Grogros commited on
Commit
2a5d2e2
·
verified ·
1 Parent(s): 9142006

Training in progress, step 1500, checkpoint

Browse files
checkpoint-1500/config.json ADDED
@@ -0,0 +1,40 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "_name_or_path": "huihui-ai/Llama-3.2-1B-Instruct-abliterated",
3
+ "architectures": [
4
+ "LlamaForCausalLM"
5
+ ],
6
+ "attention_bias": false,
7
+ "attention_dropout": 0.0,
8
+ "bos_token_id": 128000,
9
+ "eos_token_id": [
10
+ 128001,
11
+ 128008,
12
+ 128009
13
+ ],
14
+ "head_dim": 64,
15
+ "hidden_act": "silu",
16
+ "hidden_size": 2048,
17
+ "initializer_range": 0.02,
18
+ "intermediate_size": 8192,
19
+ "max_position_embeddings": 131072,
20
+ "mlp_bias": false,
21
+ "model_type": "llama",
22
+ "num_attention_heads": 32,
23
+ "num_hidden_layers": 16,
24
+ "num_key_value_heads": 8,
25
+ "pretraining_tp": 1,
26
+ "rms_norm_eps": 1e-05,
27
+ "rope_scaling": {
28
+ "factor": 32.0,
29
+ "high_freq_factor": 4.0,
30
+ "low_freq_factor": 1.0,
31
+ "original_max_position_embeddings": 8192,
32
+ "rope_type": "llama3"
33
+ },
34
+ "rope_theta": 500000.0,
35
+ "tie_word_embeddings": true,
36
+ "torch_dtype": "bfloat16",
37
+ "transformers_version": "4.46.3",
38
+ "use_cache": true,
39
+ "vocab_size": 128256
40
+ }
checkpoint-1500/generation_config.json ADDED
@@ -0,0 +1,12 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "bos_token_id": 128000,
3
+ "do_sample": true,
4
+ "eos_token_id": [
5
+ 128001,
6
+ 128008,
7
+ 128009
8
+ ],
9
+ "temperature": 0.6,
10
+ "top_p": 0.9,
11
+ "transformers_version": "4.46.3"
12
+ }
checkpoint-1500/model.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:83ba4efc99b4b28543f558d6dc094d1cbc785e7640ee26f16ad3b7e7336525d1
3
+ size 2471645608
checkpoint-1500/optimizer.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:a54bf5b79c202abeaa78cf6d7b466cf071b8576405d652d2291b226a0c5ac5d3
3
+ size 3724602
checkpoint-1500/rng_state.pth ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:beb08987971240a9cec116c03cb626dee2fa851cac068f6bd3afdd37847bf7c2
3
+ size 14244
checkpoint-1500/scheduler.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:50e0fa899b9afe4b9f410a947b2e82b267a415e5e533180013057e8a556bdbd5
3
+ size 1064
checkpoint-1500/trainer_state.json ADDED
@@ -0,0 +1,1083 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "best_metric": null,
3
+ "best_model_checkpoint": null,
4
+ "epoch": 0.6546644844517185,
5
+ "eval_steps": 500,
6
+ "global_step": 1500,
7
+ "is_hyper_param_search": false,
8
+ "is_local_process_zero": true,
9
+ "is_world_process_zero": true,
10
+ "log_history": [
11
+ {
12
+ "epoch": 0.00436442989634479,
13
+ "grad_norm": 74.5,
14
+ "learning_rate": 8.000000000000001e-07,
15
+ "loss": 1.044,
16
+ "step": 10
17
+ },
18
+ {
19
+ "epoch": 0.00872885979268958,
20
+ "grad_norm": 166.0,
21
+ "learning_rate": 1.6000000000000001e-06,
22
+ "loss": 1.0152,
23
+ "step": 20
24
+ },
25
+ {
26
+ "epoch": 0.01309328968903437,
27
+ "grad_norm": 249.0,
28
+ "learning_rate": 2.4000000000000003e-06,
29
+ "loss": 1.0905,
30
+ "step": 30
31
+ },
32
+ {
33
+ "epoch": 0.01745771958537916,
34
+ "grad_norm": 210.0,
35
+ "learning_rate": 3.2000000000000003e-06,
36
+ "loss": 1.1291,
37
+ "step": 40
38
+ },
39
+ {
40
+ "epoch": 0.02182214948172395,
41
+ "grad_norm": 151.0,
42
+ "learning_rate": 4.000000000000001e-06,
43
+ "loss": 1.384,
44
+ "step": 50
45
+ },
46
+ {
47
+ "epoch": 0.02618657937806874,
48
+ "grad_norm": 516.0,
49
+ "learning_rate": 4.800000000000001e-06,
50
+ "loss": 1.5235,
51
+ "step": 60
52
+ },
53
+ {
54
+ "epoch": 0.03055100927441353,
55
+ "grad_norm": 684.0,
56
+ "learning_rate": 5.600000000000001e-06,
57
+ "loss": 1.888,
58
+ "step": 70
59
+ },
60
+ {
61
+ "epoch": 0.03491543917075832,
62
+ "grad_norm": 504.0,
63
+ "learning_rate": 6.4000000000000006e-06,
64
+ "loss": 1.9856,
65
+ "step": 80
66
+ },
67
+ {
68
+ "epoch": 0.03927986906710311,
69
+ "grad_norm": 175.0,
70
+ "learning_rate": 7.2000000000000005e-06,
71
+ "loss": 1.628,
72
+ "step": 90
73
+ },
74
+ {
75
+ "epoch": 0.0436442989634479,
76
+ "grad_norm": 964.0,
77
+ "learning_rate": 8.000000000000001e-06,
78
+ "loss": 1.8794,
79
+ "step": 100
80
+ },
81
+ {
82
+ "epoch": 0.04800872885979269,
83
+ "grad_norm": 151.0,
84
+ "learning_rate": 8.8e-06,
85
+ "loss": 1.4958,
86
+ "step": 110
87
+ },
88
+ {
89
+ "epoch": 0.05237315875613748,
90
+ "grad_norm": 45.25,
91
+ "learning_rate": 9.600000000000001e-06,
92
+ "loss": 1.741,
93
+ "step": 120
94
+ },
95
+ {
96
+ "epoch": 0.05673758865248227,
97
+ "grad_norm": 50.5,
98
+ "learning_rate": 1.04e-05,
99
+ "loss": 1.5346,
100
+ "step": 130
101
+ },
102
+ {
103
+ "epoch": 0.06110201854882706,
104
+ "grad_norm": 116.5,
105
+ "learning_rate": 1.1200000000000001e-05,
106
+ "loss": 1.498,
107
+ "step": 140
108
+ },
109
+ {
110
+ "epoch": 0.06546644844517185,
111
+ "grad_norm": 35.0,
112
+ "learning_rate": 1.2e-05,
113
+ "loss": 1.3559,
114
+ "step": 150
115
+ },
116
+ {
117
+ "epoch": 0.06983087834151663,
118
+ "grad_norm": 57.5,
119
+ "learning_rate": 1.2800000000000001e-05,
120
+ "loss": 1.3762,
121
+ "step": 160
122
+ },
123
+ {
124
+ "epoch": 0.07419530823786143,
125
+ "grad_norm": 40.25,
126
+ "learning_rate": 1.3600000000000002e-05,
127
+ "loss": 1.5166,
128
+ "step": 170
129
+ },
130
+ {
131
+ "epoch": 0.07855973813420622,
132
+ "grad_norm": 22.5,
133
+ "learning_rate": 1.4400000000000001e-05,
134
+ "loss": 1.3441,
135
+ "step": 180
136
+ },
137
+ {
138
+ "epoch": 0.08292416803055101,
139
+ "grad_norm": 14.875,
140
+ "learning_rate": 1.5200000000000002e-05,
141
+ "loss": 1.6288,
142
+ "step": 190
143
+ },
144
+ {
145
+ "epoch": 0.0872885979268958,
146
+ "grad_norm": 203.0,
147
+ "learning_rate": 1.6000000000000003e-05,
148
+ "loss": 3.1069,
149
+ "step": 200
150
+ },
151
+ {
152
+ "epoch": 0.09165302782324058,
153
+ "grad_norm": 57.5,
154
+ "learning_rate": 1.6800000000000002e-05,
155
+ "loss": 1.5319,
156
+ "step": 210
157
+ },
158
+ {
159
+ "epoch": 0.09601745771958538,
160
+ "grad_norm": 26.0,
161
+ "learning_rate": 1.76e-05,
162
+ "loss": 1.5801,
163
+ "step": 220
164
+ },
165
+ {
166
+ "epoch": 0.10038188761593017,
167
+ "grad_norm": 24.125,
168
+ "learning_rate": 1.8400000000000003e-05,
169
+ "loss": 1.3185,
170
+ "step": 230
171
+ },
172
+ {
173
+ "epoch": 0.10474631751227496,
174
+ "grad_norm": 29.875,
175
+ "learning_rate": 1.9200000000000003e-05,
176
+ "loss": 1.3118,
177
+ "step": 240
178
+ },
179
+ {
180
+ "epoch": 0.10911074740861974,
181
+ "grad_norm": 18.625,
182
+ "learning_rate": 2e-05,
183
+ "loss": 1.1643,
184
+ "step": 250
185
+ },
186
+ {
187
+ "epoch": 0.11347517730496454,
188
+ "grad_norm": 25.5,
189
+ "learning_rate": 1.9999025240093045e-05,
190
+ "loss": 1.3403,
191
+ "step": 260
192
+ },
193
+ {
194
+ "epoch": 0.11783960720130933,
195
+ "grad_norm": 20.75,
196
+ "learning_rate": 1.9996101150403543e-05,
197
+ "loss": 1.2914,
198
+ "step": 270
199
+ },
200
+ {
201
+ "epoch": 0.12220403709765412,
202
+ "grad_norm": 17.625,
203
+ "learning_rate": 1.9991228300988586e-05,
204
+ "loss": 1.2328,
205
+ "step": 280
206
+ },
207
+ {
208
+ "epoch": 0.1265684669939989,
209
+ "grad_norm": 37.25,
210
+ "learning_rate": 1.9984407641819812e-05,
211
+ "loss": 1.2335,
212
+ "step": 290
213
+ },
214
+ {
215
+ "epoch": 0.1309328968903437,
216
+ "grad_norm": 18.75,
217
+ "learning_rate": 1.9975640502598243e-05,
218
+ "loss": 1.1872,
219
+ "step": 300
220
+ },
221
+ {
222
+ "epoch": 0.1352973267866885,
223
+ "grad_norm": 22.625,
224
+ "learning_rate": 1.9964928592495046e-05,
225
+ "loss": 1.1274,
226
+ "step": 310
227
+ },
228
+ {
229
+ "epoch": 0.13966175668303327,
230
+ "grad_norm": 15.8125,
231
+ "learning_rate": 1.9952273999818312e-05,
232
+ "loss": 1.235,
233
+ "step": 320
234
+ },
235
+ {
236
+ "epoch": 0.14402618657937807,
237
+ "grad_norm": 16.5,
238
+ "learning_rate": 1.9937679191605964e-05,
239
+ "loss": 1.1159,
240
+ "step": 330
241
+ },
242
+ {
243
+ "epoch": 0.14839061647572285,
244
+ "grad_norm": 17.5,
245
+ "learning_rate": 1.9921147013144782e-05,
246
+ "loss": 1.0252,
247
+ "step": 340
248
+ },
249
+ {
250
+ "epoch": 0.15275504637206766,
251
+ "grad_norm": 6.78125,
252
+ "learning_rate": 1.9902680687415704e-05,
253
+ "loss": 1.0231,
254
+ "step": 350
255
+ },
256
+ {
257
+ "epoch": 0.15711947626841244,
258
+ "grad_norm": 20.375,
259
+ "learning_rate": 1.988228381446553e-05,
260
+ "loss": 1.086,
261
+ "step": 360
262
+ },
263
+ {
264
+ "epoch": 0.16148390616475722,
265
+ "grad_norm": 29.375,
266
+ "learning_rate": 1.985996037070505e-05,
267
+ "loss": 1.0803,
268
+ "step": 370
269
+ },
270
+ {
271
+ "epoch": 0.16584833606110203,
272
+ "grad_norm": 10.9375,
273
+ "learning_rate": 1.983571470813386e-05,
274
+ "loss": 0.987,
275
+ "step": 380
276
+ },
277
+ {
278
+ "epoch": 0.1702127659574468,
279
+ "grad_norm": 12.625,
280
+ "learning_rate": 1.9809551553491918e-05,
281
+ "loss": 1.2151,
282
+ "step": 390
283
+ },
284
+ {
285
+ "epoch": 0.1745771958537916,
286
+ "grad_norm": 14.125,
287
+ "learning_rate": 1.9781476007338058e-05,
288
+ "loss": 0.9436,
289
+ "step": 400
290
+ },
291
+ {
292
+ "epoch": 0.1789416257501364,
293
+ "grad_norm": 56.0,
294
+ "learning_rate": 1.9751493543055634e-05,
295
+ "loss": 1.0427,
296
+ "step": 410
297
+ },
298
+ {
299
+ "epoch": 0.18330605564648117,
300
+ "grad_norm": 17.5,
301
+ "learning_rate": 1.9719610005785466e-05,
302
+ "loss": 1.0463,
303
+ "step": 420
304
+ },
305
+ {
306
+ "epoch": 0.18767048554282598,
307
+ "grad_norm": 14.4375,
308
+ "learning_rate": 1.9685831611286312e-05,
309
+ "loss": 1.0375,
310
+ "step": 430
311
+ },
312
+ {
313
+ "epoch": 0.19203491543917076,
314
+ "grad_norm": 14.3125,
315
+ "learning_rate": 1.9650164944723116e-05,
316
+ "loss": 1.0024,
317
+ "step": 440
318
+ },
319
+ {
320
+ "epoch": 0.19639934533551553,
321
+ "grad_norm": 10.75,
322
+ "learning_rate": 1.961261695938319e-05,
323
+ "loss": 1.0858,
324
+ "step": 450
325
+ },
326
+ {
327
+ "epoch": 0.20076377523186034,
328
+ "grad_norm": 8.1875,
329
+ "learning_rate": 1.9573194975320672e-05,
330
+ "loss": 0.9583,
331
+ "step": 460
332
+ },
333
+ {
334
+ "epoch": 0.20512820512820512,
335
+ "grad_norm": 16.75,
336
+ "learning_rate": 1.9531906677929472e-05,
337
+ "loss": 1.048,
338
+ "step": 470
339
+ },
340
+ {
341
+ "epoch": 0.20949263502454993,
342
+ "grad_norm": 24.25,
343
+ "learning_rate": 1.9488760116444966e-05,
344
+ "loss": 0.9226,
345
+ "step": 480
346
+ },
347
+ {
348
+ "epoch": 0.2138570649208947,
349
+ "grad_norm": 12.125,
350
+ "learning_rate": 1.944376370237481e-05,
351
+ "loss": 1.0239,
352
+ "step": 490
353
+ },
354
+ {
355
+ "epoch": 0.2182214948172395,
356
+ "grad_norm": 15.9375,
357
+ "learning_rate": 1.9396926207859085e-05,
358
+ "loss": 0.9481,
359
+ "step": 500
360
+ },
361
+ {
362
+ "epoch": 0.2225859247135843,
363
+ "grad_norm": 8.0,
364
+ "learning_rate": 1.9348256763960146e-05,
365
+ "loss": 0.9285,
366
+ "step": 510
367
+ },
368
+ {
369
+ "epoch": 0.22695035460992907,
370
+ "grad_norm": 16.625,
371
+ "learning_rate": 1.9297764858882516e-05,
372
+ "loss": 0.9811,
373
+ "step": 520
374
+ },
375
+ {
376
+ "epoch": 0.23131478450627388,
377
+ "grad_norm": 7.34375,
378
+ "learning_rate": 1.9245460336123136e-05,
379
+ "loss": 0.8843,
380
+ "step": 530
381
+ },
382
+ {
383
+ "epoch": 0.23567921440261866,
384
+ "grad_norm": 12.5,
385
+ "learning_rate": 1.9191353392552346e-05,
386
+ "loss": 0.869,
387
+ "step": 540
388
+ },
389
+ {
390
+ "epoch": 0.24004364429896344,
391
+ "grad_norm": 10.25,
392
+ "learning_rate": 1.913545457642601e-05,
393
+ "loss": 0.9146,
394
+ "step": 550
395
+ },
396
+ {
397
+ "epoch": 0.24440807419530824,
398
+ "grad_norm": 20.75,
399
+ "learning_rate": 1.907777478532909e-05,
400
+ "loss": 0.904,
401
+ "step": 560
402
+ },
403
+ {
404
+ "epoch": 0.24877250409165302,
405
+ "grad_norm": 13.9375,
406
+ "learning_rate": 1.901832526405114e-05,
407
+ "loss": 0.8561,
408
+ "step": 570
409
+ },
410
+ {
411
+ "epoch": 0.2531369339879978,
412
+ "grad_norm": 9.25,
413
+ "learning_rate": 1.895711760239413e-05,
414
+ "loss": 0.9274,
415
+ "step": 580
416
+ },
417
+ {
418
+ "epoch": 0.2575013638843426,
419
+ "grad_norm": 9.5625,
420
+ "learning_rate": 1.889416373291298e-05,
421
+ "loss": 0.9191,
422
+ "step": 590
423
+ },
424
+ {
425
+ "epoch": 0.2618657937806874,
426
+ "grad_norm": 6.75,
427
+ "learning_rate": 1.8829475928589272e-05,
428
+ "loss": 0.9027,
429
+ "step": 600
430
+ },
431
+ {
432
+ "epoch": 0.2662302236770322,
433
+ "grad_norm": 9.5,
434
+ "learning_rate": 1.8763066800438638e-05,
435
+ "loss": 0.9427,
436
+ "step": 610
437
+ },
438
+ {
439
+ "epoch": 0.270594653573377,
440
+ "grad_norm": 13.125,
441
+ "learning_rate": 1.869494929505219e-05,
442
+ "loss": 0.8664,
443
+ "step": 620
444
+ },
445
+ {
446
+ "epoch": 0.27495908346972175,
447
+ "grad_norm": 19.375,
448
+ "learning_rate": 1.8625136692072577e-05,
449
+ "loss": 0.9021,
450
+ "step": 630
451
+ },
452
+ {
453
+ "epoch": 0.27932351336606653,
454
+ "grad_norm": 14.5,
455
+ "learning_rate": 1.855364260160507e-05,
456
+ "loss": 0.9041,
457
+ "step": 640
458
+ },
459
+ {
460
+ "epoch": 0.28368794326241137,
461
+ "grad_norm": 8.75,
462
+ "learning_rate": 1.848048096156426e-05,
463
+ "loss": 0.8482,
464
+ "step": 650
465
+ },
466
+ {
467
+ "epoch": 0.28805237315875615,
468
+ "grad_norm": 18.5,
469
+ "learning_rate": 1.8405666034956842e-05,
470
+ "loss": 0.9267,
471
+ "step": 660
472
+ },
473
+ {
474
+ "epoch": 0.2924168030551009,
475
+ "grad_norm": 5.875,
476
+ "learning_rate": 1.8329212407100996e-05,
477
+ "loss": 0.8678,
478
+ "step": 670
479
+ },
480
+ {
481
+ "epoch": 0.2967812329514457,
482
+ "grad_norm": 12.625,
483
+ "learning_rate": 1.8251134982782952e-05,
484
+ "loss": 0.8268,
485
+ "step": 680
486
+ },
487
+ {
488
+ "epoch": 0.3011456628477905,
489
+ "grad_norm": 12.875,
490
+ "learning_rate": 1.8171448983351284e-05,
491
+ "loss": 0.8265,
492
+ "step": 690
493
+ },
494
+ {
495
+ "epoch": 0.3055100927441353,
496
+ "grad_norm": 8.6875,
497
+ "learning_rate": 1.8090169943749477e-05,
498
+ "loss": 0.8773,
499
+ "step": 700
500
+ },
501
+ {
502
+ "epoch": 0.3098745226404801,
503
+ "grad_norm": 16.75,
504
+ "learning_rate": 1.8007313709487334e-05,
505
+ "loss": 0.8472,
506
+ "step": 710
507
+ },
508
+ {
509
+ "epoch": 0.3142389525368249,
510
+ "grad_norm": 7.53125,
511
+ "learning_rate": 1.792289643355191e-05,
512
+ "loss": 0.8534,
513
+ "step": 720
514
+ },
515
+ {
516
+ "epoch": 0.31860338243316966,
517
+ "grad_norm": 15.125,
518
+ "learning_rate": 1.78369345732584e-05,
519
+ "loss": 0.8069,
520
+ "step": 730
521
+ },
522
+ {
523
+ "epoch": 0.32296781232951444,
524
+ "grad_norm": 11.5,
525
+ "learning_rate": 1.7749444887041797e-05,
526
+ "loss": 0.8533,
527
+ "step": 740
528
+ },
529
+ {
530
+ "epoch": 0.32733224222585927,
531
+ "grad_norm": 14.75,
532
+ "learning_rate": 1.766044443118978e-05,
533
+ "loss": 0.8662,
534
+ "step": 750
535
+ },
536
+ {
537
+ "epoch": 0.33169667212220405,
538
+ "grad_norm": 15.75,
539
+ "learning_rate": 1.7569950556517566e-05,
540
+ "loss": 0.8397,
541
+ "step": 760
542
+ },
543
+ {
544
+ "epoch": 0.33606110201854883,
545
+ "grad_norm": 19.875,
546
+ "learning_rate": 1.747798090498532e-05,
547
+ "loss": 0.7535,
548
+ "step": 770
549
+ },
550
+ {
551
+ "epoch": 0.3404255319148936,
552
+ "grad_norm": 11.75,
553
+ "learning_rate": 1.7384553406258842e-05,
554
+ "loss": 0.8573,
555
+ "step": 780
556
+ },
557
+ {
558
+ "epoch": 0.3447899618112384,
559
+ "grad_norm": 6.9375,
560
+ "learning_rate": 1.7289686274214116e-05,
561
+ "loss": 0.8559,
562
+ "step": 790
563
+ },
564
+ {
565
+ "epoch": 0.3491543917075832,
566
+ "grad_norm": 10.9375,
567
+ "learning_rate": 1.7193398003386514e-05,
568
+ "loss": 0.8459,
569
+ "step": 800
570
+ },
571
+ {
572
+ "epoch": 0.353518821603928,
573
+ "grad_norm": 12.0625,
574
+ "learning_rate": 1.709570736536521e-05,
575
+ "loss": 0.8352,
576
+ "step": 810
577
+ },
578
+ {
579
+ "epoch": 0.3578832515002728,
580
+ "grad_norm": 4.59375,
581
+ "learning_rate": 1.6996633405133656e-05,
582
+ "loss": 0.8146,
583
+ "step": 820
584
+ },
585
+ {
586
+ "epoch": 0.36224768139661756,
587
+ "grad_norm": 15.0,
588
+ "learning_rate": 1.68961954373567e-05,
589
+ "loss": 0.78,
590
+ "step": 830
591
+ },
592
+ {
593
+ "epoch": 0.36661211129296234,
594
+ "grad_norm": 5.6875,
595
+ "learning_rate": 1.6794413042615168e-05,
596
+ "loss": 0.8002,
597
+ "step": 840
598
+ },
599
+ {
600
+ "epoch": 0.3709765411893072,
601
+ "grad_norm": 18.5,
602
+ "learning_rate": 1.6691306063588583e-05,
603
+ "loss": 0.8296,
604
+ "step": 850
605
+ },
606
+ {
607
+ "epoch": 0.37534097108565195,
608
+ "grad_norm": 42.0,
609
+ "learning_rate": 1.6586894601186804e-05,
610
+ "loss": 0.8742,
611
+ "step": 860
612
+ },
613
+ {
614
+ "epoch": 0.37970540098199673,
615
+ "grad_norm": 9.5,
616
+ "learning_rate": 1.6481199010631312e-05,
617
+ "loss": 0.779,
618
+ "step": 870
619
+ },
620
+ {
621
+ "epoch": 0.3840698308783415,
622
+ "grad_norm": 6.0625,
623
+ "learning_rate": 1.63742398974869e-05,
624
+ "loss": 0.747,
625
+ "step": 880
626
+ },
627
+ {
628
+ "epoch": 0.3884342607746863,
629
+ "grad_norm": 9.375,
630
+ "learning_rate": 1.6266038113644605e-05,
631
+ "loss": 0.8371,
632
+ "step": 890
633
+ },
634
+ {
635
+ "epoch": 0.39279869067103107,
636
+ "grad_norm": 7.9375,
637
+ "learning_rate": 1.6156614753256583e-05,
638
+ "loss": 0.7598,
639
+ "step": 900
640
+ },
641
+ {
642
+ "epoch": 0.3971631205673759,
643
+ "grad_norm": 14.75,
644
+ "learning_rate": 1.6045991148623752e-05,
645
+ "loss": 0.7371,
646
+ "step": 910
647
+ },
648
+ {
649
+ "epoch": 0.4015275504637207,
650
+ "grad_norm": 5.09375,
651
+ "learning_rate": 1.5934188866037017e-05,
652
+ "loss": 0.7388,
653
+ "step": 920
654
+ },
655
+ {
656
+ "epoch": 0.40589198036006546,
657
+ "grad_norm": 7.96875,
658
+ "learning_rate": 1.5821229701572897e-05,
659
+ "loss": 0.7652,
660
+ "step": 930
661
+ },
662
+ {
663
+ "epoch": 0.41025641025641024,
664
+ "grad_norm": 12.3125,
665
+ "learning_rate": 1.570713567684432e-05,
666
+ "loss": 0.82,
667
+ "step": 940
668
+ },
669
+ {
670
+ "epoch": 0.414620840152755,
671
+ "grad_norm": 9.5,
672
+ "learning_rate": 1.5591929034707468e-05,
673
+ "loss": 0.7934,
674
+ "step": 950
675
+ },
676
+ {
677
+ "epoch": 0.41898527004909986,
678
+ "grad_norm": 8.5625,
679
+ "learning_rate": 1.5475632234925505e-05,
680
+ "loss": 0.7752,
681
+ "step": 960
682
+ },
683
+ {
684
+ "epoch": 0.42334969994544464,
685
+ "grad_norm": 8.375,
686
+ "learning_rate": 1.5358267949789968e-05,
687
+ "loss": 0.7438,
688
+ "step": 970
689
+ },
690
+ {
691
+ "epoch": 0.4277141298417894,
692
+ "grad_norm": 7.6875,
693
+ "learning_rate": 1.5239859059700794e-05,
694
+ "loss": 0.7512,
695
+ "step": 980
696
+ },
697
+ {
698
+ "epoch": 0.4320785597381342,
699
+ "grad_norm": 9.1875,
700
+ "learning_rate": 1.5120428648705716e-05,
701
+ "loss": 0.6996,
702
+ "step": 990
703
+ },
704
+ {
705
+ "epoch": 0.436442989634479,
706
+ "grad_norm": 6.0625,
707
+ "learning_rate": 1.5000000000000002e-05,
708
+ "loss": 0.7038,
709
+ "step": 1000
710
+ },
711
+ {
712
+ "epoch": 0.4408074195308238,
713
+ "grad_norm": 8.9375,
714
+ "learning_rate": 1.4878596591387329e-05,
715
+ "loss": 0.7495,
716
+ "step": 1010
717
+ },
718
+ {
719
+ "epoch": 0.4451718494271686,
720
+ "grad_norm": 6.59375,
721
+ "learning_rate": 1.4756242090702756e-05,
722
+ "loss": 0.7794,
723
+ "step": 1020
724
+ },
725
+ {
726
+ "epoch": 0.44953627932351337,
727
+ "grad_norm": 10.0625,
728
+ "learning_rate": 1.463296035119862e-05,
729
+ "loss": 0.747,
730
+ "step": 1030
731
+ },
732
+ {
733
+ "epoch": 0.45390070921985815,
734
+ "grad_norm": 11.4375,
735
+ "learning_rate": 1.4508775406894308e-05,
736
+ "loss": 0.7394,
737
+ "step": 1040
738
+ },
739
+ {
740
+ "epoch": 0.4582651391162029,
741
+ "grad_norm": 8.375,
742
+ "learning_rate": 1.4383711467890776e-05,
743
+ "loss": 0.6999,
744
+ "step": 1050
745
+ },
746
+ {
747
+ "epoch": 0.46262956901254776,
748
+ "grad_norm": 11.9375,
749
+ "learning_rate": 1.4257792915650728e-05,
750
+ "loss": 0.7475,
751
+ "step": 1060
752
+ },
753
+ {
754
+ "epoch": 0.46699399890889254,
755
+ "grad_norm": 4.125,
756
+ "learning_rate": 1.413104429824542e-05,
757
+ "loss": 0.6836,
758
+ "step": 1070
759
+ },
760
+ {
761
+ "epoch": 0.4713584288052373,
762
+ "grad_norm": 6.5,
763
+ "learning_rate": 1.4003490325568953e-05,
764
+ "loss": 0.6974,
765
+ "step": 1080
766
+ },
767
+ {
768
+ "epoch": 0.4757228587015821,
769
+ "grad_norm": 8.375,
770
+ "learning_rate": 1.3875155864521031e-05,
771
+ "loss": 0.7776,
772
+ "step": 1090
773
+ },
774
+ {
775
+ "epoch": 0.4800872885979269,
776
+ "grad_norm": 11.125,
777
+ "learning_rate": 1.3746065934159123e-05,
778
+ "loss": 0.7633,
779
+ "step": 1100
780
+ },
781
+ {
782
+ "epoch": 0.4844517184942717,
783
+ "grad_norm": 6.09375,
784
+ "learning_rate": 1.3616245700820922e-05,
785
+ "loss": 0.7146,
786
+ "step": 1110
787
+ },
788
+ {
789
+ "epoch": 0.4888161483906165,
790
+ "grad_norm": 54.25,
791
+ "learning_rate": 1.3485720473218153e-05,
792
+ "loss": 0.7503,
793
+ "step": 1120
794
+ },
795
+ {
796
+ "epoch": 0.49318057828696127,
797
+ "grad_norm": 9.0,
798
+ "learning_rate": 1.3354515697502552e-05,
799
+ "loss": 0.6919,
800
+ "step": 1130
801
+ },
802
+ {
803
+ "epoch": 0.49754500818330605,
804
+ "grad_norm": 5.65625,
805
+ "learning_rate": 1.3222656952305113e-05,
806
+ "loss": 0.702,
807
+ "step": 1140
808
+ },
809
+ {
810
+ "epoch": 0.5019094380796508,
811
+ "grad_norm": 8.125,
812
+ "learning_rate": 1.3090169943749475e-05,
813
+ "loss": 0.7013,
814
+ "step": 1150
815
+ },
816
+ {
817
+ "epoch": 0.5062738679759956,
818
+ "grad_norm": 7.125,
819
+ "learning_rate": 1.2957080500440469e-05,
820
+ "loss": 0.6457,
821
+ "step": 1160
822
+ },
823
+ {
824
+ "epoch": 0.5106382978723404,
825
+ "grad_norm": 3.59375,
826
+ "learning_rate": 1.2823414568428767e-05,
827
+ "loss": 0.7375,
828
+ "step": 1170
829
+ },
830
+ {
831
+ "epoch": 0.5150027277686852,
832
+ "grad_norm": 7.53125,
833
+ "learning_rate": 1.2689198206152657e-05,
834
+ "loss": 0.6741,
835
+ "step": 1180
836
+ },
837
+ {
838
+ "epoch": 0.51936715766503,
839
+ "grad_norm": 5.5625,
840
+ "learning_rate": 1.2554457579357906e-05,
841
+ "loss": 0.6753,
842
+ "step": 1190
843
+ },
844
+ {
845
+ "epoch": 0.5237315875613748,
846
+ "grad_norm": 15.5625,
847
+ "learning_rate": 1.2419218955996677e-05,
848
+ "loss": 0.715,
849
+ "step": 1200
850
+ },
851
+ {
852
+ "epoch": 0.5280960174577196,
853
+ "grad_norm": 15.4375,
854
+ "learning_rate": 1.2283508701106559e-05,
855
+ "loss": 0.6916,
856
+ "step": 1210
857
+ },
858
+ {
859
+ "epoch": 0.5324604473540644,
860
+ "grad_norm": 5.90625,
861
+ "learning_rate": 1.2147353271670634e-05,
862
+ "loss": 0.7213,
863
+ "step": 1220
864
+ },
865
+ {
866
+ "epoch": 0.5368248772504092,
867
+ "grad_norm": 3.3125,
868
+ "learning_rate": 1.2010779211459649e-05,
869
+ "loss": 0.6672,
870
+ "step": 1230
871
+ },
872
+ {
873
+ "epoch": 0.541189307146754,
874
+ "grad_norm": 4.46875,
875
+ "learning_rate": 1.187381314585725e-05,
876
+ "loss": 0.6987,
877
+ "step": 1240
878
+ },
879
+ {
880
+ "epoch": 0.5455537370430987,
881
+ "grad_norm": 9.3125,
882
+ "learning_rate": 1.1736481776669307e-05,
883
+ "loss": 0.6964,
884
+ "step": 1250
885
+ },
886
+ {
887
+ "epoch": 0.5499181669394435,
888
+ "grad_norm": 11.625,
889
+ "learning_rate": 1.159881187691835e-05,
890
+ "loss": 0.689,
891
+ "step": 1260
892
+ },
893
+ {
894
+ "epoch": 0.5542825968357883,
895
+ "grad_norm": 8.4375,
896
+ "learning_rate": 1.1460830285624119e-05,
897
+ "loss": 0.662,
898
+ "step": 1270
899
+ },
900
+ {
901
+ "epoch": 0.5586470267321331,
902
+ "grad_norm": 12.875,
903
+ "learning_rate": 1.1322563902571227e-05,
904
+ "loss": 0.6413,
905
+ "step": 1280
906
+ },
907
+ {
908
+ "epoch": 0.563011456628478,
909
+ "grad_norm": 5.78125,
910
+ "learning_rate": 1.1184039683065014e-05,
911
+ "loss": 0.6609,
912
+ "step": 1290
913
+ },
914
+ {
915
+ "epoch": 0.5673758865248227,
916
+ "grad_norm": 6.0,
917
+ "learning_rate": 1.1045284632676535e-05,
918
+ "loss": 0.6716,
919
+ "step": 1300
920
+ },
921
+ {
922
+ "epoch": 0.5717403164211675,
923
+ "grad_norm": 9.0,
924
+ "learning_rate": 1.0906325801977804e-05,
925
+ "loss": 0.692,
926
+ "step": 1310
927
+ },
928
+ {
929
+ "epoch": 0.5761047463175123,
930
+ "grad_norm": 4.1875,
931
+ "learning_rate": 1.0767190281268187e-05,
932
+ "loss": 0.5993,
933
+ "step": 1320
934
+ },
935
+ {
936
+ "epoch": 0.5804691762138571,
937
+ "grad_norm": 3.953125,
938
+ "learning_rate": 1.0627905195293135e-05,
939
+ "loss": 0.6846,
940
+ "step": 1330
941
+ },
942
+ {
943
+ "epoch": 0.5848336061102019,
944
+ "grad_norm": 5.65625,
945
+ "learning_rate": 1.0488497697956134e-05,
946
+ "loss": 0.6426,
947
+ "step": 1340
948
+ },
949
+ {
950
+ "epoch": 0.5891980360065466,
951
+ "grad_norm": 3.796875,
952
+ "learning_rate": 1.0348994967025012e-05,
953
+ "loss": 0.7003,
954
+ "step": 1350
955
+ },
956
+ {
957
+ "epoch": 0.5935624659028914,
958
+ "grad_norm": 4.53125,
959
+ "learning_rate": 1.0209424198833571e-05,
960
+ "loss": 0.6536,
961
+ "step": 1360
962
+ },
963
+ {
964
+ "epoch": 0.5979268957992362,
965
+ "grad_norm": 6.0625,
966
+ "learning_rate": 1.0069812602979617e-05,
967
+ "loss": 0.6849,
968
+ "step": 1370
969
+ },
970
+ {
971
+ "epoch": 0.602291325695581,
972
+ "grad_norm": 4.34375,
973
+ "learning_rate": 9.930187397020385e-06,
974
+ "loss": 0.672,
975
+ "step": 1380
976
+ },
977
+ {
978
+ "epoch": 0.6066557555919257,
979
+ "grad_norm": 7.84375,
980
+ "learning_rate": 9.790575801166432e-06,
981
+ "loss": 0.715,
982
+ "step": 1390
983
+ },
984
+ {
985
+ "epoch": 0.6110201854882706,
986
+ "grad_norm": 3.078125,
987
+ "learning_rate": 9.651005032974994e-06,
988
+ "loss": 0.7157,
989
+ "step": 1400
990
+ },
991
+ {
992
+ "epoch": 0.6153846153846154,
993
+ "grad_norm": 9.1875,
994
+ "learning_rate": 9.511502302043867e-06,
995
+ "loss": 0.6274,
996
+ "step": 1410
997
+ },
998
+ {
999
+ "epoch": 0.6197490452809602,
1000
+ "grad_norm": 3.84375,
1001
+ "learning_rate": 9.372094804706867e-06,
1002
+ "loss": 0.6586,
1003
+ "step": 1420
1004
+ },
1005
+ {
1006
+ "epoch": 0.624113475177305,
1007
+ "grad_norm": 4.15625,
1008
+ "learning_rate": 9.232809718731815e-06,
1009
+ "loss": 0.623,
1010
+ "step": 1430
1011
+ },
1012
+ {
1013
+ "epoch": 0.6284779050736498,
1014
+ "grad_norm": 4.4375,
1015
+ "learning_rate": 9.093674198022201e-06,
1016
+ "loss": 0.6606,
1017
+ "step": 1440
1018
+ },
1019
+ {
1020
+ "epoch": 0.6328423349699945,
1021
+ "grad_norm": 8.25,
1022
+ "learning_rate": 8.954715367323468e-06,
1023
+ "loss": 0.6573,
1024
+ "step": 1450
1025
+ },
1026
+ {
1027
+ "epoch": 0.6372067648663393,
1028
+ "grad_norm": 3.671875,
1029
+ "learning_rate": 8.815960316934991e-06,
1030
+ "loss": 0.6806,
1031
+ "step": 1460
1032
+ },
1033
+ {
1034
+ "epoch": 0.6415711947626841,
1035
+ "grad_norm": 9.6875,
1036
+ "learning_rate": 8.677436097428775e-06,
1037
+ "loss": 0.6407,
1038
+ "step": 1470
1039
+ },
1040
+ {
1041
+ "epoch": 0.6459356246590289,
1042
+ "grad_norm": 8.5,
1043
+ "learning_rate": 8.539169714375885e-06,
1044
+ "loss": 0.7079,
1045
+ "step": 1480
1046
+ },
1047
+ {
1048
+ "epoch": 0.6503000545553737,
1049
+ "grad_norm": 4.3125,
1050
+ "learning_rate": 8.401188123081653e-06,
1051
+ "loss": 0.6743,
1052
+ "step": 1490
1053
+ },
1054
+ {
1055
+ "epoch": 0.6546644844517185,
1056
+ "grad_norm": 5.28125,
1057
+ "learning_rate": 8.263518223330698e-06,
1058
+ "loss": 0.5891,
1059
+ "step": 1500
1060
+ }
1061
+ ],
1062
+ "logging_steps": 10,
1063
+ "max_steps": 2500,
1064
+ "num_input_tokens_seen": 0,
1065
+ "num_train_epochs": 2,
1066
+ "save_steps": 500,
1067
+ "stateful_callbacks": {
1068
+ "TrainerControl": {
1069
+ "args": {
1070
+ "should_epoch_stop": false,
1071
+ "should_evaluate": false,
1072
+ "should_log": false,
1073
+ "should_save": true,
1074
+ "should_training_stop": false
1075
+ },
1076
+ "attributes": {}
1077
+ }
1078
+ },
1079
+ "total_flos": 5.73984932364288e+17,
1080
+ "train_batch_size": 4,
1081
+ "trial_name": null,
1082
+ "trial_params": null
1083
+ }
checkpoint-1500/training_args.bin ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:0c5793f646fe6f22e93a8017b9608e8c23d00d4edcc0f719ae635844fcd92f55
3
+ size 5368