googlefan commited on
Commit
d36c31f
1 Parent(s): 1d61963

Upload folder using huggingface_hub

Browse files
config.json CHANGED
@@ -15,8 +15,7 @@
15
  },
16
  "auto_map": {
17
  "AutoConfig": "ultravox_config.UltravoxConfig",
18
- "AutoModel": "ultravox_model.UltravoxModel",
19
- "AutoProcessor": "ultravox_processing.UltravoxProcessor"
20
  },
21
  "hidden_size": 4096,
22
  "ignore_index": -100,
 
15
  },
16
  "auto_map": {
17
  "AutoConfig": "ultravox_config.UltravoxConfig",
18
+ "AutoModel": "ultravox_model.UltravoxModel"
 
19
  },
20
  "hidden_size": 4096,
21
  "ignore_index": -100,
model.safetensors CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:7287a044c945ad4f1903cef5887e534c70cddb3852b1a2bd80a230e61d00b22f
3
  size 98594264
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:3f96b114dc5aaf0babd3bfc1d6402c665d35b272794739c2c995e1b5b48613a6
3
  size 98594264
optimizer.pt CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:2b3fde1382a141c8b9075d0038d9b4ecfa526a15bca7e0b376ced5c550dce6db
3
  size 197192018
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:9212dce4de8f1f94cb675d3840a56c5d110b2a9e48a630d260d16bfea658fac8
3
  size 197192018
rng_state.pth CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:50406e2683fae845a67d2522407cfd71c13ce88867a1aac0dc9d26a8b3a5f840
3
  size 14244
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:82e130c98e7ed5e266ae7ac7b9364697dbb958f068a5dbf05247b00b39414fbf
3
  size 14244
scheduler.pt CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:ae7d1df7fbe282a2a1f6e0263b266c20c101d0f96ff5d77217fca16dab8166f9
3
  size 1064
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:b31b9209f3a41ee0b631b36612edf799c6d53b46d246c0ad39a98cc7d2469425
3
  size 1064
tokenizer_config.json CHANGED
@@ -2000,7 +2000,7 @@
2000
  "<end_of_turn>"
2001
  ],
2002
  "bos_token": "<bos>",
2003
- "chat_template": "{{ bos_token }}{% if messages[0]['role'] == 'system' %}{{ messages[0]['content'] + '\n' }}{% endif %}{% for message in messages %}{% if (message['role'] == 'assistant') %}{% set role = 'model' %}{% else %}{% set role = message['role'] %}{% endif %}{% if message['role'] != 'system' %}{{ '<start_of_turn>' + role + '\n' + message['content'] | trim + '<end_of_turn>\n' }}{% endif %}{% endfor %}{% if add_generation_prompt %}{{'<start_of_turn>model\n'}}{% endif %}",
2004
  "clean_up_tokenization_spaces": false,
2005
  "eos_token": "<eos>",
2006
  "model_max_length": 1000000000000000019884624838656,
 
2000
  "<end_of_turn>"
2001
  ],
2002
  "bos_token": "<bos>",
2003
+ "chat_template": "{{ bos_token }}{% if messages[0]['role'] == 'system' %}{{ raise_exception('System role not supported') }}{% endif %}{% for message in messages %}{% if (message['role'] == 'user') != (loop.index0 % 2 == 0) %}{{ raise_exception('Conversation roles must alternate user/assistant/user/assistant/...') }}{% endif %}{% if (message['role'] == 'assistant') %}{% set role = 'model' %}{% else %}{% set role = message['role'] %}{% endif %}{{ '<start_of_turn>' + role + '\n' + message['content'] | trim + '<end_of_turn>\n' }}{% endfor %}{% if add_generation_prompt %}{{'<start_of_turn>model\n'}}{% endif %}",
2004
  "clean_up_tokenization_spaces": false,
2005
  "eos_token": "<eos>",
2006
  "model_max_length": 1000000000000000019884624838656,
trainer_state.json CHANGED
@@ -1,9 +1,9 @@
1
  {
2
  "best_metric": null,
3
  "best_model_checkpoint": null,
4
- "epoch": 3.5842293906810037,
5
  "eval_steps": 1000,
6
- "global_step": 6000,
7
  "is_hyper_param_search": false,
8
  "is_local_process_zero": true,
9
  "is_world_process_zero": true,
@@ -17,423 +17,1263 @@
17
  },
18
  {
19
  "epoch": 0.05973715651135006,
20
- "grad_norm": 0.625,
21
  "learning_rate": 0.0002,
22
- "loss": 0.5498,
23
  "step": 100
24
  },
25
  {
26
  "epoch": 0.11947431302270012,
27
- "grad_norm": 0.322265625,
28
  "learning_rate": 0.0004,
29
- "loss": 0.3884,
30
  "step": 200
31
  },
32
  {
33
  "epoch": 0.17921146953405018,
34
- "grad_norm": 0.302734375,
35
  "learning_rate": 0.0006,
36
- "loss": 0.376,
37
  "step": 300
38
  },
39
  {
40
  "epoch": 0.23894862604540024,
41
- "grad_norm": 0.12353515625,
42
  "learning_rate": 0.0008,
43
- "loss": 0.3635,
44
  "step": 400
45
  },
46
  {
47
  "epoch": 0.2986857825567503,
48
- "grad_norm": 0.1826171875,
49
  "learning_rate": 0.001,
50
- "loss": 0.3559,
51
  "step": 500
52
  },
53
  {
54
  "epoch": 0.35842293906810035,
55
- "grad_norm": 0.10888671875,
56
  "learning_rate": 0.0012,
57
- "loss": 0.3438,
58
  "step": 600
59
  },
60
  {
61
  "epoch": 0.41816009557945044,
62
- "grad_norm": 0.11572265625,
63
  "learning_rate": 0.0014,
64
- "loss": 0.3412,
65
  "step": 700
66
  },
67
  {
68
  "epoch": 0.4778972520908005,
69
- "grad_norm": 0.08984375,
70
  "learning_rate": 0.0016,
71
- "loss": 0.3392,
72
  "step": 800
73
  },
74
  {
75
  "epoch": 0.5376344086021505,
76
- "grad_norm": 0.11376953125,
77
  "learning_rate": 0.0018000000000000002,
78
- "loss": 0.346,
79
  "step": 900
80
  },
81
  {
82
  "epoch": 0.5973715651135006,
83
- "grad_norm": 0.08251953125,
84
  "learning_rate": 0.002,
85
- "loss": 0.3174,
86
  "step": 1000
87
  },
88
  {
89
  "epoch": 0.6571087216248507,
90
- "grad_norm": 0.0703125,
91
  "learning_rate": 0.0019998292504580526,
92
- "loss": 0.3309,
93
  "step": 1100
94
  },
95
  {
96
  "epoch": 0.7168458781362007,
97
- "grad_norm": 0.1533203125,
98
  "learning_rate": 0.001999317060143023,
99
- "loss": 0.323,
100
  "step": 1200
101
  },
102
  {
103
  "epoch": 0.7765830346475507,
104
- "grad_norm": 0.09228515625,
105
  "learning_rate": 0.001998463603967434,
106
- "loss": 0.3157,
107
  "step": 1300
108
  },
109
  {
110
  "epoch": 0.8363201911589009,
111
- "grad_norm": 0.072265625,
112
  "learning_rate": 0.0019972691733857882,
113
- "loss": 0.3155,
114
  "step": 1400
115
  },
116
  {
117
  "epoch": 0.8960573476702509,
118
- "grad_norm": 0.12890625,
119
  "learning_rate": 0.0019957341762950344,
120
- "loss": 0.3148,
121
  "step": 1500
122
  },
123
  {
124
  "epoch": 0.955794504181601,
125
- "grad_norm": 0.10205078125,
126
  "learning_rate": 0.001993859136895274,
127
- "loss": 0.3013,
128
  "step": 1600
129
  },
130
  {
131
  "epoch": 1.015531660692951,
132
- "grad_norm": 0.11474609375,
133
  "learning_rate": 0.0019916446955107426,
134
- "loss": 0.3009,
135
  "step": 1700
136
  },
137
  {
138
  "epoch": 1.075268817204301,
139
- "grad_norm": 0.12890625,
140
  "learning_rate": 0.001989091608371146,
141
- "loss": 0.3005,
142
  "step": 1800
143
  },
144
  {
145
  "epoch": 1.135005973715651,
146
- "grad_norm": 0.087890625,
147
  "learning_rate": 0.0019862007473534027,
148
- "loss": 0.3028,
149
  "step": 1900
150
  },
151
  {
152
  "epoch": 1.1947431302270013,
153
- "grad_norm": 0.10546875,
154
  "learning_rate": 0.001982973099683902,
155
- "loss": 0.3006,
156
  "step": 2000
157
  },
158
  {
159
  "epoch": 1.2544802867383513,
160
- "grad_norm": 0.1044921875,
161
  "learning_rate": 0.001979409767601366,
162
- "loss": 0.2981,
163
  "step": 2100
164
  },
165
  {
166
  "epoch": 1.3142174432497014,
167
- "grad_norm": 0.10888671875,
168
  "learning_rate": 0.001975511967980437,
169
- "loss": 0.2917,
170
  "step": 2200
171
  },
172
  {
173
  "epoch": 1.3739545997610514,
174
- "grad_norm": 0.10107421875,
175
  "learning_rate": 0.001971281031916114,
176
- "loss": 0.2816,
177
  "step": 2300
178
  },
179
  {
180
  "epoch": 1.4336917562724014,
181
- "grad_norm": 0.1337890625,
182
  "learning_rate": 0.0019667184042691877,
183
- "loss": 0.2754,
184
  "step": 2400
185
  },
186
  {
187
  "epoch": 1.4934289127837514,
188
- "grad_norm": 0.08203125,
189
  "learning_rate": 0.001961825643172819,
190
- "loss": 0.2775,
191
  "step": 2500
192
  },
193
  {
194
  "epoch": 1.5531660692951015,
195
- "grad_norm": 0.0634765625,
196
  "learning_rate": 0.0019566044195004407,
197
- "loss": 0.2756,
198
  "step": 2600
199
  },
200
  {
201
  "epoch": 1.6129032258064515,
202
- "grad_norm": 0.1142578125,
203
  "learning_rate": 0.0019510565162951536,
204
- "loss": 0.27,
205
  "step": 2700
206
  },
207
  {
208
  "epoch": 1.6726403823178018,
209
- "grad_norm": 0.09423828125,
210
  "learning_rate": 0.0019451838281608197,
211
- "loss": 0.2616,
212
  "step": 2800
213
  },
214
  {
215
  "epoch": 1.7323775388291518,
216
- "grad_norm": 0.095703125,
217
  "learning_rate": 0.0019389883606150567,
218
- "loss": 0.2589,
219
  "step": 2900
220
  },
221
  {
222
  "epoch": 1.7921146953405018,
223
- "grad_norm": 0.12353515625,
224
  "learning_rate": 0.0019324722294043557,
225
- "loss": 0.2567,
226
  "step": 3000
227
  },
228
  {
229
  "epoch": 1.8518518518518519,
230
- "grad_norm": 0.138671875,
231
  "learning_rate": 0.0019256376597815564,
232
- "loss": 0.2462,
233
  "step": 3100
234
  },
235
  {
236
  "epoch": 1.911589008363202,
237
- "grad_norm": 0.10205078125,
238
  "learning_rate": 0.001918486985745923,
239
- "loss": 0.246,
240
  "step": 3200
241
  },
242
  {
243
  "epoch": 1.971326164874552,
244
- "grad_norm": 0.07861328125,
245
  "learning_rate": 0.0019110226492460884,
246
- "loss": 0.2378,
247
  "step": 3300
248
  },
249
  {
250
  "epoch": 2.031063321385902,
251
- "grad_norm": 0.091796875,
252
  "learning_rate": 0.0019032471993461289,
253
- "loss": 0.2322,
254
  "step": 3400
255
  },
256
  {
257
  "epoch": 2.090800477897252,
258
- "grad_norm": 0.09814453125,
259
  "learning_rate": 0.0018951632913550625,
260
- "loss": 0.2341,
261
  "step": 3500
262
  },
263
  {
264
  "epoch": 2.150537634408602,
265
- "grad_norm": 0.0859375,
266
  "learning_rate": 0.0018867736859200619,
267
- "loss": 0.2314,
268
  "step": 3600
269
  },
270
  {
271
  "epoch": 2.2102747909199523,
272
- "grad_norm": 0.09765625,
273
  "learning_rate": 0.0018780812480836979,
274
- "loss": 0.2297,
275
  "step": 3700
276
  },
277
  {
278
  "epoch": 2.270011947431302,
279
- "grad_norm": 0.087890625,
280
  "learning_rate": 0.0018690889463055284,
281
- "loss": 0.224,
282
  "step": 3800
283
  },
284
  {
285
  "epoch": 2.3297491039426523,
286
- "grad_norm": 0.0693359375,
287
  "learning_rate": 0.0018597998514483724,
288
- "loss": 0.2193,
289
  "step": 3900
290
  },
291
  {
292
  "epoch": 2.3894862604540026,
293
- "grad_norm": 0.11669921875,
294
  "learning_rate": 0.0018502171357296143,
295
- "loss": 0.2158,
296
  "step": 4000
297
  },
298
  {
299
  "epoch": 2.4492234169653524,
300
- "grad_norm": 0.07568359375,
301
  "learning_rate": 0.0018403440716378927,
302
- "loss": 0.2081,
303
  "step": 4100
304
  },
305
  {
306
  "epoch": 2.5089605734767026,
307
- "grad_norm": 0.10498046875,
308
  "learning_rate": 0.0018301840308155505,
309
- "loss": 0.2092,
310
  "step": 4200
311
  },
312
  {
313
  "epoch": 2.5686977299880525,
314
- "grad_norm": 0.10107421875,
315
  "learning_rate": 0.0018197404829072212,
316
- "loss": 0.2038,
317
  "step": 4300
318
  },
319
  {
320
  "epoch": 2.6284348864994027,
321
- "grad_norm": 0.103515625,
322
  "learning_rate": 0.0018090169943749475,
323
- "loss": 0.199,
324
  "step": 4400
325
  },
326
  {
327
  "epoch": 2.6881720430107525,
328
- "grad_norm": 0.087890625,
329
  "learning_rate": 0.0017980172272802398,
330
- "loss": 0.1955,
331
  "step": 4500
332
  },
333
  {
334
  "epoch": 2.7479091995221028,
335
- "grad_norm": 0.0986328125,
336
  "learning_rate": 0.0017867449380334832,
337
- "loss": 0.1937,
338
  "step": 4600
339
  },
340
  {
341
  "epoch": 2.807646356033453,
342
- "grad_norm": 0.1044921875,
343
  "learning_rate": 0.0017752039761111298,
344
- "loss": 0.1914,
345
  "step": 4700
346
  },
347
  {
348
  "epoch": 2.867383512544803,
349
- "grad_norm": 0.0947265625,
350
  "learning_rate": 0.001763398282741103,
351
- "loss": 0.1875,
352
  "step": 4800
353
  },
354
  {
355
  "epoch": 2.927120669056153,
356
- "grad_norm": 0.09765625,
357
  "learning_rate": 0.0017513318895568735,
358
- "loss": 0.1855,
359
  "step": 4900
360
  },
361
  {
362
  "epoch": 2.986857825567503,
363
- "grad_norm": 0.08056640625,
364
  "learning_rate": 0.001739008917220659,
365
- "loss": 0.1795,
366
  "step": 5000
367
  },
368
  {
369
  "epoch": 3.046594982078853,
370
- "grad_norm": 0.1259765625,
371
  "learning_rate": 0.0017264335740162242,
372
- "loss": 0.1742,
373
  "step": 5100
374
  },
375
  {
376
  "epoch": 3.106332138590203,
377
- "grad_norm": 0.09228515625,
378
  "learning_rate": 0.0017136101544117524,
379
- "loss": 0.1759,
380
  "step": 5200
381
  },
382
  {
383
  "epoch": 3.166069295101553,
384
- "grad_norm": 0.1259765625,
385
  "learning_rate": 0.0017005430375932908,
386
- "loss": 0.1756,
387
  "step": 5300
388
  },
389
  {
390
  "epoch": 3.225806451612903,
391
- "grad_norm": 0.07470703125,
392
  "learning_rate": 0.0016872366859692627,
393
- "loss": 0.1766,
394
  "step": 5400
395
  },
396
  {
397
  "epoch": 3.2855436081242533,
398
- "grad_norm": 0.087890625,
399
  "learning_rate": 0.0016736956436465573,
400
- "loss": 0.1666,
401
  "step": 5500
402
  },
403
  {
404
  "epoch": 3.3452807646356035,
405
- "grad_norm": 0.0869140625,
406
  "learning_rate": 0.0016599245348787228,
407
- "loss": 0.169,
408
  "step": 5600
409
  },
410
  {
411
  "epoch": 3.4050179211469533,
412
- "grad_norm": 0.08740234375,
413
  "learning_rate": 0.0016459280624867873,
414
- "loss": 0.1616,
415
  "step": 5700
416
  },
417
  {
418
  "epoch": 3.4647550776583036,
419
- "grad_norm": 0.158203125,
420
  "learning_rate": 0.001631711006253251,
421
- "loss": 0.1595,
422
  "step": 5800
423
  },
424
  {
425
  "epoch": 3.5244922341696534,
426
- "grad_norm": 0.078125,
427
  "learning_rate": 0.001617278221289793,
428
- "loss": 0.1626,
429
  "step": 5900
430
  },
431
  {
432
  "epoch": 3.5842293906810037,
433
- "grad_norm": 0.0908203125,
434
  "learning_rate": 0.0016026346363792565,
435
- "loss": 0.1546,
436
  "step": 6000
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
437
  }
438
  ],
439
  "logging_steps": 100,
@@ -448,12 +1288,12 @@
448
  "should_evaluate": true,
449
  "should_log": false,
450
  "should_save": true,
451
- "should_training_stop": false
452
  },
453
  "attributes": {}
454
  }
455
  },
456
- "total_flos": 1.1559951680731546e+17,
457
  "train_batch_size": 6,
458
  "trial_name": null,
459
  "trial_params": null
 
1
  {
2
  "best_metric": null,
3
  "best_model_checkpoint": null,
4
+ "epoch": 3.752688172043011,
5
  "eval_steps": 1000,
6
+ "global_step": 18000,
7
  "is_hyper_param_search": false,
8
  "is_local_process_zero": true,
9
  "is_world_process_zero": true,
 
17
  },
18
  {
19
  "epoch": 0.05973715651135006,
20
+ "grad_norm": 0.61328125,
21
  "learning_rate": 0.0002,
22
+ "loss": 0.5503,
23
  "step": 100
24
  },
25
  {
26
  "epoch": 0.11947431302270012,
27
+ "grad_norm": 0.314453125,
28
  "learning_rate": 0.0004,
29
+ "loss": 0.3885,
30
  "step": 200
31
  },
32
  {
33
  "epoch": 0.17921146953405018,
34
+ "grad_norm": 0.298828125,
35
  "learning_rate": 0.0006,
36
+ "loss": 0.3758,
37
  "step": 300
38
  },
39
  {
40
  "epoch": 0.23894862604540024,
41
+ "grad_norm": 0.12890625,
42
  "learning_rate": 0.0008,
43
+ "loss": 0.363,
44
  "step": 400
45
  },
46
  {
47
  "epoch": 0.2986857825567503,
48
+ "grad_norm": 0.1943359375,
49
  "learning_rate": 0.001,
50
+ "loss": 0.3567,
51
  "step": 500
52
  },
53
  {
54
  "epoch": 0.35842293906810035,
55
+ "grad_norm": 0.11181640625,
56
  "learning_rate": 0.0012,
57
+ "loss": 0.3435,
58
  "step": 600
59
  },
60
  {
61
  "epoch": 0.41816009557945044,
62
+ "grad_norm": 0.12109375,
63
  "learning_rate": 0.0014,
64
+ "loss": 0.3415,
65
  "step": 700
66
  },
67
  {
68
  "epoch": 0.4778972520908005,
69
+ "grad_norm": 0.09130859375,
70
  "learning_rate": 0.0016,
71
+ "loss": 0.3394,
72
  "step": 800
73
  },
74
  {
75
  "epoch": 0.5376344086021505,
76
+ "grad_norm": 0.11572265625,
77
  "learning_rate": 0.0018000000000000002,
78
+ "loss": 0.345,
79
  "step": 900
80
  },
81
  {
82
  "epoch": 0.5973715651135006,
83
+ "grad_norm": 0.07421875,
84
  "learning_rate": 0.002,
85
+ "loss": 0.3177,
86
  "step": 1000
87
  },
88
  {
89
  "epoch": 0.6571087216248507,
90
+ "grad_norm": 0.06494140625,
91
  "learning_rate": 0.0019998292504580526,
92
+ "loss": 0.3318,
93
  "step": 1100
94
  },
95
  {
96
  "epoch": 0.7168458781362007,
97
+ "grad_norm": 0.1455078125,
98
  "learning_rate": 0.001999317060143023,
99
+ "loss": 0.3229,
100
  "step": 1200
101
  },
102
  {
103
  "epoch": 0.7765830346475507,
104
+ "grad_norm": 0.1298828125,
105
  "learning_rate": 0.001998463603967434,
106
+ "loss": 0.3145,
107
  "step": 1300
108
  },
109
  {
110
  "epoch": 0.8363201911589009,
111
+ "grad_norm": 0.0732421875,
112
  "learning_rate": 0.0019972691733857882,
113
+ "loss": 0.3158,
114
  "step": 1400
115
  },
116
  {
117
  "epoch": 0.8960573476702509,
118
+ "grad_norm": 0.13671875,
119
  "learning_rate": 0.0019957341762950344,
120
+ "loss": 0.3132,
121
  "step": 1500
122
  },
123
  {
124
  "epoch": 0.955794504181601,
125
+ "grad_norm": 0.11572265625,
126
  "learning_rate": 0.001993859136895274,
127
+ "loss": 0.3011,
128
  "step": 1600
129
  },
130
  {
131
  "epoch": 1.015531660692951,
132
+ "grad_norm": 0.125,
133
  "learning_rate": 0.0019916446955107426,
134
+ "loss": 0.3014,
135
  "step": 1700
136
  },
137
  {
138
  "epoch": 1.075268817204301,
139
+ "grad_norm": 0.11279296875,
140
  "learning_rate": 0.001989091608371146,
141
+ "loss": 0.3,
142
  "step": 1800
143
  },
144
  {
145
  "epoch": 1.135005973715651,
146
+ "grad_norm": 0.08544921875,
147
  "learning_rate": 0.0019862007473534027,
148
+ "loss": 0.3016,
149
  "step": 1900
150
  },
151
  {
152
  "epoch": 1.1947431302270013,
153
+ "grad_norm": 0.10791015625,
154
  "learning_rate": 0.001982973099683902,
155
+ "loss": 0.2991,
156
  "step": 2000
157
  },
158
  {
159
  "epoch": 1.2544802867383513,
160
+ "grad_norm": 0.103515625,
161
  "learning_rate": 0.001979409767601366,
162
+ "loss": 0.2974,
163
  "step": 2100
164
  },
165
  {
166
  "epoch": 1.3142174432497014,
167
+ "grad_norm": 0.08056640625,
168
  "learning_rate": 0.001975511967980437,
169
+ "loss": 0.2903,
170
  "step": 2200
171
  },
172
  {
173
  "epoch": 1.3739545997610514,
174
+ "grad_norm": 0.109375,
175
  "learning_rate": 0.001971281031916114,
176
+ "loss": 0.2796,
177
  "step": 2300
178
  },
179
  {
180
  "epoch": 1.4336917562724014,
181
+ "grad_norm": 0.125,
182
  "learning_rate": 0.0019667184042691877,
183
+ "loss": 0.2746,
184
  "step": 2400
185
  },
186
  {
187
  "epoch": 1.4934289127837514,
188
+ "grad_norm": 0.10400390625,
189
  "learning_rate": 0.001961825643172819,
190
+ "loss": 0.2766,
191
  "step": 2500
192
  },
193
  {
194
  "epoch": 1.5531660692951015,
195
+ "grad_norm": 0.06494140625,
196
  "learning_rate": 0.0019566044195004407,
197
+ "loss": 0.2739,
198
  "step": 2600
199
  },
200
  {
201
  "epoch": 1.6129032258064515,
202
+ "grad_norm": 0.12255859375,
203
  "learning_rate": 0.0019510565162951536,
204
+ "loss": 0.2677,
205
  "step": 2700
206
  },
207
  {
208
  "epoch": 1.6726403823178018,
209
+ "grad_norm": 0.0869140625,
210
  "learning_rate": 0.0019451838281608197,
211
+ "loss": 0.2605,
212
  "step": 2800
213
  },
214
  {
215
  "epoch": 1.7323775388291518,
216
+ "grad_norm": 0.08837890625,
217
  "learning_rate": 0.0019389883606150567,
218
+ "loss": 0.259,
219
  "step": 2900
220
  },
221
  {
222
  "epoch": 1.7921146953405018,
223
+ "grad_norm": 0.1279296875,
224
  "learning_rate": 0.0019324722294043557,
225
+ "loss": 0.256,
226
  "step": 3000
227
  },
228
  {
229
  "epoch": 1.8518518518518519,
230
+ "grad_norm": 0.1279296875,
231
  "learning_rate": 0.0019256376597815564,
232
+ "loss": 0.246,
233
  "step": 3100
234
  },
235
  {
236
  "epoch": 1.911589008363202,
237
+ "grad_norm": 0.12890625,
238
  "learning_rate": 0.001918486985745923,
239
+ "loss": 0.2431,
240
  "step": 3200
241
  },
242
  {
243
  "epoch": 1.971326164874552,
244
+ "grad_norm": 0.083984375,
245
  "learning_rate": 0.0019110226492460884,
246
+ "loss": 0.2364,
247
  "step": 3300
248
  },
249
  {
250
  "epoch": 2.031063321385902,
251
+ "grad_norm": 0.0927734375,
252
  "learning_rate": 0.0019032471993461289,
253
+ "loss": 0.2306,
254
  "step": 3400
255
  },
256
  {
257
  "epoch": 2.090800477897252,
258
+ "grad_norm": 0.087890625,
259
  "learning_rate": 0.0018951632913550625,
260
+ "loss": 0.2331,
261
  "step": 3500
262
  },
263
  {
264
  "epoch": 2.150537634408602,
265
+ "grad_norm": 0.091796875,
266
  "learning_rate": 0.0018867736859200619,
267
+ "loss": 0.2291,
268
  "step": 3600
269
  },
270
  {
271
  "epoch": 2.2102747909199523,
272
+ "grad_norm": 0.0849609375,
273
  "learning_rate": 0.0018780812480836979,
274
+ "loss": 0.227,
275
  "step": 3700
276
  },
277
  {
278
  "epoch": 2.270011947431302,
279
+ "grad_norm": 0.1015625,
280
  "learning_rate": 0.0018690889463055284,
281
+ "loss": 0.2255,
282
  "step": 3800
283
  },
284
  {
285
  "epoch": 2.3297491039426523,
286
+ "grad_norm": 0.0703125,
287
  "learning_rate": 0.0018597998514483724,
288
+ "loss": 0.2165,
289
  "step": 3900
290
  },
291
  {
292
  "epoch": 2.3894862604540026,
293
+ "grad_norm": 0.091796875,
294
  "learning_rate": 0.0018502171357296143,
295
+ "loss": 0.2157,
296
  "step": 4000
297
  },
298
  {
299
  "epoch": 2.4492234169653524,
300
+ "grad_norm": 0.08740234375,
301
  "learning_rate": 0.0018403440716378927,
302
+ "loss": 0.2086,
303
  "step": 4100
304
  },
305
  {
306
  "epoch": 2.5089605734767026,
307
+ "grad_norm": 0.111328125,
308
  "learning_rate": 0.0018301840308155505,
309
+ "loss": 0.2077,
310
  "step": 4200
311
  },
312
  {
313
  "epoch": 2.5686977299880525,
314
+ "grad_norm": 0.09619140625,
315
  "learning_rate": 0.0018197404829072212,
316
+ "loss": 0.201,
317
  "step": 4300
318
  },
319
  {
320
  "epoch": 2.6284348864994027,
321
+ "grad_norm": 0.11083984375,
322
  "learning_rate": 0.0018090169943749475,
323
+ "loss": 0.1991,
324
  "step": 4400
325
  },
326
  {
327
  "epoch": 2.6881720430107525,
328
+ "grad_norm": 0.0830078125,
329
  "learning_rate": 0.0017980172272802398,
330
+ "loss": 0.1961,
331
  "step": 4500
332
  },
333
  {
334
  "epoch": 2.7479091995221028,
335
+ "grad_norm": 0.10107421875,
336
  "learning_rate": 0.0017867449380334832,
337
+ "loss": 0.1934,
338
  "step": 4600
339
  },
340
  {
341
  "epoch": 2.807646356033453,
342
+ "grad_norm": 0.1064453125,
343
  "learning_rate": 0.0017752039761111298,
344
+ "loss": 0.1862,
345
  "step": 4700
346
  },
347
  {
348
  "epoch": 2.867383512544803,
349
+ "grad_norm": 0.08154296875,
350
  "learning_rate": 0.001763398282741103,
351
+ "loss": 0.1849,
352
  "step": 4800
353
  },
354
  {
355
  "epoch": 2.927120669056153,
356
+ "grad_norm": 0.0869140625,
357
  "learning_rate": 0.0017513318895568735,
358
+ "loss": 0.1828,
359
  "step": 4900
360
  },
361
  {
362
  "epoch": 2.986857825567503,
363
+ "grad_norm": 0.09619140625,
364
  "learning_rate": 0.001739008917220659,
365
+ "loss": 0.1769,
366
  "step": 5000
367
  },
368
  {
369
  "epoch": 3.046594982078853,
370
+ "grad_norm": 0.107421875,
371
  "learning_rate": 0.0017264335740162242,
372
+ "loss": 0.1729,
373
  "step": 5100
374
  },
375
  {
376
  "epoch": 3.106332138590203,
377
+ "grad_norm": 0.0859375,
378
  "learning_rate": 0.0017136101544117524,
379
+ "loss": 0.1745,
380
  "step": 5200
381
  },
382
  {
383
  "epoch": 3.166069295101553,
384
+ "grad_norm": 0.107421875,
385
  "learning_rate": 0.0017005430375932908,
386
+ "loss": 0.1742,
387
  "step": 5300
388
  },
389
  {
390
  "epoch": 3.225806451612903,
391
+ "grad_norm": 0.08837890625,
392
  "learning_rate": 0.0016872366859692627,
393
+ "loss": 0.175,
394
  "step": 5400
395
  },
396
  {
397
  "epoch": 3.2855436081242533,
398
+ "grad_norm": 0.126953125,
399
  "learning_rate": 0.0016736956436465573,
400
+ "loss": 0.1643,
401
  "step": 5500
402
  },
403
  {
404
  "epoch": 3.3452807646356035,
405
+ "grad_norm": 0.07666015625,
406
  "learning_rate": 0.0016599245348787228,
407
+ "loss": 0.1651,
408
  "step": 5600
409
  },
410
  {
411
  "epoch": 3.4050179211469533,
412
+ "grad_norm": 0.12158203125,
413
  "learning_rate": 0.0016459280624867873,
414
+ "loss": 0.1613,
415
  "step": 5700
416
  },
417
  {
418
  "epoch": 3.4647550776583036,
419
+ "grad_norm": 0.1396484375,
420
  "learning_rate": 0.001631711006253251,
421
+ "loss": 0.1586,
422
  "step": 5800
423
  },
424
  {
425
  "epoch": 3.5244922341696534,
426
+ "grad_norm": 0.09228515625,
427
  "learning_rate": 0.001617278221289793,
428
+ "loss": 0.1601,
429
  "step": 5900
430
  },
431
  {
432
  "epoch": 3.5842293906810037,
433
+ "grad_norm": 0.0888671875,
434
  "learning_rate": 0.0016026346363792565,
435
+ "loss": 0.1515,
436
  "step": 6000
437
+ },
438
+ {
439
+ "epoch": 3.6439665471923535,
440
+ "grad_norm": 0.1259765625,
441
+ "learning_rate": 0.0015877852522924731,
442
+ "loss": 0.1492,
443
+ "step": 6100
444
+ },
445
+ {
446
+ "epoch": 3.7037037037037037,
447
+ "grad_norm": 0.0771484375,
448
+ "learning_rate": 0.0015727351400805052,
449
+ "loss": 0.1514,
450
+ "step": 6200
451
+ },
452
+ {
453
+ "epoch": 3.763440860215054,
454
+ "grad_norm": 0.0771484375,
455
+ "learning_rate": 0.0015574894393428856,
456
+ "loss": 0.1462,
457
+ "step": 6300
458
+ },
459
+ {
460
+ "epoch": 3.823178016726404,
461
+ "grad_norm": 0.078125,
462
+ "learning_rate": 0.0015420533564724495,
463
+ "loss": 0.143,
464
+ "step": 6400
465
+ },
466
+ {
467
+ "epoch": 3.882915173237754,
468
+ "grad_norm": 0.0673828125,
469
+ "learning_rate": 0.0015264321628773558,
470
+ "loss": 0.1426,
471
+ "step": 6500
472
+ },
473
+ {
474
+ "epoch": 3.942652329749104,
475
+ "grad_norm": 0.07421875,
476
+ "learning_rate": 0.001510631193180907,
477
+ "loss": 0.1391,
478
+ "step": 6600
479
+ },
480
+ {
481
+ "epoch": 4.002389486260454,
482
+ "grad_norm": 0.10107421875,
483
+ "learning_rate": 0.001494655843399779,
484
+ "loss": 0.1355,
485
+ "step": 6700
486
+ },
487
+ {
488
+ "epoch": 4.062126642771804,
489
+ "grad_norm": 0.0791015625,
490
+ "learning_rate": 0.0014785115691012866,
491
+ "loss": 0.1335,
492
+ "step": 6800
493
+ },
494
+ {
495
+ "epoch": 4.121863799283154,
496
+ "grad_norm": 0.09375,
497
+ "learning_rate": 0.0014622038835403132,
498
+ "loss": 0.1398,
499
+ "step": 6900
500
+ },
501
+ {
502
+ "epoch": 4.181600955794504,
503
+ "grad_norm": 0.08203125,
504
+ "learning_rate": 0.0014457383557765385,
505
+ "loss": 0.1342,
506
+ "step": 7000
507
+ },
508
+ {
509
+ "epoch": 4.241338112305854,
510
+ "grad_norm": 0.1064453125,
511
+ "learning_rate": 0.001429120608772609,
512
+ "loss": 0.135,
513
+ "step": 7100
514
+ },
515
+ {
516
+ "epoch": 4.301075268817204,
517
+ "grad_norm": 0.07958984375,
518
+ "learning_rate": 0.0014123563174739035,
519
+ "loss": 0.1297,
520
+ "step": 7200
521
+ },
522
+ {
523
+ "epoch": 4.360812425328555,
524
+ "grad_norm": 0.0927734375,
525
+ "learning_rate": 0.0013954512068705424,
526
+ "loss": 0.1294,
527
+ "step": 7300
528
+ },
529
+ {
530
+ "epoch": 4.4205495818399045,
531
+ "grad_norm": 0.09423828125,
532
+ "learning_rate": 0.0013784110500423103,
533
+ "loss": 0.1257,
534
+ "step": 7400
535
+ },
536
+ {
537
+ "epoch": 4.480286738351254,
538
+ "grad_norm": 0.076171875,
539
+ "learning_rate": 0.0013612416661871532,
540
+ "loss": 0.1249,
541
+ "step": 7500
542
+ },
543
+ {
544
+ "epoch": 4.540023894862604,
545
+ "grad_norm": 0.064453125,
546
+ "learning_rate": 0.0013439489186339282,
547
+ "loss": 0.1243,
548
+ "step": 7600
549
+ },
550
+ {
551
+ "epoch": 4.599761051373955,
552
+ "grad_norm": 0.07275390625,
553
+ "learning_rate": 0.0013265387128400831,
554
+ "loss": 0.1191,
555
+ "step": 7700
556
+ },
557
+ {
558
+ "epoch": 4.659498207885305,
559
+ "grad_norm": 0.06591796875,
560
+ "learning_rate": 0.0013090169943749475,
561
+ "loss": 0.1188,
562
+ "step": 7800
563
+ },
564
+ {
565
+ "epoch": 4.7192353643966545,
566
+ "grad_norm": 0.0703125,
567
+ "learning_rate": 0.0012913897468893247,
568
+ "loss": 0.1174,
569
+ "step": 7900
570
+ },
571
+ {
572
+ "epoch": 4.778972520908005,
573
+ "grad_norm": 0.061279296875,
574
+ "learning_rate": 0.0012736629900720832,
575
+ "loss": 0.1168,
576
+ "step": 8000
577
+ },
578
+ {
579
+ "epoch": 4.838709677419355,
580
+ "grad_norm": 0.06640625,
581
+ "learning_rate": 0.0012558427775944357,
582
+ "loss": 0.1135,
583
+ "step": 8100
584
+ },
585
+ {
586
+ "epoch": 4.898446833930705,
587
+ "grad_norm": 0.09912109375,
588
+ "learning_rate": 0.0012379351950426187,
589
+ "loss": 0.1117,
590
+ "step": 8200
591
+ },
592
+ {
593
+ "epoch": 4.958183990442055,
594
+ "grad_norm": 0.05712890625,
595
+ "learning_rate": 0.0012199463578396689,
596
+ "loss": 0.1092,
597
+ "step": 8300
598
+ },
599
+ {
600
+ "epoch": 5.017921146953405,
601
+ "grad_norm": 0.07861328125,
602
+ "learning_rate": 0.0012018824091570102,
603
+ "loss": 0.1091,
604
+ "step": 8400
605
+ },
606
+ {
607
+ "epoch": 5.077658303464755,
608
+ "grad_norm": 0.0859375,
609
+ "learning_rate": 0.0011837495178165704,
610
+ "loss": 0.1086,
611
+ "step": 8500
612
+ },
613
+ {
614
+ "epoch": 5.137395459976105,
615
+ "grad_norm": 0.06103515625,
616
+ "learning_rate": 0.00116555387618413,
617
+ "loss": 0.109,
618
+ "step": 8600
619
+ },
620
+ {
621
+ "epoch": 5.197132616487456,
622
+ "grad_norm": 0.0869140625,
623
+ "learning_rate": 0.0011473016980546376,
624
+ "loss": 0.1071,
625
+ "step": 8700
626
+ },
627
+ {
628
+ "epoch": 5.256869772998805,
629
+ "grad_norm": 0.0703125,
630
+ "learning_rate": 0.0011289992165302034,
631
+ "loss": 0.1081,
632
+ "step": 8800
633
+ },
634
+ {
635
+ "epoch": 5.316606929510155,
636
+ "grad_norm": 0.0869140625,
637
+ "learning_rate": 0.001110652681891501,
638
+ "loss": 0.1061,
639
+ "step": 8900
640
+ },
641
+ {
642
+ "epoch": 5.376344086021505,
643
+ "grad_norm": 0.123046875,
644
+ "learning_rate": 0.001092268359463302,
645
+ "loss": 0.1044,
646
+ "step": 9000
647
+ },
648
+ {
649
+ "epoch": 5.436081242532856,
650
+ "grad_norm": 0.07177734375,
651
+ "learning_rate": 0.001073852527474874,
652
+ "loss": 0.1004,
653
+ "step": 9100
654
+ },
655
+ {
656
+ "epoch": 5.4958183990442055,
657
+ "grad_norm": 0.09130859375,
658
+ "learning_rate": 0.00105541147491597,
659
+ "loss": 0.1004,
660
+ "step": 9200
661
+ },
662
+ {
663
+ "epoch": 5.555555555555555,
664
+ "grad_norm": 0.058837890625,
665
+ "learning_rate": 0.0010369514993891452,
666
+ "loss": 0.098,
667
+ "step": 9300
668
+ },
669
+ {
670
+ "epoch": 5.615292712066906,
671
+ "grad_norm": 0.057861328125,
672
+ "learning_rate": 0.00101847890495913,
673
+ "loss": 0.0989,
674
+ "step": 9400
675
+ },
676
+ {
677
+ "epoch": 5.675029868578256,
678
+ "grad_norm": 0.07421875,
679
+ "learning_rate": 0.001,
680
+ "loss": 0.0954,
681
+ "step": 9500
682
+ },
683
+ {
684
+ "epoch": 5.734767025089606,
685
+ "grad_norm": 0.09375,
686
+ "learning_rate": 0.0009815210950408703,
687
+ "loss": 0.0955,
688
+ "step": 9600
689
+ },
690
+ {
691
+ "epoch": 5.7945041816009555,
692
+ "grad_norm": 0.0654296875,
693
+ "learning_rate": 0.0009630485006108553,
694
+ "loss": 0.094,
695
+ "step": 9700
696
+ },
697
+ {
698
+ "epoch": 5.854241338112306,
699
+ "grad_norm": 0.06787109375,
700
+ "learning_rate": 0.0009445885250840301,
701
+ "loss": 0.0924,
702
+ "step": 9800
703
+ },
704
+ {
705
+ "epoch": 5.913978494623656,
706
+ "grad_norm": 0.06298828125,
707
+ "learning_rate": 0.0009261474725251261,
708
+ "loss": 0.0911,
709
+ "step": 9900
710
+ },
711
+ {
712
+ "epoch": 5.973715651135006,
713
+ "grad_norm": 0.054443359375,
714
+ "learning_rate": 0.0009077316405366981,
715
+ "loss": 0.0902,
716
+ "step": 10000
717
+ },
718
+ {
719
+ "epoch": 6.033452807646356,
720
+ "grad_norm": 0.0673828125,
721
+ "learning_rate": 0.0008893473181084994,
722
+ "loss": 0.0891,
723
+ "step": 10100
724
+ },
725
+ {
726
+ "epoch": 6.093189964157706,
727
+ "grad_norm": 0.06103515625,
728
+ "learning_rate": 0.000871000783469797,
729
+ "loss": 0.0914,
730
+ "step": 10200
731
+ },
732
+ {
733
+ "epoch": 6.152927120669056,
734
+ "grad_norm": 0.053466796875,
735
+ "learning_rate": 0.0008526983019453623,
736
+ "loss": 0.0872,
737
+ "step": 10300
738
+ },
739
+ {
740
+ "epoch": 6.212664277180406,
741
+ "grad_norm": 0.08740234375,
742
+ "learning_rate": 0.00083444612381587,
743
+ "loss": 0.0903,
744
+ "step": 10400
745
+ },
746
+ {
747
+ "epoch": 6.272401433691757,
748
+ "grad_norm": 0.0712890625,
749
+ "learning_rate": 0.0008162504821834296,
750
+ "loss": 0.0858,
751
+ "step": 10500
752
+ },
753
+ {
754
+ "epoch": 6.332138590203106,
755
+ "grad_norm": 0.051513671875,
756
+ "learning_rate": 0.00079811759084299,
757
+ "loss": 0.0876,
758
+ "step": 10600
759
+ },
760
+ {
761
+ "epoch": 6.391875746714456,
762
+ "grad_norm": 0.05810546875,
763
+ "learning_rate": 0.0007800536421603317,
764
+ "loss": 0.0874,
765
+ "step": 10700
766
+ },
767
+ {
768
+ "epoch": 6.451612903225806,
769
+ "grad_norm": 0.060546875,
770
+ "learning_rate": 0.0007620648049573815,
771
+ "loss": 0.0834,
772
+ "step": 10800
773
+ },
774
+ {
775
+ "epoch": 6.511350059737157,
776
+ "grad_norm": 0.05859375,
777
+ "learning_rate": 0.0007441572224055644,
778
+ "loss": 0.0814,
779
+ "step": 10900
780
+ },
781
+ {
782
+ "epoch": 6.571087216248507,
783
+ "grad_norm": 0.06494140625,
784
+ "learning_rate": 0.0007263370099279172,
785
+ "loss": 0.0822,
786
+ "step": 11000
787
+ },
788
+ {
789
+ "epoch": 6.630824372759856,
790
+ "grad_norm": 0.04833984375,
791
+ "learning_rate": 0.0007086102531106754,
792
+ "loss": 0.0814,
793
+ "step": 11100
794
+ },
795
+ {
796
+ "epoch": 6.690561529271207,
797
+ "grad_norm": 0.072265625,
798
+ "learning_rate": 0.0006909830056250527,
799
+ "loss": 0.0803,
800
+ "step": 11200
801
+ },
802
+ {
803
+ "epoch": 6.750298685782557,
804
+ "grad_norm": 0.052734375,
805
+ "learning_rate": 0.0006734612871599168,
806
+ "loss": 0.0802,
807
+ "step": 11300
808
+ },
809
+ {
810
+ "epoch": 6.810035842293907,
811
+ "grad_norm": 0.053955078125,
812
+ "learning_rate": 0.0006560510813660718,
813
+ "loss": 0.0781,
814
+ "step": 11400
815
+ },
816
+ {
817
+ "epoch": 6.8697729988052565,
818
+ "grad_norm": 0.051025390625,
819
+ "learning_rate": 0.0006387583338128471,
820
+ "loss": 0.0779,
821
+ "step": 11500
822
+ },
823
+ {
824
+ "epoch": 6.929510155316607,
825
+ "grad_norm": 0.08544921875,
826
+ "learning_rate": 0.0006215889499576897,
827
+ "loss": 0.0783,
828
+ "step": 11600
829
+ },
830
+ {
831
+ "epoch": 6.989247311827957,
832
+ "grad_norm": 0.0732421875,
833
+ "learning_rate": 0.0006045487931294575,
834
+ "loss": 0.0765,
835
+ "step": 11700
836
+ },
837
+ {
838
+ "epoch": 7.048984468339307,
839
+ "grad_norm": 0.062255859375,
840
+ "learning_rate": 0.0005876436825260967,
841
+ "loss": 0.0764,
842
+ "step": 11800
843
+ },
844
+ {
845
+ "epoch": 7.1087216248506575,
846
+ "grad_norm": 0.06298828125,
847
+ "learning_rate": 0.000570879391227391,
848
+ "loss": 0.0773,
849
+ "step": 11900
850
+ },
851
+ {
852
+ "epoch": 7.168458781362007,
853
+ "grad_norm": 0.056640625,
854
+ "learning_rate": 0.0005542616442234618,
855
+ "loss": 0.0759,
856
+ "step": 12000
857
+ },
858
+ {
859
+ "epoch": 7.228195937873357,
860
+ "grad_norm": 0.042236328125,
861
+ "learning_rate": 0.0005377961164596869,
862
+ "loss": 0.0783,
863
+ "step": 12100
864
+ },
865
+ {
866
+ "epoch": 7.287933094384707,
867
+ "grad_norm": 0.053955078125,
868
+ "learning_rate": 0.0005214884308987136,
869
+ "loss": 0.0757,
870
+ "step": 12200
871
+ },
872
+ {
873
+ "epoch": 7.347670250896058,
874
+ "grad_norm": 0.053955078125,
875
+ "learning_rate": 0.0005053441566002214,
876
+ "loss": 0.0768,
877
+ "step": 12300
878
+ },
879
+ {
880
+ "epoch": 7.407407407407407,
881
+ "grad_norm": 0.0556640625,
882
+ "learning_rate": 0.0004893688068190932,
883
+ "loss": 0.0746,
884
+ "step": 12400
885
+ },
886
+ {
887
+ "epoch": 7.467144563918757,
888
+ "grad_norm": 0.0537109375,
889
+ "learning_rate": 0.0004735678371226441,
890
+ "loss": 0.0749,
891
+ "step": 12500
892
+ },
893
+ {
894
+ "epoch": 7.526881720430108,
895
+ "grad_norm": 0.057861328125,
896
+ "learning_rate": 0.00045794664352755057,
897
+ "loss": 0.0728,
898
+ "step": 12600
899
+ },
900
+ {
901
+ "epoch": 7.586618876941458,
902
+ "grad_norm": 0.05419921875,
903
+ "learning_rate": 0.0004425105606571145,
904
+ "loss": 0.0732,
905
+ "step": 12700
906
+ },
907
+ {
908
+ "epoch": 7.646356033452808,
909
+ "grad_norm": 0.0498046875,
910
+ "learning_rate": 0.00042726485991949483,
911
+ "loss": 0.0731,
912
+ "step": 12800
913
+ },
914
+ {
915
+ "epoch": 7.706093189964157,
916
+ "grad_norm": 0.05859375,
917
+ "learning_rate": 0.00041221474770752696,
918
+ "loss": 0.0728,
919
+ "step": 12900
920
+ },
921
+ {
922
+ "epoch": 7.765830346475508,
923
+ "grad_norm": 0.046630859375,
924
+ "learning_rate": 0.0003973653636207437,
925
+ "loss": 0.072,
926
+ "step": 13000
927
+ },
928
+ {
929
+ "epoch": 7.825567502986858,
930
+ "grad_norm": 0.043212890625,
931
+ "learning_rate": 0.0003827217787102072,
932
+ "loss": 0.0718,
933
+ "step": 13100
934
+ },
935
+ {
936
+ "epoch": 7.885304659498208,
937
+ "grad_norm": 0.0400390625,
938
+ "learning_rate": 0.0003682889937467493,
939
+ "loss": 0.0721,
940
+ "step": 13200
941
+ },
942
+ {
943
+ "epoch": 7.945041816009558,
944
+ "grad_norm": 0.047607421875,
945
+ "learning_rate": 0.00035407193751321286,
946
+ "loss": 0.0703,
947
+ "step": 13300
948
+ },
949
+ {
950
+ "epoch": 1.004778972520908,
951
+ "grad_norm": 0.057861328125,
952
+ "learning_rate": 0.0003400754651212776,
953
+ "loss": 0.0716,
954
+ "step": 13400
955
+ },
956
+ {
957
+ "epoch": 1.064516129032258,
958
+ "grad_norm": 0.049072265625,
959
+ "learning_rate": 0.0003263043563534428,
960
+ "loss": 0.0711,
961
+ "step": 13500
962
+ },
963
+ {
964
+ "epoch": 1.124253285543608,
965
+ "grad_norm": 0.039794921875,
966
+ "learning_rate": 0.0003127633140307373,
967
+ "loss": 0.0714,
968
+ "step": 13600
969
+ },
970
+ {
971
+ "epoch": 1.183990442054958,
972
+ "grad_norm": 0.046630859375,
973
+ "learning_rate": 0.00029945696240670904,
974
+ "loss": 0.0728,
975
+ "step": 13700
976
+ },
977
+ {
978
+ "epoch": 1.2437275985663083,
979
+ "grad_norm": 0.06005859375,
980
+ "learning_rate": 0.00028638984558824776,
981
+ "loss": 0.0733,
982
+ "step": 13800
983
+ },
984
+ {
985
+ "epoch": 1.3034647550776584,
986
+ "grad_norm": 0.04833984375,
987
+ "learning_rate": 0.000273566425983776,
988
+ "loss": 0.071,
989
+ "step": 13900
990
+ },
991
+ {
992
+ "epoch": 1.3632019115890084,
993
+ "grad_norm": 0.042724609375,
994
+ "learning_rate": 0.000260991082779341,
995
+ "loss": 0.0729,
996
+ "step": 14000
997
+ },
998
+ {
999
+ "epoch": 1.4229390681003584,
1000
+ "grad_norm": 0.0380859375,
1001
+ "learning_rate": 0.00024866811044312666,
1002
+ "loss": 0.0711,
1003
+ "step": 14100
1004
+ },
1005
+ {
1006
+ "epoch": 1.4826762246117084,
1007
+ "grad_norm": 0.043701171875,
1008
+ "learning_rate": 0.00023660171725889702,
1009
+ "loss": 0.0704,
1010
+ "step": 14200
1011
+ },
1012
+ {
1013
+ "epoch": 1.5424133811230587,
1014
+ "grad_norm": 0.04931640625,
1015
+ "learning_rate": 0.0002247960238888701,
1016
+ "loss": 0.0705,
1017
+ "step": 14300
1018
+ },
1019
+ {
1020
+ "epoch": 1.6021505376344085,
1021
+ "grad_norm": 0.04638671875,
1022
+ "learning_rate": 0.00021325506196651677,
1023
+ "loss": 0.0709,
1024
+ "step": 14400
1025
+ },
1026
+ {
1027
+ "epoch": 1.6618876941457588,
1028
+ "grad_norm": 0.034423828125,
1029
+ "learning_rate": 0.0002019827727197605,
1030
+ "loss": 0.0708,
1031
+ "step": 14500
1032
+ },
1033
+ {
1034
+ "epoch": 1.7216248506571086,
1035
+ "grad_norm": 0.04541015625,
1036
+ "learning_rate": 0.00019098300562505265,
1037
+ "loss": 0.0692,
1038
+ "step": 14600
1039
+ },
1040
+ {
1041
+ "epoch": 1.7813620071684588,
1042
+ "grad_norm": 0.056396484375,
1043
+ "learning_rate": 0.000180259517092779,
1044
+ "loss": 0.0702,
1045
+ "step": 14700
1046
+ },
1047
+ {
1048
+ "epoch": 1.8410991636798089,
1049
+ "grad_norm": 0.050048828125,
1050
+ "learning_rate": 0.00016981596918444952,
1051
+ "loss": 0.0701,
1052
+ "step": 14800
1053
+ },
1054
+ {
1055
+ "epoch": 1.900836320191159,
1056
+ "grad_norm": 0.049560546875,
1057
+ "learning_rate": 0.0001596559283621074,
1058
+ "loss": 0.0695,
1059
+ "step": 14900
1060
+ },
1061
+ {
1062
+ "epoch": 1.960573476702509,
1063
+ "grad_norm": 0.05224609375,
1064
+ "learning_rate": 0.00014978286427038602,
1065
+ "loss": 0.0701,
1066
+ "step": 15000
1067
+ },
1068
+ {
1069
+ "epoch": 2.020310633213859,
1070
+ "grad_norm": 0.05419921875,
1071
+ "learning_rate": 0.00014020014855162756,
1072
+ "loss": 0.0684,
1073
+ "step": 15100
1074
+ },
1075
+ {
1076
+ "epoch": 2.080047789725209,
1077
+ "grad_norm": 0.040283203125,
1078
+ "learning_rate": 0.00013091105369447166,
1079
+ "loss": 0.072,
1080
+ "step": 15200
1081
+ },
1082
+ {
1083
+ "epoch": 2.139784946236559,
1084
+ "grad_norm": 0.040283203125,
1085
+ "learning_rate": 0.00012191875191630208,
1086
+ "loss": 0.0695,
1087
+ "step": 15300
1088
+ },
1089
+ {
1090
+ "epoch": 2.1995221027479093,
1091
+ "grad_norm": 0.051025390625,
1092
+ "learning_rate": 0.00011322631407993811,
1093
+ "loss": 0.0717,
1094
+ "step": 15400
1095
+ },
1096
+ {
1097
+ "epoch": 2.2592592592592595,
1098
+ "grad_norm": 0.046142578125,
1099
+ "learning_rate": 0.00010483670864493777,
1100
+ "loss": 0.0721,
1101
+ "step": 15500
1102
+ },
1103
+ {
1104
+ "epoch": 2.3189964157706093,
1105
+ "grad_norm": 0.036865234375,
1106
+ "learning_rate": 9.675280065387115e-05,
1107
+ "loss": 0.0704,
1108
+ "step": 15600
1109
+ },
1110
+ {
1111
+ "epoch": 2.378733572281959,
1112
+ "grad_norm": 0.040771484375,
1113
+ "learning_rate": 8.897735075391155e-05,
1114
+ "loss": 0.0711,
1115
+ "step": 15700
1116
+ },
1117
+ {
1118
+ "epoch": 2.4384707287933094,
1119
+ "grad_norm": 0.0322265625,
1120
+ "learning_rate": 8.151301425407698e-05,
1121
+ "loss": 0.0698,
1122
+ "step": 15800
1123
+ },
1124
+ {
1125
+ "epoch": 2.4982078853046596,
1126
+ "grad_norm": 0.043701171875,
1127
+ "learning_rate": 7.43623402184438e-05,
1128
+ "loss": 0.07,
1129
+ "step": 15900
1130
+ },
1131
+ {
1132
+ "epoch": 2.5579450418160095,
1133
+ "grad_norm": 0.041015625,
1134
+ "learning_rate": 6.75277705956443e-05,
1135
+ "loss": 0.0694,
1136
+ "step": 16000
1137
+ },
1138
+ {
1139
+ "epoch": 2.6176821983273597,
1140
+ "grad_norm": 0.046630859375,
1141
+ "learning_rate": 6.1011639384943586e-05,
1142
+ "loss": 0.0713,
1143
+ "step": 16100
1144
+ },
1145
+ {
1146
+ "epoch": 2.67741935483871,
1147
+ "grad_norm": 0.037353515625,
1148
+ "learning_rate": 5.481617183918053e-05,
1149
+ "loss": 0.069,
1150
+ "step": 16200
1151
+ },
1152
+ {
1153
+ "epoch": 2.7371565113500598,
1154
+ "grad_norm": 0.0380859375,
1155
+ "learning_rate": 4.894348370484647e-05,
1156
+ "loss": 0.0701,
1157
+ "step": 16300
1158
+ },
1159
+ {
1160
+ "epoch": 2.7968936678614096,
1161
+ "grad_norm": 0.04296875,
1162
+ "learning_rate": 4.339558049955927e-05,
1163
+ "loss": 0.0692,
1164
+ "step": 16400
1165
+ },
1166
+ {
1167
+ "epoch": 2.85663082437276,
1168
+ "grad_norm": 0.043212890625,
1169
+ "learning_rate": 3.817435682718096e-05,
1170
+ "loss": 0.0692,
1171
+ "step": 16500
1172
+ },
1173
+ {
1174
+ "epoch": 2.91636798088411,
1175
+ "grad_norm": 0.05078125,
1176
+ "learning_rate": 3.3281595730812576e-05,
1177
+ "loss": 0.0696,
1178
+ "step": 16600
1179
+ },
1180
+ {
1181
+ "epoch": 2.97610513739546,
1182
+ "grad_norm": 0.041015625,
1183
+ "learning_rate": 2.8718968083886077e-05,
1184
+ "loss": 0.0695,
1185
+ "step": 16700
1186
+ },
1187
+ {
1188
+ "epoch": 3.03584229390681,
1189
+ "grad_norm": 0.053955078125,
1190
+ "learning_rate": 2.44880320195634e-05,
1191
+ "loss": 0.0686,
1192
+ "step": 16800
1193
+ },
1194
+ {
1195
+ "epoch": 3.09557945041816,
1196
+ "grad_norm": 0.041015625,
1197
+ "learning_rate": 2.059023239863411e-05,
1198
+ "loss": 0.0713,
1199
+ "step": 16900
1200
+ },
1201
+ {
1202
+ "epoch": 3.15531660692951,
1203
+ "grad_norm": 0.04248046875,
1204
+ "learning_rate": 1.7026900316098216e-05,
1205
+ "loss": 0.069,
1206
+ "step": 17000
1207
+ },
1208
+ {
1209
+ "epoch": 3.21505376344086,
1210
+ "grad_norm": 0.0498046875,
1211
+ "learning_rate": 1.3799252646597427e-05,
1212
+ "loss": 0.0726,
1213
+ "step": 17100
1214
+ },
1215
+ {
1216
+ "epoch": 3.2747909199522103,
1217
+ "grad_norm": 0.04248046875,
1218
+ "learning_rate": 1.0908391628854042e-05,
1219
+ "loss": 0.0703,
1220
+ "step": 17200
1221
+ },
1222
+ {
1223
+ "epoch": 3.3345280764635605,
1224
+ "grad_norm": 0.049560546875,
1225
+ "learning_rate": 8.355304489257254e-06,
1226
+ "loss": 0.0711,
1227
+ "step": 17300
1228
+ },
1229
+ {
1230
+ "epoch": 3.3942652329749103,
1231
+ "grad_norm": 0.059814453125,
1232
+ "learning_rate": 6.140863104726391e-06,
1233
+ "loss": 0.072,
1234
+ "step": 17400
1235
+ },
1236
+ {
1237
+ "epoch": 3.4540023894862606,
1238
+ "grad_norm": 0.046142578125,
1239
+ "learning_rate": 4.265823704965532e-06,
1240
+ "loss": 0.0696,
1241
+ "step": 17500
1242
+ },
1243
+ {
1244
+ "epoch": 3.5137395459976104,
1245
+ "grad_norm": 0.0693359375,
1246
+ "learning_rate": 2.730826614211979e-06,
1247
+ "loss": 0.0685,
1248
+ "step": 17600
1249
+ },
1250
+ {
1251
+ "epoch": 3.5734767025089607,
1252
+ "grad_norm": 0.055419921875,
1253
+ "learning_rate": 1.5363960325660565e-06,
1254
+ "loss": 0.07,
1255
+ "step": 17700
1256
+ },
1257
+ {
1258
+ "epoch": 3.6332138590203105,
1259
+ "grad_norm": 0.046630859375,
1260
+ "learning_rate": 6.829398569770939e-07,
1261
+ "loss": 0.0708,
1262
+ "step": 17800
1263
+ },
1264
+ {
1265
+ "epoch": 3.6929510155316607,
1266
+ "grad_norm": 0.049560546875,
1267
+ "learning_rate": 1.7074954194729043e-07,
1268
+ "loss": 0.0694,
1269
+ "step": 17900
1270
+ },
1271
+ {
1272
+ "epoch": 3.752688172043011,
1273
+ "grad_norm": 0.036865234375,
1274
+ "learning_rate": 0.0,
1275
+ "loss": 0.0694,
1276
+ "step": 18000
1277
  }
1278
  ],
1279
  "logging_steps": 100,
 
1288
  "should_evaluate": true,
1289
  "should_log": false,
1290
  "should_save": true,
1291
+ "should_training_stop": true
1292
  },
1293
  "attributes": {}
1294
  }
1295
  },
1296
+ "total_flos": 3.467487355285709e+17,
1297
  "train_batch_size": 6,
1298
  "trial_name": null,
1299
  "trial_params": null
training_args.bin CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:91ac30b0e25fa5ad714bc4133367d871d355eb2cd2241f526ea5e733bbb709de
3
  size 5560
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:803d9c574bf03a7931f19bf84028283f9d1e82b62a62d69b1c4846f9c7da7b65
3
  size 5560