Hugging Face's logo

RikkiXu
/
zephyr-7b-sft-full

Text Generation
Transformers
TensorBoard
Safetensors
llama
trl
sft
Generated from Trainer
conversational
Inference Endpoints
text-generation-inference
Model card Files Metrics Community
RikkiXu commited on
Commit
759ed41
•
1 Parent(s): 416c563

Model save

Browse files
Files changed (3) hide show
  1. all_results.json +5 -5
  2. train_results.json +5 -5
  3. trainer_state.json +1440 -187
all_results.json CHANGED
@@ -1,8 +1,8 @@
1
  {
2
  "epoch": 1.0,
3
- "train_loss": 0.8946974671111917,
4
- "train_runtime": 2355.8103,
5
- "train_samples": 100000,
6
- "train_samples_per_second": 5.733,
7
- "train_steps_per_second": 0.09
8
  }
 
1
  {
2
  "epoch": 1.0,
3
+ "train_loss": 0.9518642601066596,
4
+ "train_runtime": 13070.0796,
5
+ "train_samples": 207864,
6
+ "train_samples_per_second": 10.839,
7
+ "train_steps_per_second": 0.085
8
  }
train_results.json CHANGED
@@ -1,8 +1,8 @@
1
  {
2
  "epoch": 1.0,
3
- "train_loss": 0.8946974671111917,
4
- "train_runtime": 2355.8103,
5
- "train_samples": 100000,
6
- "train_samples_per_second": 5.733,
7
- "train_steps_per_second": 0.09
8
  }
 
1
  {
2
  "epoch": 1.0,
3
+ "train_loss": 0.9518642601066596,
4
+ "train_runtime": 13070.0796,
5
+ "train_samples": 207864,
6
+ "train_samples_per_second": 10.839,
7
+ "train_steps_per_second": 0.085
8
  }
trainer_state.json CHANGED
@@ -3,337 +3,1590 @@
3
  "best_model_checkpoint": null,
4
  "epoch": 1.0,
5
  "eval_steps": 500,
6
- "global_step": 212,
7
  "is_hyper_param_search": false,
8
  "is_local_process_zero": true,
9
  "is_world_process_zero": true,
10
  "log_history": [
11
  {
12
  "epoch": 0.0,
13
- "grad_norm": 56.48654000341791,
14
- "learning_rate": 2.2727272727272729e-07,
15
- "loss": 1.4277,
16
  "step": 1
17
  },
18
  {
19
- "epoch": 0.02,
20
- "grad_norm": 32.7318603335361,
21
- "learning_rate": 1.1363636363636364e-06,
22
- "loss": 1.3555,
23
  "step": 5
24
  },
25
  {
26
- "epoch": 0.05,
27
- "grad_norm": 18.95719710067938,
28
- "learning_rate": 2.2727272727272728e-06,
29
- "loss": 1.1538,
30
  "step": 10
31
  },
32
  {
33
- "epoch": 0.07,
34
- "grad_norm": 6.971441284336922,
35
- "learning_rate": 3.409090909090909e-06,
36
- "loss": 1.0874,
37
  "step": 15
38
  },
39
  {
40
- "epoch": 0.09,
41
- "grad_norm": 5.3823086685656945,
42
- "learning_rate": 4.5454545454545455e-06,
43
- "loss": 1.0206,
44
  "step": 20
45
  },
46
  {
47
- "epoch": 0.12,
48
- "grad_norm": 4.089694965230436,
49
- "learning_rate": 4.9969249228707625e-06,
50
- "loss": 1.003,
51
  "step": 25
52
  },
53
  {
54
- "epoch": 0.14,
55
- "grad_norm": 3.6458232095508354,
56
- "learning_rate": 4.978160173317439e-06,
57
- "loss": 0.9657,
58
  "step": 30
59
  },
60
  {
61
- "epoch": 0.17,
62
- "grad_norm": 3.2176159938800266,
63
- "learning_rate": 4.942467076958999e-06,
64
- "loss": 0.9435,
65
  "step": 35
66
  },
67
  {
68
- "epoch": 0.19,
69
- "grad_norm": 2.977792325165127,
70
- "learning_rate": 4.890089453835894e-06,
71
- "loss": 0.9368,
72
  "step": 40
73
  },
74
  {
75
- "epoch": 0.21,
76
- "grad_norm": 3.1118165842382504,
77
- "learning_rate": 4.821385096224268e-06,
78
- "loss": 0.9261,
79
  "step": 45
80
  },
81
  {
82
- "epoch": 0.24,
83
- "grad_norm": 3.3730430715720994,
84
- "learning_rate": 4.736823324551909e-06,
85
- "loss": 0.9112,
86
  "step": 50
87
  },
88
  {
89
- "epoch": 0.26,
90
- "grad_norm": 3.1574351012684474,
91
- "learning_rate": 4.636981781463848e-06,
92
- "loss": 0.9145,
93
  "step": 55
94
  },
95
  {
96
- "epoch": 0.28,
97
- "grad_norm": 3.0769386301276183,
98
- "learning_rate": 4.522542485937369e-06,
99
- "loss": 0.8814,
100
  "step": 60
101
  },
102
  {
103
- "epoch": 0.31,
104
- "grad_norm": 3.055053060479708,
105
- "learning_rate": 4.394287174400838e-06,
106
- "loss": 0.8751,
107
  "step": 65
108
  },
109
  {
110
- "epoch": 0.33,
111
- "grad_norm": 5.4232076990901,
112
- "learning_rate": 4.253091960681222e-06,
113
- "loss": 0.8909,
114
  "step": 70
115
  },
116
  {
117
- "epoch": 0.35,
118
- "grad_norm": 3.2020191326645273,
119
- "learning_rate": 4.099921351258292e-06,
120
- "loss": 0.8597,
121
  "step": 75
122
  },
123
  {
124
- "epoch": 0.38,
125
- "grad_norm": 3.290349939143738,
126
- "learning_rate": 3.935821656707359e-06,
127
- "loss": 0.8913,
128
  "step": 80
129
  },
130
  {
131
- "epoch": 0.4,
132
- "grad_norm": 3.4145364641074396,
133
- "learning_rate": 3.76191384433711e-06,
134
- "loss": 0.8613,
135
  "step": 85
136
  },
137
  {
138
- "epoch": 0.42,
139
- "grad_norm": 3.263977821412841,
140
- "learning_rate": 3.579385880846232e-06,
141
- "loss": 0.8759,
142
  "step": 90
143
  },
144
  {
145
- "epoch": 0.45,
146
- "grad_norm": 10.655125505241807,
147
- "learning_rate": 3.3894846173062917e-06,
148
- "loss": 0.8519,
149
  "step": 95
150
  },
151
  {
152
- "epoch": 0.47,
153
- "grad_norm": 250.56147092672978,
154
- "learning_rate": 3.193507271904612e-06,
155
- "loss": 0.8517,
156
  "step": 100
157
  },
158
  {
159
- "epoch": 0.5,
160
- "grad_norm": 3.419289400374,
161
- "learning_rate": 2.9927925686287006e-06,
162
- "loss": 0.8594,
163
  "step": 105
164
  },
165
  {
166
- "epoch": 0.52,
167
- "grad_norm": 3.263295130566819,
168
- "learning_rate": 2.788711592423966e-06,
169
- "loss": 0.849,
170
  "step": 110
171
  },
172
  {
173
- "epoch": 0.54,
174
- "grad_norm": 3.466296576337294,
175
- "learning_rate": 2.5826584232932707e-06,
176
- "loss": 0.8876,
177
  "step": 115
178
  },
179
  {
180
- "epoch": 0.57,
181
- "grad_norm": 3.464842972592367,
182
- "learning_rate": 2.376040613316944e-06,
183
- "loss": 0.8605,
184
  "step": 120
185
  },
186
  {
187
- "epoch": 0.59,
188
- "grad_norm": 3.301898509073216,
189
- "learning_rate": 2.1702695716448276e-06,
190
- "loss": 0.8485,
191
  "step": 125
192
  },
193
  {
194
- "epoch": 0.61,
195
- "grad_norm": 3.3264276395542414,
196
- "learning_rate": 1.9667509231406332e-06,
197
- "loss": 0.84,
198
  "step": 130
199
  },
200
  {
201
- "epoch": 0.64,
202
- "grad_norm": 3.2310503156717245,
203
- "learning_rate": 1.7668749065388385e-06,
204
- "loss": 0.835,
205
  "step": 135
206
  },
207
  {
208
- "epoch": 0.66,
209
- "grad_norm": 3.361233531250762,
210
- "learning_rate": 1.5720068777044479e-06,
211
- "loss": 0.8413,
212
  "step": 140
213
  },
214
  {
215
- "epoch": 0.68,
216
- "grad_norm": 3.229898438412517,
217
- "learning_rate": 1.383477982867984e-06,
218
- "loss": 0.8471,
219
  "step": 145
220
  },
221
  {
222
- "epoch": 0.71,
223
- "grad_norm": 3.082051479835871,
224
- "learning_rate": 1.2025760655469629e-06,
225
- "loss": 0.8435,
226
  "step": 150
227
  },
228
  {
229
- "epoch": 0.73,
230
- "grad_norm": 3.392093980236051,
231
- "learning_rate": 1.0305368692688175e-06,
232
- "loss": 0.8405,
233
  "step": 155
234
  },
235
  {
236
- "epoch": 0.75,
237
- "grad_norm": 3.3593945187641316,
238
- "learning_rate": 8.685355961895783e-07,
239
- "loss": 0.8475,
240
  "step": 160
241
  },
242
  {
243
- "epoch": 0.78,
244
- "grad_norm": 3.183216674509743,
245
- "learning_rate": 7.176788792715076e-07,
246
- "loss": 0.8551,
247
  "step": 165
248
  },
249
  {
250
- "epoch": 0.8,
251
- "grad_norm": 3.1059443889529974,
252
- "learning_rate": 5.78997222857853e-07,
253
- "loss": 0.8347,
254
  "step": 170
255
  },
256
  {
257
- "epoch": 0.83,
258
- "grad_norm": 3.1090745975048404,
259
- "learning_rate": 4.534379632832692e-07,
260
- "loss": 0.8251,
261
  "step": 175
262
  },
263
  {
264
- "epoch": 0.85,
265
- "grad_norm": 3.1186904842842007,
266
- "learning_rate": 3.4185879760606525e-07,
267
- "loss": 0.816,
268
  "step": 180
269
  },
270
  {
271
- "epoch": 0.87,
272
- "grad_norm": 13.97461867271608,
273
- "learning_rate": 2.450219246676028e-07,
274
- "loss": 0.8235,
275
  "step": 185
276
  },
277
  {
278
- "epoch": 0.9,
279
- "grad_norm": 3.0246555374160558,
280
- "learning_rate": 1.6358883850134815e-07,
281
- "loss": 0.8356,
282
  "step": 190
283
  },
284
  {
285
- "epoch": 0.92,
286
- "grad_norm": 2.969712888409742,
287
- "learning_rate": 9.811580965787965e-08,
288
- "loss": 0.8327,
289
  "step": 195
290
  },
291
  {
292
- "epoch": 0.94,
293
- "grad_norm": 2.9445398637597826,
294
- "learning_rate": 4.905008531297661e-08,
295
- "loss": 0.86,
296
  "step": 200
297
  },
298
  {
299
- "epoch": 0.97,
300
- "grad_norm": 3.096896769727839,
301
- "learning_rate": 1.6726834115904645e-08,
302
- "loss": 0.8216,
303
  "step": 205
304
  },
305
  {
306
- "epoch": 0.99,
307
- "grad_norm": 2.9776364501353694,
308
- "learning_rate": 1.3668566476848777e-09,
309
- "loss": 0.8468,
310
  "step": 210
311
  },
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
312
  {
313
  "epoch": 1.0,
314
- "eval_loss": 1.3747081756591797,
315
- "eval_runtime": 1.4843,
316
- "eval_samples_per_second": 9.432,
317
- "eval_steps_per_second": 0.674,
318
- "step": 212
319
  },
320
  {
321
  "epoch": 1.0,
322
- "step": 212,
323
- "total_flos": 88776974008320.0,
324
- "train_loss": 0.8946974671111917,
325
- "train_runtime": 2355.8103,
326
- "train_samples_per_second": 5.733,
327
- "train_steps_per_second": 0.09
328
  }
329
  ],
330
  "logging_steps": 5,
331
- "max_steps": 212,
332
  "num_input_tokens_seen": 0,
333
  "num_train_epochs": 1,
334
- "save_steps": 500,
335
- "total_flos": 88776974008320.0,
336
- "train_batch_size": 8,
337
  "trial_name": null,
338
  "trial_params": null
339
  }
 
3
  "best_model_checkpoint": null,
4
  "epoch": 1.0,
5
  "eval_steps": 500,
6
+ "global_step": 1107,
7
  "is_hyper_param_search": false,
8
  "is_local_process_zero": true,
9
  "is_world_process_zero": true,
10
  "log_history": [
11
  {
12
  "epoch": 0.0,
13
+ "grad_norm": 7.434891346998684,
14
+ "learning_rate": 1.801801801801802e-07,
15
+ "loss": 1.1387,
16
  "step": 1
17
  },
18
  {
19
+ "epoch": 0.0,
20
+ "grad_norm": 6.054112603313701,
21
+ "learning_rate": 9.00900900900901e-07,
22
+ "loss": 1.0767,
23
  "step": 5
24
  },
25
  {
26
+ "epoch": 0.01,
27
+ "grad_norm": 2.720345942108079,
28
+ "learning_rate": 1.801801801801802e-06,
29
+ "loss": 1.0177,
30
  "step": 10
31
  },
32
  {
33
+ "epoch": 0.01,
34
+ "grad_norm": 3.1804731128565717,
35
+ "learning_rate": 2.702702702702703e-06,
36
+ "loss": 1.0144,
37
  "step": 15
38
  },
39
  {
40
+ "epoch": 0.02,
41
+ "grad_norm": 2.301849657938967,
42
+ "learning_rate": 3.603603603603604e-06,
43
+ "loss": 0.9791,
44
  "step": 20
45
  },
46
  {
47
+ "epoch": 0.02,
48
+ "grad_norm": 2.4337369010470637,
49
+ "learning_rate": 4.504504504504505e-06,
50
+ "loss": 0.9802,
51
  "step": 25
52
  },
53
  {
54
+ "epoch": 0.03,
55
+ "grad_norm": 1.9302145168774039,
56
+ "learning_rate": 5.405405405405406e-06,
57
+ "loss": 0.986,
58
  "step": 30
59
  },
60
  {
61
+ "epoch": 0.03,
62
+ "grad_norm": 1.8524872666819012,
63
+ "learning_rate": 6.3063063063063065e-06,
64
+ "loss": 0.9589,
65
  "step": 35
66
  },
67
  {
68
+ "epoch": 0.04,
69
+ "grad_norm": 1.8144517560069848,
70
+ "learning_rate": 7.207207207207208e-06,
71
+ "loss": 0.9633,
72
  "step": 40
73
  },
74
  {
75
+ "epoch": 0.04,
76
+ "grad_norm": 2.010937969129206,
77
+ "learning_rate": 8.108108108108109e-06,
78
+ "loss": 0.9687,
79
  "step": 45
80
  },
81
  {
82
+ "epoch": 0.05,
83
+ "grad_norm": 2.004039674415607,
84
+ "learning_rate": 9.00900900900901e-06,
85
+ "loss": 0.9419,
86
  "step": 50
87
  },
88
  {
89
+ "epoch": 0.05,
90
+ "grad_norm": 1.9619351066753616,
91
+ "learning_rate": 9.90990990990991e-06,
92
+ "loss": 0.9751,
93
  "step": 55
94
  },
95
  {
96
+ "epoch": 0.05,
97
+ "grad_norm": 2.117833280361577,
98
+ "learning_rate": 1.0810810810810812e-05,
99
+ "loss": 0.9631,
100
  "step": 60
101
  },
102
  {
103
+ "epoch": 0.06,
104
+ "grad_norm": 3.1049875306196606,
105
+ "learning_rate": 1.1711711711711713e-05,
106
+ "loss": 0.9668,
107
  "step": 65
108
  },
109
  {
110
+ "epoch": 0.06,
111
+ "grad_norm": 2.214128699414308,
112
+ "learning_rate": 1.2612612612612613e-05,
113
+ "loss": 0.9635,
114
  "step": 70
115
  },
116
  {
117
+ "epoch": 0.07,
118
+ "grad_norm": 1.737941228120663,
119
+ "learning_rate": 1.3513513513513515e-05,
120
+ "loss": 0.976,
121
  "step": 75
122
  },
123
  {
124
+ "epoch": 0.07,
125
+ "grad_norm": 1.8456119696518833,
126
+ "learning_rate": 1.4414414414414416e-05,
127
+ "loss": 0.9846,
128
  "step": 80
129
  },
130
  {
131
+ "epoch": 0.08,
132
+ "grad_norm": 1.7815352366071144,
133
+ "learning_rate": 1.5315315315315316e-05,
134
+ "loss": 0.9783,
135
  "step": 85
136
  },
137
  {
138
+ "epoch": 0.08,
139
+ "grad_norm": 2.0381827638345844,
140
+ "learning_rate": 1.6216216216216218e-05,
141
+ "loss": 0.9701,
142
  "step": 90
143
  },
144
  {
145
+ "epoch": 0.09,
146
+ "grad_norm": 3.8119526094398513,
147
+ "learning_rate": 1.711711711711712e-05,
148
+ "loss": 0.9867,
149
  "step": 95
150
  },
151
  {
152
+ "epoch": 0.09,
153
+ "grad_norm": 2.3244802021753816,
154
+ "learning_rate": 1.801801801801802e-05,
155
+ "loss": 0.9799,
156
  "step": 100
157
  },
158
  {
159
+ "epoch": 0.09,
160
+ "grad_norm": 2.193951315428562,
161
+ "learning_rate": 1.891891891891892e-05,
162
+ "loss": 1.0084,
163
  "step": 105
164
  },
165
  {
166
+ "epoch": 0.1,
167
+ "grad_norm": 2.2121294806446365,
168
+ "learning_rate": 1.981981981981982e-05,
169
+ "loss": 0.9731,
170
  "step": 110
171
  },
172
  {
173
+ "epoch": 0.1,
174
+ "grad_norm": 2.105745826240662,
175
+ "learning_rate": 1.999920408755684e-05,
176
+ "loss": 0.9968,
177
  "step": 115
178
  },
179
  {
180
+ "epoch": 0.11,
181
+ "grad_norm": 2.0624008245016165,
182
+ "learning_rate": 1.9995970910394228e-05,
183
+ "loss": 1.0007,
184
  "step": 120
185
  },
186
  {
187
+ "epoch": 0.11,
188
+ "grad_norm": 1.9216817238638413,
189
+ "learning_rate": 1.9990251527524178e-05,
190
+ "loss": 0.9864,
191
  "step": 125
192
  },
193
  {
194
+ "epoch": 0.12,
195
+ "grad_norm": 1.9416400115913934,
196
+ "learning_rate": 1.998204736147608e-05,
197
+ "loss": 0.982,
198
  "step": 130
199
  },
200
  {
201
+ "epoch": 0.12,
202
+ "grad_norm": 1.992023353075518,
203
+ "learning_rate": 1.9971360452796523e-05,
204
+ "loss": 0.9901,
205
  "step": 135
206
  },
207
  {
208
+ "epoch": 0.13,
209
+ "grad_norm": 2.082117848176388,
210
+ "learning_rate": 1.9958193459541804e-05,
211
+ "loss": 1.0065,
212
  "step": 140
213
  },
214
  {
215
+ "epoch": 0.13,
216
+ "grad_norm": 1.9564902575341125,
217
+ "learning_rate": 1.994254965661679e-05,
218
+ "loss": 1.0058,
219
  "step": 145
220
  },
221
  {
222
+ "epoch": 0.14,
223
+ "grad_norm": 25.696326656261338,
224
+ "learning_rate": 1.9924432934960384e-05,
225
+ "loss": 1.1023,
226
  "step": 150
227
  },
228
  {
229
+ "epoch": 0.14,
230
+ "grad_norm": 15.718602388026698,
231
+ "learning_rate": 1.9903847800577777e-05,
232
+ "loss": 1.1922,
233
  "step": 155
234
  },
235
  {
236
+ "epoch": 0.14,
237
+ "grad_norm": 6.361436623798567,
238
+ "learning_rate": 1.9880799373419698e-05,
239
+ "loss": 1.1022,
240
  "step": 160
241
  },
242
  {
243
+ "epoch": 0.15,
244
+ "grad_norm": 14.018808638392496,
245
+ "learning_rate": 1.9855293386108995e-05,
246
+ "loss": 1.0509,
247
  "step": 165
248
  },
249
  {
250
+ "epoch": 0.15,
251
+ "grad_norm": 2.396996729925699,
252
+ "learning_rate": 1.982733618251478e-05,
253
+ "loss": 1.0454,
254
  "step": 170
255
  },
256
  {
257
+ "epoch": 0.16,
258
+ "grad_norm": 2.0783372576733012,
259
+ "learning_rate": 1.979693471617462e-05,
260
+ "loss": 1.0209,
261
  "step": 175
262
  },
263
  {
264
+ "epoch": 0.16,
265
+ "grad_norm": 2.3172426926555207,
266
+ "learning_rate": 1.976409654856501e-05,
267
+ "loss": 1.0314,
268
  "step": 180
269
  },
270
  {
271
+ "epoch": 0.17,
272
+ "grad_norm": 2.0444492620330186,
273
+ "learning_rate": 1.97288298472207e-05,
274
+ "loss": 1.003,
275
  "step": 185
276
  },
277
  {
278
+ "epoch": 0.17,
279
+ "grad_norm": 1.881906410360296,
280
+ "learning_rate": 1.969114338370324e-05,
281
+ "loss": 1.0024,
282
  "step": 190
283
  },
284
  {
285
+ "epoch": 0.18,
286
+ "grad_norm": 1.6740200356982151,
287
+ "learning_rate": 1.9651046531419335e-05,
288
+ "loss": 1.0041,
289
  "step": 195
290
  },
291
  {
292
+ "epoch": 0.18,
293
+ "grad_norm": 1.744682737538121,
294
+ "learning_rate": 1.960854926328946e-05,
295
+ "loss": 1.0108,
296
  "step": 200
297
  },
298
  {
299
+ "epoch": 0.19,
300
+ "grad_norm": 1.7487489845353397,
301
+ "learning_rate": 1.9563662149267405e-05,
302
+ "loss": 1.0009,
303
  "step": 205
304
  },
305
  {
306
+ "epoch": 0.19,
307
+ "grad_norm": 1.7482821787721043,
308
+ "learning_rate": 1.9516396353711297e-05,
309
+ "loss": 1.008,
310
  "step": 210
311
  },
312
+ {
313
+ "epoch": 0.19,
314
+ "grad_norm": 1.7209209399401664,
315
+ "learning_rate": 1.946676363260679e-05,
316
+ "loss": 0.9967,
317
+ "step": 215
318
+ },
319
+ {
320
+ "epoch": 0.2,
321
+ "grad_norm": 1.6520280771666889,
322
+ "learning_rate": 1.9414776330643126e-05,
323
+ "loss": 0.991,
324
+ "step": 220
325
+ },
326
+ {
327
+ "epoch": 0.2,
328
+ "grad_norm": 1.899715362613167,
329
+ "learning_rate": 1.936044737814273e-05,
330
+ "loss": 1.0021,
331
+ "step": 225
332
+ },
333
+ {
334
+ "epoch": 0.21,
335
+ "grad_norm": 1.6989493561259785,
336
+ "learning_rate": 1.9303790287845183e-05,
337
+ "loss": 0.9902,
338
+ "step": 230
339
+ },
340
+ {
341
+ "epoch": 0.21,
342
+ "grad_norm": 1.7638925291836884,
343
+ "learning_rate": 1.9244819151546325e-05,
344
+ "loss": 0.9976,
345
+ "step": 235
346
+ },
347
+ {
348
+ "epoch": 0.22,
349
+ "grad_norm": 1.564735067537867,
350
+ "learning_rate": 1.9183548636593322e-05,
351
+ "loss": 0.9787,
352
+ "step": 240
353
+ },
354
+ {
355
+ "epoch": 0.22,
356
+ "grad_norm": 1.7635286854536336,
357
+ "learning_rate": 1.9119993982236608e-05,
358
+ "loss": 0.9937,
359
+ "step": 245
360
+ },
361
+ {
362
+ "epoch": 0.23,
363
+ "grad_norm": 1.5741256061070714,
364
+ "learning_rate": 1.9054170995839546e-05,
365
+ "loss": 0.9648,
366
+ "step": 250
367
+ },
368
+ {
369
+ "epoch": 0.23,
370
+ "grad_norm": 1.7091189941765559,
371
+ "learning_rate": 1.8986096048946826e-05,
372
+ "loss": 0.9818,
373
+ "step": 255
374
+ },
375
+ {
376
+ "epoch": 0.23,
377
+ "grad_norm": 1.5405427151225155,
378
+ "learning_rate": 1.8915786073212508e-05,
379
+ "loss": 0.9958,
380
+ "step": 260
381
+ },
382
+ {
383
+ "epoch": 0.24,
384
+ "grad_norm": 1.5492700267103499,
385
+ "learning_rate": 1.8843258556188787e-05,
386
+ "loss": 0.9924,
387
+ "step": 265
388
+ },
389
+ {
390
+ "epoch": 0.24,
391
+ "grad_norm": 1.6206079313144814,
392
+ "learning_rate": 1.8768531536976452e-05,
393
+ "loss": 0.9804,
394
+ "step": 270
395
+ },
396
+ {
397
+ "epoch": 0.25,
398
+ "grad_norm": 1.5730537338272383,
399
+ "learning_rate": 1.86916236017382e-05,
400
+ "loss": 0.9847,
401
+ "step": 275
402
+ },
403
+ {
404
+ "epoch": 0.25,
405
+ "grad_norm": 1.6501336616437525,
406
+ "learning_rate": 1.8612553879075875e-05,
407
+ "loss": 0.983,
408
+ "step": 280
409
+ },
410
+ {
411
+ "epoch": 0.26,
412
+ "grad_norm": 1.5113060463908174,
413
+ "learning_rate": 1.8531342035272768e-05,
414
+ "loss": 0.981,
415
+ "step": 285
416
+ },
417
+ {
418
+ "epoch": 0.26,
419
+ "grad_norm": 1.5929173745457115,
420
+ "learning_rate": 1.844800826940223e-05,
421
+ "loss": 0.9789,
422
+ "step": 290
423
+ },
424
+ {
425
+ "epoch": 0.27,
426
+ "grad_norm": 1.598141976179977,
427
+ "learning_rate": 1.836257330830372e-05,
428
+ "loss": 1.0036,
429
+ "step": 295
430
+ },
431
+ {
432
+ "epoch": 0.27,
433
+ "grad_norm": 1.8798731248860165,
434
+ "learning_rate": 1.8275058401427622e-05,
435
+ "loss": 0.9704,
436
+ "step": 300
437
+ },
438
+ {
439
+ "epoch": 0.28,
440
+ "grad_norm": 2.195099845300529,
441
+ "learning_rate": 1.8185485315550062e-05,
442
+ "loss": 0.968,
443
+ "step": 305
444
+ },
445
+ {
446
+ "epoch": 0.28,
447
+ "grad_norm": 1.6333370138254395,
448
+ "learning_rate": 1.809387632935906e-05,
449
+ "loss": 0.9884,
450
+ "step": 310
451
+ },
452
+ {
453
+ "epoch": 0.28,
454
+ "grad_norm": 1.6383788031138038,
455
+ "learning_rate": 1.8000254227913346e-05,
456
+ "loss": 0.976,
457
+ "step": 315
458
+ },
459
+ {
460
+ "epoch": 0.29,
461
+ "grad_norm": 1.59375997270626,
462
+ "learning_rate": 1.7904642296975263e-05,
463
+ "loss": 0.9752,
464
+ "step": 320
465
+ },
466
+ {
467
+ "epoch": 0.29,
468
+ "grad_norm": 1.5470092112631526,
469
+ "learning_rate": 1.7807064317219096e-05,
470
+ "loss": 0.9684,
471
+ "step": 325
472
+ },
473
+ {
474
+ "epoch": 0.3,
475
+ "grad_norm": 1.5761252347776267,
476
+ "learning_rate": 1.7707544558316332e-05,
477
+ "loss": 0.983,
478
+ "step": 330
479
+ },
480
+ {
481
+ "epoch": 0.3,
482
+ "grad_norm": 1.4924968164344596,
483
+ "learning_rate": 1.760610777289929e-05,
484
+ "loss": 0.9843,
485
+ "step": 335
486
+ },
487
+ {
488
+ "epoch": 0.31,
489
+ "grad_norm": 1.4359693377296177,
490
+ "learning_rate": 1.7502779190404615e-05,
491
+ "loss": 0.9634,
492
+ "step": 340
493
+ },
494
+ {
495
+ "epoch": 0.31,
496
+ "grad_norm": 1.5878997655742073,
497
+ "learning_rate": 1.7397584510798208e-05,
498
+ "loss": 0.9758,
499
+ "step": 345
500
+ },
501
+ {
502
+ "epoch": 0.32,
503
+ "grad_norm": 1.6814035465687263,
504
+ "learning_rate": 1.7290549898183113e-05,
505
+ "loss": 0.967,
506
+ "step": 350
507
+ },
508
+ {
509
+ "epoch": 0.32,
510
+ "grad_norm": 1.4871838892694749,
511
+ "learning_rate": 1.7181701974291927e-05,
512
+ "loss": 0.953,
513
+ "step": 355
514
+ },
515
+ {
516
+ "epoch": 0.33,
517
+ "grad_norm": 1.677903630677117,
518
+ "learning_rate": 1.7071067811865477e-05,
519
+ "loss": 0.9638,
520
+ "step": 360
521
+ },
522
+ {
523
+ "epoch": 0.33,
524
+ "grad_norm": 1.7990895100977344,
525
+ "learning_rate": 1.6958674927919213e-05,
526
+ "loss": 0.9904,
527
+ "step": 365
528
+ },
529
+ {
530
+ "epoch": 0.33,
531
+ "grad_norm": 1.577389298855143,
532
+ "learning_rate": 1.6844551276899184e-05,
533
+ "loss": 0.9714,
534
+ "step": 370
535
+ },
536
+ {
537
+ "epoch": 0.34,
538
+ "grad_norm": 1.6428970048321387,
539
+ "learning_rate": 1.672872524372919e-05,
540
+ "loss": 0.9925,
541
+ "step": 375
542
+ },
543
+ {
544
+ "epoch": 0.34,
545
+ "grad_norm": 1.6313220262054722,
546
+ "learning_rate": 1.6611225636750838e-05,
547
+ "loss": 0.9579,
548
+ "step": 380
549
+ },
550
+ {
551
+ "epoch": 0.35,
552
+ "grad_norm": 1.4972553968322877,
553
+ "learning_rate": 1.649208168055833e-05,
554
+ "loss": 0.9747,
555
+ "step": 385
556
+ },
557
+ {
558
+ "epoch": 0.35,
559
+ "grad_norm": 1.5447739295558658,
560
+ "learning_rate": 1.637132300872969e-05,
561
+ "loss": 0.9805,
562
+ "step": 390
563
+ },
564
+ {
565
+ "epoch": 0.36,
566
+ "grad_norm": 1.5876140330526054,
567
+ "learning_rate": 1.6248979656456273e-05,
568
+ "loss": 0.9684,
569
+ "step": 395
570
+ },
571
+ {
572
+ "epoch": 0.36,
573
+ "grad_norm": 1.5582777256666884,
574
+ "learning_rate": 1.6125082053072408e-05,
575
+ "loss": 0.957,
576
+ "step": 400
577
+ },
578
+ {
579
+ "epoch": 0.37,
580
+ "grad_norm": 1.5476213766084626,
581
+ "learning_rate": 1.5999661014486956e-05,
582
+ "loss": 0.9861,
583
+ "step": 405
584
+ },
585
+ {
586
+ "epoch": 0.37,
587
+ "grad_norm": 1.6383105709040227,
588
+ "learning_rate": 1.58727477355188e-05,
589
+ "loss": 0.9793,
590
+ "step": 410
591
+ },
592
+ {
593
+ "epoch": 0.37,
594
+ "grad_norm": 1.5270291237304714,
595
+ "learning_rate": 1.5744373782137993e-05,
596
+ "loss": 0.9608,
597
+ "step": 415
598
+ },
599
+ {
600
+ "epoch": 0.38,
601
+ "grad_norm": 1.5686224715893557,
602
+ "learning_rate": 1.5614571083614683e-05,
603
+ "loss": 0.975,
604
+ "step": 420
605
+ },
606
+ {
607
+ "epoch": 0.38,
608
+ "grad_norm": 1.5293178485058705,
609
+ "learning_rate": 1.5483371924577633e-05,
610
+ "loss": 0.9632,
611
+ "step": 425
612
+ },
613
+ {
614
+ "epoch": 0.39,
615
+ "grad_norm": 1.4815279637987373,
616
+ "learning_rate": 1.535080893698435e-05,
617
+ "loss": 0.9689,
618
+ "step": 430
619
+ },
620
+ {
621
+ "epoch": 0.39,
622
+ "grad_norm": 1.5169260213036269,
623
+ "learning_rate": 1.5216915092004847e-05,
624
+ "loss": 0.9809,
625
+ "step": 435
626
+ },
627
+ {
628
+ "epoch": 0.4,
629
+ "grad_norm": 1.4976729343178568,
630
+ "learning_rate": 1.5081723691821029e-05,
631
+ "loss": 0.9712,
632
+ "step": 440
633
+ },
634
+ {
635
+ "epoch": 0.4,
636
+ "grad_norm": 1.4442693064244245,
637
+ "learning_rate": 1.4945268361343747e-05,
638
+ "loss": 0.9815,
639
+ "step": 445
640
+ },
641
+ {
642
+ "epoch": 0.41,
643
+ "grad_norm": 1.55780608281581,
644
+ "learning_rate": 1.4807583039849589e-05,
645
+ "loss": 0.9872,
646
+ "step": 450
647
+ },
648
+ {
649
+ "epoch": 0.41,
650
+ "grad_norm": 1.454973247549993,
651
+ "learning_rate": 1.4668701972539459e-05,
652
+ "loss": 0.953,
653
+ "step": 455
654
+ },
655
+ {
656
+ "epoch": 0.42,
657
+ "grad_norm": 1.5744354457111398,
658
+ "learning_rate": 1.4528659702021108e-05,
659
+ "loss": 0.9569,
660
+ "step": 460
661
+ },
662
+ {
663
+ "epoch": 0.42,
664
+ "grad_norm": 1.5374683976132577,
665
+ "learning_rate": 1.4387491059717653e-05,
666
+ "loss": 0.9544,
667
+ "step": 465
668
+ },
669
+ {
670
+ "epoch": 0.42,
671
+ "grad_norm": 1.5356499554288368,
672
+ "learning_rate": 1.4245231157204282e-05,
673
+ "loss": 0.9762,
674
+ "step": 470
675
+ },
676
+ {
677
+ "epoch": 0.43,
678
+ "grad_norm": 1.451816573803636,
679
+ "learning_rate": 1.4101915377475275e-05,
680
+ "loss": 0.9484,
681
+ "step": 475
682
+ },
683
+ {
684
+ "epoch": 0.43,
685
+ "grad_norm": 1.4189149204667209,
686
+ "learning_rate": 1.3957579366143521e-05,
687
+ "loss": 0.9568,
688
+ "step": 480
689
+ },
690
+ {
691
+ "epoch": 0.44,
692
+ "grad_norm": 1.4438844550057277,
693
+ "learning_rate": 1.3812259022574717e-05,
694
+ "loss": 0.9678,
695
+ "step": 485
696
+ },
697
+ {
698
+ "epoch": 0.44,
699
+ "grad_norm": 1.4944146133812288,
700
+ "learning_rate": 1.3665990490958438e-05,
701
+ "loss": 0.9684,
702
+ "step": 490
703
+ },
704
+ {
705
+ "epoch": 0.45,
706
+ "grad_norm": 1.567533543577245,
707
+ "learning_rate": 1.351881015131833e-05,
708
+ "loss": 0.9523,
709
+ "step": 495
710
+ },
711
+ {
712
+ "epoch": 0.45,
713
+ "grad_norm": 1.4298833694464113,
714
+ "learning_rate": 1.3370754610463655e-05,
715
+ "loss": 0.9547,
716
+ "step": 500
717
+ },
718
+ {
719
+ "epoch": 0.46,
720
+ "grad_norm": 1.586075303116762,
721
+ "learning_rate": 1.3221860692884396e-05,
722
+ "loss": 0.9621,
723
+ "step": 505
724
+ },
725
+ {
726
+ "epoch": 0.46,
727
+ "grad_norm": 1.4390997480170529,
728
+ "learning_rate": 1.307216543159225e-05,
729
+ "loss": 0.9361,
730
+ "step": 510
731
+ },
732
+ {
733
+ "epoch": 0.47,
734
+ "grad_norm": 1.5962792654589735,
735
+ "learning_rate": 1.2921706058909757e-05,
736
+ "loss": 0.952,
737
+ "step": 515
738
+ },
739
+ {
740
+ "epoch": 0.47,
741
+ "grad_norm": 1.5112017838877818,
742
+ "learning_rate": 1.2770519997209837e-05,
743
+ "loss": 0.9501,
744
+ "step": 520
745
+ },
746
+ {
747
+ "epoch": 0.47,
748
+ "grad_norm": 1.46430521195488,
749
+ "learning_rate": 1.2618644849608068e-05,
750
+ "loss": 0.9656,
751
+ "step": 525
752
+ },
753
+ {
754
+ "epoch": 0.48,
755
+ "grad_norm": 1.472561859950697,
756
+ "learning_rate": 1.246611839061002e-05,
757
+ "loss": 0.9545,
758
+ "step": 530
759
+ },
760
+ {
761
+ "epoch": 0.48,
762
+ "grad_norm": 1.5276258975583332,
763
+ "learning_rate": 1.2312978556715934e-05,
764
+ "loss": 0.9502,
765
+ "step": 535
766
+ },
767
+ {
768
+ "epoch": 0.49,
769
+ "grad_norm": 1.5767752712595098,
770
+ "learning_rate": 1.2159263436985139e-05,
771
+ "loss": 0.9497,
772
+ "step": 540
773
+ },
774
+ {
775
+ "epoch": 0.49,
776
+ "grad_norm": 1.4814084910286585,
777
+ "learning_rate": 1.2005011263562514e-05,
778
+ "loss": 0.953,
779
+ "step": 545
780
+ },
781
+ {
782
+ "epoch": 0.5,
783
+ "grad_norm": 1.4811159993074694,
784
+ "learning_rate": 1.185026040216934e-05,
785
+ "loss": 0.9517,
786
+ "step": 550
787
+ },
788
+ {
789
+ "epoch": 0.5,
790
+ "grad_norm": 1.5425020495972324,
791
+ "learning_rate": 1.1695049342560969e-05,
792
+ "loss": 0.9536,
793
+ "step": 555
794
+ },
795
+ {
796
+ "epoch": 0.51,
797
+ "grad_norm": 1.4813121305496708,
798
+ "learning_rate": 1.1539416688953613e-05,
799
+ "loss": 0.9566,
800
+ "step": 560
801
+ },
802
+ {
803
+ "epoch": 0.51,
804
+ "grad_norm": 1.5386521152381667,
805
+ "learning_rate": 1.138340115042267e-05,
806
+ "loss": 0.968,
807
+ "step": 565
808
+ },
809
+ {
810
+ "epoch": 0.51,
811
+ "grad_norm": 1.4985329754887164,
812
+ "learning_rate": 1.1227041531274978e-05,
813
+ "loss": 0.9536,
814
+ "step": 570
815
+ },
816
+ {
817
+ "epoch": 0.52,
818
+ "grad_norm": 1.7426405055010268,
819
+ "learning_rate": 1.1070376721397374e-05,
820
+ "loss": 0.9387,
821
+ "step": 575
822
+ },
823
+ {
824
+ "epoch": 0.52,
825
+ "grad_norm": 1.4876962547232626,
826
+ "learning_rate": 1.0913445686583974e-05,
827
+ "loss": 0.9479,
828
+ "step": 580
829
+ },
830
+ {
831
+ "epoch": 0.53,
832
+ "grad_norm": 1.4005947575155968,
833
+ "learning_rate": 1.075628745884457e-05,
834
+ "loss": 0.94,
835
+ "step": 585
836
+ },
837
+ {
838
+ "epoch": 0.53,
839
+ "grad_norm": 1.4806276567215155,
840
+ "learning_rate": 1.0598941126696545e-05,
841
+ "loss": 0.9537,
842
+ "step": 590
843
+ },
844
+ {
845
+ "epoch": 0.54,
846
+ "grad_norm": 1.3867191265630952,
847
+ "learning_rate": 1.0441445825442773e-05,
848
+ "loss": 0.9362,
849
+ "step": 595
850
+ },
851
+ {
852
+ "epoch": 0.54,
853
+ "grad_norm": 1.4082068202931468,
854
+ "learning_rate": 1.0283840727437832e-05,
855
+ "loss": 0.9391,
856
+ "step": 600
857
+ },
858
+ {
859
+ "epoch": 0.55,
860
+ "grad_norm": 1.4462569599659194,
861
+ "learning_rate": 1.012616503234504e-05,
862
+ "loss": 0.9655,
863
+ "step": 605
864
+ },
865
+ {
866
+ "epoch": 0.55,
867
+ "grad_norm": 1.4280342668958195,
868
+ "learning_rate": 9.968457957386663e-06,
869
+ "loss": 0.9297,
870
+ "step": 610
871
+ },
872
+ {
873
+ "epoch": 0.56,
874
+ "grad_norm": 1.4502290522153605,
875
+ "learning_rate": 9.810758727589814e-06,
876
+ "loss": 0.9486,
877
+ "step": 615
878
+ },
879
+ {
880
+ "epoch": 0.56,
881
+ "grad_norm": 1.3531947713121897,
882
+ "learning_rate": 9.65310656603033e-06,
883
+ "loss": 0.9374,
884
+ "step": 620
885
+ },
886
+ {
887
+ "epoch": 0.56,
888
+ "grad_norm": 1.5003587863241752,
889
+ "learning_rate": 9.495540684077215e-06,
890
+ "loss": 0.952,
891
+ "step": 625
892
+ },
893
+ {
894
+ "epoch": 0.57,
895
+ "grad_norm": 1.4085207232358623,
896
+ "learning_rate": 9.338100271639932e-06,
897
+ "loss": 0.9211,
898
+ "step": 630
899
+ },
900
+ {
901
+ "epoch": 0.57,
902
+ "grad_norm": 1.4167180963126849,
903
+ "learning_rate": 9.180824487421077e-06,
904
+ "loss": 0.9291,
905
+ "step": 635
906
+ },
907
+ {
908
+ "epoch": 0.58,
909
+ "grad_norm": 1.444242534136093,
910
+ "learning_rate": 9.023752449176773e-06,
911
+ "loss": 0.9338,
912
+ "step": 640
913
+ },
914
+ {
915
+ "epoch": 0.58,
916
+ "grad_norm": 1.474627058570353,
917
+ "learning_rate": 8.866923223987303e-06,
918
+ "loss": 0.932,
919
+ "step": 645
920
+ },
921
+ {
922
+ "epoch": 0.59,
923
+ "grad_norm": 1.3786541919625397,
924
+ "learning_rate": 8.71037581854028e-06,
925
+ "loss": 0.9287,
926
+ "step": 650
927
+ },
928
+ {
929
+ "epoch": 0.59,
930
+ "grad_norm": 1.4174870026814845,
931
+ "learning_rate": 8.554149169428894e-06,
932
+ "loss": 0.9396,
933
+ "step": 655
934
+ },
935
+ {
936
+ "epoch": 0.6,
937
+ "grad_norm": 1.4346984007547974,
938
+ "learning_rate": 8.398282133467579e-06,
939
+ "loss": 0.9353,
940
+ "step": 660
941
+ },
942
+ {
943
+ "epoch": 0.6,
944
+ "grad_norm": 1.3995616959967054,
945
+ "learning_rate": 8.242813478027491e-06,
946
+ "loss": 0.9451,
947
+ "step": 665
948
+ },
949
+ {
950
+ "epoch": 0.61,
951
+ "grad_norm": 1.4298853430595138,
952
+ "learning_rate": 8.087781871394281e-06,
953
+ "loss": 0.9294,
954
+ "step": 670
955
+ },
956
+ {
957
+ "epoch": 0.61,
958
+ "grad_norm": 4.062626134056569,
959
+ "learning_rate": 7.93322587315047e-06,
960
+ "loss": 0.9486,
961
+ "step": 675
962
+ },
963
+ {
964
+ "epoch": 0.61,
965
+ "grad_norm": 2.593063209369072,
966
+ "learning_rate": 7.7791839245849e-06,
967
+ "loss": 0.9323,
968
+ "step": 680
969
+ },
970
+ {
971
+ "epoch": 0.62,
972
+ "grad_norm": 1.4402410302679418,
973
+ "learning_rate": 7.625694339131564e-06,
974
+ "loss": 0.9208,
975
+ "step": 685
976
+ },
977
+ {
978
+ "epoch": 0.62,
979
+ "grad_norm": 1.427719727713364,
980
+ "learning_rate": 7.4727952928402695e-06,
981
+ "loss": 0.9432,
982
+ "step": 690
983
+ },
984
+ {
985
+ "epoch": 0.63,
986
+ "grad_norm": 1.4229770799618295,
987
+ "learning_rate": 7.320524814881471e-06,
988
+ "loss": 0.926,
989
+ "step": 695
990
+ },
991
+ {
992
+ "epoch": 0.63,
993
+ "grad_norm": 1.4126909774748309,
994
+ "learning_rate": 7.1689207780876026e-06,
995
+ "loss": 0.9282,
996
+ "step": 700
997
+ },
998
+ {
999
+ "epoch": 0.64,
1000
+ "grad_norm": 1.4661343172163777,
1001
+ "learning_rate": 7.018020889533348e-06,
1002
+ "loss": 0.9245,
1003
+ "step": 705
1004
+ },
1005
+ {
1006
+ "epoch": 0.64,
1007
+ "grad_norm": 1.5393466522100154,
1008
+ "learning_rate": 6.867862681157067e-06,
1009
+ "loss": 0.9215,
1010
+ "step": 710
1011
+ },
1012
+ {
1013
+ "epoch": 0.65,
1014
+ "grad_norm": 1.412508168571422,
1015
+ "learning_rate": 6.718483500425868e-06,
1016
+ "loss": 0.9247,
1017
+ "step": 715
1018
+ },
1019
+ {
1020
+ "epoch": 0.65,
1021
+ "grad_norm": 1.4776147088210356,
1022
+ "learning_rate": 6.569920501046474e-06,
1023
+ "loss": 0.9219,
1024
+ "step": 720
1025
+ },
1026
+ {
1027
+ "epoch": 0.65,
1028
+ "grad_norm": 1.3425578015608433,
1029
+ "learning_rate": 6.42221063372436e-06,
1030
+ "loss": 0.9258,
1031
+ "step": 725
1032
+ },
1033
+ {
1034
+ "epoch": 0.66,
1035
+ "grad_norm": 1.4129522183319783,
1036
+ "learning_rate": 6.275390636973315e-06,
1037
+ "loss": 0.9192,
1038
+ "step": 730
1039
+ },
1040
+ {
1041
+ "epoch": 0.66,
1042
+ "grad_norm": 1.4289303694831434,
1043
+ "learning_rate": 6.129497027977829e-06,
1044
+ "loss": 0.9189,
1045
+ "step": 735
1046
+ },
1047
+ {
1048
+ "epoch": 0.67,
1049
+ "grad_norm": 1.3710954157535182,
1050
+ "learning_rate": 5.9845660935105084e-06,
1051
+ "loss": 0.9164,
1052
+ "step": 740
1053
+ },
1054
+ {
1055
+ "epoch": 0.67,
1056
+ "grad_norm": 1.4387200450753754,
1057
+ "learning_rate": 5.8406338809067874e-06,
1058
+ "loss": 0.9369,
1059
+ "step": 745
1060
+ },
1061
+ {
1062
+ "epoch": 0.68,
1063
+ "grad_norm": 1.4010474045719385,
1064
+ "learning_rate": 5.69773618909923e-06,
1065
+ "loss": 0.9244,
1066
+ "step": 750
1067
+ },
1068
+ {
1069
+ "epoch": 0.68,
1070
+ "grad_norm": 1.409153799110607,
1071
+ "learning_rate": 5.555908559713561e-06,
1072
+ "loss": 0.9118,
1073
+ "step": 755
1074
+ },
1075
+ {
1076
+ "epoch": 0.69,
1077
+ "grad_norm": 1.3657173927666795,
1078
+ "learning_rate": 5.4151862682287624e-06,
1079
+ "loss": 0.9142,
1080
+ "step": 760
1081
+ },
1082
+ {
1083
+ "epoch": 0.69,
1084
+ "grad_norm": 1.3963533437536293,
1085
+ "learning_rate": 5.2756043152032934e-06,
1086
+ "loss": 0.9176,
1087
+ "step": 765
1088
+ },
1089
+ {
1090
+ "epoch": 0.7,
1091
+ "grad_norm": 1.3216489386400923,
1092
+ "learning_rate": 5.137197417569739e-06,
1093
+ "loss": 0.908,
1094
+ "step": 770
1095
+ },
1096
+ {
1097
+ "epoch": 0.7,
1098
+ "grad_norm": 1.4412475309656017,
1099
+ "learning_rate": 5.000000000000003e-06,
1100
+ "loss": 0.9165,
1101
+ "step": 775
1102
+ },
1103
+ {
1104
+ "epoch": 0.7,
1105
+ "grad_norm": 1.4134533602820125,
1106
+ "learning_rate": 4.86404618634314e-06,
1107
+ "loss": 0.9279,
1108
+ "step": 780
1109
+ },
1110
+ {
1111
+ "epoch": 0.71,
1112
+ "grad_norm": 1.3837626714460547,
1113
+ "learning_rate": 4.729369791138085e-06,
1114
+ "loss": 0.9189,
1115
+ "step": 785
1116
+ },
1117
+ {
1118
+ "epoch": 0.71,
1119
+ "grad_norm": 1.397147185881214,
1120
+ "learning_rate": 4.596004311203243e-06,
1121
+ "loss": 0.9421,
1122
+ "step": 790
1123
+ },
1124
+ {
1125
+ "epoch": 0.72,
1126
+ "grad_norm": 1.3486469288795642,
1127
+ "learning_rate": 4.463982917305155e-06,
1128
+ "loss": 0.9156,
1129
+ "step": 795
1130
+ },
1131
+ {
1132
+ "epoch": 0.72,
1133
+ "grad_norm": 1.3475341827233354,
1134
+ "learning_rate": 4.333338445908225e-06,
1135
+ "loss": 0.9292,
1136
+ "step": 800
1137
+ },
1138
+ {
1139
+ "epoch": 0.73,
1140
+ "grad_norm": 1.3536202190201114,
1141
+ "learning_rate": 4.2041033910076235e-06,
1142
+ "loss": 0.8996,
1143
+ "step": 805
1144
+ },
1145
+ {
1146
+ "epoch": 0.73,
1147
+ "grad_norm": 1.3534435686443709,
1148
+ "learning_rate": 4.076309896047337e-06,
1149
+ "loss": 0.9357,
1150
+ "step": 810
1151
+ },
1152
+ {
1153
+ "epoch": 0.74,
1154
+ "grad_norm": 1.3961829341566565,
1155
+ "learning_rate": 3.9499897459254375e-06,
1156
+ "loss": 0.9233,
1157
+ "step": 815
1158
+ },
1159
+ {
1160
+ "epoch": 0.74,
1161
+ "grad_norm": 1.348649115175699,
1162
+ "learning_rate": 3.825174359088526e-06,
1163
+ "loss": 0.9097,
1164
+ "step": 820
1165
+ },
1166
+ {
1167
+ "epoch": 0.75,
1168
+ "grad_norm": 1.4476303062234663,
1169
+ "learning_rate": 3.7018947797172864e-06,
1170
+ "loss": 0.9274,
1171
+ "step": 825
1172
+ },
1173
+ {
1174
+ "epoch": 0.75,
1175
+ "grad_norm": 1.390535701834856,
1176
+ "learning_rate": 3.580181670005183e-06,
1177
+ "loss": 0.9184,
1178
+ "step": 830
1179
+ },
1180
+ {
1181
+ "epoch": 0.75,
1182
+ "grad_norm": 1.3785793159092763,
1183
+ "learning_rate": 3.4600653025321085e-06,
1184
+ "loss": 0.9055,
1185
+ "step": 835
1186
+ },
1187
+ {
1188
+ "epoch": 0.76,
1189
+ "grad_norm": 1.354137886395205,
1190
+ "learning_rate": 3.341575552734978e-06,
1191
+ "loss": 0.9109,
1192
+ "step": 840
1193
+ },
1194
+ {
1195
+ "epoch": 0.76,
1196
+ "grad_norm": 1.4067176317883785,
1197
+ "learning_rate": 3.224741891477096e-06,
1198
+ "loss": 0.9241,
1199
+ "step": 845
1200
+ },
1201
+ {
1202
+ "epoch": 0.77,
1203
+ "grad_norm": 1.374060716621096,
1204
+ "learning_rate": 3.1095933777181165e-06,
1205
+ "loss": 0.9118,
1206
+ "step": 850
1207
+ },
1208
+ {
1209
+ "epoch": 0.77,
1210
+ "grad_norm": 1.3942906283270295,
1211
+ "learning_rate": 2.9961586512864947e-06,
1212
+ "loss": 0.9,
1213
+ "step": 855
1214
+ },
1215
+ {
1216
+ "epoch": 0.78,
1217
+ "grad_norm": 1.4465506221807978,
1218
+ "learning_rate": 2.884465925756159e-06,
1219
+ "loss": 0.9242,
1220
+ "step": 860
1221
+ },
1222
+ {
1223
+ "epoch": 0.78,
1224
+ "grad_norm": 1.3396029078248526,
1225
+ "learning_rate": 2.7745429814292147e-06,
1226
+ "loss": 0.9241,
1227
+ "step": 865
1228
+ },
1229
+ {
1230
+ "epoch": 0.79,
1231
+ "grad_norm": 1.3098914704831672,
1232
+ "learning_rate": 2.666417158426393e-06,
1233
+ "loss": 0.9228,
1234
+ "step": 870
1235
+ },
1236
+ {
1237
+ "epoch": 0.79,
1238
+ "grad_norm": 1.434445807607541,
1239
+ "learning_rate": 2.5601153498870137e-06,
1240
+ "loss": 0.9191,
1241
+ "step": 875
1242
+ },
1243
+ {
1244
+ "epoch": 0.79,
1245
+ "grad_norm": 1.3696234318588858,
1246
+ "learning_rate": 2.4556639952800786e-06,
1247
+ "loss": 0.9216,
1248
+ "step": 880
1249
+ },
1250
+ {
1251
+ "epoch": 0.8,
1252
+ "grad_norm": 1.3799439546633658,
1253
+ "learning_rate": 2.353089073828255e-06,
1254
+ "loss": 0.9066,
1255
+ "step": 885
1256
+ },
1257
+ {
1258
+ "epoch": 0.8,
1259
+ "grad_norm": 1.3783696426493683,
1260
+ "learning_rate": 2.252416098046275e-06,
1261
+ "loss": 0.9102,
1262
+ "step": 890
1263
+ },
1264
+ {
1265
+ "epoch": 0.81,
1266
+ "grad_norm": 1.3692307220867967,
1267
+ "learning_rate": 2.153670107395456e-06,
1268
+ "loss": 0.8958,
1269
+ "step": 895
1270
+ },
1271
+ {
1272
+ "epoch": 0.81,
1273
+ "grad_norm": 1.3494441419685654,
1274
+ "learning_rate": 2.056875662055874e-06,
1275
+ "loss": 0.9144,
1276
+ "step": 900
1277
+ },
1278
+ {
1279
+ "epoch": 0.82,
1280
+ "grad_norm": 1.356794216227153,
1281
+ "learning_rate": 1.9620568368177183e-06,
1282
+ "loss": 0.8964,
1283
+ "step": 905
1284
+ },
1285
+ {
1286
+ "epoch": 0.82,
1287
+ "grad_norm": 1.3429980043030805,
1288
+ "learning_rate": 1.8692372150934113e-06,
1289
+ "loss": 0.9194,
1290
+ "step": 910
1291
+ },
1292
+ {
1293
+ "epoch": 0.83,
1294
+ "grad_norm": 1.339376480893687,
1295
+ "learning_rate": 1.7784398830519002e-06,
1296
+ "loss": 0.9093,
1297
+ "step": 915
1298
+ },
1299
+ {
1300
+ "epoch": 0.83,
1301
+ "grad_norm": 1.3355920171762052,
1302
+ "learning_rate": 1.6896874238766703e-06,
1303
+ "loss": 0.8913,
1304
+ "step": 920
1305
+ },
1306
+ {
1307
+ "epoch": 0.84,
1308
+ "grad_norm": 1.293120986101463,
1309
+ "learning_rate": 1.6030019121488227e-06,
1310
+ "loss": 0.9182,
1311
+ "step": 925
1312
+ },
1313
+ {
1314
+ "epoch": 0.84,
1315
+ "grad_norm": 1.4443458217709528,
1316
+ "learning_rate": 1.5184049083566688e-06,
1317
+ "loss": 0.9123,
1318
+ "step": 930
1319
+ },
1320
+ {
1321
+ "epoch": 0.84,
1322
+ "grad_norm": 1.3695869429477345,
1323
+ "learning_rate": 1.4359174535331998e-06,
1324
+ "loss": 0.9092,
1325
+ "step": 935
1326
+ },
1327
+ {
1328
+ "epoch": 0.85,
1329
+ "grad_norm": 1.339902530091704,
1330
+ "learning_rate": 1.3555600640227284e-06,
1331
+ "loss": 0.9254,
1332
+ "step": 940
1333
+ },
1334
+ {
1335
+ "epoch": 0.85,
1336
+ "grad_norm": 1.4388242190037084,
1337
+ "learning_rate": 1.2773527263780626e-06,
1338
+ "loss": 0.8972,
1339
+ "step": 945
1340
+ },
1341
+ {
1342
+ "epoch": 0.86,
1343
+ "grad_norm": 1.430786276692823,
1344
+ "learning_rate": 1.2013148923894213e-06,
1345
+ "loss": 0.9197,
1346
+ "step": 950
1347
+ },
1348
+ {
1349
+ "epoch": 0.86,
1350
+ "grad_norm": 1.3197356728810627,
1351
+ "learning_rate": 1.1274654742463842e-06,
1352
+ "loss": 0.897,
1353
+ "step": 955
1354
+ },
1355
+ {
1356
+ "epoch": 0.87,
1357
+ "grad_norm": 1.3361601981054119,
1358
+ "learning_rate": 1.0558228398340188e-06,
1359
+ "loss": 0.9094,
1360
+ "step": 960
1361
+ },
1362
+ {
1363
+ "epoch": 0.87,
1364
+ "grad_norm": 1.3263397846241956,
1365
+ "learning_rate": 9.86404808164426e-07,
1366
+ "loss": 0.8958,
1367
+ "step": 965
1368
+ },
1369
+ {
1370
+ "epoch": 0.88,
1371
+ "grad_norm": 1.3279928456335177,
1372
+ "learning_rate": 9.192286449447684e-07,
1373
+ "loss": 0.8967,
1374
+ "step": 970
1375
+ },
1376
+ {
1377
+ "epoch": 0.88,
1378
+ "grad_norm": 1.3258111261619026,
1379
+ "learning_rate": 8.543110582829272e-07,
1380
+ "loss": 0.9021,
1381
+ "step": 975
1382
+ },
1383
+ {
1384
+ "epoch": 0.89,
1385
+ "grad_norm": 1.2957246118366699,
1386
+ "learning_rate": 7.916681945318649e-07,
1387
+ "loss": 0.9083,
1388
+ "step": 980
1389
+ },
1390
+ {
1391
+ "epoch": 0.89,
1392
+ "grad_norm": 1.3406921495185697,
1393
+ "learning_rate": 7.313156342736738e-07,
1394
+ "loss": 0.898,
1395
+ "step": 985
1396
+ },
1397
+ {
1398
+ "epoch": 0.89,
1399
+ "grad_norm": 1.3197731448045178,
1400
+ "learning_rate": 6.732683884443736e-07,
1401
+ "loss": 0.8957,
1402
+ "step": 990
1403
+ },
1404
+ {
1405
+ "epoch": 0.9,
1406
+ "grad_norm": 1.3468993226799906,
1407
+ "learning_rate": 6.175408946003703e-07,
1408
+ "loss": 0.9035,
1409
+ "step": 995
1410
+ },
1411
+ {
1412
+ "epoch": 0.9,
1413
+ "grad_norm": 1.3353098250857058,
1414
+ "learning_rate": 5.641470133275473e-07,
1415
+ "loss": 0.894,
1416
+ "step": 1000
1417
+ },
1418
+ {
1419
+ "epoch": 0.91,
1420
+ "grad_norm": 1.3632991670701868,
1421
+ "learning_rate": 5.131000247938367e-07,
1422
+ "loss": 0.9149,
1423
+ "step": 1005
1424
+ },
1425
+ {
1426
+ "epoch": 0.91,
1427
+ "grad_norm": 1.3014286387338503,
1428
+ "learning_rate": 4.644126254461756e-07,
1429
+ "loss": 0.8919,
1430
+ "step": 1010
1431
+ },
1432
+ {
1433
+ "epoch": 0.92,
1434
+ "grad_norm": 1.3376939712764266,
1435
+ "learning_rate": 4.180969248526334e-07,
1436
+ "loss": 0.9151,
1437
+ "step": 1015
1438
+ },
1439
+ {
1440
+ "epoch": 0.92,
1441
+ "grad_norm": 1.316358705323747,
1442
+ "learning_rate": 3.7416444269050335e-07,
1443
+ "loss": 0.9109,
1444
+ "step": 1020
1445
+ },
1446
+ {
1447
+ "epoch": 0.93,
1448
+ "grad_norm": 1.351422091186982,
1449
+ "learning_rate": 3.326261058811331e-07,
1450
+ "loss": 0.9046,
1451
+ "step": 1025
1452
+ },
1453
+ {
1454
+ "epoch": 0.93,
1455
+ "grad_norm": 1.3147060361947367,
1456
+ "learning_rate": 2.9349224587215786e-07,
1457
+ "loss": 0.9036,
1458
+ "step": 1030
1459
+ },
1460
+ {
1461
+ "epoch": 0.93,
1462
+ "grad_norm": 1.376354495442706,
1463
+ "learning_rate": 2.5677259606786686e-07,
1464
+ "loss": 0.9109,
1465
+ "step": 1035
1466
+ },
1467
+ {
1468
+ "epoch": 0.94,
1469
+ "grad_norm": 1.318239765222968,
1470
+ "learning_rate": 2.2247628940829214e-07,
1471
+ "loss": 0.9011,
1472
+ "step": 1040
1473
+ },
1474
+ {
1475
+ "epoch": 0.94,
1476
+ "grad_norm": 1.334098597139642,
1477
+ "learning_rate": 1.9061185609766996e-07,
1478
+ "loss": 0.9158,
1479
+ "step": 1045
1480
+ },
1481
+ {
1482
+ "epoch": 0.95,
1483
+ "grad_norm": 1.311535708505789,
1484
+ "learning_rate": 1.6118722148278586e-07,
1485
+ "loss": 0.9069,
1486
+ "step": 1050
1487
+ },
1488
+ {
1489
+ "epoch": 0.95,
1490
+ "grad_norm": 1.3399145129480423,
1491
+ "learning_rate": 1.3420970408178912e-07,
1492
+ "loss": 0.9108,
1493
+ "step": 1055
1494
+ },
1495
+ {
1496
+ "epoch": 0.96,
1497
+ "grad_norm": 1.3369774366097849,
1498
+ "learning_rate": 1.0968601376391996e-07,
1499
+ "loss": 0.9022,
1500
+ "step": 1060
1501
+ },
1502
+ {
1503
+ "epoch": 0.96,
1504
+ "grad_norm": 1.327397654041781,
1505
+ "learning_rate": 8.762225008062675e-08,
1506
+ "loss": 0.9063,
1507
+ "step": 1065
1508
+ },
1509
+ {
1510
+ "epoch": 0.97,
1511
+ "grad_norm": 1.3398923885085607,
1512
+ "learning_rate": 6.802390074847731e-08,
1513
+ "loss": 0.92,
1514
+ "step": 1070
1515
+ },
1516
+ {
1517
+ "epoch": 0.97,
1518
+ "grad_norm": 1.3031847902685991,
1519
+ "learning_rate": 5.0895840284257424e-08,
1520
+ "loss": 0.8971,
1521
+ "step": 1075
1522
+ },
1523
+ {
1524
+ "epoch": 0.98,
1525
+ "grad_norm": 1.34860295723008,
1526
+ "learning_rate": 3.6242328792567286e-08,
1527
+ "loss": 0.9018,
1528
+ "step": 1080
1529
+ },
1530
+ {
1531
+ "epoch": 0.98,
1532
+ "grad_norm": 1.314620543755103,
1533
+ "learning_rate": 2.406701090625463e-08,
1534
+ "loss": 0.8847,
1535
+ "step": 1085
1536
+ },
1537
+ {
1538
+ "epoch": 0.98,
1539
+ "grad_norm": 1.315793020724077,
1540
+ "learning_rate": 1.4372914879909882e-08,
1541
+ "loss": 0.896,
1542
+ "step": 1090
1543
+ },
1544
+ {
1545
+ "epoch": 0.99,
1546
+ "grad_norm": 1.3160311821089112,
1547
+ "learning_rate": 7.162451836685291e-09,
1548
+ "loss": 0.9092,
1549
+ "step": 1095
1550
+ },
1551
+ {
1552
+ "epoch": 0.99,
1553
+ "grad_norm": 1.381770390973591,
1554
+ "learning_rate": 2.4374151685913063e-09,
1555
+ "loss": 0.9208,
1556
+ "step": 1100
1557
+ },
1558
+ {
1559
+ "epoch": 1.0,
1560
+ "grad_norm": 1.3304085867307816,
1561
+ "learning_rate": 1.989800904445005e-10,
1562
+ "loss": 0.9189,
1563
+ "step": 1105
1564
+ },
1565
  {
1566
  "epoch": 1.0,
1567
+ "eval_loss": 0.9156445264816284,
1568
+ "eval_runtime": 344.7542,
1569
+ "eval_samples_per_second": 45.482,
1570
+ "eval_steps_per_second": 0.711,
1571
+ "step": 1107
1572
  },
1573
  {
1574
  "epoch": 1.0,
1575
+ "step": 1107,
1576
+ "total_flos": 463566557675520.0,
1577
+ "train_loss": 0.9518642601066596,
1578
+ "train_runtime": 13070.0796,
1579
+ "train_samples_per_second": 10.839,
1580
+ "train_steps_per_second": 0.085
1581
  }
1582
  ],
1583
  "logging_steps": 5,
1584
+ "max_steps": 1107,
1585
  "num_input_tokens_seen": 0,
1586
  "num_train_epochs": 1,
1587
+ "save_steps": 100,
1588
+ "total_flos": 463566557675520.0,
1589
+ "train_batch_size": 16,
1590
  "trial_name": null,
1591
  "trial_params": null
1592
  }