": 3.6,
321
+ "grad_norm": 2.45711350440979,
322
+ "learning_rate": 1.989632675428582e-05,
323
+ "loss": 0.4811,
324
  "step": 450
325
  },
326
  {
327
+ "epoch": 3.68,
328
+ "grad_norm": 2.5552310943603516,
329
+ "learning_rate": 1.9879663945474416e-05,
330
+ "loss": 0.4255,
331
  "step": 460
332
  },
333
  {
334
+ "epoch": 3.76,
335
+ "grad_norm": 1.763244867324829,
336
+ "learning_rate": 1.9861768012613435e-05,
337
+ "loss": 0.4579,
338
  "step": 470
339
  },
340
  {
341
+ "epoch": 3.84,
342
+ "grad_norm": 1.0951169729232788,
343
+ "learning_rate": 1.984264118937249e-05,
344
+ "loss": 0.398,
345
  "step": 480
346
  },
347
  {
348
+ "epoch": 3.92,
349
+ "grad_norm": 8.235908508300781,
350
+ "learning_rate": 1.9822285863054e-05,
351
+ "loss": 0.4101,
352
  "step": 490
353
  },
354
  {
355
+ "epoch": 4.0,
356
+ "grad_norm": 19.39371681213379,
357
+ "learning_rate": 1.9800704574295246e-05,
358
+ "loss": 0.4645,
359
  "step": 500
360
  },
361
  {
362
+ "epoch": 4.08,
363
+ "grad_norm": 1.2594205141067505,
364
+ "learning_rate": 1.9777900016751224e-05,
365
+ "loss": 0.3632,
366
  "step": 510
367
  },
368
  {
369
+ "epoch": 4.16,
370
+ "grad_norm": 1.7697833776474,
371
+ "learning_rate": 1.9753875036758464e-05,
372
+ "loss": 0.3356,
373
  "step": 520
374
  },
375
  {
376
+ "epoch": 4.24,
377
+ "grad_norm": 1.66884446144104,
378
+ "learning_rate": 1.9728632632979746e-05,
379
+ "loss": 0.382,
380
  "step": 530
381
  },
382
  {
383
+ "epoch": 4.32,
384
+ "grad_norm": 1.7648805379867554,
385
+ "learning_rate": 1.970217595602985e-05,
386
+ "loss": 0.3997,
387
  "step": 540
388
  },
389
  {
390
+ "epoch": 4.4,
391
+ "grad_norm": 1.6132514476776123,
392
+ "learning_rate": 1.967450830808228e-05,
393
+ "loss": 0.3476,
394
  "step": 550
395
  },
396
  {
397
+ "epoch": 4.48,
398
+ "grad_norm": 1.3217155933380127,
399
+ "learning_rate": 1.9645633142457143e-05,
400
+ "loss": 0.3362,
401
  "step": 560
402
  },
403
  {
404
+ "epoch": 4.5600000000000005,
405
+ "grad_norm": 1.5668702125549316,
406
+ "learning_rate": 1.9615554063190098e-05,
407
+ "loss": 0.3996,
408
  "step": 570
409
  },
410
  {
411
+ "epoch": 4.64,
412
+ "grad_norm": 2.051454544067383,
413
+ "learning_rate": 1.958427482458253e-05,
414
+ "loss": 0.4104,
415
  "step": 580
416
  },
417
  {
418
+ "epoch": 4.72,
419
+ "grad_norm": 1.84868586063385,
420
+ "learning_rate": 1.9551799330732954e-05,
421
+ "loss": 0.374,
422
  "step": 590
423
  },
424
  {
425
+ "epoch": 4.8,
426
+ "grad_norm": 1.3498121500015259,
427
+ "learning_rate": 1.9518131635049745e-05,
428
+ "loss": 0.3444,
429
  "step": 600
430
  },
431
  {
432
+ "epoch": 4.88,
433
+ "grad_norm": 1.962553858757019,
434
+ "learning_rate": 1.9483275939745184e-05,
435
+ "loss": 0.3153,
436
  "step": 610
437
  },
438
  {
439
+ "epoch": 4.96,
440
+ "grad_norm": 1.7132457494735718,
441
+ "learning_rate": 1.944723659531099e-05,
442
+ "loss": 0.383,
443
  "step": 620
444
  },
445
  {
446
+ "epoch": 5.04,
447
+ "grad_norm": 1.8860396146774292,
448
+ "learning_rate": 1.9410018099975297e-05,
449
+ "loss": 0.327,
450
  "step": 630
451
  },
452
  {
453
+ "epoch": 5.12,
454
+ "grad_norm": 1.9154820442199707,
455
+ "learning_rate": 1.9371625099141223e-05,
456
+ "loss": 0.2704,
457
  "step": 640
458
  },
459
  {
460
+ "epoch": 5.2,
461
+ "grad_norm": 1.4077069759368896,
462
+ "learning_rate": 1.9332062384807058e-05,
463
+ "loss": 0.2659,
464
  "step": 650
465
  },
466
  {
467
+ "epoch": 5.28,
468
+ "grad_norm": 1.8489642143249512,
469
+ "learning_rate": 1.9291334894968133e-05,
470
+ "loss": 0.2523,
471
  "step": 660
472
  },
473
  {
474
+ "epoch": 5.36,
475
+ "grad_norm": 2.3917829990386963,
476
+ "learning_rate": 1.9249447713000515e-05,
477
+ "loss": 0.2885,
478
  "step": 670
479
  },
480
  {
481
+ "epoch": 5.44,
482
+ "grad_norm": 2.5429599285125732,
483
+ "learning_rate": 1.9206406067026506e-05,
484
+ "loss": 0.3097,
485
  "step": 680
486
  },
487
  {
488
+ "epoch": 5.52,
489
+ "grad_norm": 3.4684863090515137,
490
+ "learning_rate": 1.9162215329262115e-05,
491
+ "loss": 0.2868,
492
  "step": 690
493
  },
494
  {
495
+ "epoch": 5.6,
496
+ "grad_norm": 2.662013292312622,
497
+ "learning_rate": 1.9116881015346517e-05,
498
+ "loss": 0.3281,
499
  "step": 700
500
  },
501
  {
502
+ "epoch": 5.68,
503
+ "grad_norm": 2.223334312438965,
504
+ "learning_rate": 1.9070408783653627e-05,
505
+ "loss": 0.2799,
506
  "step": 710
507
  },
508
  {
509
+ "epoch": 5.76,
510
+ "grad_norm": 2.992532730102539,
511
+ "learning_rate": 1.9022804434585854e-05,
512
+ "loss": 0.3252,
513
  "step": 720
514
  },
515
  {
516
+ "epoch": 5.84,
517
+ "grad_norm": 1.6079481840133667,
518
+ "learning_rate": 1.8974073909850125e-05,
519
+ "loss": 0.2791,
520
  "step": 730
521
  },
522
  {
523
+ "epoch": 5.92,
524
+ "grad_norm": 2.97104549407959,
525
+ "learning_rate": 1.8924223291716274e-05,
526
+ "loss": 0.3029,
527
  "step": 740
528
  },
529
  {
530
+ "epoch": 6.0,
531
+ "grad_norm": 2.1889212131500244,
532
+ "learning_rate": 1.887325880225789e-05,
533
+ "loss": 0.293,
534
  "step": 750
535
  },
536
  {
537
+ "epoch": 6.08,
538
+ "grad_norm": 2.123610019683838,
539
+ "learning_rate": 1.882118680257572e-05,
540
+ "loss": 0.2308,
541
  "step": 760
542
  },
543
  {
544
+ "epoch": 6.16,
545
+ "grad_norm": 3.71563720703125,
546
+ "learning_rate": 1.8768013792003683e-05,
547
+ "loss": 0.216,
548
  "step": 770
549
  },
550
  {
551
+ "epoch": 6.24,
552
+ "grad_norm": 2.459091901779175,
553
+ "learning_rate": 1.8713746407297703e-05,
554
+ "loss": 0.1972,
555
  "step": 780
556
  },
557
  {
558
+ "epoch": 6.32,
559
+ "grad_norm": 2.669945240020752,
560
+ "learning_rate": 1.8658391421807313e-05,
561
+ "loss": 0.2478,
562
  "step": 790
563
  },
564
  {
565
+ "epoch": 6.4,
566
+ "grad_norm": 2.1581931114196777,
567
+ "learning_rate": 1.8601955744630255e-05,
568
+ "loss": 0.2295,
569
  "step": 800
570
  },
571
  {
572
+ "epoch": 6.48,
573
+ "grad_norm": 2.567908763885498,
574
+ "learning_rate": 1.8544446419750125e-05,
575
+ "loss": 0.2239,
576
  "step": 810
577
  },
578
  {
579
+ "epoch": 6.5600000000000005,
580
+ "grad_norm": 2.0205740928649902,
581
+ "learning_rate": 1.8485870625157186e-05,
582
+ "loss": 0.216,
583
  "step": 820
584
  },
585
  {
586
+ "epoch": 6.64,
587
+ "grad_norm": 2.2898030281066895,
588
+ "learning_rate": 1.8426235671952452e-05,
589
+ "loss": 0.2343,
590
  "step": 830
591
  },
592
  {
593
+ "epoch": 6.72,
594
+ "grad_norm": 2.2003941535949707,
595
+ "learning_rate": 1.836554900343514e-05,
596
+ "loss": 0.2076,
597
  "step": 840
598
  },
599
  {
600
+ "epoch": 6.8,
601
+ "grad_norm": 1.5117710828781128,
602
+ "learning_rate": 1.8303818194173665e-05,
603
+ "loss": 0.1904,
604
  "step": 850
605
  },
606
  {
607
+ "epoch": 6.88,
608
+ "grad_norm": 1.2647337913513184,
609
+ "learning_rate": 1.824105094906021e-05,
610
+ "loss": 0.2341,
611
  "step": 860
612
  },
613
  {
614
+ "epoch": 6.96,
615
+ "grad_norm": 2.00254225730896,
616
+ "learning_rate": 1.8177255102349047e-05,
617
+ "loss": 0.2288,
618
  "step": 870
619
  },
620
  {
621
+ "epoch": 7.04,
622
+ "grad_norm": 2.516129493713379,
623
+ "learning_rate": 1.8112438616678712e-05,
624
+ "loss": 0.1685,
625
  "step": 880
626
  },
627
  {
628
+ "epoch": 7.12,
629
+ "grad_norm": 3.6129150390625,
630
+ "learning_rate": 1.8046609582078147e-05,
631
+ "loss": 0.1625,
632
  "step": 890
633
  },
634
  {
635
+ "epoch": 7.2,
636
+ "grad_norm": 3.1554768085479736,
637
+ "learning_rate": 1.797977621495696e-05,
638
+ "loss": 0.2011,
639
  "step": 900
640
  },
641
  {
642
+ "epoch": 7.28,
643
+ "grad_norm": 3.8677282333374023,
644
+ "learning_rate": 1.7911946857079886e-05,
645
+ "loss": 0.1579,
646
  "step": 910
647
  },
648
  {
649
+ "epoch": 7.36,
650
+ "grad_norm": 3.062568426132202,
651
+ "learning_rate": 1.784312997452562e-05,
652
+ "loss": 0.1705,
653
  "step": 920
654
  },
655
  {
656
+ "epoch": 7.44,
657
+ "grad_norm": 1.8281478881835938,
658
+ "learning_rate": 1.777333415663014e-05,
659
+ "loss": 0.1708,
660
  "step": 930
661
  },
662
  {
663
+ "epoch": 7.52,
664
+ "grad_norm": 1.1312201023101807,
665
+ "learning_rate": 1.7702568114914607e-05,
666
+ "loss": 0.1644,
667
  "step": 940
668
  },
669
  {
670
+ "epoch": 7.6,
671
+ "grad_norm": 1.1482549905776978,
672
+ "learning_rate": 1.7630840681998068e-05,
673
+ "loss": 0.1297,
674
  "step": 950
675
  },
676
  {
677
+ "epoch": 7.68,
678
+ "grad_norm": 3.109949827194214,
679
+ "learning_rate": 1.755816081049501e-05,
680
+ "loss": 0.1812,
681
  "step": 960
682
  },
683
  {
684
+ "epoch": 7.76,
685
+ "grad_norm": 1.2507814168930054,
686
+ "learning_rate": 1.7484537571897943e-05,
687
+ "loss": 0.1331,
688
  "step": 970
689
  },
690
  {
691
+ "epoch": 7.84,
692
+ "grad_norm": 2.898353338241577,
693
+ "learning_rate": 1.740998015544514e-05,
694
+ "loss": 0.1625,
695
  "step": 980
696
  },
697
  {
698
+ "epoch": 7.92,
699
+ "grad_norm": 2.2018096446990967,
700
+ "learning_rate": 1.7334497866973716e-05,
701
+ "loss": 0.1441,
702
  "step": 990
703
  },
704
  {
705
+ "epoch": 8.0,
706
+ "grad_norm": 2.1120376586914062,
707
+ "learning_rate": 1.725810012775808e-05,
708
+ "loss": 0.1804,
709
  "step": 1000
710
  },
711
  {
712
+ "epoch": 8.08,
713
+ "grad_norm": 1.815804362297058,
714
+ "learning_rate": 1.7180796473334075e-05,
715
+ "loss": 0.1145,
716
  "step": 1010
717
  },
718
  {
719
+ "epoch": 8.16,
720
+ "grad_norm": 2.5671639442443848,
721
+ "learning_rate": 1.7102596552308765e-05,
722
+ "loss": 0.1094,
723
  "step": 1020
724
  },
725
  {
726
+ "epoch": 8.24,
727
+ "grad_norm": 0.9062528610229492,
728
+ "learning_rate": 1.7023510125156173e-05,
729
+ "loss": 0.0924,
730
  "step": 1030
731
  },
732
  {
733
+ "epoch": 8.32,
734
+ "grad_norm": 3.5460164546966553,
735
+ "learning_rate": 1.6943547062999027e-05,
736
+ "loss": 0.1323,
737
  "step": 1040
738
  },
739
  {
740
+ "epoch": 8.4,
741
+ "grad_norm": 2.856273889541626,
742
+ "learning_rate": 1.6862717346376706e-05,
743
+ "loss": 0.0938,
744
  "step": 1050
745
  },
746
  {
747
+ "epoch": 8.48,
748
+ "grad_norm": 3.177924633026123,
749
+ "learning_rate": 1.6781031063999515e-05,
750
+ "loss": 0.1204,
751
  "step": 1060
752
  },
753
  {
754
+ "epoch": 8.56,
755
+ "grad_norm": 1.3371851444244385,
756
+ "learning_rate": 1.6698498411489477e-05,
757
+ "loss": 0.1342,
758
  "step": 1070
759
  },
760
  {
761
+ "epoch": 8.64,
762
+ "grad_norm": 2.0833916664123535,
763
+ "learning_rate": 1.6615129690107773e-05,
764
+ "loss": 0.1337,
765
  "step": 1080
766
  },
767
  {
768
+ "epoch": 8.72,
769
+ "grad_norm": 2.4857378005981445,
770
+ "learning_rate": 1.6530935305469e-05,
771
+ "loss": 0.1233,
772
  "step": 1090
773
  },
774
  {
775
+ "epoch": 8.8,
776
+ "grad_norm": 1.7040643692016602,
777
+ "learning_rate": 1.6445925766242392e-05,
778
+ "loss": 0.121,
779
  "step": 1100
780
  },
781
  {
782
+ "epoch": 8.88,
783
+ "grad_norm": 1.2290680408477783,
784
+ "learning_rate": 1.6360111682840184e-05,
785
+ "loss": 0.1037,
786
  "step": 1110
787
  },
788
  {
789
+ "epoch": 8.96,
790
+ "grad_norm": 3.9770867824554443,
791
+ "learning_rate": 1.62735037660933e-05,
792
+ "loss": 0.1133,
793
  "step": 1120
794
  },
795
  {
796
+ "epoch": 9.04,
797
+ "grad_norm": 1.0533943176269531,
798
+ "learning_rate": 1.618611282591446e-05,
799
+ "loss": 0.0943,
800
  "step": 1130
801
  },
802
  {
803
+ "epoch": 9.12,
804
+ "grad_norm": 1.9520882368087769,
805
+ "learning_rate": 1.609794976994897e-05,
806
+ "loss": 0.0725,
807
  "step": 1140
808
  },
809
  {
810
+ "epoch": 9.2,
811
+ "grad_norm": 3.610280990600586,
812
+ "learning_rate": 1.600902560221329e-05,
813
+ "loss": 0.1015,
814
  "step": 1150
815
  },
816
  {
817
+ "epoch": 9.28,
818
+ "grad_norm": 1.3682546615600586,
819
+ "learning_rate": 1.5919351421721548e-05,
820
+ "loss": 0.0945,
821
  "step": 1160
822
  },
823
  {
824
+ "epoch": 9.36,
825
+ "grad_norm": 2.125917673110962,
826
+ "learning_rate": 1.5828938421100266e-05,
827
+ "loss": 0.0733,
828
  "step": 1170
829
  },
830
  {
831
+ "epoch": 9.44,
832
+ "grad_norm": 2.5206658840179443,
833
+ "learning_rate": 1.5737797885191316e-05,
834
+ "loss": 0.0646,
835
  "step": 1180
836
  },
837
  {
838
+ "epoch": 9.52,
839
+ "grad_norm": 2.5717523097991943,
840
+ "learning_rate": 1.5645941189643444e-05,
841
+ "loss": 0.0836,
842
  "step": 1190
843
  },
844
  {
845
+ "epoch": 9.6,
846
+ "grad_norm": 2.1724965572357178,
847
+ "learning_rate": 1.55533797994924e-05,
848
+ "loss": 0.0933,
849
  "step": 1200
850
  },
851
  {
852
+ "epoch": 9.68,
853
+ "grad_norm": 3.0643951892852783,
854
+ "learning_rate": 1.546012526772996e-05,
855
+ "loss": 0.083,
856
  "step": 1210
857
  },
858
  {
859
+ "epoch": 9.76,
860
+ "grad_norm": 1.3585115671157837,
861
+ "learning_rate": 1.5366189233861933e-05,
862
+ "loss": 0.0804,
863
  "step": 1220
864
  },
865
  {
866
+ "epoch": 9.84,
867
+ "grad_norm": 2.106403112411499,
868
+ "learning_rate": 1.5271583422455373e-05,
869
+ "loss": 0.0941,
870
  "step": 1230
871
  },
872
  {
873
+ "epoch": 9.92,
874
+ "grad_norm": 1.7128456830978394,
875
+ "learning_rate": 1.5176319641675213e-05,
876
+ "loss": 0.0668,
877
  "step": 1240
878
  },
879
  {
880
+ "epoch": 10.0,
881
+ "grad_norm": 1.7037241458892822,
882
+ "learning_rate": 1.5080409781810406e-05,
883
+ "loss": 0.1043,
884
  "step": 1250
885
  },
886
  {
887
+ "epoch": 10.08,
888
+ "grad_norm": 0.8989495038986206,
889
+ "learning_rate": 1.4983865813789869e-05,
890
+ "loss": 0.0521,
891
  "step": 1260
892
  },
893
  {
894
+ "epoch": 10.16,
895
+ "grad_norm": 1.3013043403625488,
896
+ "learning_rate": 1.488669978768833e-05,
897
+ "loss": 0.0647,
898
  "step": 1270
899
  },
900
  {
901
+ "epoch": 10.24,
902
+ "grad_norm": 0.6377755999565125,
903
+ "learning_rate": 1.47889238312223e-05,
904
+ "loss": 0.0367,
905
  "step": 1280
906
  },
907
  {
908
+ "epoch": 10.32,
909
+ "grad_norm": 2.2679171562194824,
910
+ "learning_rate": 1.4690550148236371e-05,
911
+ "loss": 0.0769,
912
  "step": 1290
913
  },
914
  {
915
+ "epoch": 10.4,
916
+ "grad_norm": 2.3292899131774902,
917
+ "learning_rate": 1.4591591017179993e-05,
918
+ "loss": 0.0908,
919
  "step": 1300
920
  },
921
  {
922
+ "epoch": 10.48,
923
+ "grad_norm": 1.8322817087173462,
924
+ "learning_rate": 1.4492058789574948e-05,
925
+ "loss": 0.0551,
926
  "step": 1310
927
  },
928
  {
929
+ "epoch": 10.56,
930
+ "grad_norm": 5.285430431365967,
931
+ "learning_rate": 1.4391965888473705e-05,
932
+ "loss": 0.0584,
933
  "step": 1320
934
  },
935
  {
936
+ "epoch": 10.64,
937
+ "grad_norm": 0.768731415271759,
938
+ "learning_rate": 1.4291324806908846e-05,
939
+ "loss": 0.0833,
940
  "step": 1330
941
  },
942
  {
943
+ "epoch": 10.72,
944
+ "grad_norm": 1.1357346773147583,
945
+ "learning_rate": 1.419014810633374e-05,
946
+ "loss": 0.0596,
947
  "step": 1340
948
  },
949
  {
950
+ "epoch": 10.8,
951
+ "grad_norm": 0.9390424489974976,
952
+ "learning_rate": 1.408844841505473e-05,
953
+ "loss": 0.0436,
954
  "step": 1350
955
  },
956
  {
957
+ "epoch": 10.88,
958
+ "grad_norm": 0.775092363357544,
959
+ "learning_rate": 1.3986238426654894e-05,
960
+ "loss": 0.0663,
961
  "step": 1360
962
  },
963
  {
964
+ "epoch": 10.96,
965
+ "grad_norm": 1.6158931255340576,
966
+ "learning_rate": 1.3883530898409736e-05,
967
+ "loss": 0.0438,
968
  "step": 1370
969
  },
970
  {
971
+ "epoch": 11.04,
972
+ "grad_norm": 1.8279812335968018,
973
+ "learning_rate": 1.3780338649694874e-05,
974
+ "loss": 0.0944,
975
  "step": 1380
976
  },
977
  {
978
+ "epoch": 11.12,
979
+ "grad_norm": 1.8661407232284546,
980
+ "learning_rate": 1.3676674560386018e-05,
981
+ "loss": 0.0398,
982
  "step": 1390
983
  },
984
  {
985
+ "epoch": 11.2,
986
+ "grad_norm": 1.8592077493667603,
987
+ "learning_rate": 1.357255156925136e-05,
988
+ "loss": 0.0561,
989
  "step": 1400
990
  },
991
  {
992
+ "epoch": 11.28,
993
+ "grad_norm": 3.479532241821289,
994
+ "learning_rate": 1.3467982672336633e-05,
995
+ "loss": 0.0408,
996
  "step": 1410
997
  },
998
  {
999
+ "epoch": 11.36,
1000
+ "grad_norm": 2.4556691646575928,
1001
+ "learning_rate": 1.336298092134302e-05,
1002
+ "loss": 0.0603,
1003
  "step": 1420
1004
  },
1005
  {
1006
+ "epoch": 11.44,
1007
+ "grad_norm": 2.028334379196167,
1008
+ "learning_rate": 1.325755942199812e-05,
1009
+ "loss": 0.0423,
1010
  "step": 1430
1011
  },
1012
  {
1013
+ "epoch": 11.52,
1014
+ "grad_norm": 0.6319141387939453,
1015
+ "learning_rate": 1.3151731332420152e-05,
1016
+ "loss": 0.0717,
1017
  "step": 1440
1018
  },
1019
  {
1020
+ "epoch": 11.6,
1021
+ "grad_norm": 1.3395280838012695,
1022
+ "learning_rate": 1.3045509861475645e-05,
1023
+ "loss": 0.0505,
1024
  "step": 1450
1025
  },
1026
  {
1027
+ "epoch": 11.68,
1028
+ "grad_norm": 2.619269847869873,
1029
+ "learning_rate": 1.293890826713077e-05,
1030
+ "loss": 0.0477,
1031
  "step": 1460
1032
  },
1033
  {
1034
+ "epoch": 11.76,
1035
+ "grad_norm": 2.6248464584350586,
1036
+ "learning_rate": 1.2831939854796567e-05,
1037
+ "loss": 0.0426,
1038
  "step": 1470
1039
  },
1040
  {
1041
+ "epoch": 11.84,
1042
+ "grad_norm": 1.672989845275879,
1043
+ "learning_rate": 1.2724617975668229e-05,
1044
+ "loss": 0.0533,
1045
  "step": 1480
1046
  },
1047
  {
1048
+ "epoch": 11.92,
1049
+ "grad_norm": 5.669078350067139,
1050
+ "learning_rate": 1.2616956025058688e-05,
1051
+ "loss": 0.0413,
1052
  "step": 1490
1053
  },
1054
  {
1055
+ "epoch": 12.0,
1056
+ "grad_norm": 2.987652063369751,
1057
+ "learning_rate": 1.2508967440726689e-05,
1058
+ "loss": 0.0442,
1059
  "step": 1500
1060
  },
1061
  {
1062
+ "epoch": 12.08,
1063
+ "grad_norm": 1.2909622192382812,
1064
+ "learning_rate": 1.2400665701199541e-05,
1065
+ "loss": 0.0565,
1066
  "step": 1510
1067
  },
1068
  {
1069
+ "epoch": 12.16,
1070
+ "grad_norm": 0.264878511428833,
1071
+ "learning_rate": 1.2292064324090842e-05,
1072
+ "loss": 0.0243,
1073
  "step": 1520
1074
  },
1075
  {
1076
+ "epoch": 12.24,
1077
+ "grad_norm": 2.0698373317718506,
1078
+ "learning_rate": 1.2183176864413262e-05,
1079
+ "loss": 0.0241,
1080
  "step": 1530
1081
  },
1082
  {
1083
+ "epoch": 12.32,
1084
+ "grad_norm": 0.20062156021595,
1085
+ "learning_rate": 1.2074016912886683e-05,
1086
+ "loss": 0.0204,
1087
  "step": 1540
1088
  },
1089
  {
1090
+ "epoch": 12.4,
1091
+ "grad_norm": 0.5076002478599548,
1092
+ "learning_rate": 1.1964598094241884e-05,
1093
+ "loss": 0.0462,
1094
  "step": 1550
1095
  },
1096
  {
1097
+ "epoch": 12.48,
1098
+ "grad_norm": 1.1509472131729126,
1099
+ "learning_rate": 1.1854934065519986e-05,
1100
+ "loss": 0.0302,
1101
  "step": 1560
1102
  },
1103
  {
1104
+ "epoch": 12.56,
1105
+ "grad_norm": 1.603930950164795,
1106
+ "learning_rate": 1.1745038514367853e-05,
1107
+ "loss": 0.0282,
1108
  "step": 1570
1109
  },
1110
  {
1111
+ "epoch": 12.64,
1112
+ "grad_norm": 0.9757488965988159,
1113
+ "learning_rate": 1.1634925157329668e-05,
1114
+ "loss": 0.0387,
1115
  "step": 1580
1116
  },
1117
  {
1118
+ "epoch": 12.72,
1119
+ "grad_norm": 1.058209776878357,
1120
+ "learning_rate": 1.1524607738134928e-05,
1121
+ "loss": 0.0445,
1122
  "step": 1590
1123
  },
1124
  {
1125
+ "epoch": 12.8,
1126
+ "grad_norm": 2.3291449546813965,
1127
+ "learning_rate": 1.141410002598301e-05,
1128
+ "loss": 0.0572,
1129
  "step": 1600
1130
  },
1131
  {
1132
+ "epoch": 12.88,
1133
+ "grad_norm": 0.8108656406402588,
1134
+ "learning_rate": 1.1303415813824599e-05,
1135
+ "loss": 0.0337,
1136
  "step": 1610
1137
  },
1138
  {
1139
+ "epoch": 12.96,
1140
+ "grad_norm": 6.551490783691406,
1141
+ "learning_rate": 1.1192568916640107e-05,
1142
+ "loss": 0.0337,
1143
  "step": 1620
1144
  },
1145
  {
1146
+ "epoch": 13.04,
1147
+ "grad_norm": 0.2524939477443695,
1148
+ "learning_rate": 1.1081573169715379e-05,
1149
+ "loss": 0.0218,
1150
  "step": 1630
1151
  },
1152
  {
1153
+ "epoch": 13.12,
1154
+ "grad_norm": 0.17596031725406647,
1155
+ "learning_rate": 1.0970442426914847e-05,
1156
+ "loss": 0.0132,
1157
  "step": 1640
1158
  },
1159
  {
1160
+ "epoch": 13.2,
1161
+ "grad_norm": 0.7680679559707642,
1162
+ "learning_rate": 1.085919055895237e-05,
1163
+ "loss": 0.0207,
1164
  "step": 1650
1165
  },
1166
  {
1167
+ "epoch": 13.28,
1168
+ "grad_norm": 0.24066166579723358,
1169
+ "learning_rate": 1.0747831451659967e-05,
1170
+ "loss": 0.0142,
1171
  "step": 1660
1172
  },
1173
  {
1174
+ "epoch": 13.36,
1175
+ "grad_norm": 3.3667285442352295,
1176
+ "learning_rate": 1.0636379004254665e-05,
1177
+ "loss": 0.0183,
1178
  "step": 1670
1179
  },
1180
  {
1181
+ "epoch": 13.44,
1182
+ "grad_norm": 1.8831534385681152,
1183
+ "learning_rate": 1.0524847127603677e-05,
1184
+ "loss": 0.0253,
1185
  "step": 1680
1186
  },
1187
  {
1188
+ "epoch": 13.52,
1189
+ "grad_norm": 1.329085111618042,
1190
+ "learning_rate": 1.0413249742488132e-05,
1191
+ "loss": 0.0118,
1192
  "step": 1690
1193
  },
1194
  {
1195
+ "epoch": 13.6,
1196
+ "grad_norm": 2.0776255130767822,
1197
+ "learning_rate": 1.030160077786556e-05,
1198
+ "loss": 0.0128,
1199
  "step": 1700
1200
  },
1201
  {
1202
+ "epoch": 13.68,
1203
+ "grad_norm": 1.02082097530365,
1204
+ "learning_rate": 1.0189914169131341e-05,
1205
+ "loss": 0.0406,
1206
  "step": 1710
1207
  },
1208
  {
1209
+ "epoch": 13.76,
1210
+ "grad_norm": 0.7947062253952026,
1211
+ "learning_rate": 1.0078203856379394e-05,
1212
+ "loss": 0.0183,
1213
  "step": 1720
1214
  },
1215
  {
1216
+ "epoch": 13.84,
1217
+ "grad_norm": 0.3098564147949219,
1218
+ "learning_rate": 9.966483782662233e-06,
1219
+ "loss": 0.0251,
1220
  "step": 1730
1221
  },
1222
  {
1223
+ "epoch": 13.92,
1224
+ "grad_norm": 0.4250534772872925,
1225
+ "learning_rate": 9.854767892250692e-06,
1226
+ "loss": 0.0394,
1227
  "step": 1740
1228
  },
1229
  {
1230
+ "epoch": 14.0,
1231
+ "grad_norm": 1.0927317142486572,
1232
+ "learning_rate": 9.743070128893452e-06,
1233
+ "loss": 0.0187,
1234
  "step": 1750
1235
  },
1236
  {
1237
+ "epoch": 14.08,
1238
+ "grad_norm": 0.5906019806861877,
1239
+ "learning_rate": 9.631404434076687e-06,
1240
+ "loss": 0.0042,
1241
  "step": 1760
1242
  },
1243
  {
1244
+ "epoch": 14.16,
1245
+ "grad_norm": 0.4497601389884949,
1246
+ "learning_rate": 9.519784745283956e-06,
1247
+ "loss": 0.0402,
1248
  "step": 1770
1249
  },
1250
  {
1251
+ "epoch": 14.24,
1252
+ "grad_norm": 0.2257859855890274,
1253
+ "learning_rate": 9.408224994256603e-06,
1254
+ "loss": 0.0039,
1255
  "step": 1780
1256
  },
1257
  {
1258
+ "epoch": 14.32,
1259
+ "grad_norm": 0.9509989023208618,
1260
+ "learning_rate": 9.296739105254869e-06,
1261
+ "loss": 0.0156,
1262
  "step": 1790
1263
  },
1264
  {
1265
+ "epoch": 14.4,
1266
+ "grad_norm": 1.7652183771133423,
1267
+ "learning_rate": 9.185340993319977e-06,
1268
+ "loss": 0.0121,
1269
  "step": 1800
1270
  },
1271
  {
1272
+ "epoch": 14.48,
1273
+ "grad_norm": 8.208316802978516,
1274
+ "learning_rate": 9.074044562537284e-06,
1275
+ "loss": 0.0389,
1276
  "step": 1810
1277
  },
1278
  {
1279
+ "epoch": 14.56,
1280
+ "grad_norm": 0.13252846896648407,
1281
+ "learning_rate": 8.962863704300893e-06,
1282
+ "loss": 0.011,
1283
  "step": 1820
1284
  },
1285
  {
1286
+ "epoch": 14.64,
1287
+ "grad_norm": 0.23998035490512848,
1288
+ "learning_rate": 8.851812295579789e-06,
1289
+ "loss": 0.0248,
1290
  "step": 1830
1291
  },
1292
  {
1293
+ "epoch": 14.72,
1294
+ "grad_norm": 0.10990134626626968,
1295
+ "learning_rate": 8.740904197185794e-06,
1296
+ "loss": 0.0204,
1297
  "step": 1840
1298
  },
1299
  {
1300
+ "epoch": 14.8,
1301
+ "grad_norm": 3.542644500732422,
1302
+ "learning_rate": 8.630153252043543e-06,
1303
+ "loss": 0.0163,
1304
  "step": 1850
1305
  },
1306
  {
1307
+ "epoch": 14.88,
1308
+ "grad_norm": 0.7273716926574707,
1309
+ "learning_rate": 8.519573283462688e-06,
1310
+ "loss": 0.0068,
1311
  "step": 1860
1312
  },
1313
  {
1314
+ "epoch": 14.96,
1315
+ "grad_norm": 0.036577966064214706,
1316
+ "learning_rate": 8.409178093412549e-06,
1317
+ "loss": 0.0084,
1318
  "step": 1870
1319
  },
1320
  {
1321
+ "epoch": 15.04,
1322
+ "grad_norm": 0.1642957627773285,
1323
+ "learning_rate": 8.298981460799426e-06,
1324
+ "loss": 0.0093,
1325
  "step": 1880
1326
  },
1327
  {
1328
+ "epoch": 15.12,
1329
+ "grad_norm": 0.11017989367246628,
1330
+ "learning_rate": 8.188997139746807e-06,
1331
+ "loss": 0.0076,
1332
  "step": 1890
1333
  },
1334
  {
1335
+ "epoch": 15.2,
1336
+ "grad_norm": 0.04475900158286095,
1337
+ "learning_rate": 8.079238857878631e-06,
1338
+ "loss": 0.0059,
1339
  "step": 1900
1340
  },
1341
  {
1342
+ "epoch": 15.28,
1343
+ "grad_norm": 1.0872098207473755,
1344
+ "learning_rate": 7.969720314605915e-06,
1345
+ "loss": 0.0051,
1346
  "step": 1910
1347
  },
1348
  {
1349
+ "epoch": 15.36,
1350
+ "grad_norm": 0.32729968428611755,
1351
+ "learning_rate": 7.860455179416837e-06,
1352
+ "loss": 0.0016,
1353
  "step": 1920
1354
  },
1355
  {
1356
+ "epoch": 15.44,
1357
+ "grad_norm": 1.3896279335021973,
1358
+ "learning_rate": 7.751457090170616e-06,
1359
+ "loss": 0.0066,
1360
  "step": 1930
1361
  },
1362
  {
1363
+ "epoch": 15.52,
1364
+ "grad_norm": 0.24193909764289856,
1365
+ "learning_rate": 7.642739651395295e-06,
1366
+ "loss": 0.0049,
1367
  "step": 1940
1368
  },
1369
  {
1370
+ "epoch": 15.6,
1371
+ "grad_norm": 0.0498543456196785,
1372
+ "learning_rate": 7.534316432589706e-06,
1373
+ "loss": 0.0065,
1374
  "step": 1950
1375
  },
1376
  {
1377
+ "epoch": 15.68,
1378
+ "grad_norm": 0.053042277693748474,
1379
+ "learning_rate": 7.426200966529795e-06,
1380
+ "loss": 0.0096,
1381
  "step": 1960
1382
  },
1383
  {
1384
+ "epoch": 15.76,
1385
+ "grad_norm": 0.6754854917526245,
1386
+ "learning_rate": 7.318406747579556e-06,
1387
+ "loss": 0.0074,
1388
  "step": 1970
1389
  },
1390
  {
1391
+ "epoch": 15.84,
1392
+ "grad_norm": 0.05438900366425514,
1393
+ "learning_rate": 7.210947230006713e-06,
1394
+ "loss": 0.0025,
1395
  "step": 1980
1396
  },
1397
  {
1398
+ "epoch": 15.92,
1399
+ "grad_norm": 0.0401092991232872,
1400
+ "learning_rate": 7.103835826303451e-06,
1401
+ "loss": 0.0022,
1402
  "step": 1990
1403
  },
1404
  {
1405
+ "epoch": 16.0,
1406
+ "grad_norm": 0.05556059256196022,
1407
+ "learning_rate": 6.997085905512346e-06,
1408
+ "loss": 0.0049,
1409
  "step": 2000
1410
  },
1411
  {
1412
+ "epoch": 16.08,
1413
+ "grad_norm": 0.07605039328336716,
1414
+ "learning_rate": 6.8907107915577075e-06,
1415
+ "loss": 0.0051,
1416
  "step": 2010
1417
  },
1418
  {
1419
+ "epoch": 16.16,
1420
+ "grad_norm": 0.07472195476293564,
1421
+ "learning_rate": 6.7847237615825636e-06,
1422
+ "loss": 0.0031,
1423
  "step": 2020
1424
  },
1425
  {
1426
+ "epoch": 16.24,
1427
+ "grad_norm": 0.203807532787323,
1428
+ "learning_rate": 6.6791380442914866e-06,
1429
+ "loss": 0.0014,
1430
  "step": 2030
1431
  },
1432
  {
1433
+ "epoch": 16.32,
1434
+ "grad_norm": 0.08696573972702026,
1435
+ "learning_rate": 6.573966818299461e-06,
1436
+ "loss": 0.0024,
1437
  "step": 2040
1438
  },
1439
  {
1440
+ "epoch": 16.4,
1441
+ "grad_norm": 0.025404011830687523,
1442
+ "learning_rate": 6.469223210486992e-06,
1443
+ "loss": 0.0007,
1444
  "step": 2050
1445
  },
1446
  {
1447
+ "epoch": 16.48,
1448
+ "grad_norm": 0.05262196436524391,
1449
+ "learning_rate": 6.364920294361701e-06,
1450
+ "loss": 0.0006,
1451
  "step": 2060
1452
  },
1453
  {
1454
+ "epoch": 16.56,
1455
+ "grad_norm": 0.029066545888781548,
1456
+ "learning_rate": 6.261071088426546e-06,
1457
+ "loss": 0.0004,
1458
  "step": 2070
1459
  },
1460
  {
1461
+ "epoch": 16.64,
1462
+ "grad_norm": 0.02618943341076374,
1463
+ "learning_rate": 6.1576885545549355e-06,
1464
+ "loss": 0.0022,
1465
  "step": 2080
1466
  },
1467
  {
1468
+ "epoch": 16.72,
1469
+ "grad_norm": 0.06179804354906082,
1470
+ "learning_rate": 6.054785596372894e-06,
1471
+ "loss": 0.0005,
1472
  "step": 2090
1473
  },
1474
  {
1475
+ "epoch": 16.8,
1476
+ "grad_norm": 0.014377947896718979,
1477
+ "learning_rate": 5.952375057648509e-06,
1478
+ "loss": 0.0028,
1479
  "step": 2100
1480
  },
1481
  {
1482
+ "epoch": 16.88,
1483
+ "grad_norm": 0.039646223187446594,
1484
+ "learning_rate": 5.850469720688847e-06,
1485
+ "loss": 0.001,
1486
  "step": 2110
1487
  },
1488
  {
1489
+ "epoch": 16.96,
1490
+ "grad_norm": 0.44483599066734314,
1491
+ "learning_rate": 5.74908230474453e-06,
1492
+ "loss": 0.0044,
1493
  "step": 2120
1494
  },
1495
  {
1496
+ "epoch": 17.04,
1497
+ "grad_norm": 0.024487623944878578,
1498
+ "learning_rate": 5.648225464422189e-06,
1499
+ "loss": 0.0007,
1500
  "step": 2130
1501
  },
1502
  {
1503
+ "epoch": 17.12,
1504
+ "grad_norm": 0.0326782651245594,
1505
+ "learning_rate": 5.547911788105001e-06,
1506
+ "loss": 0.0008,
1507
  "step": 2140
1508
  },
1509
  {
1510
+ "epoch": 17.2,
1511
+ "grad_norm": 0.015867168083786964,
1512
+ "learning_rate": 5.4481537963814675e-06,
1513
+ "loss": 0.0004,
1514
  "step": 2150
1515
  },
1516
  {
1517
+ "epoch": 17.28,
1518
+ "grad_norm": 0.022902317345142365,
1519
+ "learning_rate": 5.348963940482663e-06,
1520
+ "loss": 0.0003,
1521
  "step": 2160
1522
  },
1523
  {
1524
+ "epoch": 17.36,
1525
+ "grad_norm": 0.024737635627388954,
1526
+ "learning_rate": 5.25035460072814e-06,
1527
+ "loss": 0.0023,
1528
  "step": 2170
1529
  },
1530
  {
1531
+ "epoch": 17.44,
1532
+ "grad_norm": 0.019413290545344353,
1533
+ "learning_rate": 5.15233808498071e-06,
1534
+ "loss": 0.0004,
1535
  "step": 2180
1536
  },
1537
  {
1538
+ "epoch": 17.52,
1539
+ "grad_norm": 0.01702185347676277,
1540
+ "learning_rate": 5.054926627110208e-06,
1541
+ "loss": 0.0004,
1542
  "step": 2190
1543
  },
1544
  {
1545
+ "epoch": 17.6,
1546
+ "grad_norm": 2.464230537414551,
1547
+ "learning_rate": 4.9581323854665695e-06,
1548
+ "loss": 0.0049,
1549
  "step": 2200
1550
  },
1551
  {
1552
+ "epoch": 17.68,
1553
+ "grad_norm": 0.013668897561728954,
1554
+ "learning_rate": 4.861967441362262e-06,
1555
+ "loss": 0.0004,
1556
  "step": 2210
1557
  },
1558
  {
1559
+ "epoch": 17.76,
1560
+ "grad_norm": 0.014037170447409153,
1561
+ "learning_rate": 4.766443797564375e-06,
1562
+ "loss": 0.0004,
1563
  "step": 2220
1564
  },
1565
  {
1566
+ "epoch": 17.84,
1567
+ "grad_norm": 0.011271145194768906,
1568
+ "learning_rate": 4.671573376796511e-06,
1569
+ "loss": 0.0005,
1570
  "step": 2230
1571
  },
1572
  {
1573
+ "epoch": 17.92,
1574
+ "grad_norm": 0.015100213699042797,
1575
+ "learning_rate": 4.57736802025065e-06,
1576
+ "loss": 0.0007,
1577
  "step": 2240
1578
  },
1579
  {
1580
+ "epoch": 18.0,
1581
+ "grad_norm": 0.1259368360042572,
1582
+ "learning_rate": 4.48383948610919e-06,
1583
+ "loss": 0.0003,
1584
  "step": 2250
1585
  },
1586
  {
1587
+ "epoch": 18.08,
1588
+ "grad_norm": 0.04630790278315544,
1589
+ "learning_rate": 4.390999448077375e-06,
1590
+ "loss": 0.0004,
1591
  "step": 2260
1592
  },
1593
  {
1594
+ "epoch": 18.16,
1595
+ "grad_norm": 0.03695525601506233,
1596
+ "learning_rate": 4.298859493926235e-06,
1597
+ "loss": 0.0003,
1598
  "step": 2270
1599
  },
1600
  {
1601
+ "epoch": 18.24,
1602
+ "grad_norm": 0.00454062782227993,
1603
+ "learning_rate": 4.207431124046267e-06,
1604
+ "loss": 0.0003,
1605
  "step": 2280
1606
  },
1607
  {
1608
+ "epoch": 18.32,
1609
+ "grad_norm": 0.016318323090672493,
1610
+ "learning_rate": 4.116725750012035e-06,
1611
+ "loss": 0.0003,
1612
  "step": 2290
1613
  },
1614
  {
1615
+ "epoch": 18.4,
1616
+ "grad_norm": 0.02545909956097603,
1617
+ "learning_rate": 4.026754693157816e-06,
1618
+ "loss": 0.0003,
1619
  "step": 2300
1620
  },
1621
  {
1622
+ "epoch": 18.48,
1623
+ "grad_norm": 0.014473868533968925,
1624
+ "learning_rate": 3.937529183164562e-06,
1625
+ "loss": 0.0003,
1626
  "step": 2310
1627
  },
1628
  {
1629
+ "epoch": 18.56,
1630
+ "grad_norm": 0.014056684449315071,
1631
+ "learning_rate": 3.84906035665826e-06,
1632
  "loss": 0.0002,
1633
  "step": 2320
1634
  },
1635
  {
1636
+ "epoch": 18.64,
1637
+ "grad_norm": 0.013438849709928036,
1638
+ "learning_rate": 3.7613592558199162e-06,
1639
+ "loss": 0.0002,
1640
  "step": 2330
1641
  },
1642
  {
1643
+ "epoch": 18.72,
1644
+ "grad_norm": 0.01617852970957756,
1645
+ "learning_rate": 3.6744368270073393e-06,
1646
+ "loss": 0.0007,
1647
  "step": 2340
1648
  },
1649
  {
1650
+ "epoch": 18.8,
1651
+ "grad_norm": 0.013655665330588818,
1652
+ "learning_rate": 3.5883039193888914e-06,
1653
+ "loss": 0.0007,
1654
  "step": 2350
1655
  },
1656
  {
1657
+ "epoch": 18.88,
1658
+ "grad_norm": 0.014764860272407532,
1659
+ "learning_rate": 3.502971283589326e-06,
1660
+ "loss": 0.0005,
1661
  "step": 2360
1662
  },
1663
  {
1664
+ "epoch": 18.96,
1665
+ "grad_norm": 0.03937000408768654,
1666
+ "learning_rate": 3.418449570347986e-06,
1667
+ "loss": 0.0003,
1668
  "step": 2370
1669
  },
1670
  {
1671
+ "epoch": 19.04,
1672
+ "grad_norm": 0.014079142361879349,
1673
+ "learning_rate": 3.334749329189415e-06,
1674
+ "loss": 0.0003,
1675
  "step": 2380
1676
  },
1677
  {
1678
+ "epoch": 19.12,
1679
+ "grad_norm": 0.016212107613682747,
1680
+ "learning_rate": 3.2518810071066363e-06,
1681
+ "loss": 0.0002,
1682
  "step": 2390
1683
  },
1684
  {
1685
+ "epoch": 19.2,
1686
+ "grad_norm": 0.015338711440563202,
1687
+ "learning_rate": 3.1698549472572203e-06,
1688
+ "loss": 0.0003,
1689
  "step": 2400
1690
  },
1691
  {
1692
+ "epoch": 19.28,
1693
+ "grad_norm": 0.02485400065779686,
1694
+ "learning_rate": 3.0886813876723075e-06,
1695
+ "loss": 0.0002,
1696
  "step": 2410
1697
  },
1698
  {
1699
+ "epoch": 19.36,
1700
+ "grad_norm": 0.017927484586834908,
1701
+ "learning_rate": 3.0083704599787423e-06,
1702
+ "loss": 0.0003,
1703
  "step": 2420
1704
  },
1705
  {
1706
+ "epoch": 19.44,
1707
+ "grad_norm": 0.012980809435248375,
1708
+ "learning_rate": 2.9289321881345257e-06,
1709
+ "loss": 0.0005,
1710
  "step": 2430
1711
  },
1712
  {
1713
+ "epoch": 19.52,
1714
+ "grad_norm": 0.03192094713449478,
1715
+ "learning_rate": 2.850376487177656e-06,
1716
+ "loss": 0.0002,
1717
  "step": 2440
1718
  },
1719
  {
1720
+ "epoch": 19.6,
1721
+ "grad_norm": 0.019058849662542343,
1722
+ "learning_rate": 2.7727131619886017e-06,
1723
  "loss": 0.0002,
1724
  "step": 2450
1725
  },
1726
  {
1727
+ "epoch": 19.68,
1728
+ "grad_norm": 0.012099268846213818,
1729
+ "learning_rate": 2.6959519060665195e-06,
1730
  "loss": 0.0002,
1731
  "step": 2460
1732
  },
1733
  {
1734
+ "epoch": 19.76,
1735
+ "grad_norm": 0.010341337881982327,
1736
+ "learning_rate": 2.6201023003193437e-06,
1737
+ "loss": 0.0002,
1738
  "step": 2470
1739
  },
1740
  {
1741
+ "epoch": 19.84,
1742
+ "grad_norm": 0.014265513978898525,
1743
+ "learning_rate": 2.545173811867977e-06,
1744
+ "loss": 0.0002,
1745
  "step": 2480
1746
  },
1747
  {
1748
+ "epoch": 19.92,
1749
+ "grad_norm": 0.01814926601946354,
1750
+ "learning_rate": 2.471175792864642e-06,
1751
+ "loss": 0.001,
1752
  "step": 2490
1753
  },
1754
  {
1755
+ "epoch": 20.0,
1756
+ "grad_norm": 0.011018522083759308,
1757
+ "learning_rate": 2.3981174793255956e-06,
1758
+ "loss": 0.0002,
1759
  "step": 2500
1760
  },
1761
  {
1762
+ "epoch": 20.08,
1763
+ "grad_norm": 0.015308992005884647,
1764
+ "learning_rate": 2.3260079899783492e-06,
1765
+ "loss": 0.0008,
1766
  "step": 2510
1767
  },
1768
  {
1769
+ "epoch": 20.16,
1770
+ "grad_norm": 0.010280388407409191,
1771
+ "learning_rate": 2.254856325123529e-06,
1772
+ "loss": 0.0002,
1773
  "step": 2520
1774
  },
1775
  {
1776
+ "epoch": 20.24,
1777
+ "grad_norm": 0.011187535710632801,
1778
+ "learning_rate": 2.1846713655114836e-06,
1779
+ "loss": 0.0002,
1780
  "step": 2530
1781
  },
1782
  {
1783
+ "epoch": 20.32,
1784
+ "grad_norm": 0.015644947066903114,
1785
+ "learning_rate": 2.115461871233867e-06,
1786
+ "loss": 0.0002,
1787
  "step": 2540
1788
  },
1789
  {
1790
+ "epoch": 20.4,
1791
+ "grad_norm": 0.00875458400696516,
1792
+ "learning_rate": 2.0472364806302313e-06,
1793
  "loss": 0.0002,
1794
  "step": 2550
1795
  },
1796
  {
1797
+ "epoch": 20.48,
1798
+ "grad_norm": 0.017235323786735535,
1799
+ "learning_rate": 1.9800037092098477e-06,
1800
+ "loss": 0.0002,
1801
  "step": 2560
1802
  },
1803
  {
1804
+ "epoch": 20.56,
1805
+ "grad_norm": 0.03400786966085434,
1806
+ "learning_rate": 1.9137719485888527e-06,
1807
+ "loss": 0.0002,
1808
  "step": 2570
1809
  },
1810
  {
1811
+ "epoch": 20.64,
1812
+ "grad_norm": 0.011186143383383751,
1813
+ "learning_rate": 1.8485494654428482e-06,
1814
+ "loss": 0.0002,
1815
  "step": 2580
1816
  },
1817
  {
1818
+ "epoch": 20.72,
1819
+ "grad_norm": 0.008517293259501457,
1820
+ "learning_rate": 1.784344400475093e-06,
1821
+ "loss": 0.0002,
1822
  "step": 2590
1823
  },
1824
  {
1825
+ "epoch": 20.8,
1826
+ "grad_norm": 0.010979237034916878,
1827
+ "learning_rate": 1.7211647674004483e-06,
1828
+ "loss": 0.0002,
1829
  "step": 2600
1830
  },
1831
  {
1832
+ "epoch": 20.88,
1833
+ "grad_norm": 0.01790277473628521,
1834
+ "learning_rate": 1.659018451945128e-06,
1835
+ "loss": 0.0002,
1836
  "step": 2610
1837
  },
1838
  {
1839
+ "epoch": 20.96,
1840
+ "grad_norm": 0.01020762138068676,
1841
+ "learning_rate": 1.5979132108624572e-06,
1842
+ "loss": 0.0002,
1843
  "step": 2620
1844
  },
1845
  {
1846
+ "epoch": 21.04,
1847
+ "grad_norm": 0.014280925504863262,
1848
+ "learning_rate": 1.5378566709647225e-06,
1849
+ "loss": 0.0003,
1850
  "step": 2630
1851
  },
1852
  {
1853
+ "epoch": 21.12,
1854
+ "grad_norm": 0.012542898766696453,
1855
+ "learning_rate": 1.4788563281712253e-06,
1856
+ "loss": 0.0002,
1857
  "step": 2640
1858
  },
1859
  {
1860
+ "epoch": 21.2,
1861
+ "grad_norm": 0.0072167362086474895,
1862
+ "learning_rate": 1.420919546572691e-06,
1863
+ "loss": 0.0002,
1864
  "step": 2650
1865
  },
1866
  {
1867
+ "epoch": 21.28,
1868
+ "grad_norm": 0.010835397057235241,
1869
+ "learning_rate": 1.364053557512126e-06,
1870
+ "loss": 0.0002,
1871
  "step": 2660
1872
  },
1873
  {
1874
+ "epoch": 21.36,
1875
+ "grad_norm": 0.010230830870568752,
1876
+ "learning_rate": 1.308265458682234e-06,
1877
+ "loss": 0.0004,
1878
  "step": 2670
1879
  },
1880
  {
1881
+ "epoch": 21.44,
1882
+ "grad_norm": 0.00813596136868,
1883
+ "learning_rate": 1.2535622132395242e-06,
1884
+ "loss": 0.0005,
1885
  "step": 2680
1886
  },
1887
  {
1888
+ "epoch": 21.52,
1889
+ "grad_norm": 0.02058684267103672,
1890
+ "learning_rate": 1.1999506489352208e-06,
1891
+ "loss": 0.0002,
1892
  "step": 2690
1893
  },
1894
  {
1895
+ "epoch": 21.6,
1896
+ "grad_norm": 0.022641103714704514,
1897
+ "learning_rate": 1.1474374572630432e-06,
1898
+ "loss": 0.0002,
1899
  "step": 2700
1900
  },
1901
  {
1902
+ "epoch": 21.68,
1903
+ "grad_norm": 0.015892324969172478,
1904
+ "learning_rate": 1.0960291926240263e-06,
1905
+ "loss": 0.0002,
1906
  "step": 2710
1907
  },
1908
  {
1909
+ "epoch": 21.76,
1910
+ "grad_norm": 0.01219907309859991,
1911
+ "learning_rate": 1.0457322715084305e-06,
1912
+ "loss": 0.0002,
1913
  "step": 2720
1914
  },
1915
  {
1916
+ "epoch": 21.84,
1917
+ "grad_norm": 0.009200166910886765,
1918
+ "learning_rate": 9.965529716948684e-07,
1919
+ "loss": 0.0002,
1920
  "step": 2730
1921
  },
1922
  {
1923
+ "epoch": 21.92,
1924
+ "grad_norm": 0.00976441241800785,
1925
+ "learning_rate": 9.484974314667561e-07,
1926
+ "loss": 0.0002,
1927
  "step": 2740
1928
  },
1929
  {
1930
+ "epoch": 22.0,
1931
+ "grad_norm": 0.010954627767205238,
1932
+ "learning_rate": 9.015716488461656e-07,
1933
+ "loss": 0.0002,
1934
  "step": 2750
1935
  },
1936
  {
1937
+ "epoch": 22.08,
1938
+ "grad_norm": 0.016307951882481575,
1939
+ "learning_rate": 8.557814808451737e-07,
1940
+ "loss": 0.0002,
1941
  "step": 2760
1942
  },
1943
  {
1944
+ "epoch": 22.16,
1945
+ "grad_norm": 0.009793213568627834,
1946
+ "learning_rate": 8.11132642734841e-07,
1947
  "loss": 0.0001,
1948
  "step": 2770
1949
  },
1950
  {
1951
+ "epoch": 22.24,
1952
+ "grad_norm": 0.023596106097102165,
1953
+ "learning_rate": 7.676307073318479e-07,
1954
+ "loss": 0.0002,
1955
  "step": 2780
1956
  },
1957
  {
1958
+ "epoch": 22.32,
1959
+ "grad_norm": 0.011904028244316578,
1960
+ "learning_rate": 7.252811043029373e-07,
1961
+ "loss": 0.0004,
1962
  "step": 2790
1963
  },
1964
  {
1965
+ "epoch": 22.4,
1966
+ "grad_norm": 0.013149023056030273,
1967
+ "learning_rate": 6.840891194872112e-07,
1968
+ "loss": 0.0002,
1969
  "step": 2800
1970
  },
1971
  {
1972
+ "epoch": 22.48,
1973
+ "grad_norm": 0.011410288512706757,
1974
+ "learning_rate": 6.440598942363796e-07,
1975
+ "loss": 0.0002,
1976
  "step": 2810
1977
  },
1978
  {
1979
+ "epoch": 22.56,
1980
+ "grad_norm": 0.008377713151276112,
1981
+ "learning_rate": 6.051984247730447e-07,
1982
+ "loss": 0.0002,
1983
  "step": 2820
1984
  },
1985
  {
1986
+ "epoch": 22.64,
1987
+ "grad_norm": 0.19204267859458923,
1988
+ "learning_rate": 5.675095615671144e-07,
1989
+ "loss": 0.0005,
1990
  "step": 2830
1991
  },
1992
  {
1993
+ "epoch": 22.72,
1994
+ "grad_norm": 0.017059409990906715,
1995
+ "learning_rate": 5.309980087303713e-07,
1996
+ "loss": 0.0002,
1997
  "step": 2840
1998
  },
1999
  {
2000
+ "epoch": 22.8,
2001
+ "grad_norm": 0.00994216650724411,
2002
+ "learning_rate": 4.956683234293491e-07,
2003
+ "loss": 0.0002,
2004
  "step": 2850
2005
  },
2006
  {
2007
+ "epoch": 22.88,
2008
+ "grad_norm": 0.017344018444418907,
2009
+ "learning_rate": 4.61524915316528e-07,
2010
  "loss": 0.0001,
2011
  "step": 2860
2012
  },
2013
  {
2014
+ "epoch": 22.96,
2015
+ "grad_norm": 0.009385120123624802,
2016
+ "learning_rate": 4.285720459799425e-07,
2017
+ "loss": 0.0002,
2018
  "step": 2870
2019
  },
2020
  {
2021
+ "epoch": 23.04,
2022
+ "grad_norm": 0.009774768725037575,
2023
+ "learning_rate": 3.9681382841128323e-07,
2024
+ "loss": 0.0002,
2025
  "step": 2880
2026
  },
2027
  {
2028
+ "epoch": 23.12,
2029
+ "grad_norm": 0.13636986911296844,
2030
+ "learning_rate": 3.6625422649252617e-07,
2031
+ "loss": 0.0005,
2032
  "step": 2890
2033
  },
2034
  {
2035
+ "epoch": 23.2,
2036
+ "grad_norm": 0.011067167855799198,
2037
+ "learning_rate": 3.368970545011874e-07,
2038
+ "loss": 0.0002,
2039
  "step": 2900
2040
  },
2041
  {
2042
+ "epoch": 23.28,
2043
+ "grad_norm": 0.008906451985239983,
2044
+ "learning_rate": 3.0874597663425046e-07,
2045
+ "loss": 0.0002,
2046
  "step": 2910
2047
  },
2048
  {
2049
+ "epoch": 23.36,
2050
+ "grad_norm": 0.011505813337862492,
2051
+ "learning_rate": 2.818045065508168e-07,
2052
  "loss": 0.0001,
2053
  "step": 2920
2054
  },
2055
  {
2056
+ "epoch": 23.44,
2057
+ "grad_norm": 0.01766866073012352,
2058
+ "learning_rate": 2.560760069335511e-07,
2059
+ "loss": 0.0002,
2060
  "step": 2930
2061
  },
2062
  {
2063
+ "epoch": 23.52,
2064
+ "grad_norm": 0.01433554571121931,
2065
+ "learning_rate": 2.315636890689743e-07,
2066
+ "loss": 0.0002,
2067
  "step": 2940
2068
  },
2069
  {
2070
+ "epoch": 23.6,
2071
+ "grad_norm": 0.009151219390332699,
2072
+ "learning_rate": 2.082706124466416e-07,
2073
+ "loss": 0.0002,
2074
  "step": 2950
2075
  },
2076
  {
2077
+ "epoch": 23.68,
2078
+ "grad_norm": 0.014051095582544804,
2079
+ "learning_rate": 1.8619968437727954e-07,
2080
+ "loss": 0.0002,
2081
  "step": 2960
2082
  },
2083
  {
2084
+ "epoch": 23.76,
2085
+ "grad_norm": 0.01479571033269167,
2086
+ "learning_rate": 1.6535365962991322e-07,
2087
+ "loss": 0.0002,
2088
  "step": 2970
2089
  },
2090
  {
2091
+ "epoch": 23.84,
2092
+ "grad_norm": 0.17555338144302368,
2093
+ "learning_rate": 1.4573514008802693e-07,
2094
+ "loss": 0.0005,
2095
  "step": 2980
2096
  },
2097
  {
2098
+ "epoch": 23.92,
2099
+ "grad_norm": 0.01707664132118225,
2100
+ "learning_rate": 1.2734657442481368e-07,
2101
+ "loss": 0.0002,
2102
  "step": 2990
2103
  },
2104
  {
2105
+ "epoch": 24.0,
2106
+ "grad_norm": 0.023542573675513268,
2107
+ "learning_rate": 1.1019025779754666e-07,
2108
+ "loss": 0.0002,
2109
  "step": 3000
2110
  }
2111
  ],
2112
  "logging_steps": 10,
2113
+ "max_steps": 3125,
2114
  "num_input_tokens_seen": 0,
2115
+ "num_train_epochs": 25,
2116
  "save_steps": 1000,
2117
  "stateful_callbacks": {
2118
  "TrainerControl": {
 
2126
  "attributes": {}
2127
  }
2128
  },
2129
+ "total_flos": 1.1967026057917235e+17,
2130
+ "train_batch_size": 2,
2131
  "trial_name": null,
2132
  "trial_params": null
2133
  }
checkpoint-3000/training_args.bin CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:fe16ad85cf67a1d616e8a664749a043a7e42845bd6dfe8881e3e3c947d23497a
3
  size 5624
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:27ff69d62d45c0f05227ab8bfdfb70f5de878e29a775a5568fe676d60d677a54
3
  size 5624
checkpoint-3125/README.md ADDED
@@ -0,0 +1,202 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ base_model: Qwen/Qwen2-VL-2B-Instruct
3
+ library_name: peft
4
+ ---
5
+
6
+ # Model Card for Model ID
7
+
8
+ <!-- Provide a quick summary of what the model is/does. -->
9
+
10
+
11
+
12
+ ## Model Details
13
+
14
+ ### Model Description
15
+
16
+ <!-- Provide a longer summary of what this model is. -->
17
+
18
+
19
+
20
+ - **Developed by:** [More Information Needed]
21
+ - **Funded by [optional]:** [More Information Needed]
22
+ - **Shared by [optional]:** [More Information Needed]
23
+ - **Model type:** [More Information Needed]
24
+ - **Language(s) (NLP):** [More Information Needed]
25
+ - **License:** [More Information Needed]
26
+ - **Finetuned from model [optional]:** [More Information Needed]
27
+
28
+ ### Model Sources [optional]
29
+
30
+ <!-- Provide the basic links for the model. -->
31
+
32
+ - **Repository:** [More Information Needed]
33
+ - **Paper [optional]:** [More Information Needed]
34
+ - **Demo [optional]:** [More Information Needed]
35
+
36
+ ## Uses
37
+
38
+ <!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
39
+
40
+ ### Direct Use
41
+
42
+ <!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
43
+
44
+ [More Information Needed]
45
+
46
+ ### Downstream Use [optional]
47
+
48
+ <!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
49
+
50
+ [More Information Needed]
51
+
52
+ ### Out-of-Scope Use
53
+
54
+ <!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
55
+
56
+ [More Information Needed]
57
+
58
+ ## Bias, Risks, and Limitations
59
+
60
+ <!-- This section is meant to convey both technical and sociotechnical limitations. -->
61
+
62
+ [More Information Needed]
63
+
64
+ ### Recommendations
65
+
66
+ <!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
67
+
68
+ Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
69
+
70
+ ## How to Get Started with the Model
71
+
72
+ Use the code below to get started with the model.
73
+
74
+ [More Information Needed]
75
+
76
+ ## Training Details
77
+
78
+ ### Training Data
79
+
80
+ <!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
81
+
82
+ [More Information Needed]
83
+
84
+ ### Training Procedure
85
+
86
+ <!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
87
+
88
+ #### Preprocessing [optional]
89
+
90
+ [More Information Needed]
91
+
92
+
93
+ #### Training Hyperparameters
94
+
95
+ - **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
96
+
97
+ #### Speeds, Sizes, Times [optional]
98
+
99
+ <!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
100
+
101
+ [More Information Needed]
102
+
103
+ ## Evaluation
104
+
105
+ <!-- This section describes the evaluation protocols and provides the results. -->
106
+
107
+ ### Testing Data, Factors & Metrics
108
+
109
+ #### Testing Data
110
+
111
+ <!-- This should link to a Dataset Card if possible. -->
112
+
113
+ [More Information Needed]
114
+
115
+ #### Factors
116
+
117
+ <!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
118
+
119
+ [More Information Needed]
120
+
121
+ #### Metrics
122
+
123
+ <!-- These are the evaluation metrics being used, ideally with a description of why. -->
124
+
125
+ [More Information Needed]
126
+
127
+ ### Results
128
+
129
+ [More Information Needed]
130
+
131
+ #### Summary
132
+
133
+
134
+
135
+ ## Model Examination [optional]
136
+
137
+ <!-- Relevant interpretability work for the model goes here -->
138
+
139
+ [More Information Needed]
140
+
141
+ ## Environmental Impact
142
+
143
+ <!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
144
+
145
+ Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
146
+
147
+ - **Hardware Type:** [More Information Needed]
148
+ - **Hours used:** [More Information Needed]
149
+ - **Cloud Provider:** [More Information Needed]
150
+ - **Compute Region:** [More Information Needed]
151
+ - **Carbon Emitted:** [More Information Needed]
152
+
153
+ ## Technical Specifications [optional]
154
+
155
+ ### Model Architecture and Objective
156
+
157
+ [More Information Needed]
158
+
159
+ ### Compute Infrastructure
160
+
161
+ [More Information Needed]
162
+
163
+ #### Hardware
164
+
165
+ [More Information Needed]
166
+
167
+ #### Software
168
+
169
+ [More Information Needed]
170
+
171
+ ## Citation [optional]
172
+
173
+ <!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
174
+
175
+ **BibTeX:**
176
+
177
+ [More Information Needed]
178
+
179
+ **APA:**
180
+
181
+ [More Information Needed]
182
+
183
+ ## Glossary [optional]
184
+
185
+ <!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
186
+
187
+ [More Information Needed]
188
+
189
+ ## More Information [optional]
190
+
191
+ [More Information Needed]
192
+
193
+ ## Model Card Authors [optional]
194
+
195
+ [More Information Needed]
196
+
197
+ ## Model Card Contact
198
+
199
+ [More Information Needed]
200
+ ### Framework versions
201
+
202
+ - PEFT 0.12.0
checkpoint-3125/adapter_config.json ADDED
@@ -0,0 +1,26 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "alpha_pattern": {},
3
+ "auto_mapping": null,
4
+ "base_model_name_or_path": "Qwen/Qwen2-VL-2B-Instruct",
5
+ "bias": "none",
6
+ "fan_in_fan_out": false,
7
+ "inference_mode": true,
8
+ "init_lora_weights": true,
9
+ "layer_replication": null,
10
+ "layers_pattern": null,
11
+ "layers_to_transform": null,
12
+ "loftq_config": {},
13
+ "lora_alpha": 32,
14
+ "lora_dropout": 0.0,
15
+ "megatron_config": null,
16
+ "megatron_core": "megatron.core",
17
+ "modules_to_save": null,
18
+ "peft_type": "LORA",
19
+ "r": 16,
20
+ "rank_pattern": {},
21
+ "revision": null,
22
+ "target_modules": "^(?!.*patch_embed).*(?:o_proj|fc2|v_proj|down_proj|q_proj|up_proj|proj|qkv|gate_proj|k_proj|fc1).*",
23
+ "task_type": "CAUSAL_LM",
24
+ "use_dora": false,
25
+ "use_rslora": false
26
+ }
checkpoint-3125/adapter_model.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:335a6122bd58fa538a3cd935a28d1549f4413046e91785944e52bf8e4db9e78c
3
+ size 115886968
checkpoint-3125/added_tokens.json ADDED
@@ -0,0 +1,16 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "<|box_end|>": 151649,
3
+ "<|box_start|>": 151648,
4
+ "<|endoftext|>": 151643,
5
+ "<|im_end|>": 151645,
6
+ "<|im_start|>": 151644,
7
+ "<|image_pad|>": 151655,
8
+ "<|object_ref_end|>": 151647,
9
+ "<|object_ref_start|>": 151646,
10
+ "<|quad_end|>": 151651,
11
+ "<|quad_start|>": 151650,
12
+ "<|video_pad|>": 151656,
13
+ "<|vision_end|>": 151653,
14
+ "<|vision_pad|>": 151654,
15
+ "<|vision_start|>": 151652
16
+ }
checkpoint-3125/chat_template.json ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ {
2
+ "chat_template": "{% set image_count = namespace(value=0) %}{% set video_count = namespace(value=0) %}{% for message in messages %}{% if loop.first and message['role'] != 'system' %}<|im_start|>system\nYou are a helpful assistant.<|im_end|>\n{% endif %}<|im_start|>{{ message['role'] }}\n{% if message['content'] is string %}{{ message['content'] }}<|im_end|>\n{% else %}{% for content in message['content'] %}{% if content['type'] == 'image' or 'image' in content or 'image_url' in content %}{% set image_count.value = image_count.value + 1 %}{% if add_vision_id %}Picture {{ image_count.value }}: {% endif %}<|vision_start|><|image_pad|><|vision_end|>{% elif content['type'] == 'video' or 'video' in content %}{% set video_count.value = video_count.value + 1 %}{% if add_vision_id %}Video {{ video_count.value }}: {% endif %}<|vision_start|><|video_pad|><|vision_end|>{% elif 'text' in content %}{{ content['text'] }}{% endif %}{% endfor %}<|im_end|>\n{% endif %}{% endfor %}{% if add_generation_prompt %}<|im_start|>assistant\n{% endif %}"
3
+ }
checkpoint-3125/merges.txt ADDED
The diff for this file is too large to render. See raw diff
 
checkpoint-3125/optimizer.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:7d80185f62b32382c1dab8164cb318abf02068a90e0f3d2b092228736d673a9a
3
+ size 232150634
checkpoint-3125/preprocessor_config.json ADDED
@@ -0,0 +1,29 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "do_convert_rgb": true,
3
+ "do_normalize": true,
4
+ "do_rescale": true,
5
+ "do_resize": true,
6
+ "image_mean": [
7
+ 0.48145466,
8
+ 0.4578275,
9
+ 0.40821073
10
+ ],
11
+ "image_processor_type": "Qwen2VLImageProcessor",
12
+ "image_std": [
13
+ 0.26862954,
14
+ 0.26130258,
15
+ 0.27577711
16
+ ],
17
+ "max_pixels": 12845056,
18
+ "merge_size": 2,
19
+ "min_pixels": 3136,
20
+ "patch_size": 14,
21
+ "processor_class": "Qwen2VLProcessor",
22
+ "resample": 3,
23
+ "rescale_factor": 0.00392156862745098,
24
+ "size": {
25
+ "max_pixels": 12845056,
26
+ "min_pixels": 3136
27
+ },
28
+ "temporal_patch_size": 2
29
+ }
checkpoint-3125/rng_state_0.pth ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:a456c014f4062602da085bb6b03f6bda6f86ab8b2dcf1a268dd903ed52f3fb91
3
+ size 14512
checkpoint-3125/rng_state_1.pth ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:74ea19755c03db3ccc062fd1a38208daf2c7258b6dd1d567b2257f1306869548
3
+ size 14512
checkpoint-3125/scheduler.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:9b69f9c3415ac518a3b27c7070592ae3338684506b156bdee0616aee17a8205b
3
+ size 1064
checkpoint-3125/special_tokens_map.json ADDED
@@ -0,0 +1,31 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "additional_special_tokens": [
3
+ "<|im_start|>",
4
+ "<|im_end|>",
5
+ "<|object_ref_start|>",
6
+ "<|object_ref_end|>",
7
+ "<|box_start|>",
8
+ "<|box_end|>",
9
+ "<|quad_start|>",
10
+ "<|quad_end|>",
11
+ "<|vision_start|>",
12
+ "<|vision_end|>",
13
+ "<|vision_pad|>",
14
+ "<|image_pad|>",
15
+ "<|video_pad|>"
16
+ ],
17
+ "eos_token": {
18
+ "content": "<|im_end|>",
19
+ "lstrip": false,
20
+ "normalized": false,
21
+ "rstrip": false,
22
+ "single_word": false
23
+ },
24
+ "pad_token": {
25
+ "content": "<|endoftext|>",
26
+ "lstrip": false,
27
+ "normalized": false,
28
+ "rstrip": false,
29
+ "single_word": false
30
+ }
31
+ }
checkpoint-3125/tokenizer.json ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:091aa7594dc2fcfbfa06b9e3c22a5f0562ac14f30375c13af7309407a0e67b8a
3
+ size 11420371
checkpoint-3125/tokenizer_config.json ADDED
@@ -0,0 +1,145 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "add_prefix_space": false,
3
+ "added_tokens_decoder": {
4
+ "151643": {
5
+ "content": "<|endoftext|>",
6
+ "lstrip": false,
7
+ "normalized": false,
8
+ "rstrip": false,
9
+ "single_word": false,
10
+ "special": true
11
+ },
12
+ "151644": {
13
+ "content": "<|im_start|>",
14
+ "lstrip": false,
15
+ "normalized": false,
16
+ "rstrip": false,
17
+ "single_word": false,
18
+ "special": true
19
+ },
20
+ "151645": {
21
+ "content": "<|im_end|>",
22
+ "lstrip": false,
23
+ "normalized": false,
24
+ "rstrip": false,
25
+ "single_word": false,
26
+ "special": true
27
+ },
28
+ "151646": {
29
+ "content": "<|object_ref_start|>",
30
+ "lstrip": false,
31
+ "normalized": false,
32
+ "rstrip": false,
33
+ "single_word": false,
34
+ "special": true
35
+ },
36
+ "151647": {
37
+ "content": "<|object_ref_end|>",
38
+ "lstrip": false,
39
+ "normalized": false,
40
+ "rstrip": false,
41
+ "single_word": false,
42
+ "special": true
43
+ },
44
+ "151648": {
45
+ "content": "<|box_start|>",
46
+ "lstrip": false,
47
+ "normalized": false,
48
+ "rstrip": false,
49
+ "single_word": false,
50
+ "special": true
51
+ },
52
+ "151649": {
53
+ "content": "<|box_end|>",
54
+ "lstrip": false,
55
+ "normalized": false,
56
+ "rstrip": false,
57
+ "single_word": false,
58
+ "special": true
59
+ },
60
+ "151650": {
61
+ "content": "<|quad_start|>",
62
+ "lstrip": false,
63
+ "normalized": false,
64
+ "rstrip": false,
65
+ "single_word": false,
66
+ "special": true
67
+ },
68
+ "151651": {
69
+ "content": "<|quad_end|>",
70
+ "lstrip": false,
71
+ "normalized": false,
72
+ "rstrip": false,
73
+ "single_word": false,
74
+ "special": true
75
+ },
76
+ "151652": {
77
+ "content": "<|vision_start|>",
78
+ "lstrip": false,
79
+ "normalized": false,
80
+ "rstrip": false,
81
+ "single_word": false,
82
+ "special": true
83
+ },
84
+ "151653": {
85
+ "content": "<|vision_end|>",
86
+ "lstrip": false,
87
+ "normalized": false,
88
+ "rstrip": false,
89
+ "single_word": false,
90
+ "special": true
91
+ },
92
+ "151654": {
93
+ "content": "<|vision_pad|>",
94
+ "lstrip": false,
95
+ "normalized": false,
96
+ "rstrip": false,
97
+ "single_word": false,
98
+ "special": true
99
+ },
100
+ "151655": {
101
+ "content": "<|image_pad|>",
102
+ "lstrip": false,
103
+ "normalized": false,
104
+ "rstrip": false,
105
+ "single_word": false,
106
+ "special": true
107
+ },
108
+ "151656": {
109
+ "content": "<|video_pad|>",
110
+ "lstrip": false,
111
+ "normalized": false,
112
+ "rstrip": false,
113
+ "single_word": false,
114
+ "special": true
115
+ }
116
+ },
117
+ "additional_special_tokens": [
118
+ "<|im_start|>",
119
+ "<|im_end|>",
120
+ "<|object_ref_start|>",
121
+ "<|object_ref_end|>",
122
+ "<|box_start|>",
123
+ "<|box_end|>",
124
+ "<|quad_start|>",
125
+ "<|quad_end|>",
126
+ "<|vision_start|>",
127
+ "<|vision_end|>",
128
+ "<|vision_pad|>",
129
+ "<|image_pad|>",
130
+ "<|video_pad|>"
131
+ ],
132
+ "bos_token": null,
133
+ "chat_template": "{% set image_count = namespace(value=0) %}{% set video_count = namespace(value=0) %}{% for message in messages %}{% if loop.first and message['role'] != 'system' %}<|im_start|>system\nYou are a helpful assistant.<|im_end|>\n{% endif %}<|im_start|>{{ message['role'] }}\n{% if message['content'] is string %}{{ message['content'] }}<|im_end|>\n{% else %}{% for content in message['content'] %}{% if content['type'] == 'image' or 'image' in content or 'image_url' in content %}{% set image_count.value = image_count.value + 1 %}{% if add_vision_id %}Picture {{ image_count.value }}: {% endif %}<|vision_start|><|image_pad|><|vision_end|>{% elif content['type'] == 'video' or 'video' in content %}{% set video_count.value = video_count.value + 1 %}{% if add_vision_id %}Video {{ video_count.value }}: {% endif %}<|vision_start|><|video_pad|><|vision_end|>{% elif 'text' in content %}{{ content['text'] }}{% endif %}{% endfor %}<|im_end|>\n{% endif %}{% endfor %}{% if add_generation_prompt %}<|im_start|>assistant\n{% endif %}",
134
+ "clean_up_tokenization_spaces": false,
135
+ "eos_token": "<|im_end|>",
136
+ "errors": "replace",
137
+ "extra_special_tokens": {},
138
+ "model_max_length": 2048,
139
+ "pad_token": "<|endoftext|>",
140
+ "padding_side": "right",
141
+ "processor_class": "Qwen2VLProcessor",
142
+ "split_special_tokens": false,
143
+ "tokenizer_class": "Qwen2Tokenizer",
144
+ "unk_token": null
145
+ }
checkpoint-3125/trainer_state.json ADDED
@@ -0,0 +1,2217 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "best_metric": null,
3
+ "best_model_checkpoint": null,
4
+ "epoch": 25.0,
5
+ "eval_steps": 500,
6
+ "global_step": 3125,
7
+ "is_hyper_param_search": false,
8
+ "is_local_process_zero": true,
9
+ "is_world_process_zero": true,
10
+ "log_history": [
11
+ {
12
+ "epoch": 0.08,
13
+ "grad_norm": 6.874109268188477,
14
+ "learning_rate": 2.5559105431309904e-07,
15
+ "loss": 1.5574,
16
+ "step": 10
17
+ },
18
+ {
19
+ "epoch": 0.16,
20
+ "grad_norm": 6.183030128479004,
21
+ "learning_rate": 8.306709265175719e-07,
22
+ "loss": 1.4724,
23
+ "step": 20
24
+ },
25
+ {
26
+ "epoch": 0.24,
27
+ "grad_norm": 2.6565823554992676,
28
+ "learning_rate": 1.4696485623003196e-06,
29
+ "loss": 1.054,
30
+ "step": 30
31
+ },
32
+ {
33
+ "epoch": 0.32,
34
+ "grad_norm": 3.227349042892456,
35
+ "learning_rate": 2.1086261980830672e-06,
36
+ "loss": 0.8732,
37
+ "step": 40
38
+ },
39
+ {
40
+ "epoch": 0.4,
41
+ "grad_norm": 1.1049450635910034,
42
+ "learning_rate": 2.747603833865815e-06,
43
+ "loss": 0.7617,
44
+ "step": 50
45
+ },
46
+ {
47
+ "epoch": 0.48,
48
+ "grad_norm": 1.8624663352966309,
49
+ "learning_rate": 3.386581469648563e-06,
50
+ "loss": 0.7144,
51
+ "step": 60
52
+ },
53
+ {
54
+ "epoch": 0.56,
55
+ "grad_norm": 2.1665029525756836,
56
+ "learning_rate": 4.02555910543131e-06,
57
+ "loss": 0.7612,
58
+ "step": 70
59
+ },
60
+ {
61
+ "epoch": 0.64,
62
+ "grad_norm": 1.0787376165390015,
63
+ "learning_rate": 4.664536741214058e-06,
64
+ "loss": 0.6864,
65
+ "step": 80
66
+ },
67
+ {
68
+ "epoch": 0.72,
69
+ "grad_norm": 1.4465893507003784,
70
+ "learning_rate": 5.303514376996806e-06,
71
+ "loss": 0.6558,
72
+ "step": 90
73
+ },
74
+ {
75
+ "epoch": 0.8,
76
+ "grad_norm": 1.1737278699874878,
77
+ "learning_rate": 5.8785942492012785e-06,
78
+ "loss": 0.6295,
79
+ "step": 100
80
+ },
81
+ {
82
+ "epoch": 0.88,
83
+ "grad_norm": 1.7027487754821777,
84
+ "learning_rate": 6.517571884984026e-06,
85
+ "loss": 0.638,
86
+ "step": 110
87
+ },
88
+ {
89
+ "epoch": 0.96,
90
+ "grad_norm": 1.5483449697494507,
91
+ "learning_rate": 7.156549520766773e-06,
92
+ "loss": 0.6173,
93
+ "step": 120
94
+ },
95
+ {
96
+ "epoch": 1.04,
97
+ "grad_norm": 1.9829121828079224,
98
+ "learning_rate": 7.795527156549521e-06,
99
+ "loss": 0.5511,
100
+ "step": 130
101
+ },
102
+ {
103
+ "epoch": 1.12,
104
+ "grad_norm": 1.466599464416504,
105
+ "learning_rate": 8.434504792332269e-06,
106
+ "loss": 0.5366,
107
+ "step": 140
108
+ },
109
+ {
110
+ "epoch": 1.2,
111
+ "grad_norm": 1.5559449195861816,
112
+ "learning_rate": 9.073482428115017e-06,
113
+ "loss": 0.5662,
114
+ "step": 150
115
+ },
116
+ {
117
+ "epoch": 1.28,
118
+ "grad_norm": 1.33512282371521,
119
+ "learning_rate": 9.712460063897765e-06,
120
+ "loss": 0.5567,
121
+ "step": 160
122
+ },
123
+ {
124
+ "epoch": 1.3599999999999999,
125
+ "grad_norm": 0.943450927734375,
126
+ "learning_rate": 1.035143769968051e-05,
127
+ "loss": 0.5997,
128
+ "step": 170
129
+ },
130
+ {
131
+ "epoch": 1.44,
132
+ "grad_norm": 1.2756850719451904,
133
+ "learning_rate": 1.099041533546326e-05,
134
+ "loss": 0.5587,
135
+ "step": 180
136
+ },
137
+ {
138
+ "epoch": 1.52,
139
+ "grad_norm": 1.0768604278564453,
140
+ "learning_rate": 1.1629392971246008e-05,
141
+ "loss": 0.6288,
142
+ "step": 190
143
+ },
144
+ {
145
+ "epoch": 1.6,
146
+ "grad_norm": 1.082158088684082,
147
+ "learning_rate": 1.2268370607028754e-05,
148
+ "loss": 0.5475,
149
+ "step": 200
150
+ },
151
+ {
152
+ "epoch": 1.6800000000000002,
153
+ "grad_norm": 1.4872550964355469,
154
+ "learning_rate": 1.2907348242811502e-05,
155
+ "loss": 0.6002,
156
+ "step": 210
157
+ },
158
+ {
159
+ "epoch": 1.76,
160
+ "grad_norm": 2.0925095081329346,
161
+ "learning_rate": 1.3546325878594251e-05,
162
+ "loss": 0.5887,
163
+ "step": 220
164
+ },
165
+ {
166
+ "epoch": 1.8399999999999999,
167
+ "grad_norm": 1.4571633338928223,
168
+ "learning_rate": 1.4185303514376998e-05,
169
+ "loss": 0.6107,
170
+ "step": 230
171
+ },
172
+ {
173
+ "epoch": 1.92,
174
+ "grad_norm": 2.4871673583984375,
175
+ "learning_rate": 1.4824281150159745e-05,
176
+ "loss": 0.5582,
177
+ "step": 240
178
+ },
179
+ {
180
+ "epoch": 2.0,
181
+ "grad_norm": 1.3184454441070557,
182
+ "learning_rate": 1.5463258785942495e-05,
183
+ "loss": 0.582,
184
+ "step": 250
185
+ },
186
+ {
187
+ "epoch": 2.08,
188
+ "grad_norm": 1.9747495651245117,
189
+ "learning_rate": 1.610223642172524e-05,
190
+ "loss": 0.4687,
191
+ "step": 260
192
+ },
193
+ {
194
+ "epoch": 2.16,
195
+ "grad_norm": 1.7691447734832764,
196
+ "learning_rate": 1.6741214057507987e-05,
197
+ "loss": 0.5013,
198
+ "step": 270
199
+ },
200
+ {
201
+ "epoch": 2.24,
202
+ "grad_norm": 1.8902829885482788,
203
+ "learning_rate": 1.7380191693290737e-05,
204
+ "loss": 0.5008,
205
+ "step": 280
206
+ },
207
+ {
208
+ "epoch": 2.32,
209
+ "grad_norm": 1.3197141885757446,
210
+ "learning_rate": 1.8019169329073486e-05,
211
+ "loss": 0.5229,
212
+ "step": 290
213
+ },
214
+ {
215
+ "epoch": 2.4,
216
+ "grad_norm": 2.325404167175293,
217
+ "learning_rate": 1.8658146964856232e-05,
218
+ "loss": 0.4975,
219
+ "step": 300
220
+ },
221
+ {
222
+ "epoch": 2.48,
223
+ "grad_norm": 1.3323551416397095,
224
+ "learning_rate": 1.929712460063898e-05,
225
+ "loss": 0.5254,
226
+ "step": 310
227
+ },
228
+ {
229
+ "epoch": 2.56,
230
+ "grad_norm": 0.7770269513130188,
231
+ "learning_rate": 1.9936102236421725e-05,
232
+ "loss": 0.5304,
233
+ "step": 320
234
+ },
235
+ {
236
+ "epoch": 2.64,
237
+ "grad_norm": 1.887294054031372,
238
+ "learning_rate": 1.999949450079496e-05,
239
+ "loss": 0.4755,
240
+ "step": 330
241
+ },
242
+ {
243
+ "epoch": 2.7199999999999998,
244
+ "grad_norm": 1.2255381345748901,
245
+ "learning_rate": 1.9997747161747696e-05,
246
+ "loss": 0.4982,
247
+ "step": 340
248
+ },
249
+ {
250
+ "epoch": 2.8,
251
+ "grad_norm": 1.6865767240524292,
252
+ "learning_rate": 1.9994751960168383e-05,
253
+ "loss": 0.4857,
254
+ "step": 350
255
+ },
256
+ {
257
+ "epoch": 2.88,
258
+ "grad_norm": 1.2997519969940186,
259
+ "learning_rate": 1.999050926990122e-05,
260
+ "loss": 0.4987,
261
+ "step": 360
262
+ },
263
+ {
264
+ "epoch": 2.96,
265
+ "grad_norm": 1.4636356830596924,
266
+ "learning_rate": 1.9985019620494935e-05,
267
+ "loss": 0.4947,
268
+ "step": 370
269
+ },
270
+ {
271
+ "epoch": 3.04,
272
+ "grad_norm": 1.3720135688781738,
273
+ "learning_rate": 1.9978283697136662e-05,
274
+ "loss": 0.5135,
275
+ "step": 380
276
+ },
277
+ {
278
+ "epoch": 3.12,
279
+ "grad_norm": 1.6108952760696411,
280
+ "learning_rate": 1.997030234056645e-05,
281
+ "loss": 0.4405,
282
+ "step": 390
283
+ },
284
+ {
285
+ "epoch": 3.2,
286
+ "grad_norm": 2.1790759563446045,
287
+ "learning_rate": 1.9961076546972304e-05,
288
+ "loss": 0.4468,
289
+ "step": 400
290
+ },
291
+ {
292
+ "epoch": 3.2800000000000002,
293
+ "grad_norm": 4.406925678253174,
294
+ "learning_rate": 1.9950607467865856e-05,
295
+ "loss": 0.4675,
296
+ "step": 410
297
+ },
298
+ {
299
+ "epoch": 3.36,
300
+ "grad_norm": 1.7346851825714111,
301
+ "learning_rate": 1.993889640993864e-05,
302
+ "loss": 0.4243,
303
+ "step": 420
304
+ },
305
+ {
306
+ "epoch": 3.44,
307
+ "grad_norm": 1.2865455150604248,
308
+ "learning_rate": 1.9925944834898992e-05,
309
+ "loss": 0.3787,
310
+ "step": 430
311
+ },
312
+ {
313
+ "epoch": 3.52,
314
+ "grad_norm": 1.321824550628662,
315
+ "learning_rate": 1.991175435928962e-05,
316
+ "loss": 0.3982,
317
+ "step": 440
318
+ },
319
+ {
320
+ "epoch": 3.6,
321
+ "grad_norm": 2.45711350440979,
322
+ "learning_rate": 1.989632675428582e-05,
323
+ "loss": 0.4811,
324
+ "step": 450
325
+ },
326
+ {
327
+ "epoch": 3.68,
328
+ "grad_norm": 2.5552310943603516,
329
+ "learning_rate": 1.9879663945474416e-05,
330
+ "loss": 0.4255,
331
+ "step": 460
332
+ },
333
+ {
334
+ "epoch": 3.76,
335
+ "grad_norm": 1.763244867324829,
336
+ "learning_rate": 1.9861768012613435e-05,
337
+ "loss": 0.4579,
338
+ "step": 470
339
+ },
340
+ {
341
+ "epoch": 3.84,
342
+ "grad_norm": 1.0951169729232788,
343
+ "learning_rate": 1.984264118937249e-05,
344
+ "loss": 0.398,
345
+ "step": 480
346
+ },
347
+ {
348
+ "epoch": 3.92,
349
+ "grad_norm": 8.235908508300781,
350
+ "learning_rate": 1.9822285863054e-05,
351
+ "loss": 0.4101,
352
+ "step": 490
353
+ },
354
+ {
355
+ "epoch": 4.0,
356
+ "grad_norm": 19.39371681213379,
357
+ "learning_rate": 1.9800704574295246e-05,
358
+ "loss": 0.4645,
359
+ "step": 500
360
+ },
361
+ {
362
+ "epoch": 4.08,
363
+ "grad_norm": 1.2594205141067505,
364
+ "learning_rate": 1.9777900016751224e-05,
365
+ "loss": 0.3632,
366
+ "step": 510
367
+ },
368
+ {
369
+ "epoch": 4.16,
370
+ "grad_norm": 1.7697833776474,
371
+ "learning_rate": 1.9753875036758464e-05,
372
+ "loss": 0.3356,
373
+ "step": 520
374
+ },
375
+ {
376
+ "epoch": 4.24,
377
+ "grad_norm": 1.66884446144104,
378
+ "learning_rate": 1.9728632632979746e-05,
379
+ "loss": 0.382,
380
+ "step": 530
381
+ },
382
+ {
383
+ "epoch": 4.32,
384
+ "grad_norm": 1.7648805379867554,
385
+ "learning_rate": 1.970217595602985e-05,
386
+ "loss": 0.3997,
387
+ "step": 540
388
+ },
389
+ {
390
+ "epoch": 4.4,
391
+ "grad_norm": 1.6132514476776123,
392
+ "learning_rate": 1.967450830808228e-05,
393
+ "loss": 0.3476,
394
+ "step": 550
395
+ },
396
+ {
397
+ "epoch": 4.48,
398
+ "grad_norm": 1.3217155933380127,
399
+ "learning_rate": 1.9645633142457143e-05,
400
+ "loss": 0.3362,
401
+ "step": 560
402
+ },
403
+ {
404
+ "epoch": 4.5600000000000005,
405
+ "grad_norm": 1.5668702125549316,
406
+ "learning_rate": 1.9615554063190098e-05,
407
+ "loss": 0.3996,
408
+ "step": 570
409
+ },
410
+ {
411
+ "epoch": 4.64,
412
+ "grad_norm": 2.051454544067383,
413
+ "learning_rate": 1.958427482458253e-05,
414
+ "loss": 0.4104,
415
+ "step": 580
416
+ },
417
+ {
418
+ "epoch": 4.72,
419
+ "grad_norm": 1.84868586063385,
420
+ "learning_rate": 1.9551799330732954e-05,
421
+ "loss": 0.374,
422
+ "step": 590
423
+ },
424
+ {
425
+ "epoch": 4.8,
426
+ "grad_norm": 1.3498121500015259,
427
+ "learning_rate": 1.9518131635049745e-05,
428
+ "loss": 0.3444,
429
+ "step": 600
430
+ },
431
+ {
432
+ "epoch": 4.88,
433
+ "grad_norm": 1.962553858757019,
434
+ "learning_rate": 1.9483275939745184e-05,
435
+ "loss": 0.3153,
436
+ "step": 610
437
+ },
438
+ {
439
+ "epoch": 4.96,
440
+ "grad_norm": 1.7132457494735718,
441
+ "learning_rate": 1.944723659531099e-05,
442
+ "loss": 0.383,
443
+ "step": 620
444
+ },
445
+ {
446
+ "epoch": 5.04,
447
+ "grad_norm": 1.8860396146774292,
448
+ "learning_rate": 1.9410018099975297e-05,
449
+ "loss": 0.327,
450
+ "step": 630
451
+ },
452
+ {
453
+ "epoch": 5.12,
454
+ "grad_norm": 1.9154820442199707,
455
+ "learning_rate": 1.9371625099141223e-05,
456
+ "loss": 0.2704,
457
+ "step": 640
458
+ },
459
+ {
460
+ "epoch": 5.2,
461
+ "grad_norm": 1.4077069759368896,
462
+ "learning_rate": 1.9332062384807058e-05,
463
+ "loss": 0.2659,
464
+ "step": 650
465
+ },
466
+ {
467
+ "epoch": 5.28,
468
+ "grad_norm": 1.8489642143249512,
469
+ "learning_rate": 1.9291334894968133e-05,
470
+ "loss": 0.2523,
471
+ "step": 660
472
+ },
473
+ {
474
+ "epoch": 5.36,
475
+ "grad_norm": 2.3917829990386963,
476
+ "learning_rate": 1.9249447713000515e-05,
477
+ "loss": 0.2885,
478
+ "step": 670
479
+ },
480
+ {
481
+ "epoch": 5.44,
482
+ "grad_norm": 2.5429599285125732,
483
+ "learning_rate": 1.9206406067026506e-05,
484
+ "loss": 0.3097,
485
+ "step": 680
486
+ },
487
+ {
488
+ "epoch": 5.52,
489
+ "grad_norm": 3.4684863090515137,
490
+ "learning_rate": 1.9162215329262115e-05,
491
+ "loss": 0.2868,
492
+ "step": 690
493
+ },
494
+ {
495
+ "epoch": 5.6,
496
+ "grad_norm": 2.662013292312622,
497
+ "learning_rate": 1.9116881015346517e-05,
498
+ "loss": 0.3281,
499
+ "step": 700
500
+ },
501
+ {
502
+ "epoch": 5.68,
503
+ "grad_norm": 2.223334312438965,
504
+ "learning_rate": 1.9070408783653627e-05,
505
+ "loss": 0.2799,
506
+ "step": 710
507
+ },
508
+ {
509
+ "epoch": 5.76,
510
+ "grad_norm": 2.992532730102539,
511
+ "learning_rate": 1.9022804434585854e-05,
512
+ "loss": 0.3252,
513
+ "step": 720
514
+ },
515
+ {
516
+ "epoch": 5.84,
517
+ "grad_norm": 1.6079481840133667,
518
+ "learning_rate": 1.8974073909850125e-05,
519
+ "loss": 0.2791,
520
+ "step": 730
521
+ },
522
+ {
523
+ "epoch": 5.92,
524
+ "grad_norm": 2.97104549407959,
525
+ "learning_rate": 1.8924223291716274e-05,
526
+ "loss": 0.3029,
527
+ "step": 740
528
+ },
529
+ {
530
+ "epoch": 6.0,
531
+ "grad_norm": 2.1889212131500244,
532
+ "learning_rate": 1.887325880225789e-05,
533
+ "loss": 0.293,
534
+ "step": 750
535
+ },
536
+ {
537
+ "epoch": 6.08,
538
+ "grad_norm": 2.123610019683838,
539
+ "learning_rate": 1.882118680257572e-05,
540
+ "loss": 0.2308,
541
+ "step": 760
542
+ },
543
+ {
544
+ "epoch": 6.16,
545
+ "grad_norm": 3.71563720703125,
546
+ "learning_rate": 1.8768013792003683e-05,
547
+ "loss": 0.216,
548
+ "step": 770
549
+ },
550
+ {
551
+ "epoch": 6.24,
552
+ "grad_norm": 2.459091901779175,
553
+ "learning_rate": 1.8713746407297703e-05,
554
+ "loss": 0.1972,
555
+ "step": 780
556
+ },
557
+ {
558
+ "epoch": 6.32,
559
+ "grad_norm": 2.669945240020752,
560
+ "learning_rate": 1.8658391421807313e-05,
561
+ "loss": 0.2478,
562
+ "step": 790
563
+ },
564
+ {
565
+ "epoch": 6.4,
566
+ "grad_norm": 2.1581931114196777,
567
+ "learning_rate": 1.8601955744630255e-05,
568
+ "loss": 0.2295,
569
+ "step": 800
570
+ },
571
+ {
572
+ "epoch": 6.48,
573
+ "grad_norm": 2.567908763885498,
574
+ "learning_rate": 1.8544446419750125e-05,
575
+ "loss": 0.2239,
576
+ "step": 810
577
+ },
578
+ {
579
+ "epoch": 6.5600000000000005,
580
+ "grad_norm": 2.0205740928649902,
581
+ "learning_rate": 1.8485870625157186e-05,
582
+ "loss": 0.216,
583
+ "step": 820
584
+ },
585
+ {
586
+ "epoch": 6.64,
587
+ "grad_norm": 2.2898030281066895,
588
+ "learning_rate": 1.8426235671952452e-05,
589
+ "loss": 0.2343,
590
+ "step": 830
591
+ },
592
+ {
593
+ "epoch": 6.72,
594
+ "grad_norm": 2.2003941535949707,
595
+ "learning_rate": 1.836554900343514e-05,
596
+ "loss": 0.2076,
597
+ "step": 840
598
+ },
599
+ {
600
+ "epoch": 6.8,
601
+ "grad_norm": 1.5117710828781128,
602
+ "learning_rate": 1.8303818194173665e-05,
603
+ "loss": 0.1904,
604
+ "step": 850
605
+ },
606
+ {
607
+ "epoch": 6.88,
608
+ "grad_norm": 1.2647337913513184,
609
+ "learning_rate": 1.824105094906021e-05,
610
+ "loss": 0.2341,
611
+ "step": 860
612
+ },
613
+ {
614
+ "epoch": 6.96,
615
+ "grad_norm": 2.00254225730896,
616
+ "learning_rate": 1.8177255102349047e-05,
617
+ "loss": 0.2288,
618
+ "step": 870
619
+ },
620
+ {
621
+ "epoch": 7.04,
622
+ "grad_norm": 2.516129493713379,
623
+ "learning_rate": 1.8112438616678712e-05,
624
+ "loss": 0.1685,
625
+ "step": 880
626
+ },
627
+ {
628
+ "epoch": 7.12,
629
+ "grad_norm": 3.6129150390625,
630
+ "learning_rate": 1.8046609582078147e-05,
631
+ "loss": 0.1625,
632
+ "step": 890
633
+ },
634
+ {
635
+ "epoch": 7.2,
636
+ "grad_norm": 3.1554768085479736,
637
+ "learning_rate": 1.797977621495696e-05,
638
+ "loss": 0.2011,
639
+ "step": 900
640
+ },
641
+ {
642
+ "epoch": 7.28,
643
+ "grad_norm": 3.8677282333374023,
644
+ "learning_rate": 1.7911946857079886e-05,
645
+ "loss": 0.1579,
646
+ "step": 910
647
+ },
648
+ {
649
+ "epoch": 7.36,
650
+ "grad_norm": 3.062568426132202,
651
+ "learning_rate": 1.784312997452562e-05,
652
+ "loss": 0.1705,
653
+ "step": 920
654
+ },
655
+ {
656
+ "epoch": 7.44,
657
+ "grad_norm": 1.8281478881835938,
658
+ "learning_rate": 1.777333415663014e-05,
659
+ "loss": 0.1708,
660
+ "step": 930
661
+ },
662
+ {
663
+ "epoch": 7.52,
664
+ "grad_norm": 1.1312201023101807,
665
+ "learning_rate": 1.7702568114914607e-05,
666
+ "loss": 0.1644,
667
+ "step": 940
668
+ },
669
+ {
670
+ "epoch": 7.6,
671
+ "grad_norm": 1.1482549905776978,
672
+ "learning_rate": 1.7630840681998068e-05,
673
+ "loss": 0.1297,
674
+ "step": 950
675
+ },
676
+ {
677
+ "epoch": 7.68,
678
+ "grad_norm": 3.109949827194214,
679
+ "learning_rate": 1.755816081049501e-05,
680
+ "loss": 0.1812,
681
+ "step": 960
682
+ },
683
+ {
684
+ "epoch": 7.76,
685
+ "grad_norm": 1.2507814168930054,
686
+ "learning_rate": 1.7484537571897943e-05,
687
+ "loss": 0.1331,
688
+ "step": 970
689
+ },
690
+ {
691
+ "epoch": 7.84,
692
+ "grad_norm": 2.898353338241577,
693
+ "learning_rate": 1.740998015544514e-05,
694
+ "loss": 0.1625,
695
+ "step": 980
696
+ },
697
+ {
698
+ "epoch": 7.92,
699
+ "grad_norm": 2.2018096446990967,
700
+ "learning_rate": 1.7334497866973716e-05,
701
+ "loss": 0.1441,
702
+ "step": 990
703
+ },
704
+ {
705
+ "epoch": 8.0,
706
+ "grad_norm": 2.1120376586914062,
707
+ "learning_rate": 1.725810012775808e-05,
708
+ "loss": 0.1804,
709
+ "step": 1000
710
+ },
711
+ {
712
+ "epoch": 8.08,
713
+ "grad_norm": 1.815804362297058,
714
+ "learning_rate": 1.7180796473334075e-05,
715
+ "loss": 0.1145,
716
+ "step": 1010
717
+ },
718
+ {
719
+ "epoch": 8.16,
720
+ "grad_norm": 2.5671639442443848,
721
+ "learning_rate": 1.7102596552308765e-05,
722
+ "loss": 0.1094,
723
+ "step": 1020
724
+ },
725
+ {
726
+ "epoch": 8.24,
727
+ "grad_norm": 0.9062528610229492,
728
+ "learning_rate": 1.7023510125156173e-05,
729
+ "loss": 0.0924,
730
+ "step": 1030
731
+ },
732
+ {
733
+ "epoch": 8.32,
734
+ "grad_norm": 3.5460164546966553,
735
+ "learning_rate": 1.6943547062999027e-05,
736
+ "loss": 0.1323,
737
+ "step": 1040
738
+ },
739
+ {
740
+ "epoch": 8.4,
741
+ "grad_norm": 2.856273889541626,
742
+ "learning_rate": 1.6862717346376706e-05,
743
+ "loss": 0.0938,
744
+ "step": 1050
745
+ },
746
+ {
747
+ "epoch": 8.48,
748
+ "grad_norm": 3.177924633026123,
749
+ "learning_rate": 1.6781031063999515e-05,
750
+ "loss": 0.1204,
751
+ "step": 1060
752
+ },
753
+ {
754
+ "epoch": 8.56,
755
+ "grad_norm": 1.3371851444244385,
756
+ "learning_rate": 1.6698498411489477e-05,
757
+ "loss": 0.1342,
758
+ "step": 1070
759
+ },
760
+ {
761
+ "epoch": 8.64,
762
+ "grad_norm": 2.0833916664123535,
763
+ "learning_rate": 1.6615129690107773e-05,
764
+ "loss": 0.1337,
765
+ "step": 1080
766
+ },
767
+ {
768
+ "epoch": 8.72,
769
+ "grad_norm": 2.4857378005981445,
770
+ "learning_rate": 1.6530935305469e-05,
771
+ "loss": 0.1233,
772
+ "step": 1090
773
+ },
774
+ {
775
+ "epoch": 8.8,
776
+ "grad_norm": 1.7040643692016602,
777
+ "learning_rate": 1.6445925766242392e-05,
778
+ "loss": 0.121,
779
+ "step": 1100
780
+ },
781
+ {
782
+ "epoch": 8.88,
783
+ "grad_norm": 1.2290680408477783,
784
+ "learning_rate": 1.6360111682840184e-05,
785
+ "loss": 0.1037,
786
+ "step": 1110
787
+ },
788
+ {
789
+ "epoch": 8.96,
790
+ "grad_norm": 3.9770867824554443,
791
+ "learning_rate": 1.62735037660933e-05,
792
+ "loss": 0.1133,
793
+ "step": 1120
794
+ },
795
+ {
796
+ "epoch": 9.04,
797
+ "grad_norm": 1.0533943176269531,
798
+ "learning_rate": 1.618611282591446e-05,
799
+ "loss": 0.0943,
800
+ "step": 1130
801
+ },
802
+ {
803
+ "epoch": 9.12,
804
+ "grad_norm": 1.9520882368087769,
805
+ "learning_rate": 1.609794976994897e-05,
806
+ "loss": 0.0725,
807
+ "step": 1140
808
+ },
809
+ {
810
+ "epoch": 9.2,
811
+ "grad_norm": 3.610280990600586,
812
+ "learning_rate": 1.600902560221329e-05,
813
+ "loss": 0.1015,
814
+ "step": 1150
815
+ },
816
+ {
817
+ "epoch": 9.28,
818
+ "grad_norm": 1.3682546615600586,
819
+ "learning_rate": 1.5919351421721548e-05,
820
+ "loss": 0.0945,
821
+ "step": 1160
822
+ },
823
+ {
824
+ "epoch": 9.36,
825
+ "grad_norm": 2.125917673110962,
826
+ "learning_rate": 1.5828938421100266e-05,
827
+ "loss": 0.0733,
828
+ "step": 1170
829
+ },
830
+ {
831
+ "epoch": 9.44,
832
+ "grad_norm": 2.5206658840179443,
833
+ "learning_rate": 1.5737797885191316e-05,
834
+ "loss": 0.0646,
835
+ "step": 1180
836
+ },
837
+ {
838
+ "epoch": 9.52,
839
+ "grad_norm": 2.5717523097991943,
840
+ "learning_rate": 1.5645941189643444e-05,
841
+ "loss": 0.0836,
842
+ "step": 1190
843
+ },
844
+ {
845
+ "epoch": 9.6,
846
+ "grad_norm": 2.1724965572357178,
847
+ "learning_rate": 1.55533797994924e-05,
848
+ "loss": 0.0933,
849
+ "step": 1200
850
+ },
851
+ {
852
+ "epoch": 9.68,
853
+ "grad_norm": 3.0643951892852783,
854
+ "learning_rate": 1.546012526772996e-05,
855
+ "loss": 0.083,
856
+ "step": 1210
857
+ },
858
+ {
859
+ "epoch": 9.76,
860
+ "grad_norm": 1.3585115671157837,
861
+ "learning_rate": 1.5366189233861933e-05,
862
+ "loss": 0.0804,
863
+ "step": 1220
864
+ },
865
+ {
866
+ "epoch": 9.84,
867
+ "grad_norm": 2.106403112411499,
868
+ "learning_rate": 1.5271583422455373e-05,
869
+ "loss": 0.0941,
870
+ "step": 1230
871
+ },
872
+ {
873
+ "epoch": 9.92,
874
+ "grad_norm": 1.7128456830978394,
875
+ "learning_rate": 1.5176319641675213e-05,
876
+ "loss": 0.0668,
877
+ "step": 1240
878
+ },
879
+ {
880
+ "epoch": 10.0,
881
+ "grad_norm": 1.7037241458892822,
882
+ "learning_rate": 1.5080409781810406e-05,
883
+ "loss": 0.1043,
884
+ "step": 1250
885
+ },
886
+ {
887
+ "epoch": 10.08,
888
+ "grad_norm": 0.8989495038986206,
889
+ "learning_rate": 1.4983865813789869e-05,
890
+ "loss": 0.0521,
891
+ "step": 1260
892
+ },
893
+ {
894
+ "epoch": 10.16,
895
+ "grad_norm": 1.3013043403625488,
896
+ "learning_rate": 1.488669978768833e-05,
897
+ "loss": 0.0647,
898
+ "step": 1270
899
+ },
900
+ {
901
+ "epoch": 10.24,
902
+ "grad_norm": 0.6377755999565125,
903
+ "learning_rate": 1.47889238312223e-05,
904
+ "loss": 0.0367,
905
+ "step": 1280
906
+ },
907
+ {
908
+ "epoch": 10.32,
909
+ "grad_norm": 2.2679171562194824,
910
+ "learning_rate": 1.4690550148236371e-05,
911
+ "loss": 0.0769,
912
+ "step": 1290
913
+ },
914
+ {
915
+ "epoch": 10.4,
916
+ "grad_norm": 2.3292899131774902,
917
+ "learning_rate": 1.4591591017179993e-05,
918
+ "loss": 0.0908,
919
+ "step": 1300
920
+ },
921
+ {
922
+ "epoch": 10.48,
923
+ "grad_norm": 1.8322817087173462,
924
+ "learning_rate": 1.4492058789574948e-05,
925
+ "loss": 0.0551,
926
+ "step": 1310
927
+ },
928
+ {
929
+ "epoch": 10.56,
930
+ "grad_norm": 5.285430431365967,
931
+ "learning_rate": 1.4391965888473705e-05,
932
+ "loss": 0.0584,
933
+ "step": 1320
934
+ },
935
+ {
936
+ "epoch": 10.64,
937
+ "grad_norm": 0.768731415271759,
938
+ "learning_rate": 1.4291324806908846e-05,
939
+ "loss": 0.0833,
940
+ "step": 1330
941
+ },
942
+ {
943
+ "epoch": 10.72,
944
+ "grad_norm": 1.1357346773147583,
945
+ "learning_rate": 1.419014810633374e-05,
946
+ "loss": 0.0596,
947
+ "step": 1340
948
+ },
949
+ {
950
+ "epoch": 10.8,
951
+ "grad_norm": 0.9390424489974976,
952
+ "learning_rate": 1.408844841505473e-05,
953
+ "loss": 0.0436,
954
+ "step": 1350
955
+ },
956
+ {
957
+ "epoch": 10.88,
958
+ "grad_norm": 0.775092363357544,
959
+ "learning_rate": 1.3986238426654894e-05,
960
+ "loss": 0.0663,
961
+ "step": 1360
962
+ },
963
+ {
964
+ "epoch": 10.96,
965
+ "grad_norm": 1.6158931255340576,
966
+ "learning_rate": 1.3883530898409736e-05,
967
+ "loss": 0.0438,
968
+ "step": 1370
969
+ },
970
+ {
971
+ "epoch": 11.04,
972
+ "grad_norm": 1.8279812335968018,
973
+ "learning_rate": 1.3780338649694874e-05,
974
+ "loss": 0.0944,
975
+ "step": 1380
976
+ },
977
+ {
978
+ "epoch": 11.12,
979
+ "grad_norm": 1.8661407232284546,
980
+ "learning_rate": 1.3676674560386018e-05,
981
+ "loss": 0.0398,
982
+ "step": 1390
983
+ },
984
+ {
985
+ "epoch": 11.2,
986
+ "grad_norm": 1.8592077493667603,
987
+ "learning_rate": 1.357255156925136e-05,
988
+ "loss": 0.0561,
989
+ "step": 1400
990
+ },
991
+ {
992
+ "epoch": 11.28,
993
+ "grad_norm": 3.479532241821289,
994
+ "learning_rate": 1.3467982672336633e-05,
995
+ "loss": 0.0408,
996
+ "step": 1410
997
+ },
998
+ {
999
+ "epoch": 11.36,
1000
+ "grad_norm": 2.4556691646575928,
1001
+ "learning_rate": 1.336298092134302e-05,
1002
+ "loss": 0.0603,
1003
+ "step": 1420
1004
+ },
1005
+ {
1006
+ "epoch": 11.44,
1007
+ "grad_norm": 2.028334379196167,
1008
+ "learning_rate": 1.325755942199812e-05,
1009
+ "loss": 0.0423,
1010
+ "step": 1430
1011
+ },
1012
+ {
1013
+ "epoch": 11.52,
1014
+ "grad_norm": 0.6319141387939453,
1015
+ "learning_rate": 1.3151731332420152e-05,
1016
+ "loss": 0.0717,
1017
+ "step": 1440
1018
+ },
1019
+ {
1020
+ "epoch": 11.6,
1021
+ "grad_norm": 1.3395280838012695,
1022
+ "learning_rate": 1.3045509861475645e-05,
1023
+ "loss": 0.0505,
1024
+ "step": 1450
1025
+ },
1026
+ {
1027
+ "epoch": 11.68,
1028
+ "grad_norm": 2.619269847869873,
1029
+ "learning_rate": 1.293890826713077e-05,
1030
+ "loss": 0.0477,
1031
+ "step": 1460
1032
+ },
1033
+ {
1034
+ "epoch": 11.76,
1035
+ "grad_norm": 2.6248464584350586,
1036
+ "learning_rate": 1.2831939854796567e-05,
1037
+ "loss": 0.0426,
1038
+ "step": 1470
1039
+ },
1040
+ {
1041
+ "epoch": 11.84,
1042
+ "grad_norm": 1.672989845275879,
1043
+ "learning_rate": 1.2724617975668229e-05,
1044
+ "loss": 0.0533,
1045
+ "step": 1480
1046
+ },
1047
+ {
1048
+ "epoch": 11.92,
1049
+ "grad_norm": 5.669078350067139,
1050
+ "learning_rate": 1.2616956025058688e-05,
1051
+ "loss": 0.0413,
1052
+ "step": 1490
1053
+ },
1054
+ {
1055
+ "epoch": 12.0,
1056
+ "grad_norm": 2.987652063369751,
1057
+ "learning_rate": 1.2508967440726689e-05,
1058
+ "loss": 0.0442,
1059
+ "step": 1500
1060
+ },
1061
+ {
1062
+ "epoch": 12.08,
1063
+ "grad_norm": 1.2909622192382812,
1064
+ "learning_rate": 1.2400665701199541e-05,
1065
+ "loss": 0.0565,
1066
+ "step": 1510
1067
+ },
1068
+ {
1069
+ "epoch": 12.16,
1070
+ "grad_norm": 0.264878511428833,
1071
+ "learning_rate": 1.2292064324090842e-05,
1072
+ "loss": 0.0243,
1073
+ "step": 1520
1074
+ },
1075
+ {
1076
+ "epoch": 12.24,
1077
+ "grad_norm": 2.0698373317718506,
1078
+ "learning_rate": 1.2183176864413262e-05,
1079
+ "loss": 0.0241,
1080
+ "step": 1530
1081
+ },
1082
+ {
1083
+ "epoch": 12.32,
1084
+ "grad_norm": 0.20062156021595,
1085
+ "learning_rate": 1.2074016912886683e-05,
1086
+ "loss": 0.0204,
1087
+ "step": 1540
1088
+ },
1089
+ {
1090
+ "epoch": 12.4,
1091
+ "grad_norm": 0.5076002478599548,
1092
+ "learning_rate": 1.1964598094241884e-05,
1093
+ "loss": 0.0462,
1094
+ "step": 1550
1095
+ },
1096
+ {
1097
+ "epoch": 12.48,
1098
+ "grad_norm": 1.1509472131729126,
1099
+ "learning_rate": 1.1854934065519986e-05,
1100
+ "loss": 0.0302,
1101
+ "step": 1560
1102
+ },
1103
+ {
1104
+ "epoch": 12.56,
1105
+ "grad_norm": 1.603930950164795,
1106
+ "learning_rate": 1.1745038514367853e-05,
1107
+ "loss": 0.0282,
1108
+ "step": 1570
1109
+ },
1110
+ {
1111
+ "epoch": 12.64,
1112
+ "grad_norm": 0.9757488965988159,
1113
+ "learning_rate": 1.1634925157329668e-05,
1114
+ "loss": 0.0387,
1115
+ "step": 1580
1116
+ },
1117
+ {
1118
+ "epoch": 12.72,
1119
+ "grad_norm": 1.058209776878357,
1120
+ "learning_rate": 1.1524607738134928e-05,
1121
+ "loss": 0.0445,
1122
+ "step": 1590
1123
+ },
1124
+ {
1125
+ "epoch": 12.8,
1126
+ "grad_norm": 2.3291449546813965,
1127
+ "learning_rate": 1.141410002598301e-05,
1128
+ "loss": 0.0572,
1129
+ "step": 1600
1130
+ },
1131
+ {
1132
+ "epoch": 12.88,
1133
+ "grad_norm": 0.8108656406402588,
1134
+ "learning_rate": 1.1303415813824599e-05,
1135
+ "loss": 0.0337,
1136
+ "step": 1610
1137
+ },
1138
+ {
1139
+ "epoch": 12.96,
1140
+ "grad_norm": 6.551490783691406,
1141
+ "learning_rate": 1.1192568916640107e-05,
1142
+ "loss": 0.0337,
1143
+ "step": 1620
1144
+ },
1145
+ {
1146
+ "epoch": 13.04,
1147
+ "grad_norm": 0.2524939477443695,
1148
+ "learning_rate": 1.1081573169715379e-05,
1149
+ "loss": 0.0218,
1150
+ "step": 1630
1151
+ },
1152
+ {
1153
+ "epoch": 13.12,
1154
+ "grad_norm": 0.17596031725406647,
1155
+ "learning_rate": 1.0970442426914847e-05,
1156
+ "loss": 0.0132,
1157
+ "step": 1640
1158
+ },
1159
+ {
1160
+ "epoch": 13.2,
1161
+ "grad_norm": 0.7680679559707642,
1162
+ "learning_rate": 1.085919055895237e-05,
1163
+ "loss": 0.0207,
1164
+ "step": 1650
1165
+ },
1166
+ {
1167
+ "epoch": 13.28,
1168
+ "grad_norm": 0.24066166579723358,
1169
+ "learning_rate": 1.0747831451659967e-05,
1170
+ "loss": 0.0142,
1171
+ "step": 1660
1172
+ },
1173
+ {
1174
+ "epoch": 13.36,
1175
+ "grad_norm": 3.3667285442352295,
1176
+ "learning_rate": 1.0636379004254665e-05,
1177
+ "loss": 0.0183,
1178
+ "step": 1670
1179
+ },
1180
+ {
1181
+ "epoch": 13.44,
1182
+ "grad_norm": 1.8831534385681152,
1183
+ "learning_rate": 1.0524847127603677e-05,
1184
+ "loss": 0.0253,
1185
+ "step": 1680
1186
+ },
1187
+ {
1188
+ "epoch": 13.52,
1189
+ "grad_norm": 1.329085111618042,
1190
+ "learning_rate": 1.0413249742488132e-05,
1191
+ "loss": 0.0118,
1192
+ "step": 1690
1193
+ },
1194
+ {
1195
+ "epoch": 13.6,
1196
+ "grad_norm": 2.0776255130767822,
1197
+ "learning_rate": 1.030160077786556e-05,
1198
+ "loss": 0.0128,
1199
+ "step": 1700
1200
+ },
1201
+ {
1202
+ "epoch": 13.68,
1203
+ "grad_norm": 1.02082097530365,
1204
+ "learning_rate": 1.0189914169131341e-05,
1205
+ "loss": 0.0406,
1206
+ "step": 1710
1207
+ },
1208
+ {
1209
+ "epoch": 13.76,
1210
+ "grad_norm": 0.7947062253952026,
1211
+ "learning_rate": 1.0078203856379394e-05,
1212
+ "loss": 0.0183,
1213
+ "step": 1720
1214
+ },
1215
+ {
1216
+ "epoch": 13.84,
1217
+ "grad_norm": 0.3098564147949219,
1218
+ "learning_rate": 9.966483782662233e-06,
1219
+ "loss": 0.0251,
1220
+ "step": 1730
1221
+ },
1222
+ {
1223
+ "epoch": 13.92,
1224
+ "grad_norm": 0.4250534772872925,
1225
+ "learning_rate": 9.854767892250692e-06,
1226
+ "loss": 0.0394,
1227
+ "step": 1740
1228
+ },
1229
+ {
1230
+ "epoch": 14.0,
1231
+ "grad_norm": 1.0927317142486572,
1232
+ "learning_rate": 9.743070128893452e-06,
1233
+ "loss": 0.0187,
1234
+ "step": 1750
1235
+ },
1236
+ {
1237
+ "epoch": 14.08,
1238
+ "grad_norm": 0.5906019806861877,
1239
+ "learning_rate": 9.631404434076687e-06,
1240
+ "loss": 0.0042,
1241
+ "step": 1760
1242
+ },
1243
+ {
1244
+ "epoch": 14.16,
1245
+ "grad_norm": 0.4497601389884949,
1246
+ "learning_rate": 9.519784745283956e-06,
1247
+ "loss": 0.0402,
1248
+ "step": 1770
1249
+ },
1250
+ {
1251
+ "epoch": 14.24,
1252
+ "grad_norm": 0.2257859855890274,
1253
+ "learning_rate": 9.408224994256603e-06,
1254
+ "loss": 0.0039,
1255
+ "step": 1780
1256
+ },
1257
+ {
1258
+ "epoch": 14.32,
1259
+ "grad_norm": 0.9509989023208618,
1260
+ "learning_rate": 9.296739105254869e-06,
1261
+ "loss": 0.0156,
1262
+ "step": 1790
1263
+ },
1264
+ {
1265
+ "epoch": 14.4,
1266
+ "grad_norm": 1.7652183771133423,
1267
+ "learning_rate": 9.185340993319977e-06,
1268
+ "loss": 0.0121,
1269
+ "step": 1800
1270
+ },
1271
+ {
1272
+ "epoch": 14.48,
1273
+ "grad_norm": 8.208316802978516,
1274
+ "learning_rate": 9.074044562537284e-06,
1275
+ "loss": 0.0389,
1276
+ "step": 1810
1277
+ },
1278
+ {
1279
+ "epoch": 14.56,
1280
+ "grad_norm": 0.13252846896648407,
1281
+ "learning_rate": 8.962863704300893e-06,
1282
+ "loss": 0.011,
1283
+ "step": 1820
1284
+ },
1285
+ {
1286
+ "epoch": 14.64,
1287
+ "grad_norm": 0.23998035490512848,
1288
+ "learning_rate": 8.851812295579789e-06,
1289
+ "loss": 0.0248,
1290
+ "step": 1830
1291
+ },
1292
+ {
1293
+ "epoch": 14.72,
1294
+ "grad_norm": 0.10990134626626968,
1295
+ "learning_rate": 8.740904197185794e-06,
1296
+ "loss": 0.0204,
1297
+ "step": 1840
1298
+ },
1299
+ {
1300
+ "epoch": 14.8,
1301
+ "grad_norm": 3.542644500732422,
1302
+ "learning_rate": 8.630153252043543e-06,
1303
+ "loss": 0.0163,
1304
+ "step": 1850
1305
+ },
1306
+ {
1307
+ "epoch": 14.88,
1308
+ "grad_norm": 0.7273716926574707,
1309
+ "learning_rate": 8.519573283462688e-06,
1310
+ "loss": 0.0068,
1311
+ "step": 1860
1312
+ },
1313
+ {
1314
+ "epoch": 14.96,
1315
+ "grad_norm": 0.036577966064214706,
1316
+ "learning_rate": 8.409178093412549e-06,
1317
+ "loss": 0.0084,
1318
+ "step": 1870
1319
+ },
1320
+ {
1321
+ "epoch": 15.04,
1322
+ "grad_norm": 0.1642957627773285,
1323
+ "learning_rate": 8.298981460799426e-06,
1324
+ "loss": 0.0093,
1325
+ "step": 1880
1326
+ },
1327
+ {
1328
+ "epoch": 15.12,
1329
+ "grad_norm": 0.11017989367246628,
1330
+ "learning_rate": 8.188997139746807e-06,
1331
+ "loss": 0.0076,
1332
+ "step": 1890
1333
+ },
1334
+ {
1335
+ "epoch": 15.2,
1336
+ "grad_norm": 0.04475900158286095,
1337
+ "learning_rate": 8.079238857878631e-06,
1338
+ "loss": 0.0059,
1339
+ "step": 1900
1340
+ },
1341
+ {
1342
+ "epoch": 15.28,
1343
+ "grad_norm": 1.0872098207473755,
1344
+ "learning_rate": 7.969720314605915e-06,
1345
+ "loss": 0.0051,
1346
+ "step": 1910
1347
+ },
1348
+ {
1349
+ "epoch": 15.36,
1350
+ "grad_norm": 0.32729968428611755,
1351
+ "learning_rate": 7.860455179416837e-06,
1352
+ "loss": 0.0016,
1353
+ "step": 1920
1354
+ },
1355
+ {
1356
+ "epoch": 15.44,
1357
+ "grad_norm": 1.3896279335021973,
1358
+ "learning_rate": 7.751457090170616e-06,
1359
+ "loss": 0.0066,
1360
+ "step": 1930
1361
+ },
1362
+ {
1363
+ "epoch": 15.52,
1364
+ "grad_norm": 0.24193909764289856,
1365
+ "learning_rate": 7.642739651395295e-06,
1366
+ "loss": 0.0049,
1367
+ "step": 1940
1368
+ },
1369
+ {
1370
+ "epoch": 15.6,
1371
+ "grad_norm": 0.0498543456196785,
1372
+ "learning_rate": 7.534316432589706e-06,
1373
+ "loss": 0.0065,
1374
+ "step": 1950
1375
+ },
1376
+ {
1377
+ "epoch": 15.68,
1378
+ "grad_norm": 0.053042277693748474,
1379
+ "learning_rate": 7.426200966529795e-06,
1380
+ "loss": 0.0096,
1381
+ "step": 1960
1382
+ },
1383
+ {
1384
+ "epoch": 15.76,
1385
+ "grad_norm": 0.6754854917526245,
1386
+ "learning_rate": 7.318406747579556e-06,
1387
+ "loss": 0.0074,
1388
+ "step": 1970
1389
+ },
1390
+ {
1391
+ "epoch": 15.84,
1392
+ "grad_norm": 0.05438900366425514,
1393
+ "learning_rate": 7.210947230006713e-06,
1394
+ "loss": 0.0025,
1395
+ "step": 1980
1396
+ },
1397
+ {
1398
+ "epoch": 15.92,
1399
+ "grad_norm": 0.0401092991232872,
1400
+ "learning_rate": 7.103835826303451e-06,
1401
+ "loss": 0.0022,
1402
+ "step": 1990
1403
+ },
1404
+ {
1405
+ "epoch": 16.0,
1406
+ "grad_norm": 0.05556059256196022,
1407
+ "learning_rate": 6.997085905512346e-06,
1408
+ "loss": 0.0049,
1409
+ "step": 2000
1410
+ },
1411
+ {
1412
+ "epoch": 16.08,
1413
+ "grad_norm": 0.07605039328336716,
1414
+ "learning_rate": 6.8907107915577075e-06,
1415
+ "loss": 0.0051,
1416
+ "step": 2010
1417
+ },
1418
+ {
1419
+ "epoch": 16.16,
1420
+ "grad_norm": 0.07472195476293564,
1421
+ "learning_rate": 6.7847237615825636e-06,
1422
+ "loss": 0.0031,
1423
+ "step": 2020
1424
+ },
1425
+ {
1426
+ "epoch": 16.24,
1427
+ "grad_norm": 0.203807532787323,
1428
+ "learning_rate": 6.6791380442914866e-06,
1429
+ "loss": 0.0014,
1430
+ "step": 2030
1431
+ },
1432
+ {
1433
+ "epoch": 16.32,
1434
+ "grad_norm": 0.08696573972702026,
1435
+ "learning_rate": 6.573966818299461e-06,
1436
+ "loss": 0.0024,
1437
+ "step": 2040
1438
+ },
1439
+ {
1440
+ "epoch": 16.4,
1441
+ "grad_norm": 0.025404011830687523,
1442
+ "learning_rate": 6.469223210486992e-06,
1443
+ "loss": 0.0007,
1444
+ "step": 2050
1445
+ },
1446
+ {
1447
+ "epoch": 16.48,
1448
+ "grad_norm": 0.05262196436524391,
1449
+ "learning_rate": 6.364920294361701e-06,
1450
+ "loss": 0.0006,
1451
+ "step": 2060
1452
+ },
1453
+ {
1454
+ "epoch": 16.56,
1455
+ "grad_norm": 0.029066545888781548,
1456
+ "learning_rate": 6.261071088426546e-06,
1457
+ "loss": 0.0004,
1458
+ "step": 2070
1459
+ },
1460
+ {
1461
+ "epoch": 16.64,
1462
+ "grad_norm": 0.02618943341076374,
1463
+ "learning_rate": 6.1576885545549355e-06,
1464
+ "loss": 0.0022,
1465
+ "step": 2080
1466
+ },
1467
+ {
1468
+ "epoch": 16.72,
1469
+ "grad_norm": 0.06179804354906082,
1470
+ "learning_rate": 6.054785596372894e-06,
1471
+ "loss": 0.0005,
1472
+ "step": 2090
1473
+ },
1474
+ {
1475
+ "epoch": 16.8,
1476
+ "grad_norm": 0.014377947896718979,
1477
+ "learning_rate": 5.952375057648509e-06,
1478
+ "loss": 0.0028,
1479
+ "step": 2100
1480
+ },
1481
+ {
1482
+ "epoch": 16.88,
1483
+ "grad_norm": 0.039646223187446594,
1484
+ "learning_rate": 5.850469720688847e-06,
1485
+ "loss": 0.001,
1486
+ "step": 2110
1487
+ },
1488
+ {
1489
+ "epoch": 16.96,
1490
+ "grad_norm": 0.44483599066734314,
1491
+ "learning_rate": 5.74908230474453e-06,
1492
+ "loss": 0.0044,
1493
+ "step": 2120
1494
+ },
1495
+ {
1496
+ "epoch": 17.04,
1497
+ "grad_norm": 0.024487623944878578,
1498
+ "learning_rate": 5.648225464422189e-06,
1499
+ "loss": 0.0007,
1500
+ "step": 2130
1501
+ },
1502
+ {
1503
+ "epoch": 17.12,
1504
+ "grad_norm": 0.0326782651245594,
1505
+ "learning_rate": 5.547911788105001e-06,
1506
+ "loss": 0.0008,
1507
+ "step": 2140
1508
+ },
1509
+ {
1510
+ "epoch": 17.2,
1511
+ "grad_norm": 0.015867168083786964,
1512
+ "learning_rate": 5.4481537963814675e-06,
1513
+ "loss": 0.0004,
1514
+ "step": 2150
1515
+ },
1516
+ {
1517
+ "epoch": 17.28,
1518
+ "grad_norm": 0.022902317345142365,
1519
+ "learning_rate": 5.348963940482663e-06,
1520
+ "loss": 0.0003,
1521
+ "step": 2160
1522
+ },
1523
+ {
1524
+ "epoch": 17.36,
1525
+ "grad_norm": 0.024737635627388954,
1526
+ "learning_rate": 5.25035460072814e-06,
1527
+ "loss": 0.0023,
1528
+ "step": 2170
1529
+ },
1530
+ {
1531
+ "epoch": 17.44,
1532
+ "grad_norm": 0.019413290545344353,
1533
+ "learning_rate": 5.15233808498071e-06,
1534
+ "loss": 0.0004,
1535
+ "step": 2180
1536
+ },
1537
+ {
1538
+ "epoch": 17.52,
1539
+ "grad_norm": 0.01702185347676277,
1540
+ "learning_rate": 5.054926627110208e-06,
1541
+ "loss": 0.0004,
1542
+ "step": 2190
1543
+ },
1544
+ {
1545
+ "epoch": 17.6,
1546
+ "grad_norm": 2.464230537414551,
1547
+ "learning_rate": 4.9581323854665695e-06,
1548
+ "loss": 0.0049,
1549
+ "step": 2200
1550
+ },
1551
+ {
1552
+ "epoch": 17.68,
1553
+ "grad_norm": 0.013668897561728954,
1554
+ "learning_rate": 4.861967441362262e-06,
1555
+ "loss": 0.0004,
1556
+ "step": 2210
1557
+ },
1558
+ {
1559
+ "epoch": 17.76,
1560
+ "grad_norm": 0.014037170447409153,
1561
+ "learning_rate": 4.766443797564375e-06,
1562
+ "loss": 0.0004,
1563
+ "step": 2220
1564
+ },
1565
+ {
1566
+ "epoch": 17.84,
1567
+ "grad_norm": 0.011271145194768906,
1568
+ "learning_rate": 4.671573376796511e-06,
1569
+ "loss": 0.0005,
1570
+ "step": 2230
1571
+ },
1572
+ {
1573
+ "epoch": 17.92,
1574
+ "grad_norm": 0.015100213699042797,
1575
+ "learning_rate": 4.57736802025065e-06,
1576
+ "loss": 0.0007,
1577
+ "step": 2240
1578
+ },
1579
+ {
1580
+ "epoch": 18.0,
1581
+ "grad_norm": 0.1259368360042572,
1582
+ "learning_rate": 4.48383948610919e-06,
1583
+ "loss": 0.0003,
1584
+ "step": 2250
1585
+ },
1586
+ {
1587
+ "epoch": 18.08,
1588
+ "grad_norm": 0.04630790278315544,
1589
+ "learning_rate": 4.390999448077375e-06,
1590
+ "loss": 0.0004,
1591
+ "step": 2260
1592
+ },
1593
+ {
1594
+ "epoch": 18.16,
1595
+ "grad_norm": 0.03695525601506233,
1596
+ "learning_rate": 4.298859493926235e-06,
1597
+ "loss": 0.0003,
1598
+ "step": 2270
1599
+ },
1600
+ {
1601
+ "epoch": 18.24,
1602
+ "grad_norm": 0.00454062782227993,
1603
+ "learning_rate": 4.207431124046267e-06,
1604
+ "loss": 0.0003,
1605
+ "step": 2280
1606
+ },
1607
+ {
1608
+ "epoch": 18.32,
1609
+ "grad_norm": 0.016318323090672493,
1610
+ "learning_rate": 4.116725750012035e-06,
1611
+ "loss": 0.0003,
1612
+ "step": 2290
1613
+ },
1614
+ {
1615
+ "epoch": 18.4,
1616
+ "grad_norm": 0.02545909956097603,
1617
+ "learning_rate": 4.026754693157816e-06,
1618
+ "loss": 0.0003,
1619
+ "step": 2300
1620
+ },
1621
+ {
1622
+ "epoch": 18.48,
1623
+ "grad_norm": 0.014473868533968925,
1624
+ "learning_rate": 3.937529183164562e-06,
1625
+ "loss": 0.0003,
1626
+ "step": 2310
1627
+ },
1628
+ {
1629
+ "epoch": 18.56,
1630
+ "grad_norm": 0.014056684449315071,
1631
+ "learning_rate": 3.84906035665826e-06,
1632
+ "loss": 0.0002,
1633
+ "step": 2320
1634
+ },
1635
+ {
1636
+ "epoch": 18.64,
1637
+ "grad_norm": 0.013438849709928036,
1638
+ "learning_rate": 3.7613592558199162e-06,
1639
+ "loss": 0.0002,
1640
+ "step": 2330
1641
+ },
1642
+ {
1643
+ "epoch": 18.72,
1644
+ "grad_norm": 0.01617852970957756,
1645
+ "learning_rate": 3.6744368270073393e-06,
1646
+ "loss": 0.0007,
1647
+ "step": 2340
1648
+ },
1649
+ {
1650
+ "epoch": 18.8,
1651
+ "grad_norm": 0.013655665330588818,
1652
+ "learning_rate": 3.5883039193888914e-06,
1653
+ "loss": 0.0007,
1654
+ "step": 2350
1655
+ },
1656
+ {
1657
+ "epoch": 18.88,
1658
+ "grad_norm": 0.014764860272407532,
1659
+ "learning_rate": 3.502971283589326e-06,
1660
+ "loss": 0.0005,
1661
+ "step": 2360
1662
+ },
1663
+ {
1664
+ "epoch": 18.96,
1665
+ "grad_norm": 0.03937000408768654,
1666
+ "learning_rate": 3.418449570347986e-06,
1667
+ "loss": 0.0003,
1668
+ "step": 2370
1669
+ },
1670
+ {
1671
+ "epoch": 19.04,
1672
+ "grad_norm": 0.014079142361879349,
1673
+ "learning_rate": 3.334749329189415e-06,
1674
+ "loss": 0.0003,
1675
+ "step": 2380
1676
+ },
1677
+ {
1678
+ "epoch": 19.12,
1679
+ "grad_norm": 0.016212107613682747,
1680
+ "learning_rate": 3.2518810071066363e-06,
1681
+ "loss": 0.0002,
1682
+ "step": 2390
1683
+ },
1684
+ {
1685
+ "epoch": 19.2,
1686
+ "grad_norm": 0.015338711440563202,
1687
+ "learning_rate": 3.1698549472572203e-06,
1688
+ "loss": 0.0003,
1689
+ "step": 2400
1690
+ },
1691
+ {
1692
+ "epoch": 19.28,
1693
+ "grad_norm": 0.02485400065779686,
1694
+ "learning_rate": 3.0886813876723075e-06,
1695
+ "loss": 0.0002,
1696
+ "step": 2410
1697
+ },
1698
+ {
1699
+ "epoch": 19.36,
1700
+ "grad_norm": 0.017927484586834908,
1701
+ "learning_rate": 3.0083704599787423e-06,
1702
+ "loss": 0.0003,
1703
+ "step": 2420
1704
+ },
1705
+ {
1706
+ "epoch": 19.44,
1707
+ "grad_norm": 0.012980809435248375,
1708
+ "learning_rate": 2.9289321881345257e-06,
1709
+ "loss": 0.0005,
1710
+ "step": 2430
1711
+ },
1712
+ {
1713
+ "epoch": 19.52,
1714
+ "grad_norm": 0.03192094713449478,
1715
+ "learning_rate": 2.850376487177656e-06,
1716
+ "loss": 0.0002,
1717
+ "step": 2440
1718
+ },
1719
+ {
1720
+ "epoch": 19.6,
1721
+ "grad_norm": 0.019058849662542343,
1722
+ "learning_rate": 2.7727131619886017e-06,
1723
+ "loss": 0.0002,
1724
+ "step": 2450
1725
+ },
1726
+ {
1727
+ "epoch": 19.68,
1728
+ "grad_norm": 0.012099268846213818,
1729
+ "learning_rate": 2.6959519060665195e-06,
1730
+ "loss": 0.0002,
1731
+ "step": 2460
1732
+ },
1733
+ {
1734
+ "epoch": 19.76,
1735
+ "grad_norm": 0.010341337881982327,
1736
+ "learning_rate": 2.6201023003193437e-06,
1737
+ "loss": 0.0002,
1738
+ "step": 2470
1739
+ },
1740
+ {
1741
+ "epoch": 19.84,
1742
+ "grad_norm": 0.014265513978898525,
1743
+ "learning_rate": 2.545173811867977e-06,
1744
+ "loss": 0.0002,
1745
+ "step": 2480
1746
+ },
1747
+ {
1748
+ "epoch": 19.92,
1749
+ "grad_norm": 0.01814926601946354,
1750
+ "learning_rate": 2.471175792864642e-06,
1751
+ "loss": 0.001,
1752
+ "step": 2490
1753
+ },
1754
+ {
1755
+ "epoch": 20.0,
1756
+ "grad_norm": 0.011018522083759308,
1757
+ "learning_rate": 2.3981174793255956e-06,
1758
+ "loss": 0.0002,
1759
+ "step": 2500
1760
+ },
1761
+ {
1762
+ "epoch": 20.08,
1763
+ "grad_norm": 0.015308992005884647,
1764
+ "learning_rate": 2.3260079899783492e-06,
1765
+ "loss": 0.0008,
1766
+ "step": 2510
1767
+ },
1768
+ {
1769
+ "epoch": 20.16,
1770
+ "grad_norm": 0.010280388407409191,
1771
+ "learning_rate": 2.254856325123529e-06,
1772
+ "loss": 0.0002,
1773
+ "step": 2520
1774
+ },
1775
+ {
1776
+ "epoch": 20.24,
1777
+ "grad_norm": 0.011187535710632801,
1778
+ "learning_rate": 2.1846713655114836e-06,
1779
+ "loss": 0.0002,
1780
+ "step": 2530
1781
+ },
1782
+ {
1783
+ "epoch": 20.32,
1784
+ "grad_norm": 0.015644947066903114,
1785
+ "learning_rate": 2.115461871233867e-06,
1786
+ "loss": 0.0002,
1787
+ "step": 2540
1788
+ },
1789
+ {
1790
+ "epoch": 20.4,
1791
+ "grad_norm": 0.00875458400696516,
1792
+ "learning_rate": 2.0472364806302313e-06,
1793
+ "loss": 0.0002,
1794
+ "step": 2550
1795
+ },
1796
+ {
1797
+ "epoch": 20.48,
1798
+ "grad_norm": 0.017235323786735535,
1799
+ "learning_rate": 1.9800037092098477e-06,
1800
+ "loss": 0.0002,
1801
+ "step": 2560
1802
+ },
1803
+ {
1804
+ "epoch": 20.56,
1805
+ "grad_norm": 0.03400786966085434,
1806
+ "learning_rate": 1.9137719485888527e-06,
1807
+ "loss": 0.0002,
1808
+ "step": 2570
1809
+ },
1810
+ {
1811
+ "epoch": 20.64,
1812
+ "grad_norm": 0.011186143383383751,
1813
+ "learning_rate": 1.8485494654428482e-06,
1814
+ "loss": 0.0002,
1815
+ "step": 2580
1816
+ },
1817
+ {
1818
+ "epoch": 20.72,
1819
+ "grad_norm": 0.008517293259501457,
1820
+ "learning_rate": 1.784344400475093e-06,
1821
+ "loss": 0.0002,
1822
+ "step": 2590
1823
+ },
1824
+ {
1825
+ "epoch": 20.8,
1826
+ "grad_norm": 0.010979237034916878,
1827
+ "learning_rate": 1.7211647674004483e-06,
1828
+ "loss": 0.0002,
1829
+ "step": 2600
1830
+ },
1831
+ {
1832
+ "epoch": 20.88,
1833
+ "grad_norm": 0.01790277473628521,
1834
+ "learning_rate": 1.659018451945128e-06,
1835
+ "loss": 0.0002,
1836
+ "step": 2610
1837
+ },
1838
+ {
1839
+ "epoch": 20.96,
1840
+ "grad_norm": 0.01020762138068676,
1841
+ "learning_rate": 1.5979132108624572e-06,
1842
+ "loss": 0.0002,
1843
+ "step": 2620
1844
+ },
1845
+ {
1846
+ "epoch": 21.04,
1847
+ "grad_norm": 0.014280925504863262,
1848
+ "learning_rate": 1.5378566709647225e-06,
1849
+ "loss": 0.0003,
1850
+ "step": 2630
1851
+ },
1852
+ {
1853
+ "epoch": 21.12,
1854
+ "grad_norm": 0.012542898766696453,
1855
+ "learning_rate": 1.4788563281712253e-06,
1856
+ "loss": 0.0002,
1857
+ "step": 2640
1858
+ },
1859
+ {
1860
+ "epoch": 21.2,
1861
+ "grad_norm": 0.0072167362086474895,
1862
+ "learning_rate": 1.420919546572691e-06,
1863
+ "loss": 0.0002,
1864
+ "step": 2650
1865
+ },
1866
+ {
1867
+ "epoch": 21.28,
1868
+ "grad_norm": 0.010835397057235241,
1869
+ "learning_rate": 1.364053557512126e-06,
1870
+ "loss": 0.0002,
1871
+ "step": 2660
1872
+ },
1873
+ {
1874
+ "epoch": 21.36,
1875
+ "grad_norm": 0.010230830870568752,
1876
+ "learning_rate": 1.308265458682234e-06,
1877
+ "loss": 0.0004,
1878
+ "step": 2670
1879
+ },
1880
+ {
1881
+ "epoch": 21.44,
1882
+ "grad_norm": 0.00813596136868,
1883
+ "learning_rate": 1.2535622132395242e-06,
1884
+ "loss": 0.0005,
1885
+ "step": 2680
1886
+ },
1887
+ {
1888
+ "epoch": 21.52,
1889
+ "grad_norm": 0.02058684267103672,
1890
+ "learning_rate": 1.1999506489352208e-06,
1891
+ "loss": 0.0002,
1892
+ "step": 2690
1893
+ },
1894
+ {
1895
+ "epoch": 21.6,
1896
+ "grad_norm": 0.022641103714704514,
1897
+ "learning_rate": 1.1474374572630432e-06,
1898
+ "loss": 0.0002,
1899
+ "step": 2700
1900
+ },
1901
+ {
1902
+ "epoch": 21.68,
1903
+ "grad_norm": 0.015892324969172478,
1904
+ "learning_rate": 1.0960291926240263e-06,
1905
+ "loss": 0.0002,
1906
+ "step": 2710
1907
+ },
1908
+ {
1909
+ "epoch": 21.76,
1910
+ "grad_norm": 0.01219907309859991,
1911
+ "learning_rate": 1.0457322715084305e-06,
1912
+ "loss": 0.0002,
1913
+ "step": 2720
1914
+ },
1915
+ {
1916
+ "epoch": 21.84,
1917
+ "grad_norm": 0.009200166910886765,
1918
+ "learning_rate": 9.965529716948684e-07,
1919
+ "loss": 0.0002,
1920
+ "step": 2730
1921
+ },
1922
+ {
1923
+ "epoch": 21.92,
1924
+ "grad_norm": 0.00976441241800785,
1925
+ "learning_rate": 9.484974314667561e-07,
1926
+ "loss": 0.0002,
1927
+ "step": 2740
1928
+ },
1929
+ {
1930
+ "epoch": 22.0,
1931
+ "grad_norm": 0.010954627767205238,
1932
+ "learning_rate": 9.015716488461656e-07,
1933
+ "loss": 0.0002,
1934
+ "step": 2750
1935
+ },
1936
+ {
1937
+ "epoch": 22.08,
1938
+ "grad_norm": 0.016307951882481575,
1939
+ "learning_rate": 8.557814808451737e-07,
1940
+ "loss": 0.0002,
1941
+ "step": 2760
1942
+ },
1943
+ {
1944
+ "epoch": 22.16,
1945
+ "grad_norm": 0.009793213568627834,
1946
+ "learning_rate": 8.11132642734841e-07,
1947
+ "loss": 0.0001,
1948
+ "step": 2770
1949
+ },
1950
+ {
1951
+ "epoch": 22.24,
1952
+ "grad_norm": 0.023596106097102165,
1953
+ "learning_rate": 7.676307073318479e-07,
1954
+ "loss": 0.0002,
1955
+ "step": 2780
1956
+ },
1957
+ {
1958
+ "epoch": 22.32,
1959
+ "grad_norm": 0.011904028244316578,
1960
+ "learning_rate": 7.252811043029373e-07,
1961
+ "loss": 0.0004,
1962
+ "step": 2790
1963
+ },
1964
+ {
1965
+ "epoch": 22.4,
1966
+ "grad_norm": 0.013149023056030273,
1967
+ "learning_rate": 6.840891194872112e-07,
1968
+ "loss": 0.0002,
1969
+ "step": 2800
1970
+ },
1971
+ {
1972
+ "epoch": 22.48,
1973
+ "grad_norm": 0.011410288512706757,
1974
+ "learning_rate": 6.440598942363796e-07,
1975
+ "loss": 0.0002,
1976
+ "step": 2810
1977
+ },
1978
+ {
1979
+ "epoch": 22.56,
1980
+ "grad_norm": 0.008377713151276112,
1981
+ "learning_rate": 6.051984247730447e-07,
1982
+ "loss": 0.0002,
1983
+ "step": 2820
1984
+ },
1985
+ {
1986
+ "epoch": 22.64,
1987
+ "grad_norm": 0.19204267859458923,
1988
+ "learning_rate": 5.675095615671144e-07,
1989
+ "loss": 0.0005,
1990
+ "step": 2830
1991
+ },
1992
+ {
1993
+ "epoch": 22.72,
1994
+ "grad_norm": 0.017059409990906715,
1995
+ "learning_rate": 5.309980087303713e-07,
1996
+ "loss": 0.0002,
1997
+ "step": 2840
1998
+ },
1999
+ {
2000
+ "epoch": 22.8,
2001
+ "grad_norm": 0.00994216650724411,
2002
+ "learning_rate": 4.956683234293491e-07,
2003
+ "loss": 0.0002,
2004
+ "step": 2850
2005
+ },
2006
+ {
2007
+ "epoch": 22.88,
2008
+ "grad_norm": 0.017344018444418907,
2009
+ "learning_rate": 4.61524915316528e-07,
2010
+ "loss": 0.0001,
2011
+ "step": 2860
2012
+ },
2013
+ {
2014
+ "epoch": 22.96,
2015
+ "grad_norm": 0.009385120123624802,
2016
+ "learning_rate": 4.285720459799425e-07,
2017
+ "loss": 0.0002,
2018
+ "step": 2870
2019
+ },
2020
+ {
2021
+ "epoch": 23.04,
2022
+ "grad_norm": 0.009774768725037575,
2023
+ "learning_rate": 3.9681382841128323e-07,
2024
+ "loss": 0.0002,
2025
+ "step": 2880
2026
+ },
2027
+ {
2028
+ "epoch": 23.12,
2029
+ "grad_norm": 0.13636986911296844,
2030
+ "learning_rate": 3.6625422649252617e-07,
2031
+ "loss": 0.0005,
2032
+ "step": 2890
2033
+ },
2034
+ {
2035
+ "epoch": 23.2,
2036
+ "grad_norm": 0.011067167855799198,
2037
+ "learning_rate": 3.368970545011874e-07,
2038
+ "loss": 0.0002,
2039
+ "step": 2900
2040
+ },
2041
+ {
2042
+ "epoch": 23.28,
2043
+ "grad_norm": 0.008906451985239983,
2044
+ "learning_rate": 3.0874597663425046e-07,
2045
+ "loss": 0.0002,
2046
+ "step": 2910
2047
+ },
2048
+ {
2049
+ "epoch": 23.36,
2050
+ "grad_norm": 0.011505813337862492,
2051
+ "learning_rate": 2.818045065508168e-07,
2052
+ "loss": 0.0001,
2053
+ "step": 2920
2054
+ },
2055
+ {
2056
+ "epoch": 23.44,
2057
+ "grad_norm": 0.01766866073012352,
2058
+ "learning_rate": 2.560760069335511e-07,
2059
+ "loss": 0.0002,
2060
+ "step": 2930
2061
+ },
2062
+ {
2063
+ "epoch": 23.52,
2064
+ "grad_norm": 0.01433554571121931,
2065
+ "learning_rate": 2.315636890689743e-07,
2066
+ "loss": 0.0002,
2067
+ "step": 2940
2068
+ },
2069
+ {
2070
+ "epoch": 23.6,
2071
+ "grad_norm": 0.009151219390332699,
2072
+ "learning_rate": 2.082706124466416e-07,
2073
+ "loss": 0.0002,
2074
+ "step": 2950
2075
+ },
2076
+ {
2077
+ "epoch": 23.68,
2078
+ "grad_norm": 0.014051095582544804,
2079
+ "learning_rate": 1.8619968437727954e-07,
2080
+ "loss": 0.0002,
2081
+ "step": 2960
2082
+ },
2083
+ {
2084
+ "epoch": 23.76,
2085
+ "grad_norm": 0.01479571033269167,
2086
+ "learning_rate": 1.6535365962991322e-07,
2087
+ "loss": 0.0002,
2088
+ "step": 2970
2089
+ },
2090
+ {
2091
+ "epoch": 23.84,
2092
+ "grad_norm": 0.17555338144302368,
2093
+ "learning_rate": 1.4573514008802693e-07,
2094
+ "loss": 0.0005,
2095
+ "step": 2980
2096
+ },
2097
+ {
2098
+ "epoch": 23.92,
2099
+ "grad_norm": 0.01707664132118225,
2100
+ "learning_rate": 1.2734657442481368e-07,
2101
+ "loss": 0.0002,
2102
+ "step": 2990
2103
+ },
2104
+ {
2105
+ "epoch": 24.0,
2106
+ "grad_norm": 0.023542573675513268,
2107
+ "learning_rate": 1.1019025779754666e-07,
2108
+ "loss": 0.0002,
2109
+ "step": 3000
2110
+ },
2111
+ {
2112
+ "epoch": 24.08,
2113
+ "grad_norm": 0.008962544612586498,
2114
+ "learning_rate": 9.426833156111038e-08,
2115
+ "loss": 0.0002,
2116
+ "step": 3010
2117
+ },
2118
+ {
2119
+ "epoch": 24.16,
2120
+ "grad_norm": 0.009664681740105152,
2121
+ "learning_rate": 7.958278300072453e-08,
2122
+ "loss": 0.0002,
2123
+ "step": 3020
2124
+ },
2125
+ {
2126
+ "epoch": 24.24,
2127
+ "grad_norm": 0.0093710171058774,
2128
+ "learning_rate": 6.613544508391024e-08,
2129
+ "loss": 0.0001,
2130
+ "step": 3030
2131
+ },
2132
+ {
2133
+ "epoch": 24.32,
2134
+ "grad_norm": 0.016148734837770462,
2135
+ "learning_rate": 5.392799623170186e-08,
2136
+ "loss": 0.0002,
2137
+ "step": 3040
2138
+ },
2139
+ {
2140
+ "epoch": 24.4,
2141
+ "grad_norm": 0.018474169075489044,
2142
+ "learning_rate": 4.2961960109160205e-08,
2143
+ "loss": 0.0002,
2144
+ "step": 3050
2145
+ },
2146
+ {
2147
+ "epoch": 24.48,
2148
+ "grad_norm": 0.007969476282596588,
2149
+ "learning_rate": 3.323870543519458e-08,
2150
+ "loss": 0.0001,
2151
+ "step": 3060
2152
+ },
2153
+ {
2154
+ "epoch": 24.56,
2155
+ "grad_norm": 0.009026006795465946,
2156
+ "learning_rate": 2.475944581173173e-08,
2157
+ "loss": 0.0002,
2158
+ "step": 3070
2159
+ },
2160
+ {
2161
+ "epoch": 24.64,
2162
+ "grad_norm": 0.016195567324757576,
2163
+ "learning_rate": 1.752523957223362e-08,
2164
+ "loss": 0.0002,
2165
+ "step": 3080
2166
+ },
2167
+ {
2168
+ "epoch": 24.72,
2169
+ "grad_norm": 0.004977029282599688,
2170
+ "learning_rate": 1.1536989649608699e-08,
2171
+ "loss": 0.0002,
2172
+ "step": 3090
2173
+ },
2174
+ {
2175
+ "epoch": 24.8,
2176
+ "grad_norm": 0.008723889477550983,
2177
+ "learning_rate": 6.795443463509799e-09,
2178
+ "loss": 0.0002,
2179
+ "step": 3100
2180
+ },
2181
+ {
2182
+ "epoch": 24.88,
2183
+ "grad_norm": 0.01094700489193201,
2184
+ "learning_rate": 3.3011928270454406e-09,
2185
+ "loss": 0.0004,
2186
+ "step": 3110
2187
+ },
2188
+ {
2189
+ "epoch": 24.96,
2190
+ "grad_norm": 0.009475067257881165,
2191
+ "learning_rate": 1.054673872914469e-09,
2192
+ "loss": 0.0005,
2193
+ "step": 3120
2194
+ }
2195
+ ],
2196
+ "logging_steps": 10,
2197
+ "max_steps": 3125,
2198
+ "num_input_tokens_seen": 0,
2199
+ "num_train_epochs": 25,
2200
+ "save_steps": 1000,
2201
+ "stateful_callbacks": {
2202
+ "TrainerControl": {
2203
+ "args": {
2204
+ "should_epoch_stop": false,
2205
+ "should_evaluate": false,
2206
+ "should_log": false,
2207
+ "should_save": true,
2208
+ "should_training_stop": true
2209
+ },
2210
+ "attributes": {}
2211
+ }
2212
+ },
2213
+ "total_flos": 1.246578685771776e+17,
2214
+ "train_batch_size": 2,
2215
+ "trial_name": null,
2216
+ "trial_params": null
2217
+ }
checkpoint-3125/training_args.bin ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:27ff69d62d45c0f05227ab8bfdfb70f5de878e29a775a5568fe676d60d677a54
3
+ size 5624
checkpoint-3125/vocab.json ADDED
The diff for this file is too large to render. See raw diff
 
runs/Feb03_13-32-36_120d765cf84c/events.out.tfevents.1738589701.120d765cf84c.147.0 ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:35ffeed6dc698c81e2bd120d8f2ff3b857e5ef0107f92145f59a7f1b1d2b04db
3
+ size 72496
train_results.json CHANGED
@@ -1,8 +1,8 @@
1
  {
2
- "epoch": 40.0,
3
- "total_flos": 2.004550530755461e+17,
4
- "train_loss": 0.09495920253196216,
5
- "train_runtime": 28025.0795,
6
- "train_samples_per_second": 0.714,
7
- "train_steps_per_second": 0.12
8
  }
 
1
  {
2
+ "epoch": 25.0,
3
+ "total_flos": 1.246578685771776e+17,
4
+ "train_loss": 0.15220735648881645,
5
+ "train_runtime": 31927.3271,
6
+ "train_samples_per_second": 0.392,
7
+ "train_steps_per_second": 0.098
8
  }
trainer_log.jsonl CHANGED
The diff for this file is too large to render. See raw diff
 
trainer_state.json CHANGED
The diff for this file is too large to render. See raw diff
 
training_args.bin CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:fe16ad85cf67a1d616e8a664749a043a7e42845bd6dfe8881e3e3c947d23497a
3
  size 5624
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:27ff69d62d45c0f05227ab8bfdfb70f5de878e29a775a5568fe676d60d677a54
3
  size 5624