Grogros commited on
Commit
f2e7fce
·
verified ·
1 Parent(s): 59a57f7

Training in progress, step 2000, checkpoint

Browse files
checkpoint-2000/config.json ADDED
@@ -0,0 +1,39 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "architectures": [
3
+ "LlamaForCausalLM"
4
+ ],
5
+ "attention_bias": false,
6
+ "attention_dropout": 0.0,
7
+ "bos_token_id": 128000,
8
+ "eos_token_id": [
9
+ 128001,
10
+ 128008,
11
+ 128009
12
+ ],
13
+ "head_dim": 64,
14
+ "hidden_act": "silu",
15
+ "hidden_size": 2048,
16
+ "initializer_range": 0.02,
17
+ "intermediate_size": 8192,
18
+ "max_position_embeddings": 131072,
19
+ "mlp_bias": false,
20
+ "model_type": "llama",
21
+ "num_attention_heads": 32,
22
+ "num_hidden_layers": 16,
23
+ "num_key_value_heads": 8,
24
+ "pretraining_tp": 1,
25
+ "rms_norm_eps": 1e-05,
26
+ "rope_scaling": {
27
+ "factor": 32.0,
28
+ "high_freq_factor": 4.0,
29
+ "low_freq_factor": 1.0,
30
+ "original_max_position_embeddings": 8192,
31
+ "rope_type": "llama3"
32
+ },
33
+ "rope_theta": 500000.0,
34
+ "tie_word_embeddings": true,
35
+ "torch_dtype": "bfloat16",
36
+ "transformers_version": "4.51.3",
37
+ "use_cache": true,
38
+ "vocab_size": 128256
39
+ }
checkpoint-2000/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": "4.51.3"
12
+ }
checkpoint-2000/model.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:0376df7a50c87a3cfcabf9936d1c1a7c607b6fcdd06cf21f0f305410719f47c5
3
+ size 2471645608
checkpoint-2000/optimizer.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:2c6b4d2430fb415befe9825a749ff5c0c01ecb0b2573a4bdf98dc72c9e71b94e
3
+ size 3724602
checkpoint-2000/rng_state.pth ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:72699c98ff774edfda1e2ca07ac6fee6bb35ef78b5e8d27ba2903d37a4968363
3
+ size 14244
checkpoint-2000/scheduler.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:ec54b58fc9956250e312b35ee776338d7eeeb2e73eda4ecadc6702134123107e
3
+ size 1064
checkpoint-2000/trainer_state.json ADDED
@@ -0,0 +1,1434 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "best_global_step": null,
3
+ "best_metric": null,
4
+ "best_model_checkpoint": null,
5
+ "epoch": 1.0,
6
+ "eval_steps": 500,
7
+ "global_step": 2000,
8
+ "is_hyper_param_search": false,
9
+ "is_local_process_zero": true,
10
+ "is_world_process_zero": true,
11
+ "log_history": [
12
+ {
13
+ "epoch": 0.005,
14
+ "grad_norm": 181.0,
15
+ "learning_rate": 2.2500000000000002e-07,
16
+ "loss": 0.5867,
17
+ "step": 10
18
+ },
19
+ {
20
+ "epoch": 0.01,
21
+ "grad_norm": 178.0,
22
+ "learning_rate": 4.7500000000000006e-07,
23
+ "loss": 0.5369,
24
+ "step": 20
25
+ },
26
+ {
27
+ "epoch": 0.015,
28
+ "grad_norm": 185.0,
29
+ "learning_rate": 7.25e-07,
30
+ "loss": 0.6025,
31
+ "step": 30
32
+ },
33
+ {
34
+ "epoch": 0.02,
35
+ "grad_norm": 159.0,
36
+ "learning_rate": 9.750000000000002e-07,
37
+ "loss": 0.5992,
38
+ "step": 40
39
+ },
40
+ {
41
+ "epoch": 0.025,
42
+ "grad_norm": 145.0,
43
+ "learning_rate": 1.2250000000000001e-06,
44
+ "loss": 0.6375,
45
+ "step": 50
46
+ },
47
+ {
48
+ "epoch": 0.03,
49
+ "grad_norm": 7.96875,
50
+ "learning_rate": 1.475e-06,
51
+ "loss": 0.6512,
52
+ "step": 60
53
+ },
54
+ {
55
+ "epoch": 0.035,
56
+ "grad_norm": 5.9375,
57
+ "learning_rate": 1.725e-06,
58
+ "loss": 0.5709,
59
+ "step": 70
60
+ },
61
+ {
62
+ "epoch": 0.04,
63
+ "grad_norm": 37.75,
64
+ "learning_rate": 1.975e-06,
65
+ "loss": 0.5432,
66
+ "step": 80
67
+ },
68
+ {
69
+ "epoch": 0.045,
70
+ "grad_norm": 4.03125,
71
+ "learning_rate": 2.2250000000000003e-06,
72
+ "loss": 0.5684,
73
+ "step": 90
74
+ },
75
+ {
76
+ "epoch": 0.05,
77
+ "grad_norm": 81.5,
78
+ "learning_rate": 2.475e-06,
79
+ "loss": 0.5439,
80
+ "step": 100
81
+ },
82
+ {
83
+ "epoch": 0.055,
84
+ "grad_norm": 4.125,
85
+ "learning_rate": 2.7250000000000006e-06,
86
+ "loss": 0.5053,
87
+ "step": 110
88
+ },
89
+ {
90
+ "epoch": 0.06,
91
+ "grad_norm": 4.03125,
92
+ "learning_rate": 2.9750000000000003e-06,
93
+ "loss": 0.5178,
94
+ "step": 120
95
+ },
96
+ {
97
+ "epoch": 0.065,
98
+ "grad_norm": 2.296875,
99
+ "learning_rate": 3.2250000000000005e-06,
100
+ "loss": 0.5037,
101
+ "step": 130
102
+ },
103
+ {
104
+ "epoch": 0.07,
105
+ "grad_norm": 31.125,
106
+ "learning_rate": 3.475e-06,
107
+ "loss": 0.4973,
108
+ "step": 140
109
+ },
110
+ {
111
+ "epoch": 0.075,
112
+ "grad_norm": 4.09375,
113
+ "learning_rate": 3.7250000000000003e-06,
114
+ "loss": 0.5008,
115
+ "step": 150
116
+ },
117
+ {
118
+ "epoch": 0.08,
119
+ "grad_norm": 2.890625,
120
+ "learning_rate": 3.975000000000001e-06,
121
+ "loss": 0.5391,
122
+ "step": 160
123
+ },
124
+ {
125
+ "epoch": 0.085,
126
+ "grad_norm": 2.984375,
127
+ "learning_rate": 4.225e-06,
128
+ "loss": 0.4914,
129
+ "step": 170
130
+ },
131
+ {
132
+ "epoch": 0.09,
133
+ "grad_norm": 2.609375,
134
+ "learning_rate": 4.475e-06,
135
+ "loss": 0.5141,
136
+ "step": 180
137
+ },
138
+ {
139
+ "epoch": 0.095,
140
+ "grad_norm": 3.25,
141
+ "learning_rate": 4.7250000000000005e-06,
142
+ "loss": 0.4637,
143
+ "step": 190
144
+ },
145
+ {
146
+ "epoch": 0.1,
147
+ "grad_norm": 11.625,
148
+ "learning_rate": 4.975000000000001e-06,
149
+ "loss": 0.4551,
150
+ "step": 200
151
+ },
152
+ {
153
+ "epoch": 0.105,
154
+ "grad_norm": 2.21875,
155
+ "learning_rate": 4.9996915812041515e-06,
156
+ "loss": 0.4729,
157
+ "step": 210
158
+ },
159
+ {
160
+ "epoch": 0.11,
161
+ "grad_norm": 2.09375,
162
+ "learning_rate": 4.998625539854394e-06,
163
+ "loss": 0.4775,
164
+ "step": 220
165
+ },
166
+ {
167
+ "epoch": 0.115,
168
+ "grad_norm": 2.46875,
169
+ "learning_rate": 4.996798392960466e-06,
170
+ "loss": 0.4682,
171
+ "step": 230
172
+ },
173
+ {
174
+ "epoch": 0.12,
175
+ "grad_norm": 1.9375,
176
+ "learning_rate": 4.9942106970890136e-06,
177
+ "loss": 0.4684,
178
+ "step": 240
179
+ },
180
+ {
181
+ "epoch": 0.125,
182
+ "grad_norm": 2.375,
183
+ "learning_rate": 4.990863240477266e-06,
184
+ "loss": 0.4549,
185
+ "step": 250
186
+ },
187
+ {
188
+ "epoch": 0.13,
189
+ "grad_norm": 2.21875,
190
+ "learning_rate": 4.9867570427929356e-06,
191
+ "loss": 0.4758,
192
+ "step": 260
193
+ },
194
+ {
195
+ "epoch": 0.135,
196
+ "grad_norm": 2.328125,
197
+ "learning_rate": 4.981893354823614e-06,
198
+ "loss": 0.4531,
199
+ "step": 270
200
+ },
201
+ {
202
+ "epoch": 0.14,
203
+ "grad_norm": 2.1875,
204
+ "learning_rate": 4.976273658095772e-06,
205
+ "loss": 0.4609,
206
+ "step": 280
207
+ },
208
+ {
209
+ "epoch": 0.145,
210
+ "grad_norm": 1.984375,
211
+ "learning_rate": 4.969899664423473e-06,
212
+ "loss": 0.4629,
213
+ "step": 290
214
+ },
215
+ {
216
+ "epoch": 0.15,
217
+ "grad_norm": 1.96875,
218
+ "learning_rate": 4.962773315386935e-06,
219
+ "loss": 0.4469,
220
+ "step": 300
221
+ },
222
+ {
223
+ "epoch": 0.155,
224
+ "grad_norm": 2.359375,
225
+ "learning_rate": 4.95489678174111e-06,
226
+ "loss": 0.5066,
227
+ "step": 310
228
+ },
229
+ {
230
+ "epoch": 0.16,
231
+ "grad_norm": 1.9375,
232
+ "learning_rate": 4.946272462754447e-06,
233
+ "loss": 0.4465,
234
+ "step": 320
235
+ },
236
+ {
237
+ "epoch": 0.165,
238
+ "grad_norm": 2.0625,
239
+ "learning_rate": 4.936902985478055e-06,
240
+ "loss": 0.4424,
241
+ "step": 330
242
+ },
243
+ {
244
+ "epoch": 0.17,
245
+ "grad_norm": 1.8671875,
246
+ "learning_rate": 4.926791203945477e-06,
247
+ "loss": 0.4582,
248
+ "step": 340
249
+ },
250
+ {
251
+ "epoch": 0.175,
252
+ "grad_norm": 2.921875,
253
+ "learning_rate": 4.915940198303324e-06,
254
+ "loss": 0.4988,
255
+ "step": 350
256
+ },
257
+ {
258
+ "epoch": 0.18,
259
+ "grad_norm": 2.0625,
260
+ "learning_rate": 4.904353273873029e-06,
261
+ "loss": 0.458,
262
+ "step": 360
263
+ },
264
+ {
265
+ "epoch": 0.185,
266
+ "grad_norm": 2.03125,
267
+ "learning_rate": 4.89203396014402e-06,
268
+ "loss": 0.4227,
269
+ "step": 370
270
+ },
271
+ {
272
+ "epoch": 0.19,
273
+ "grad_norm": 2.046875,
274
+ "learning_rate": 4.878986009698596e-06,
275
+ "loss": 0.4311,
276
+ "step": 380
277
+ },
278
+ {
279
+ "epoch": 0.195,
280
+ "grad_norm": 2.0,
281
+ "learning_rate": 4.865213397068864e-06,
282
+ "loss": 0.4676,
283
+ "step": 390
284
+ },
285
+ {
286
+ "epoch": 0.2,
287
+ "grad_norm": 1.8984375,
288
+ "learning_rate": 4.850720317526047e-06,
289
+ "loss": 0.4613,
290
+ "step": 400
291
+ },
292
+ {
293
+ "epoch": 0.205,
294
+ "grad_norm": 2.359375,
295
+ "learning_rate": 4.835511185802574e-06,
296
+ "loss": 0.4557,
297
+ "step": 410
298
+ },
299
+ {
300
+ "epoch": 0.21,
301
+ "grad_norm": 2.375,
302
+ "learning_rate": 4.8195906347473e-06,
303
+ "loss": 0.4549,
304
+ "step": 420
305
+ },
306
+ {
307
+ "epoch": 0.215,
308
+ "grad_norm": 2.0,
309
+ "learning_rate": 4.802963513914304e-06,
310
+ "loss": 0.4551,
311
+ "step": 430
312
+ },
313
+ {
314
+ "epoch": 0.22,
315
+ "grad_norm": 1.984375,
316
+ "learning_rate": 4.7856348880856595e-06,
317
+ "loss": 0.4965,
318
+ "step": 440
319
+ },
320
+ {
321
+ "epoch": 0.225,
322
+ "grad_norm": 1.8046875,
323
+ "learning_rate": 4.767610035728663e-06,
324
+ "loss": 0.4322,
325
+ "step": 450
326
+ },
327
+ {
328
+ "epoch": 0.23,
329
+ "grad_norm": 1.84375,
330
+ "learning_rate": 4.7488944473879515e-06,
331
+ "loss": 0.4396,
332
+ "step": 460
333
+ },
334
+ {
335
+ "epoch": 0.235,
336
+ "grad_norm": 2.484375,
337
+ "learning_rate": 4.729493824013036e-06,
338
+ "loss": 0.4568,
339
+ "step": 470
340
+ },
341
+ {
342
+ "epoch": 0.24,
343
+ "grad_norm": 1.6328125,
344
+ "learning_rate": 4.709414075221734e-06,
345
+ "loss": 0.4641,
346
+ "step": 480
347
+ },
348
+ {
349
+ "epoch": 0.245,
350
+ "grad_norm": 2.890625,
351
+ "learning_rate": 4.688661317500045e-06,
352
+ "loss": 0.4416,
353
+ "step": 490
354
+ },
355
+ {
356
+ "epoch": 0.25,
357
+ "grad_norm": 1.875,
358
+ "learning_rate": 4.667241872339007e-06,
359
+ "loss": 0.4379,
360
+ "step": 500
361
+ },
362
+ {
363
+ "epoch": 0.255,
364
+ "grad_norm": 1.9140625,
365
+ "learning_rate": 4.645162264309112e-06,
366
+ "loss": 0.4637,
367
+ "step": 510
368
+ },
369
+ {
370
+ "epoch": 0.26,
371
+ "grad_norm": 2.0625,
372
+ "learning_rate": 4.622429219072854e-06,
373
+ "loss": 0.4652,
374
+ "step": 520
375
+ },
376
+ {
377
+ "epoch": 0.265,
378
+ "grad_norm": 1.59375,
379
+ "learning_rate": 4.599049661336033e-06,
380
+ "loss": 0.4172,
381
+ "step": 530
382
+ },
383
+ {
384
+ "epoch": 0.27,
385
+ "grad_norm": 1.625,
386
+ "learning_rate": 4.5750307127384194e-06,
387
+ "loss": 0.4557,
388
+ "step": 540
389
+ },
390
+ {
391
+ "epoch": 0.275,
392
+ "grad_norm": 2.109375,
393
+ "learning_rate": 4.550379689684431e-06,
394
+ "loss": 0.46,
395
+ "step": 550
396
+ },
397
+ {
398
+ "epoch": 0.28,
399
+ "grad_norm": 2.296875,
400
+ "learning_rate": 4.5251041011144905e-06,
401
+ "loss": 0.4445,
402
+ "step": 560
403
+ },
404
+ {
405
+ "epoch": 0.285,
406
+ "grad_norm": 2.203125,
407
+ "learning_rate": 4.4992116462177274e-06,
408
+ "loss": 0.4826,
409
+ "step": 570
410
+ },
411
+ {
412
+ "epoch": 0.29,
413
+ "grad_norm": 1.765625,
414
+ "learning_rate": 4.4727102120867274e-06,
415
+ "loss": 0.4227,
416
+ "step": 580
417
+ },
418
+ {
419
+ "epoch": 0.295,
420
+ "grad_norm": 1.5546875,
421
+ "learning_rate": 4.445607871315053e-06,
422
+ "loss": 0.4412,
423
+ "step": 590
424
+ },
425
+ {
426
+ "epoch": 0.3,
427
+ "grad_norm": 2.46875,
428
+ "learning_rate": 4.41791287953825e-06,
429
+ "loss": 0.457,
430
+ "step": 600
431
+ },
432
+ {
433
+ "epoch": 0.305,
434
+ "grad_norm": 1.8515625,
435
+ "learning_rate": 4.389633672919099e-06,
436
+ "loss": 0.4262,
437
+ "step": 610
438
+ },
439
+ {
440
+ "epoch": 0.31,
441
+ "grad_norm": 2.3125,
442
+ "learning_rate": 4.360778865577885e-06,
443
+ "loss": 0.466,
444
+ "step": 620
445
+ },
446
+ {
447
+ "epoch": 0.315,
448
+ "grad_norm": 1.65625,
449
+ "learning_rate": 4.331357246968447e-06,
450
+ "loss": 0.4111,
451
+ "step": 630
452
+ },
453
+ {
454
+ "epoch": 0.32,
455
+ "grad_norm": 2.15625,
456
+ "learning_rate": 4.301377779200826e-06,
457
+ "loss": 0.476,
458
+ "step": 640
459
+ },
460
+ {
461
+ "epoch": 0.325,
462
+ "grad_norm": 2.234375,
463
+ "learning_rate": 4.270849594311323e-06,
464
+ "loss": 0.4396,
465
+ "step": 650
466
+ },
467
+ {
468
+ "epoch": 0.33,
469
+ "grad_norm": 2.71875,
470
+ "learning_rate": 4.239781991480786e-06,
471
+ "loss": 0.4193,
472
+ "step": 660
473
+ },
474
+ {
475
+ "epoch": 0.335,
476
+ "grad_norm": 2.640625,
477
+ "learning_rate": 4.208184434201999e-06,
478
+ "loss": 0.4551,
479
+ "step": 670
480
+ },
481
+ {
482
+ "epoch": 0.34,
483
+ "grad_norm": 2.234375,
484
+ "learning_rate": 4.176066547396998e-06,
485
+ "loss": 0.4352,
486
+ "step": 680
487
+ },
488
+ {
489
+ "epoch": 0.345,
490
+ "grad_norm": 2.203125,
491
+ "learning_rate": 4.14343811448524e-06,
492
+ "loss": 0.4195,
493
+ "step": 690
494
+ },
495
+ {
496
+ "epoch": 0.35,
497
+ "grad_norm": 2.1875,
498
+ "learning_rate": 4.110309074403467e-06,
499
+ "loss": 0.4426,
500
+ "step": 700
501
+ },
502
+ {
503
+ "epoch": 0.355,
504
+ "grad_norm": 2.9375,
505
+ "learning_rate": 4.076689518578217e-06,
506
+ "loss": 0.434,
507
+ "step": 710
508
+ },
509
+ {
510
+ "epoch": 0.36,
511
+ "grad_norm": 2.0625,
512
+ "learning_rate": 4.0425896878518725e-06,
513
+ "loss": 0.4283,
514
+ "step": 720
515
+ },
516
+ {
517
+ "epoch": 0.365,
518
+ "grad_norm": 1.90625,
519
+ "learning_rate": 4.008019969363206e-06,
520
+ "loss": 0.4023,
521
+ "step": 730
522
+ },
523
+ {
524
+ "epoch": 0.37,
525
+ "grad_norm": 1.859375,
526
+ "learning_rate": 3.972990893383356e-06,
527
+ "loss": 0.3877,
528
+ "step": 740
529
+ },
530
+ {
531
+ "epoch": 0.375,
532
+ "grad_norm": 1.9140625,
533
+ "learning_rate": 3.9375131301081974e-06,
534
+ "loss": 0.4492,
535
+ "step": 750
536
+ },
537
+ {
538
+ "epoch": 0.38,
539
+ "grad_norm": 2.296875,
540
+ "learning_rate": 3.901597486408105e-06,
541
+ "loss": 0.4469,
542
+ "step": 760
543
+ },
544
+ {
545
+ "epoch": 0.385,
546
+ "grad_norm": 2.140625,
547
+ "learning_rate": 3.865254902536073e-06,
548
+ "loss": 0.4564,
549
+ "step": 770
550
+ },
551
+ {
552
+ "epoch": 0.39,
553
+ "grad_norm": 1.7109375,
554
+ "learning_rate": 3.828496448795208e-06,
555
+ "loss": 0.416,
556
+ "step": 780
557
+ },
558
+ {
559
+ "epoch": 0.395,
560
+ "grad_norm": 2.59375,
561
+ "learning_rate": 3.791333322166605e-06,
562
+ "loss": 0.4373,
563
+ "step": 790
564
+ },
565
+ {
566
+ "epoch": 0.4,
567
+ "grad_norm": 2.4375,
568
+ "learning_rate": 3.753776842898644e-06,
569
+ "loss": 0.4203,
570
+ "step": 800
571
+ },
572
+ {
573
+ "epoch": 0.405,
574
+ "grad_norm": 4.6875,
575
+ "learning_rate": 3.7158384510587264e-06,
576
+ "loss": 0.4467,
577
+ "step": 810
578
+ },
579
+ {
580
+ "epoch": 0.41,
581
+ "grad_norm": 2.078125,
582
+ "learning_rate": 3.677529703048525e-06,
583
+ "loss": 0.4469,
584
+ "step": 820
585
+ },
586
+ {
587
+ "epoch": 0.415,
588
+ "grad_norm": 1.7265625,
589
+ "learning_rate": 3.6388622680837893e-06,
590
+ "loss": 0.4213,
591
+ "step": 830
592
+ },
593
+ {
594
+ "epoch": 0.42,
595
+ "grad_norm": 2.1875,
596
+ "learning_rate": 3.599847924639788e-06,
597
+ "loss": 0.4578,
598
+ "step": 840
599
+ },
600
+ {
601
+ "epoch": 0.425,
602
+ "grad_norm": 2.375,
603
+ "learning_rate": 3.5604985568634754e-06,
604
+ "loss": 0.4281,
605
+ "step": 850
606
+ },
607
+ {
608
+ "epoch": 0.43,
609
+ "grad_norm": 2.171875,
610
+ "learning_rate": 3.5208261509534627e-06,
611
+ "loss": 0.4145,
612
+ "step": 860
613
+ },
614
+ {
615
+ "epoch": 0.435,
616
+ "grad_norm": 2.125,
617
+ "learning_rate": 3.480842791508904e-06,
618
+ "loss": 0.468,
619
+ "step": 870
620
+ },
621
+ {
622
+ "epoch": 0.44,
623
+ "grad_norm": 1.8828125,
624
+ "learning_rate": 3.440560657848414e-06,
625
+ "loss": 0.4195,
626
+ "step": 880
627
+ },
628
+ {
629
+ "epoch": 0.445,
630
+ "grad_norm": 1.8984375,
631
+ "learning_rate": 3.3999920203001287e-06,
632
+ "loss": 0.4541,
633
+ "step": 890
634
+ },
635
+ {
636
+ "epoch": 0.45,
637
+ "grad_norm": 1.78125,
638
+ "learning_rate": 3.359149236464041e-06,
639
+ "loss": 0.4076,
640
+ "step": 900
641
+ },
642
+ {
643
+ "epoch": 0.455,
644
+ "grad_norm": 2.828125,
645
+ "learning_rate": 3.31804474744776e-06,
646
+ "loss": 0.4088,
647
+ "step": 910
648
+ },
649
+ {
650
+ "epoch": 0.46,
651
+ "grad_norm": 1.7109375,
652
+ "learning_rate": 3.27669107407683e-06,
653
+ "loss": 0.4096,
654
+ "step": 920
655
+ },
656
+ {
657
+ "epoch": 0.465,
658
+ "grad_norm": 2.84375,
659
+ "learning_rate": 3.23510081308076e-06,
660
+ "loss": 0.4359,
661
+ "step": 930
662
+ },
663
+ {
664
+ "epoch": 0.47,
665
+ "grad_norm": 2.59375,
666
+ "learning_rate": 3.1932866332559455e-06,
667
+ "loss": 0.418,
668
+ "step": 940
669
+ },
670
+ {
671
+ "epoch": 0.475,
672
+ "grad_norm": 2.34375,
673
+ "learning_rate": 3.1512612716066217e-06,
674
+ "loss": 0.4348,
675
+ "step": 950
676
+ },
677
+ {
678
+ "epoch": 0.48,
679
+ "grad_norm": 1.8359375,
680
+ "learning_rate": 3.1090375294650565e-06,
681
+ "loss": 0.4035,
682
+ "step": 960
683
+ },
684
+ {
685
+ "epoch": 0.485,
686
+ "grad_norm": 2.0,
687
+ "learning_rate": 3.066628268592138e-06,
688
+ "loss": 0.4248,
689
+ "step": 970
690
+ },
691
+ {
692
+ "epoch": 0.49,
693
+ "grad_norm": 1.984375,
694
+ "learning_rate": 3.0240464072595547e-06,
695
+ "loss": 0.4254,
696
+ "step": 980
697
+ },
698
+ {
699
+ "epoch": 0.495,
700
+ "grad_norm": 2.203125,
701
+ "learning_rate": 2.981304916314769e-06,
702
+ "loss": 0.4197,
703
+ "step": 990
704
+ },
705
+ {
706
+ "epoch": 0.5,
707
+ "grad_norm": 1.6328125,
708
+ "learning_rate": 2.938416815229968e-06,
709
+ "loss": 0.4479,
710
+ "step": 1000
711
+ },
712
+ {
713
+ "epoch": 0.505,
714
+ "grad_norm": 1.765625,
715
+ "learning_rate": 2.8953951681362098e-06,
716
+ "loss": 0.4039,
717
+ "step": 1010
718
+ },
719
+ {
720
+ "epoch": 0.51,
721
+ "grad_norm": 2.578125,
722
+ "learning_rate": 2.852253079843957e-06,
723
+ "loss": 0.4723,
724
+ "step": 1020
725
+ },
726
+ {
727
+ "epoch": 0.515,
728
+ "grad_norm": 2.6875,
729
+ "learning_rate": 2.809003691851232e-06,
730
+ "loss": 0.423,
731
+ "step": 1030
732
+ },
733
+ {
734
+ "epoch": 0.52,
735
+ "grad_norm": 1.7578125,
736
+ "learning_rate": 2.7656601783405833e-06,
737
+ "loss": 0.4002,
738
+ "step": 1040
739
+ },
740
+ {
741
+ "epoch": 0.525,
742
+ "grad_norm": 2.34375,
743
+ "learning_rate": 2.7222357421661042e-06,
744
+ "loss": 0.426,
745
+ "step": 1050
746
+ },
747
+ {
748
+ "epoch": 0.53,
749
+ "grad_norm": 1.6875,
750
+ "learning_rate": 2.678743610831715e-06,
751
+ "loss": 0.4447,
752
+ "step": 1060
753
+ },
754
+ {
755
+ "epoch": 0.535,
756
+ "grad_norm": 1.8203125,
757
+ "learning_rate": 2.635197032461939e-06,
758
+ "loss": 0.4324,
759
+ "step": 1070
760
+ },
761
+ {
762
+ "epoch": 0.54,
763
+ "grad_norm": 2.03125,
764
+ "learning_rate": 2.591609271766391e-06,
765
+ "loss": 0.3877,
766
+ "step": 1080
767
+ },
768
+ {
769
+ "epoch": 0.545,
770
+ "grad_norm": 1.8125,
771
+ "learning_rate": 2.547993605999225e-06,
772
+ "loss": 0.3949,
773
+ "step": 1090
774
+ },
775
+ {
776
+ "epoch": 0.55,
777
+ "grad_norm": 2.0625,
778
+ "learning_rate": 2.504363320914746e-06,
779
+ "loss": 0.4428,
780
+ "step": 1100
781
+ },
782
+ {
783
+ "epoch": 0.555,
784
+ "grad_norm": 2.03125,
785
+ "learning_rate": 2.460731706720449e-06,
786
+ "loss": 0.4383,
787
+ "step": 1110
788
+ },
789
+ {
790
+ "epoch": 0.56,
791
+ "grad_norm": 2.140625,
792
+ "learning_rate": 2.4171120540286848e-06,
793
+ "loss": 0.4244,
794
+ "step": 1120
795
+ },
796
+ {
797
+ "epoch": 0.565,
798
+ "grad_norm": 2.890625,
799
+ "learning_rate": 2.373517649808217e-06,
800
+ "loss": 0.4281,
801
+ "step": 1130
802
+ },
803
+ {
804
+ "epoch": 0.57,
805
+ "grad_norm": 2.109375,
806
+ "learning_rate": 2.3299617733368805e-06,
807
+ "loss": 0.4238,
808
+ "step": 1140
809
+ },
810
+ {
811
+ "epoch": 0.575,
812
+ "grad_norm": 2.015625,
813
+ "learning_rate": 2.2864576921565816e-06,
814
+ "loss": 0.45,
815
+ "step": 1150
816
+ },
817
+ {
818
+ "epoch": 0.58,
819
+ "grad_norm": 1.9296875,
820
+ "learning_rate": 2.2430186580318833e-06,
821
+ "loss": 0.4137,
822
+ "step": 1160
823
+ },
824
+ {
825
+ "epoch": 0.585,
826
+ "grad_norm": 1.96875,
827
+ "learning_rate": 2.1996579029133826e-06,
828
+ "loss": 0.402,
829
+ "step": 1170
830
+ },
831
+ {
832
+ "epoch": 0.59,
833
+ "grad_norm": 2.453125,
834
+ "learning_rate": 2.156388634907134e-06,
835
+ "loss": 0.4365,
836
+ "step": 1180
837
+ },
838
+ {
839
+ "epoch": 0.595,
840
+ "grad_norm": 1.7421875,
841
+ "learning_rate": 2.1132240342513304e-06,
842
+ "loss": 0.4355,
843
+ "step": 1190
844
+ },
845
+ {
846
+ "epoch": 0.6,
847
+ "grad_norm": 1.6875,
848
+ "learning_rate": 2.070177249301476e-06,
849
+ "loss": 0.4039,
850
+ "step": 1200
851
+ },
852
+ {
853
+ "epoch": 0.605,
854
+ "grad_norm": 1.734375,
855
+ "learning_rate": 2.0272613925252716e-06,
856
+ "loss": 0.426,
857
+ "step": 1210
858
+ },
859
+ {
860
+ "epoch": 0.61,
861
+ "grad_norm": 1.96875,
862
+ "learning_rate": 1.9844895365084264e-06,
863
+ "loss": 0.4178,
864
+ "step": 1220
865
+ },
866
+ {
867
+ "epoch": 0.615,
868
+ "grad_norm": 2.0625,
869
+ "learning_rate": 1.941874709972622e-06,
870
+ "loss": 0.4367,
871
+ "step": 1230
872
+ },
873
+ {
874
+ "epoch": 0.62,
875
+ "grad_norm": 1.9921875,
876
+ "learning_rate": 1.899429893806841e-06,
877
+ "loss": 0.4021,
878
+ "step": 1240
879
+ },
880
+ {
881
+ "epoch": 0.625,
882
+ "grad_norm": 2.046875,
883
+ "learning_rate": 1.8571680171132603e-06,
884
+ "loss": 0.4404,
885
+ "step": 1250
886
+ },
887
+ {
888
+ "epoch": 0.63,
889
+ "grad_norm": 2.578125,
890
+ "learning_rate": 1.815101953268919e-06,
891
+ "loss": 0.4143,
892
+ "step": 1260
893
+ },
894
+ {
895
+ "epoch": 0.635,
896
+ "grad_norm": 2.359375,
897
+ "learning_rate": 1.7732445160043687e-06,
898
+ "loss": 0.4301,
899
+ "step": 1270
900
+ },
901
+ {
902
+ "epoch": 0.64,
903
+ "grad_norm": 1.984375,
904
+ "learning_rate": 1.7316084555004825e-06,
905
+ "loss": 0.4666,
906
+ "step": 1280
907
+ },
908
+ {
909
+ "epoch": 0.645,
910
+ "grad_norm": 2.46875,
911
+ "learning_rate": 1.6902064545046271e-06,
912
+ "loss": 0.4102,
913
+ "step": 1290
914
+ },
915
+ {
916
+ "epoch": 0.65,
917
+ "grad_norm": 2.546875,
918
+ "learning_rate": 1.6490511244673752e-06,
919
+ "loss": 0.4322,
920
+ "step": 1300
921
+ },
922
+ {
923
+ "epoch": 0.655,
924
+ "grad_norm": 1.8046875,
925
+ "learning_rate": 1.6081550017009368e-06,
926
+ "loss": 0.4273,
927
+ "step": 1310
928
+ },
929
+ {
930
+ "epoch": 0.66,
931
+ "grad_norm": 2.421875,
932
+ "learning_rate": 1.5675305435604776e-06,
933
+ "loss": 0.424,
934
+ "step": 1320
935
+ },
936
+ {
937
+ "epoch": 0.665,
938
+ "grad_norm": 2.140625,
939
+ "learning_rate": 1.5271901246494847e-06,
940
+ "loss": 0.4307,
941
+ "step": 1330
942
+ },
943
+ {
944
+ "epoch": 0.67,
945
+ "grad_norm": 2.109375,
946
+ "learning_rate": 1.487146033050344e-06,
947
+ "loss": 0.4164,
948
+ "step": 1340
949
+ },
950
+ {
951
+ "epoch": 0.675,
952
+ "grad_norm": 2.375,
953
+ "learning_rate": 1.4474104665812727e-06,
954
+ "loss": 0.4736,
955
+ "step": 1350
956
+ },
957
+ {
958
+ "epoch": 0.68,
959
+ "grad_norm": 2.0,
960
+ "learning_rate": 1.4079955290807452e-06,
961
+ "loss": 0.4191,
962
+ "step": 1360
963
+ },
964
+ {
965
+ "epoch": 0.685,
966
+ "grad_norm": 2.359375,
967
+ "learning_rate": 1.3689132267205432e-06,
968
+ "loss": 0.4023,
969
+ "step": 1370
970
+ },
971
+ {
972
+ "epoch": 0.69,
973
+ "grad_norm": 2.21875,
974
+ "learning_rate": 1.3301754643485671e-06,
975
+ "loss": 0.4332,
976
+ "step": 1380
977
+ },
978
+ {
979
+ "epoch": 0.695,
980
+ "grad_norm": 2.265625,
981
+ "learning_rate": 1.2917940418624942e-06,
982
+ "loss": 0.427,
983
+ "step": 1390
984
+ },
985
+ {
986
+ "epoch": 0.7,
987
+ "grad_norm": 2.09375,
988
+ "learning_rate": 1.2537806506154246e-06,
989
+ "loss": 0.4205,
990
+ "step": 1400
991
+ },
992
+ {
993
+ "epoch": 0.705,
994
+ "grad_norm": 2.53125,
995
+ "learning_rate": 1.2161468698545755e-06,
996
+ "loss": 0.4221,
997
+ "step": 1410
998
+ },
999
+ {
1000
+ "epoch": 0.71,
1001
+ "grad_norm": 2.421875,
1002
+ "learning_rate": 1.1789041631941326e-06,
1003
+ "loss": 0.4084,
1004
+ "step": 1420
1005
+ },
1006
+ {
1007
+ "epoch": 0.715,
1008
+ "grad_norm": 1.90625,
1009
+ "learning_rate": 1.142063875123323e-06,
1010
+ "loss": 0.4139,
1011
+ "step": 1430
1012
+ },
1013
+ {
1014
+ "epoch": 0.72,
1015
+ "grad_norm": 1.796875,
1016
+ "learning_rate": 1.1056372275507748e-06,
1017
+ "loss": 0.3836,
1018
+ "step": 1440
1019
+ },
1020
+ {
1021
+ "epoch": 0.725,
1022
+ "grad_norm": 1.6953125,
1023
+ "learning_rate": 1.06963531638621e-06,
1024
+ "loss": 0.3801,
1025
+ "step": 1450
1026
+ },
1027
+ {
1028
+ "epoch": 0.73,
1029
+ "grad_norm": 1.8515625,
1030
+ "learning_rate": 1.0340691081605267e-06,
1031
+ "loss": 0.4252,
1032
+ "step": 1460
1033
+ },
1034
+ {
1035
+ "epoch": 0.735,
1036
+ "grad_norm": 2.765625,
1037
+ "learning_rate": 9.989494366852904e-07,
1038
+ "loss": 0.4455,
1039
+ "step": 1470
1040
+ },
1041
+ {
1042
+ "epoch": 0.74,
1043
+ "grad_norm": 1.8203125,
1044
+ "learning_rate": 9.64286999752642e-07,
1045
+ "loss": 0.4498,
1046
+ "step": 1480
1047
+ },
1048
+ {
1049
+ "epoch": 0.745,
1050
+ "grad_norm": 1.6171875,
1051
+ "learning_rate": 9.300923558766556e-07,
1052
+ "loss": 0.3969,
1053
+ "step": 1490
1054
+ },
1055
+ {
1056
+ "epoch": 0.75,
1057
+ "grad_norm": 1.7578125,
1058
+ "learning_rate": 8.963759210771053e-07,
1059
+ "loss": 0.423,
1060
+ "step": 1500
1061
+ },
1062
+ {
1063
+ "epoch": 0.755,
1064
+ "grad_norm": 2.125,
1065
+ "learning_rate": 8.631479657066508e-07,
1066
+ "loss": 0.393,
1067
+ "step": 1510
1068
+ },
1069
+ {
1070
+ "epoch": 0.76,
1071
+ "grad_norm": 1.984375,
1072
+ "learning_rate": 8.304186113223839e-07,
1073
+ "loss": 0.4291,
1074
+ "step": 1520
1075
+ },
1076
+ {
1077
+ "epoch": 0.765,
1078
+ "grad_norm": 2.5,
1079
+ "learning_rate": 7.981978276027055e-07,
1080
+ "loss": 0.4395,
1081
+ "step": 1530
1082
+ },
1083
+ {
1084
+ "epoch": 0.77,
1085
+ "grad_norm": 1.8671875,
1086
+ "learning_rate": 7.664954293104674e-07,
1087
+ "loss": 0.4242,
1088
+ "step": 1540
1089
+ },
1090
+ {
1091
+ "epoch": 0.775,
1092
+ "grad_norm": 2.3125,
1093
+ "learning_rate": 7.353210733032976e-07,
1094
+ "loss": 0.4441,
1095
+ "step": 1550
1096
+ },
1097
+ {
1098
+ "epoch": 0.78,
1099
+ "grad_norm": 3.234375,
1100
+ "learning_rate": 7.046842555920283e-07,
1101
+ "loss": 0.4115,
1102
+ "step": 1560
1103
+ },
1104
+ {
1105
+ "epoch": 0.785,
1106
+ "grad_norm": 1.6171875,
1107
+ "learning_rate": 6.74594308448119e-07,
1108
+ "loss": 0.4098,
1109
+ "step": 1570
1110
+ },
1111
+ {
1112
+ "epoch": 0.79,
1113
+ "grad_norm": 2.40625,
1114
+ "learning_rate": 6.450603975609593e-07,
1115
+ "loss": 0.4529,
1116
+ "step": 1580
1117
+ },
1118
+ {
1119
+ "epoch": 0.795,
1120
+ "grad_norm": 2.265625,
1121
+ "learning_rate": 6.160915192459058e-07,
1122
+ "loss": 0.4076,
1123
+ "step": 1590
1124
+ },
1125
+ {
1126
+ "epoch": 0.8,
1127
+ "grad_norm": 2.140625,
1128
+ "learning_rate": 5.876964977039207e-07,
1129
+ "loss": 0.4283,
1130
+ "step": 1600
1131
+ },
1132
+ {
1133
+ "epoch": 0.805,
1134
+ "grad_norm": 1.796875,
1135
+ "learning_rate": 5.598839823336349e-07,
1136
+ "loss": 0.4104,
1137
+ "step": 1610
1138
+ },
1139
+ {
1140
+ "epoch": 0.81,
1141
+ "grad_norm": 2.015625,
1142
+ "learning_rate": 5.32662445096657e-07,
1143
+ "loss": 0.3916,
1144
+ "step": 1620
1145
+ },
1146
+ {
1147
+ "epoch": 0.815,
1148
+ "grad_norm": 2.03125,
1149
+ "learning_rate": 5.060401779369292e-07,
1150
+ "loss": 0.4215,
1151
+ "step": 1630
1152
+ },
1153
+ {
1154
+ "epoch": 0.82,
1155
+ "grad_norm": 1.7578125,
1156
+ "learning_rate": 4.800252902549243e-07,
1157
+ "loss": 0.4039,
1158
+ "step": 1640
1159
+ },
1160
+ {
1161
+ "epoch": 0.825,
1162
+ "grad_norm": 3.0,
1163
+ "learning_rate": 4.54625706437441e-07,
1164
+ "loss": 0.4115,
1165
+ "step": 1650
1166
+ },
1167
+ {
1168
+ "epoch": 0.83,
1169
+ "grad_norm": 1.5703125,
1170
+ "learning_rate": 4.2984916344376404e-07,
1171
+ "loss": 0.4311,
1172
+ "step": 1660
1173
+ },
1174
+ {
1175
+ "epoch": 0.835,
1176
+ "grad_norm": 2.03125,
1177
+ "learning_rate": 4.057032084489032e-07,
1178
+ "loss": 0.3963,
1179
+ "step": 1670
1180
+ },
1181
+ {
1182
+ "epoch": 0.84,
1183
+ "grad_norm": 1.7734375,
1184
+ "learning_rate": 3.821951965446577e-07,
1185
+ "loss": 0.4145,
1186
+ "step": 1680
1187
+ },
1188
+ {
1189
+ "epoch": 0.845,
1190
+ "grad_norm": 1.828125,
1191
+ "learning_rate": 3.5933228849917956e-07,
1192
+ "loss": 0.4207,
1193
+ "step": 1690
1194
+ },
1195
+ {
1196
+ "epoch": 0.85,
1197
+ "grad_norm": 2.03125,
1198
+ "learning_rate": 3.371214485757393e-07,
1199
+ "loss": 0.4111,
1200
+ "step": 1700
1201
+ },
1202
+ {
1203
+ "epoch": 0.855,
1204
+ "grad_norm": 1.890625,
1205
+ "learning_rate": 3.1556944241133704e-07,
1206
+ "loss": 0.4266,
1207
+ "step": 1710
1208
+ },
1209
+ {
1210
+ "epoch": 0.86,
1211
+ "grad_norm": 2.15625,
1212
+ "learning_rate": 2.946828349558309e-07,
1213
+ "loss": 0.4121,
1214
+ "step": 1720
1215
+ },
1216
+ {
1217
+ "epoch": 0.865,
1218
+ "grad_norm": 3.34375,
1219
+ "learning_rate": 2.7446798847218376e-07,
1220
+ "loss": 0.4457,
1221
+ "step": 1730
1222
+ },
1223
+ {
1224
+ "epoch": 0.87,
1225
+ "grad_norm": 1.796875,
1226
+ "learning_rate": 2.549310605984612e-07,
1227
+ "loss": 0.4207,
1228
+ "step": 1740
1229
+ },
1230
+ {
1231
+ "epoch": 0.875,
1232
+ "grad_norm": 1.7109375,
1233
+ "learning_rate": 2.360780024721515e-07,
1234
+ "loss": 0.4021,
1235
+ "step": 1750
1236
+ },
1237
+ {
1238
+ "epoch": 0.88,
1239
+ "grad_norm": 2.0625,
1240
+ "learning_rate": 2.1791455691739323e-07,
1241
+ "loss": 0.4209,
1242
+ "step": 1760
1243
+ },
1244
+ {
1245
+ "epoch": 0.885,
1246
+ "grad_norm": 2.140625,
1247
+ "learning_rate": 2.0044625669565582e-07,
1248
+ "loss": 0.448,
1249
+ "step": 1770
1250
+ },
1251
+ {
1252
+ "epoch": 0.89,
1253
+ "grad_norm": 1.546875,
1254
+ "learning_rate": 1.8367842282040692e-07,
1255
+ "loss": 0.4229,
1256
+ "step": 1780
1257
+ },
1258
+ {
1259
+ "epoch": 0.895,
1260
+ "grad_norm": 2.640625,
1261
+ "learning_rate": 1.676161629362777e-07,
1262
+ "loss": 0.3854,
1263
+ "step": 1790
1264
+ },
1265
+ {
1266
+ "epoch": 0.9,
1267
+ "grad_norm": 1.65625,
1268
+ "learning_rate": 1.5226436976322728e-07,
1269
+ "loss": 0.3898,
1270
+ "step": 1800
1271
+ },
1272
+ {
1273
+ "epoch": 0.905,
1274
+ "grad_norm": 1.765625,
1275
+ "learning_rate": 1.3762771960617315e-07,
1276
+ "loss": 0.4404,
1277
+ "step": 1810
1278
+ },
1279
+ {
1280
+ "epoch": 0.91,
1281
+ "grad_norm": 2.265625,
1282
+ "learning_rate": 1.237106709305408e-07,
1283
+ "loss": 0.4217,
1284
+ "step": 1820
1285
+ },
1286
+ {
1287
+ "epoch": 0.915,
1288
+ "grad_norm": 1.8984375,
1289
+ "learning_rate": 1.105174630041747e-07,
1290
+ "loss": 0.4096,
1291
+ "step": 1830
1292
+ },
1293
+ {
1294
+ "epoch": 0.92,
1295
+ "grad_norm": 2.3125,
1296
+ "learning_rate": 9.805211460601455e-08,
1297
+ "loss": 0.434,
1298
+ "step": 1840
1299
+ },
1300
+ {
1301
+ "epoch": 0.925,
1302
+ "grad_norm": 1.6640625,
1303
+ "learning_rate": 8.631842280193759e-08,
1304
+ "loss": 0.4104,
1305
+ "step": 1850
1306
+ },
1307
+ {
1308
+ "epoch": 0.93,
1309
+ "grad_norm": 3.359375,
1310
+ "learning_rate": 7.531996178813311e-08,
1311
+ "loss": 0.4564,
1312
+ "step": 1860
1313
+ },
1314
+ {
1315
+ "epoch": 0.935,
1316
+ "grad_norm": 1.734375,
1317
+ "learning_rate": 6.506008180237111e-08,
1318
+ "loss": 0.4031,
1319
+ "step": 1870
1320
+ },
1321
+ {
1322
+ "epoch": 0.94,
1323
+ "grad_norm": 2.03125,
1324
+ "learning_rate": 5.554190810348442e-08,
1325
+ "loss": 0.4025,
1326
+ "step": 1880
1327
+ },
1328
+ {
1329
+ "epoch": 0.945,
1330
+ "grad_norm": 2.265625,
1331
+ "learning_rate": 4.676834001938718e-08,
1332
+ "loss": 0.4193,
1333
+ "step": 1890
1334
+ },
1335
+ {
1336
+ "epoch": 0.95,
1337
+ "grad_norm": 1.7421875,
1338
+ "learning_rate": 3.874205006390852e-08,
1339
+ "loss": 0.4361,
1340
+ "step": 1900
1341
+ },
1342
+ {
1343
+ "epoch": 0.955,
1344
+ "grad_norm": 2.125,
1345
+ "learning_rate": 3.146548312272152e-08,
1346
+ "loss": 0.4219,
1347
+ "step": 1910
1348
+ },
1349
+ {
1350
+ "epoch": 0.96,
1351
+ "grad_norm": 2.609375,
1352
+ "learning_rate": 2.494085570860616e-08,
1353
+ "loss": 0.4297,
1354
+ "step": 1920
1355
+ },
1356
+ {
1357
+ "epoch": 0.965,
1358
+ "grad_norm": 2.28125,
1359
+ "learning_rate": 1.91701552862783e-08,
1360
+ "loss": 0.4148,
1361
+ "step": 1930
1362
+ },
1363
+ {
1364
+ "epoch": 0.97,
1365
+ "grad_norm": 1.734375,
1366
+ "learning_rate": 1.4155139666988393e-08,
1367
+ "loss": 0.4313,
1368
+ "step": 1940
1369
+ },
1370
+ {
1371
+ "epoch": 0.975,
1372
+ "grad_norm": 1.671875,
1373
+ "learning_rate": 9.897336473076168e-09,
1374
+ "loss": 0.4064,
1375
+ "step": 1950
1376
+ },
1377
+ {
1378
+ "epoch": 0.98,
1379
+ "grad_norm": 2.34375,
1380
+ "learning_rate": 6.398042672640104e-09,
1381
+ "loss": 0.4498,
1382
+ "step": 1960
1383
+ },
1384
+ {
1385
+ "epoch": 0.985,
1386
+ "grad_norm": 1.78125,
1387
+ "learning_rate": 3.6583241844706517e-09,
1388
+ "loss": 0.4023,
1389
+ "step": 1970
1390
+ },
1391
+ {
1392
+ "epoch": 0.99,
1393
+ "grad_norm": 2.421875,
1394
+ "learning_rate": 1.6790155533594198e-09,
1395
+ "loss": 0.4588,
1396
+ "step": 1980
1397
+ },
1398
+ {
1399
+ "epoch": 0.995,
1400
+ "grad_norm": 1.7734375,
1401
+ "learning_rate": 4.6071969588945555e-10,
1402
+ "loss": 0.4533,
1403
+ "step": 1990
1404
+ },
1405
+ {
1406
+ "epoch": 1.0,
1407
+ "grad_norm": 1.890625,
1408
+ "learning_rate": 3.807716780768189e-12,
1409
+ "loss": 0.4279,
1410
+ "step": 2000
1411
+ }
1412
+ ],
1413
+ "logging_steps": 10,
1414
+ "max_steps": 2000,
1415
+ "num_input_tokens_seen": 0,
1416
+ "num_train_epochs": 9223372036854775807,
1417
+ "save_steps": 4000,
1418
+ "stateful_callbacks": {
1419
+ "TrainerControl": {
1420
+ "args": {
1421
+ "should_epoch_stop": false,
1422
+ "should_evaluate": false,
1423
+ "should_log": false,
1424
+ "should_save": true,
1425
+ "should_training_stop": true
1426
+ },
1427
+ "attributes": {}
1428
+ }
1429
+ },
1430
+ "total_flos": 1.91328310788096e+17,
1431
+ "train_batch_size": 32,
1432
+ "trial_name": null,
1433
+ "trial_params": null
1434
+ }
checkpoint-2000/training_args.bin ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:bbd60ee4651254bec576bbf9d0eb0600206a2b63f12999f682ec638d8fa85eed
3
+ size 5368