gsmyrnis commited on
Commit
74e48dd
·
verified ·
1 Parent(s): faefd19

End of training

Browse files
Files changed (5) hide show
  1. README.md +2 -1
  2. all_results.json +6 -10
  3. train_results.json +6 -6
  4. trainer_state.json +1507 -40
  5. training_loss.png +0 -0
README.md CHANGED
@@ -4,6 +4,7 @@ license: apache-2.0
4
  base_model: Qwen/Qwen2.5-7B-Instruct
5
  tags:
6
  - llama-factory
 
7
  - generated_from_trainer
8
  model-index:
9
  - name: llama3-1_8b_r1_annotated_math
@@ -15,7 +16,7 @@ should probably proofread and complete it, then remove this comment. -->
15
 
16
  # llama3-1_8b_r1_annotated_math
17
 
18
- This model is a fine-tuned version of [Qwen/Qwen2.5-7B-Instruct](https://huggingface.co/Qwen/Qwen2.5-7B-Instruct) on an unknown dataset.
19
 
20
  ## Model description
21
 
 
4
  base_model: Qwen/Qwen2.5-7B-Instruct
5
  tags:
6
  - llama-factory
7
+ - full
8
  - generated_from_trainer
9
  model-index:
10
  - name: llama3-1_8b_r1_annotated_math
 
16
 
17
  # llama3-1_8b_r1_annotated_math
18
 
19
+ This model is a fine-tuned version of [Qwen/Qwen2.5-7B-Instruct](https://huggingface.co/Qwen/Qwen2.5-7B-Instruct) on the mlfoundations-dev/r1_annotated_math dataset.
20
 
21
  ## Model description
22
 
all_results.json CHANGED
@@ -1,12 +1,8 @@
1
  {
2
- "epoch": 3.0,
3
- "eval_loss": 0.5833666324615479,
4
- "eval_runtime": 1.8372,
5
- "eval_samples_per_second": 162.199,
6
- "eval_steps_per_second": 1.089,
7
- "total_flos": 60301340835840.0,
8
- "train_loss": 0.6545261012183295,
9
- "train_runtime": 627.5494,
10
- "train_samples_per_second": 27.043,
11
- "train_steps_per_second": 0.057
12
  }
 
1
  {
2
+ "epoch": 2.986175115207373,
3
+ "total_flos": 130743839866880.0,
4
+ "train_loss": 0.4320466773653472,
5
+ "train_runtime": 2638.4795,
6
+ "train_samples_per_second": 7.869,
7
+ "train_steps_per_second": 0.082
 
 
 
 
8
  }
train_results.json CHANGED
@@ -1,8 +1,8 @@
1
  {
2
- "epoch": 3.0,
3
- "total_flos": 60301340835840.0,
4
- "train_loss": 0.6545261012183295,
5
- "train_runtime": 627.5494,
6
- "train_samples_per_second": 27.043,
7
- "train_steps_per_second": 0.057
8
  }
 
1
  {
2
+ "epoch": 2.986175115207373,
3
+ "total_flos": 130743839866880.0,
4
+ "train_loss": 0.4320466773653472,
5
+ "train_runtime": 2638.4795,
6
+ "train_samples_per_second": 7.869,
7
+ "train_steps_per_second": 0.082
8
  }
trainer_state.json CHANGED
@@ -1,70 +1,1537 @@
1
  {
2
  "best_metric": null,
3
  "best_model_checkpoint": null,
4
- "epoch": 3.0,
5
  "eval_steps": 500,
6
- "global_step": 36,
7
  "is_hyper_param_search": false,
8
  "is_local_process_zero": true,
9
  "is_world_process_zero": true,
10
  "log_history": [
11
  {
12
- "epoch": 0.8333333333333334,
13
- "grad_norm": 1.4976304773590843,
14
- "learning_rate": 5e-06,
15
- "loss": 0.8098,
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
16
  "step": 10
17
  },
18
  {
19
- "epoch": 1.0,
20
- "eval_loss": 0.6748126149177551,
21
- "eval_runtime": 1.8628,
22
- "eval_samples_per_second": 159.978,
23
- "eval_steps_per_second": 1.074,
 
 
 
 
 
 
24
  "step": 12
25
  },
26
  {
27
- "epoch": 1.6666666666666665,
28
- "grad_norm": 0.9879481748413028,
29
- "learning_rate": 5e-06,
30
- "loss": 0.6489,
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
31
  "step": 20
32
  },
33
  {
34
- "epoch": 2.0,
35
- "eval_loss": 0.6079719066619873,
36
- "eval_runtime": 1.8561,
37
- "eval_samples_per_second": 160.554,
38
- "eval_steps_per_second": 1.078,
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
39
  "step": 24
40
  },
41
  {
42
- "epoch": 2.5,
43
- "grad_norm": 1.0949275268721474,
44
- "learning_rate": 5e-06,
45
- "loss": 0.5774,
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
46
  "step": 30
47
  },
48
  {
49
- "epoch": 3.0,
50
- "eval_loss": 0.5833666324615479,
51
- "eval_runtime": 1.8239,
52
- "eval_samples_per_second": 163.388,
53
- "eval_steps_per_second": 1.097,
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
54
  "step": 36
55
  },
56
  {
57
- "epoch": 3.0,
58
- "step": 36,
59
- "total_flos": 60301340835840.0,
60
- "train_loss": 0.6545261012183295,
61
- "train_runtime": 627.5494,
62
- "train_samples_per_second": 27.043,
63
- "train_steps_per_second": 0.057
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
64
  }
65
  ],
66
- "logging_steps": 10,
67
- "max_steps": 36,
68
  "num_input_tokens_seen": 0,
69
  "num_train_epochs": 3,
70
  "save_steps": 500,
@@ -80,8 +1547,8 @@
80
  "attributes": {}
81
  }
82
  },
83
- "total_flos": 60301340835840.0,
84
- "train_batch_size": 16,
85
  "trial_name": null,
86
  "trial_params": null
87
  }
 
1
  {
2
  "best_metric": null,
3
  "best_model_checkpoint": null,
4
+ "epoch": 2.986175115207373,
5
  "eval_steps": 500,
6
+ "global_step": 216,
7
  "is_hyper_param_search": false,
8
  "is_local_process_zero": true,
9
  "is_world_process_zero": true,
10
  "log_history": [
11
  {
12
+ "epoch": 0.013824884792626729,
13
+ "grad_norm": 7.633559049394886,
14
+ "learning_rate": 4.5454545454545457e-07,
15
+ "loss": 0.8764,
16
+ "step": 1
17
+ },
18
+ {
19
+ "epoch": 0.027649769585253458,
20
+ "grad_norm": 7.6469224678917,
21
+ "learning_rate": 9.090909090909091e-07,
22
+ "loss": 0.8712,
23
+ "step": 2
24
+ },
25
+ {
26
+ "epoch": 0.041474654377880185,
27
+ "grad_norm": 7.5089460233893295,
28
+ "learning_rate": 1.3636363636363636e-06,
29
+ "loss": 0.8804,
30
+ "step": 3
31
+ },
32
+ {
33
+ "epoch": 0.055299539170506916,
34
+ "grad_norm": 7.691760047071555,
35
+ "learning_rate": 1.8181818181818183e-06,
36
+ "loss": 0.8775,
37
+ "step": 4
38
+ },
39
+ {
40
+ "epoch": 0.06912442396313365,
41
+ "grad_norm": 7.294963661060089,
42
+ "learning_rate": 2.2727272727272728e-06,
43
+ "loss": 0.8709,
44
+ "step": 5
45
+ },
46
+ {
47
+ "epoch": 0.08294930875576037,
48
+ "grad_norm": 5.443182880842741,
49
+ "learning_rate": 2.7272727272727272e-06,
50
+ "loss": 0.7907,
51
+ "step": 6
52
+ },
53
+ {
54
+ "epoch": 0.0967741935483871,
55
+ "grad_norm": 5.047617387545844,
56
+ "learning_rate": 3.181818181818182e-06,
57
+ "loss": 0.7836,
58
+ "step": 7
59
+ },
60
+ {
61
+ "epoch": 0.11059907834101383,
62
+ "grad_norm": 3.2253892674265594,
63
+ "learning_rate": 3.6363636363636366e-06,
64
+ "loss": 0.7531,
65
+ "step": 8
66
+ },
67
+ {
68
+ "epoch": 0.12442396313364056,
69
+ "grad_norm": 2.8131888920599657,
70
+ "learning_rate": 4.0909090909090915e-06,
71
+ "loss": 0.7417,
72
+ "step": 9
73
+ },
74
+ {
75
+ "epoch": 0.1382488479262673,
76
+ "grad_norm": 3.6977429779435815,
77
+ "learning_rate": 4.5454545454545455e-06,
78
+ "loss": 0.723,
79
  "step": 10
80
  },
81
  {
82
+ "epoch": 0.15207373271889402,
83
+ "grad_norm": 4.281498675575116,
84
+ "learning_rate": 5e-06,
85
+ "loss": 0.7246,
86
+ "step": 11
87
+ },
88
+ {
89
+ "epoch": 0.16589861751152074,
90
+ "grad_norm": 3.6996387465448373,
91
+ "learning_rate": 5.4545454545454545e-06,
92
+ "loss": 0.6595,
93
  "step": 12
94
  },
95
  {
96
+ "epoch": 0.17972350230414746,
97
+ "grad_norm": 3.78830344374056,
98
+ "learning_rate": 5.90909090909091e-06,
99
+ "loss": 0.6694,
100
+ "step": 13
101
+ },
102
+ {
103
+ "epoch": 0.1935483870967742,
104
+ "grad_norm": 3.3490463351493713,
105
+ "learning_rate": 6.363636363636364e-06,
106
+ "loss": 0.6227,
107
+ "step": 14
108
+ },
109
+ {
110
+ "epoch": 0.2073732718894009,
111
+ "grad_norm": 3.374405746489098,
112
+ "learning_rate": 6.818181818181818e-06,
113
+ "loss": 0.6273,
114
+ "step": 15
115
+ },
116
+ {
117
+ "epoch": 0.22119815668202766,
118
+ "grad_norm": 2.4954014740266195,
119
+ "learning_rate": 7.272727272727273e-06,
120
+ "loss": 0.6062,
121
+ "step": 16
122
+ },
123
+ {
124
+ "epoch": 0.2350230414746544,
125
+ "grad_norm": 1.8239008983436202,
126
+ "learning_rate": 7.727272727272727e-06,
127
+ "loss": 0.593,
128
+ "step": 17
129
+ },
130
+ {
131
+ "epoch": 0.2488479262672811,
132
+ "grad_norm": 1.635875724467323,
133
+ "learning_rate": 8.181818181818183e-06,
134
+ "loss": 0.5959,
135
+ "step": 18
136
+ },
137
+ {
138
+ "epoch": 0.2626728110599078,
139
+ "grad_norm": 2.0334716933077166,
140
+ "learning_rate": 8.636363636363637e-06,
141
+ "loss": 0.5672,
142
+ "step": 19
143
+ },
144
+ {
145
+ "epoch": 0.2764976958525346,
146
+ "grad_norm": 1.9168703834196266,
147
+ "learning_rate": 9.090909090909091e-06,
148
+ "loss": 0.5729,
149
  "step": 20
150
  },
151
  {
152
+ "epoch": 0.2903225806451613,
153
+ "grad_norm": 1.4466966621915673,
154
+ "learning_rate": 9.545454545454547e-06,
155
+ "loss": 0.5534,
156
+ "step": 21
157
+ },
158
+ {
159
+ "epoch": 0.30414746543778803,
160
+ "grad_norm": 1.0267896741626927,
161
+ "learning_rate": 1e-05,
162
+ "loss": 0.5628,
163
+ "step": 22
164
+ },
165
+ {
166
+ "epoch": 0.31797235023041476,
167
+ "grad_norm": 1.262616482011545,
168
+ "learning_rate": 9.999344418328161e-06,
169
+ "loss": 0.5572,
170
+ "step": 23
171
+ },
172
+ {
173
+ "epoch": 0.3317972350230415,
174
+ "grad_norm": 1.0906634918414906,
175
+ "learning_rate": 9.997377845227577e-06,
176
+ "loss": 0.5146,
177
  "step": 24
178
  },
179
  {
180
+ "epoch": 0.3456221198156682,
181
+ "grad_norm": 1.0415845965231307,
182
+ "learning_rate": 9.994100796397954e-06,
183
+ "loss": 0.4973,
184
+ "step": 25
185
+ },
186
+ {
187
+ "epoch": 0.35944700460829493,
188
+ "grad_norm": 0.815227789570975,
189
+ "learning_rate": 9.98951413118856e-06,
190
+ "loss": 0.5218,
191
+ "step": 26
192
+ },
193
+ {
194
+ "epoch": 0.37327188940092165,
195
+ "grad_norm": 0.8903243919241076,
196
+ "learning_rate": 9.983619052372847e-06,
197
+ "loss": 0.5167,
198
+ "step": 27
199
+ },
200
+ {
201
+ "epoch": 0.3870967741935484,
202
+ "grad_norm": 1.0991055385660755,
203
+ "learning_rate": 9.97641710583307e-06,
204
+ "loss": 0.5003,
205
+ "step": 28
206
+ },
207
+ {
208
+ "epoch": 0.4009216589861751,
209
+ "grad_norm": 0.9348867416996314,
210
+ "learning_rate": 9.96791018015489e-06,
211
+ "loss": 0.5078,
212
+ "step": 29
213
+ },
214
+ {
215
+ "epoch": 0.4147465437788018,
216
+ "grad_norm": 0.7138578554042776,
217
+ "learning_rate": 9.958100506132127e-06,
218
+ "loss": 0.5075,
219
  "step": 30
220
  },
221
  {
222
+ "epoch": 0.42857142857142855,
223
+ "grad_norm": 0.8349505675026924,
224
+ "learning_rate": 9.946990656181782e-06,
225
+ "loss": 0.511,
226
+ "step": 31
227
+ },
228
+ {
229
+ "epoch": 0.4423963133640553,
230
+ "grad_norm": 1.052621697155694,
231
+ "learning_rate": 9.934583543669454e-06,
232
+ "loss": 0.4824,
233
+ "step": 32
234
+ },
235
+ {
236
+ "epoch": 0.45622119815668205,
237
+ "grad_norm": 0.7618516629853732,
238
+ "learning_rate": 9.920882422145372e-06,
239
+ "loss": 0.4842,
240
+ "step": 33
241
+ },
242
+ {
243
+ "epoch": 0.4700460829493088,
244
+ "grad_norm": 0.8523130797451687,
245
+ "learning_rate": 9.905890884491196e-06,
246
+ "loss": 0.4843,
247
+ "step": 34
248
+ },
249
+ {
250
+ "epoch": 0.4838709677419355,
251
+ "grad_norm": 0.7854821313979271,
252
+ "learning_rate": 9.889612861977855e-06,
253
+ "loss": 0.4893,
254
+ "step": 35
255
+ },
256
+ {
257
+ "epoch": 0.4976958525345622,
258
+ "grad_norm": 0.9098881831155701,
259
+ "learning_rate": 9.872052623234632e-06,
260
+ "loss": 0.5024,
261
  "step": 36
262
  },
263
  {
264
+ "epoch": 0.511520737327189,
265
+ "grad_norm": 0.9133145917535354,
266
+ "learning_rate": 9.853214773129796e-06,
267
+ "loss": 0.4828,
268
+ "step": 37
269
+ },
270
+ {
271
+ "epoch": 0.5253456221198156,
272
+ "grad_norm": 0.7771673181178608,
273
+ "learning_rate": 9.833104251563058e-06,
274
+ "loss": 0.478,
275
+ "step": 38
276
+ },
277
+ {
278
+ "epoch": 0.5391705069124424,
279
+ "grad_norm": 0.6686414859648492,
280
+ "learning_rate": 9.811726332170153e-06,
281
+ "loss": 0.4945,
282
+ "step": 39
283
+ },
284
+ {
285
+ "epoch": 0.5529953917050692,
286
+ "grad_norm": 0.7697756564970056,
287
+ "learning_rate": 9.789086620939936e-06,
288
+ "loss": 0.4704,
289
+ "step": 40
290
+ },
291
+ {
292
+ "epoch": 0.5668202764976958,
293
+ "grad_norm": 0.8065645041682494,
294
+ "learning_rate": 9.765191054744305e-06,
295
+ "loss": 0.4512,
296
+ "step": 41
297
+ },
298
+ {
299
+ "epoch": 0.5806451612903226,
300
+ "grad_norm": 0.7428419868914332,
301
+ "learning_rate": 9.740045899781353e-06,
302
+ "loss": 0.4631,
303
+ "step": 42
304
+ },
305
+ {
306
+ "epoch": 0.5944700460829493,
307
+ "grad_norm": 0.7370318617715542,
308
+ "learning_rate": 9.713657749932172e-06,
309
+ "loss": 0.4834,
310
+ "step": 43
311
+ },
312
+ {
313
+ "epoch": 0.6082949308755761,
314
+ "grad_norm": 0.5647708496508699,
315
+ "learning_rate": 9.68603352503172e-06,
316
+ "loss": 0.4774,
317
+ "step": 44
318
+ },
319
+ {
320
+ "epoch": 0.6221198156682027,
321
+ "grad_norm": 0.6790318592831586,
322
+ "learning_rate": 9.657180469054213e-06,
323
+ "loss": 0.4964,
324
+ "step": 45
325
+ },
326
+ {
327
+ "epoch": 0.6359447004608295,
328
+ "grad_norm": 0.6728496708051098,
329
+ "learning_rate": 9.627106148213521e-06,
330
+ "loss": 0.5024,
331
+ "step": 46
332
+ },
333
+ {
334
+ "epoch": 0.6497695852534562,
335
+ "grad_norm": 0.6158261913286127,
336
+ "learning_rate": 9.595818448979061e-06,
337
+ "loss": 0.4656,
338
+ "step": 47
339
+ },
340
+ {
341
+ "epoch": 0.663594470046083,
342
+ "grad_norm": 0.6396258914140209,
343
+ "learning_rate": 9.563325576007702e-06,
344
+ "loss": 0.4593,
345
+ "step": 48
346
+ },
347
+ {
348
+ "epoch": 0.6774193548387096,
349
+ "grad_norm": 0.55574623840064,
350
+ "learning_rate": 9.529636049992235e-06,
351
+ "loss": 0.4848,
352
+ "step": 49
353
+ },
354
+ {
355
+ "epoch": 0.6912442396313364,
356
+ "grad_norm": 0.5960107153575481,
357
+ "learning_rate": 9.494758705426978e-06,
358
+ "loss": 0.4606,
359
+ "step": 50
360
+ },
361
+ {
362
+ "epoch": 0.7050691244239631,
363
+ "grad_norm": 0.6187132785014103,
364
+ "learning_rate": 9.458702688291072e-06,
365
+ "loss": 0.4706,
366
+ "step": 51
367
+ },
368
+ {
369
+ "epoch": 0.7188940092165899,
370
+ "grad_norm": 0.596391894724299,
371
+ "learning_rate": 9.421477453650118e-06,
372
+ "loss": 0.4552,
373
+ "step": 52
374
+ },
375
+ {
376
+ "epoch": 0.7327188940092166,
377
+ "grad_norm": 0.5828525328792538,
378
+ "learning_rate": 9.38309276317674e-06,
379
+ "loss": 0.4419,
380
+ "step": 53
381
+ },
382
+ {
383
+ "epoch": 0.7465437788018433,
384
+ "grad_norm": 0.5574488192167764,
385
+ "learning_rate": 9.343558682590757e-06,
386
+ "loss": 0.4482,
387
+ "step": 54
388
+ },
389
+ {
390
+ "epoch": 0.7603686635944701,
391
+ "grad_norm": 0.5810226590826171,
392
+ "learning_rate": 9.302885579019626e-06,
393
+ "loss": 0.4574,
394
+ "step": 55
395
+ },
396
+ {
397
+ "epoch": 0.7741935483870968,
398
+ "grad_norm": 0.6373464198147619,
399
+ "learning_rate": 9.261084118279846e-06,
400
+ "loss": 0.5021,
401
+ "step": 56
402
+ },
403
+ {
404
+ "epoch": 0.7880184331797235,
405
+ "grad_norm": 0.6473035563889259,
406
+ "learning_rate": 9.218165262080024e-06,
407
+ "loss": 0.4688,
408
+ "step": 57
409
+ },
410
+ {
411
+ "epoch": 0.8018433179723502,
412
+ "grad_norm": 0.5496475133999553,
413
+ "learning_rate": 9.174140265146356e-06,
414
+ "loss": 0.4614,
415
+ "step": 58
416
+ },
417
+ {
418
+ "epoch": 0.815668202764977,
419
+ "grad_norm": 0.5633672236656848,
420
+ "learning_rate": 9.129020672271283e-06,
421
+ "loss": 0.4631,
422
+ "step": 59
423
+ },
424
+ {
425
+ "epoch": 0.8294930875576036,
426
+ "grad_norm": 0.5996514489934792,
427
+ "learning_rate": 9.082818315286054e-06,
428
+ "loss": 0.4709,
429
+ "step": 60
430
+ },
431
+ {
432
+ "epoch": 0.8433179723502304,
433
+ "grad_norm": 0.5800489851592833,
434
+ "learning_rate": 9.035545309958048e-06,
435
+ "loss": 0.4529,
436
+ "step": 61
437
+ },
438
+ {
439
+ "epoch": 0.8571428571428571,
440
+ "grad_norm": 0.5728565615351726,
441
+ "learning_rate": 8.987214052813605e-06,
442
+ "loss": 0.4689,
443
+ "step": 62
444
+ },
445
+ {
446
+ "epoch": 0.8709677419354839,
447
+ "grad_norm": 0.635495629930791,
448
+ "learning_rate": 8.937837217887273e-06,
449
+ "loss": 0.4577,
450
+ "step": 63
451
+ },
452
+ {
453
+ "epoch": 0.8847926267281107,
454
+ "grad_norm": 0.6832859023985155,
455
+ "learning_rate": 8.887427753398249e-06,
456
+ "loss": 0.4551,
457
+ "step": 64
458
+ },
459
+ {
460
+ "epoch": 0.8986175115207373,
461
+ "grad_norm": 0.6250704218728608,
462
+ "learning_rate": 8.83599887835493e-06,
463
+ "loss": 0.4412,
464
+ "step": 65
465
+ },
466
+ {
467
+ "epoch": 0.9124423963133641,
468
+ "grad_norm": 0.4941669349016631,
469
+ "learning_rate": 8.783564079088478e-06,
470
+ "loss": 0.452,
471
+ "step": 66
472
+ },
473
+ {
474
+ "epoch": 0.9262672811059908,
475
+ "grad_norm": 0.6418218067831372,
476
+ "learning_rate": 8.730137105716231e-06,
477
+ "loss": 0.4374,
478
+ "step": 67
479
+ },
480
+ {
481
+ "epoch": 0.9400921658986175,
482
+ "grad_norm": 0.4944995367016919,
483
+ "learning_rate": 8.675731968536004e-06,
484
+ "loss": 0.4538,
485
+ "step": 68
486
+ },
487
+ {
488
+ "epoch": 0.9539170506912442,
489
+ "grad_norm": 0.5351112267428982,
490
+ "learning_rate": 8.620362934352109e-06,
491
+ "loss": 0.456,
492
+ "step": 69
493
+ },
494
+ {
495
+ "epoch": 0.967741935483871,
496
+ "grad_norm": 0.550190483341697,
497
+ "learning_rate": 8.564044522734147e-06,
498
+ "loss": 0.4696,
499
+ "step": 70
500
+ },
501
+ {
502
+ "epoch": 0.9815668202764977,
503
+ "grad_norm": 0.5416241347883733,
504
+ "learning_rate": 8.506791502209497e-06,
505
+ "loss": 0.4351,
506
+ "step": 71
507
+ },
508
+ {
509
+ "epoch": 0.9953917050691244,
510
+ "grad_norm": 0.5646211850916225,
511
+ "learning_rate": 8.448618886390523e-06,
512
+ "loss": 0.4395,
513
+ "step": 72
514
+ },
515
+ {
516
+ "epoch": 1.0092165898617511,
517
+ "grad_norm": 1.0497545314797017,
518
+ "learning_rate": 8.389541930037516e-06,
519
+ "loss": 0.7044,
520
+ "step": 73
521
+ },
522
+ {
523
+ "epoch": 1.023041474654378,
524
+ "grad_norm": 0.6050451609110485,
525
+ "learning_rate": 8.329576125058406e-06,
526
+ "loss": 0.4019,
527
+ "step": 74
528
+ },
529
+ {
530
+ "epoch": 1.0368663594470047,
531
+ "grad_norm": 0.668170300230099,
532
+ "learning_rate": 8.268737196446264e-06,
533
+ "loss": 0.4301,
534
+ "step": 75
535
+ },
536
+ {
537
+ "epoch": 1.0506912442396312,
538
+ "grad_norm": 0.5614144997152238,
539
+ "learning_rate": 8.207041098155701e-06,
540
+ "loss": 0.3802,
541
+ "step": 76
542
+ },
543
+ {
544
+ "epoch": 1.064516129032258,
545
+ "grad_norm": 0.6517654191953809,
546
+ "learning_rate": 8.144504008919224e-06,
547
+ "loss": 0.4079,
548
+ "step": 77
549
+ },
550
+ {
551
+ "epoch": 1.0783410138248848,
552
+ "grad_norm": 0.5437003169084378,
553
+ "learning_rate": 8.081142328004638e-06,
554
+ "loss": 0.3725,
555
+ "step": 78
556
+ },
557
+ {
558
+ "epoch": 1.0921658986175116,
559
+ "grad_norm": 0.6545122792430208,
560
+ "learning_rate": 8.016972670914624e-06,
561
+ "loss": 0.4282,
562
+ "step": 79
563
+ },
564
+ {
565
+ "epoch": 1.1059907834101383,
566
+ "grad_norm": 0.6048618596815725,
567
+ "learning_rate": 7.952011865029614e-06,
568
+ "loss": 0.4037,
569
+ "step": 80
570
+ },
571
+ {
572
+ "epoch": 1.119815668202765,
573
+ "grad_norm": 0.6691394124559549,
574
+ "learning_rate": 7.886276945195098e-06,
575
+ "loss": 0.3643,
576
+ "step": 81
577
+ },
578
+ {
579
+ "epoch": 1.1336405529953917,
580
+ "grad_norm": 0.7270083026506138,
581
+ "learning_rate": 7.819785149254534e-06,
582
+ "loss": 0.4131,
583
+ "step": 82
584
+ },
585
+ {
586
+ "epoch": 1.1474654377880185,
587
+ "grad_norm": 0.5582455458120927,
588
+ "learning_rate": 7.752553913529019e-06,
589
+ "loss": 0.368,
590
+ "step": 83
591
+ },
592
+ {
593
+ "epoch": 1.1612903225806452,
594
+ "grad_norm": 0.6782237172690554,
595
+ "learning_rate": 7.68460086824492e-06,
596
+ "loss": 0.4308,
597
+ "step": 84
598
+ },
599
+ {
600
+ "epoch": 1.1751152073732718,
601
+ "grad_norm": 0.6173590398850847,
602
+ "learning_rate": 7.61594383291065e-06,
603
+ "loss": 0.3943,
604
+ "step": 85
605
+ },
606
+ {
607
+ "epoch": 1.1889400921658986,
608
+ "grad_norm": 0.5909172461024365,
609
+ "learning_rate": 7.546600811643816e-06,
610
+ "loss": 0.4211,
611
+ "step": 86
612
+ },
613
+ {
614
+ "epoch": 1.2027649769585254,
615
+ "grad_norm": 0.5670563053696445,
616
+ "learning_rate": 7.476589988449939e-06,
617
+ "loss": 0.3704,
618
+ "step": 87
619
+ },
620
+ {
621
+ "epoch": 1.2165898617511521,
622
+ "grad_norm": 0.5698865538843494,
623
+ "learning_rate": 7.405929722454026e-06,
624
+ "loss": 0.3912,
625
+ "step": 88
626
+ },
627
+ {
628
+ "epoch": 1.230414746543779,
629
+ "grad_norm": 0.5331376655462895,
630
+ "learning_rate": 7.334638543086203e-06,
631
+ "loss": 0.3836,
632
+ "step": 89
633
+ },
634
+ {
635
+ "epoch": 1.2442396313364055,
636
+ "grad_norm": 0.6325980757371303,
637
+ "learning_rate": 7.262735145222696e-06,
638
+ "loss": 0.3915,
639
+ "step": 90
640
+ },
641
+ {
642
+ "epoch": 1.2580645161290323,
643
+ "grad_norm": 0.5952012132281536,
644
+ "learning_rate": 7.190238384283413e-06,
645
+ "loss": 0.4005,
646
+ "step": 91
647
+ },
648
+ {
649
+ "epoch": 1.271889400921659,
650
+ "grad_norm": 0.5581313424613799,
651
+ "learning_rate": 7.117167271287453e-06,
652
+ "loss": 0.3991,
653
+ "step": 92
654
+ },
655
+ {
656
+ "epoch": 1.2857142857142856,
657
+ "grad_norm": 0.5043175284762973,
658
+ "learning_rate": 7.043540967867782e-06,
659
+ "loss": 0.3643,
660
+ "step": 93
661
+ },
662
+ {
663
+ "epoch": 1.2995391705069124,
664
+ "grad_norm": 0.5598625221287372,
665
+ "learning_rate": 6.969378781246436e-06,
666
+ "loss": 0.4334,
667
+ "step": 94
668
+ },
669
+ {
670
+ "epoch": 1.3133640552995391,
671
+ "grad_norm": 0.5503414372922475,
672
+ "learning_rate": 6.894700159171535e-06,
673
+ "loss": 0.415,
674
+ "step": 95
675
+ },
676
+ {
677
+ "epoch": 1.327188940092166,
678
+ "grad_norm": 0.5592853069996093,
679
+ "learning_rate": 6.819524684817439e-06,
680
+ "loss": 0.407,
681
+ "step": 96
682
+ },
683
+ {
684
+ "epoch": 1.3410138248847927,
685
+ "grad_norm": 0.5431739142566777,
686
+ "learning_rate": 6.743872071649411e-06,
687
+ "loss": 0.401,
688
+ "step": 97
689
+ },
690
+ {
691
+ "epoch": 1.3548387096774195,
692
+ "grad_norm": 0.5168119415119301,
693
+ "learning_rate": 6.667762158254104e-06,
694
+ "loss": 0.3713,
695
+ "step": 98
696
+ },
697
+ {
698
+ "epoch": 1.368663594470046,
699
+ "grad_norm": 0.6207479541491344,
700
+ "learning_rate": 6.591214903137221e-06,
701
+ "loss": 0.389,
702
+ "step": 99
703
+ },
704
+ {
705
+ "epoch": 1.3824884792626728,
706
+ "grad_norm": 0.5364662180025893,
707
+ "learning_rate": 6.514250379489754e-06,
708
+ "loss": 0.3954,
709
+ "step": 100
710
+ },
711
+ {
712
+ "epoch": 1.3963133640552996,
713
+ "grad_norm": 0.5066869500264701,
714
+ "learning_rate": 6.436888769924142e-06,
715
+ "loss": 0.3939,
716
+ "step": 101
717
+ },
718
+ {
719
+ "epoch": 1.4101382488479262,
720
+ "grad_norm": 0.5780367268696035,
721
+ "learning_rate": 6.3591503611817155e-06,
722
+ "loss": 0.4076,
723
+ "step": 102
724
+ },
725
+ {
726
+ "epoch": 1.423963133640553,
727
+ "grad_norm": 0.5070490336891309,
728
+ "learning_rate": 6.281055538812861e-06,
729
+ "loss": 0.3604,
730
+ "step": 103
731
+ },
732
+ {
733
+ "epoch": 1.4377880184331797,
734
+ "grad_norm": 0.612079276266504,
735
+ "learning_rate": 6.202624781831269e-06,
736
+ "loss": 0.4067,
737
+ "step": 104
738
+ },
739
+ {
740
+ "epoch": 1.4516129032258065,
741
+ "grad_norm": 0.5734702647572101,
742
+ "learning_rate": 6.123878657343648e-06,
743
+ "loss": 0.4255,
744
+ "step": 105
745
+ },
746
+ {
747
+ "epoch": 1.4654377880184333,
748
+ "grad_norm": 0.5354043563850319,
749
+ "learning_rate": 6.044837815156377e-06,
750
+ "loss": 0.4074,
751
+ "step": 106
752
+ },
753
+ {
754
+ "epoch": 1.4792626728110598,
755
+ "grad_norm": 0.5791707978034125,
756
+ "learning_rate": 5.965522982360441e-06,
757
+ "loss": 0.3895,
758
+ "step": 107
759
+ },
760
+ {
761
+ "epoch": 1.4930875576036866,
762
+ "grad_norm": 0.5703922296796244,
763
+ "learning_rate": 5.885954957896115e-06,
764
+ "loss": 0.4086,
765
+ "step": 108
766
+ },
767
+ {
768
+ "epoch": 1.5069124423963134,
769
+ "grad_norm": 0.5065784500919478,
770
+ "learning_rate": 5.806154607098799e-06,
771
+ "loss": 0.3693,
772
+ "step": 109
773
+ },
774
+ {
775
+ "epoch": 1.52073732718894,
776
+ "grad_norm": 0.556653840063644,
777
+ "learning_rate": 5.726142856227453e-06,
778
+ "loss": 0.4299,
779
+ "step": 110
780
+ },
781
+ {
782
+ "epoch": 1.5345622119815667,
783
+ "grad_norm": 0.5474174976443692,
784
+ "learning_rate": 5.645940686977033e-06,
785
+ "loss": 0.4148,
786
+ "step": 111
787
+ },
788
+ {
789
+ "epoch": 1.5483870967741935,
790
+ "grad_norm": 0.4806025731120033,
791
+ "learning_rate": 5.5655691309764225e-06,
792
+ "loss": 0.3924,
793
+ "step": 112
794
+ },
795
+ {
796
+ "epoch": 1.5622119815668203,
797
+ "grad_norm": 0.5324709526786499,
798
+ "learning_rate": 5.485049264273241e-06,
799
+ "loss": 0.3789,
800
+ "step": 113
801
+ },
802
+ {
803
+ "epoch": 1.576036866359447,
804
+ "grad_norm": 0.4659634896990746,
805
+ "learning_rate": 5.404402201807022e-06,
806
+ "loss": 0.3609,
807
+ "step": 114
808
+ },
809
+ {
810
+ "epoch": 1.5898617511520738,
811
+ "grad_norm": 0.5808920912192916,
812
+ "learning_rate": 5.323649091872179e-06,
813
+ "loss": 0.4271,
814
+ "step": 115
815
+ },
816
+ {
817
+ "epoch": 1.6036866359447006,
818
+ "grad_norm": 0.5993918356229727,
819
+ "learning_rate": 5.242811110572243e-06,
820
+ "loss": 0.3989,
821
+ "step": 116
822
+ },
823
+ {
824
+ "epoch": 1.6175115207373272,
825
+ "grad_norm": 0.5109808802516896,
826
+ "learning_rate": 5.161909456266781e-06,
827
+ "loss": 0.3814,
828
+ "step": 117
829
+ },
830
+ {
831
+ "epoch": 1.631336405529954,
832
+ "grad_norm": 0.5423585025364462,
833
+ "learning_rate": 5.080965344012509e-06,
834
+ "loss": 0.3985,
835
+ "step": 118
836
+ },
837
+ {
838
+ "epoch": 1.6451612903225805,
839
+ "grad_norm": 0.5004581821839913,
840
+ "learning_rate": 5e-06,
841
+ "loss": 0.3523,
842
+ "step": 119
843
+ },
844
+ {
845
+ "epoch": 1.6589861751152073,
846
+ "grad_norm": 0.555371258631358,
847
+ "learning_rate": 4.919034655987493e-06,
848
+ "loss": 0.3977,
849
+ "step": 120
850
+ },
851
+ {
852
+ "epoch": 1.672811059907834,
853
+ "grad_norm": 0.6223674774033777,
854
+ "learning_rate": 4.838090543733222e-06,
855
+ "loss": 0.4122,
856
+ "step": 121
857
+ },
858
+ {
859
+ "epoch": 1.6866359447004609,
860
+ "grad_norm": 0.49317572687267214,
861
+ "learning_rate": 4.757188889427761e-06,
862
+ "loss": 0.3936,
863
+ "step": 122
864
+ },
865
+ {
866
+ "epoch": 1.7004608294930876,
867
+ "grad_norm": 0.5116997857144775,
868
+ "learning_rate": 4.6763509081278215e-06,
869
+ "loss": 0.3812,
870
+ "step": 123
871
+ },
872
+ {
873
+ "epoch": 1.7142857142857144,
874
+ "grad_norm": 0.5050661891764184,
875
+ "learning_rate": 4.59559779819298e-06,
876
+ "loss": 0.3744,
877
+ "step": 124
878
+ },
879
+ {
880
+ "epoch": 1.728110599078341,
881
+ "grad_norm": 0.5660294368372862,
882
+ "learning_rate": 4.51495073572676e-06,
883
+ "loss": 0.4133,
884
+ "step": 125
885
+ },
886
+ {
887
+ "epoch": 1.7419354838709677,
888
+ "grad_norm": 0.5414753484738278,
889
+ "learning_rate": 4.434430869023579e-06,
890
+ "loss": 0.4059,
891
+ "step": 126
892
+ },
893
+ {
894
+ "epoch": 1.7557603686635943,
895
+ "grad_norm": 0.5274523827704818,
896
+ "learning_rate": 4.3540593130229695e-06,
897
+ "loss": 0.4058,
898
+ "step": 127
899
+ },
900
+ {
901
+ "epoch": 1.769585253456221,
902
+ "grad_norm": 0.48723665989341286,
903
+ "learning_rate": 4.27385714377255e-06,
904
+ "loss": 0.4261,
905
+ "step": 128
906
+ },
907
+ {
908
+ "epoch": 1.7834101382488479,
909
+ "grad_norm": 0.5557410469617062,
910
+ "learning_rate": 4.1938453929012014e-06,
911
+ "loss": 0.3799,
912
+ "step": 129
913
+ },
914
+ {
915
+ "epoch": 1.7972350230414746,
916
+ "grad_norm": 0.5628455615899114,
917
+ "learning_rate": 4.1140450421038865e-06,
918
+ "loss": 0.4027,
919
+ "step": 130
920
+ },
921
+ {
922
+ "epoch": 1.8110599078341014,
923
+ "grad_norm": 0.5407292854180369,
924
+ "learning_rate": 4.034477017639561e-06,
925
+ "loss": 0.3414,
926
+ "step": 131
927
+ },
928
+ {
929
+ "epoch": 1.8248847926267282,
930
+ "grad_norm": 0.5110854563369478,
931
+ "learning_rate": 3.955162184843625e-06,
932
+ "loss": 0.4054,
933
+ "step": 132
934
+ },
935
+ {
936
+ "epoch": 1.838709677419355,
937
+ "grad_norm": 0.573230524328855,
938
+ "learning_rate": 3.8761213426563546e-06,
939
+ "loss": 0.4111,
940
+ "step": 133
941
+ },
942
+ {
943
+ "epoch": 1.8525345622119815,
944
+ "grad_norm": 0.5266763198269163,
945
+ "learning_rate": 3.7973752181687336e-06,
946
+ "loss": 0.3725,
947
+ "step": 134
948
+ },
949
+ {
950
+ "epoch": 1.8663594470046083,
951
+ "grad_norm": 0.5053510634901648,
952
+ "learning_rate": 3.7189444611871383e-06,
953
+ "loss": 0.3989,
954
+ "step": 135
955
+ },
956
+ {
957
+ "epoch": 1.8801843317972349,
958
+ "grad_norm": 0.5767131796506803,
959
+ "learning_rate": 3.6408496388182857e-06,
960
+ "loss": 0.456,
961
+ "step": 136
962
+ },
963
+ {
964
+ "epoch": 1.8940092165898617,
965
+ "grad_norm": 0.4875050911999597,
966
+ "learning_rate": 3.5631112300758595e-06,
967
+ "loss": 0.3442,
968
+ "step": 137
969
+ },
970
+ {
971
+ "epoch": 1.9078341013824884,
972
+ "grad_norm": 0.5145298101038916,
973
+ "learning_rate": 3.4857496205102475e-06,
974
+ "loss": 0.4418,
975
+ "step": 138
976
+ },
977
+ {
978
+ "epoch": 1.9216589861751152,
979
+ "grad_norm": 0.46196122478440654,
980
+ "learning_rate": 3.4087850968627823e-06,
981
+ "loss": 0.3882,
982
+ "step": 139
983
+ },
984
+ {
985
+ "epoch": 1.935483870967742,
986
+ "grad_norm": 0.4906484173808876,
987
+ "learning_rate": 3.3322378417458985e-06,
988
+ "loss": 0.3947,
989
+ "step": 140
990
+ },
991
+ {
992
+ "epoch": 1.9493087557603688,
993
+ "grad_norm": 0.4989650080684894,
994
+ "learning_rate": 3.2561279283505888e-06,
995
+ "loss": 0.3399,
996
+ "step": 141
997
+ },
998
+ {
999
+ "epoch": 1.9631336405529956,
1000
+ "grad_norm": 0.46237620183821815,
1001
+ "learning_rate": 3.180475315182563e-06,
1002
+ "loss": 0.405,
1003
+ "step": 142
1004
+ },
1005
+ {
1006
+ "epoch": 1.976958525345622,
1007
+ "grad_norm": 0.4637781556805509,
1008
+ "learning_rate": 3.1052998408284664e-06,
1009
+ "loss": 0.4145,
1010
+ "step": 143
1011
+ },
1012
+ {
1013
+ "epoch": 1.9907834101382489,
1014
+ "grad_norm": 0.4775120449168806,
1015
+ "learning_rate": 3.0306212187535653e-06,
1016
+ "loss": 0.3988,
1017
+ "step": 144
1018
+ },
1019
+ {
1020
+ "epoch": 2.0046082949308754,
1021
+ "grad_norm": 0.9379207114350527,
1022
+ "learning_rate": 2.9564590321322206e-06,
1023
+ "loss": 0.6161,
1024
+ "step": 145
1025
+ },
1026
+ {
1027
+ "epoch": 2.0184331797235022,
1028
+ "grad_norm": 0.4763284552212227,
1029
+ "learning_rate": 2.882832728712551e-06,
1030
+ "loss": 0.3328,
1031
+ "step": 146
1032
+ },
1033
+ {
1034
+ "epoch": 2.032258064516129,
1035
+ "grad_norm": 0.5068219788042055,
1036
+ "learning_rate": 2.8097616157165886e-06,
1037
+ "loss": 0.3514,
1038
+ "step": 147
1039
+ },
1040
+ {
1041
+ "epoch": 2.046082949308756,
1042
+ "grad_norm": 0.5004585471417805,
1043
+ "learning_rate": 2.7372648547773063e-06,
1044
+ "loss": 0.3663,
1045
+ "step": 148
1046
+ },
1047
+ {
1048
+ "epoch": 2.0599078341013826,
1049
+ "grad_norm": 0.48776506523697694,
1050
+ "learning_rate": 2.665361456913797e-06,
1051
+ "loss": 0.3557,
1052
+ "step": 149
1053
+ },
1054
+ {
1055
+ "epoch": 2.0737327188940093,
1056
+ "grad_norm": 0.4565469138842312,
1057
+ "learning_rate": 2.594070277545975e-06,
1058
+ "loss": 0.3325,
1059
+ "step": 150
1060
+ },
1061
+ {
1062
+ "epoch": 2.087557603686636,
1063
+ "grad_norm": 0.5109481857152307,
1064
+ "learning_rate": 2.5234100115500643e-06,
1065
+ "loss": 0.3723,
1066
+ "step": 151
1067
+ },
1068
+ {
1069
+ "epoch": 2.1013824884792625,
1070
+ "grad_norm": 0.503915544983321,
1071
+ "learning_rate": 2.4533991883561868e-06,
1072
+ "loss": 0.3454,
1073
+ "step": 152
1074
+ },
1075
+ {
1076
+ "epoch": 2.1152073732718892,
1077
+ "grad_norm": 0.4920495646406938,
1078
+ "learning_rate": 2.38405616708935e-06,
1079
+ "loss": 0.3491,
1080
+ "step": 153
1081
+ },
1082
+ {
1083
+ "epoch": 2.129032258064516,
1084
+ "grad_norm": 0.46249250593080354,
1085
+ "learning_rate": 2.315399131755081e-06,
1086
+ "loss": 0.3714,
1087
+ "step": 154
1088
+ },
1089
+ {
1090
+ "epoch": 2.142857142857143,
1091
+ "grad_norm": 0.4756209372804288,
1092
+ "learning_rate": 2.2474460864709825e-06,
1093
+ "loss": 0.341,
1094
+ "step": 155
1095
+ },
1096
+ {
1097
+ "epoch": 2.1566820276497696,
1098
+ "grad_norm": 0.49448121204775364,
1099
+ "learning_rate": 2.1802148507454675e-06,
1100
+ "loss": 0.3362,
1101
+ "step": 156
1102
+ },
1103
+ {
1104
+ "epoch": 2.1705069124423964,
1105
+ "grad_norm": 0.49087318600689567,
1106
+ "learning_rate": 2.1137230548049042e-06,
1107
+ "loss": 0.3712,
1108
+ "step": 157
1109
+ },
1110
+ {
1111
+ "epoch": 2.184331797235023,
1112
+ "grad_norm": 0.4394130757578322,
1113
+ "learning_rate": 2.0479881349703885e-06,
1114
+ "loss": 0.3289,
1115
+ "step": 158
1116
+ },
1117
+ {
1118
+ "epoch": 2.19815668202765,
1119
+ "grad_norm": 0.49665520195966556,
1120
+ "learning_rate": 1.983027329085377e-06,
1121
+ "loss": 0.3648,
1122
+ "step": 159
1123
+ },
1124
+ {
1125
+ "epoch": 2.2119815668202767,
1126
+ "grad_norm": 0.4113878764187301,
1127
+ "learning_rate": 1.9188576719953635e-06,
1128
+ "loss": 0.2849,
1129
+ "step": 160
1130
+ },
1131
+ {
1132
+ "epoch": 2.225806451612903,
1133
+ "grad_norm": 0.4999778833865041,
1134
+ "learning_rate": 1.8554959910807773e-06,
1135
+ "loss": 0.3695,
1136
+ "step": 161
1137
+ },
1138
+ {
1139
+ "epoch": 2.23963133640553,
1140
+ "grad_norm": 0.5189860690853264,
1141
+ "learning_rate": 1.7929589018443016e-06,
1142
+ "loss": 0.3504,
1143
+ "step": 162
1144
+ },
1145
+ {
1146
+ "epoch": 2.2534562211981566,
1147
+ "grad_norm": 0.48043530531701206,
1148
+ "learning_rate": 1.7312628035537388e-06,
1149
+ "loss": 0.332,
1150
+ "step": 163
1151
+ },
1152
+ {
1153
+ "epoch": 2.2672811059907834,
1154
+ "grad_norm": 0.5324429864403901,
1155
+ "learning_rate": 1.6704238749415958e-06,
1156
+ "loss": 0.3387,
1157
+ "step": 164
1158
+ },
1159
+ {
1160
+ "epoch": 2.28110599078341,
1161
+ "grad_norm": 0.5029628675101937,
1162
+ "learning_rate": 1.6104580699624839e-06,
1163
+ "loss": 0.3592,
1164
+ "step": 165
1165
+ },
1166
+ {
1167
+ "epoch": 2.294930875576037,
1168
+ "grad_norm": 0.45804507081769336,
1169
+ "learning_rate": 1.5513811136094786e-06,
1170
+ "loss": 0.3063,
1171
+ "step": 166
1172
+ },
1173
+ {
1174
+ "epoch": 2.3087557603686637,
1175
+ "grad_norm": 0.4647295250952741,
1176
+ "learning_rate": 1.4932084977905043e-06,
1177
+ "loss": 0.3254,
1178
+ "step": 167
1179
+ },
1180
+ {
1181
+ "epoch": 2.3225806451612905,
1182
+ "grad_norm": 0.46453643921934695,
1183
+ "learning_rate": 1.4359554772658551e-06,
1184
+ "loss": 0.3774,
1185
+ "step": 168
1186
+ },
1187
+ {
1188
+ "epoch": 2.3364055299539173,
1189
+ "grad_norm": 0.45343468728467234,
1190
+ "learning_rate": 1.3796370656478936e-06,
1191
+ "loss": 0.3325,
1192
+ "step": 169
1193
+ },
1194
+ {
1195
+ "epoch": 2.3502304147465436,
1196
+ "grad_norm": 0.4732479977172983,
1197
+ "learning_rate": 1.3242680314639995e-06,
1198
+ "loss": 0.3373,
1199
+ "step": 170
1200
+ },
1201
+ {
1202
+ "epoch": 2.3640552995391704,
1203
+ "grad_norm": 0.47208675924851334,
1204
+ "learning_rate": 1.2698628942837698e-06,
1205
+ "loss": 0.3427,
1206
+ "step": 171
1207
+ },
1208
+ {
1209
+ "epoch": 2.377880184331797,
1210
+ "grad_norm": 0.4454283369984702,
1211
+ "learning_rate": 1.2164359209115235e-06,
1212
+ "loss": 0.3257,
1213
+ "step": 172
1214
+ },
1215
+ {
1216
+ "epoch": 2.391705069124424,
1217
+ "grad_norm": 0.49273795207826715,
1218
+ "learning_rate": 1.164001121645069e-06,
1219
+ "loss": 0.3772,
1220
+ "step": 173
1221
+ },
1222
+ {
1223
+ "epoch": 2.4055299539170507,
1224
+ "grad_norm": 0.46302965066920987,
1225
+ "learning_rate": 1.1125722466017547e-06,
1226
+ "loss": 0.326,
1227
+ "step": 174
1228
+ },
1229
+ {
1230
+ "epoch": 2.4193548387096775,
1231
+ "grad_norm": 0.44068332388009435,
1232
+ "learning_rate": 1.062162782112729e-06,
1233
+ "loss": 0.3947,
1234
+ "step": 175
1235
+ },
1236
+ {
1237
+ "epoch": 2.4331797235023043,
1238
+ "grad_norm": 0.41223830858735033,
1239
+ "learning_rate": 1.012785947186397e-06,
1240
+ "loss": 0.3027,
1241
+ "step": 176
1242
+ },
1243
+ {
1244
+ "epoch": 2.447004608294931,
1245
+ "grad_norm": 0.44346734726903453,
1246
+ "learning_rate": 9.644546900419533e-07,
1247
+ "loss": 0.3306,
1248
+ "step": 177
1249
+ },
1250
+ {
1251
+ "epoch": 2.460829493087558,
1252
+ "grad_norm": 0.45992313059173967,
1253
+ "learning_rate": 9.171816847139447e-07,
1254
+ "loss": 0.3596,
1255
+ "step": 178
1256
+ },
1257
+ {
1258
+ "epoch": 2.474654377880184,
1259
+ "grad_norm": 0.4558549067269223,
1260
+ "learning_rate": 8.709793277287182e-07,
1261
+ "loss": 0.3684,
1262
+ "step": 179
1263
+ },
1264
+ {
1265
+ "epoch": 2.488479262672811,
1266
+ "grad_norm": 0.41803380966992537,
1267
+ "learning_rate": 8.258597348536452e-07,
1268
+ "loss": 0.32,
1269
+ "step": 180
1270
+ },
1271
+ {
1272
+ "epoch": 2.5023041474654377,
1273
+ "grad_norm": 0.42403429172582247,
1274
+ "learning_rate": 7.818347379199781e-07,
1275
+ "loss": 0.3273,
1276
+ "step": 181
1277
+ },
1278
+ {
1279
+ "epoch": 2.5161290322580645,
1280
+ "grad_norm": 0.46169239465826656,
1281
+ "learning_rate": 7.389158817201541e-07,
1282
+ "loss": 0.3313,
1283
+ "step": 182
1284
+ },
1285
+ {
1286
+ "epoch": 2.5299539170506913,
1287
+ "grad_norm": 0.49336488496655645,
1288
+ "learning_rate": 6.971144209803738e-07,
1289
+ "loss": 0.369,
1290
+ "step": 183
1291
+ },
1292
+ {
1293
+ "epoch": 2.543778801843318,
1294
+ "grad_norm": 0.47648081907983714,
1295
+ "learning_rate": 6.564413174092443e-07,
1296
+ "loss": 0.3464,
1297
+ "step": 184
1298
+ },
1299
+ {
1300
+ "epoch": 2.557603686635945,
1301
+ "grad_norm": 0.4151223124023637,
1302
+ "learning_rate": 6.16907236823262e-07,
1303
+ "loss": 0.3334,
1304
+ "step": 185
1305
+ },
1306
+ {
1307
+ "epoch": 2.571428571428571,
1308
+ "grad_norm": 0.4214581402580324,
1309
+ "learning_rate": 5.785225463498828e-07,
1310
+ "loss": 0.3336,
1311
+ "step": 186
1312
+ },
1313
+ {
1314
+ "epoch": 2.5852534562211984,
1315
+ "grad_norm": 0.4404488936158401,
1316
+ "learning_rate": 5.412973117089288e-07,
1317
+ "loss": 0.357,
1318
+ "step": 187
1319
+ },
1320
+ {
1321
+ "epoch": 2.5990783410138247,
1322
+ "grad_norm": 0.4694197953797554,
1323
+ "learning_rate": 5.05241294573024e-07,
1324
+ "loss": 0.3389,
1325
+ "step": 188
1326
+ },
1327
+ {
1328
+ "epoch": 2.6129032258064515,
1329
+ "grad_norm": 0.43224485913576827,
1330
+ "learning_rate": 4.7036395000776556e-07,
1331
+ "loss": 0.3816,
1332
+ "step": 189
1333
+ },
1334
+ {
1335
+ "epoch": 2.6267281105990783,
1336
+ "grad_norm": 0.46915895400007485,
1337
+ "learning_rate": 4.3667442399229985e-07,
1338
+ "loss": 0.3212,
1339
+ "step": 190
1340
+ },
1341
+ {
1342
+ "epoch": 2.640552995391705,
1343
+ "grad_norm": 0.41455265232398864,
1344
+ "learning_rate": 4.041815510209396e-07,
1345
+ "loss": 0.349,
1346
+ "step": 191
1347
+ },
1348
+ {
1349
+ "epoch": 2.654377880184332,
1350
+ "grad_norm": 0.4138588542457796,
1351
+ "learning_rate": 3.728938517864794e-07,
1352
+ "loss": 0.3155,
1353
+ "step": 192
1354
+ },
1355
+ {
1356
+ "epoch": 2.6682027649769586,
1357
+ "grad_norm": 0.44949477525662346,
1358
+ "learning_rate": 3.4281953094578877e-07,
1359
+ "loss": 0.3608,
1360
+ "step": 193
1361
+ },
1362
+ {
1363
+ "epoch": 2.6820276497695854,
1364
+ "grad_norm": 0.45077480983651,
1365
+ "learning_rate": 3.1396647496828245e-07,
1366
+ "loss": 0.366,
1367
+ "step": 194
1368
+ },
1369
+ {
1370
+ "epoch": 2.6958525345622117,
1371
+ "grad_norm": 0.43977571760655004,
1372
+ "learning_rate": 2.8634225006782867e-07,
1373
+ "loss": 0.3447,
1374
+ "step": 195
1375
+ },
1376
+ {
1377
+ "epoch": 2.709677419354839,
1378
+ "grad_norm": 0.4531801768047782,
1379
+ "learning_rate": 2.599541002186479e-07,
1380
+ "loss": 0.3791,
1381
+ "step": 196
1382
+ },
1383
+ {
1384
+ "epoch": 2.7235023041474653,
1385
+ "grad_norm": 0.40108250009956564,
1386
+ "learning_rate": 2.3480894525569564e-07,
1387
+ "loss": 0.303,
1388
+ "step": 197
1389
+ },
1390
+ {
1391
+ "epoch": 2.737327188940092,
1392
+ "grad_norm": 0.40980648885083565,
1393
+ "learning_rate": 2.109133790600648e-07,
1394
+ "loss": 0.3427,
1395
+ "step": 198
1396
+ },
1397
+ {
1398
+ "epoch": 2.751152073732719,
1399
+ "grad_norm": 0.41297927531361595,
1400
+ "learning_rate": 1.8827366782984913e-07,
1401
+ "loss": 0.325,
1402
+ "step": 199
1403
+ },
1404
+ {
1405
+ "epoch": 2.7649769585253456,
1406
+ "grad_norm": 0.429935038739435,
1407
+ "learning_rate": 1.6689574843694433e-07,
1408
+ "loss": 0.3852,
1409
+ "step": 200
1410
+ },
1411
+ {
1412
+ "epoch": 2.7788018433179724,
1413
+ "grad_norm": 0.4020104101913433,
1414
+ "learning_rate": 1.4678522687020414e-07,
1415
+ "loss": 0.3513,
1416
+ "step": 201
1417
+ },
1418
+ {
1419
+ "epoch": 2.792626728110599,
1420
+ "grad_norm": 0.39219309859175616,
1421
+ "learning_rate": 1.2794737676536993e-07,
1422
+ "loss": 0.331,
1423
+ "step": 202
1424
+ },
1425
+ {
1426
+ "epoch": 2.806451612903226,
1427
+ "grad_norm": 0.4097391288745776,
1428
+ "learning_rate": 1.1038713802214718e-07,
1429
+ "loss": 0.3411,
1430
+ "step": 203
1431
+ },
1432
+ {
1433
+ "epoch": 2.8202764976958523,
1434
+ "grad_norm": 0.4309769162405172,
1435
+ "learning_rate": 9.410911550880474e-08,
1436
+ "loss": 0.3244,
1437
+ "step": 204
1438
+ },
1439
+ {
1440
+ "epoch": 2.8341013824884795,
1441
+ "grad_norm": 0.4449894760218027,
1442
+ "learning_rate": 7.911757785462882e-08,
1443
+ "loss": 0.3417,
1444
+ "step": 205
1445
+ },
1446
+ {
1447
+ "epoch": 2.847926267281106,
1448
+ "grad_norm": 0.4303644153909974,
1449
+ "learning_rate": 6.54164563305465e-08,
1450
+ "loss": 0.3391,
1451
+ "step": 206
1452
+ },
1453
+ {
1454
+ "epoch": 2.8617511520737327,
1455
+ "grad_norm": 0.42194746880922407,
1456
+ "learning_rate": 5.3009343818219985e-08,
1457
+ "loss": 0.3707,
1458
+ "step": 207
1459
+ },
1460
+ {
1461
+ "epoch": 2.8755760368663594,
1462
+ "grad_norm": 0.4196594211589896,
1463
+ "learning_rate": 4.189949386787462e-08,
1464
+ "loss": 0.351,
1465
+ "step": 208
1466
+ },
1467
+ {
1468
+ "epoch": 2.889400921658986,
1469
+ "grad_norm": 0.40539018280598516,
1470
+ "learning_rate": 3.2089819845111946e-08,
1471
+ "loss": 0.3168,
1472
+ "step": 209
1473
+ },
1474
+ {
1475
+ "epoch": 2.903225806451613,
1476
+ "grad_norm": 0.4658694331536401,
1477
+ "learning_rate": 2.358289416693027e-08,
1478
+ "loss": 0.4198,
1479
+ "step": 210
1480
+ },
1481
+ {
1482
+ "epoch": 2.9170506912442398,
1483
+ "grad_norm": 0.41736837046748076,
1484
+ "learning_rate": 1.6380947627153143e-08,
1485
+ "loss": 0.3174,
1486
+ "step": 211
1487
+ },
1488
+ {
1489
+ "epoch": 2.9308755760368665,
1490
+ "grad_norm": 0.42675226026620394,
1491
+ "learning_rate": 1.0485868811441757e-08,
1492
+ "loss": 0.3538,
1493
+ "step": 212
1494
+ },
1495
+ {
1496
+ "epoch": 2.944700460829493,
1497
+ "grad_norm": 0.4160647077774905,
1498
+ "learning_rate": 5.899203602046655e-09,
1499
+ "loss": 0.3474,
1500
+ "step": 213
1501
+ },
1502
+ {
1503
+ "epoch": 2.9585253456221197,
1504
+ "grad_norm": 0.40018027901405295,
1505
+ "learning_rate": 2.6221547724253337e-09,
1506
+ "loss": 0.319,
1507
+ "step": 214
1508
+ },
1509
+ {
1510
+ "epoch": 2.9723502304147464,
1511
+ "grad_norm": 0.4334075522903773,
1512
+ "learning_rate": 6.555816718389896e-10,
1513
+ "loss": 0.3884,
1514
+ "step": 215
1515
+ },
1516
+ {
1517
+ "epoch": 2.986175115207373,
1518
+ "grad_norm": 0.4031378328194389,
1519
+ "learning_rate": 0.0,
1520
+ "loss": 0.331,
1521
+ "step": 216
1522
+ },
1523
+ {
1524
+ "epoch": 2.986175115207373,
1525
+ "step": 216,
1526
+ "total_flos": 130743839866880.0,
1527
+ "train_loss": 0.4320466773653472,
1528
+ "train_runtime": 2638.4795,
1529
+ "train_samples_per_second": 7.869,
1530
+ "train_steps_per_second": 0.082
1531
  }
1532
  ],
1533
+ "logging_steps": 1,
1534
+ "max_steps": 216,
1535
  "num_input_tokens_seen": 0,
1536
  "num_train_epochs": 3,
1537
  "save_steps": 500,
 
1547
  "attributes": {}
1548
  }
1549
  },
1550
+ "total_flos": 130743839866880.0,
1551
+ "train_batch_size": 1,
1552
  "trial_name": null,
1553
  "trial_params": null
1554
  }
training_loss.png CHANGED