chansung commited on
Commit
4345b3d
1 Parent(s): 56218e1

Model save

Browse files
README.md CHANGED
@@ -2,13 +2,12 @@
2
  license: gemma
3
  library_name: peft
4
  tags:
5
- - alignment-handbook
6
  - trl
7
  - sft
8
  - generated_from_trainer
9
  base_model: google/gemma-7b
10
  datasets:
11
- - llama-duo/synth_summarize_dataset
12
  model-index:
13
  - name: gemma7b-summarize-claude3sonnet-30k
14
  results: []
@@ -17,12 +16,12 @@ model-index:
17
  <!-- This model card has been generated automatically according to the information the Trainer had access to. You
18
  should probably proofread and complete it, then remove this comment. -->
19
 
20
- [<img src="https://raw.githubusercontent.com/wandb/assets/main/wandb-github-badge-28.svg" alt="Visualize in Weights & Biases" width="200" height="32"/>](https://wandb.ai/chansung18/huggingface/runs/62awpxu7)
21
  # gemma7b-summarize-claude3sonnet-30k
22
 
23
- This model is a fine-tuned version of [google/gemma-7b](https://huggingface.co/google/gemma-7b) on the llama-duo/synth_summarize_dataset dataset.
24
  It achieves the following results on the evaluation set:
25
- - Loss: 2.3268
26
 
27
  ## Model description
28
 
@@ -53,17 +52,22 @@ The following hyperparameters were used during training:
53
  - optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
54
  - lr_scheduler_type: cosine
55
  - lr_scheduler_warmup_ratio: 0.1
56
- - num_epochs: 5
57
 
58
  ### Training results
59
 
60
  | Training Loss | Epoch | Step | Validation Loss |
61
  |:-------------:|:-----:|:----:|:---------------:|
62
- | 0.7892 | 1.0 | 205 | 2.2406 |
63
- | 0.7033 | 2.0 | 410 | 2.2004 |
64
- | 0.6346 | 3.0 | 615 | 2.2458 |
65
- | 0.5755 | 4.0 | 820 | 2.2935 |
66
- | 0.5144 | 5.0 | 1025 | 2.3268 |
 
 
 
 
 
67
 
68
 
69
  ### Framework versions
 
2
  license: gemma
3
  library_name: peft
4
  tags:
 
5
  - trl
6
  - sft
7
  - generated_from_trainer
8
  base_model: google/gemma-7b
9
  datasets:
10
+ - generator
11
  model-index:
12
  - name: gemma7b-summarize-claude3sonnet-30k
13
  results: []
 
16
  <!-- This model card has been generated automatically according to the information the Trainer had access to. You
17
  should probably proofread and complete it, then remove this comment. -->
18
 
19
+ [<img src="https://raw.githubusercontent.com/wandb/assets/main/wandb-github-badge-28.svg" alt="Visualize in Weights & Biases" width="200" height="32"/>](https://wandb.ai/chansung18/huggingface/runs/7bdtvabz)
20
  # gemma7b-summarize-claude3sonnet-30k
21
 
22
+ This model is a fine-tuned version of [google/gemma-7b](https://huggingface.co/google/gemma-7b) on the generator dataset.
23
  It achieves the following results on the evaluation set:
24
+ - Loss: 3.0044
25
 
26
  ## Model description
27
 
 
52
  - optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
53
  - lr_scheduler_type: cosine
54
  - lr_scheduler_warmup_ratio: 0.1
55
+ - num_epochs: 10
56
 
57
  ### Training results
58
 
59
  | Training Loss | Epoch | Step | Validation Loss |
60
  |:-------------:|:-----:|:----:|:---------------:|
61
+ | 1.0072 | 1.0 | 148 | 2.2672 |
62
+ | 0.8705 | 2.0 | 296 | 2.1745 |
63
+ | 0.7957 | 3.0 | 444 | 2.1914 |
64
+ | 0.731 | 4.0 | 592 | 2.2511 |
65
+ | 0.634 | 5.0 | 740 | 2.3409 |
66
+ | 0.5418 | 6.0 | 888 | 2.4841 |
67
+ | 0.4578 | 7.0 | 1036 | 2.6822 |
68
+ | 0.3886 | 8.0 | 1184 | 2.8497 |
69
+ | 0.3611 | 9.0 | 1332 | 2.9868 |
70
+ | 0.3501 | 10.0 | 1480 | 3.0044 |
71
 
72
 
73
  ### Framework versions
adapter_model.safetensors CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:dad991653bac09b52d163fbec1c43bda26259e79bdb54f9e5540d668e943a8ce
3
  size 50056096
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:f4986c382bd11a9859e7e4b4bfb564bd3755dd1372fc11f9cff8bc07726c7ba1
3
  size 50056096
all_results.json CHANGED
@@ -1,14 +1,9 @@
1
  {
2
- "epoch": 5.0,
3
- "eval_loss": 2.326845407485962,
4
- "eval_runtime": 1.0399,
5
- "eval_samples": 25,
6
- "eval_samples_per_second": 4.808,
7
- "eval_steps_per_second": 1.923,
8
- "total_flos": 1.5670950020754964e+18,
9
- "train_loss": 1.4066738275202308,
10
- "train_runtime": 8016.4558,
11
- "train_samples": 29489,
12
- "train_samples_per_second": 2.044,
13
- "train_steps_per_second": 0.128
14
  }
 
1
  {
2
+ "epoch": 10.0,
3
+ "total_flos": 2.2627322959180595e+18,
4
+ "train_loss": 1.3636517344294368,
5
+ "train_runtime": 11668.1199,
6
+ "train_samples": 30867,
7
+ "train_samples_per_second": 2.027,
8
+ "train_steps_per_second": 0.127
 
 
 
 
 
9
  }
runs/May21_20-36-36_deep-diver-main-dangerous-sloth-1-0-0/events.out.tfevents.1716338340.deep-diver-main-dangerous-sloth-1-0-0.385.0 CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:d1450c93681b04a98350cd594a0475c27bbdae9bdb1a03e5d851f39a3489820f
3
- size 67253
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:212b78a85df7517fb282cb76388b9079333746497e930e83d10ef11f9c70718a
3
+ size 71254
train_results.json CHANGED
@@ -1,9 +1,9 @@
1
  {
2
- "epoch": 5.0,
3
- "total_flos": 1.5670950020754964e+18,
4
- "train_loss": 1.4066738275202308,
5
- "train_runtime": 8016.4558,
6
- "train_samples": 29489,
7
- "train_samples_per_second": 2.044,
8
- "train_steps_per_second": 0.128
9
  }
 
1
  {
2
+ "epoch": 10.0,
3
+ "total_flos": 2.2627322959180595e+18,
4
+ "train_loss": 1.3636517344294368,
5
+ "train_runtime": 11668.1199,
6
+ "train_samples": 30867,
7
+ "train_samples_per_second": 2.027,
8
+ "train_steps_per_second": 0.127
9
  }
trainer_state.json CHANGED
@@ -1,1509 +1,2186 @@
1
  {
2
  "best_metric": null,
3
  "best_model_checkpoint": null,
4
- "epoch": 5.0,
5
  "eval_steps": 500,
6
- "global_step": 1025,
7
  "is_hyper_param_search": false,
8
  "is_local_process_zero": true,
9
  "is_world_process_zero": true,
10
  "log_history": [
11
  {
12
- "epoch": 0.004878048780487805,
13
- "grad_norm": 616.0,
14
- "learning_rate": 1.941747572815534e-06,
15
- "loss": 39.9034,
16
  "step": 1
17
  },
18
  {
19
- "epoch": 0.024390243902439025,
20
- "grad_norm": 466.0,
21
- "learning_rate": 9.70873786407767e-06,
22
- "loss": 34.9341,
23
  "step": 5
24
  },
25
  {
26
- "epoch": 0.04878048780487805,
27
- "grad_norm": 251.0,
28
- "learning_rate": 1.941747572815534e-05,
29
- "loss": 29.4015,
30
  "step": 10
31
  },
32
  {
33
- "epoch": 0.07317073170731707,
34
- "grad_norm": 56.25,
35
- "learning_rate": 2.912621359223301e-05,
36
- "loss": 17.9289,
37
  "step": 15
38
  },
39
  {
40
- "epoch": 0.0975609756097561,
41
- "grad_norm": 39.75,
42
- "learning_rate": 3.883495145631068e-05,
43
- "loss": 14.9717,
44
  "step": 20
45
  },
46
  {
47
- "epoch": 0.12195121951219512,
48
- "grad_norm": 15.75,
49
- "learning_rate": 4.854368932038835e-05,
50
- "loss": 12.4657,
51
  "step": 25
52
  },
53
  {
54
- "epoch": 0.14634146341463414,
55
- "grad_norm": 5.15625,
56
- "learning_rate": 5.825242718446602e-05,
57
- "loss": 10.6748,
58
  "step": 30
59
  },
60
  {
61
- "epoch": 0.17073170731707318,
62
- "grad_norm": 3.890625,
63
- "learning_rate": 6.79611650485437e-05,
64
- "loss": 9.811,
65
  "step": 35
66
  },
67
  {
68
- "epoch": 0.1951219512195122,
69
- "grad_norm": 6.75,
70
- "learning_rate": 7.766990291262136e-05,
71
- "loss": 9.6309,
72
  "step": 40
73
  },
74
  {
75
- "epoch": 0.21951219512195122,
76
- "grad_norm": 15.3125,
77
- "learning_rate": 8.737864077669902e-05,
78
- "loss": 8.5315,
79
  "step": 45
80
  },
81
  {
82
- "epoch": 0.24390243902439024,
83
- "grad_norm": 31.75,
84
- "learning_rate": 9.70873786407767e-05,
85
- "loss": 6.2519,
86
  "step": 50
87
  },
88
  {
89
- "epoch": 0.2682926829268293,
90
- "grad_norm": 5.25,
91
- "learning_rate": 0.00010679611650485437,
92
- "loss": 2.2665,
93
  "step": 55
94
  },
95
  {
96
- "epoch": 0.2926829268292683,
97
- "grad_norm": 2.40625,
98
- "learning_rate": 0.00011650485436893204,
99
- "loss": 1.5008,
100
  "step": 60
101
  },
102
  {
103
- "epoch": 0.3170731707317073,
104
- "grad_norm": 1.2578125,
105
- "learning_rate": 0.00012621359223300972,
106
- "loss": 1.3289,
107
  "step": 65
108
  },
109
  {
110
- "epoch": 0.34146341463414637,
111
- "grad_norm": 2.796875,
112
- "learning_rate": 0.0001359223300970874,
113
- "loss": 1.2198,
114
  "step": 70
115
  },
116
  {
117
- "epoch": 0.36585365853658536,
118
- "grad_norm": 1.4375,
119
- "learning_rate": 0.00014563106796116506,
120
- "loss": 1.1493,
121
  "step": 75
122
  },
123
  {
124
- "epoch": 0.3902439024390244,
125
- "grad_norm": 8.1875,
126
- "learning_rate": 0.0001553398058252427,
127
- "loss": 1.0862,
128
  "step": 80
129
  },
130
  {
131
- "epoch": 0.4146341463414634,
132
- "grad_norm": 3.0625,
133
- "learning_rate": 0.0001650485436893204,
134
- "loss": 1.0768,
135
  "step": 85
136
  },
137
  {
138
- "epoch": 0.43902439024390244,
139
- "grad_norm": 1.2109375,
140
- "learning_rate": 0.00017475728155339805,
141
- "loss": 1.0468,
142
  "step": 90
143
  },
144
  {
145
- "epoch": 0.4634146341463415,
146
- "grad_norm": 4.21875,
147
- "learning_rate": 0.00018446601941747576,
148
- "loss": 0.9991,
149
  "step": 95
150
  },
151
  {
152
- "epoch": 0.4878048780487805,
153
- "grad_norm": 2.0625,
154
- "learning_rate": 0.0001941747572815534,
155
- "loss": 0.9596,
156
  "step": 100
157
  },
158
  {
159
- "epoch": 0.5121951219512195,
160
- "grad_norm": 1.5078125,
161
- "learning_rate": 0.00019999767797859854,
162
- "loss": 0.9712,
163
  "step": 105
164
  },
165
  {
166
- "epoch": 0.5365853658536586,
167
- "grad_norm": 0.890625,
168
- "learning_rate": 0.0001999715564762413,
169
- "loss": 0.9139,
170
  "step": 110
171
  },
172
  {
173
- "epoch": 0.5609756097560976,
174
- "grad_norm": 1.1875,
175
- "learning_rate": 0.00019991641855173097,
176
- "loss": 0.9175,
177
  "step": 115
178
  },
179
  {
180
- "epoch": 0.5853658536585366,
181
- "grad_norm": 1.25,
182
- "learning_rate": 0.00019983228020867242,
183
- "loss": 0.8959,
184
  "step": 120
185
  },
186
  {
187
- "epoch": 0.6097560975609756,
188
- "grad_norm": 1.3203125,
189
- "learning_rate": 0.00019971916586794867,
190
- "loss": 0.8605,
191
  "step": 125
192
  },
193
  {
194
- "epoch": 0.6341463414634146,
195
- "grad_norm": 0.98828125,
196
- "learning_rate": 0.00019957710836063263,
197
- "loss": 0.8857,
198
  "step": 130
199
  },
200
  {
201
- "epoch": 0.6585365853658537,
202
- "grad_norm": 1.8046875,
203
- "learning_rate": 0.00019940614891845809,
204
- "loss": 0.8533,
205
  "step": 135
206
  },
207
  {
208
- "epoch": 0.6829268292682927,
209
- "grad_norm": 1.9140625,
210
- "learning_rate": 0.00019920633716185226,
211
- "loss": 0.8332,
212
  "step": 140
213
  },
214
  {
215
- "epoch": 0.7073170731707317,
216
- "grad_norm": 0.9140625,
217
- "learning_rate": 0.00019897773108553378,
218
- "loss": 0.8485,
219
  "step": 145
220
  },
221
  {
222
- "epoch": 0.7317073170731707,
223
- "grad_norm": 1.9296875,
224
- "learning_rate": 0.00019872039704167964,
225
- "loss": 0.8479,
 
 
 
 
 
 
 
 
226
  "step": 150
227
  },
228
  {
229
- "epoch": 0.7560975609756098,
230
- "grad_norm": 1.4921875,
231
- "learning_rate": 0.00019843440972066697,
232
- "loss": 0.8418,
233
  "step": 155
234
  },
235
  {
236
- "epoch": 0.7804878048780488,
237
- "grad_norm": 1.9296875,
238
- "learning_rate": 0.00019811985212939416,
239
- "loss": 0.8513,
240
  "step": 160
241
  },
242
  {
243
- "epoch": 0.8048780487804879,
244
- "grad_norm": 1.5234375,
245
- "learning_rate": 0.00019777681556718864,
246
- "loss": 0.8217,
247
  "step": 165
248
  },
249
  {
250
- "epoch": 0.8292682926829268,
251
- "grad_norm": 1.8671875,
252
- "learning_rate": 0.00019740539959930725,
253
- "loss": 0.8169,
254
  "step": 170
255
  },
256
  {
257
- "epoch": 0.8536585365853658,
258
- "grad_norm": 1.046875,
259
- "learning_rate": 0.00019700571202803797,
260
- "loss": 0.8386,
261
  "step": 175
262
  },
263
  {
264
- "epoch": 0.8780487804878049,
265
- "grad_norm": 1.2421875,
266
- "learning_rate": 0.00019657786886141052,
267
- "loss": 0.8144,
268
  "step": 180
269
  },
270
  {
271
- "epoch": 0.9024390243902439,
272
- "grad_norm": 4.15625,
273
- "learning_rate": 0.00019612199427952552,
274
- "loss": 0.8229,
275
  "step": 185
276
  },
277
  {
278
- "epoch": 0.926829268292683,
279
- "grad_norm": 0.70703125,
280
- "learning_rate": 0.00019563822059851145,
281
- "loss": 0.8128,
282
  "step": 190
283
  },
284
  {
285
- "epoch": 0.9512195121951219,
286
- "grad_norm": 0.97265625,
287
- "learning_rate": 0.00019512668823212055,
288
- "loss": 0.8072,
289
  "step": 195
290
  },
291
  {
292
- "epoch": 0.975609756097561,
293
- "grad_norm": 0.6484375,
294
- "learning_rate": 0.0001945875456509739,
295
- "loss": 0.8097,
296
  "step": 200
297
  },
298
  {
299
- "epoch": 1.0,
300
- "grad_norm": 0.6953125,
301
- "learning_rate": 0.00019402094933946857,
302
- "loss": 0.7892,
303
- "step": 205
304
- },
305
- {
306
- "epoch": 1.0,
307
- "eval_loss": 2.2406020164489746,
308
- "eval_runtime": 0.9998,
309
- "eval_samples_per_second": 5.001,
310
- "eval_steps_per_second": 2.0,
311
  "step": 205
312
  },
313
  {
314
- "epoch": 1.024390243902439,
315
- "grad_norm": 0.6796875,
316
- "learning_rate": 0.0001934270637503584,
317
- "loss": 0.7753,
318
  "step": 210
319
  },
320
  {
321
- "epoch": 1.048780487804878,
322
- "grad_norm": 0.71875,
323
- "learning_rate": 0.00019280606125702203,
324
- "loss": 0.7289,
325
  "step": 215
326
  },
327
  {
328
- "epoch": 1.0731707317073171,
329
- "grad_norm": 0.625,
330
- "learning_rate": 0.00019215812210343226,
331
- "loss": 0.7656,
332
  "step": 220
333
  },
334
  {
335
- "epoch": 1.0975609756097562,
336
- "grad_norm": 0.65625,
337
- "learning_rate": 0.00019148343435184079,
338
- "loss": 0.7603,
339
  "step": 225
340
  },
341
  {
342
- "epoch": 1.1219512195121952,
343
- "grad_norm": 0.671875,
344
- "learning_rate": 0.00019078219382819353,
345
- "loss": 0.7491,
346
  "step": 230
347
  },
348
  {
349
- "epoch": 1.146341463414634,
350
- "grad_norm": 0.56640625,
351
- "learning_rate": 0.00019005460406529311,
352
- "loss": 0.7249,
353
  "step": 235
354
  },
355
  {
356
- "epoch": 1.170731707317073,
357
- "grad_norm": 0.57421875,
358
- "learning_rate": 0.00018930087624372387,
359
- "loss": 0.7755,
360
  "step": 240
361
  },
362
  {
363
- "epoch": 1.1951219512195121,
364
- "grad_norm": 0.703125,
365
- "learning_rate": 0.00018852122913055742,
366
- "loss": 0.7553,
367
  "step": 245
368
  },
369
  {
370
- "epoch": 1.2195121951219512,
371
- "grad_norm": 0.890625,
372
- "learning_rate": 0.00018771588901585635,
373
- "loss": 0.7482,
374
  "step": 250
375
  },
376
  {
377
- "epoch": 1.2439024390243902,
378
- "grad_norm": 0.77734375,
379
- "learning_rate": 0.00018688508964699404,
380
- "loss": 0.7402,
381
  "step": 255
382
  },
383
  {
384
- "epoch": 1.2682926829268293,
385
- "grad_norm": 1.015625,
386
- "learning_rate": 0.00018602907216081044,
387
- "loss": 0.7263,
388
  "step": 260
389
  },
390
  {
391
- "epoch": 1.2926829268292683,
392
- "grad_norm": 1.3828125,
393
- "learning_rate": 0.0001851480850136228,
394
- "loss": 0.7435,
395
  "step": 265
396
  },
397
  {
398
- "epoch": 1.3170731707317074,
399
- "grad_norm": 0.58984375,
400
- "learning_rate": 0.00018424238390911198,
401
- "loss": 0.7457,
402
  "step": 270
403
  },
404
  {
405
- "epoch": 1.3414634146341464,
406
- "grad_norm": 1.0078125,
407
- "learning_rate": 0.00018331223172410535,
408
- "loss": 0.7483,
409
  "step": 275
410
  },
411
  {
412
- "epoch": 1.3658536585365852,
413
- "grad_norm": 0.609375,
414
- "learning_rate": 0.00018235789843227756,
415
- "loss": 0.7486,
416
  "step": 280
417
  },
418
  {
419
- "epoch": 1.3902439024390243,
420
- "grad_norm": 1.265625,
421
- "learning_rate": 0.00018137966102579176,
422
- "loss": 0.7548,
423
  "step": 285
424
  },
425
  {
426
- "epoch": 1.4146341463414633,
427
- "grad_norm": 1.1953125,
428
- "learning_rate": 0.00018037780343490312,
429
- "loss": 0.7282,
430
  "step": 290
431
  },
432
  {
433
- "epoch": 1.4390243902439024,
434
- "grad_norm": 0.6875,
435
- "learning_rate": 0.00017935261644554942,
436
- "loss": 0.7415,
437
  "step": 295
438
  },
439
  {
440
- "epoch": 1.4634146341463414,
441
- "grad_norm": 0.73046875,
442
- "learning_rate": 0.0001783043976149511,
443
- "loss": 0.7362,
 
 
 
 
 
 
 
 
444
  "step": 300
445
  },
446
  {
447
- "epoch": 1.4878048780487805,
448
- "grad_norm": 0.578125,
449
- "learning_rate": 0.0001772334511852463,
450
- "loss": 0.724,
451
  "step": 305
452
  },
453
  {
454
- "epoch": 1.5121951219512195,
455
- "grad_norm": 0.59765625,
456
- "learning_rate": 0.0001761400879951856,
457
- "loss": 0.7249,
458
  "step": 310
459
  },
460
  {
461
- "epoch": 1.5365853658536586,
462
- "grad_norm": 1.6953125,
463
- "learning_rate": 0.00017502462538991205,
464
- "loss": 0.7212,
465
  "step": 315
466
  },
467
  {
468
- "epoch": 1.5609756097560976,
469
- "grad_norm": 0.6328125,
470
- "learning_rate": 0.00017388738712885275,
471
- "loss": 0.7194,
472
  "step": 320
473
  },
474
  {
475
- "epoch": 1.5853658536585367,
476
- "grad_norm": 1.3125,
477
- "learning_rate": 0.0001727287032917487,
478
- "loss": 0.765,
479
  "step": 325
480
  },
481
  {
482
- "epoch": 1.6097560975609757,
483
- "grad_norm": 1.578125,
484
- "learning_rate": 0.00017154891018285028,
485
- "loss": 0.7318,
486
  "step": 330
487
  },
488
  {
489
- "epoch": 1.6341463414634148,
490
- "grad_norm": 0.50390625,
491
- "learning_rate": 0.00017034835023330597,
492
- "loss": 0.7396,
493
  "step": 335
494
  },
495
  {
496
- "epoch": 1.6585365853658538,
497
- "grad_norm": 2.0625,
498
- "learning_rate": 0.00016912737190177292,
499
- "loss": 0.737,
500
  "step": 340
501
  },
502
  {
503
- "epoch": 1.6829268292682928,
504
- "grad_norm": 0.59765625,
505
- "learning_rate": 0.00016788632957327772,
506
- "loss": 0.7302,
507
  "step": 345
508
  },
509
  {
510
- "epoch": 1.7073170731707317,
511
- "grad_norm": 0.65234375,
512
- "learning_rate": 0.00016662558345635753,
513
- "loss": 0.7193,
514
  "step": 350
515
  },
516
  {
517
- "epoch": 1.7317073170731707,
518
- "grad_norm": 0.76953125,
519
- "learning_rate": 0.00016534549947851062,
520
- "loss": 0.7147,
521
  "step": 355
522
  },
523
  {
524
- "epoch": 1.7560975609756098,
525
- "grad_norm": 0.73828125,
526
- "learning_rate": 0.00016404644917998698,
527
- "loss": 0.7539,
528
  "step": 360
529
  },
530
  {
531
- "epoch": 1.7804878048780488,
532
- "grad_norm": 0.9453125,
533
- "learning_rate": 0.00016272880960595024,
534
- "loss": 0.754,
535
  "step": 365
536
  },
537
  {
538
- "epoch": 1.8048780487804879,
539
- "grad_norm": 0.90625,
540
- "learning_rate": 0.00016139296319704117,
541
- "loss": 0.7346,
542
  "step": 370
543
  },
544
  {
545
- "epoch": 1.8292682926829267,
546
- "grad_norm": 0.89453125,
547
- "learning_rate": 0.00016003929767837588,
548
- "loss": 0.6961,
549
  "step": 375
550
  },
551
  {
552
- "epoch": 1.8536585365853657,
553
- "grad_norm": 1.078125,
554
- "learning_rate": 0.00015866820594700944,
555
- "loss": 0.7259,
556
  "step": 380
557
  },
558
  {
559
- "epoch": 1.8780487804878048,
560
- "grad_norm": 0.65625,
561
- "learning_rate": 0.00015728008595789926,
562
- "loss": 0.7198,
563
  "step": 385
564
  },
565
  {
566
- "epoch": 1.9024390243902438,
567
- "grad_norm": 0.515625,
568
- "learning_rate": 0.0001558753406083995,
569
- "loss": 0.6858,
570
  "step": 390
571
  },
572
  {
573
- "epoch": 1.9268292682926829,
574
- "grad_norm": 0.60546875,
575
- "learning_rate": 0.00015445437762132174,
576
- "loss": 0.6987,
577
  "step": 395
578
  },
579
  {
580
- "epoch": 1.951219512195122,
581
- "grad_norm": 0.5625,
582
- "learning_rate": 0.0001530176094265945,
583
- "loss": 0.7054,
584
  "step": 400
585
  },
586
  {
587
- "epoch": 1.975609756097561,
588
- "grad_norm": 0.62890625,
589
- "learning_rate": 0.00015156545304155698,
590
- "loss": 0.7226,
591
  "step": 405
592
  },
593
  {
594
- "epoch": 2.0,
595
- "grad_norm": 0.52734375,
596
- "learning_rate": 0.00015009832994992102,
597
- "loss": 0.7033,
598
- "step": 410
599
- },
600
- {
601
- "epoch": 2.0,
602
- "eval_loss": 2.2004494667053223,
603
- "eval_runtime": 1.0005,
604
- "eval_samples_per_second": 4.998,
605
- "eval_steps_per_second": 1.999,
606
  "step": 410
607
  },
608
  {
609
- "epoch": 2.024390243902439,
610
- "grad_norm": 0.6328125,
611
- "learning_rate": 0.0001486166659794368,
612
- "loss": 0.6398,
613
  "step": 415
614
  },
615
  {
616
- "epoch": 2.048780487804878,
617
- "grad_norm": 0.65234375,
618
- "learning_rate": 0.00014712089117829776,
619
- "loss": 0.647,
620
  "step": 420
621
  },
622
  {
623
- "epoch": 2.073170731707317,
624
- "grad_norm": 0.66796875,
625
- "learning_rate": 0.0001456114396903204,
626
- "loss": 0.6333,
627
  "step": 425
628
  },
629
  {
630
- "epoch": 2.097560975609756,
631
- "grad_norm": 0.60546875,
632
- "learning_rate": 0.0001440887496289356,
633
- "loss": 0.6296,
634
  "step": 430
635
  },
636
  {
637
- "epoch": 2.1219512195121952,
638
- "grad_norm": 0.546875,
639
- "learning_rate": 0.00014255326295002754,
640
- "loss": 0.6516,
641
  "step": 435
642
  },
643
  {
644
- "epoch": 2.1463414634146343,
645
- "grad_norm": 0.671875,
646
- "learning_rate": 0.00014100542532365724,
647
- "loss": 0.6468,
648
  "step": 440
649
  },
650
  {
651
- "epoch": 2.1707317073170733,
652
- "grad_norm": 0.578125,
653
- "learning_rate": 0.0001394456860047086,
654
- "loss": 0.6472,
 
 
 
 
 
 
 
 
655
  "step": 445
656
  },
657
  {
658
- "epoch": 2.1951219512195124,
659
- "grad_norm": 0.77734375,
660
- "learning_rate": 0.00013787449770249336,
661
- "loss": 0.6587,
662
  "step": 450
663
  },
664
  {
665
- "epoch": 2.2195121951219514,
666
- "grad_norm": 0.70703125,
667
- "learning_rate": 0.0001362923164493538,
668
- "loss": 0.6505,
669
  "step": 455
670
  },
671
  {
672
- "epoch": 2.2439024390243905,
673
- "grad_norm": 0.69140625,
674
- "learning_rate": 0.00013469960146830073,
675
- "loss": 0.6463,
676
  "step": 460
677
  },
678
  {
679
- "epoch": 2.2682926829268295,
680
- "grad_norm": 0.72265625,
681
- "learning_rate": 0.00013309681503972565,
682
- "loss": 0.6372,
683
  "step": 465
684
  },
685
  {
686
- "epoch": 2.292682926829268,
687
- "grad_norm": 0.80078125,
688
- "learning_rate": 0.00013148442236722506,
689
- "loss": 0.6285,
690
  "step": 470
691
  },
692
  {
693
- "epoch": 2.317073170731707,
694
- "grad_norm": 0.7578125,
695
- "learning_rate": 0.00012986289144257705,
696
- "loss": 0.6645,
697
  "step": 475
698
  },
699
  {
700
- "epoch": 2.341463414634146,
701
- "grad_norm": 1.0234375,
702
- "learning_rate": 0.00012823269290990777,
703
- "loss": 0.6663,
704
  "step": 480
705
  },
706
  {
707
- "epoch": 2.3658536585365852,
708
- "grad_norm": 1.0625,
709
- "learning_rate": 0.0001265942999290887,
710
- "loss": 0.6495,
711
  "step": 485
712
  },
713
  {
714
- "epoch": 2.3902439024390243,
715
- "grad_norm": 1.0078125,
716
- "learning_rate": 0.00012494818803840367,
717
- "loss": 0.6429,
718
  "step": 490
719
  },
720
  {
721
- "epoch": 2.4146341463414633,
722
- "grad_norm": 1.796875,
723
- "learning_rate": 0.00012329483501652492,
724
- "loss": 0.6573,
725
  "step": 495
726
  },
727
  {
728
- "epoch": 2.4390243902439024,
729
- "grad_norm": 0.921875,
730
- "learning_rate": 0.00012163472074383994,
731
- "loss": 0.6487,
732
  "step": 500
733
  },
734
  {
735
- "epoch": 2.4634146341463414,
736
- "grad_norm": 0.59375,
737
- "learning_rate": 0.00011996832706316739,
738
- "loss": 0.6726,
739
  "step": 505
740
  },
741
  {
742
- "epoch": 2.4878048780487805,
743
- "grad_norm": 0.5546875,
744
- "learning_rate": 0.00011829613763990384,
745
- "loss": 0.6546,
746
  "step": 510
747
  },
748
  {
749
- "epoch": 2.5121951219512195,
750
- "grad_norm": 0.57421875,
751
- "learning_rate": 0.00011661863782164153,
752
- "loss": 0.6306,
753
  "step": 515
754
  },
755
  {
756
- "epoch": 2.5365853658536586,
757
- "grad_norm": 0.61328125,
758
- "learning_rate": 0.00011493631449729767,
759
- "loss": 0.6591,
760
  "step": 520
761
  },
762
  {
763
- "epoch": 2.5609756097560976,
764
- "grad_norm": 0.578125,
765
- "learning_rate": 0.00011324965595579666,
766
- "loss": 0.6387,
767
  "step": 525
768
  },
769
  {
770
- "epoch": 2.5853658536585367,
771
- "grad_norm": 0.60546875,
772
- "learning_rate": 0.00011155915174434561,
773
- "loss": 0.6539,
774
  "step": 530
775
  },
776
  {
777
- "epoch": 2.6097560975609757,
778
- "grad_norm": 0.87109375,
779
- "learning_rate": 0.00010986529252634503,
780
- "loss": 0.6654,
781
  "step": 535
782
  },
783
  {
784
- "epoch": 2.6341463414634148,
785
- "grad_norm": 0.859375,
786
- "learning_rate": 0.00010816856993897522,
787
- "loss": 0.6471,
788
  "step": 540
789
  },
790
  {
791
- "epoch": 2.658536585365854,
792
- "grad_norm": 0.5703125,
793
- "learning_rate": 0.00010646947645050023,
794
- "loss": 0.6482,
795
  "step": 545
796
  },
797
  {
798
- "epoch": 2.682926829268293,
799
- "grad_norm": 0.5703125,
800
- "learning_rate": 0.00010476850521733048,
801
- "loss": 0.6252,
802
  "step": 550
803
  },
804
  {
805
- "epoch": 2.7073170731707314,
806
- "grad_norm": 0.69140625,
807
- "learning_rate": 0.00010306614994088582,
808
- "loss": 0.6605,
809
  "step": 555
810
  },
811
  {
812
- "epoch": 2.7317073170731705,
813
- "grad_norm": 0.52734375,
814
- "learning_rate": 0.00010136290472430013,
815
- "loss": 0.646,
816
  "step": 560
817
  },
818
  {
819
- "epoch": 2.7560975609756095,
820
- "grad_norm": 0.64453125,
821
- "learning_rate": 9.965926392900956e-05,
822
- "loss": 0.6492,
823
  "step": 565
824
  },
825
  {
826
- "epoch": 2.7804878048780486,
827
- "grad_norm": 0.63671875,
828
- "learning_rate": 9.795572203126573e-05,
829
- "loss": 0.6477,
830
  "step": 570
831
  },
832
  {
833
- "epoch": 2.8048780487804876,
834
- "grad_norm": 0.703125,
835
- "learning_rate": 9.625277347861553e-05,
836
- "loss": 0.6268,
837
  "step": 575
838
  },
839
  {
840
- "epoch": 2.8292682926829267,
841
- "grad_norm": 0.80078125,
842
- "learning_rate": 9.455091254638939e-05,
843
- "loss": 0.6364,
844
  "step": 580
845
  },
846
  {
847
- "epoch": 2.8536585365853657,
848
- "grad_norm": 0.62890625,
849
- "learning_rate": 9.285063319423939e-05,
850
- "loss": 0.6497,
851
  "step": 585
852
  },
853
  {
854
- "epoch": 2.8780487804878048,
855
- "grad_norm": 0.58203125,
856
- "learning_rate": 9.115242892276909e-05,
857
- "loss": 0.6446,
858
  "step": 590
859
  },
860
  {
861
- "epoch": 2.902439024390244,
862
- "grad_norm": 0.70703125,
863
- "learning_rate": 8.945679263029661e-05,
864
- "loss": 0.6657,
 
 
 
 
 
 
 
 
865
  "step": 595
866
  },
867
  {
868
- "epoch": 2.926829268292683,
869
- "grad_norm": 0.64453125,
870
- "learning_rate": 8.776421646979232e-05,
871
- "loss": 0.6498,
872
  "step": 600
873
  },
874
  {
875
- "epoch": 2.951219512195122,
876
- "grad_norm": 0.5546875,
877
- "learning_rate": 8.607519170603328e-05,
878
- "loss": 0.6623,
879
  "step": 605
880
  },
881
  {
882
- "epoch": 2.975609756097561,
883
- "grad_norm": 0.55078125,
884
- "learning_rate": 8.439020857301503e-05,
885
- "loss": 0.6467,
886
  "step": 610
887
  },
888
  {
889
- "epoch": 3.0,
890
- "grad_norm": 0.6875,
891
- "learning_rate": 8.270975613166281e-05,
892
- "loss": 0.6346,
893
- "step": 615
894
- },
895
- {
896
- "epoch": 3.0,
897
- "eval_loss": 2.245811939239502,
898
- "eval_runtime": 1.001,
899
- "eval_samples_per_second": 4.995,
900
- "eval_steps_per_second": 1.998,
901
  "step": 615
902
  },
903
  {
904
- "epoch": 3.024390243902439,
905
- "grad_norm": 0.609375,
906
- "learning_rate": 8.103432212788323e-05,
907
- "loss": 0.5718,
908
  "step": 620
909
  },
910
  {
911
- "epoch": 3.048780487804878,
912
- "grad_norm": 0.66796875,
913
- "learning_rate": 7.936439285099752e-05,
914
- "loss": 0.5989,
915
  "step": 625
916
  },
917
  {
918
- "epoch": 3.073170731707317,
919
  "grad_norm": 0.62890625,
920
- "learning_rate": 7.770045299259774e-05,
921
- "loss": 0.598,
922
  "step": 630
923
  },
924
  {
925
- "epoch": 3.097560975609756,
926
- "grad_norm": 0.5625,
927
- "learning_rate": 7.60429855058664e-05,
928
- "loss": 0.5731,
929
  "step": 635
930
  },
931
  {
932
- "epoch": 3.1219512195121952,
933
- "grad_norm": 0.56640625,
934
- "learning_rate": 7.439247146540109e-05,
935
- "loss": 0.6034,
936
  "step": 640
937
  },
938
  {
939
- "epoch": 3.1463414634146343,
940
- "grad_norm": 0.6015625,
941
- "learning_rate": 7.274938992758403e-05,
942
- "loss": 0.5669,
943
  "step": 645
944
  },
945
  {
946
- "epoch": 3.1707317073170733,
947
- "grad_norm": 0.59765625,
948
- "learning_rate": 7.111421779153745e-05,
949
- "loss": 0.5873,
950
  "step": 650
951
  },
952
  {
953
- "epoch": 3.1951219512195124,
954
- "grad_norm": 0.60546875,
955
- "learning_rate": 6.94874296607052e-05,
956
- "loss": 0.5674,
957
  "step": 655
958
  },
959
  {
960
- "epoch": 3.2195121951219514,
961
- "grad_norm": 0.54296875,
962
- "learning_rate": 6.786949770510071e-05,
963
- "loss": 0.5726,
964
  "step": 660
965
  },
966
  {
967
- "epoch": 3.2439024390243905,
968
- "grad_norm": 0.578125,
969
- "learning_rate": 6.626089152426097e-05,
970
- "loss": 0.6006,
971
  "step": 665
972
  },
973
  {
974
- "epoch": 3.2682926829268295,
975
- "grad_norm": 0.58984375,
976
- "learning_rate": 6.4662078010947e-05,
977
- "loss": 0.5729,
978
  "step": 670
979
  },
980
  {
981
- "epoch": 3.292682926829268,
982
- "grad_norm": 0.546875,
983
- "learning_rate": 6.307352121562949e-05,
984
- "loss": 0.5719,
985
  "step": 675
986
  },
987
  {
988
- "epoch": 3.317073170731707,
989
- "grad_norm": 0.5859375,
990
- "learning_rate": 6.149568221179993e-05,
991
- "loss": 0.571,
992
  "step": 680
993
  },
994
  {
995
- "epoch": 3.341463414634146,
996
- "grad_norm": 0.5859375,
997
- "learning_rate": 5.992901896214526e-05,
998
- "loss": 0.5674,
999
  "step": 685
1000
  },
1001
  {
1002
- "epoch": 3.3658536585365852,
1003
- "grad_norm": 0.6640625,
1004
- "learning_rate": 5.837398618562584e-05,
1005
- "loss": 0.5772,
1006
  "step": 690
1007
  },
1008
  {
1009
- "epoch": 3.3902439024390243,
1010
- "grad_norm": 0.58203125,
1011
- "learning_rate": 5.68310352254946e-05,
1012
- "loss": 0.5893,
1013
  "step": 695
1014
  },
1015
  {
1016
- "epoch": 3.4146341463414633,
1017
- "grad_norm": 0.62109375,
1018
- "learning_rate": 5.5300613918296295e-05,
1019
- "loss": 0.5771,
1020
  "step": 700
1021
  },
1022
  {
1023
- "epoch": 3.4390243902439024,
1024
- "grad_norm": 0.58203125,
1025
- "learning_rate": 5.378316646388424e-05,
1026
- "loss": 0.5721,
1027
  "step": 705
1028
  },
1029
  {
1030
- "epoch": 3.4634146341463414,
1031
- "grad_norm": 0.578125,
1032
- "learning_rate": 5.227913329649271e-05,
1033
- "loss": 0.5788,
1034
  "step": 710
1035
  },
1036
  {
1037
- "epoch": 3.4878048780487805,
1038
- "grad_norm": 0.59765625,
1039
- "learning_rate": 5.078895095690249e-05,
1040
- "loss": 0.5693,
1041
  "step": 715
1042
  },
1043
  {
1044
- "epoch": 3.5121951219512195,
1045
- "grad_norm": 0.5703125,
1046
- "learning_rate": 4.931305196573621e-05,
1047
- "loss": 0.6107,
1048
  "step": 720
1049
  },
1050
  {
1051
- "epoch": 3.5365853658536586,
1052
- "grad_norm": 0.58203125,
1053
- "learning_rate": 4.78518646979206e-05,
1054
- "loss": 0.5695,
1055
  "step": 725
1056
  },
1057
  {
1058
- "epoch": 3.5609756097560976,
1059
- "grad_norm": 0.58984375,
1060
- "learning_rate": 4.6405813258352135e-05,
1061
- "loss": 0.5707,
1062
  "step": 730
1063
  },
1064
  {
1065
- "epoch": 3.5853658536585367,
1066
- "grad_norm": 0.62109375,
1067
- "learning_rate": 4.4975317358801885e-05,
1068
- "loss": 0.5635,
1069
  "step": 735
1070
  },
1071
  {
1072
- "epoch": 3.6097560975609757,
1073
- "grad_norm": 0.57421875,
1074
- "learning_rate": 4.3560792196095543e-05,
1075
- "loss": 0.5747,
1076
  "step": 740
1077
  },
1078
  {
1079
- "epoch": 3.6341463414634148,
1080
- "grad_norm": 0.57421875,
1081
- "learning_rate": 4.216264833160396e-05,
1082
- "loss": 0.5856,
 
 
 
 
 
 
 
 
1083
  "step": 745
1084
  },
1085
  {
1086
- "epoch": 3.658536585365854,
1087
- "grad_norm": 0.6015625,
1088
- "learning_rate": 4.0781291572078806e-05,
1089
- "loss": 0.5797,
1090
  "step": 750
1091
  },
1092
  {
1093
- "epoch": 3.682926829268293,
1094
- "grad_norm": 0.55859375,
1095
- "learning_rate": 3.941712285186878e-05,
1096
- "loss": 0.5643,
1097
  "step": 755
1098
  },
1099
  {
1100
- "epoch": 3.7073170731707314,
1101
- "grad_norm": 0.5859375,
1102
- "learning_rate": 3.807053811654948e-05,
1103
- "loss": 0.5723,
1104
  "step": 760
1105
  },
1106
  {
1107
- "epoch": 3.7317073170731705,
1108
- "grad_norm": 0.59375,
1109
- "learning_rate": 3.674192820800156e-05,
1110
- "loss": 0.5894,
1111
  "step": 765
1112
  },
1113
  {
1114
- "epoch": 3.7560975609756095,
1115
- "grad_norm": 0.55859375,
1116
- "learning_rate": 3.543167875097013e-05,
1117
- "loss": 0.5505,
1118
  "step": 770
1119
  },
1120
  {
1121
- "epoch": 3.7804878048780486,
1122
- "grad_norm": 0.5859375,
1123
- "learning_rate": 3.4140170041138385e-05,
1124
- "loss": 0.5752,
1125
  "step": 775
1126
  },
1127
  {
1128
- "epoch": 3.8048780487804876,
1129
- "grad_norm": 0.55859375,
1130
- "learning_rate": 3.286777693474803e-05,
1131
- "loss": 0.5629,
1132
  "step": 780
1133
  },
1134
  {
1135
- "epoch": 3.8292682926829267,
1136
- "grad_norm": 0.59765625,
1137
- "learning_rate": 3.1614868739798495e-05,
1138
- "loss": 0.5787,
1139
  "step": 785
1140
  },
1141
  {
1142
- "epoch": 3.8536585365853657,
1143
- "grad_norm": 0.6015625,
1144
- "learning_rate": 3.0381809108856398e-05,
1145
- "loss": 0.574,
1146
  "step": 790
1147
  },
1148
  {
1149
- "epoch": 3.8780487804878048,
1150
- "grad_norm": 0.6171875,
1151
- "learning_rate": 2.9168955933506648e-05,
1152
- "loss": 0.5826,
1153
  "step": 795
1154
  },
1155
  {
1156
- "epoch": 3.902439024390244,
1157
- "grad_norm": 0.578125,
1158
- "learning_rate": 2.79766612404755e-05,
1159
- "loss": 0.5725,
1160
  "step": 800
1161
  },
1162
  {
1163
- "epoch": 3.926829268292683,
1164
- "grad_norm": 0.55859375,
1165
- "learning_rate": 2.6805271089455986e-05,
1166
- "loss": 0.5612,
1167
  "step": 805
1168
  },
1169
  {
1170
- "epoch": 3.951219512195122,
1171
- "grad_norm": 0.6328125,
1172
- "learning_rate": 2.565512547266511e-05,
1173
- "loss": 0.5721,
1174
  "step": 810
1175
  },
1176
  {
1177
- "epoch": 3.975609756097561,
1178
- "grad_norm": 0.5625,
1179
- "learning_rate": 2.4526558216162322e-05,
1180
- "loss": 0.5725,
1181
  "step": 815
1182
  },
1183
  {
1184
- "epoch": 4.0,
1185
- "grad_norm": 0.56640625,
1186
- "learning_rate": 2.3419896882957527e-05,
1187
- "loss": 0.5755,
1188
- "step": 820
1189
- },
1190
- {
1191
- "epoch": 4.0,
1192
- "eval_loss": 2.2934529781341553,
1193
- "eval_runtime": 0.9995,
1194
- "eval_samples_per_second": 5.002,
1195
- "eval_steps_per_second": 2.001,
1196
  "step": 820
1197
  },
1198
  {
1199
- "epoch": 4.024390243902439,
1200
- "grad_norm": 0.55859375,
1201
- "learning_rate": 2.2335462677936957e-05,
1202
- "loss": 0.5324,
1203
  "step": 825
1204
  },
1205
  {
1206
- "epoch": 4.048780487804878,
1207
- "grad_norm": 0.61328125,
1208
- "learning_rate": 2.1273570354634508e-05,
1209
- "loss": 0.5386,
1210
  "step": 830
1211
  },
1212
  {
1213
- "epoch": 4.073170731707317,
1214
- "grad_norm": 0.55859375,
1215
- "learning_rate": 2.023452812387555e-05,
1216
- "loss": 0.5296,
1217
  "step": 835
1218
  },
1219
  {
1220
- "epoch": 4.097560975609756,
1221
- "grad_norm": 0.62890625,
1222
- "learning_rate": 1.9218637564319696e-05,
1223
- "loss": 0.5304,
1224
  "step": 840
1225
  },
1226
  {
1227
- "epoch": 4.121951219512195,
1228
- "grad_norm": 0.57421875,
1229
- "learning_rate": 1.8226193534928604e-05,
1230
- "loss": 0.5554,
1231
  "step": 845
1232
  },
1233
  {
1234
- "epoch": 4.146341463414634,
1235
- "grad_norm": 0.578125,
1236
- "learning_rate": 1.725748408938408e-05,
1237
- "loss": 0.5376,
1238
  "step": 850
1239
  },
1240
  {
1241
- "epoch": 4.170731707317073,
1242
- "grad_norm": 0.58984375,
1243
- "learning_rate": 1.63127903924815e-05,
1244
- "loss": 0.5335,
1245
  "step": 855
1246
  },
1247
  {
1248
- "epoch": 4.195121951219512,
1249
- "grad_norm": 0.5625,
1250
- "learning_rate": 1.5392386638522482e-05,
1251
- "loss": 0.5298,
1252
  "step": 860
1253
  },
1254
  {
1255
- "epoch": 4.219512195121951,
1256
- "grad_norm": 0.60546875,
1257
- "learning_rate": 1.4496539971731026e-05,
1258
- "loss": 0.5239,
1259
  "step": 865
1260
  },
1261
  {
1262
- "epoch": 4.2439024390243905,
1263
- "grad_norm": 0.57421875,
1264
- "learning_rate": 1.3625510408715714e-05,
1265
- "loss": 0.5255,
1266
  "step": 870
1267
  },
1268
  {
1269
- "epoch": 4.2682926829268295,
1270
- "grad_norm": 0.5859375,
1271
- "learning_rate": 1.2779550763000703e-05,
1272
- "loss": 0.5382,
1273
  "step": 875
1274
  },
1275
  {
1276
- "epoch": 4.2926829268292686,
1277
- "grad_norm": 0.609375,
1278
- "learning_rate": 1.1958906571647421e-05,
1279
- "loss": 0.5469,
1280
  "step": 880
1281
  },
1282
  {
1283
- "epoch": 4.317073170731708,
1284
- "grad_norm": 0.59375,
1285
- "learning_rate": 1.1163816023988261e-05,
1286
- "loss": 0.547,
1287
  "step": 885
1288
  },
1289
  {
1290
- "epoch": 4.341463414634147,
1291
- "grad_norm": 0.56640625,
1292
- "learning_rate": 1.0394509892492833e-05,
1293
- "loss": 0.5387,
 
 
 
 
 
 
 
 
1294
  "step": 890
1295
  },
1296
  {
1297
- "epoch": 4.365853658536586,
1298
- "grad_norm": 0.58203125,
1299
- "learning_rate": 9.65121146578709e-06,
1300
- "loss": 0.5487,
1301
  "step": 895
1302
  },
1303
  {
1304
- "epoch": 4.390243902439025,
1305
- "grad_norm": 0.57421875,
1306
- "learning_rate": 8.934136483844391e-06,
1307
- "loss": 0.5306,
1308
  "step": 900
1309
  },
1310
  {
1311
- "epoch": 4.414634146341464,
1312
- "grad_norm": 0.58984375,
1313
- "learning_rate": 8.243493075367813e-06,
1314
- "loss": 0.5259,
1315
  "step": 905
1316
  },
1317
  {
1318
- "epoch": 4.439024390243903,
1319
- "grad_norm": 0.5703125,
1320
- "learning_rate": 7.579481697381363e-06,
1321
- "loss": 0.5473,
1322
  "step": 910
1323
  },
1324
  {
1325
- "epoch": 4.463414634146342,
1326
- "grad_norm": 0.55859375,
1327
- "learning_rate": 6.942295077048011e-06,
1328
- "loss": 0.5329,
1329
  "step": 915
1330
  },
1331
  {
1332
- "epoch": 4.487804878048781,
1333
- "grad_norm": 0.60546875,
1334
- "learning_rate": 6.3321181557312815e-06,
1335
- "loss": 0.551,
1336
  "step": 920
1337
  },
1338
  {
1339
- "epoch": 4.512195121951219,
1340
- "grad_norm": 0.5703125,
1341
- "learning_rate": 5.749128035316553e-06,
1342
- "loss": 0.5508,
1343
  "step": 925
1344
  },
1345
  {
1346
- "epoch": 4.536585365853659,
1347
- "grad_norm": 0.59375,
1348
- "learning_rate": 5.193493926807835e-06,
1349
- "loss": 0.5348,
1350
  "step": 930
1351
  },
1352
  {
1353
- "epoch": 4.560975609756097,
1354
- "grad_norm": 0.58984375,
1355
- "learning_rate": 4.665377101214863e-06,
1356
- "loss": 0.5168,
1357
  "step": 935
1358
  },
1359
  {
1360
- "epoch": 4.585365853658536,
1361
- "grad_norm": 0.61328125,
1362
- "learning_rate": 4.164930842744608e-06,
1363
- "loss": 0.5442,
1364
  "step": 940
1365
  },
1366
  {
1367
- "epoch": 4.609756097560975,
1368
- "grad_norm": 0.59375,
1369
- "learning_rate": 3.6923004043111444e-06,
1370
- "loss": 0.5296,
1371
  "step": 945
1372
  },
1373
  {
1374
- "epoch": 4.634146341463414,
1375
- "grad_norm": 0.546875,
1376
- "learning_rate": 3.2476229653763734e-06,
1377
- "loss": 0.5566,
1378
  "step": 950
1379
  },
1380
  {
1381
- "epoch": 4.658536585365853,
1382
- "grad_norm": 0.58984375,
1383
- "learning_rate": 2.8310275921341944e-06,
1384
- "loss": 0.5219,
1385
  "step": 955
1386
  },
1387
  {
1388
- "epoch": 4.682926829268292,
1389
- "grad_norm": 0.578125,
1390
- "learning_rate": 2.44263520004937e-06,
1391
- "loss": 0.5577,
1392
  "step": 960
1393
  },
1394
  {
1395
- "epoch": 4.7073170731707314,
1396
- "grad_norm": 0.57421875,
1397
- "learning_rate": 2.0825585187623007e-06,
1398
- "loss": 0.5235,
1399
  "step": 965
1400
  },
1401
  {
1402
- "epoch": 4.7317073170731705,
1403
- "grad_norm": 0.578125,
1404
- "learning_rate": 1.7509020593695302e-06,
1405
- "loss": 0.547,
1406
  "step": 970
1407
  },
1408
  {
1409
- "epoch": 4.7560975609756095,
1410
- "grad_norm": 0.57421875,
1411
- "learning_rate": 1.4477620840897766e-06,
1412
- "loss": 0.5331,
1413
  "step": 975
1414
  },
1415
  {
1416
- "epoch": 4.780487804878049,
1417
- "grad_norm": 0.59375,
1418
- "learning_rate": 1.1732265783241492e-06,
1419
- "loss": 0.5251,
1420
  "step": 980
1421
  },
1422
  {
1423
- "epoch": 4.804878048780488,
1424
- "grad_norm": 0.60546875,
1425
- "learning_rate": 9.273752251186096e-07,
1426
- "loss": 0.527,
1427
  "step": 985
1428
  },
1429
  {
1430
- "epoch": 4.829268292682927,
1431
- "grad_norm": 0.57421875,
1432
- "learning_rate": 7.102793820362829e-07,
1433
- "loss": 0.5353,
1434
  "step": 990
1435
  },
1436
  {
1437
- "epoch": 4.853658536585366,
1438
- "grad_norm": 0.58203125,
1439
- "learning_rate": 5.22002060446125e-07,
1440
- "loss": 0.5347,
1441
  "step": 995
1442
  },
1443
  {
1444
- "epoch": 4.878048780487805,
1445
- "grad_norm": 0.58984375,
1446
- "learning_rate": 3.6259790723409683e-07,
1447
- "loss": 0.5436,
1448
  "step": 1000
1449
  },
1450
  {
1451
- "epoch": 4.902439024390244,
1452
- "grad_norm": 0.5625,
1453
- "learning_rate": 2.3211318894205136e-07,
1454
- "loss": 0.535,
1455
  "step": 1005
1456
  },
1457
  {
1458
- "epoch": 4.926829268292683,
1459
- "grad_norm": 0.58984375,
1460
- "learning_rate": 1.3058577833905404e-07,
1461
- "loss": 0.5213,
1462
  "step": 1010
1463
  },
1464
  {
1465
- "epoch": 4.951219512195122,
1466
- "grad_norm": 0.55859375,
1467
- "learning_rate": 5.804514342889755e-08,
1468
- "loss": 0.5247,
1469
  "step": 1015
1470
  },
1471
  {
1472
- "epoch": 4.975609756097561,
1473
- "grad_norm": 0.60546875,
1474
- "learning_rate": 1.4512338897121335e-08,
1475
- "loss": 0.5333,
1476
  "step": 1020
1477
  },
1478
  {
1479
- "epoch": 5.0,
1480
- "grad_norm": 0.546875,
1481
- "learning_rate": 0.0,
1482
- "loss": 0.5144,
1483
  "step": 1025
1484
  },
1485
  {
1486
- "epoch": 5.0,
1487
- "eval_loss": 2.326845407485962,
1488
- "eval_runtime": 1.0003,
1489
- "eval_samples_per_second": 4.998,
1490
- "eval_steps_per_second": 1.999,
1491
- "step": 1025
1492
  },
1493
  {
1494
- "epoch": 5.0,
1495
- "step": 1025,
1496
- "total_flos": 1.5670950020754964e+18,
1497
- "train_loss": 1.4066738275202308,
1498
- "train_runtime": 8016.4558,
1499
- "train_samples_per_second": 2.044,
1500
- "train_steps_per_second": 0.128
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1501
  }
1502
  ],
1503
  "logging_steps": 5,
1504
- "max_steps": 1025,
1505
  "num_input_tokens_seen": 0,
1506
- "num_train_epochs": 5,
1507
  "save_steps": 100,
1508
  "stateful_callbacks": {
1509
  "TrainerControl": {
@@ -1517,7 +2194,7 @@
1517
  "attributes": {}
1518
  }
1519
  },
1520
- "total_flos": 1.5670950020754964e+18,
1521
  "train_batch_size": 4,
1522
  "trial_name": null,
1523
  "trial_params": null
 
1
  {
2
  "best_metric": null,
3
  "best_model_checkpoint": null,
4
+ "epoch": 10.0,
5
  "eval_steps": 500,
6
+ "global_step": 1480,
7
  "is_hyper_param_search": false,
8
  "is_local_process_zero": true,
9
  "is_world_process_zero": true,
10
  "log_history": [
11
  {
12
+ "epoch": 0.006756756756756757,
13
+ "grad_norm": 474.0,
14
+ "learning_rate": 1.3513513513513515e-06,
15
+ "loss": 38.1848,
16
  "step": 1
17
  },
18
  {
19
+ "epoch": 0.033783783783783786,
20
+ "grad_norm": 456.0,
21
+ "learning_rate": 6.7567567567567575e-06,
22
+ "loss": 40.5702,
23
  "step": 5
24
  },
25
  {
26
+ "epoch": 0.06756756756756757,
27
+ "grad_norm": 388.0,
28
+ "learning_rate": 1.3513513513513515e-05,
29
+ "loss": 37.3485,
30
  "step": 10
31
  },
32
  {
33
+ "epoch": 0.10135135135135136,
34
+ "grad_norm": 256.0,
35
+ "learning_rate": 2.0270270270270273e-05,
36
+ "loss": 27.8365,
37
  "step": 15
38
  },
39
  {
40
+ "epoch": 0.13513513513513514,
41
+ "grad_norm": 41.75,
42
+ "learning_rate": 2.702702702702703e-05,
43
+ "loss": 19.8335,
44
  "step": 20
45
  },
46
  {
47
+ "epoch": 0.16891891891891891,
48
+ "grad_norm": 32.25,
49
+ "learning_rate": 3.3783783783783784e-05,
50
+ "loss": 17.7565,
51
  "step": 25
52
  },
53
  {
54
+ "epoch": 0.20270270270270271,
55
+ "grad_norm": 18.125,
56
+ "learning_rate": 4.0540540540540545e-05,
57
+ "loss": 16.1239,
58
  "step": 30
59
  },
60
  {
61
+ "epoch": 0.23648648648648649,
62
+ "grad_norm": 5.78125,
63
+ "learning_rate": 4.72972972972973e-05,
64
+ "loss": 14.7405,
65
  "step": 35
66
  },
67
  {
68
+ "epoch": 0.2702702702702703,
69
+ "grad_norm": 5.375,
70
+ "learning_rate": 5.405405405405406e-05,
71
+ "loss": 13.8327,
72
  "step": 40
73
  },
74
  {
75
+ "epoch": 0.30405405405405406,
76
+ "grad_norm": 8.375,
77
+ "learning_rate": 6.0810810810810814e-05,
78
+ "loss": 13.4116,
79
  "step": 45
80
  },
81
  {
82
+ "epoch": 0.33783783783783783,
83
+ "grad_norm": 14.625,
84
+ "learning_rate": 6.756756756756757e-05,
85
+ "loss": 12.5308,
86
  "step": 50
87
  },
88
  {
89
+ "epoch": 0.3716216216216216,
90
+ "grad_norm": 31.5,
91
+ "learning_rate": 7.432432432432433e-05,
92
+ "loss": 10.6509,
93
  "step": 55
94
  },
95
  {
96
+ "epoch": 0.40540540540540543,
97
+ "grad_norm": 35.0,
98
+ "learning_rate": 8.108108108108109e-05,
99
+ "loss": 6.0675,
100
  "step": 60
101
  },
102
  {
103
+ "epoch": 0.4391891891891892,
104
+ "grad_norm": 8.75,
105
+ "learning_rate": 8.783783783783784e-05,
106
+ "loss": 2.1561,
107
  "step": 65
108
  },
109
  {
110
+ "epoch": 0.47297297297297297,
111
+ "grad_norm": 3.515625,
112
+ "learning_rate": 9.45945945945946e-05,
113
+ "loss": 1.7484,
114
  "step": 70
115
  },
116
  {
117
+ "epoch": 0.5067567567567568,
118
+ "grad_norm": 2.203125,
119
+ "learning_rate": 0.00010135135135135136,
120
+ "loss": 1.565,
121
  "step": 75
122
  },
123
  {
124
+ "epoch": 0.5405405405405406,
125
+ "grad_norm": 3.265625,
126
+ "learning_rate": 0.00010810810810810812,
127
+ "loss": 1.4507,
128
  "step": 80
129
  },
130
  {
131
+ "epoch": 0.5743243243243243,
132
+ "grad_norm": 1.484375,
133
+ "learning_rate": 0.00011486486486486487,
134
+ "loss": 1.3348,
135
  "step": 85
136
  },
137
  {
138
+ "epoch": 0.6081081081081081,
139
+ "grad_norm": 2.484375,
140
+ "learning_rate": 0.00012162162162162163,
141
+ "loss": 1.2468,
142
  "step": 90
143
  },
144
  {
145
+ "epoch": 0.6418918918918919,
146
+ "grad_norm": 4.6875,
147
+ "learning_rate": 0.0001283783783783784,
148
+ "loss": 1.2245,
149
  "step": 95
150
  },
151
  {
152
+ "epoch": 0.6756756756756757,
153
+ "grad_norm": 3.28125,
154
+ "learning_rate": 0.00013513513513513514,
155
+ "loss": 1.1992,
156
  "step": 100
157
  },
158
  {
159
+ "epoch": 0.7094594594594594,
160
+ "grad_norm": 4.40625,
161
+ "learning_rate": 0.00014189189189189188,
162
+ "loss": 1.1334,
163
  "step": 105
164
  },
165
  {
166
+ "epoch": 0.7432432432432432,
167
+ "grad_norm": 2.53125,
168
+ "learning_rate": 0.00014864864864864866,
169
+ "loss": 1.1272,
170
  "step": 110
171
  },
172
  {
173
+ "epoch": 0.777027027027027,
174
+ "grad_norm": 2.015625,
175
+ "learning_rate": 0.0001554054054054054,
176
+ "loss": 1.0732,
177
  "step": 115
178
  },
179
  {
180
+ "epoch": 0.8108108108108109,
181
+ "grad_norm": 2.125,
182
+ "learning_rate": 0.00016216216216216218,
183
+ "loss": 1.0813,
184
  "step": 120
185
  },
186
  {
187
+ "epoch": 0.8445945945945946,
188
+ "grad_norm": 3.734375,
189
+ "learning_rate": 0.00016891891891891893,
190
+ "loss": 1.0315,
191
  "step": 125
192
  },
193
  {
194
+ "epoch": 0.8783783783783784,
195
+ "grad_norm": 4.09375,
196
+ "learning_rate": 0.00017567567567567568,
197
+ "loss": 1.0636,
198
  "step": 130
199
  },
200
  {
201
+ "epoch": 0.9121621621621622,
202
+ "grad_norm": 1.390625,
203
+ "learning_rate": 0.00018243243243243245,
204
+ "loss": 0.9858,
205
  "step": 135
206
  },
207
  {
208
+ "epoch": 0.9459459459459459,
209
+ "grad_norm": 11.0,
210
+ "learning_rate": 0.0001891891891891892,
211
+ "loss": 0.9889,
212
  "step": 140
213
  },
214
  {
215
+ "epoch": 0.9797297297297297,
216
+ "grad_norm": 30.25,
217
+ "learning_rate": 0.00019594594594594594,
218
+ "loss": 1.0072,
219
  "step": 145
220
  },
221
  {
222
+ "epoch": 1.0,
223
+ "eval_loss": 2.267159938812256,
224
+ "eval_runtime": 1.0098,
225
+ "eval_samples_per_second": 4.951,
226
+ "eval_steps_per_second": 1.981,
227
+ "step": 148
228
+ },
229
+ {
230
+ "epoch": 1.0135135135135136,
231
+ "grad_norm": 3.546875,
232
+ "learning_rate": 0.00019999888744757143,
233
+ "loss": 1.0602,
234
  "step": 150
235
  },
236
  {
237
+ "epoch": 1.0472972972972974,
238
+ "grad_norm": 3.234375,
239
+ "learning_rate": 0.000199986371517049,
240
+ "loss": 0.9825,
241
  "step": 155
242
  },
243
  {
244
+ "epoch": 1.0810810810810811,
245
+ "grad_norm": 0.9609375,
246
+ "learning_rate": 0.0001999599507118322,
247
+ "loss": 0.9573,
248
  "step": 160
249
  },
250
  {
251
+ "epoch": 1.114864864864865,
252
+ "grad_norm": 3.125,
253
+ "learning_rate": 0.00019991962870620153,
254
+ "loss": 0.9675,
255
  "step": 165
256
  },
257
  {
258
+ "epoch": 1.1486486486486487,
259
+ "grad_norm": 3.34375,
260
+ "learning_rate": 0.00019986541110764565,
261
+ "loss": 0.9773,
262
  "step": 170
263
  },
264
  {
265
+ "epoch": 1.1824324324324325,
266
+ "grad_norm": 2.125,
267
+ "learning_rate": 0.00019979730545608126,
268
+ "loss": 0.9352,
269
  "step": 175
270
  },
271
  {
272
+ "epoch": 1.2162162162162162,
273
+ "grad_norm": 0.96484375,
274
+ "learning_rate": 0.00019971532122280464,
275
+ "loss": 0.9498,
276
  "step": 180
277
  },
278
  {
279
+ "epoch": 1.25,
280
+ "grad_norm": 1.25,
281
+ "learning_rate": 0.00019961946980917456,
282
+ "loss": 0.9255,
283
  "step": 185
284
  },
285
  {
286
+ "epoch": 1.2837837837837838,
287
+ "grad_norm": 1.1640625,
288
+ "learning_rate": 0.0001995097645450266,
289
+ "loss": 0.9378,
290
  "step": 190
291
  },
292
  {
293
+ "epoch": 1.3175675675675675,
294
+ "grad_norm": 0.7421875,
295
+ "learning_rate": 0.00019938622068681953,
296
+ "loss": 0.9164,
297
  "step": 195
298
  },
299
  {
300
+ "epoch": 1.3513513513513513,
301
+ "grad_norm": 1.703125,
302
+ "learning_rate": 0.0001992488554155135,
303
+ "loss": 0.8986,
304
  "step": 200
305
  },
306
  {
307
+ "epoch": 1.385135135135135,
308
+ "grad_norm": 6.59375,
309
+ "learning_rate": 0.00019909768783418086,
310
+ "loss": 0.9389,
 
 
 
 
 
 
 
 
311
  "step": 205
312
  },
313
  {
314
+ "epoch": 1.4189189189189189,
315
+ "grad_norm": 1.6171875,
316
+ "learning_rate": 0.00019893273896534936,
317
+ "loss": 0.9261,
318
  "step": 210
319
  },
320
  {
321
+ "epoch": 1.4527027027027026,
322
+ "grad_norm": 0.80859375,
323
+ "learning_rate": 0.00019875403174807882,
324
+ "loss": 0.8863,
325
  "step": 215
326
  },
327
  {
328
+ "epoch": 1.4864864864864864,
329
+ "grad_norm": 0.7890625,
330
+ "learning_rate": 0.00019856159103477086,
331
+ "loss": 0.8941,
332
  "step": 220
333
  },
334
  {
335
+ "epoch": 1.5202702702702702,
336
+ "grad_norm": 0.88671875,
337
+ "learning_rate": 0.0001983554435877128,
338
+ "loss": 0.8607,
339
  "step": 225
340
  },
341
  {
342
+ "epoch": 1.554054054054054,
343
+ "grad_norm": 0.78125,
344
+ "learning_rate": 0.00019813561807535598,
345
+ "loss": 0.9002,
346
  "step": 230
347
  },
348
  {
349
+ "epoch": 1.5878378378378377,
350
+ "grad_norm": 1.5,
351
+ "learning_rate": 0.00019790214506832868,
352
+ "loss": 0.894,
353
  "step": 235
354
  },
355
  {
356
+ "epoch": 1.6216216216216215,
357
+ "grad_norm": 1.890625,
358
+ "learning_rate": 0.00019765505703518496,
359
+ "loss": 0.8619,
360
  "step": 240
361
  },
362
  {
363
+ "epoch": 1.6554054054054053,
364
+ "grad_norm": 0.7109375,
365
+ "learning_rate": 0.0001973943883378892,
366
+ "loss": 0.8988,
367
  "step": 245
368
  },
369
  {
370
+ "epoch": 1.689189189189189,
371
+ "grad_norm": 1.2109375,
372
+ "learning_rate": 0.00019712017522703764,
373
+ "loss": 0.8851,
374
  "step": 250
375
  },
376
  {
377
+ "epoch": 1.722972972972973,
378
+ "grad_norm": 1.5,
379
+ "learning_rate": 0.00019683245583681675,
380
+ "loss": 0.8613,
381
  "step": 255
382
  },
383
  {
384
+ "epoch": 1.7567567567567568,
385
+ "grad_norm": 0.66796875,
386
+ "learning_rate": 0.00019653127017970034,
387
+ "loss": 0.8816,
388
  "step": 260
389
  },
390
  {
391
+ "epoch": 1.7905405405405406,
392
+ "grad_norm": 0.7421875,
393
+ "learning_rate": 0.00019621666014088494,
394
+ "loss": 0.9012,
395
  "step": 265
396
  },
397
  {
398
+ "epoch": 1.8243243243243243,
399
+ "grad_norm": 0.85546875,
400
+ "learning_rate": 0.00019588866947246498,
401
+ "loss": 0.8495,
402
  "step": 270
403
  },
404
  {
405
+ "epoch": 1.8581081081081081,
406
+ "grad_norm": 0.98046875,
407
+ "learning_rate": 0.00019554734378734824,
408
+ "loss": 0.8473,
409
  "step": 275
410
  },
411
  {
412
+ "epoch": 1.8918918918918919,
413
+ "grad_norm": 1.296875,
414
+ "learning_rate": 0.00019519273055291266,
415
+ "loss": 0.9128,
416
  "step": 280
417
  },
418
  {
419
+ "epoch": 1.9256756756756757,
420
+ "grad_norm": 0.7265625,
421
+ "learning_rate": 0.000194824879084405,
422
+ "loss": 0.8479,
423
  "step": 285
424
  },
425
  {
426
+ "epoch": 1.9594594594594594,
427
+ "grad_norm": 2.03125,
428
+ "learning_rate": 0.00019444384053808288,
429
+ "loss": 0.8502,
430
  "step": 290
431
  },
432
  {
433
+ "epoch": 1.9932432432432432,
434
+ "grad_norm": 6.03125,
435
+ "learning_rate": 0.00019404966790410047,
436
+ "loss": 0.8705,
437
  "step": 295
438
  },
439
  {
440
+ "epoch": 2.0,
441
+ "eval_loss": 2.1745002269744873,
442
+ "eval_runtime": 1.0121,
443
+ "eval_samples_per_second": 4.94,
444
+ "eval_steps_per_second": 1.976,
445
+ "step": 296
446
+ },
447
+ {
448
+ "epoch": 2.027027027027027,
449
+ "grad_norm": 1.6875,
450
+ "learning_rate": 0.00019364241599913924,
451
+ "loss": 0.7692,
452
  "step": 300
453
  },
454
  {
455
+ "epoch": 2.060810810810811,
456
+ "grad_norm": 0.75390625,
457
+ "learning_rate": 0.00019322214145878487,
458
+ "loss": 0.7726,
459
  "step": 305
460
  },
461
  {
462
+ "epoch": 2.0945945945945947,
463
+ "grad_norm": 1.0859375,
464
+ "learning_rate": 0.00019278890272965096,
465
+ "loss": 0.7765,
466
  "step": 310
467
  },
468
  {
469
+ "epoch": 2.1283783783783785,
470
+ "grad_norm": 4.625,
471
+ "learning_rate": 0.000192342760061251,
472
+ "loss": 0.7588,
473
  "step": 315
474
  },
475
  {
476
+ "epoch": 2.1621621621621623,
477
+ "grad_norm": 0.92578125,
478
+ "learning_rate": 0.00019188377549761963,
479
+ "loss": 0.7696,
480
  "step": 320
481
  },
482
  {
483
+ "epoch": 2.195945945945946,
484
+ "grad_norm": 0.71875,
485
+ "learning_rate": 0.00019141201286868435,
486
+ "loss": 0.7883,
487
  "step": 325
488
  },
489
  {
490
+ "epoch": 2.22972972972973,
491
+ "grad_norm": 0.6796875,
492
+ "learning_rate": 0.00019092753778138886,
493
+ "loss": 0.7835,
494
  "step": 330
495
  },
496
  {
497
+ "epoch": 2.2635135135135136,
498
+ "grad_norm": 0.8671875,
499
+ "learning_rate": 0.00019043041761056907,
500
+ "loss": 0.7848,
501
  "step": 335
502
  },
503
  {
504
+ "epoch": 2.2972972972972974,
505
+ "grad_norm": 0.765625,
506
+ "learning_rate": 0.00018992072148958368,
507
+ "loss": 0.7891,
508
  "step": 340
509
  },
510
  {
511
+ "epoch": 2.331081081081081,
512
+ "grad_norm": 0.96875,
513
+ "learning_rate": 0.00018939852030069981,
514
+ "loss": 0.7922,
515
  "step": 345
516
  },
517
  {
518
+ "epoch": 2.364864864864865,
519
+ "grad_norm": 1.328125,
520
+ "learning_rate": 0.0001888638866652356,
521
+ "loss": 0.7857,
522
  "step": 350
523
  },
524
  {
525
+ "epoch": 2.3986486486486487,
526
+ "grad_norm": 2.40625,
527
+ "learning_rate": 0.00018831689493346095,
528
+ "loss": 0.7839,
529
  "step": 355
530
  },
531
  {
532
+ "epoch": 2.4324324324324325,
533
+ "grad_norm": 2.0625,
534
+ "learning_rate": 0.00018775762117425777,
535
+ "loss": 0.8031,
536
  "step": 360
537
  },
538
  {
539
+ "epoch": 2.4662162162162162,
540
+ "grad_norm": 0.83203125,
541
+ "learning_rate": 0.00018718614316454133,
542
+ "loss": 0.7835,
543
  "step": 365
544
  },
545
  {
546
+ "epoch": 2.5,
547
+ "grad_norm": 2.5,
548
+ "learning_rate": 0.00018660254037844388,
549
+ "loss": 0.7749,
550
  "step": 370
551
  },
552
  {
553
+ "epoch": 2.5337837837837838,
554
+ "grad_norm": 0.9140625,
555
+ "learning_rate": 0.00018600689397626246,
556
+ "loss": 0.7973,
557
  "step": 375
558
  },
559
  {
560
+ "epoch": 2.5675675675675675,
561
+ "grad_norm": 1.1953125,
562
+ "learning_rate": 0.0001853992867931721,
563
+ "loss": 0.8086,
564
  "step": 380
565
  },
566
  {
567
+ "epoch": 2.6013513513513513,
568
+ "grad_norm": 0.84375,
569
+ "learning_rate": 0.00018477980332770607,
570
+ "loss": 0.7994,
571
  "step": 385
572
  },
573
  {
574
+ "epoch": 2.635135135135135,
575
+ "grad_norm": 0.921875,
576
+ "learning_rate": 0.00018414852973000503,
577
+ "loss": 0.7872,
578
  "step": 390
579
  },
580
  {
581
+ "epoch": 2.668918918918919,
582
+ "grad_norm": 0.78125,
583
+ "learning_rate": 0.00018350555378983608,
584
+ "loss": 0.7871,
585
  "step": 395
586
  },
587
  {
588
+ "epoch": 2.7027027027027026,
589
+ "grad_norm": 0.640625,
590
+ "learning_rate": 0.00018285096492438424,
591
+ "loss": 0.7913,
592
  "step": 400
593
  },
594
  {
595
+ "epoch": 2.7364864864864864,
596
+ "grad_norm": 1.109375,
597
+ "learning_rate": 0.00018218485416581726,
598
+ "loss": 0.7861,
599
  "step": 405
600
  },
601
  {
602
+ "epoch": 2.77027027027027,
603
+ "grad_norm": 0.921875,
604
+ "learning_rate": 0.00018150731414862622,
605
+ "loss": 0.7757,
 
 
 
 
 
 
 
 
606
  "step": 410
607
  },
608
  {
609
+ "epoch": 2.804054054054054,
610
+ "grad_norm": 1.15625,
611
+ "learning_rate": 0.00018081843909674276,
612
+ "loss": 0.7893,
613
  "step": 415
614
  },
615
  {
616
+ "epoch": 2.8378378378378377,
617
+ "grad_norm": 1.0390625,
618
+ "learning_rate": 0.00018011832481043576,
619
+ "loss": 0.795,
620
  "step": 420
621
  },
622
  {
623
+ "epoch": 2.8716216216216215,
624
+ "grad_norm": 0.65234375,
625
+ "learning_rate": 0.0001794070686529886,
626
+ "loss": 0.7933,
627
  "step": 425
628
  },
629
  {
630
+ "epoch": 2.9054054054054053,
631
+ "grad_norm": 0.625,
632
+ "learning_rate": 0.000178684769537159,
633
+ "loss": 0.7832,
634
  "step": 430
635
  },
636
  {
637
+ "epoch": 2.939189189189189,
638
+ "grad_norm": 0.69921875,
639
+ "learning_rate": 0.0001779515279114236,
640
+ "loss": 0.7833,
641
  "step": 435
642
  },
643
  {
644
+ "epoch": 2.972972972972973,
645
+ "grad_norm": 0.78125,
646
+ "learning_rate": 0.00017720744574600863,
647
+ "loss": 0.7957,
648
  "step": 440
649
  },
650
  {
651
+ "epoch": 3.0,
652
+ "eval_loss": 2.191443920135498,
653
+ "eval_runtime": 1.0106,
654
+ "eval_samples_per_second": 4.947,
655
+ "eval_steps_per_second": 1.979,
656
+ "step": 444
657
+ },
658
+ {
659
+ "epoch": 3.0067567567567566,
660
+ "grad_norm": 0.71484375,
661
+ "learning_rate": 0.00017645262651870926,
662
+ "loss": 0.7603,
663
  "step": 445
664
  },
665
  {
666
+ "epoch": 3.0405405405405403,
667
+ "grad_norm": 0.64453125,
668
+ "learning_rate": 0.0001756871752004992,
669
+ "loss": 0.6923,
670
  "step": 450
671
  },
672
  {
673
+ "epoch": 3.074324324324324,
674
+ "grad_norm": 0.7109375,
675
+ "learning_rate": 0.0001749111982409325,
676
+ "loss": 0.6958,
677
  "step": 455
678
  },
679
  {
680
+ "epoch": 3.108108108108108,
681
+ "grad_norm": 0.69921875,
682
+ "learning_rate": 0.00017412480355334005,
683
+ "loss": 0.6803,
684
  "step": 460
685
  },
686
  {
687
+ "epoch": 3.141891891891892,
688
+ "grad_norm": 0.84765625,
689
+ "learning_rate": 0.00017332810049982208,
690
+ "loss": 0.7161,
691
  "step": 465
692
  },
693
  {
694
+ "epoch": 3.175675675675676,
695
+ "grad_norm": 1.0234375,
696
+ "learning_rate": 0.00017252119987603973,
697
+ "loss": 0.68,
698
  "step": 470
699
  },
700
  {
701
+ "epoch": 3.2094594594594597,
702
+ "grad_norm": 0.70703125,
703
+ "learning_rate": 0.00017170421389580667,
704
+ "loss": 0.6976,
705
  "step": 475
706
  },
707
  {
708
+ "epoch": 3.2432432432432434,
709
+ "grad_norm": 0.765625,
710
+ "learning_rate": 0.00017087725617548385,
711
+ "loss": 0.6836,
712
  "step": 480
713
  },
714
  {
715
+ "epoch": 3.277027027027027,
716
+ "grad_norm": 0.69921875,
717
+ "learning_rate": 0.00017004044171817925,
718
+ "loss": 0.7073,
719
  "step": 485
720
  },
721
  {
722
+ "epoch": 3.310810810810811,
723
+ "grad_norm": 0.72265625,
724
+ "learning_rate": 0.00016919388689775464,
725
+ "loss": 0.7111,
726
  "step": 490
727
  },
728
  {
729
+ "epoch": 3.3445945945945947,
730
+ "grad_norm": 0.6953125,
731
+ "learning_rate": 0.00016833770944264153,
732
+ "loss": 0.716,
733
  "step": 495
734
  },
735
  {
736
+ "epoch": 3.3783783783783785,
737
+ "grad_norm": 0.8046875,
738
+ "learning_rate": 0.00016747202841946928,
739
+ "loss": 0.7096,
740
  "step": 500
741
  },
742
  {
743
+ "epoch": 3.4121621621621623,
744
+ "grad_norm": 0.859375,
745
+ "learning_rate": 0.00016659696421650645,
746
+ "loss": 0.6776,
747
  "step": 505
748
  },
749
  {
750
+ "epoch": 3.445945945945946,
751
+ "grad_norm": 1.40625,
752
+ "learning_rate": 0.00016571263852691888,
753
+ "loss": 0.7246,
754
  "step": 510
755
  },
756
  {
757
+ "epoch": 3.47972972972973,
758
+ "grad_norm": 1.09375,
759
+ "learning_rate": 0.00016481917433184607,
760
+ "loss": 0.706,
761
  "step": 515
762
  },
763
  {
764
+ "epoch": 3.5135135135135136,
765
+ "grad_norm": 0.8828125,
766
+ "learning_rate": 0.0001639166958832985,
767
+ "loss": 0.694,
768
  "step": 520
769
  },
770
  {
771
+ "epoch": 3.5472972972972974,
772
+ "grad_norm": 0.921875,
773
+ "learning_rate": 0.00016300532868687806,
774
+ "loss": 0.7063,
775
  "step": 525
776
  },
777
  {
778
+ "epoch": 3.581081081081081,
779
+ "grad_norm": 0.84375,
780
+ "learning_rate": 0.0001620851994843244,
781
+ "loss": 0.7071,
782
  "step": 530
783
  },
784
  {
785
+ "epoch": 3.614864864864865,
786
+ "grad_norm": 0.8671875,
787
+ "learning_rate": 0.00016115643623588915,
788
+ "loss": 0.7343,
789
  "step": 535
790
  },
791
  {
792
+ "epoch": 3.6486486486486487,
793
+ "grad_norm": 1.109375,
794
+ "learning_rate": 0.00016021916810254097,
795
+ "loss": 0.6788,
796
  "step": 540
797
  },
798
  {
799
+ "epoch": 3.6824324324324325,
800
+ "grad_norm": 0.75390625,
801
+ "learning_rate": 0.00015927352542800317,
802
+ "loss": 0.7099,
803
  "step": 545
804
  },
805
  {
806
+ "epoch": 3.7162162162162162,
807
+ "grad_norm": 0.7265625,
808
+ "learning_rate": 0.00015831963972062733,
809
+ "loss": 0.7176,
810
  "step": 550
811
  },
812
  {
813
+ "epoch": 3.75,
814
+ "grad_norm": 0.7109375,
815
+ "learning_rate": 0.0001573576436351046,
816
+ "loss": 0.7099,
817
  "step": 555
818
  },
819
  {
820
+ "epoch": 3.7837837837837838,
821
+ "grad_norm": 0.703125,
822
+ "learning_rate": 0.0001563876709540178,
823
+ "loss": 0.7117,
824
  "step": 560
825
  },
826
  {
827
+ "epoch": 3.8175675675675675,
828
+ "grad_norm": 0.8203125,
829
+ "learning_rate": 0.00015540985656923645,
830
+ "loss": 0.7227,
831
  "step": 565
832
  },
833
  {
834
+ "epoch": 3.8513513513513513,
835
+ "grad_norm": 0.703125,
836
+ "learning_rate": 0.0001544243364631579,
837
+ "loss": 0.7077,
838
  "step": 570
839
  },
840
  {
841
+ "epoch": 3.885135135135135,
842
+ "grad_norm": 1.15625,
843
+ "learning_rate": 0.00015343124768979637,
844
+ "loss": 0.7162,
845
  "step": 575
846
  },
847
  {
848
+ "epoch": 3.918918918918919,
849
+ "grad_norm": 0.921875,
850
+ "learning_rate": 0.00015243072835572318,
851
+ "loss": 0.7164,
852
  "step": 580
853
  },
854
  {
855
+ "epoch": 3.9527027027027026,
856
+ "grad_norm": 2.703125,
857
+ "learning_rate": 0.0001514229176008607,
858
+ "loss": 0.7312,
859
  "step": 585
860
  },
861
  {
862
+ "epoch": 3.9864864864864864,
863
+ "grad_norm": 0.78125,
864
+ "learning_rate": 0.00015040795557913245,
865
+ "loss": 0.731,
866
  "step": 590
867
  },
868
  {
869
+ "epoch": 4.0,
870
+ "eval_loss": 2.2511258125305176,
871
+ "eval_runtime": 1.0139,
872
+ "eval_samples_per_second": 4.932,
873
+ "eval_steps_per_second": 1.973,
874
+ "step": 592
875
+ },
876
+ {
877
+ "epoch": 4.02027027027027,
878
+ "grad_norm": 0.88671875,
879
+ "learning_rate": 0.00014938598343897214,
880
+ "loss": 0.653,
881
  "step": 595
882
  },
883
  {
884
+ "epoch": 4.054054054054054,
885
+ "grad_norm": 0.8828125,
886
+ "learning_rate": 0.00014835714330369446,
887
+ "loss": 0.6058,
888
  "step": 600
889
  },
890
  {
891
+ "epoch": 4.087837837837838,
892
+ "grad_norm": 0.9609375,
893
+ "learning_rate": 0.00014732157825173044,
894
+ "loss": 0.6245,
895
  "step": 605
896
  },
897
  {
898
+ "epoch": 4.121621621621622,
899
+ "grad_norm": 0.94921875,
900
+ "learning_rate": 0.0001462794322967299,
901
+ "loss": 0.6074,
902
  "step": 610
903
  },
904
  {
905
+ "epoch": 4.155405405405405,
906
+ "grad_norm": 0.8515625,
907
+ "learning_rate": 0.00014523085036753354,
908
+ "loss": 0.6067,
 
 
 
 
 
 
 
 
909
  "step": 615
910
  },
911
  {
912
+ "epoch": 4.1891891891891895,
913
+ "grad_norm": 0.90234375,
914
+ "learning_rate": 0.00014417597828801832,
915
+ "loss": 0.608,
916
  "step": 620
917
  },
918
  {
919
+ "epoch": 4.222972972972973,
920
+ "grad_norm": 0.81640625,
921
+ "learning_rate": 0.00014311496275681783,
922
+ "loss": 0.6079,
923
  "step": 625
924
  },
925
  {
926
+ "epoch": 4.256756756756757,
927
  "grad_norm": 0.62890625,
928
+ "learning_rate": 0.00014204795132692144,
929
+ "loss": 0.6234,
930
  "step": 630
931
  },
932
  {
933
+ "epoch": 4.29054054054054,
934
+ "grad_norm": 0.67578125,
935
+ "learning_rate": 0.00014097509238515432,
936
+ "loss": 0.6139,
937
  "step": 635
938
  },
939
  {
940
+ "epoch": 4.324324324324325,
941
+ "grad_norm": 0.62109375,
942
+ "learning_rate": 0.00013989653513154165,
943
+ "loss": 0.6203,
944
  "step": 640
945
  },
946
  {
947
+ "epoch": 4.358108108108108,
948
+ "grad_norm": 0.66796875,
949
+ "learning_rate": 0.00013881242955855974,
950
+ "loss": 0.6053,
951
  "step": 645
952
  },
953
  {
954
+ "epoch": 4.391891891891892,
955
+ "grad_norm": 0.72265625,
956
+ "learning_rate": 0.000137722926430277,
957
+ "loss": 0.6032,
958
  "step": 650
959
  },
960
  {
961
+ "epoch": 4.425675675675675,
962
+ "grad_norm": 0.93359375,
963
+ "learning_rate": 0.00013662817726138728,
964
+ "loss": 0.6301,
965
  "step": 655
966
  },
967
  {
968
+ "epoch": 4.45945945945946,
969
+ "grad_norm": 0.7265625,
970
+ "learning_rate": 0.00013552833429613938,
971
+ "loss": 0.6211,
972
  "step": 660
973
  },
974
  {
975
+ "epoch": 4.493243243243243,
976
+ "grad_norm": 0.65625,
977
+ "learning_rate": 0.0001344235504871645,
978
+ "loss": 0.625,
979
  "step": 665
980
  },
981
  {
982
+ "epoch": 4.527027027027027,
983
+ "grad_norm": 0.828125,
984
+ "learning_rate": 0.00013331397947420576,
985
+ "loss": 0.6075,
986
  "step": 670
987
  },
988
  {
989
+ "epoch": 4.5608108108108105,
990
+ "grad_norm": 0.6875,
991
+ "learning_rate": 0.00013219977556275163,
992
+ "loss": 0.6016,
993
  "step": 675
994
  },
995
  {
996
+ "epoch": 4.594594594594595,
997
+ "grad_norm": 0.70703125,
998
+ "learning_rate": 0.00013108109370257712,
999
+ "loss": 0.6126,
1000
  "step": 680
1001
  },
1002
  {
1003
+ "epoch": 4.628378378378378,
1004
+ "grad_norm": 0.62109375,
1005
+ "learning_rate": 0.0001299580894661953,
1006
+ "loss": 0.6229,
1007
  "step": 685
1008
  },
1009
  {
1010
+ "epoch": 4.662162162162162,
1011
+ "grad_norm": 0.66015625,
1012
+ "learning_rate": 0.0001288309190272222,
1013
+ "loss": 0.6368,
1014
  "step": 690
1015
  },
1016
  {
1017
+ "epoch": 4.695945945945946,
1018
+ "grad_norm": 0.640625,
1019
+ "learning_rate": 0.00012769973913865794,
1020
+ "loss": 0.6246,
1021
  "step": 695
1022
  },
1023
  {
1024
+ "epoch": 4.72972972972973,
1025
+ "grad_norm": 0.87890625,
1026
+ "learning_rate": 0.00012656470711108764,
1027
+ "loss": 0.6279,
1028
  "step": 700
1029
  },
1030
  {
1031
+ "epoch": 4.763513513513513,
1032
+ "grad_norm": 0.70703125,
1033
+ "learning_rate": 0.00012542598079080456,
1034
+ "loss": 0.6183,
1035
  "step": 705
1036
  },
1037
  {
1038
+ "epoch": 4.797297297297297,
1039
+ "grad_norm": 0.76953125,
1040
+ "learning_rate": 0.0001242837185378587,
1041
+ "loss": 0.6299,
1042
  "step": 710
1043
  },
1044
  {
1045
+ "epoch": 4.831081081081081,
1046
+ "grad_norm": 0.96484375,
1047
+ "learning_rate": 0.00012313807920403419,
1048
+ "loss": 0.6327,
1049
  "step": 715
1050
  },
1051
  {
1052
+ "epoch": 4.864864864864865,
1053
+ "grad_norm": 0.796875,
1054
+ "learning_rate": 0.00012198922211075778,
1055
+ "loss": 0.6425,
1056
  "step": 720
1057
  },
1058
  {
1059
+ "epoch": 4.898648648648649,
1060
+ "grad_norm": 0.71875,
1061
+ "learning_rate": 0.00012083730702694291,
1062
+ "loss": 0.6226,
1063
  "step": 725
1064
  },
1065
  {
1066
+ "epoch": 4.9324324324324325,
1067
+ "grad_norm": 0.7109375,
1068
+ "learning_rate": 0.00011968249414677055,
1069
+ "loss": 0.6108,
1070
  "step": 730
1071
  },
1072
  {
1073
+ "epoch": 4.966216216216216,
1074
+ "grad_norm": 0.71484375,
1075
+ "learning_rate": 0.00011852494406741165,
1076
+ "loss": 0.6359,
1077
  "step": 735
1078
  },
1079
  {
1080
+ "epoch": 5.0,
1081
+ "grad_norm": 0.703125,
1082
+ "learning_rate": 0.00011736481776669306,
1083
+ "loss": 0.634,
1084
  "step": 740
1085
  },
1086
  {
1087
+ "epoch": 5.0,
1088
+ "eval_loss": 2.340895175933838,
1089
+ "eval_runtime": 1.0117,
1090
+ "eval_samples_per_second": 4.942,
1091
+ "eval_steps_per_second": 1.977,
1092
+ "step": 740
1093
+ },
1094
+ {
1095
+ "epoch": 5.033783783783784,
1096
+ "grad_norm": 0.9453125,
1097
+ "learning_rate": 0.00011620227658071087,
1098
+ "loss": 0.53,
1099
  "step": 745
1100
  },
1101
  {
1102
+ "epoch": 5.0675675675675675,
1103
+ "grad_norm": 0.6796875,
1104
+ "learning_rate": 0.00011503748218139369,
1105
+ "loss": 0.5224,
1106
  "step": 750
1107
  },
1108
  {
1109
+ "epoch": 5.101351351351352,
1110
+ "grad_norm": 0.75390625,
1111
+ "learning_rate": 0.00011387059655401932,
1112
+ "loss": 0.5205,
1113
  "step": 755
1114
  },
1115
  {
1116
+ "epoch": 5.135135135135135,
1117
+ "grad_norm": 0.80859375,
1118
+ "learning_rate": 0.00011270178197468789,
1119
+ "loss": 0.5171,
1120
  "step": 760
1121
  },
1122
  {
1123
+ "epoch": 5.168918918918919,
1124
+ "grad_norm": 0.79296875,
1125
+ "learning_rate": 0.00011153120098775434,
1126
+ "loss": 0.5174,
1127
  "step": 765
1128
  },
1129
  {
1130
+ "epoch": 5.202702702702703,
1131
+ "grad_norm": 0.765625,
1132
+ "learning_rate": 0.00011035901638322392,
1133
+ "loss": 0.5288,
1134
  "step": 770
1135
  },
1136
  {
1137
+ "epoch": 5.236486486486487,
1138
+ "grad_norm": 0.921875,
1139
+ "learning_rate": 0.00010918539117411333,
1140
+ "loss": 0.5389,
1141
  "step": 775
1142
  },
1143
  {
1144
+ "epoch": 5.27027027027027,
1145
+ "grad_norm": 0.796875,
1146
+ "learning_rate": 0.00010801048857378071,
1147
+ "loss": 0.5254,
1148
  "step": 780
1149
  },
1150
  {
1151
+ "epoch": 5.304054054054054,
1152
+ "grad_norm": 0.734375,
1153
+ "learning_rate": 0.00010683447197322817,
1154
+ "loss": 0.535,
1155
  "step": 785
1156
  },
1157
  {
1158
+ "epoch": 5.337837837837838,
1159
+ "grad_norm": 0.76171875,
1160
+ "learning_rate": 0.00010565750491837925,
1161
+ "loss": 0.5152,
1162
  "step": 790
1163
  },
1164
  {
1165
+ "epoch": 5.371621621621622,
1166
+ "grad_norm": 0.7734375,
1167
+ "learning_rate": 0.00010447975108733492,
1168
+ "loss": 0.5387,
1169
  "step": 795
1170
  },
1171
  {
1172
+ "epoch": 5.405405405405405,
1173
+ "grad_norm": 0.8515625,
1174
+ "learning_rate": 0.00010330137426761135,
1175
+ "loss": 0.5245,
1176
  "step": 800
1177
  },
1178
  {
1179
+ "epoch": 5.4391891891891895,
1180
+ "grad_norm": 0.69140625,
1181
+ "learning_rate": 0.00010212253833336237,
1182
+ "loss": 0.5284,
1183
  "step": 805
1184
  },
1185
  {
1186
+ "epoch": 5.472972972972973,
1187
+ "grad_norm": 0.7890625,
1188
+ "learning_rate": 0.00010094340722258969,
1189
+ "loss": 0.5501,
1190
  "step": 810
1191
  },
1192
  {
1193
+ "epoch": 5.506756756756757,
1194
+ "grad_norm": 0.78515625,
1195
+ "learning_rate": 9.976414491434463e-05,
1196
+ "loss": 0.5526,
1197
  "step": 815
1198
  },
1199
  {
1200
+ "epoch": 5.54054054054054,
1201
+ "grad_norm": 0.76953125,
1202
+ "learning_rate": 9.858491540592382e-05,
1203
+ "loss": 0.5313,
 
 
 
 
 
 
 
 
1204
  "step": 820
1205
  },
1206
  {
1207
+ "epoch": 5.574324324324325,
1208
+ "grad_norm": 0.828125,
1209
+ "learning_rate": 9.740588269006246e-05,
1210
+ "loss": 0.5398,
1211
  "step": 825
1212
  },
1213
  {
1214
+ "epoch": 5.608108108108108,
1215
+ "grad_norm": 0.7578125,
1216
+ "learning_rate": 9.622721073212832e-05,
1217
+ "loss": 0.5228,
1218
  "step": 830
1219
  },
1220
  {
1221
+ "epoch": 5.641891891891892,
1222
+ "grad_norm": 0.796875,
1223
+ "learning_rate": 9.504906344731932e-05,
1224
+ "loss": 0.5509,
1225
  "step": 835
1226
  },
1227
  {
1228
+ "epoch": 5.675675675675675,
1229
+ "grad_norm": 0.8125,
1230
+ "learning_rate": 9.38716046778684e-05,
1231
+ "loss": 0.547,
1232
  "step": 840
1233
  },
1234
  {
1235
+ "epoch": 5.70945945945946,
1236
+ "grad_norm": 0.7109375,
1237
+ "learning_rate": 9.269499817025814e-05,
1238
+ "loss": 0.5351,
1239
  "step": 845
1240
  },
1241
  {
1242
+ "epoch": 5.743243243243243,
1243
+ "grad_norm": 0.75390625,
1244
+ "learning_rate": 9.151940755244912e-05,
1245
+ "loss": 0.5352,
1246
  "step": 850
1247
  },
1248
  {
1249
+ "epoch": 5.777027027027027,
1250
+ "grad_norm": 0.69921875,
1251
+ "learning_rate": 9.034499631112437e-05,
1252
+ "loss": 0.5371,
1253
  "step": 855
1254
  },
1255
  {
1256
+ "epoch": 5.8108108108108105,
1257
+ "grad_norm": 0.6796875,
1258
+ "learning_rate": 8.917192776895382e-05,
1259
+ "loss": 0.5459,
1260
  "step": 860
1261
  },
1262
  {
1263
+ "epoch": 5.844594594594595,
1264
+ "grad_norm": 0.80859375,
1265
+ "learning_rate": 8.800036506188129e-05,
1266
+ "loss": 0.5324,
1267
  "step": 865
1268
  },
1269
  {
1270
+ "epoch": 5.878378378378378,
1271
+ "grad_norm": 0.7578125,
1272
+ "learning_rate": 8.683047111643763e-05,
1273
+ "loss": 0.5478,
1274
  "step": 870
1275
  },
1276
  {
1277
+ "epoch": 5.912162162162162,
1278
+ "grad_norm": 0.82421875,
1279
+ "learning_rate": 8.566240862708274e-05,
1280
+ "loss": 0.5465,
1281
  "step": 875
1282
  },
1283
  {
1284
+ "epoch": 5.945945945945946,
1285
+ "grad_norm": 0.90234375,
1286
+ "learning_rate": 8.449634003358022e-05,
1287
+ "loss": 0.5341,
1288
  "step": 880
1289
  },
1290
  {
1291
+ "epoch": 5.97972972972973,
1292
+ "grad_norm": 0.79296875,
1293
+ "learning_rate": 8.33324274984071e-05,
1294
+ "loss": 0.5418,
1295
  "step": 885
1296
  },
1297
  {
1298
+ "epoch": 6.0,
1299
+ "eval_loss": 2.4840505123138428,
1300
+ "eval_runtime": 1.0123,
1301
+ "eval_samples_per_second": 4.939,
1302
+ "eval_steps_per_second": 1.976,
1303
+ "step": 888
1304
+ },
1305
+ {
1306
+ "epoch": 6.013513513513513,
1307
+ "grad_norm": 0.8515625,
1308
+ "learning_rate": 8.217083288420241e-05,
1309
+ "loss": 0.5205,
1310
  "step": 890
1311
  },
1312
  {
1313
+ "epoch": 6.047297297297297,
1314
+ "grad_norm": 0.8828125,
1315
+ "learning_rate": 8.101171773125716e-05,
1316
+ "loss": 0.4562,
1317
  "step": 895
1318
  },
1319
  {
1320
+ "epoch": 6.081081081081081,
1321
+ "grad_norm": 0.83984375,
1322
+ "learning_rate": 7.985524323504948e-05,
1323
+ "loss": 0.4565,
1324
  "step": 900
1325
  },
1326
  {
1327
+ "epoch": 6.114864864864865,
1328
+ "grad_norm": 0.7578125,
1329
+ "learning_rate": 7.870157022382735e-05,
1330
+ "loss": 0.4372,
1331
  "step": 905
1332
  },
1333
  {
1334
+ "epoch": 6.148648648648648,
1335
+ "grad_norm": 0.76171875,
1336
+ "learning_rate": 7.755085913624274e-05,
1337
+ "loss": 0.4503,
1338
  "step": 910
1339
  },
1340
  {
1341
+ "epoch": 6.1824324324324325,
1342
+ "grad_norm": 0.7578125,
1343
+ "learning_rate": 7.640326999903967e-05,
1344
+ "loss": 0.449,
1345
  "step": 915
1346
  },
1347
  {
1348
+ "epoch": 6.216216216216216,
1349
+ "grad_norm": 0.734375,
1350
+ "learning_rate": 7.525896240479976e-05,
1351
+ "loss": 0.4404,
1352
  "step": 920
1353
  },
1354
  {
1355
+ "epoch": 6.25,
1356
+ "grad_norm": 0.8359375,
1357
+ "learning_rate": 7.411809548974792e-05,
1358
+ "loss": 0.4534,
1359
  "step": 925
1360
  },
1361
  {
1362
+ "epoch": 6.283783783783784,
1363
+ "grad_norm": 0.76953125,
1364
+ "learning_rate": 7.29808279116218e-05,
1365
+ "loss": 0.4564,
1366
  "step": 930
1367
  },
1368
  {
1369
+ "epoch": 6.3175675675675675,
1370
+ "grad_norm": 0.7890625,
1371
+ "learning_rate": 7.184731782760746e-05,
1372
+ "loss": 0.4515,
1373
  "step": 935
1374
  },
1375
  {
1376
+ "epoch": 6.351351351351352,
1377
+ "grad_norm": 0.91796875,
1378
+ "learning_rate": 7.071772287234497e-05,
1379
+ "loss": 0.4481,
1380
  "step": 940
1381
  },
1382
  {
1383
+ "epoch": 6.385135135135135,
1384
+ "grad_norm": 0.8046875,
1385
+ "learning_rate": 6.959220013600641e-05,
1386
+ "loss": 0.45,
1387
  "step": 945
1388
  },
1389
  {
1390
+ "epoch": 6.418918918918919,
1391
+ "grad_norm": 0.8125,
1392
+ "learning_rate": 6.847090614244977e-05,
1393
+ "loss": 0.4494,
1394
  "step": 950
1395
  },
1396
  {
1397
+ "epoch": 6.452702702702703,
1398
+ "grad_norm": 0.9140625,
1399
+ "learning_rate": 6.735399682745145e-05,
1400
+ "loss": 0.4462,
1401
  "step": 955
1402
  },
1403
  {
1404
+ "epoch": 6.486486486486487,
1405
+ "grad_norm": 0.83203125,
1406
+ "learning_rate": 6.624162751702076e-05,
1407
+ "loss": 0.4597,
1408
  "step": 960
1409
  },
1410
  {
1411
+ "epoch": 6.52027027027027,
1412
+ "grad_norm": 0.83203125,
1413
+ "learning_rate": 6.513395290579901e-05,
1414
+ "loss": 0.4488,
1415
  "step": 965
1416
  },
1417
  {
1418
+ "epoch": 6.554054054054054,
1419
+ "grad_norm": 0.78125,
1420
+ "learning_rate": 6.403112703554643e-05,
1421
+ "loss": 0.4635,
1422
  "step": 970
1423
  },
1424
  {
1425
+ "epoch": 6.587837837837838,
1426
+ "grad_norm": 0.84375,
1427
+ "learning_rate": 6.293330327372005e-05,
1428
+ "loss": 0.4549,
1429
  "step": 975
1430
  },
1431
  {
1432
+ "epoch": 6.621621621621622,
1433
+ "grad_norm": 0.83984375,
1434
+ "learning_rate": 6.184063429214515e-05,
1435
+ "loss": 0.4614,
1436
  "step": 980
1437
  },
1438
  {
1439
+ "epoch": 6.655405405405405,
1440
+ "grad_norm": 0.76171875,
1441
+ "learning_rate": 6.0753272045783625e-05,
1442
+ "loss": 0.449,
1443
  "step": 985
1444
  },
1445
  {
1446
+ "epoch": 6.6891891891891895,
1447
+ "grad_norm": 0.75390625,
1448
+ "learning_rate": 5.967136775160187e-05,
1449
+ "loss": 0.4631,
1450
  "step": 990
1451
  },
1452
  {
1453
+ "epoch": 6.722972972972973,
1454
+ "grad_norm": 0.7890625,
1455
+ "learning_rate": 5.859507186754146e-05,
1456
+ "loss": 0.4672,
1457
  "step": 995
1458
  },
1459
  {
1460
+ "epoch": 6.756756756756757,
1461
+ "grad_norm": 0.82421875,
1462
+ "learning_rate": 5.752453407159522e-05,
1463
+ "loss": 0.4552,
1464
  "step": 1000
1465
  },
1466
  {
1467
+ "epoch": 6.79054054054054,
1468
+ "grad_norm": 0.84765625,
1469
+ "learning_rate": 5.645990324099197e-05,
1470
+ "loss": 0.4807,
1471
  "step": 1005
1472
  },
1473
  {
1474
+ "epoch": 6.824324324324325,
1475
+ "grad_norm": 0.73828125,
1476
+ "learning_rate": 5.540132743149242e-05,
1477
+ "loss": 0.4661,
1478
  "step": 1010
1479
  },
1480
  {
1481
+ "epoch": 6.858108108108108,
1482
+ "grad_norm": 0.75,
1483
+ "learning_rate": 5.434895385679937e-05,
1484
+ "loss": 0.4595,
1485
  "step": 1015
1486
  },
1487
  {
1488
+ "epoch": 6.891891891891892,
1489
+ "grad_norm": 0.75,
1490
+ "learning_rate": 5.33029288680852e-05,
1491
+ "loss": 0.4457,
1492
  "step": 1020
1493
  },
1494
  {
1495
+ "epoch": 6.925675675675675,
1496
+ "grad_norm": 0.8046875,
1497
+ "learning_rate": 5.226339793363898e-05,
1498
+ "loss": 0.4594,
1499
  "step": 1025
1500
  },
1501
  {
1502
+ "epoch": 6.95945945945946,
1503
+ "grad_norm": 0.87890625,
1504
+ "learning_rate": 5.123050561863657e-05,
1505
+ "loss": 0.479,
1506
+ "step": 1030
 
1507
  },
1508
  {
1509
+ "epoch": 6.993243243243243,
1510
+ "grad_norm": 0.8984375,
1511
+ "learning_rate": 5.020439556503629e-05,
1512
+ "loss": 0.4578,
1513
+ "step": 1035
1514
+ },
1515
+ {
1516
+ "epoch": 7.0,
1517
+ "eval_loss": 2.6821556091308594,
1518
+ "eval_runtime": 1.013,
1519
+ "eval_samples_per_second": 4.936,
1520
+ "eval_steps_per_second": 1.974,
1521
+ "step": 1036
1522
+ },
1523
+ {
1524
+ "epoch": 7.027027027027027,
1525
+ "grad_norm": 0.8125,
1526
+ "learning_rate": 4.918521047160308e-05,
1527
+ "loss": 0.3997,
1528
+ "step": 1040
1529
+ },
1530
+ {
1531
+ "epoch": 7.0608108108108105,
1532
+ "grad_norm": 0.8671875,
1533
+ "learning_rate": 4.817309207406346e-05,
1534
+ "loss": 0.392,
1535
+ "step": 1045
1536
+ },
1537
+ {
1538
+ "epoch": 7.094594594594595,
1539
+ "grad_norm": 0.8359375,
1540
+ "learning_rate": 4.716818112539485e-05,
1541
+ "loss": 0.3883,
1542
+ "step": 1050
1543
+ },
1544
+ {
1545
+ "epoch": 7.128378378378378,
1546
+ "grad_norm": 0.7421875,
1547
+ "learning_rate": 4.617061737625139e-05,
1548
+ "loss": 0.3929,
1549
+ "step": 1055
1550
+ },
1551
+ {
1552
+ "epoch": 7.162162162162162,
1553
+ "grad_norm": 0.8125,
1554
+ "learning_rate": 4.518053955552903e-05,
1555
+ "loss": 0.3921,
1556
+ "step": 1060
1557
+ },
1558
+ {
1559
+ "epoch": 7.195945945945946,
1560
+ "grad_norm": 0.80078125,
1561
+ "learning_rate": 4.419808535107287e-05,
1562
+ "loss": 0.3865,
1563
+ "step": 1065
1564
+ },
1565
+ {
1566
+ "epoch": 7.22972972972973,
1567
+ "grad_norm": 0.79296875,
1568
+ "learning_rate": 4.322339139052921e-05,
1569
+ "loss": 0.3812,
1570
+ "step": 1070
1571
+ },
1572
+ {
1573
+ "epoch": 7.263513513513513,
1574
+ "grad_norm": 0.8046875,
1575
+ "learning_rate": 4.2256593222345185e-05,
1576
+ "loss": 0.3915,
1577
+ "step": 1075
1578
+ },
1579
+ {
1580
+ "epoch": 7.297297297297297,
1581
+ "grad_norm": 0.76953125,
1582
+ "learning_rate": 4.129782529691815e-05,
1583
+ "loss": 0.396,
1584
+ "step": 1080
1585
+ },
1586
+ {
1587
+ "epoch": 7.331081081081081,
1588
+ "grad_norm": 0.79296875,
1589
+ "learning_rate": 4.034722094789809e-05,
1590
+ "loss": 0.408,
1591
+ "step": 1085
1592
+ },
1593
+ {
1594
+ "epoch": 7.364864864864865,
1595
+ "grad_norm": 0.75390625,
1596
+ "learning_rate": 3.9404912373645185e-05,
1597
+ "loss": 0.3819,
1598
+ "step": 1090
1599
+ },
1600
+ {
1601
+ "epoch": 7.398648648648648,
1602
+ "grad_norm": 0.80078125,
1603
+ "learning_rate": 3.8471030618845375e-05,
1604
+ "loss": 0.3925,
1605
+ "step": 1095
1606
+ },
1607
+ {
1608
+ "epoch": 7.4324324324324325,
1609
+ "grad_norm": 0.80859375,
1610
+ "learning_rate": 3.7545705556286126e-05,
1611
+ "loss": 0.3932,
1612
+ "step": 1100
1613
+ },
1614
+ {
1615
+ "epoch": 7.466216216216216,
1616
+ "grad_norm": 0.79296875,
1617
+ "learning_rate": 3.662906586879542e-05,
1618
+ "loss": 0.3964,
1619
+ "step": 1105
1620
+ },
1621
+ {
1622
+ "epoch": 7.5,
1623
+ "grad_norm": 0.81640625,
1624
+ "learning_rate": 3.5721239031346066e-05,
1625
+ "loss": 0.4121,
1626
+ "step": 1110
1627
+ },
1628
+ {
1629
+ "epoch": 7.533783783783784,
1630
+ "grad_norm": 0.76953125,
1631
+ "learning_rate": 3.48223512933282e-05,
1632
+ "loss": 0.3911,
1633
+ "step": 1115
1634
+ },
1635
+ {
1636
+ "epoch": 7.5675675675675675,
1637
+ "grad_norm": 0.76171875,
1638
+ "learning_rate": 3.393252766099187e-05,
1639
+ "loss": 0.3941,
1640
+ "step": 1120
1641
+ },
1642
+ {
1643
+ "epoch": 7.601351351351351,
1644
+ "grad_norm": 0.80078125,
1645
+ "learning_rate": 3.305189188006281e-05,
1646
+ "loss": 0.4026,
1647
+ "step": 1125
1648
+ },
1649
+ {
1650
+ "epoch": 7.635135135135135,
1651
+ "grad_norm": 0.80078125,
1652
+ "learning_rate": 3.218056641853337e-05,
1653
+ "loss": 0.3947,
1654
+ "step": 1130
1655
+ },
1656
+ {
1657
+ "epoch": 7.668918918918919,
1658
+ "grad_norm": 0.7890625,
1659
+ "learning_rate": 3.1318672449631284e-05,
1660
+ "loss": 0.3939,
1661
+ "step": 1135
1662
+ },
1663
+ {
1664
+ "epoch": 7.702702702702703,
1665
+ "grad_norm": 0.75390625,
1666
+ "learning_rate": 3.0466329834968233e-05,
1667
+ "loss": 0.3939,
1668
+ "step": 1140
1669
+ },
1670
+ {
1671
+ "epoch": 7.736486486486487,
1672
+ "grad_norm": 0.80859375,
1673
+ "learning_rate": 2.9623657107870996e-05,
1674
+ "loss": 0.3925,
1675
+ "step": 1145
1676
+ },
1677
+ {
1678
+ "epoch": 7.77027027027027,
1679
+ "grad_norm": 0.7890625,
1680
+ "learning_rate": 2.879077145689746e-05,
1681
+ "loss": 0.3896,
1682
+ "step": 1150
1683
+ },
1684
+ {
1685
+ "epoch": 7.804054054054054,
1686
+ "grad_norm": 0.7890625,
1687
+ "learning_rate": 2.7967788709539233e-05,
1688
+ "loss": 0.3823,
1689
+ "step": 1155
1690
+ },
1691
+ {
1692
+ "epoch": 7.837837837837838,
1693
+ "grad_norm": 0.79296875,
1694
+ "learning_rate": 2.7154823316113932e-05,
1695
+ "loss": 0.4006,
1696
+ "step": 1160
1697
+ },
1698
+ {
1699
+ "epoch": 7.871621621621622,
1700
+ "grad_norm": 0.7578125,
1701
+ "learning_rate": 2.6351988333848788e-05,
1702
+ "loss": 0.3879,
1703
+ "step": 1165
1704
+ },
1705
+ {
1706
+ "epoch": 7.905405405405405,
1707
+ "grad_norm": 0.86328125,
1708
+ "learning_rate": 2.5559395411158115e-05,
1709
+ "loss": 0.3861,
1710
+ "step": 1170
1711
+ },
1712
+ {
1713
+ "epoch": 7.9391891891891895,
1714
+ "grad_norm": 0.76171875,
1715
+ "learning_rate": 2.4777154772116496e-05,
1716
+ "loss": 0.4031,
1717
+ "step": 1175
1718
+ },
1719
+ {
1720
+ "epoch": 7.972972972972973,
1721
+ "grad_norm": 0.80859375,
1722
+ "learning_rate": 2.4005375201130274e-05,
1723
+ "loss": 0.3886,
1724
+ "step": 1180
1725
+ },
1726
+ {
1727
+ "epoch": 8.0,
1728
+ "eval_loss": 2.8496615886688232,
1729
+ "eval_runtime": 1.0122,
1730
+ "eval_samples_per_second": 4.94,
1731
+ "eval_steps_per_second": 1.976,
1732
+ "step": 1184
1733
+ },
1734
+ {
1735
+ "epoch": 8.006756756756756,
1736
+ "grad_norm": 0.7109375,
1737
+ "learning_rate": 2.324416402780907e-05,
1738
+ "loss": 0.3975,
1739
+ "step": 1185
1740
+ },
1741
+ {
1742
+ "epoch": 8.04054054054054,
1743
+ "grad_norm": 0.7109375,
1744
+ "learning_rate": 2.249362711203985e-05,
1745
+ "loss": 0.3604,
1746
+ "step": 1190
1747
+ },
1748
+ {
1749
+ "epoch": 8.074324324324325,
1750
+ "grad_norm": 0.7734375,
1751
+ "learning_rate": 2.1753868829265046e-05,
1752
+ "loss": 0.361,
1753
+ "step": 1195
1754
+ },
1755
+ {
1756
+ "epoch": 8.108108108108109,
1757
+ "grad_norm": 0.85546875,
1758
+ "learning_rate": 2.102499205596743e-05,
1759
+ "loss": 0.3573,
1760
+ "step": 1200
1761
+ },
1762
+ {
1763
+ "epoch": 8.141891891891891,
1764
+ "grad_norm": 0.75390625,
1765
+ "learning_rate": 2.0307098155363236e-05,
1766
+ "loss": 0.3567,
1767
+ "step": 1205
1768
+ },
1769
+ {
1770
+ "epoch": 8.175675675675675,
1771
+ "grad_norm": 0.734375,
1772
+ "learning_rate": 1.9600286963305957e-05,
1773
+ "loss": 0.3561,
1774
+ "step": 1210
1775
+ },
1776
+ {
1777
+ "epoch": 8.20945945945946,
1778
+ "grad_norm": 0.75,
1779
+ "learning_rate": 1.8904656774402208e-05,
1780
+ "loss": 0.3661,
1781
+ "step": 1215
1782
+ },
1783
+ {
1784
+ "epoch": 8.243243243243244,
1785
+ "grad_norm": 0.76953125,
1786
+ "learning_rate": 1.8220304328342252e-05,
1787
+ "loss": 0.3729,
1788
+ "step": 1220
1789
+ },
1790
+ {
1791
+ "epoch": 8.277027027027026,
1792
+ "grad_norm": 0.7578125,
1793
+ "learning_rate": 1.754732479644655e-05,
1794
+ "loss": 0.3614,
1795
+ "step": 1225
1796
+ },
1797
+ {
1798
+ "epoch": 8.31081081081081,
1799
+ "grad_norm": 0.734375,
1800
+ "learning_rate": 1.688581176843066e-05,
1801
+ "loss": 0.3608,
1802
+ "step": 1230
1803
+ },
1804
+ {
1805
+ "epoch": 8.344594594594595,
1806
+ "grad_norm": 0.77734375,
1807
+ "learning_rate": 1.6235857239389696e-05,
1808
+ "loss": 0.368,
1809
+ "step": 1235
1810
+ },
1811
+ {
1812
+ "epoch": 8.378378378378379,
1813
+ "grad_norm": 0.7734375,
1814
+ "learning_rate": 1.5597551597004966e-05,
1815
+ "loss": 0.3453,
1816
+ "step": 1240
1817
+ },
1818
+ {
1819
+ "epoch": 8.412162162162161,
1820
+ "grad_norm": 0.71484375,
1821
+ "learning_rate": 1.4970983608973942e-05,
1822
+ "loss": 0.3594,
1823
+ "step": 1245
1824
+ },
1825
+ {
1826
+ "epoch": 8.445945945945946,
1827
+ "grad_norm": 0.74609375,
1828
+ "learning_rate": 1.4356240410665433e-05,
1829
+ "loss": 0.3673,
1830
+ "step": 1250
1831
+ },
1832
+ {
1833
+ "epoch": 8.47972972972973,
1834
+ "grad_norm": 0.73828125,
1835
+ "learning_rate": 1.3753407493001968e-05,
1836
+ "loss": 0.3618,
1837
+ "step": 1255
1838
+ },
1839
+ {
1840
+ "epoch": 8.513513513513514,
1841
+ "grad_norm": 0.74609375,
1842
+ "learning_rate": 1.3162568690570743e-05,
1843
+ "loss": 0.3462,
1844
+ "step": 1260
1845
+ },
1846
+ {
1847
+ "epoch": 8.547297297297296,
1848
+ "grad_norm": 0.79296875,
1849
+ "learning_rate": 1.2583806169964961e-05,
1850
+ "loss": 0.3717,
1851
+ "step": 1265
1852
+ },
1853
+ {
1854
+ "epoch": 8.58108108108108,
1855
+ "grad_norm": 0.73046875,
1856
+ "learning_rate": 1.2017200418357078e-05,
1857
+ "loss": 0.3506,
1858
+ "step": 1270
1859
+ },
1860
+ {
1861
+ "epoch": 8.614864864864865,
1862
+ "grad_norm": 0.7578125,
1863
+ "learning_rate": 1.14628302323056e-05,
1864
+ "loss": 0.3653,
1865
+ "step": 1275
1866
+ },
1867
+ {
1868
+ "epoch": 8.64864864864865,
1869
+ "grad_norm": 0.80078125,
1870
+ "learning_rate": 1.0920772706797167e-05,
1871
+ "loss": 0.3672,
1872
+ "step": 1280
1873
+ },
1874
+ {
1875
+ "epoch": 8.682432432432432,
1876
+ "grad_norm": 0.7421875,
1877
+ "learning_rate": 1.0391103224524956e-05,
1878
+ "loss": 0.3627,
1879
+ "step": 1285
1880
+ },
1881
+ {
1882
+ "epoch": 8.716216216216216,
1883
+ "grad_norm": 0.73046875,
1884
+ "learning_rate": 9.873895445405523e-06,
1885
+ "loss": 0.3608,
1886
+ "step": 1290
1887
+ },
1888
+ {
1889
+ "epoch": 8.75,
1890
+ "grad_norm": 0.796875,
1891
+ "learning_rate": 9.369221296335006e-06,
1892
+ "loss": 0.3657,
1893
+ "step": 1295
1894
+ },
1895
+ {
1896
+ "epoch": 8.783783783783784,
1897
+ "grad_norm": 0.72265625,
1898
+ "learning_rate": 8.87715096118642e-06,
1899
+ "loss": 0.3412,
1900
+ "step": 1300
1901
+ },
1902
+ {
1903
+ "epoch": 8.817567567567568,
1904
+ "grad_norm": 0.7421875,
1905
+ "learning_rate": 8.397752871049436e-06,
1906
+ "loss": 0.3612,
1907
+ "step": 1305
1908
+ },
1909
+ {
1910
+ "epoch": 8.85135135135135,
1911
+ "grad_norm": 0.76171875,
1912
+ "learning_rate": 7.931093694713687e-06,
1913
+ "loss": 0.3552,
1914
+ "step": 1310
1915
+ },
1916
+ {
1917
+ "epoch": 8.885135135135135,
1918
+ "grad_norm": 0.7734375,
1919
+ "learning_rate": 7.477238329397418e-06,
1920
+ "loss": 0.3623,
1921
+ "step": 1315
1922
+ },
1923
+ {
1924
+ "epoch": 8.91891891891892,
1925
+ "grad_norm": 0.75390625,
1926
+ "learning_rate": 7.03624989172228e-06,
1927
+ "loss": 0.3563,
1928
+ "step": 1320
1929
+ },
1930
+ {
1931
+ "epoch": 8.952702702702704,
1932
+ "grad_norm": 0.82421875,
1933
+ "learning_rate": 6.608189708935964e-06,
1934
+ "loss": 0.3574,
1935
+ "step": 1325
1936
+ },
1937
+ {
1938
+ "epoch": 8.986486486486486,
1939
+ "grad_norm": 0.75390625,
1940
+ "learning_rate": 6.1931173103834115e-06,
1941
+ "loss": 0.3611,
1942
+ "step": 1330
1943
+ },
1944
+ {
1945
+ "epoch": 9.0,
1946
+ "eval_loss": 2.9868175983428955,
1947
+ "eval_runtime": 1.0121,
1948
+ "eval_samples_per_second": 4.94,
1949
+ "eval_steps_per_second": 1.976,
1950
+ "step": 1332
1951
+ },
1952
+ {
1953
+ "epoch": 9.02027027027027,
1954
+ "grad_norm": 0.75390625,
1955
+ "learning_rate": 5.791090419228351e-06,
1956
+ "loss": 0.3557,
1957
+ "step": 1335
1958
+ },
1959
+ {
1960
+ "epoch": 9.054054054054054,
1961
+ "grad_norm": 0.71484375,
1962
+ "learning_rate": 5.402164944425758e-06,
1963
+ "loss": 0.3503,
1964
+ "step": 1340
1965
+ },
1966
+ {
1967
+ "epoch": 9.087837837837839,
1968
+ "grad_norm": 0.7265625,
1969
+ "learning_rate": 5.026394972946813e-06,
1970
+ "loss": 0.3513,
1971
+ "step": 1345
1972
+ },
1973
+ {
1974
+ "epoch": 9.121621621621621,
1975
+ "grad_norm": 0.703125,
1976
+ "learning_rate": 4.66383276225707e-06,
1977
+ "loss": 0.354,
1978
+ "step": 1350
1979
+ },
1980
+ {
1981
+ "epoch": 9.155405405405405,
1982
+ "grad_norm": 0.796875,
1983
+ "learning_rate": 4.314528733049206e-06,
1984
+ "loss": 0.3452,
1985
+ "step": 1355
1986
+ },
1987
+ {
1988
+ "epoch": 9.18918918918919,
1989
+ "grad_norm": 0.73046875,
1990
+ "learning_rate": 3.9785314622310495e-06,
1991
+ "loss": 0.3555,
1992
+ "step": 1360
1993
+ },
1994
+ {
1995
+ "epoch": 9.222972972972974,
1996
+ "grad_norm": 0.73828125,
1997
+ "learning_rate": 3.655887676170222e-06,
1998
+ "loss": 0.3437,
1999
+ "step": 1365
2000
+ },
2001
+ {
2002
+ "epoch": 9.256756756756756,
2003
+ "grad_norm": 0.71484375,
2004
+ "learning_rate": 3.3466422441958634e-06,
2005
+ "loss": 0.3448,
2006
+ "step": 1370
2007
+ },
2008
+ {
2009
+ "epoch": 9.29054054054054,
2010
+ "grad_norm": 0.765625,
2011
+ "learning_rate": 3.050838172358883e-06,
2012
+ "loss": 0.3558,
2013
+ "step": 1375
2014
+ },
2015
+ {
2016
+ "epoch": 9.324324324324325,
2017
+ "grad_norm": 0.76171875,
2018
+ "learning_rate": 2.7685165974510986e-06,
2019
+ "loss": 0.3604,
2020
+ "step": 1380
2021
+ },
2022
+ {
2023
+ "epoch": 9.358108108108109,
2024
+ "grad_norm": 0.75,
2025
+ "learning_rate": 2.499716781284556e-06,
2026
+ "loss": 0.3458,
2027
+ "step": 1385
2028
+ },
2029
+ {
2030
+ "epoch": 9.391891891891891,
2031
+ "grad_norm": 0.765625,
2032
+ "learning_rate": 2.2444761052313856e-06,
2033
+ "loss": 0.3534,
2034
+ "step": 1390
2035
+ },
2036
+ {
2037
+ "epoch": 9.425675675675675,
2038
+ "grad_norm": 0.75390625,
2039
+ "learning_rate": 2.002830065025263e-06,
2040
+ "loss": 0.3618,
2041
+ "step": 1395
2042
+ },
2043
+ {
2044
+ "epoch": 9.45945945945946,
2045
+ "grad_norm": 0.734375,
2046
+ "learning_rate": 1.7748122658251876e-06,
2047
+ "loss": 0.3562,
2048
+ "step": 1400
2049
+ },
2050
+ {
2051
+ "epoch": 9.493243243243244,
2052
+ "grad_norm": 0.73828125,
2053
+ "learning_rate": 1.56045441754199e-06,
2054
+ "loss": 0.341,
2055
+ "step": 1405
2056
+ },
2057
+ {
2058
+ "epoch": 9.527027027027026,
2059
+ "grad_norm": 0.8125,
2060
+ "learning_rate": 1.3597863304285475e-06,
2061
+ "loss": 0.3735,
2062
+ "step": 1410
2063
+ },
2064
+ {
2065
+ "epoch": 9.56081081081081,
2066
+ "grad_norm": 0.70703125,
2067
+ "learning_rate": 1.1728359109341446e-06,
2068
+ "loss": 0.3583,
2069
+ "step": 1415
2070
+ },
2071
+ {
2072
+ "epoch": 9.594594594594595,
2073
+ "grad_norm": 0.76171875,
2074
+ "learning_rate": 9.996291578236228e-07,
2075
+ "loss": 0.3456,
2076
+ "step": 1420
2077
+ },
2078
+ {
2079
+ "epoch": 9.628378378378379,
2080
+ "grad_norm": 0.7734375,
2081
+ "learning_rate": 8.401901585616823e-07,
2082
+ "loss": 0.3568,
2083
+ "step": 1425
2084
+ },
2085
+ {
2086
+ "epoch": 9.662162162162161,
2087
+ "grad_norm": 0.75,
2088
+ "learning_rate": 6.945410859632295e-07,
2089
+ "loss": 0.3518,
2090
+ "step": 1430
2091
+ },
2092
+ {
2093
+ "epoch": 9.695945945945946,
2094
+ "grad_norm": 0.7421875,
2095
+ "learning_rate": 5.627021951097545e-07,
2096
+ "loss": 0.3585,
2097
+ "step": 1435
2098
+ },
2099
+ {
2100
+ "epoch": 9.72972972972973,
2101
+ "grad_norm": 0.7578125,
2102
+ "learning_rate": 4.44691820532539e-07,
2103
+ "loss": 0.3471,
2104
+ "step": 1440
2105
+ },
2106
+ {
2107
+ "epoch": 9.763513513513514,
2108
+ "grad_norm": 0.7421875,
2109
+ "learning_rate": 3.405263736629416e-07,
2110
+ "loss": 0.3496,
2111
+ "step": 1445
2112
+ },
2113
+ {
2114
+ "epoch": 9.797297297297296,
2115
+ "grad_norm": 0.7421875,
2116
+ "learning_rate": 2.5022034055003364e-07,
2117
+ "loss": 0.3512,
2118
+ "step": 1450
2119
+ },
2120
+ {
2121
+ "epoch": 9.83108108108108,
2122
+ "grad_norm": 0.7421875,
2123
+ "learning_rate": 1.7378627984612207e-07,
2124
+ "loss": 0.3498,
2125
+ "step": 1455
2126
+ },
2127
+ {
2128
+ "epoch": 9.864864864864865,
2129
+ "grad_norm": 0.7421875,
2130
+ "learning_rate": 1.1123482106021322e-07,
2131
+ "loss": 0.3514,
2132
+ "step": 1460
2133
+ },
2134
+ {
2135
+ "epoch": 9.89864864864865,
2136
+ "grad_norm": 0.75,
2137
+ "learning_rate": 6.25746630798063e-08,
2138
+ "loss": 0.3613,
2139
+ "step": 1465
2140
+ },
2141
+ {
2142
+ "epoch": 9.932432432432432,
2143
+ "grad_norm": 0.75390625,
2144
+ "learning_rate": 2.7812572961127824e-08,
2145
+ "loss": 0.3546,
2146
+ "step": 1470
2147
+ },
2148
+ {
2149
+ "epoch": 9.966216216216216,
2150
+ "grad_norm": 0.75390625,
2151
+ "learning_rate": 6.953384988095391e-09,
2152
+ "loss": 0.3575,
2153
+ "step": 1475
2154
+ },
2155
+ {
2156
+ "epoch": 10.0,
2157
+ "grad_norm": 0.71484375,
2158
+ "learning_rate": 0.0,
2159
+ "loss": 0.3501,
2160
+ "step": 1480
2161
+ },
2162
+ {
2163
+ "epoch": 10.0,
2164
+ "eval_loss": 3.0043678283691406,
2165
+ "eval_runtime": 1.0152,
2166
+ "eval_samples_per_second": 4.925,
2167
+ "eval_steps_per_second": 1.97,
2168
+ "step": 1480
2169
+ },
2170
+ {
2171
+ "epoch": 10.0,
2172
+ "step": 1480,
2173
+ "total_flos": 2.2627322959180595e+18,
2174
+ "train_loss": 1.3636517344294368,
2175
+ "train_runtime": 11668.1199,
2176
+ "train_samples_per_second": 2.027,
2177
+ "train_steps_per_second": 0.127
2178
  }
2179
  ],
2180
  "logging_steps": 5,
2181
+ "max_steps": 1480,
2182
  "num_input_tokens_seen": 0,
2183
+ "num_train_epochs": 10,
2184
  "save_steps": 100,
2185
  "stateful_callbacks": {
2186
  "TrainerControl": {
 
2194
  "attributes": {}
2195
  }
2196
  },
2197
+ "total_flos": 2.2627322959180595e+18,
2198
  "train_batch_size": 4,
2199
  "trial_name": null,
2200
  "trial_params": null