RikkiXu commited on
Commit
6eeb6b9
1 Parent(s): c3f25ad

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 +1665 -65
all_results.json CHANGED
@@ -1,8 +1,8 @@
1
  {
2
- "epoch": 1.0,
3
- "train_loss": 0.908859632783017,
4
- "train_runtime": 1626.0912,
5
  "train_samples": 6599,
6
- "train_samples_per_second": 4.625,
7
- "train_steps_per_second": 0.036
8
  }
 
1
  {
2
+ "epoch": 5.0,
3
+ "train_loss": 0.1592100728509274,
4
+ "train_runtime": 9397.2382,
5
  "train_samples": 6599,
6
+ "train_samples_per_second": 3.994,
7
+ "train_steps_per_second": 0.125
8
  }
train_results.json CHANGED
@@ -1,8 +1,8 @@
1
  {
2
- "epoch": 1.0,
3
- "train_loss": 0.908859632783017,
4
- "train_runtime": 1626.0912,
5
  "train_samples": 6599,
6
- "train_samples_per_second": 4.625,
7
- "train_steps_per_second": 0.036
8
  }
 
1
  {
2
+ "epoch": 5.0,
3
+ "train_loss": 0.1592100728509274,
4
+ "train_runtime": 9397.2382,
5
  "train_samples": 6599,
6
+ "train_samples_per_second": 3.994,
7
+ "train_steps_per_second": 0.125
8
  }
trainer_state.json CHANGED
@@ -1,122 +1,1722 @@
1
  {
2
  "best_metric": null,
3
  "best_model_checkpoint": null,
4
- "epoch": 1.0,
5
  "eval_steps": 500,
6
- "global_step": 59,
7
  "is_hyper_param_search": false,
8
  "is_local_process_zero": true,
9
  "is_world_process_zero": true,
10
  "log_history": [
11
  {
12
- "epoch": 0.02,
13
- "grad_norm": 4.72473337945939,
14
- "learning_rate": 3.3333333333333333e-06,
15
- "loss": 0.9545,
16
  "step": 1
17
  },
18
  {
19
- "epoch": 0.08,
20
- "grad_norm": 15.606619949426117,
21
- "learning_rate": 1.6666666666666667e-05,
22
- "loss": 0.9459,
23
  "step": 5
24
  },
25
  {
26
- "epoch": 0.17,
27
- "grad_norm": 97.5546177933576,
28
- "learning_rate": 1.972022914080411e-05,
29
- "loss": 1.309,
30
  "step": 10
31
  },
32
  {
33
- "epoch": 0.25,
34
- "grad_norm": 7.451729536881689,
35
- "learning_rate": 1.8610436117673557e-05,
36
- "loss": 1.1022,
37
  "step": 15
38
  },
39
  {
40
- "epoch": 0.34,
41
- "grad_norm": 4.013413307542302,
42
- "learning_rate": 1.6749830015182106e-05,
43
- "loss": 0.9536,
44
  "step": 20
45
  },
46
  {
47
- "epoch": 0.42,
48
- "grad_norm": 2.5281389594304153,
49
- "learning_rate": 1.4300652022765207e-05,
50
- "loss": 0.9122,
51
  "step": 25
52
  },
53
  {
54
- "epoch": 0.51,
55
- "grad_norm": 2.4087072108196184,
56
- "learning_rate": 1.1476465640024814e-05,
57
- "loss": 0.8548,
58
  "step": 30
59
  },
60
  {
61
- "epoch": 0.59,
62
- "grad_norm": 2.127180353057553,
63
- "learning_rate": 8.52353435997519e-06,
64
- "loss": 0.8386,
65
  "step": 35
66
  },
67
  {
68
- "epoch": 0.68,
69
- "grad_norm": 1.9083395084428687,
70
- "learning_rate": 5.699347977234799e-06,
71
- "loss": 0.8324,
72
  "step": 40
73
  },
74
  {
75
- "epoch": 0.76,
76
- "grad_norm": 1.7712149272954687,
77
- "learning_rate": 3.250169984817897e-06,
78
- "loss": 0.7792,
79
  "step": 45
80
  },
81
  {
82
- "epoch": 0.85,
83
- "grad_norm": 1.6965019256204015,
84
- "learning_rate": 1.3895638823264447e-06,
85
- "loss": 0.77,
86
  "step": 50
87
  },
88
  {
89
- "epoch": 0.93,
90
- "grad_norm": 1.6261522519462899,
91
- "learning_rate": 2.7977085919589253e-07,
92
- "loss": 0.7609,
93
  "step": 55
94
  },
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
95
  {
96
  "epoch": 1.0,
97
- "eval_loss": 0.7076106071472168,
98
- "eval_runtime": 338.673,
99
- "eval_samples_per_second": 22.207,
100
- "eval_steps_per_second": 0.348,
101
- "step": 59
102
  },
103
  {
104
  "epoch": 1.0,
105
- "step": 59,
106
- "total_flos": 49413598740480.0,
107
- "train_loss": 0.908859632783017,
108
- "train_runtime": 1626.0912,
109
- "train_samples_per_second": 4.625,
110
- "train_steps_per_second": 0.036
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
111
  }
112
  ],
113
  "logging_steps": 5,
114
- "max_steps": 59,
115
  "num_input_tokens_seen": 0,
116
- "num_train_epochs": 1,
117
  "save_steps": 100,
118
- "total_flos": 49413598740480.0,
119
- "train_batch_size": 16,
120
  "trial_name": null,
121
  "trial_params": null
122
  }
 
1
  {
2
  "best_metric": null,
3
  "best_model_checkpoint": null,
4
+ "epoch": 5.0,
5
  "eval_steps": 500,
6
+ "global_step": 1175,
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": 3.3722460551119577,
14
+ "learning_rate": 1.6949152542372883e-07,
15
+ "loss": 0.4424,
16
  "step": 1
17
  },
18
  {
19
+ "epoch": 0.02,
20
+ "grad_norm": 3.3291982249982532,
21
+ "learning_rate": 8.474576271186441e-07,
22
+ "loss": 0.521,
23
  "step": 5
24
  },
25
  {
26
+ "epoch": 0.04,
27
+ "grad_norm": 3.366718903705411,
28
+ "learning_rate": 1.6949152542372882e-06,
29
+ "loss": 0.5267,
30
  "step": 10
31
  },
32
  {
33
+ "epoch": 0.06,
34
+ "grad_norm": 3.743534573457503,
35
+ "learning_rate": 2.5423728813559323e-06,
36
+ "loss": 0.5054,
37
  "step": 15
38
  },
39
  {
40
+ "epoch": 0.09,
41
+ "grad_norm": 2.6726105105201854,
42
+ "learning_rate": 3.3898305084745763e-06,
43
+ "loss": 0.5138,
44
  "step": 20
45
  },
46
  {
47
+ "epoch": 0.11,
48
+ "grad_norm": 2.5592230066929766,
49
+ "learning_rate": 4.23728813559322e-06,
50
+ "loss": 0.4656,
51
  "step": 25
52
  },
53
  {
54
+ "epoch": 0.13,
55
+ "grad_norm": 2.5555759369628186,
56
+ "learning_rate": 5.084745762711865e-06,
57
+ "loss": 0.4712,
58
  "step": 30
59
  },
60
  {
61
+ "epoch": 0.15,
62
+ "grad_norm": 2.5046298800000617,
63
+ "learning_rate": 5.932203389830509e-06,
64
+ "loss": 0.4611,
65
  "step": 35
66
  },
67
  {
68
+ "epoch": 0.17,
69
+ "grad_norm": 2.7136428613817194,
70
+ "learning_rate": 6.779661016949153e-06,
71
+ "loss": 0.4541,
72
  "step": 40
73
  },
74
  {
75
+ "epoch": 0.19,
76
+ "grad_norm": 3.010241923008894,
77
+ "learning_rate": 7.627118644067797e-06,
78
+ "loss": 0.4781,
79
  "step": 45
80
  },
81
  {
82
+ "epoch": 0.21,
83
+ "grad_norm": 2.475959902357205,
84
+ "learning_rate": 8.47457627118644e-06,
85
+ "loss": 0.442,
86
  "step": 50
87
  },
88
  {
89
+ "epoch": 0.23,
90
+ "grad_norm": 3.128773493576812,
91
+ "learning_rate": 9.322033898305085e-06,
92
+ "loss": 0.426,
93
  "step": 55
94
  },
95
+ {
96
+ "epoch": 0.26,
97
+ "grad_norm": 3.1009164137521252,
98
+ "learning_rate": 1.016949152542373e-05,
99
+ "loss": 0.4958,
100
+ "step": 60
101
+ },
102
+ {
103
+ "epoch": 0.28,
104
+ "grad_norm": 2.808266593808473,
105
+ "learning_rate": 1.1016949152542374e-05,
106
+ "loss": 0.46,
107
+ "step": 65
108
+ },
109
+ {
110
+ "epoch": 0.3,
111
+ "grad_norm": 2.6760608859412054,
112
+ "learning_rate": 1.1864406779661018e-05,
113
+ "loss": 0.4929,
114
+ "step": 70
115
+ },
116
+ {
117
+ "epoch": 0.32,
118
+ "grad_norm": 3.900783318495709,
119
+ "learning_rate": 1.2711864406779661e-05,
120
+ "loss": 0.4618,
121
+ "step": 75
122
+ },
123
+ {
124
+ "epoch": 0.34,
125
+ "grad_norm": 2.524216832199052,
126
+ "learning_rate": 1.3559322033898305e-05,
127
+ "loss": 0.4588,
128
+ "step": 80
129
+ },
130
+ {
131
+ "epoch": 0.36,
132
+ "grad_norm": 60.979865346510024,
133
+ "learning_rate": 1.440677966101695e-05,
134
+ "loss": 0.6492,
135
+ "step": 85
136
+ },
137
+ {
138
+ "epoch": 0.38,
139
+ "grad_norm": 15.672327487548742,
140
+ "learning_rate": 1.5254237288135594e-05,
141
+ "loss": 0.6082,
142
+ "step": 90
143
+ },
144
+ {
145
+ "epoch": 0.4,
146
+ "grad_norm": 4.6849329984217345,
147
+ "learning_rate": 1.6101694915254237e-05,
148
+ "loss": 0.5659,
149
+ "step": 95
150
+ },
151
+ {
152
+ "epoch": 0.43,
153
+ "grad_norm": 7.354203430908385,
154
+ "learning_rate": 1.694915254237288e-05,
155
+ "loss": 0.5671,
156
+ "step": 100
157
+ },
158
+ {
159
+ "epoch": 0.45,
160
+ "grad_norm": 13.233658769725766,
161
+ "learning_rate": 1.7796610169491526e-05,
162
+ "loss": 0.6546,
163
+ "step": 105
164
+ },
165
+ {
166
+ "epoch": 0.47,
167
+ "grad_norm": 8.816187325352404,
168
+ "learning_rate": 1.864406779661017e-05,
169
+ "loss": 0.6247,
170
+ "step": 110
171
+ },
172
+ {
173
+ "epoch": 0.49,
174
+ "grad_norm": 110.07956321249138,
175
+ "learning_rate": 1.9491525423728814e-05,
176
+ "loss": 0.6283,
177
+ "step": 115
178
+ },
179
+ {
180
+ "epoch": 0.51,
181
+ "grad_norm": 9.519856297335076,
182
+ "learning_rate": 1.999982332362188e-05,
183
+ "loss": 0.5585,
184
+ "step": 120
185
+ },
186
+ {
187
+ "epoch": 0.53,
188
+ "grad_norm": 7.0894189196386375,
189
+ "learning_rate": 1.999783578606323e-05,
190
+ "loss": 0.5264,
191
+ "step": 125
192
+ },
193
+ {
194
+ "epoch": 0.55,
195
+ "grad_norm": 5.230200581327651,
196
+ "learning_rate": 1.999364030586835e-05,
197
+ "loss": 0.5317,
198
+ "step": 130
199
+ },
200
+ {
201
+ "epoch": 0.57,
202
+ "grad_norm": 4.219668396076119,
203
+ "learning_rate": 1.9987237809575722e-05,
204
+ "loss": 0.5372,
205
+ "step": 135
206
+ },
207
+ {
208
+ "epoch": 0.6,
209
+ "grad_norm": 5.10590490035955,
210
+ "learning_rate": 1.997862971112581e-05,
211
+ "loss": 0.5236,
212
+ "step": 140
213
+ },
214
+ {
215
+ "epoch": 0.62,
216
+ "grad_norm": 9.22741963642746,
217
+ "learning_rate": 1.9967817911548796e-05,
218
+ "loss": 0.5351,
219
+ "step": 145
220
+ },
221
+ {
222
+ "epoch": 0.64,
223
+ "grad_norm": 3.2116021264594767,
224
+ "learning_rate": 1.9954804798544748e-05,
225
+ "loss": 0.477,
226
+ "step": 150
227
+ },
228
+ {
229
+ "epoch": 0.66,
230
+ "grad_norm": 4.7279336245257255,
231
+ "learning_rate": 1.993959324595634e-05,
232
+ "loss": 0.5063,
233
+ "step": 155
234
+ },
235
+ {
236
+ "epoch": 0.68,
237
+ "grad_norm": 2.6783717640693663,
238
+ "learning_rate": 1.9922186613134152e-05,
239
+ "loss": 0.4547,
240
+ "step": 160
241
+ },
242
+ {
243
+ "epoch": 0.7,
244
+ "grad_norm": 4.6102800388264,
245
+ "learning_rate": 1.9902588744194815e-05,
246
+ "loss": 0.4618,
247
+ "step": 165
248
+ },
249
+ {
250
+ "epoch": 0.72,
251
+ "grad_norm": 2.5223162544544246,
252
+ "learning_rate": 1.9880803967172048e-05,
253
+ "loss": 0.4532,
254
+ "step": 170
255
+ },
256
+ {
257
+ "epoch": 0.74,
258
+ "grad_norm": 2.2521576224935687,
259
+ "learning_rate": 1.985683709306085e-05,
260
+ "loss": 0.4627,
261
+ "step": 175
262
+ },
263
+ {
264
+ "epoch": 0.77,
265
+ "grad_norm": 2.363613216496382,
266
+ "learning_rate": 1.983069341475504e-05,
267
+ "loss": 0.4266,
268
+ "step": 180
269
+ },
270
+ {
271
+ "epoch": 0.79,
272
+ "grad_norm": 2.319077028125871,
273
+ "learning_rate": 1.9802378705878354e-05,
274
+ "loss": 0.4294,
275
+ "step": 185
276
+ },
277
+ {
278
+ "epoch": 0.81,
279
+ "grad_norm": 2.5983523553598675,
280
+ "learning_rate": 1.9771899219509388e-05,
281
+ "loss": 0.4463,
282
+ "step": 190
283
+ },
284
+ {
285
+ "epoch": 0.83,
286
+ "grad_norm": 2.697869053891185,
287
+ "learning_rate": 1.9739261686800662e-05,
288
+ "loss": 0.3995,
289
+ "step": 195
290
+ },
291
+ {
292
+ "epoch": 0.85,
293
+ "grad_norm": 2.2299084670054166,
294
+ "learning_rate": 1.9704473315492072e-05,
295
+ "loss": 0.4145,
296
+ "step": 200
297
+ },
298
+ {
299
+ "epoch": 0.87,
300
+ "grad_norm": 2.144013684497221,
301
+ "learning_rate": 1.966754178831916e-05,
302
+ "loss": 0.4047,
303
+ "step": 205
304
+ },
305
+ {
306
+ "epoch": 0.89,
307
+ "grad_norm": 2.165636444751405,
308
+ "learning_rate": 1.962847526131642e-05,
309
+ "loss": 0.429,
310
+ "step": 210
311
+ },
312
+ {
313
+ "epoch": 0.91,
314
+ "grad_norm": 2.5009764776602066,
315
+ "learning_rate": 1.9587282362016083e-05,
316
+ "loss": 0.3987,
317
+ "step": 215
318
+ },
319
+ {
320
+ "epoch": 0.94,
321
+ "grad_norm": 3.2454133915018946,
322
+ "learning_rate": 1.9543972187542833e-05,
323
+ "loss": 0.4196,
324
+ "step": 220
325
+ },
326
+ {
327
+ "epoch": 0.96,
328
+ "grad_norm": 2.273670978657276,
329
+ "learning_rate": 1.9498554302604768e-05,
330
+ "loss": 0.4029,
331
+ "step": 225
332
+ },
333
+ {
334
+ "epoch": 0.98,
335
+ "grad_norm": 3.004323662088254,
336
+ "learning_rate": 1.9451038737381078e-05,
337
+ "loss": 0.4268,
338
+ "step": 230
339
+ },
340
  {
341
  "epoch": 1.0,
342
+ "grad_norm": 2.3784052825284845,
343
+ "learning_rate": 1.940143598530701e-05,
344
+ "loss": 0.4066,
345
+ "step": 235
 
346
  },
347
  {
348
  "epoch": 1.0,
349
+ "eval_loss": 0.7733179926872253,
350
+ "eval_runtime": 347.9521,
351
+ "eval_samples_per_second": 21.575,
352
+ "eval_steps_per_second": 0.339,
353
+ "step": 235
354
+ },
355
+ {
356
+ "epoch": 1.02,
357
+ "grad_norm": 2.2553533504002683,
358
+ "learning_rate": 1.9349757000756442e-05,
359
+ "loss": 0.2356,
360
+ "step": 240
361
+ },
362
+ {
363
+ "epoch": 1.04,
364
+ "grad_norm": 2.5450743747403717,
365
+ "learning_rate": 1.929601319662271e-05,
366
+ "loss": 0.2119,
367
+ "step": 245
368
+ },
369
+ {
370
+ "epoch": 1.06,
371
+ "grad_norm": 3.778262007789548,
372
+ "learning_rate": 1.9240216441798145e-05,
373
+ "loss": 0.2295,
374
+ "step": 250
375
+ },
376
+ {
377
+ "epoch": 1.09,
378
+ "grad_norm": 2.2858585923056536,
379
+ "learning_rate": 1.918237905855295e-05,
380
+ "loss": 0.2251,
381
+ "step": 255
382
+ },
383
+ {
384
+ "epoch": 1.11,
385
+ "grad_norm": 1.7784488699701833,
386
+ "learning_rate": 1.91225138198139e-05,
387
+ "loss": 0.2145,
388
+ "step": 260
389
+ },
390
+ {
391
+ "epoch": 1.13,
392
+ "grad_norm": 1.8254402227157958,
393
+ "learning_rate": 1.906063394634356e-05,
394
+ "loss": 0.2191,
395
+ "step": 265
396
+ },
397
+ {
398
+ "epoch": 1.15,
399
+ "grad_norm": 1.9464300006155193,
400
+ "learning_rate": 1.899675310382057e-05,
401
+ "loss": 0.2273,
402
+ "step": 270
403
+ },
404
+ {
405
+ "epoch": 1.17,
406
+ "grad_norm": 2.2936148112741757,
407
+ "learning_rate": 1.8930885399821693e-05,
408
+ "loss": 0.2246,
409
+ "step": 275
410
+ },
411
+ {
412
+ "epoch": 1.19,
413
+ "grad_norm": 2.3799646925913245,
414
+ "learning_rate": 1.8863045380706275e-05,
415
+ "loss": 0.2303,
416
+ "step": 280
417
+ },
418
+ {
419
+ "epoch": 1.21,
420
+ "grad_norm": 2.190554922996115,
421
+ "learning_rate": 1.879324802840379e-05,
422
+ "loss": 0.2318,
423
+ "step": 285
424
+ },
425
+ {
426
+ "epoch": 1.23,
427
+ "grad_norm": 2.7452325011160035,
428
+ "learning_rate": 1.8721508757105203e-05,
429
+ "loss": 0.2276,
430
+ "step": 290
431
+ },
432
+ {
433
+ "epoch": 1.26,
434
+ "grad_norm": 1.8163744421498897,
435
+ "learning_rate": 1.864784340985887e-05,
436
+ "loss": 0.24,
437
+ "step": 295
438
+ },
439
+ {
440
+ "epoch": 1.28,
441
+ "grad_norm": 1.885101115960248,
442
+ "learning_rate": 1.8572268255071718e-05,
443
+ "loss": 0.2399,
444
+ "step": 300
445
+ },
446
+ {
447
+ "epoch": 1.3,
448
+ "grad_norm": 1.760073718971802,
449
+ "learning_rate": 1.8494799982916512e-05,
450
+ "loss": 0.1955,
451
+ "step": 305
452
+ },
453
+ {
454
+ "epoch": 1.32,
455
+ "grad_norm": 1.8829516569820812,
456
+ "learning_rate": 1.8415455701645942e-05,
457
+ "loss": 0.2035,
458
+ "step": 310
459
+ },
460
+ {
461
+ "epoch": 1.34,
462
+ "grad_norm": 2.146880306353409,
463
+ "learning_rate": 1.833425293381443e-05,
464
+ "loss": 0.2271,
465
+ "step": 315
466
+ },
467
+ {
468
+ "epoch": 1.36,
469
+ "grad_norm": 2.0534275082799684,
470
+ "learning_rate": 1.8251209612408375e-05,
471
+ "loss": 0.197,
472
+ "step": 320
473
+ },
474
+ {
475
+ "epoch": 1.38,
476
+ "grad_norm": 2.0857973185832743,
477
+ "learning_rate": 1.816634407688583e-05,
478
+ "loss": 0.2337,
479
+ "step": 325
480
+ },
481
+ {
482
+ "epoch": 1.4,
483
+ "grad_norm": 1.7319986338118718,
484
+ "learning_rate": 1.807967506912636e-05,
485
+ "loss": 0.1909,
486
+ "step": 330
487
+ },
488
+ {
489
+ "epoch": 1.43,
490
+ "grad_norm": 1.7148628603629519,
491
+ "learning_rate": 1.799122172929206e-05,
492
+ "loss": 0.2103,
493
+ "step": 335
494
+ },
495
+ {
496
+ "epoch": 1.45,
497
+ "grad_norm": 2.0404565885174786,
498
+ "learning_rate": 1.7901003591600575e-05,
499
+ "loss": 0.2045,
500
+ "step": 340
501
+ },
502
+ {
503
+ "epoch": 1.47,
504
+ "grad_norm": 2.132532760084515,
505
+ "learning_rate": 1.780904058001116e-05,
506
+ "loss": 0.2234,
507
+ "step": 345
508
+ },
509
+ {
510
+ "epoch": 1.49,
511
+ "grad_norm": 2.2388854318437774,
512
+ "learning_rate": 1.7715353003824613e-05,
513
+ "loss": 0.222,
514
+ "step": 350
515
+ },
516
+ {
517
+ "epoch": 1.51,
518
+ "grad_norm": 2.1625096071245986,
519
+ "learning_rate": 1.761996155319811e-05,
520
+ "loss": 0.2278,
521
+ "step": 355
522
+ },
523
+ {
524
+ "epoch": 1.53,
525
+ "grad_norm": 1.8994643849098976,
526
+ "learning_rate": 1.7522887294575978e-05,
527
+ "loss": 0.2114,
528
+ "step": 360
529
+ },
530
+ {
531
+ "epoch": 1.55,
532
+ "grad_norm": 2.0232265086023213,
533
+ "learning_rate": 1.742415166603733e-05,
534
+ "loss": 0.2276,
535
+ "step": 365
536
+ },
537
+ {
538
+ "epoch": 1.57,
539
+ "grad_norm": 2.0393231755488737,
540
+ "learning_rate": 1.7323776472561625e-05,
541
+ "loss": 0.2065,
542
+ "step": 370
543
+ },
544
+ {
545
+ "epoch": 1.6,
546
+ "grad_norm": 1.7156091041056394,
547
+ "learning_rate": 1.7221783881213222e-05,
548
+ "loss": 0.2201,
549
+ "step": 375
550
+ },
551
+ {
552
+ "epoch": 1.62,
553
+ "grad_norm": 2.0249954720249645,
554
+ "learning_rate": 1.7118196416245947e-05,
555
+ "loss": 0.1986,
556
+ "step": 380
557
+ },
558
+ {
559
+ "epoch": 1.64,
560
+ "grad_norm": 1.860444641374077,
561
+ "learning_rate": 1.701303695412881e-05,
562
+ "loss": 0.2224,
563
+ "step": 385
564
+ },
565
+ {
566
+ "epoch": 1.66,
567
+ "grad_norm": 9.195021611366267,
568
+ "learning_rate": 1.6906328718493906e-05,
569
+ "loss": 0.2152,
570
+ "step": 390
571
+ },
572
+ {
573
+ "epoch": 1.68,
574
+ "grad_norm": 2.790540417351318,
575
+ "learning_rate": 1.679809527500765e-05,
576
+ "loss": 0.2014,
577
+ "step": 395
578
+ },
579
+ {
580
+ "epoch": 1.7,
581
+ "grad_norm": 1.9401261875054212,
582
+ "learning_rate": 1.6688360526166514e-05,
583
+ "loss": 0.2305,
584
+ "step": 400
585
+ },
586
+ {
587
+ "epoch": 1.72,
588
+ "grad_norm": 2.3499169909762903,
589
+ "learning_rate": 1.657714870601833e-05,
590
+ "loss": 0.2163,
591
+ "step": 405
592
+ },
593
+ {
594
+ "epoch": 1.74,
595
+ "grad_norm": 1.889841729135842,
596
+ "learning_rate": 1.646448437481039e-05,
597
+ "loss": 0.2285,
598
+ "step": 410
599
+ },
600
+ {
601
+ "epoch": 1.77,
602
+ "grad_norm": 1.9100327330280613,
603
+ "learning_rate": 1.635039241356553e-05,
604
+ "loss": 0.2066,
605
+ "step": 415
606
+ },
607
+ {
608
+ "epoch": 1.79,
609
+ "grad_norm": 2.0395782940541913,
610
+ "learning_rate": 1.6234898018587336e-05,
611
+ "loss": 0.2159,
612
+ "step": 420
613
+ },
614
+ {
615
+ "epoch": 1.81,
616
+ "grad_norm": 2.0672756907905647,
617
+ "learning_rate": 1.611802669589575e-05,
618
+ "loss": 0.2211,
619
+ "step": 425
620
+ },
621
+ {
622
+ "epoch": 1.83,
623
+ "grad_norm": 1.8644628412570703,
624
+ "learning_rate": 1.5999804255594262e-05,
625
+ "loss": 0.2123,
626
+ "step": 430
627
+ },
628
+ {
629
+ "epoch": 1.85,
630
+ "grad_norm": 2.109246262952444,
631
+ "learning_rate": 1.5880256806169954e-05,
632
+ "loss": 0.2179,
633
+ "step": 435
634
+ },
635
+ {
636
+ "epoch": 1.87,
637
+ "grad_norm": 2.0987747021618537,
638
+ "learning_rate": 1.5759410748727663e-05,
639
+ "loss": 0.2165,
640
+ "step": 440
641
+ },
642
+ {
643
+ "epoch": 1.89,
644
+ "grad_norm": 1.8615315847750686,
645
+ "learning_rate": 1.563729277115947e-05,
646
+ "loss": 0.2115,
647
+ "step": 445
648
+ },
649
+ {
650
+ "epoch": 1.91,
651
+ "grad_norm": 1.91470907138453,
652
+ "learning_rate": 1.551392984225094e-05,
653
+ "loss": 0.2226,
654
+ "step": 450
655
+ },
656
+ {
657
+ "epoch": 1.94,
658
+ "grad_norm": 2.1673896851739,
659
+ "learning_rate": 1.5389349205725244e-05,
660
+ "loss": 0.2034,
661
+ "step": 455
662
+ },
663
+ {
664
+ "epoch": 1.96,
665
+ "grad_norm": 1.9561220925261051,
666
+ "learning_rate": 1.5263578374226607e-05,
667
+ "loss": 0.2079,
668
+ "step": 460
669
+ },
670
+ {
671
+ "epoch": 1.98,
672
+ "grad_norm": 1.8490634437024727,
673
+ "learning_rate": 1.5136645123244366e-05,
674
+ "loss": 0.2011,
675
+ "step": 465
676
+ },
677
+ {
678
+ "epoch": 2.0,
679
+ "grad_norm": 1.8370535270484316,
680
+ "learning_rate": 1.5008577484978966e-05,
681
+ "loss": 0.1901,
682
+ "step": 470
683
+ },
684
+ {
685
+ "epoch": 2.0,
686
+ "eval_loss": 0.6736623048782349,
687
+ "eval_runtime": 345.8336,
688
+ "eval_samples_per_second": 21.707,
689
+ "eval_steps_per_second": 0.341,
690
+ "step": 470
691
+ },
692
+ {
693
+ "epoch": 2.02,
694
+ "grad_norm": 3.2591869308084505,
695
+ "learning_rate": 1.4879403742151283e-05,
696
+ "loss": 0.0766,
697
+ "step": 475
698
+ },
699
+ {
700
+ "epoch": 2.04,
701
+ "grad_norm": 1.5781740604686478,
702
+ "learning_rate": 1.4749152421756596e-05,
703
+ "loss": 0.0809,
704
+ "step": 480
705
+ },
706
+ {
707
+ "epoch": 2.06,
708
+ "grad_norm": 1.6789209629444652,
709
+ "learning_rate": 1.4617852288764624e-05,
710
+ "loss": 0.0728,
711
+ "step": 485
712
+ },
713
+ {
714
+ "epoch": 2.09,
715
+ "grad_norm": 1.524275026933269,
716
+ "learning_rate": 1.4485532339767036e-05,
717
+ "loss": 0.0671,
718
+ "step": 490
719
+ },
720
+ {
721
+ "epoch": 2.11,
722
+ "grad_norm": 1.911681176460919,
723
+ "learning_rate": 1.4352221796573758e-05,
724
+ "loss": 0.0829,
725
+ "step": 495
726
+ },
727
+ {
728
+ "epoch": 2.13,
729
+ "grad_norm": 1.5314000006841326,
730
+ "learning_rate": 1.421795009975957e-05,
731
+ "loss": 0.0739,
732
+ "step": 500
733
+ },
734
+ {
735
+ "epoch": 2.15,
736
+ "grad_norm": 1.8497162969927008,
737
+ "learning_rate": 1.4082746902162414e-05,
738
+ "loss": 0.0782,
739
+ "step": 505
740
+ },
741
+ {
742
+ "epoch": 2.17,
743
+ "grad_norm": 1.5153675955770316,
744
+ "learning_rate": 1.3946642062334765e-05,
745
+ "loss": 0.0717,
746
+ "step": 510
747
+ },
748
+ {
749
+ "epoch": 2.19,
750
+ "grad_norm": 1.4798583086240507,
751
+ "learning_rate": 1.3809665637949636e-05,
752
+ "loss": 0.0683,
753
+ "step": 515
754
+ },
755
+ {
756
+ "epoch": 2.21,
757
+ "grad_norm": 1.5185665743912256,
758
+ "learning_rate": 1.3671847879162562e-05,
759
+ "loss": 0.0647,
760
+ "step": 520
761
+ },
762
+ {
763
+ "epoch": 2.23,
764
+ "grad_norm": 1.6655269792696328,
765
+ "learning_rate": 1.3533219221931102e-05,
766
+ "loss": 0.0756,
767
+ "step": 525
768
+ },
769
+ {
770
+ "epoch": 2.26,
771
+ "grad_norm": 1.4623074396711568,
772
+ "learning_rate": 1.3393810281293294e-05,
773
+ "loss": 0.079,
774
+ "step": 530
775
+ },
776
+ {
777
+ "epoch": 2.28,
778
+ "grad_norm": 1.4769143347541074,
779
+ "learning_rate": 1.3253651844606571e-05,
780
+ "loss": 0.0644,
781
+ "step": 535
782
+ },
783
+ {
784
+ "epoch": 2.3,
785
+ "grad_norm": 1.5924771297209446,
786
+ "learning_rate": 1.311277486474862e-05,
787
+ "loss": 0.0676,
788
+ "step": 540
789
+ },
790
+ {
791
+ "epoch": 2.32,
792
+ "grad_norm": 1.4963368374266843,
793
+ "learning_rate": 1.2971210453281675e-05,
794
+ "loss": 0.0685,
795
+ "step": 545
796
+ },
797
+ {
798
+ "epoch": 2.34,
799
+ "grad_norm": 1.3660066623187117,
800
+ "learning_rate": 1.2828989873581786e-05,
801
+ "loss": 0.0643,
802
+ "step": 550
803
+ },
804
+ {
805
+ "epoch": 2.36,
806
+ "grad_norm": 1.5417608458745726,
807
+ "learning_rate": 1.268614453393454e-05,
808
+ "loss": 0.0704,
809
+ "step": 555
810
+ },
811
+ {
812
+ "epoch": 2.38,
813
+ "grad_norm": 1.3386793668448367,
814
+ "learning_rate": 1.2542705980598813e-05,
815
+ "loss": 0.0725,
816
+ "step": 560
817
+ },
818
+ {
819
+ "epoch": 2.4,
820
+ "grad_norm": 4.538383693449207,
821
+ "learning_rate": 1.2398705890839988e-05,
822
+ "loss": 0.0724,
823
+ "step": 565
824
+ },
825
+ {
826
+ "epoch": 2.43,
827
+ "grad_norm": 1.5161704188772478,
828
+ "learning_rate": 1.2254176065934332e-05,
829
+ "loss": 0.069,
830
+ "step": 570
831
+ },
832
+ {
833
+ "epoch": 2.45,
834
+ "grad_norm": 1.5190424960322728,
835
+ "learning_rate": 1.2109148424145897e-05,
836
+ "loss": 0.0675,
837
+ "step": 575
838
+ },
839
+ {
840
+ "epoch": 2.47,
841
+ "grad_norm": 1.6418241298329592,
842
+ "learning_rate": 1.1963654993677645e-05,
843
+ "loss": 0.0706,
844
+ "step": 580
845
+ },
846
+ {
847
+ "epoch": 2.49,
848
+ "grad_norm": 2.7746288115630415,
849
+ "learning_rate": 1.1817727905598268e-05,
850
+ "loss": 0.0661,
851
+ "step": 585
852
+ },
853
+ {
854
+ "epoch": 2.51,
855
+ "grad_norm": 1.613430655848529,
856
+ "learning_rate": 1.1671399386746301e-05,
857
+ "loss": 0.0685,
858
+ "step": 590
859
+ },
860
+ {
861
+ "epoch": 2.53,
862
+ "grad_norm": 1.652377714904139,
863
+ "learning_rate": 1.1524701752613074e-05,
864
+ "loss": 0.0685,
865
+ "step": 595
866
+ },
867
+ {
868
+ "epoch": 2.55,
869
+ "grad_norm": 1.6108396859937457,
870
+ "learning_rate": 1.13776674002061e-05,
871
+ "loss": 0.0736,
872
+ "step": 600
873
+ },
874
+ {
875
+ "epoch": 2.57,
876
+ "grad_norm": 1.4308175997531103,
877
+ "learning_rate": 1.1230328800894437e-05,
878
+ "loss": 0.0699,
879
+ "step": 605
880
+ },
881
+ {
882
+ "epoch": 2.6,
883
+ "grad_norm": 1.529984546400154,
884
+ "learning_rate": 1.108271849323767e-05,
885
+ "loss": 0.0647,
886
+ "step": 610
887
+ },
888
+ {
889
+ "epoch": 2.62,
890
+ "grad_norm": 1.7364294117046162,
891
+ "learning_rate": 1.09348690758e-05,
892
+ "loss": 0.0691,
893
+ "step": 615
894
+ },
895
+ {
896
+ "epoch": 2.64,
897
+ "grad_norm": 1.4526326385257828,
898
+ "learning_rate": 1.0786813199951145e-05,
899
+ "loss": 0.0623,
900
+ "step": 620
901
+ },
902
+ {
903
+ "epoch": 2.66,
904
+ "grad_norm": 1.5442010646463833,
905
+ "learning_rate": 1.0638583562655498e-05,
906
+ "loss": 0.0673,
907
+ "step": 625
908
+ },
909
+ {
910
+ "epoch": 2.68,
911
+ "grad_norm": 1.4797041462885991,
912
+ "learning_rate": 1.0490212899251308e-05,
913
+ "loss": 0.0686,
914
+ "step": 630
915
+ },
916
+ {
917
+ "epoch": 2.7,
918
+ "grad_norm": 1.650135612201066,
919
+ "learning_rate": 1.0341733976221313e-05,
920
+ "loss": 0.0715,
921
+ "step": 635
922
+ },
923
+ {
924
+ "epoch": 2.72,
925
+ "grad_norm": 1.5265081306931187,
926
+ "learning_rate": 1.0193179583956523e-05,
927
+ "loss": 0.0642,
928
+ "step": 640
929
+ },
930
+ {
931
+ "epoch": 2.74,
932
+ "grad_norm": 1.4887311962873813,
933
+ "learning_rate": 1.0044582529514739e-05,
934
+ "loss": 0.075,
935
+ "step": 645
936
+ },
937
+ {
938
+ "epoch": 2.77,
939
+ "grad_norm": 1.4387050627018103,
940
+ "learning_rate": 9.89597562937536e-06,
941
+ "loss": 0.06,
942
+ "step": 650
943
+ },
944
+ {
945
+ "epoch": 2.79,
946
+ "grad_norm": 1.58228556181931,
947
+ "learning_rate": 9.747391702192132e-06,
948
+ "loss": 0.0613,
949
+ "step": 655
950
+ },
951
+ {
952
+ "epoch": 2.81,
953
+ "grad_norm": 1.424709805936743,
954
+ "learning_rate": 9.598863561545404e-06,
955
+ "loss": 0.0728,
956
+ "step": 660
957
+ },
958
+ {
959
+ "epoch": 2.83,
960
+ "grad_norm": 1.3925788280661726,
961
+ "learning_rate": 9.45042400869551e-06,
962
+ "loss": 0.0638,
963
+ "step": 665
964
+ },
965
+ {
966
+ "epoch": 2.85,
967
+ "grad_norm": 1.8579081866696643,
968
+ "learning_rate": 9.302105825338876e-06,
969
+ "loss": 0.0729,
970
+ "step": 670
971
+ },
972
+ {
973
+ "epoch": 2.87,
974
+ "grad_norm": 1.4419650630198477,
975
+ "learning_rate": 9.153941766368439e-06,
976
+ "loss": 0.0605,
977
+ "step": 675
978
+ },
979
+ {
980
+ "epoch": 2.89,
981
+ "grad_norm": 1.4587070966850302,
982
+ "learning_rate": 9.005964552639983e-06,
983
+ "loss": 0.0637,
984
+ "step": 680
985
+ },
986
+ {
987
+ "epoch": 2.91,
988
+ "grad_norm": 1.400937857775751,
989
+ "learning_rate": 8.858206863746018e-06,
990
+ "loss": 0.0644,
991
+ "step": 685
992
+ },
993
+ {
994
+ "epoch": 2.94,
995
+ "grad_norm": 1.4194625294878935,
996
+ "learning_rate": 8.71070133079872e-06,
997
+ "loss": 0.0708,
998
+ "step": 690
999
+ },
1000
+ {
1001
+ "epoch": 2.96,
1002
+ "grad_norm": 1.511069294541966,
1003
+ "learning_rate": 8.563480529223638e-06,
1004
+ "loss": 0.0667,
1005
+ "step": 695
1006
+ },
1007
+ {
1008
+ "epoch": 2.98,
1009
+ "grad_norm": 1.3579712613828467,
1010
+ "learning_rate": 8.41657697156566e-06,
1011
+ "loss": 0.0527,
1012
+ "step": 700
1013
+ },
1014
+ {
1015
+ "epoch": 3.0,
1016
+ "grad_norm": 1.4627979346880684,
1017
+ "learning_rate": 8.270023100308865e-06,
1018
+ "loss": 0.068,
1019
+ "step": 705
1020
+ },
1021
+ {
1022
+ "epoch": 3.0,
1023
+ "eval_loss": 0.6575748324394226,
1024
+ "eval_runtime": 346.2698,
1025
+ "eval_samples_per_second": 21.68,
1026
+ "eval_steps_per_second": 0.341,
1027
+ "step": 705
1028
+ },
1029
+ {
1030
+ "epoch": 3.02,
1031
+ "grad_norm": 0.9187425982230547,
1032
+ "learning_rate": 8.123851280711877e-06,
1033
+ "loss": 0.0216,
1034
+ "step": 710
1035
+ },
1036
+ {
1037
+ "epoch": 3.04,
1038
+ "grad_norm": 1.0704321956753566,
1039
+ "learning_rate": 7.978093793660234e-06,
1040
+ "loss": 0.0187,
1041
+ "step": 715
1042
+ },
1043
+ {
1044
+ "epoch": 3.06,
1045
+ "grad_norm": 0.9253831765543917,
1046
+ "learning_rate": 7.832782828537437e-06,
1047
+ "loss": 0.0191,
1048
+ "step": 720
1049
+ },
1050
+ {
1051
+ "epoch": 3.09,
1052
+ "grad_norm": 0.9546574193581036,
1053
+ "learning_rate": 7.68795047611615e-06,
1054
+ "loss": 0.0186,
1055
+ "step": 725
1056
+ },
1057
+ {
1058
+ "epoch": 3.11,
1059
+ "grad_norm": 0.9527598588762864,
1060
+ "learning_rate": 7.543628721471234e-06,
1061
+ "loss": 0.0191,
1062
+ "step": 730
1063
+ },
1064
+ {
1065
+ "epoch": 3.13,
1066
+ "grad_norm": 1.0142369552642194,
1067
+ "learning_rate": 7.399849436916076e-06,
1068
+ "loss": 0.017,
1069
+ "step": 735
1070
+ },
1071
+ {
1072
+ "epoch": 3.15,
1073
+ "grad_norm": 0.8516277331105468,
1074
+ "learning_rate": 7.256644374963857e-06,
1075
+ "loss": 0.0169,
1076
+ "step": 740
1077
+ },
1078
+ {
1079
+ "epoch": 3.17,
1080
+ "grad_norm": 0.6998532587968802,
1081
+ "learning_rate": 7.11404516131526e-06,
1082
+ "loss": 0.0167,
1083
+ "step": 745
1084
+ },
1085
+ {
1086
+ "epoch": 3.19,
1087
+ "grad_norm": 0.9597297439764481,
1088
+ "learning_rate": 6.9720832878741776e-06,
1089
+ "loss": 0.0195,
1090
+ "step": 750
1091
+ },
1092
+ {
1093
+ "epoch": 3.21,
1094
+ "grad_norm": 0.8162423625488969,
1095
+ "learning_rate": 6.8307901057929735e-06,
1096
+ "loss": 0.0196,
1097
+ "step": 755
1098
+ },
1099
+ {
1100
+ "epoch": 3.23,
1101
+ "grad_norm": 1.0809022540934834,
1102
+ "learning_rate": 6.690196818548846e-06,
1103
+ "loss": 0.0202,
1104
+ "step": 760
1105
+ },
1106
+ {
1107
+ "epoch": 3.26,
1108
+ "grad_norm": 0.9451137823905494,
1109
+ "learning_rate": 6.550334475052767e-06,
1110
+ "loss": 0.0173,
1111
+ "step": 765
1112
+ },
1113
+ {
1114
+ "epoch": 3.28,
1115
+ "grad_norm": 1.0194236278105107,
1116
+ "learning_rate": 6.411233962792593e-06,
1117
+ "loss": 0.0217,
1118
+ "step": 770
1119
+ },
1120
+ {
1121
+ "epoch": 3.3,
1122
+ "grad_norm": 0.9428518118269854,
1123
+ "learning_rate": 6.2729260010117995e-06,
1124
+ "loss": 0.0205,
1125
+ "step": 775
1126
+ },
1127
+ {
1128
+ "epoch": 3.32,
1129
+ "grad_norm": 0.8298491758909723,
1130
+ "learning_rate": 6.135441133925382e-06,
1131
+ "loss": 0.0156,
1132
+ "step": 780
1133
+ },
1134
+ {
1135
+ "epoch": 3.34,
1136
+ "grad_norm": 0.9988001166823972,
1137
+ "learning_rate": 5.998809723974407e-06,
1138
+ "loss": 0.018,
1139
+ "step": 785
1140
+ },
1141
+ {
1142
+ "epoch": 3.36,
1143
+ "grad_norm": 0.8874287620178792,
1144
+ "learning_rate": 5.863061945120719e-06,
1145
+ "loss": 0.0172,
1146
+ "step": 790
1147
+ },
1148
+ {
1149
+ "epoch": 3.38,
1150
+ "grad_norm": 0.9021252322116792,
1151
+ "learning_rate": 5.728227776183244e-06,
1152
+ "loss": 0.0184,
1153
+ "step": 795
1154
+ },
1155
+ {
1156
+ "epoch": 3.4,
1157
+ "grad_norm": 0.899901563579598,
1158
+ "learning_rate": 5.594336994217416e-06,
1159
+ "loss": 0.0207,
1160
+ "step": 800
1161
+ },
1162
+ {
1163
+ "epoch": 3.43,
1164
+ "grad_norm": 0.7294818546149708,
1165
+ "learning_rate": 5.461419167939145e-06,
1166
+ "loss": 0.0154,
1167
+ "step": 805
1168
+ },
1169
+ {
1170
+ "epoch": 3.45,
1171
+ "grad_norm": 0.9595599651370788,
1172
+ "learning_rate": 5.329503651194805e-06,
1173
+ "loss": 0.0182,
1174
+ "step": 810
1175
+ },
1176
+ {
1177
+ "epoch": 3.47,
1178
+ "grad_norm": 0.7263044743513145,
1179
+ "learning_rate": 5.198619576478678e-06,
1180
+ "loss": 0.017,
1181
+ "step": 815
1182
+ },
1183
+ {
1184
+ "epoch": 3.49,
1185
+ "grad_norm": 0.9300662832706075,
1186
+ "learning_rate": 5.068795848499257e-06,
1187
+ "loss": 0.0182,
1188
+ "step": 820
1189
+ },
1190
+ {
1191
+ "epoch": 3.51,
1192
+ "grad_norm": 1.044597812236674,
1193
+ "learning_rate": 4.940061137795876e-06,
1194
+ "loss": 0.0154,
1195
+ "step": 825
1196
+ },
1197
+ {
1198
+ "epoch": 3.53,
1199
+ "grad_norm": 0.9405246884194852,
1200
+ "learning_rate": 4.812443874407059e-06,
1201
+ "loss": 0.0187,
1202
+ "step": 830
1203
+ },
1204
+ {
1205
+ "epoch": 3.55,
1206
+ "grad_norm": 0.8135204435661318,
1207
+ "learning_rate": 4.685972241591956e-06,
1208
+ "loss": 0.0156,
1209
+ "step": 835
1210
+ },
1211
+ {
1212
+ "epoch": 3.57,
1213
+ "grad_norm": 0.806855367473174,
1214
+ "learning_rate": 4.560674169606317e-06,
1215
+ "loss": 0.0165,
1216
+ "step": 840
1217
+ },
1218
+ {
1219
+ "epoch": 3.6,
1220
+ "grad_norm": 1.1283393882651565,
1221
+ "learning_rate": 4.436577329534291e-06,
1222
+ "loss": 0.0161,
1223
+ "step": 845
1224
+ },
1225
+ {
1226
+ "epoch": 3.62,
1227
+ "grad_norm": 1.0316895831074073,
1228
+ "learning_rate": 4.3137091271775e-06,
1229
+ "loss": 0.018,
1230
+ "step": 850
1231
+ },
1232
+ {
1233
+ "epoch": 3.64,
1234
+ "grad_norm": 0.83713135691551,
1235
+ "learning_rate": 4.192096697002686e-06,
1236
+ "loss": 0.0154,
1237
+ "step": 855
1238
+ },
1239
+ {
1240
+ "epoch": 3.66,
1241
+ "grad_norm": 0.8862197834996656,
1242
+ "learning_rate": 4.0717668961492725e-06,
1243
+ "loss": 0.014,
1244
+ "step": 860
1245
+ },
1246
+ {
1247
+ "epoch": 3.68,
1248
+ "grad_norm": 0.9409064835179552,
1249
+ "learning_rate": 3.9527462984981954e-06,
1250
+ "loss": 0.0185,
1251
+ "step": 865
1252
+ },
1253
+ {
1254
+ "epoch": 3.7,
1255
+ "grad_norm": 0.8141414364863905,
1256
+ "learning_rate": 3.8350611888032474e-06,
1257
+ "loss": 0.0167,
1258
+ "step": 870
1259
+ },
1260
+ {
1261
+ "epoch": 3.72,
1262
+ "grad_norm": 0.9491167080432446,
1263
+ "learning_rate": 3.718737556886316e-06,
1264
+ "loss": 0.0166,
1265
+ "step": 875
1266
+ },
1267
+ {
1268
+ "epoch": 3.74,
1269
+ "grad_norm": 0.8939541066316343,
1270
+ "learning_rate": 3.6038010918977308e-06,
1271
+ "loss": 0.0157,
1272
+ "step": 880
1273
+ },
1274
+ {
1275
+ "epoch": 3.77,
1276
+ "grad_norm": 1.1762213442708835,
1277
+ "learning_rate": 3.490277176643033e-06,
1278
+ "loss": 0.0176,
1279
+ "step": 885
1280
+ },
1281
+ {
1282
+ "epoch": 3.79,
1283
+ "grad_norm": 0.8409814896300148,
1284
+ "learning_rate": 3.378190881977359e-06,
1285
+ "loss": 0.0152,
1286
+ "step": 890
1287
+ },
1288
+ {
1289
+ "epoch": 3.81,
1290
+ "grad_norm": 1.0408829490759663,
1291
+ "learning_rate": 3.2675669612687565e-06,
1292
+ "loss": 0.0191,
1293
+ "step": 895
1294
+ },
1295
+ {
1296
+ "epoch": 3.83,
1297
+ "grad_norm": 0.9464969671060768,
1298
+ "learning_rate": 3.158429844931611e-06,
1299
+ "loss": 0.0156,
1300
+ "step": 900
1301
+ },
1302
+ {
1303
+ "epoch": 3.85,
1304
+ "grad_norm": 0.8183518112165244,
1305
+ "learning_rate": 3.0508036350313553e-06,
1306
+ "loss": 0.0155,
1307
+ "step": 905
1308
+ },
1309
+ {
1310
+ "epoch": 3.87,
1311
+ "grad_norm": 1.031893449336409,
1312
+ "learning_rate": 2.9447120999617363e-06,
1313
+ "loss": 0.0151,
1314
+ "step": 910
1315
+ },
1316
+ {
1317
+ "epoch": 3.89,
1318
+ "grad_norm": 0.9770328880326357,
1319
+ "learning_rate": 2.8401786691957632e-06,
1320
+ "loss": 0.0156,
1321
+ "step": 915
1322
+ },
1323
+ {
1324
+ "epoch": 3.91,
1325
+ "grad_norm": 0.9350326565557209,
1326
+ "learning_rate": 2.737226428111471e-06,
1327
+ "loss": 0.0176,
1328
+ "step": 920
1329
+ },
1330
+ {
1331
+ "epoch": 3.94,
1332
+ "grad_norm": 0.7498411603912395,
1333
+ "learning_rate": 2.635878112893717e-06,
1334
+ "loss": 0.0147,
1335
+ "step": 925
1336
+ },
1337
+ {
1338
+ "epoch": 3.96,
1339
+ "grad_norm": 0.890762978868864,
1340
+ "learning_rate": 2.5361561055130625e-06,
1341
+ "loss": 0.0179,
1342
+ "step": 930
1343
+ },
1344
+ {
1345
+ "epoch": 3.98,
1346
+ "grad_norm": 1.153428874102931,
1347
+ "learning_rate": 2.4380824287829073e-06,
1348
+ "loss": 0.0159,
1349
+ "step": 935
1350
+ },
1351
+ {
1352
+ "epoch": 4.0,
1353
+ "grad_norm": 0.89001029327834,
1354
+ "learning_rate": 2.3416787414959097e-06,
1355
+ "loss": 0.0183,
1356
+ "step": 940
1357
+ },
1358
+ {
1359
+ "epoch": 4.0,
1360
+ "eval_loss": 0.7071638107299805,
1361
+ "eval_runtime": 348.8132,
1362
+ "eval_samples_per_second": 21.522,
1363
+ "eval_steps_per_second": 0.338,
1364
+ "step": 940
1365
+ },
1366
+ {
1367
+ "epoch": 4.02,
1368
+ "grad_norm": 0.3600823717570703,
1369
+ "learning_rate": 2.246966333640823e-06,
1370
+ "loss": 0.0048,
1371
+ "step": 945
1372
+ },
1373
+ {
1374
+ "epoch": 4.04,
1375
+ "grad_norm": 0.33280820804988087,
1376
+ "learning_rate": 2.153966121700769e-06,
1377
+ "loss": 0.0049,
1378
+ "step": 950
1379
+ },
1380
+ {
1381
+ "epoch": 4.06,
1382
+ "grad_norm": 0.48812206732400915,
1383
+ "learning_rate": 2.0626986440340036e-06,
1384
+ "loss": 0.0058,
1385
+ "step": 955
1386
+ },
1387
+ {
1388
+ "epoch": 4.09,
1389
+ "grad_norm": 0.3662559033659471,
1390
+ "learning_rate": 1.973184056338173e-06,
1391
+ "loss": 0.0039,
1392
+ "step": 960
1393
+ },
1394
+ {
1395
+ "epoch": 4.11,
1396
+ "grad_norm": 0.3710554752839552,
1397
+ "learning_rate": 1.8854421271990964e-06,
1398
+ "loss": 0.0037,
1399
+ "step": 965
1400
+ },
1401
+ {
1402
+ "epoch": 4.13,
1403
+ "grad_norm": 0.3360503768539956,
1404
+ "learning_rate": 1.7994922337250276e-06,
1405
+ "loss": 0.0041,
1406
+ "step": 970
1407
+ },
1408
+ {
1409
+ "epoch": 4.15,
1410
+ "grad_norm": 0.32232522758989207,
1411
+ "learning_rate": 1.7153533572673708e-06,
1412
+ "loss": 0.0042,
1413
+ "step": 975
1414
+ },
1415
+ {
1416
+ "epoch": 4.17,
1417
+ "grad_norm": 0.3933577657073139,
1418
+ "learning_rate": 1.633044079228817e-06,
1419
+ "loss": 0.0042,
1420
+ "step": 980
1421
+ },
1422
+ {
1423
+ "epoch": 4.19,
1424
+ "grad_norm": 0.4133692185032901,
1425
+ "learning_rate": 1.5525825769597625e-06,
1426
+ "loss": 0.0042,
1427
+ "step": 985
1428
+ },
1429
+ {
1430
+ "epoch": 4.21,
1431
+ "grad_norm": 0.311775019674446,
1432
+ "learning_rate": 1.4739866197440046e-06,
1433
+ "loss": 0.004,
1434
+ "step": 990
1435
+ },
1436
+ {
1437
+ "epoch": 4.23,
1438
+ "grad_norm": 0.49759561364679256,
1439
+ "learning_rate": 1.3972735648745295e-06,
1440
+ "loss": 0.0049,
1441
+ "step": 995
1442
+ },
1443
+ {
1444
+ "epoch": 4.26,
1445
+ "grad_norm": 0.4477509727419261,
1446
+ "learning_rate": 1.3224603538202929e-06,
1447
+ "loss": 0.0041,
1448
+ "step": 1000
1449
+ },
1450
+ {
1451
+ "epoch": 4.28,
1452
+ "grad_norm": 0.4461256466898462,
1453
+ "learning_rate": 1.2495635084848356e-06,
1454
+ "loss": 0.0038,
1455
+ "step": 1005
1456
+ },
1457
+ {
1458
+ "epoch": 4.3,
1459
+ "grad_norm": 0.40708673174578486,
1460
+ "learning_rate": 1.1785991275575426e-06,
1461
+ "loss": 0.0039,
1462
+ "step": 1010
1463
+ },
1464
+ {
1465
+ "epoch": 4.32,
1466
+ "grad_norm": 0.4494486001323653,
1467
+ "learning_rate": 1.1095828829583844e-06,
1468
+ "loss": 0.0049,
1469
+ "step": 1015
1470
+ },
1471
+ {
1472
+ "epoch": 4.34,
1473
+ "grad_norm": 0.3778130066090292,
1474
+ "learning_rate": 1.0425300163768903e-06,
1475
+ "loss": 0.0042,
1476
+ "step": 1020
1477
+ },
1478
+ {
1479
+ "epoch": 4.36,
1480
+ "grad_norm": 0.3983653765993126,
1481
+ "learning_rate": 9.77455335906149e-07,
1482
+ "loss": 0.0053,
1483
+ "step": 1025
1484
+ },
1485
+ {
1486
+ "epoch": 4.38,
1487
+ "grad_norm": 0.3604607140996719,
1488
+ "learning_rate": 9.143732127725591e-07,
1489
+ "loss": 0.0039,
1490
+ "step": 1030
1491
+ },
1492
+ {
1493
+ "epoch": 4.4,
1494
+ "grad_norm": 0.6021048871397348,
1495
+ "learning_rate": 8.532975781620511e-07,
1496
+ "loss": 0.0035,
1497
+ "step": 1035
1498
+ },
1499
+ {
1500
+ "epoch": 4.43,
1501
+ "grad_norm": 0.3771138444026374,
1502
+ "learning_rate": 7.942419201435014e-07,
1503
+ "loss": 0.0041,
1504
+ "step": 1040
1505
+ },
1506
+ {
1507
+ "epoch": 4.45,
1508
+ "grad_norm": 0.40288900755654594,
1509
+ "learning_rate": 7.372192806899947e-07,
1510
+ "loss": 0.0045,
1511
+ "step": 1045
1512
+ },
1513
+ {
1514
+ "epoch": 4.47,
1515
+ "grad_norm": 0.4475530083250148,
1516
+ "learning_rate": 6.822422527986161e-07,
1517
+ "loss": 0.0043,
1518
+ "step": 1050
1519
+ },
1520
+ {
1521
+ "epoch": 4.49,
1522
+ "grad_norm": 0.3791478244996085,
1523
+ "learning_rate": 6.293229777093779e-07,
1524
+ "loss": 0.0046,
1525
+ "step": 1055
1526
+ },
1527
+ {
1528
+ "epoch": 4.51,
1529
+ "grad_norm": 0.3778438936995471,
1530
+ "learning_rate": 5.784731422239276e-07,
1531
+ "loss": 0.0038,
1532
+ "step": 1060
1533
+ },
1534
+ {
1535
+ "epoch": 4.53,
1536
+ "grad_norm": 0.376192442816014,
1537
+ "learning_rate": 5.297039761246137e-07,
1538
+ "loss": 0.0038,
1539
+ "step": 1065
1540
+ },
1541
+ {
1542
+ "epoch": 4.55,
1543
+ "grad_norm": 0.2960720923070219,
1544
+ "learning_rate": 4.830262496944693e-07,
1545
+ "loss": 0.0037,
1546
+ "step": 1070
1547
+ },
1548
+ {
1549
+ "epoch": 4.57,
1550
+ "grad_norm": 0.37323225545474764,
1551
+ "learning_rate": 4.384502713386918e-07,
1552
+ "loss": 0.0038,
1553
+ "step": 1075
1554
+ },
1555
+ {
1556
+ "epoch": 4.6,
1557
+ "grad_norm": 0.33848258454167024,
1558
+ "learning_rate": 3.9598588530810335e-07,
1559
+ "loss": 0.0037,
1560
+ "step": 1080
1561
+ },
1562
+ {
1563
+ "epoch": 4.62,
1564
+ "grad_norm": 0.4828649331052619,
1565
+ "learning_rate": 3.5564246952512817e-07,
1566
+ "loss": 0.0034,
1567
+ "step": 1085
1568
+ },
1569
+ {
1570
+ "epoch": 4.64,
1571
+ "grad_norm": 0.43489507591573195,
1572
+ "learning_rate": 3.174289335127612e-07,
1573
+ "loss": 0.0039,
1574
+ "step": 1090
1575
+ },
1576
+ {
1577
+ "epoch": 4.66,
1578
+ "grad_norm": 0.3646806476787612,
1579
+ "learning_rate": 2.8135371642695865e-07,
1580
+ "loss": 0.0042,
1581
+ "step": 1095
1582
+ },
1583
+ {
1584
+ "epoch": 4.68,
1585
+ "grad_norm": 0.5325215204533521,
1586
+ "learning_rate": 2.474247851929246e-07,
1587
+ "loss": 0.0048,
1588
+ "step": 1100
1589
+ },
1590
+ {
1591
+ "epoch": 4.7,
1592
+ "grad_norm": 0.40238125679038056,
1593
+ "learning_rate": 2.1564963274568028e-07,
1594
+ "loss": 0.0051,
1595
+ "step": 1105
1596
+ },
1597
+ {
1598
+ "epoch": 4.72,
1599
+ "grad_norm": 0.38702990805573456,
1600
+ "learning_rate": 1.860352763753004e-07,
1601
+ "loss": 0.0042,
1602
+ "step": 1110
1603
+ },
1604
+ {
1605
+ "epoch": 4.74,
1606
+ "grad_norm": 0.3852959902302661,
1607
+ "learning_rate": 1.585882561772112e-07,
1608
+ "loss": 0.0044,
1609
+ "step": 1115
1610
+ },
1611
+ {
1612
+ "epoch": 4.77,
1613
+ "grad_norm": 0.44717232901584403,
1614
+ "learning_rate": 1.333146336078528e-07,
1615
+ "loss": 0.0043,
1616
+ "step": 1120
1617
+ },
1618
+ {
1619
+ "epoch": 4.79,
1620
+ "grad_norm": 0.372462847171052,
1621
+ "learning_rate": 1.1021999014606322e-07,
1622
+ "loss": 0.0038,
1623
+ "step": 1125
1624
+ },
1625
+ {
1626
+ "epoch": 4.81,
1627
+ "grad_norm": 0.2928988966778955,
1628
+ "learning_rate": 8.930942606044434e-08,
1629
+ "loss": 0.0034,
1630
+ "step": 1130
1631
+ },
1632
+ {
1633
+ "epoch": 4.83,
1634
+ "grad_norm": 0.39889536364254635,
1635
+ "learning_rate": 7.058755928301631e-08,
1636
+ "loss": 0.0051,
1637
+ "step": 1135
1638
+ },
1639
+ {
1640
+ "epoch": 4.85,
1641
+ "grad_norm": 0.40786489098741363,
1642
+ "learning_rate": 5.405852438937764e-08,
1643
+ "loss": 0.0034,
1644
+ "step": 1140
1645
+ },
1646
+ {
1647
+ "epoch": 4.87,
1648
+ "grad_norm": 0.3349150763788282,
1649
+ "learning_rate": 3.972597168562131e-08,
1650
+ "loss": 0.0066,
1651
+ "step": 1145
1652
+ },
1653
+ {
1654
+ "epoch": 4.89,
1655
+ "grad_norm": 0.3597591461189066,
1656
+ "learning_rate": 2.7593066402189506e-08,
1657
+ "loss": 0.0037,
1658
+ "step": 1150
1659
+ },
1660
+ {
1661
+ "epoch": 4.91,
1662
+ "grad_norm": 0.4040878903041599,
1663
+ "learning_rate": 1.7662487994862808e-08,
1664
+ "loss": 0.0039,
1665
+ "step": 1155
1666
+ },
1667
+ {
1668
+ "epoch": 4.94,
1669
+ "grad_norm": 0.3992498945384279,
1670
+ "learning_rate": 9.93642955301799e-09,
1671
+ "loss": 0.0039,
1672
+ "step": 1160
1673
+ },
1674
+ {
1675
+ "epoch": 4.96,
1676
+ "grad_norm": 0.3821243876748289,
1677
+ "learning_rate": 4.4165973153076355e-09,
1678
+ "loss": 0.0044,
1679
+ "step": 1165
1680
+ },
1681
+ {
1682
+ "epoch": 4.98,
1683
+ "grad_norm": 0.3430681816581551,
1684
+ "learning_rate": 1.1042102928460196e-09,
1685
+ "loss": 0.0045,
1686
+ "step": 1170
1687
+ },
1688
+ {
1689
+ "epoch": 5.0,
1690
+ "grad_norm": 0.32719513702973596,
1691
+ "learning_rate": 0.0,
1692
+ "loss": 0.004,
1693
+ "step": 1175
1694
+ },
1695
+ {
1696
+ "epoch": 5.0,
1697
+ "eval_loss": 0.7659260034561157,
1698
+ "eval_runtime": 346.6983,
1699
+ "eval_samples_per_second": 21.653,
1700
+ "eval_steps_per_second": 0.34,
1701
+ "step": 1175
1702
+ },
1703
+ {
1704
+ "epoch": 5.0,
1705
+ "step": 1175,
1706
+ "total_flos": 246021095424000.0,
1707
+ "train_loss": 0.1592100728509274,
1708
+ "train_runtime": 9397.2382,
1709
+ "train_samples_per_second": 3.994,
1710
+ "train_steps_per_second": 0.125
1711
  }
1712
  ],
1713
  "logging_steps": 5,
1714
+ "max_steps": 1175,
1715
  "num_input_tokens_seen": 0,
1716
+ "num_train_epochs": 5,
1717
  "save_steps": 100,
1718
+ "total_flos": 246021095424000.0,
1719
+ "train_batch_size": 4,
1720
  "trial_name": null,
1721
  "trial_params": null
1722
  }