Hugging Face's logo

yuyijiong
/
Qwen3-SWA-adaptation

Text Generation
Transformers
Safetensors
English
Chinese
Model card Files
xet
Community
1
yuyijiong commited on
Commit
ec2daa8
·
verified ·
1 Parent(s): a9c4e12

Add files using upload-large-folder tool

Browse files
This view is limited to 50 files because it contains too many changes.   See raw diff
Files changed (50) hide show
  1. .gitattributes +14 -0
  2. Qwen3-30B-A3B-Instruct-2507-sft-fusang-swa=2k_sink=0/checkpoint-870/added_tokens.json +28 -0
  3. Qwen3-30B-A3B-Instruct-2507-sft-fusang-swa=2k_sink=0/checkpoint-870/merges.txt +0 -0
  4. Qwen3-30B-A3B-Instruct-2507-sft-fusang-swa=2k_sink=0/checkpoint-870/special_tokens_map.json +31 -0
  5. Qwen3-30B-A3B-Instruct-2507-sft-fusang-swa=2k_sink=0/checkpoint-870/tokenizer_config.json +239 -0
  6. Qwen3-30B-A3B-Instruct-2507-sft-fusang-swa=2k_sink=0/checkpoint-870/trainer_state.json +1252 -0
  7. Qwen3-30B-A3B-Instruct-2507-sft-fusang-swa=2k_sink=0/checkpoint-870/vocab.json +0 -0
  8. Qwen3-30B-A3B-Instruct-2507-sft-fusang-swa=2k_sink=0_falayer=25_from0_step2_fadec/checkpoint-994/adapter_config.json +38 -0
  9. Qwen3-30B-A3B-Instruct-2507-sft-fusang-swa=2k_sink=0_falayer=25_from0_step2_fadec/checkpoint-994/tokenizer.json +3 -0
  10. Qwen3-30B-A3B-Instruct-2507-sft-fusang-swa=2k_sink=0_falayer=25_from0_step2_fadec/checkpoint-994/vocab.json +0 -0
  11. Qwen3-30B-A3B-Instruct-2507-sft-fusang-swa=2k_sink=100_fadec/checkpoint-870/adapter_config.json +38 -0
  12. Qwen3-30B-A3B-Instruct-2507-sft-fusang-swa=2k_sink=100_fadec/checkpoint-870/chat_template.jinja +61 -0
  13. Qwen3-30B-A3B-Instruct-2507-sft-fusang-swa=2k_sink=100_fadec/checkpoint-870/special_tokens_map.json +31 -0
  14. Qwen3-30B-A3B-Instruct-2507-sft-fusang-swa=2k_sink=100_fadec/checkpoint-870/tokenizer.json +3 -0
  15. Qwen3-30B-A3B-Instruct-2507-sft-fusang-swa=2k_sink=100_fadec/checkpoint-870/tokenizer_config.json +239 -0
  16. Qwen3-30B-A3B-Instruct-2507-sft-fusang-swa=2k_sink=100_fadec/checkpoint-870/trainer_state.json +1252 -0
  17. Qwen3-30B-A3B-Instruct-2507-sft-fusang-swa=2k_sink=100_fadec/checkpoint-870/vocab.json +0 -0
  18. Qwen3-30B-A3B-Instruct-2507-sft-fusang-swa=2k_sink=100_falayer=25_from0_step2_fadec/checkpoint-870/README.md +207 -0
  19. Qwen3-30B-A3B-Instruct-2507-sft-fusang-swa=2k_sink=100_falayer=25_from0_step2_fadec/checkpoint-870/merges.txt +0 -0
  20. Qwen3-30B-A3B-Instruct-2507-sft-fusang-swa=2k_sink=100_falayer=25_from0_step2_fadec/checkpoint-870/special_tokens_map.json +31 -0
  21. Qwen3-30B-A3B-Instruct-2507-sft-fusang-swa=2k_sink=100_falayer=25_from0_step2_fadec/checkpoint-870/tokenizer_config.json +239 -0
  22. Qwen3-30B-A3B-Instruct-2507-sft-fusang-swa=2k_sink=100_falayer=25_from0_step2_fadec/checkpoint-870/trainer_state.json +1252 -0
  23. Qwen3-30B-A3B-Instruct-2507-sft-fusang-swa=2k_sink=100_falayer=25_from0_step2_fadec/checkpoint-870/training_args.bin +3 -0
  24. Qwen3-30B-A3B-Instruct-2507-sft-fusang-swa=2k_sink=100_falayer=25_from0_step2_fadec/checkpoint-870/vocab.json +0 -0
  25. Qwen3-30B-A3B-Thinking-2507-sft-fusang-swa=2k_sink=0/checkpoint-862/tokenizer.json +3 -0
  26. Qwen3-30B-A3B-Thinking-2507-sft-fusang-swa=2k_sink=100/checkpoint-862/training_args.bin +3 -0
  27. Qwen3-30B-A3B-Thinking-2507-sft-fusang-swa=2k_sink=100_falayer=25_from0_step2_fadec/checkpoint-862/merges.txt +0 -0
  28. Qwen3-30B-A3B-Thinking-2507-sft-fusang-swa=2k_sink=100_falayer=25_from0_step2_fadec/checkpoint-862/vocab.json +0 -0
  29. Qwen3-30B-A3B-Thinking-2507-sft-fusang/checkpoint-862/adapter_config.json +38 -0
  30. Qwen3-30B-A3B-Thinking-2507-sft-fusang/checkpoint-862/special_tokens_map.json +31 -0
  31. Qwen3-4B-Instruct-2507-sft-fusang-swa=2k_sink=0/checkpoint-1302/tokenizer.json +3 -0
  32. Qwen3-4B-Instruct-2507-sft-fusang-swa=2k_sink=0/checkpoint-1736/tokenizer.json +3 -0
  33. Qwen3-4B-Instruct-2507-sft-fusang-swa=2k_sink=0_fadec/checkpoint-1760/tokenizer.json +3 -0
  34. Qwen3-4B-Instruct-2507-sft-fusang-swa=2k_sink=0_fadec/checkpoint-1760/trainer_state.json +2498 -0
  35. Qwen3-4B-Instruct-2507-sft-fusang-swa=2k_sink=0_falayer=13_from1_step4_fadec/checkpoint-921/chat_template.jinja +61 -0
  36. Qwen3-4B-Instruct-2507-sft-fusang-swa=2k_sink=0_falayer=13_from1_step4_fadec/checkpoint-921/trainer_state.json +1322 -0
  37. Qwen3-4B-Instruct-2507-sft-fusang-swa=2k_sink=0_falayer=25_from1_step2/checkpoint-921/tokenizer.json +3 -0
  38. Qwen3-4B-Instruct-2507-sft-fusang-swa=2k_sink=0_falayer=25_from1_step2/checkpoint-921/trainer_state.json +1322 -0
  39. Qwen3-4B-Instruct-2507-sft-fusang-swa=2k_sink=0_falayer=7_from0_step8_fadec/checkpoint-1085/README.md +207 -0
  40. Qwen3-4B-Instruct-2507-sft-fusang-swa=2k_sink=0_falayer=7_from0_step8_fadec/checkpoint-1085/merges.txt +0 -0
  41. Qwen3-4B-Instruct-2507-sft-fusang-swa=2k_sink=0_falayer=7_from0_step8_fadec/checkpoint-1085/trainer_state.json +1553 -0
  42. Qwen3-4B-Instruct-2507-sft-fusang-swa=2k_sink=0_falayer=7_from0_step8_fadec/checkpoint-1085/vocab.json +0 -0
  43. Qwen3-4B-Instruct-2507-sft-fusang-swa=2k_sink=10/checkpoint-1760/chat_template.jinja +61 -0
  44. Qwen3-4B-Instruct-2507-sft-fusang-swa=2k_sink=10/checkpoint-1760/tokenizer.json +3 -0
  45. Qwen3-4B-Instruct-2507-sft-fusang-swa=2k_sink=100/checkpoint-1760/tokenizer.json +3 -0
  46. Qwen3-4B-Instruct-2507-sft-fusang-swa=2k_sink=100_falayer=25_from1_step2_fadec/checkpoint-921/tokenizer.json +3 -0
  47. Qwen3-4B-Instruct-2507-sft-fusang-swa=2k_sink=100_falayer=25_from1_step2_fadec/checkpoint-921/training_args.bin +3 -0
  48. Qwen3-4B-Instruct-2507-sft-fusang/checkpoint-1760/added_tokens.json +28 -0
  49. Qwen3-4B-Instruct-2507-sft-fusang/checkpoint-1760/special_tokens_map.json +31 -0
  50. Qwen3-4B-Instruct-2507-sft-fusang/checkpoint-1760/tokenizer_config.json +239 -0
.gitattributes CHANGED
@@ -33,3 +33,17 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
33
  *.zip filter=lfs diff=lfs merge=lfs -text
34
  *.zst filter=lfs diff=lfs merge=lfs -text
35
  *tfevents* filter=lfs diff=lfs merge=lfs -text
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
33
  *.zip filter=lfs diff=lfs merge=lfs -text
34
  *.zst filter=lfs diff=lfs merge=lfs -text
35
  *tfevents* filter=lfs diff=lfs merge=lfs -text
36
+ Qwen3-4B-Thinking-2507-sft-fusang-swa=2k_sink=0_falayer=13_from1_step4/checkpoint-1771/tokenizer.json filter=lfs diff=lfs merge=lfs -text
37
+ Qwen3-30B-A3B-Thinking-2507-sft-fusang-swa=2k_sink=0/checkpoint-862/tokenizer.json filter=lfs diff=lfs merge=lfs -text
38
+ Qwen3-4B-Thinking-2507-sft-fusang-swa=2k_sink=0/checkpoint-8855/tokenizer.json filter=lfs diff=lfs merge=lfs -text
39
+ Qwen3-4B-Thinking-2507-sft-fusang-swa=2k_sink=100/checkpoint-1771/tokenizer.json filter=lfs diff=lfs merge=lfs -text
40
+ Qwen3-4B-Instruct-2507-sft-fusang-swa=2k_sink=0_falayer=25_from1_step2/checkpoint-921/tokenizer.json filter=lfs diff=lfs merge=lfs -text
41
+ Qwen3-4B-Thinking-2507-sft-fusang-swa=2k_sink=0_falayer=13_from0_step4_fadec/checkpoint-1771/tokenizer.json filter=lfs diff=lfs merge=lfs -text
42
+ Qwen3-4B-Instruct-2507-sft-fusang-swa=2k_sink=0/checkpoint-1736/tokenizer.json filter=lfs diff=lfs merge=lfs -text
43
+ Qwen3-4B-Instruct-2507-sft-fusang-swa=2k_sink=100_falayer=25_from1_step2_fadec/checkpoint-921/tokenizer.json filter=lfs diff=lfs merge=lfs -text
44
+ Qwen3-4B-Instruct-2507-sft-fusang-swa=2k_sink=0/checkpoint-1302/tokenizer.json filter=lfs diff=lfs merge=lfs -text
45
+ Qwen3-4B-Instruct-2507-sft-fusang-swa=2k_sink=100/checkpoint-1760/tokenizer.json filter=lfs diff=lfs merge=lfs -text
46
+ Qwen3-30B-A3B-Instruct-2507-sft-fusang-swa=2k_sink=100_fadec/checkpoint-870/tokenizer.json filter=lfs diff=lfs merge=lfs -text
47
+ Qwen3-30B-A3B-Instruct-2507-sft-fusang-swa=2k_sink=0_falayer=25_from0_step2_fadec/checkpoint-994/tokenizer.json filter=lfs diff=lfs merge=lfs -text
48
+ Qwen3-4B-Instruct-2507-sft-fusang-swa=2k_sink=10/checkpoint-1760/tokenizer.json filter=lfs diff=lfs merge=lfs -text
49
+ Qwen3-4B-Instruct-2507-sft-fusang-swa=2k_sink=0_fadec/checkpoint-1760/tokenizer.json filter=lfs diff=lfs merge=lfs -text
Qwen3-30B-A3B-Instruct-2507-sft-fusang-swa=2k_sink=0/checkpoint-870/added_tokens.json ADDED
@@ -0,0 +1,28 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "</think>": 151668,
3
+ "</tool_call>": 151658,
4
+ "</tool_response>": 151666,
5
+ "<think>": 151667,
6
+ "<tool_call>": 151657,
7
+ "<tool_response>": 151665,
8
+ "<|box_end|>": 151649,
9
+ "<|box_start|>": 151648,
10
+ "<|endoftext|>": 151643,
11
+ "<|file_sep|>": 151664,
12
+ "<|fim_middle|>": 151660,
13
+ "<|fim_pad|>": 151662,
14
+ "<|fim_prefix|>": 151659,
15
+ "<|fim_suffix|>": 151661,
16
+ "<|im_end|>": 151645,
17
+ "<|im_start|>": 151644,
18
+ "<|image_pad|>": 151655,
19
+ "<|object_ref_end|>": 151647,
20
+ "<|object_ref_start|>": 151646,
21
+ "<|quad_end|>": 151651,
22
+ "<|quad_start|>": 151650,
23
+ "<|repo_name|>": 151663,
24
+ "<|video_pad|>": 151656,
25
+ "<|vision_end|>": 151653,
26
+ "<|vision_pad|>": 151654,
27
+ "<|vision_start|>": 151652
28
+ }
Qwen3-30B-A3B-Instruct-2507-sft-fusang-swa=2k_sink=0/checkpoint-870/merges.txt ADDED
The diff for this file is too large to render. See raw diff
 
Qwen3-30B-A3B-Instruct-2507-sft-fusang-swa=2k_sink=0/checkpoint-870/special_tokens_map.json ADDED
@@ -0,0 +1,31 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "additional_special_tokens": [
3
+ "<|im_start|>",
4
+ "<|im_end|>",
5
+ "<|object_ref_start|>",
6
+ "<|object_ref_end|>",
7
+ "<|box_start|>",
8
+ "<|box_end|>",
9
+ "<|quad_start|>",
10
+ "<|quad_end|>",
11
+ "<|vision_start|>",
12
+ "<|vision_end|>",
13
+ "<|vision_pad|>",
14
+ "<|image_pad|>",
15
+ "<|video_pad|>"
16
+ ],
17
+ "eos_token": {
18
+ "content": "<|im_end|>",
19
+ "lstrip": false,
20
+ "normalized": false,
21
+ "rstrip": false,
22
+ "single_word": false
23
+ },
24
+ "pad_token": {
25
+ "content": "<|endoftext|>",
26
+ "lstrip": false,
27
+ "normalized": false,
28
+ "rstrip": false,
29
+ "single_word": false
30
+ }
31
+ }
Qwen3-30B-A3B-Instruct-2507-sft-fusang-swa=2k_sink=0/checkpoint-870/tokenizer_config.json ADDED
@@ -0,0 +1,239 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "add_bos_token": false,
3
+ "add_prefix_space": false,
4
+ "added_tokens_decoder": {
5
+ "151643": {
6
+ "content": "<|endoftext|>",
7
+ "lstrip": false,
8
+ "normalized": false,
9
+ "rstrip": false,
10
+ "single_word": false,
11
+ "special": true
12
+ },
13
+ "151644": {
14
+ "content": "<|im_start|>",
15
+ "lstrip": false,
16
+ "normalized": false,
17
+ "rstrip": false,
18
+ "single_word": false,
19
+ "special": true
20
+ },
21
+ "151645": {
22
+ "content": "<|im_end|>",
23
+ "lstrip": false,
24
+ "normalized": false,
25
+ "rstrip": false,
26
+ "single_word": false,
27
+ "special": true
28
+ },
29
+ "151646": {
30
+ "content": "<|object_ref_start|>",
31
+ "lstrip": false,
32
+ "normalized": false,
33
+ "rstrip": false,
34
+ "single_word": false,
35
+ "special": true
36
+ },
37
+ "151647": {
38
+ "content": "<|object_ref_end|>",
39
+ "lstrip": false,
40
+ "normalized": false,
41
+ "rstrip": false,
42
+ "single_word": false,
43
+ "special": true
44
+ },
45
+ "151648": {
46
+ "content": "<|box_start|>",
47
+ "lstrip": false,
48
+ "normalized": false,
49
+ "rstrip": false,
50
+ "single_word": false,
51
+ "special": true
52
+ },
53
+ "151649": {
54
+ "content": "<|box_end|>",
55
+ "lstrip": false,
56
+ "normalized": false,
57
+ "rstrip": false,
58
+ "single_word": false,
59
+ "special": true
60
+ },
61
+ "151650": {
62
+ "content": "<|quad_start|>",
63
+ "lstrip": false,
64
+ "normalized": false,
65
+ "rstrip": false,
66
+ "single_word": false,
67
+ "special": true
68
+ },
69
+ "151651": {
70
+ "content": "<|quad_end|>",
71
+ "lstrip": false,
72
+ "normalized": false,
73
+ "rstrip": false,
74
+ "single_word": false,
75
+ "special": true
76
+ },
77
+ "151652": {
78
+ "content": "<|vision_start|>",
79
+ "lstrip": false,
80
+ "normalized": false,
81
+ "rstrip": false,
82
+ "single_word": false,
83
+ "special": true
84
+ },
85
+ "151653": {
86
+ "content": "<|vision_end|>",
87
+ "lstrip": false,
88
+ "normalized": false,
89
+ "rstrip": false,
90
+ "single_word": false,
91
+ "special": true
92
+ },
93
+ "151654": {
94
+ "content": "<|vision_pad|>",
95
+ "lstrip": false,
96
+ "normalized": false,
97
+ "rstrip": false,
98
+ "single_word": false,
99
+ "special": true
100
+ },
101
+ "151655": {
102
+ "content": "<|image_pad|>",
103
+ "lstrip": false,
104
+ "normalized": false,
105
+ "rstrip": false,
106
+ "single_word": false,
107
+ "special": true
108
+ },
109
+ "151656": {
110
+ "content": "<|video_pad|>",
111
+ "lstrip": false,
112
+ "normalized": false,
113
+ "rstrip": false,
114
+ "single_word": false,
115
+ "special": true
116
+ },
117
+ "151657": {
118
+ "content": "<tool_call>",
119
+ "lstrip": false,
120
+ "normalized": false,
121
+ "rstrip": false,
122
+ "single_word": false,
123
+ "special": false
124
+ },
125
+ "151658": {
126
+ "content": "</tool_call>",
127
+ "lstrip": false,
128
+ "normalized": false,
129
+ "rstrip": false,
130
+ "single_word": false,
131
+ "special": false
132
+ },
133
+ "151659": {
134
+ "content": "<|fim_prefix|>",
135
+ "lstrip": false,
136
+ "normalized": false,
137
+ "rstrip": false,
138
+ "single_word": false,
139
+ "special": false
140
+ },
141
+ "151660": {
142
+ "content": "<|fim_middle|>",
143
+ "lstrip": false,
144
+ "normalized": false,
145
+ "rstrip": false,
146
+ "single_word": false,
147
+ "special": false
148
+ },
149
+ "151661": {
150
+ "content": "<|fim_suffix|>",
151
+ "lstrip": false,
152
+ "normalized": false,
153
+ "rstrip": false,
154
+ "single_word": false,
155
+ "special": false
156
+ },
157
+ "151662": {
158
+ "content": "<|fim_pad|>",
159
+ "lstrip": false,
160
+ "normalized": false,
161
+ "rstrip": false,
162
+ "single_word": false,
163
+ "special": false
164
+ },
165
+ "151663": {
166
+ "content": "<|repo_name|>",
167
+ "lstrip": false,
168
+ "normalized": false,
169
+ "rstrip": false,
170
+ "single_word": false,
171
+ "special": false
172
+ },
173
+ "151664": {
174
+ "content": "<|file_sep|>",
175
+ "lstrip": false,
176
+ "normalized": false,
177
+ "rstrip": false,
178
+ "single_word": false,
179
+ "special": false
180
+ },
181
+ "151665": {
182
+ "content": "<tool_response>",
183
+ "lstrip": false,
184
+ "normalized": false,
185
+ "rstrip": false,
186
+ "single_word": false,
187
+ "special": false
188
+ },
189
+ "151666": {
190
+ "content": "</tool_response>",
191
+ "lstrip": false,
192
+ "normalized": false,
193
+ "rstrip": false,
194
+ "single_word": false,
195
+ "special": false
196
+ },
197
+ "151667": {
198
+ "content": "<think>",
199
+ "lstrip": false,
200
+ "normalized": false,
201
+ "rstrip": false,
202
+ "single_word": false,
203
+ "special": false
204
+ },
205
+ "151668": {
206
+ "content": "</think>",
207
+ "lstrip": false,
208
+ "normalized": false,
209
+ "rstrip": false,
210
+ "single_word": false,
211
+ "special": false
212
+ }
213
+ },
214
+ "additional_special_tokens": [
215
+ "<|im_start|>",
216
+ "<|im_end|>",
217
+ "<|object_ref_start|>",
218
+ "<|object_ref_end|>",
219
+ "<|box_start|>",
220
+ "<|box_end|>",
221
+ "<|quad_start|>",
222
+ "<|quad_end|>",
223
+ "<|vision_start|>",
224
+ "<|vision_end|>",
225
+ "<|vision_pad|>",
226
+ "<|image_pad|>",
227
+ "<|video_pad|>"
228
+ ],
229
+ "bos_token": null,
230
+ "clean_up_tokenization_spaces": false,
231
+ "eos_token": "<|im_end|>",
232
+ "errors": "replace",
233
+ "extra_special_tokens": {},
234
+ "model_max_length": 1010000,
235
+ "pad_token": "<|endoftext|>",
236
+ "split_special_tokens": false,
237
+ "tokenizer_class": "Qwen2Tokenizer",
238
+ "unk_token": null
239
+ }
Qwen3-30B-A3B-Instruct-2507-sft-fusang-swa=2k_sink=0/checkpoint-870/trainer_state.json ADDED
@@ -0,0 +1,1252 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "best_global_step": null,
3
+ "best_metric": null,
4
+ "best_model_checkpoint": null,
5
+ "epoch": 1.0,
6
+ "eval_steps": 500,
7
+ "global_step": 870,
8
+ "is_hyper_param_search": false,
9
+ "is_local_process_zero": true,
10
+ "is_world_process_zero": true,
11
+ "log_history": [
12
+ {
13
+ "epoch": 0.005752912411908529,
14
+ "grad_norm": 0.282577782869339,
15
+ "learning_rate": 9.090909090909091e-06,
16
+ "loss": 1.0694,
17
+ "step": 5
18
+ },
19
+ {
20
+ "epoch": 0.011505824823817058,
21
+ "grad_norm": 0.36353206634521484,
22
+ "learning_rate": 2.0454545454545457e-05,
23
+ "loss": 1.0505,
24
+ "step": 10
25
+ },
26
+ {
27
+ "epoch": 0.017258737235725585,
28
+ "grad_norm": 0.44881004095077515,
29
+ "learning_rate": 3.181818181818182e-05,
30
+ "loss": 0.9183,
31
+ "step": 15
32
+ },
33
+ {
34
+ "epoch": 0.023011649647634115,
35
+ "grad_norm": 0.3924488127231598,
36
+ "learning_rate": 4.318181818181819e-05,
37
+ "loss": 0.8695,
38
+ "step": 20
39
+ },
40
+ {
41
+ "epoch": 0.028764562059542642,
42
+ "grad_norm": 0.2559906542301178,
43
+ "learning_rate": 5.4545454545454546e-05,
44
+ "loss": 0.8112,
45
+ "step": 25
46
+ },
47
+ {
48
+ "epoch": 0.03451747447145117,
49
+ "grad_norm": 0.22150976955890656,
50
+ "learning_rate": 6.59090909090909e-05,
51
+ "loss": 0.8127,
52
+ "step": 30
53
+ },
54
+ {
55
+ "epoch": 0.0402703868833597,
56
+ "grad_norm": 0.1973196119070053,
57
+ "learning_rate": 7.727272727272727e-05,
58
+ "loss": 0.7678,
59
+ "step": 35
60
+ },
61
+ {
62
+ "epoch": 0.04602329929526823,
63
+ "grad_norm": 0.16540609300136566,
64
+ "learning_rate": 8.863636363636364e-05,
65
+ "loss": 0.7635,
66
+ "step": 40
67
+ },
68
+ {
69
+ "epoch": 0.05177621170717676,
70
+ "grad_norm": 0.13939730823040009,
71
+ "learning_rate": 0.0001,
72
+ "loss": 0.6718,
73
+ "step": 45
74
+ },
75
+ {
76
+ "epoch": 0.057529124119085284,
77
+ "grad_norm": 0.1322619915008545,
78
+ "learning_rate": 9.93946731234867e-05,
79
+ "loss": 0.7013,
80
+ "step": 50
81
+ },
82
+ {
83
+ "epoch": 0.06328203653099382,
84
+ "grad_norm": 0.12162351608276367,
85
+ "learning_rate": 9.878934624697337e-05,
86
+ "loss": 0.6941,
87
+ "step": 55
88
+ },
89
+ {
90
+ "epoch": 0.06903494894290234,
91
+ "grad_norm": 0.1044265404343605,
92
+ "learning_rate": 9.818401937046005e-05,
93
+ "loss": 0.6889,
94
+ "step": 60
95
+ },
96
+ {
97
+ "epoch": 0.07478786135481087,
98
+ "grad_norm": 0.1243564710021019,
99
+ "learning_rate": 9.757869249394673e-05,
100
+ "loss": 0.6672,
101
+ "step": 65
102
+ },
103
+ {
104
+ "epoch": 0.0805407737667194,
105
+ "grad_norm": 0.11447352916002274,
106
+ "learning_rate": 9.697336561743342e-05,
107
+ "loss": 0.6536,
108
+ "step": 70
109
+ },
110
+ {
111
+ "epoch": 0.08629368617862793,
112
+ "grad_norm": 0.10959111899137497,
113
+ "learning_rate": 9.63680387409201e-05,
114
+ "loss": 0.6615,
115
+ "step": 75
116
+ },
117
+ {
118
+ "epoch": 0.09204659859053646,
119
+ "grad_norm": 0.11831460893154144,
120
+ "learning_rate": 9.576271186440679e-05,
121
+ "loss": 0.673,
122
+ "step": 80
123
+ },
124
+ {
125
+ "epoch": 0.097799511002445,
126
+ "grad_norm": 0.11301098763942719,
127
+ "learning_rate": 9.515738498789348e-05,
128
+ "loss": 0.6649,
129
+ "step": 85
130
+ },
131
+ {
132
+ "epoch": 0.10355242341435351,
133
+ "grad_norm": 0.1104111522436142,
134
+ "learning_rate": 9.455205811138015e-05,
135
+ "loss": 0.6531,
136
+ "step": 90
137
+ },
138
+ {
139
+ "epoch": 0.10930533582626205,
140
+ "grad_norm": 0.2240690290927887,
141
+ "learning_rate": 9.394673123486683e-05,
142
+ "loss": 0.6508,
143
+ "step": 95
144
+ },
145
+ {
146
+ "epoch": 0.11505824823817057,
147
+ "grad_norm": 0.08954017609357834,
148
+ "learning_rate": 9.334140435835351e-05,
149
+ "loss": 0.6125,
150
+ "step": 100
151
+ },
152
+ {
153
+ "epoch": 0.1208111606500791,
154
+ "grad_norm": 0.09213356673717499,
155
+ "learning_rate": 9.27360774818402e-05,
156
+ "loss": 0.6874,
157
+ "step": 105
158
+ },
159
+ {
160
+ "epoch": 0.12656407306198764,
161
+ "grad_norm": 0.12646251916885376,
162
+ "learning_rate": 9.213075060532688e-05,
163
+ "loss": 0.6628,
164
+ "step": 110
165
+ },
166
+ {
167
+ "epoch": 0.13231698547389617,
168
+ "grad_norm": 0.0964658111333847,
169
+ "learning_rate": 9.152542372881357e-05,
170
+ "loss": 0.6374,
171
+ "step": 115
172
+ },
173
+ {
174
+ "epoch": 0.13806989788580468,
175
+ "grad_norm": 0.14164495468139648,
176
+ "learning_rate": 9.092009685230026e-05,
177
+ "loss": 0.6382,
178
+ "step": 120
179
+ },
180
+ {
181
+ "epoch": 0.1438228102977132,
182
+ "grad_norm": 0.0880463570356369,
183
+ "learning_rate": 9.031476997578692e-05,
184
+ "loss": 0.6212,
185
+ "step": 125
186
+ },
187
+ {
188
+ "epoch": 0.14957572270962174,
189
+ "grad_norm": 0.1216307133436203,
190
+ "learning_rate": 8.970944309927361e-05,
191
+ "loss": 0.6637,
192
+ "step": 130
193
+ },
194
+ {
195
+ "epoch": 0.15532863512153028,
196
+ "grad_norm": 0.10961570590734482,
197
+ "learning_rate": 8.910411622276029e-05,
198
+ "loss": 0.6698,
199
+ "step": 135
200
+ },
201
+ {
202
+ "epoch": 0.1610815475334388,
203
+ "grad_norm": 0.09453413635492325,
204
+ "learning_rate": 8.849878934624698e-05,
205
+ "loss": 0.6624,
206
+ "step": 140
207
+ },
208
+ {
209
+ "epoch": 0.16683445994534732,
210
+ "grad_norm": 0.08942891657352448,
211
+ "learning_rate": 8.789346246973366e-05,
212
+ "loss": 0.6134,
213
+ "step": 145
214
+ },
215
+ {
216
+ "epoch": 0.17258737235725585,
217
+ "grad_norm": 0.0847589448094368,
218
+ "learning_rate": 8.728813559322035e-05,
219
+ "loss": 0.6156,
220
+ "step": 150
221
+ },
222
+ {
223
+ "epoch": 0.1783402847691644,
224
+ "grad_norm": 0.08911514282226562,
225
+ "learning_rate": 8.668280871670703e-05,
226
+ "loss": 0.6619,
227
+ "step": 155
228
+ },
229
+ {
230
+ "epoch": 0.18409319718107292,
231
+ "grad_norm": 0.09653668105602264,
232
+ "learning_rate": 8.60774818401937e-05,
233
+ "loss": 0.684,
234
+ "step": 160
235
+ },
236
+ {
237
+ "epoch": 0.18984610959298145,
238
+ "grad_norm": 0.1439998894929886,
239
+ "learning_rate": 8.54721549636804e-05,
240
+ "loss": 0.6479,
241
+ "step": 165
242
+ },
243
+ {
244
+ "epoch": 0.19559902200489,
245
+ "grad_norm": 0.11341068148612976,
246
+ "learning_rate": 8.486682808716707e-05,
247
+ "loss": 0.63,
248
+ "step": 170
249
+ },
250
+ {
251
+ "epoch": 0.2013519344167985,
252
+ "grad_norm": 0.11294681578874588,
253
+ "learning_rate": 8.426150121065376e-05,
254
+ "loss": 0.6412,
255
+ "step": 175
256
+ },
257
+ {
258
+ "epoch": 0.20710484682870703,
259
+ "grad_norm": 0.10501556843519211,
260
+ "learning_rate": 8.365617433414044e-05,
261
+ "loss": 0.648,
262
+ "step": 180
263
+ },
264
+ {
265
+ "epoch": 0.21285775924061556,
266
+ "grad_norm": 0.16354763507843018,
267
+ "learning_rate": 8.305084745762712e-05,
268
+ "loss": 0.6786,
269
+ "step": 185
270
+ },
271
+ {
272
+ "epoch": 0.2186106716525241,
273
+ "grad_norm": 0.10617512464523315,
274
+ "learning_rate": 8.244552058111381e-05,
275
+ "loss": 0.666,
276
+ "step": 190
277
+ },
278
+ {
279
+ "epoch": 0.22436358406443263,
280
+ "grad_norm": 0.10191624611616135,
281
+ "learning_rate": 8.184019370460048e-05,
282
+ "loss": 0.6389,
283
+ "step": 195
284
+ },
285
+ {
286
+ "epoch": 0.23011649647634114,
287
+ "grad_norm": 0.08037319034337997,
288
+ "learning_rate": 8.123486682808718e-05,
289
+ "loss": 0.6147,
290
+ "step": 200
291
+ },
292
+ {
293
+ "epoch": 0.23586940888824967,
294
+ "grad_norm": 0.07923904061317444,
295
+ "learning_rate": 8.062953995157385e-05,
296
+ "loss": 0.6819,
297
+ "step": 205
298
+ },
299
+ {
300
+ "epoch": 0.2416223213001582,
301
+ "grad_norm": 0.10331271588802338,
302
+ "learning_rate": 8.002421307506054e-05,
303
+ "loss": 0.6792,
304
+ "step": 210
305
+ },
306
+ {
307
+ "epoch": 0.24737523371206674,
308
+ "grad_norm": 0.0893971174955368,
309
+ "learning_rate": 7.941888619854722e-05,
310
+ "loss": 0.6336,
311
+ "step": 215
312
+ },
313
+ {
314
+ "epoch": 0.2531281461239753,
315
+ "grad_norm": 0.10475458949804306,
316
+ "learning_rate": 7.88135593220339e-05,
317
+ "loss": 0.6492,
318
+ "step": 220
319
+ },
320
+ {
321
+ "epoch": 0.2588810585358838,
322
+ "grad_norm": 0.08506541699171066,
323
+ "learning_rate": 7.820823244552058e-05,
324
+ "loss": 0.6279,
325
+ "step": 225
326
+ },
327
+ {
328
+ "epoch": 0.26463397094779234,
329
+ "grad_norm": 0.08328807353973389,
330
+ "learning_rate": 7.760290556900727e-05,
331
+ "loss": 0.6357,
332
+ "step": 230
333
+ },
334
+ {
335
+ "epoch": 0.27038688335970085,
336
+ "grad_norm": 0.08994337916374207,
337
+ "learning_rate": 7.699757869249396e-05,
338
+ "loss": 0.6652,
339
+ "step": 235
340
+ },
341
+ {
342
+ "epoch": 0.27613979577160935,
343
+ "grad_norm": 0.0984901711344719,
344
+ "learning_rate": 7.639225181598063e-05,
345
+ "loss": 0.659,
346
+ "step": 240
347
+ },
348
+ {
349
+ "epoch": 0.2818927081835179,
350
+ "grad_norm": 0.15574225783348083,
351
+ "learning_rate": 7.578692493946733e-05,
352
+ "loss": 0.6252,
353
+ "step": 245
354
+ },
355
+ {
356
+ "epoch": 0.2876456205954264,
357
+ "grad_norm": 0.07243578135967255,
358
+ "learning_rate": 7.518159806295399e-05,
359
+ "loss": 0.6277,
360
+ "step": 250
361
+ },
362
+ {
363
+ "epoch": 0.293398533007335,
364
+ "grad_norm": 0.08968574553728104,
365
+ "learning_rate": 7.457627118644068e-05,
366
+ "loss": 0.6337,
367
+ "step": 255
368
+ },
369
+ {
370
+ "epoch": 0.2991514454192435,
371
+ "grad_norm": 0.08754673600196838,
372
+ "learning_rate": 7.397094430992736e-05,
373
+ "loss": 0.6474,
374
+ "step": 260
375
+ },
376
+ {
377
+ "epoch": 0.304904357831152,
378
+ "grad_norm": 0.07851745933294296,
379
+ "learning_rate": 7.336561743341405e-05,
380
+ "loss": 0.6467,
381
+ "step": 265
382
+ },
383
+ {
384
+ "epoch": 0.31065727024306056,
385
+ "grad_norm": 0.1870853453874588,
386
+ "learning_rate": 7.276029055690074e-05,
387
+ "loss": 0.6128,
388
+ "step": 270
389
+ },
390
+ {
391
+ "epoch": 0.31641018265496906,
392
+ "grad_norm": 0.08895161747932434,
393
+ "learning_rate": 7.215496368038742e-05,
394
+ "loss": 0.6252,
395
+ "step": 275
396
+ },
397
+ {
398
+ "epoch": 0.3221630950668776,
399
+ "grad_norm": 0.08845832943916321,
400
+ "learning_rate": 7.154963680387409e-05,
401
+ "loss": 0.6763,
402
+ "step": 280
403
+ },
404
+ {
405
+ "epoch": 0.32791600747878613,
406
+ "grad_norm": 0.08799268305301666,
407
+ "learning_rate": 7.094430992736077e-05,
408
+ "loss": 0.6594,
409
+ "step": 285
410
+ },
411
+ {
412
+ "epoch": 0.33366891989069464,
413
+ "grad_norm": 0.10060054808855057,
414
+ "learning_rate": 7.033898305084746e-05,
415
+ "loss": 0.6297,
416
+ "step": 290
417
+ },
418
+ {
419
+ "epoch": 0.3394218323026032,
420
+ "grad_norm": 0.10938215255737305,
421
+ "learning_rate": 6.973365617433414e-05,
422
+ "loss": 0.6271,
423
+ "step": 295
424
+ },
425
+ {
426
+ "epoch": 0.3451747447145117,
427
+ "grad_norm": 0.0884404182434082,
428
+ "learning_rate": 6.912832929782083e-05,
429
+ "loss": 0.6008,
430
+ "step": 300
431
+ },
432
+ {
433
+ "epoch": 0.35092765712642027,
434
+ "grad_norm": 0.08146841824054718,
435
+ "learning_rate": 6.852300242130752e-05,
436
+ "loss": 0.6709,
437
+ "step": 305
438
+ },
439
+ {
440
+ "epoch": 0.3566805695383288,
441
+ "grad_norm": 0.08434424549341202,
442
+ "learning_rate": 6.791767554479418e-05,
443
+ "loss": 0.6408,
444
+ "step": 310
445
+ },
446
+ {
447
+ "epoch": 0.36243348195023734,
448
+ "grad_norm": 0.09514341503381729,
449
+ "learning_rate": 6.731234866828087e-05,
450
+ "loss": 0.6321,
451
+ "step": 315
452
+ },
453
+ {
454
+ "epoch": 0.36818639436214584,
455
+ "grad_norm": 0.11883344501256943,
456
+ "learning_rate": 6.670702179176755e-05,
457
+ "loss": 0.6062,
458
+ "step": 320
459
+ },
460
+ {
461
+ "epoch": 0.37393930677405435,
462
+ "grad_norm": 0.0747121125459671,
463
+ "learning_rate": 6.610169491525424e-05,
464
+ "loss": 0.6261,
465
+ "step": 325
466
+ },
467
+ {
468
+ "epoch": 0.3796922191859629,
469
+ "grad_norm": 0.08501125127077103,
470
+ "learning_rate": 6.549636803874092e-05,
471
+ "loss": 0.6377,
472
+ "step": 330
473
+ },
474
+ {
475
+ "epoch": 0.3854451315978714,
476
+ "grad_norm": 0.09555582702159882,
477
+ "learning_rate": 6.489104116222761e-05,
478
+ "loss": 0.6591,
479
+ "step": 335
480
+ },
481
+ {
482
+ "epoch": 0.39119804400978,
483
+ "grad_norm": 0.10119756311178207,
484
+ "learning_rate": 6.428571428571429e-05,
485
+ "loss": 0.6425,
486
+ "step": 340
487
+ },
488
+ {
489
+ "epoch": 0.3969509564216885,
490
+ "grad_norm": 0.1302293986082077,
491
+ "learning_rate": 6.368038740920097e-05,
492
+ "loss": 0.6136,
493
+ "step": 345
494
+ },
495
+ {
496
+ "epoch": 0.402703868833597,
497
+ "grad_norm": 0.07237003743648529,
498
+ "learning_rate": 6.307506053268766e-05,
499
+ "loss": 0.6574,
500
+ "step": 350
501
+ },
502
+ {
503
+ "epoch": 0.40845678124550555,
504
+ "grad_norm": 0.0840902030467987,
505
+ "learning_rate": 6.246973365617433e-05,
506
+ "loss": 0.6456,
507
+ "step": 355
508
+ },
509
+ {
510
+ "epoch": 0.41420969365741406,
511
+ "grad_norm": 0.08987671881914139,
512
+ "learning_rate": 6.186440677966102e-05,
513
+ "loss": 0.6394,
514
+ "step": 360
515
+ },
516
+ {
517
+ "epoch": 0.4199626060693226,
518
+ "grad_norm": 0.09869033098220825,
519
+ "learning_rate": 6.12590799031477e-05,
520
+ "loss": 0.619,
521
+ "step": 365
522
+ },
523
+ {
524
+ "epoch": 0.4257155184812311,
525
+ "grad_norm": 0.09710430353879929,
526
+ "learning_rate": 6.0653753026634385e-05,
527
+ "loss": 0.6083,
528
+ "step": 370
529
+ },
530
+ {
531
+ "epoch": 0.43146843089313963,
532
+ "grad_norm": 0.07835939526557922,
533
+ "learning_rate": 6.0048426150121076e-05,
534
+ "loss": 0.584,
535
+ "step": 375
536
+ },
537
+ {
538
+ "epoch": 0.4372213433050482,
539
+ "grad_norm": 0.08729743957519531,
540
+ "learning_rate": 5.9443099273607746e-05,
541
+ "loss": 0.6493,
542
+ "step": 380
543
+ },
544
+ {
545
+ "epoch": 0.4429742557169567,
546
+ "grad_norm": 0.10821270942687988,
547
+ "learning_rate": 5.883777239709444e-05,
548
+ "loss": 0.6745,
549
+ "step": 385
550
+ },
551
+ {
552
+ "epoch": 0.44872716812886526,
553
+ "grad_norm": 0.08159793168306351,
554
+ "learning_rate": 5.8232445520581115e-05,
555
+ "loss": 0.622,
556
+ "step": 390
557
+ },
558
+ {
559
+ "epoch": 0.45448008054077377,
560
+ "grad_norm": 0.10589876025915146,
561
+ "learning_rate": 5.76271186440678e-05,
562
+ "loss": 0.6386,
563
+ "step": 395
564
+ },
565
+ {
566
+ "epoch": 0.4602329929526823,
567
+ "grad_norm": 0.08343663066625595,
568
+ "learning_rate": 5.7021791767554476e-05,
569
+ "loss": 0.6047,
570
+ "step": 400
571
+ },
572
+ {
573
+ "epoch": 0.46598590536459084,
574
+ "grad_norm": 0.09743469953536987,
575
+ "learning_rate": 5.641646489104117e-05,
576
+ "loss": 0.6111,
577
+ "step": 405
578
+ },
579
+ {
580
+ "epoch": 0.47173881777649934,
581
+ "grad_norm": 0.09691858291625977,
582
+ "learning_rate": 5.581113801452785e-05,
583
+ "loss": 0.6311,
584
+ "step": 410
585
+ },
586
+ {
587
+ "epoch": 0.4774917301884079,
588
+ "grad_norm": 0.09147972613573074,
589
+ "learning_rate": 5.520581113801453e-05,
590
+ "loss": 0.6283,
591
+ "step": 415
592
+ },
593
+ {
594
+ "epoch": 0.4832446426003164,
595
+ "grad_norm": 0.09131626039743423,
596
+ "learning_rate": 5.460048426150122e-05,
597
+ "loss": 0.5772,
598
+ "step": 420
599
+ },
600
+ {
601
+ "epoch": 0.4889975550122249,
602
+ "grad_norm": 0.07642737030982971,
603
+ "learning_rate": 5.3995157384987896e-05,
604
+ "loss": 0.6169,
605
+ "step": 425
606
+ },
607
+ {
608
+ "epoch": 0.4947504674241335,
609
+ "grad_norm": 0.09316980093717575,
610
+ "learning_rate": 5.338983050847458e-05,
611
+ "loss": 0.6591,
612
+ "step": 430
613
+ },
614
+ {
615
+ "epoch": 0.500503379836042,
616
+ "grad_norm": 0.13764190673828125,
617
+ "learning_rate": 5.278450363196126e-05,
618
+ "loss": 0.6561,
619
+ "step": 435
620
+ },
621
+ {
622
+ "epoch": 0.5062562922479505,
623
+ "grad_norm": 0.1025523766875267,
624
+ "learning_rate": 5.217917675544795e-05,
625
+ "loss": 0.6379,
626
+ "step": 440
627
+ },
628
+ {
629
+ "epoch": 0.5120092046598591,
630
+ "grad_norm": 0.10537386685609818,
631
+ "learning_rate": 5.157384987893463e-05,
632
+ "loss": 0.6057,
633
+ "step": 445
634
+ },
635
+ {
636
+ "epoch": 0.5177621170717676,
637
+ "grad_norm": 0.08331896364688873,
638
+ "learning_rate": 5.096852300242131e-05,
639
+ "loss": 0.6312,
640
+ "step": 450
641
+ },
642
+ {
643
+ "epoch": 0.5235150294836761,
644
+ "grad_norm": 0.0825723186135292,
645
+ "learning_rate": 5.036319612590799e-05,
646
+ "loss": 0.6644,
647
+ "step": 455
648
+ },
649
+ {
650
+ "epoch": 0.5292679418955847,
651
+ "grad_norm": 0.09409131109714508,
652
+ "learning_rate": 4.975786924939467e-05,
653
+ "loss": 0.6501,
654
+ "step": 460
655
+ },
656
+ {
657
+ "epoch": 0.5350208543074931,
658
+ "grad_norm": 0.10890763252973557,
659
+ "learning_rate": 4.915254237288136e-05,
660
+ "loss": 0.63,
661
+ "step": 465
662
+ },
663
+ {
664
+ "epoch": 0.5407737667194017,
665
+ "grad_norm": 0.11255718022584915,
666
+ "learning_rate": 4.8547215496368045e-05,
667
+ "loss": 0.6325,
668
+ "step": 470
669
+ },
670
+ {
671
+ "epoch": 0.5465266791313103,
672
+ "grad_norm": 0.07348914444446564,
673
+ "learning_rate": 4.794188861985472e-05,
674
+ "loss": 0.6084,
675
+ "step": 475
676
+ },
677
+ {
678
+ "epoch": 0.5522795915432187,
679
+ "grad_norm": 0.09068702161312103,
680
+ "learning_rate": 4.733656174334141e-05,
681
+ "loss": 0.6556,
682
+ "step": 480
683
+ },
684
+ {
685
+ "epoch": 0.5580325039551273,
686
+ "grad_norm": 0.0901346281170845,
687
+ "learning_rate": 4.673123486682809e-05,
688
+ "loss": 0.6526,
689
+ "step": 485
690
+ },
691
+ {
692
+ "epoch": 0.5637854163670358,
693
+ "grad_norm": 0.08321259915828705,
694
+ "learning_rate": 4.612590799031477e-05,
695
+ "loss": 0.6401,
696
+ "step": 490
697
+ },
698
+ {
699
+ "epoch": 0.5695383287789444,
700
+ "grad_norm": 0.09410235285758972,
701
+ "learning_rate": 4.552058111380145e-05,
702
+ "loss": 0.6236,
703
+ "step": 495
704
+ },
705
+ {
706
+ "epoch": 0.5752912411908528,
707
+ "grad_norm": 0.07987872511148453,
708
+ "learning_rate": 4.491525423728814e-05,
709
+ "loss": 0.6051,
710
+ "step": 500
711
+ },
712
+ {
713
+ "epoch": 0.5810441536027614,
714
+ "grad_norm": 0.08640313893556595,
715
+ "learning_rate": 4.430992736077482e-05,
716
+ "loss": 0.6621,
717
+ "step": 505
718
+ },
719
+ {
720
+ "epoch": 0.58679706601467,
721
+ "grad_norm": 0.0959148108959198,
722
+ "learning_rate": 4.3704600484261504e-05,
723
+ "loss": 0.6731,
724
+ "step": 510
725
+ },
726
+ {
727
+ "epoch": 0.5925499784265784,
728
+ "grad_norm": 0.1282995641231537,
729
+ "learning_rate": 4.309927360774819e-05,
730
+ "loss": 0.6518,
731
+ "step": 515
732
+ },
733
+ {
734
+ "epoch": 0.598302890838487,
735
+ "grad_norm": 0.11661899834871292,
736
+ "learning_rate": 4.2493946731234865e-05,
737
+ "loss": 0.5979,
738
+ "step": 520
739
+ },
740
+ {
741
+ "epoch": 0.6040558032503955,
742
+ "grad_norm": 0.0803794413805008,
743
+ "learning_rate": 4.188861985472155e-05,
744
+ "loss": 0.6159,
745
+ "step": 525
746
+ },
747
+ {
748
+ "epoch": 0.609808715662304,
749
+ "grad_norm": 0.0907973200082779,
750
+ "learning_rate": 4.1283292978208233e-05,
751
+ "loss": 0.6878,
752
+ "step": 530
753
+ },
754
+ {
755
+ "epoch": 0.6155616280742126,
756
+ "grad_norm": 0.09559116512537003,
757
+ "learning_rate": 4.067796610169492e-05,
758
+ "loss": 0.6578,
759
+ "step": 535
760
+ },
761
+ {
762
+ "epoch": 0.6213145404861211,
763
+ "grad_norm": 0.11219466477632523,
764
+ "learning_rate": 4.00726392251816e-05,
765
+ "loss": 0.6271,
766
+ "step": 540
767
+ },
768
+ {
769
+ "epoch": 0.6270674528980297,
770
+ "grad_norm": 0.1003156527876854,
771
+ "learning_rate": 3.9467312348668285e-05,
772
+ "loss": 0.6136,
773
+ "step": 545
774
+ },
775
+ {
776
+ "epoch": 0.6328203653099381,
777
+ "grad_norm": 0.07345616817474365,
778
+ "learning_rate": 3.886198547215496e-05,
779
+ "loss": 0.6215,
780
+ "step": 550
781
+ },
782
+ {
783
+ "epoch": 0.6385732777218467,
784
+ "grad_norm": 0.08116694539785385,
785
+ "learning_rate": 3.825665859564165e-05,
786
+ "loss": 0.6454,
787
+ "step": 555
788
+ },
789
+ {
790
+ "epoch": 0.6443261901337552,
791
+ "grad_norm": 0.0830857902765274,
792
+ "learning_rate": 3.765133171912833e-05,
793
+ "loss": 0.6705,
794
+ "step": 560
795
+ },
796
+ {
797
+ "epoch": 0.6500791025456637,
798
+ "grad_norm": 0.08986219763755798,
799
+ "learning_rate": 3.7046004842615015e-05,
800
+ "loss": 0.6581,
801
+ "step": 565
802
+ },
803
+ {
804
+ "epoch": 0.6558320149575723,
805
+ "grad_norm": 0.09591398388147354,
806
+ "learning_rate": 3.644067796610169e-05,
807
+ "loss": 0.6396,
808
+ "step": 570
809
+ },
810
+ {
811
+ "epoch": 0.6615849273694808,
812
+ "grad_norm": 0.06546227633953094,
813
+ "learning_rate": 3.583535108958838e-05,
814
+ "loss": 0.6131,
815
+ "step": 575
816
+ },
817
+ {
818
+ "epoch": 0.6673378397813893,
819
+ "grad_norm": 0.07436897605657578,
820
+ "learning_rate": 3.523002421307507e-05,
821
+ "loss": 0.6594,
822
+ "step": 580
823
+ },
824
+ {
825
+ "epoch": 0.6730907521932978,
826
+ "grad_norm": 0.08146259188652039,
827
+ "learning_rate": 3.4624697336561744e-05,
828
+ "loss": 0.6073,
829
+ "step": 585
830
+ },
831
+ {
832
+ "epoch": 0.6788436646052064,
833
+ "grad_norm": 0.07927431166172028,
834
+ "learning_rate": 3.401937046004843e-05,
835
+ "loss": 0.643,
836
+ "step": 590
837
+ },
838
+ {
839
+ "epoch": 0.684596577017115,
840
+ "grad_norm": 0.12594762444496155,
841
+ "learning_rate": 3.341404358353511e-05,
842
+ "loss": 0.5955,
843
+ "step": 595
844
+ },
845
+ {
846
+ "epoch": 0.6903494894290234,
847
+ "grad_norm": 0.07793177664279938,
848
+ "learning_rate": 3.280871670702179e-05,
849
+ "loss": 0.5978,
850
+ "step": 600
851
+ },
852
+ {
853
+ "epoch": 0.696102401840932,
854
+ "grad_norm": 0.07432356476783752,
855
+ "learning_rate": 3.2203389830508473e-05,
856
+ "loss": 0.6707,
857
+ "step": 605
858
+ },
859
+ {
860
+ "epoch": 0.7018553142528405,
861
+ "grad_norm": 0.08397427201271057,
862
+ "learning_rate": 3.1598062953995164e-05,
863
+ "loss": 0.6721,
864
+ "step": 610
865
+ },
866
+ {
867
+ "epoch": 0.707608226664749,
868
+ "grad_norm": 0.08302836120128632,
869
+ "learning_rate": 3.099273607748184e-05,
870
+ "loss": 0.628,
871
+ "step": 615
872
+ },
873
+ {
874
+ "epoch": 0.7133611390766575,
875
+ "grad_norm": 0.21343210339546204,
876
+ "learning_rate": 3.0387409200968526e-05,
877
+ "loss": 0.5716,
878
+ "step": 620
879
+ },
880
+ {
881
+ "epoch": 0.7191140514885661,
882
+ "grad_norm": 0.09918905794620514,
883
+ "learning_rate": 2.9782082324455206e-05,
884
+ "loss": 0.6042,
885
+ "step": 625
886
+ },
887
+ {
888
+ "epoch": 0.7248669639004747,
889
+ "grad_norm": 0.08919833600521088,
890
+ "learning_rate": 2.917675544794189e-05,
891
+ "loss": 0.6468,
892
+ "step": 630
893
+ },
894
+ {
895
+ "epoch": 0.7306198763123831,
896
+ "grad_norm": 0.09351971745491028,
897
+ "learning_rate": 2.857142857142857e-05,
898
+ "loss": 0.6401,
899
+ "step": 635
900
+ },
901
+ {
902
+ "epoch": 0.7363727887242917,
903
+ "grad_norm": 0.085872121155262,
904
+ "learning_rate": 2.7966101694915255e-05,
905
+ "loss": 0.636,
906
+ "step": 640
907
+ },
908
+ {
909
+ "epoch": 0.7421257011362002,
910
+ "grad_norm": 0.11117161065340042,
911
+ "learning_rate": 2.7360774818401942e-05,
912
+ "loss": 0.6049,
913
+ "step": 645
914
+ },
915
+ {
916
+ "epoch": 0.7478786135481087,
917
+ "grad_norm": 0.06663698703050613,
918
+ "learning_rate": 2.6755447941888623e-05,
919
+ "loss": 0.6281,
920
+ "step": 650
921
+ },
922
+ {
923
+ "epoch": 0.7536315259600173,
924
+ "grad_norm": 0.08496002107858658,
925
+ "learning_rate": 2.6150121065375304e-05,
926
+ "loss": 0.6322,
927
+ "step": 655
928
+ },
929
+ {
930
+ "epoch": 0.7593844383719258,
931
+ "grad_norm": 0.11393781751394272,
932
+ "learning_rate": 2.5544794188861988e-05,
933
+ "loss": 0.6742,
934
+ "step": 660
935
+ },
936
+ {
937
+ "epoch": 0.7651373507838343,
938
+ "grad_norm": 0.07825738936662674,
939
+ "learning_rate": 2.4939467312348668e-05,
940
+ "loss": 0.6112,
941
+ "step": 665
942
+ },
943
+ {
944
+ "epoch": 0.7708902631957428,
945
+ "grad_norm": 0.09605414420366287,
946
+ "learning_rate": 2.4334140435835352e-05,
947
+ "loss": 0.617,
948
+ "step": 670
949
+ },
950
+ {
951
+ "epoch": 0.7766431756076514,
952
+ "grad_norm": 0.07142200320959091,
953
+ "learning_rate": 2.3728813559322036e-05,
954
+ "loss": 0.6125,
955
+ "step": 675
956
+ },
957
+ {
958
+ "epoch": 0.78239608801956,
959
+ "grad_norm": 0.08845455199480057,
960
+ "learning_rate": 2.3123486682808717e-05,
961
+ "loss": 0.6434,
962
+ "step": 680
963
+ },
964
+ {
965
+ "epoch": 0.7881490004314684,
966
+ "grad_norm": 0.09362835437059402,
967
+ "learning_rate": 2.25181598062954e-05,
968
+ "loss": 0.6661,
969
+ "step": 685
970
+ },
971
+ {
972
+ "epoch": 0.793901912843377,
973
+ "grad_norm": 0.07780930399894714,
974
+ "learning_rate": 2.1912832929782085e-05,
975
+ "loss": 0.6372,
976
+ "step": 690
977
+ },
978
+ {
979
+ "epoch": 0.7996548252552855,
980
+ "grad_norm": 0.09395743161439896,
981
+ "learning_rate": 2.1307506053268766e-05,
982
+ "loss": 0.6482,
983
+ "step": 695
984
+ },
985
+ {
986
+ "epoch": 0.805407737667194,
987
+ "grad_norm": 0.08778950572013855,
988
+ "learning_rate": 2.070217917675545e-05,
989
+ "loss": 0.5997,
990
+ "step": 700
991
+ },
992
+ {
993
+ "epoch": 0.8111606500791025,
994
+ "grad_norm": 0.07783056795597076,
995
+ "learning_rate": 2.009685230024213e-05,
996
+ "loss": 0.6399,
997
+ "step": 705
998
+ },
999
+ {
1000
+ "epoch": 0.8169135624910111,
1001
+ "grad_norm": 0.0792059674859047,
1002
+ "learning_rate": 1.9491525423728814e-05,
1003
+ "loss": 0.6422,
1004
+ "step": 710
1005
+ },
1006
+ {
1007
+ "epoch": 0.8226664749029196,
1008
+ "grad_norm": 0.09166564792394638,
1009
+ "learning_rate": 1.88861985472155e-05,
1010
+ "loss": 0.6425,
1011
+ "step": 715
1012
+ },
1013
+ {
1014
+ "epoch": 0.8284193873148281,
1015
+ "grad_norm": 0.10166376829147339,
1016
+ "learning_rate": 1.828087167070218e-05,
1017
+ "loss": 0.6304,
1018
+ "step": 720
1019
+ },
1020
+ {
1021
+ "epoch": 0.8341722997267367,
1022
+ "grad_norm": 0.08811493217945099,
1023
+ "learning_rate": 1.7675544794188863e-05,
1024
+ "loss": 0.6248,
1025
+ "step": 725
1026
+ },
1027
+ {
1028
+ "epoch": 0.8399252121386452,
1029
+ "grad_norm": 0.09864843636751175,
1030
+ "learning_rate": 1.7070217917675547e-05,
1031
+ "loss": 0.6347,
1032
+ "step": 730
1033
+ },
1034
+ {
1035
+ "epoch": 0.8456781245505537,
1036
+ "grad_norm": 0.0813138335943222,
1037
+ "learning_rate": 1.6464891041162228e-05,
1038
+ "loss": 0.6568,
1039
+ "step": 735
1040
+ },
1041
+ {
1042
+ "epoch": 0.8514310369624623,
1043
+ "grad_norm": 0.07606098800897598,
1044
+ "learning_rate": 1.5859564164648912e-05,
1045
+ "loss": 0.6503,
1046
+ "step": 740
1047
+ },
1048
+ {
1049
+ "epoch": 0.8571839493743708,
1050
+ "grad_norm": 0.1486358940601349,
1051
+ "learning_rate": 1.5254237288135596e-05,
1052
+ "loss": 0.6062,
1053
+ "step": 745
1054
+ },
1055
+ {
1056
+ "epoch": 0.8629368617862793,
1057
+ "grad_norm": 0.08340667188167572,
1058
+ "learning_rate": 1.4648910411622276e-05,
1059
+ "loss": 0.594,
1060
+ "step": 750
1061
+ },
1062
+ {
1063
+ "epoch": 0.8686897741981878,
1064
+ "grad_norm": 0.07569226622581482,
1065
+ "learning_rate": 1.4043583535108959e-05,
1066
+ "loss": 0.6742,
1067
+ "step": 755
1068
+ },
1069
+ {
1070
+ "epoch": 0.8744426866100964,
1071
+ "grad_norm": 0.10132869333028793,
1072
+ "learning_rate": 1.3438256658595641e-05,
1073
+ "loss": 0.6566,
1074
+ "step": 760
1075
+ },
1076
+ {
1077
+ "epoch": 0.8801955990220048,
1078
+ "grad_norm": 0.11646664887666702,
1079
+ "learning_rate": 1.2832929782082325e-05,
1080
+ "loss": 0.6165,
1081
+ "step": 765
1082
+ },
1083
+ {
1084
+ "epoch": 0.8859485114339134,
1085
+ "grad_norm": 0.08847946673631668,
1086
+ "learning_rate": 1.2227602905569007e-05,
1087
+ "loss": 0.6005,
1088
+ "step": 770
1089
+ },
1090
+ {
1091
+ "epoch": 0.891701423845822,
1092
+ "grad_norm": 0.0643245056271553,
1093
+ "learning_rate": 1.1622276029055691e-05,
1094
+ "loss": 0.612,
1095
+ "step": 775
1096
+ },
1097
+ {
1098
+ "epoch": 0.8974543362577305,
1099
+ "grad_norm": 0.07544547319412231,
1100
+ "learning_rate": 1.1016949152542374e-05,
1101
+ "loss": 0.6364,
1102
+ "step": 780
1103
+ },
1104
+ {
1105
+ "epoch": 0.903207248669639,
1106
+ "grad_norm": 0.08121296763420105,
1107
+ "learning_rate": 1.0411622276029056e-05,
1108
+ "loss": 0.6447,
1109
+ "step": 785
1110
+ },
1111
+ {
1112
+ "epoch": 0.9089601610815475,
1113
+ "grad_norm": 0.07453462481498718,
1114
+ "learning_rate": 9.806295399515738e-06,
1115
+ "loss": 0.6407,
1116
+ "step": 790
1117
+ },
1118
+ {
1119
+ "epoch": 0.9147130734934561,
1120
+ "grad_norm": 0.12451841682195663,
1121
+ "learning_rate": 9.200968523002422e-06,
1122
+ "loss": 0.6306,
1123
+ "step": 795
1124
+ },
1125
+ {
1126
+ "epoch": 0.9204659859053645,
1127
+ "grad_norm": 0.07748465985059738,
1128
+ "learning_rate": 8.595641646489105e-06,
1129
+ "loss": 0.603,
1130
+ "step": 800
1131
+ },
1132
+ {
1133
+ "epoch": 0.9262188983172731,
1134
+ "grad_norm": 0.08191264420747757,
1135
+ "learning_rate": 7.990314769975787e-06,
1136
+ "loss": 0.6493,
1137
+ "step": 805
1138
+ },
1139
+ {
1140
+ "epoch": 0.9319718107291817,
1141
+ "grad_norm": 0.08016539365053177,
1142
+ "learning_rate": 7.3849878934624694e-06,
1143
+ "loss": 0.64,
1144
+ "step": 810
1145
+ },
1146
+ {
1147
+ "epoch": 0.9377247231410901,
1148
+ "grad_norm": 0.08719329535961151,
1149
+ "learning_rate": 6.779661016949153e-06,
1150
+ "loss": 0.6163,
1151
+ "step": 815
1152
+ },
1153
+ {
1154
+ "epoch": 0.9434776355529987,
1155
+ "grad_norm": 0.0915544256567955,
1156
+ "learning_rate": 6.174334140435836e-06,
1157
+ "loss": 0.5936,
1158
+ "step": 820
1159
+ },
1160
+ {
1161
+ "epoch": 0.9492305479649072,
1162
+ "grad_norm": 0.0850190818309784,
1163
+ "learning_rate": 5.569007263922518e-06,
1164
+ "loss": 0.6156,
1165
+ "step": 825
1166
+ },
1167
+ {
1168
+ "epoch": 0.9549834603768158,
1169
+ "grad_norm": 0.08744943141937256,
1170
+ "learning_rate": 4.963680387409201e-06,
1171
+ "loss": 0.6321,
1172
+ "step": 830
1173
+ },
1174
+ {
1175
+ "epoch": 0.9607363727887243,
1176
+ "grad_norm": 0.09916157275438309,
1177
+ "learning_rate": 4.358353510895884e-06,
1178
+ "loss": 0.6684,
1179
+ "step": 835
1180
+ },
1181
+ {
1182
+ "epoch": 0.9664892852006328,
1183
+ "grad_norm": 0.07601335644721985,
1184
+ "learning_rate": 3.7530266343825673e-06,
1185
+ "loss": 0.6212,
1186
+ "step": 840
1187
+ },
1188
+ {
1189
+ "epoch": 0.9722421976125414,
1190
+ "grad_norm": 0.09425202012062073,
1191
+ "learning_rate": 3.1476997578692496e-06,
1192
+ "loss": 0.5949,
1193
+ "step": 845
1194
+ },
1195
+ {
1196
+ "epoch": 0.9779951100244498,
1197
+ "grad_norm": 0.08275400847196579,
1198
+ "learning_rate": 2.5423728813559323e-06,
1199
+ "loss": 0.631,
1200
+ "step": 850
1201
+ },
1202
+ {
1203
+ "epoch": 0.9837480224363584,
1204
+ "grad_norm": 0.09351499378681183,
1205
+ "learning_rate": 1.937046004842615e-06,
1206
+ "loss": 0.6345,
1207
+ "step": 855
1208
+ },
1209
+ {
1210
+ "epoch": 0.989500934848267,
1211
+ "grad_norm": 0.0893460288643837,
1212
+ "learning_rate": 1.3317191283292979e-06,
1213
+ "loss": 0.6311,
1214
+ "step": 860
1215
+ },
1216
+ {
1217
+ "epoch": 0.9952538472601755,
1218
+ "grad_norm": 0.07067562639713287,
1219
+ "learning_rate": 7.263922518159807e-07,
1220
+ "loss": 0.6441,
1221
+ "step": 865
1222
+ },
1223
+ {
1224
+ "epoch": 1.0,
1225
+ "grad_norm": 0.22969621419906616,
1226
+ "learning_rate": 1.2106537530266344e-07,
1227
+ "loss": 0.5609,
1228
+ "step": 870
1229
+ }
1230
+ ],
1231
+ "logging_steps": 5,
1232
+ "max_steps": 870,
1233
+ "num_input_tokens_seen": 0,
1234
+ "num_train_epochs": 1,
1235
+ "save_steps": 500,
1236
+ "stateful_callbacks": {
1237
+ "TrainerControl": {
1238
+ "args": {
1239
+ "should_epoch_stop": false,
1240
+ "should_evaluate": false,
1241
+ "should_log": false,
1242
+ "should_save": true,
1243
+ "should_training_stop": true
1244
+ },
1245
+ "attributes": {}
1246
+ }
1247
+ },
1248
+ "total_flos": 1.1405546259399626e+20,
1249
+ "train_batch_size": 1,
1250
+ "trial_name": null,
1251
+ "trial_params": null
1252
+ }
Qwen3-30B-A3B-Instruct-2507-sft-fusang-swa=2k_sink=0/checkpoint-870/vocab.json ADDED
The diff for this file is too large to render. See raw diff
 
Qwen3-30B-A3B-Instruct-2507-sft-fusang-swa=2k_sink=0_falayer=25_from0_step2_fadec/checkpoint-994/adapter_config.json ADDED
@@ -0,0 +1,38 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "alpha_pattern": {},
3
+ "auto_mapping": null,
4
+ "base_model_name_or_path": "/share/models/Qwen3-30B-A3B-Instruct-2507",
5
+ "bias": "none",
6
+ "corda_config": null,
7
+ "eva_config": null,
8
+ "exclude_modules": null,
9
+ "fan_in_fan_out": false,
10
+ "inference_mode": true,
11
+ "init_lora_weights": true,
12
+ "layer_replication": null,
13
+ "layers_pattern": null,
14
+ "layers_to_transform": null,
15
+ "loftq_config": {},
16
+ "lora_alpha": 32,
17
+ "lora_bias": false,
18
+ "lora_dropout": 0.1,
19
+ "megatron_config": null,
20
+ "megatron_core": "megatron.core",
21
+ "modules_to_save": null,
22
+ "peft_type": "LORA",
23
+ "qalora_group_size": 16,
24
+ "r": 16,
25
+ "rank_pattern": {},
26
+ "revision": null,
27
+ "target_modules": [
28
+ "k_proj",
29
+ "v_proj",
30
+ "q_proj"
31
+ ],
32
+ "target_parameters": null,
33
+ "task_type": "CAUSAL_LM",
34
+ "trainable_token_indices": null,
35
+ "use_dora": false,
36
+ "use_qalora": false,
37
+ "use_rslora": true
38
+ }
Qwen3-30B-A3B-Instruct-2507-sft-fusang-swa=2k_sink=0_falayer=25_from0_step2_fadec/checkpoint-994/tokenizer.json ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:aeb13307a71acd8fe81861d94ad54ab689df773318809eed3cbe794b4492dae4
3
+ size 11422654
Qwen3-30B-A3B-Instruct-2507-sft-fusang-swa=2k_sink=0_falayer=25_from0_step2_fadec/checkpoint-994/vocab.json ADDED
The diff for this file is too large to render. See raw diff
 
Qwen3-30B-A3B-Instruct-2507-sft-fusang-swa=2k_sink=100_fadec/checkpoint-870/adapter_config.json ADDED
@@ -0,0 +1,38 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "alpha_pattern": {},
3
+ "auto_mapping": null,
4
+ "base_model_name_or_path": "/share/models/Qwen3-30B-A3B-Instruct-2507",
5
+ "bias": "none",
6
+ "corda_config": null,
7
+ "eva_config": null,
8
+ "exclude_modules": null,
9
+ "fan_in_fan_out": false,
10
+ "inference_mode": true,
11
+ "init_lora_weights": true,
12
+ "layer_replication": null,
13
+ "layers_pattern": null,
14
+ "layers_to_transform": null,
15
+ "loftq_config": {},
16
+ "lora_alpha": 32,
17
+ "lora_bias": false,
18
+ "lora_dropout": 0.1,
19
+ "megatron_config": null,
20
+ "megatron_core": "megatron.core",
21
+ "modules_to_save": null,
22
+ "peft_type": "LORA",
23
+ "qalora_group_size": 16,
24
+ "r": 16,
25
+ "rank_pattern": {},
26
+ "revision": null,
27
+ "target_modules": [
28
+ "q_proj",
29
+ "k_proj",
30
+ "v_proj"
31
+ ],
32
+ "target_parameters": null,
33
+ "task_type": "CAUSAL_LM",
34
+ "trainable_token_indices": null,
35
+ "use_dora": false,
36
+ "use_qalora": false,
37
+ "use_rslora": true
38
+ }
Qwen3-30B-A3B-Instruct-2507-sft-fusang-swa=2k_sink=100_fadec/checkpoint-870/chat_template.jinja ADDED
@@ -0,0 +1,61 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {%- if tools %}
2
+ {{- '<|im_start|>system\n' }}
3
+ {%- if messages[0].role == 'system' %}
4
+ {{- messages[0].content + '\n\n' }}
5
+ {%- endif %}
6
+ {{- "# Tools\n\nYou may call one or more functions to assist with the user query.\n\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>" }}
7
+ {%- for tool in tools %}
8
+ {{- "\n" }}
9
+ {{- tool | tojson }}
10
+ {%- endfor %}
11
+ {{- "\n</tools>\n\nFor each function call, return a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call><|im_end|>\n" }}
12
+ {%- else %}
13
+ {%- if messages[0].role == 'system' %}
14
+ {{- '<|im_start|>system\n' + messages[0].content + '<|im_end|>\n' }}
15
+ {%- endif %}
16
+ {%- endif %}
17
+ {%- for message in messages %}
18
+ {%- if message.content is string %}
19
+ {%- set content = message.content %}
20
+ {%- else %}
21
+ {%- set content = '' %}
22
+ {%- endif %}
23
+ {%- if (message.role == "user") or (message.role == "system" and not loop.first) %}
24
+ {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
25
+ {%- elif message.role == "assistant" %}
26
+ {{- '<|im_start|>' + message.role + '\n' + content }}
27
+ {%- if message.tool_calls %}
28
+ {%- for tool_call in message.tool_calls %}
29
+ {%- if (loop.first and content) or (not loop.first) %}
30
+ {{- '\n' }}
31
+ {%- endif %}
32
+ {%- if tool_call.function %}
33
+ {%- set tool_call = tool_call.function %}
34
+ {%- endif %}
35
+ {{- '<tool_call>\n{"name": "' }}
36
+ {{- tool_call.name }}
37
+ {{- '", "arguments": ' }}
38
+ {%- if tool_call.arguments is string %}
39
+ {{- tool_call.arguments }}
40
+ {%- else %}
41
+ {{- tool_call.arguments | tojson }}
42
+ {%- endif %}
43
+ {{- '}\n</tool_call>' }}
44
+ {%- endfor %}
45
+ {%- endif %}
46
+ {{- '<|im_end|>\n' }}
47
+ {%- elif message.role == "tool" %}
48
+ {%- if loop.first or (messages[loop.index0 - 1].role != "tool") %}
49
+ {{- '<|im_start|>user' }}
50
+ {%- endif %}
51
+ {{- '\n<tool_response>\n' }}
52
+ {{- content }}
53
+ {{- '\n</tool_response>' }}
54
+ {%- if loop.last or (messages[loop.index0 + 1].role != "tool") %}
55
+ {{- '<|im_end|>\n' }}
56
+ {%- endif %}
57
+ {%- endif %}
58
+ {%- endfor %}
59
+ {%- if add_generation_prompt %}
60
+ {{- '<|im_start|>assistant\n' }}
61
+ {%- endif %}
Qwen3-30B-A3B-Instruct-2507-sft-fusang-swa=2k_sink=100_fadec/checkpoint-870/special_tokens_map.json ADDED
@@ -0,0 +1,31 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "additional_special_tokens": [
3
+ "<|im_start|>",
4
+ "<|im_end|>",
5
+ "<|object_ref_start|>",
6
+ "<|object_ref_end|>",
7
+ "<|box_start|>",
8
+ "<|box_end|>",
9
+ "<|quad_start|>",
10
+ "<|quad_end|>",
11
+ "<|vision_start|>",
12
+ "<|vision_end|>",
13
+ "<|vision_pad|>",
14
+ "<|image_pad|>",
15
+ "<|video_pad|>"
16
+ ],
17
+ "eos_token": {
18
+ "content": "<|im_end|>",
19
+ "lstrip": false,
20
+ "normalized": false,
21
+ "rstrip": false,
22
+ "single_word": false
23
+ },
24
+ "pad_token": {
25
+ "content": "<|endoftext|>",
26
+ "lstrip": false,
27
+ "normalized": false,
28
+ "rstrip": false,
29
+ "single_word": false
30
+ }
31
+ }
Qwen3-30B-A3B-Instruct-2507-sft-fusang-swa=2k_sink=100_fadec/checkpoint-870/tokenizer.json ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:aeb13307a71acd8fe81861d94ad54ab689df773318809eed3cbe794b4492dae4
3
+ size 11422654
Qwen3-30B-A3B-Instruct-2507-sft-fusang-swa=2k_sink=100_fadec/checkpoint-870/tokenizer_config.json ADDED
@@ -0,0 +1,239 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "add_bos_token": false,
3
+ "add_prefix_space": false,
4
+ "added_tokens_decoder": {
5
+ "151643": {
6
+ "content": "<|endoftext|>",
7
+ "lstrip": false,
8
+ "normalized": false,
9
+ "rstrip": false,
10
+ "single_word": false,
11
+ "special": true
12
+ },
13
+ "151644": {
14
+ "content": "<|im_start|>",
15
+ "lstrip": false,
16
+ "normalized": false,
17
+ "rstrip": false,
18
+ "single_word": false,
19
+ "special": true
20
+ },
21
+ "151645": {
22
+ "content": "<|im_end|>",
23
+ "lstrip": false,
24
+ "normalized": false,
25
+ "rstrip": false,
26
+ "single_word": false,
27
+ "special": true
28
+ },
29
+ "151646": {
30
+ "content": "<|object_ref_start|>",
31
+ "lstrip": false,
32
+ "normalized": false,
33
+ "rstrip": false,
34
+ "single_word": false,
35
+ "special": true
36
+ },
37
+ "151647": {
38
+ "content": "<|object_ref_end|>",
39
+ "lstrip": false,
40
+ "normalized": false,
41
+ "rstrip": false,
42
+ "single_word": false,
43
+ "special": true
44
+ },
45
+ "151648": {
46
+ "content": "<|box_start|>",
47
+ "lstrip": false,
48
+ "normalized": false,
49
+ "rstrip": false,
50
+ "single_word": false,
51
+ "special": true
52
+ },
53
+ "151649": {
54
+ "content": "<|box_end|>",
55
+ "lstrip": false,
56
+ "normalized": false,
57
+ "rstrip": false,
58
+ "single_word": false,
59
+ "special": true
60
+ },
61
+ "151650": {
62
+ "content": "<|quad_start|>",
63
+ "lstrip": false,
64
+ "normalized": false,
65
+ "rstrip": false,
66
+ "single_word": false,
67
+ "special": true
68
+ },
69
+ "151651": {
70
+ "content": "<|quad_end|>",
71
+ "lstrip": false,
72
+ "normalized": false,
73
+ "rstrip": false,
74
+ "single_word": false,
75
+ "special": true
76
+ },
77
+ "151652": {
78
+ "content": "<|vision_start|>",
79
+ "lstrip": false,
80
+ "normalized": false,
81
+ "rstrip": false,
82
+ "single_word": false,
83
+ "special": true
84
+ },
85
+ "151653": {
86
+ "content": "<|vision_end|>",
87
+ "lstrip": false,
88
+ "normalized": false,
89
+ "rstrip": false,
90
+ "single_word": false,
91
+ "special": true
92
+ },
93
+ "151654": {
94
+ "content": "<|vision_pad|>",
95
+ "lstrip": false,
96
+ "normalized": false,
97
+ "rstrip": false,
98
+ "single_word": false,
99
+ "special": true
100
+ },
101
+ "151655": {
102
+ "content": "<|image_pad|>",
103
+ "lstrip": false,
104
+ "normalized": false,
105
+ "rstrip": false,
106
+ "single_word": false,
107
+ "special": true
108
+ },
109
+ "151656": {
110
+ "content": "<|video_pad|>",
111
+ "lstrip": false,
112
+ "normalized": false,
113
+ "rstrip": false,
114
+ "single_word": false,
115
+ "special": true
116
+ },
117
+ "151657": {
118
+ "content": "<tool_call>",
119
+ "lstrip": false,
120
+ "normalized": false,
121
+ "rstrip": false,
122
+ "single_word": false,
123
+ "special": false
124
+ },
125
+ "151658": {
126
+ "content": "</tool_call>",
127
+ "lstrip": false,
128
+ "normalized": false,
129
+ "rstrip": false,
130
+ "single_word": false,
131
+ "special": false
132
+ },
133
+ "151659": {
134
+ "content": "<|fim_prefix|>",
135
+ "lstrip": false,
136
+ "normalized": false,
137
+ "rstrip": false,
138
+ "single_word": false,
139
+ "special": false
140
+ },
141
+ "151660": {
142
+ "content": "<|fim_middle|>",
143
+ "lstrip": false,
144
+ "normalized": false,
145
+ "rstrip": false,
146
+ "single_word": false,
147
+ "special": false
148
+ },
149
+ "151661": {
150
+ "content": "<|fim_suffix|>",
151
+ "lstrip": false,
152
+ "normalized": false,
153
+ "rstrip": false,
154
+ "single_word": false,
155
+ "special": false
156
+ },
157
+ "151662": {
158
+ "content": "<|fim_pad|>",
159
+ "lstrip": false,
160
+ "normalized": false,
161
+ "rstrip": false,
162
+ "single_word": false,
163
+ "special": false
164
+ },
165
+ "151663": {
166
+ "content": "<|repo_name|>",
167
+ "lstrip": false,
168
+ "normalized": false,
169
+ "rstrip": false,
170
+ "single_word": false,
171
+ "special": false
172
+ },
173
+ "151664": {
174
+ "content": "<|file_sep|>",
175
+ "lstrip": false,
176
+ "normalized": false,
177
+ "rstrip": false,
178
+ "single_word": false,
179
+ "special": false
180
+ },
181
+ "151665": {
182
+ "content": "<tool_response>",
183
+ "lstrip": false,
184
+ "normalized": false,
185
+ "rstrip": false,
186
+ "single_word": false,
187
+ "special": false
188
+ },
189
+ "151666": {
190
+ "content": "</tool_response>",
191
+ "lstrip": false,
192
+ "normalized": false,
193
+ "rstrip": false,
194
+ "single_word": false,
195
+ "special": false
196
+ },
197
+ "151667": {
198
+ "content": "<think>",
199
+ "lstrip": false,
200
+ "normalized": false,
201
+ "rstrip": false,
202
+ "single_word": false,
203
+ "special": false
204
+ },
205
+ "151668": {
206
+ "content": "</think>",
207
+ "lstrip": false,
208
+ "normalized": false,
209
+ "rstrip": false,
210
+ "single_word": false,
211
+ "special": false
212
+ }
213
+ },
214
+ "additional_special_tokens": [
215
+ "<|im_start|>",
216
+ "<|im_end|>",
217
+ "<|object_ref_start|>",
218
+ "<|object_ref_end|>",
219
+ "<|box_start|>",
220
+ "<|box_end|>",
221
+ "<|quad_start|>",
222
+ "<|quad_end|>",
223
+ "<|vision_start|>",
224
+ "<|vision_end|>",
225
+ "<|vision_pad|>",
226
+ "<|image_pad|>",
227
+ "<|video_pad|>"
228
+ ],
229
+ "bos_token": null,
230
+ "clean_up_tokenization_spaces": false,
231
+ "eos_token": "<|im_end|>",
232
+ "errors": "replace",
233
+ "extra_special_tokens": {},
234
+ "model_max_length": 1010000,
235
+ "pad_token": "<|endoftext|>",
236
+ "split_special_tokens": false,
237
+ "tokenizer_class": "Qwen2Tokenizer",
238
+ "unk_token": null
239
+ }
Qwen3-30B-A3B-Instruct-2507-sft-fusang-swa=2k_sink=100_fadec/checkpoint-870/trainer_state.json ADDED
@@ -0,0 +1,1252 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "best_global_step": null,
3
+ "best_metric": null,
4
+ "best_model_checkpoint": null,
5
+ "epoch": 1.0,
6
+ "eval_steps": 500,
7
+ "global_step": 870,
8
+ "is_hyper_param_search": false,
9
+ "is_local_process_zero": true,
10
+ "is_world_process_zero": true,
11
+ "log_history": [
12
+ {
13
+ "epoch": 0.005752912411908529,
14
+ "grad_norm": 0.1586630642414093,
15
+ "learning_rate": 3.053435114503817e-06,
16
+ "loss": 0.6579,
17
+ "step": 5
18
+ },
19
+ {
20
+ "epoch": 0.011505824823817058,
21
+ "grad_norm": 0.16521619260311127,
22
+ "learning_rate": 6.870229007633589e-06,
23
+ "loss": 0.6588,
24
+ "step": 10
25
+ },
26
+ {
27
+ "epoch": 0.017258737235725585,
28
+ "grad_norm": 0.17625975608825684,
29
+ "learning_rate": 1.0687022900763359e-05,
30
+ "loss": 0.6143,
31
+ "step": 15
32
+ },
33
+ {
34
+ "epoch": 0.023011649647634115,
35
+ "grad_norm": 0.13356424868106842,
36
+ "learning_rate": 1.450381679389313e-05,
37
+ "loss": 0.6204,
38
+ "step": 20
39
+ },
40
+ {
41
+ "epoch": 0.028764562059542642,
42
+ "grad_norm": 0.2011476457118988,
43
+ "learning_rate": 1.83206106870229e-05,
44
+ "loss": 0.6355,
45
+ "step": 25
46
+ },
47
+ {
48
+ "epoch": 0.03451747447145117,
49
+ "grad_norm": 0.1653004288673401,
50
+ "learning_rate": 2.2137404580152673e-05,
51
+ "loss": 0.6685,
52
+ "step": 30
53
+ },
54
+ {
55
+ "epoch": 0.0402703868833597,
56
+ "grad_norm": 0.0881788358092308,
57
+ "learning_rate": 2.5954198473282442e-05,
58
+ "loss": 0.6626,
59
+ "step": 35
60
+ },
61
+ {
62
+ "epoch": 0.04602329929526823,
63
+ "grad_norm": 0.08161737769842148,
64
+ "learning_rate": 2.9770992366412214e-05,
65
+ "loss": 0.6833,
66
+ "step": 40
67
+ },
68
+ {
69
+ "epoch": 0.05177621170717676,
70
+ "grad_norm": 0.12486688047647476,
71
+ "learning_rate": 3.358778625954199e-05,
72
+ "loss": 0.5966,
73
+ "step": 45
74
+ },
75
+ {
76
+ "epoch": 0.057529124119085284,
77
+ "grad_norm": 0.09793131053447723,
78
+ "learning_rate": 3.7404580152671756e-05,
79
+ "loss": 0.631,
80
+ "step": 50
81
+ },
82
+ {
83
+ "epoch": 0.06328203653099382,
84
+ "grad_norm": 0.11433672904968262,
85
+ "learning_rate": 4.122137404580153e-05,
86
+ "loss": 0.6258,
87
+ "step": 55
88
+ },
89
+ {
90
+ "epoch": 0.06903494894290234,
91
+ "grad_norm": 0.08854100108146667,
92
+ "learning_rate": 4.5038167938931294e-05,
93
+ "loss": 0.6288,
94
+ "step": 60
95
+ },
96
+ {
97
+ "epoch": 0.07478786135481087,
98
+ "grad_norm": 0.08889709413051605,
99
+ "learning_rate": 4.885496183206107e-05,
100
+ "loss": 0.6121,
101
+ "step": 65
102
+ },
103
+ {
104
+ "epoch": 0.0805407737667194,
105
+ "grad_norm": 0.09466788917779922,
106
+ "learning_rate": 5.267175572519084e-05,
107
+ "loss": 0.6002,
108
+ "step": 70
109
+ },
110
+ {
111
+ "epoch": 0.08629368617862793,
112
+ "grad_norm": 0.1008005142211914,
113
+ "learning_rate": 5.648854961832062e-05,
114
+ "loss": 0.6134,
115
+ "step": 75
116
+ },
117
+ {
118
+ "epoch": 0.09204659859053646,
119
+ "grad_norm": 0.08992719650268555,
120
+ "learning_rate": 6.0305343511450384e-05,
121
+ "loss": 0.6198,
122
+ "step": 80
123
+ },
124
+ {
125
+ "epoch": 0.097799511002445,
126
+ "grad_norm": 0.0943504273891449,
127
+ "learning_rate": 6.412213740458015e-05,
128
+ "loss": 0.6168,
129
+ "step": 85
130
+ },
131
+ {
132
+ "epoch": 0.10355242341435351,
133
+ "grad_norm": 0.080809086561203,
134
+ "learning_rate": 6.793893129770993e-05,
135
+ "loss": 0.6079,
136
+ "step": 90
137
+ },
138
+ {
139
+ "epoch": 0.10930533582626205,
140
+ "grad_norm": 0.09639251977205276,
141
+ "learning_rate": 7.175572519083969e-05,
142
+ "loss": 0.6088,
143
+ "step": 95
144
+ },
145
+ {
146
+ "epoch": 0.11505824823817057,
147
+ "grad_norm": 0.08458084613084793,
148
+ "learning_rate": 7.557251908396947e-05,
149
+ "loss": 0.5698,
150
+ "step": 100
151
+ },
152
+ {
153
+ "epoch": 0.1208111606500791,
154
+ "grad_norm": 0.08075346052646637,
155
+ "learning_rate": 7.938931297709924e-05,
156
+ "loss": 0.6421,
157
+ "step": 105
158
+ },
159
+ {
160
+ "epoch": 0.12656407306198764,
161
+ "grad_norm": 0.08896135538816452,
162
+ "learning_rate": 8.320610687022902e-05,
163
+ "loss": 0.6222,
164
+ "step": 110
165
+ },
166
+ {
167
+ "epoch": 0.13231698547389617,
168
+ "grad_norm": 0.07927868515253067,
169
+ "learning_rate": 8.702290076335878e-05,
170
+ "loss": 0.5968,
171
+ "step": 115
172
+ },
173
+ {
174
+ "epoch": 0.13806989788580468,
175
+ "grad_norm": 0.07641974836587906,
176
+ "learning_rate": 9.083969465648856e-05,
177
+ "loss": 0.601,
178
+ "step": 120
179
+ },
180
+ {
181
+ "epoch": 0.1438228102977132,
182
+ "grad_norm": 0.08499161154031754,
183
+ "learning_rate": 9.465648854961833e-05,
184
+ "loss": 0.5856,
185
+ "step": 125
186
+ },
187
+ {
188
+ "epoch": 0.14957572270962174,
189
+ "grad_norm": 0.06614955514669418,
190
+ "learning_rate": 9.84732824427481e-05,
191
+ "loss": 0.6219,
192
+ "step": 130
193
+ },
194
+ {
195
+ "epoch": 0.15532863512153028,
196
+ "grad_norm": 0.08317167311906815,
197
+ "learning_rate": 9.987898346107302e-05,
198
+ "loss": 0.6316,
199
+ "step": 135
200
+ },
201
+ {
202
+ "epoch": 0.1610815475334388,
203
+ "grad_norm": 0.07750188559293747,
204
+ "learning_rate": 9.967728922952804e-05,
205
+ "loss": 0.6285,
206
+ "step": 140
207
+ },
208
+ {
209
+ "epoch": 0.16683445994534732,
210
+ "grad_norm": 0.07134247571229935,
211
+ "learning_rate": 9.947559499798306e-05,
212
+ "loss": 0.5832,
213
+ "step": 145
214
+ },
215
+ {
216
+ "epoch": 0.17258737235725585,
217
+ "grad_norm": 0.08801887184381485,
218
+ "learning_rate": 9.927390076643809e-05,
219
+ "loss": 0.5836,
220
+ "step": 150
221
+ },
222
+ {
223
+ "epoch": 0.1783402847691644,
224
+ "grad_norm": 0.0803149864077568,
225
+ "learning_rate": 9.907220653489311e-05,
226
+ "loss": 0.6242,
227
+ "step": 155
228
+ },
229
+ {
230
+ "epoch": 0.18409319718107292,
231
+ "grad_norm": 0.07454356551170349,
232
+ "learning_rate": 9.887051230334813e-05,
233
+ "loss": 0.6479,
234
+ "step": 160
235
+ },
236
+ {
237
+ "epoch": 0.18984610959298145,
238
+ "grad_norm": 0.07962540537118912,
239
+ "learning_rate": 9.866881807180315e-05,
240
+ "loss": 0.6109,
241
+ "step": 165
242
+ },
243
+ {
244
+ "epoch": 0.19559902200489,
245
+ "grad_norm": 0.09444350749254227,
246
+ "learning_rate": 9.846712384025818e-05,
247
+ "loss": 0.5989,
248
+ "step": 170
249
+ },
250
+ {
251
+ "epoch": 0.2013519344167985,
252
+ "grad_norm": 0.07151385396718979,
253
+ "learning_rate": 9.82654296087132e-05,
254
+ "loss": 0.6069,
255
+ "step": 175
256
+ },
257
+ {
258
+ "epoch": 0.20710484682870703,
259
+ "grad_norm": 0.07170318812131882,
260
+ "learning_rate": 9.806373537716822e-05,
261
+ "loss": 0.6148,
262
+ "step": 180
263
+ },
264
+ {
265
+ "epoch": 0.21285775924061556,
266
+ "grad_norm": 0.06816047430038452,
267
+ "learning_rate": 9.786204114562324e-05,
268
+ "loss": 0.645,
269
+ "step": 185
270
+ },
271
+ {
272
+ "epoch": 0.2186106716525241,
273
+ "grad_norm": 0.07456276565790176,
274
+ "learning_rate": 9.766034691407826e-05,
275
+ "loss": 0.6319,
276
+ "step": 190
277
+ },
278
+ {
279
+ "epoch": 0.22436358406443263,
280
+ "grad_norm": 0.07059594988822937,
281
+ "learning_rate": 9.745865268253328e-05,
282
+ "loss": 0.6088,
283
+ "step": 195
284
+ },
285
+ {
286
+ "epoch": 0.23011649647634114,
287
+ "grad_norm": 0.07076138257980347,
288
+ "learning_rate": 9.725695845098831e-05,
289
+ "loss": 0.584,
290
+ "step": 200
291
+ },
292
+ {
293
+ "epoch": 0.23586940888824967,
294
+ "grad_norm": 0.06562484055757523,
295
+ "learning_rate": 9.705526421944333e-05,
296
+ "loss": 0.6465,
297
+ "step": 205
298
+ },
299
+ {
300
+ "epoch": 0.2416223213001582,
301
+ "grad_norm": 0.06166617199778557,
302
+ "learning_rate": 9.685356998789835e-05,
303
+ "loss": 0.649,
304
+ "step": 210
305
+ },
306
+ {
307
+ "epoch": 0.24737523371206674,
308
+ "grad_norm": 0.06290106475353241,
309
+ "learning_rate": 9.665187575635337e-05,
310
+ "loss": 0.6021,
311
+ "step": 215
312
+ },
313
+ {
314
+ "epoch": 0.2531281461239753,
315
+ "grad_norm": 0.0670909509062767,
316
+ "learning_rate": 9.64501815248084e-05,
317
+ "loss": 0.6202,
318
+ "step": 220
319
+ },
320
+ {
321
+ "epoch": 0.2588810585358838,
322
+ "grad_norm": 0.07171754539012909,
323
+ "learning_rate": 9.624848729326342e-05,
324
+ "loss": 0.5997,
325
+ "step": 225
326
+ },
327
+ {
328
+ "epoch": 0.26463397094779234,
329
+ "grad_norm": 0.0632658302783966,
330
+ "learning_rate": 9.604679306171844e-05,
331
+ "loss": 0.6028,
332
+ "step": 230
333
+ },
334
+ {
335
+ "epoch": 0.27038688335970085,
336
+ "grad_norm": 0.061919230967760086,
337
+ "learning_rate": 9.584509883017346e-05,
338
+ "loss": 0.6366,
339
+ "step": 235
340
+ },
341
+ {
342
+ "epoch": 0.27613979577160935,
343
+ "grad_norm": 0.06847239285707474,
344
+ "learning_rate": 9.564340459862848e-05,
345
+ "loss": 0.6297,
346
+ "step": 240
347
+ },
348
+ {
349
+ "epoch": 0.2818927081835179,
350
+ "grad_norm": 0.0715722143650055,
351
+ "learning_rate": 9.54417103670835e-05,
352
+ "loss": 0.5975,
353
+ "step": 245
354
+ },
355
+ {
356
+ "epoch": 0.2876456205954264,
357
+ "grad_norm": 0.05896317958831787,
358
+ "learning_rate": 9.524001613553852e-05,
359
+ "loss": 0.6009,
360
+ "step": 250
361
+ },
362
+ {
363
+ "epoch": 0.293398533007335,
364
+ "grad_norm": 0.06568906456232071,
365
+ "learning_rate": 9.503832190399355e-05,
366
+ "loss": 0.6034,
367
+ "step": 255
368
+ },
369
+ {
370
+ "epoch": 0.2991514454192435,
371
+ "grad_norm": 0.06472741067409515,
372
+ "learning_rate": 9.483662767244857e-05,
373
+ "loss": 0.6192,
374
+ "step": 260
375
+ },
376
+ {
377
+ "epoch": 0.304904357831152,
378
+ "grad_norm": 0.07814178615808487,
379
+ "learning_rate": 9.46349334409036e-05,
380
+ "loss": 0.6199,
381
+ "step": 265
382
+ },
383
+ {
384
+ "epoch": 0.31065727024306056,
385
+ "grad_norm": 0.06985796242952347,
386
+ "learning_rate": 9.443323920935862e-05,
387
+ "loss": 0.5859,
388
+ "step": 270
389
+ },
390
+ {
391
+ "epoch": 0.31641018265496906,
392
+ "grad_norm": 0.07870235294103622,
393
+ "learning_rate": 9.423154497781364e-05,
394
+ "loss": 0.5987,
395
+ "step": 275
396
+ },
397
+ {
398
+ "epoch": 0.3221630950668776,
399
+ "grad_norm": 0.06456617265939713,
400
+ "learning_rate": 9.402985074626867e-05,
401
+ "loss": 0.6445,
402
+ "step": 280
403
+ },
404
+ {
405
+ "epoch": 0.32791600747878613,
406
+ "grad_norm": 0.06096259877085686,
407
+ "learning_rate": 9.382815651472369e-05,
408
+ "loss": 0.6332,
409
+ "step": 285
410
+ },
411
+ {
412
+ "epoch": 0.33366891989069464,
413
+ "grad_norm": 0.07445373386144638,
414
+ "learning_rate": 9.362646228317871e-05,
415
+ "loss": 0.6034,
416
+ "step": 290
417
+ },
418
+ {
419
+ "epoch": 0.3394218323026032,
420
+ "grad_norm": 0.0695696771144867,
421
+ "learning_rate": 9.342476805163373e-05,
422
+ "loss": 0.6005,
423
+ "step": 295
424
+ },
425
+ {
426
+ "epoch": 0.3451747447145117,
427
+ "grad_norm": 0.062321294099092484,
428
+ "learning_rate": 9.322307382008875e-05,
429
+ "loss": 0.578,
430
+ "step": 300
431
+ },
432
+ {
433
+ "epoch": 0.35092765712642027,
434
+ "grad_norm": 0.09297119081020355,
435
+ "learning_rate": 9.302137958854377e-05,
436
+ "loss": 0.6418,
437
+ "step": 305
438
+ },
439
+ {
440
+ "epoch": 0.3566805695383288,
441
+ "grad_norm": 0.0628850907087326,
442
+ "learning_rate": 9.28196853569988e-05,
443
+ "loss": 0.614,
444
+ "step": 310
445
+ },
446
+ {
447
+ "epoch": 0.36243348195023734,
448
+ "grad_norm": 0.05998345836997032,
449
+ "learning_rate": 9.261799112545381e-05,
450
+ "loss": 0.6064,
451
+ "step": 315
452
+ },
453
+ {
454
+ "epoch": 0.36818639436214584,
455
+ "grad_norm": 0.09244129806756973,
456
+ "learning_rate": 9.241629689390884e-05,
457
+ "loss": 0.5831,
458
+ "step": 320
459
+ },
460
+ {
461
+ "epoch": 0.37393930677405435,
462
+ "grad_norm": 0.06163156405091286,
463
+ "learning_rate": 9.221460266236386e-05,
464
+ "loss": 0.6028,
465
+ "step": 325
466
+ },
467
+ {
468
+ "epoch": 0.3796922191859629,
469
+ "grad_norm": 0.06200260668992996,
470
+ "learning_rate": 9.201290843081888e-05,
471
+ "loss": 0.611,
472
+ "step": 330
473
+ },
474
+ {
475
+ "epoch": 0.3854451315978714,
476
+ "grad_norm": 0.06909740716218948,
477
+ "learning_rate": 9.181121419927391e-05,
478
+ "loss": 0.6317,
479
+ "step": 335
480
+ },
481
+ {
482
+ "epoch": 0.39119804400978,
483
+ "grad_norm": 0.06346993148326874,
484
+ "learning_rate": 9.160951996772893e-05,
485
+ "loss": 0.6182,
486
+ "step": 340
487
+ },
488
+ {
489
+ "epoch": 0.3969509564216885,
490
+ "grad_norm": 0.06136496737599373,
491
+ "learning_rate": 9.140782573618395e-05,
492
+ "loss": 0.588,
493
+ "step": 345
494
+ },
495
+ {
496
+ "epoch": 0.402703868833597,
497
+ "grad_norm": 0.060762710869312286,
498
+ "learning_rate": 9.120613150463897e-05,
499
+ "loss": 0.6331,
500
+ "step": 350
501
+ },
502
+ {
503
+ "epoch": 0.40845678124550555,
504
+ "grad_norm": 0.06335324048995972,
505
+ "learning_rate": 9.1004437273094e-05,
506
+ "loss": 0.6208,
507
+ "step": 355
508
+ },
509
+ {
510
+ "epoch": 0.41420969365741406,
511
+ "grad_norm": 0.09197219461202621,
512
+ "learning_rate": 9.080274304154903e-05,
513
+ "loss": 0.6144,
514
+ "step": 360
515
+ },
516
+ {
517
+ "epoch": 0.4199626060693226,
518
+ "grad_norm": 0.0645226389169693,
519
+ "learning_rate": 9.060104881000403e-05,
520
+ "loss": 0.5951,
521
+ "step": 365
522
+ },
523
+ {
524
+ "epoch": 0.4257155184812311,
525
+ "grad_norm": 0.059518225491046906,
526
+ "learning_rate": 9.039935457845906e-05,
527
+ "loss": 0.5856,
528
+ "step": 370
529
+ },
530
+ {
531
+ "epoch": 0.43146843089313963,
532
+ "grad_norm": 0.08617997914552689,
533
+ "learning_rate": 9.019766034691408e-05,
534
+ "loss": 0.5628,
535
+ "step": 375
536
+ },
537
+ {
538
+ "epoch": 0.4372213433050482,
539
+ "grad_norm": 0.06207031011581421,
540
+ "learning_rate": 8.99959661153691e-05,
541
+ "loss": 0.6241,
542
+ "step": 380
543
+ },
544
+ {
545
+ "epoch": 0.4429742557169567,
546
+ "grad_norm": 0.07896573096513748,
547
+ "learning_rate": 8.979427188382412e-05,
548
+ "loss": 0.6484,
549
+ "step": 385
550
+ },
551
+ {
552
+ "epoch": 0.44872716812886526,
553
+ "grad_norm": 0.05776543542742729,
554
+ "learning_rate": 8.959257765227915e-05,
555
+ "loss": 0.5995,
556
+ "step": 390
557
+ },
558
+ {
559
+ "epoch": 0.45448008054077377,
560
+ "grad_norm": 0.0684133842587471,
561
+ "learning_rate": 8.939088342073417e-05,
562
+ "loss": 0.6157,
563
+ "step": 395
564
+ },
565
+ {
566
+ "epoch": 0.4602329929526823,
567
+ "grad_norm": 0.07692847400903702,
568
+ "learning_rate": 8.918918918918919e-05,
569
+ "loss": 0.5821,
570
+ "step": 400
571
+ },
572
+ {
573
+ "epoch": 0.46598590536459084,
574
+ "grad_norm": 0.06857931613922119,
575
+ "learning_rate": 8.898749495764421e-05,
576
+ "loss": 0.5866,
577
+ "step": 405
578
+ },
579
+ {
580
+ "epoch": 0.47173881777649934,
581
+ "grad_norm": 0.06455834954977036,
582
+ "learning_rate": 8.878580072609923e-05,
583
+ "loss": 0.6096,
584
+ "step": 410
585
+ },
586
+ {
587
+ "epoch": 0.4774917301884079,
588
+ "grad_norm": 0.06101421266794205,
589
+ "learning_rate": 8.858410649455427e-05,
590
+ "loss": 0.6064,
591
+ "step": 415
592
+ },
593
+ {
594
+ "epoch": 0.4832446426003164,
595
+ "grad_norm": 0.061161015182733536,
596
+ "learning_rate": 8.838241226300929e-05,
597
+ "loss": 0.5564,
598
+ "step": 420
599
+ },
600
+ {
601
+ "epoch": 0.4889975550122249,
602
+ "grad_norm": 0.0558997243642807,
603
+ "learning_rate": 8.818071803146431e-05,
604
+ "loss": 0.5968,
605
+ "step": 425
606
+ },
607
+ {
608
+ "epoch": 0.4947504674241335,
609
+ "grad_norm": 0.07274127751588821,
610
+ "learning_rate": 8.797902379991933e-05,
611
+ "loss": 0.6325,
612
+ "step": 430
613
+ },
614
+ {
615
+ "epoch": 0.500503379836042,
616
+ "grad_norm": 0.05725545436143875,
617
+ "learning_rate": 8.777732956837435e-05,
618
+ "loss": 0.6314,
619
+ "step": 435
620
+ },
621
+ {
622
+ "epoch": 0.5062562922479505,
623
+ "grad_norm": 0.06649309396743774,
624
+ "learning_rate": 8.757563533682937e-05,
625
+ "loss": 0.6159,
626
+ "step": 440
627
+ },
628
+ {
629
+ "epoch": 0.5120092046598591,
630
+ "grad_norm": 0.10541196167469025,
631
+ "learning_rate": 8.737394110528439e-05,
632
+ "loss": 0.5837,
633
+ "step": 445
634
+ },
635
+ {
636
+ "epoch": 0.5177621170717676,
637
+ "grad_norm": 0.0593460276722908,
638
+ "learning_rate": 8.717224687373941e-05,
639
+ "loss": 0.6117,
640
+ "step": 450
641
+ },
642
+ {
643
+ "epoch": 0.5235150294836761,
644
+ "grad_norm": 0.11490193754434586,
645
+ "learning_rate": 8.697055264219443e-05,
646
+ "loss": 0.6403,
647
+ "step": 455
648
+ },
649
+ {
650
+ "epoch": 0.5292679418955847,
651
+ "grad_norm": 0.06850075721740723,
652
+ "learning_rate": 8.676885841064945e-05,
653
+ "loss": 0.6284,
654
+ "step": 460
655
+ },
656
+ {
657
+ "epoch": 0.5350208543074931,
658
+ "grad_norm": 0.05682928115129471,
659
+ "learning_rate": 8.656716417910447e-05,
660
+ "loss": 0.6095,
661
+ "step": 465
662
+ },
663
+ {
664
+ "epoch": 0.5407737667194017,
665
+ "grad_norm": 0.07022815197706223,
666
+ "learning_rate": 8.636546994755951e-05,
667
+ "loss": 0.6125,
668
+ "step": 470
669
+ },
670
+ {
671
+ "epoch": 0.5465266791313103,
672
+ "grad_norm": 0.05265359953045845,
673
+ "learning_rate": 8.616377571601453e-05,
674
+ "loss": 0.5893,
675
+ "step": 475
676
+ },
677
+ {
678
+ "epoch": 0.5522795915432187,
679
+ "grad_norm": 0.06176578998565674,
680
+ "learning_rate": 8.596208148446955e-05,
681
+ "loss": 0.6335,
682
+ "step": 480
683
+ },
684
+ {
685
+ "epoch": 0.5580325039551273,
686
+ "grad_norm": 0.09064935892820358,
687
+ "learning_rate": 8.576038725292457e-05,
688
+ "loss": 0.6323,
689
+ "step": 485
690
+ },
691
+ {
692
+ "epoch": 0.5637854163670358,
693
+ "grad_norm": 0.05583803355693817,
694
+ "learning_rate": 8.555869302137959e-05,
695
+ "loss": 0.6206,
696
+ "step": 490
697
+ },
698
+ {
699
+ "epoch": 0.5695383287789444,
700
+ "grad_norm": 0.06663497537374496,
701
+ "learning_rate": 8.535699878983463e-05,
702
+ "loss": 0.6037,
703
+ "step": 495
704
+ },
705
+ {
706
+ "epoch": 0.5752912411908528,
707
+ "grad_norm": 0.06015989929437637,
708
+ "learning_rate": 8.515530455828963e-05,
709
+ "loss": 0.587,
710
+ "step": 500
711
+ },
712
+ {
713
+ "epoch": 0.5810441536027614,
714
+ "grad_norm": 0.06235828250646591,
715
+ "learning_rate": 8.495361032674465e-05,
716
+ "loss": 0.6413,
717
+ "step": 505
718
+ },
719
+ {
720
+ "epoch": 0.58679706601467,
721
+ "grad_norm": 0.056652236729860306,
722
+ "learning_rate": 8.475191609519967e-05,
723
+ "loss": 0.6527,
724
+ "step": 510
725
+ },
726
+ {
727
+ "epoch": 0.5925499784265784,
728
+ "grad_norm": 0.05927580967545509,
729
+ "learning_rate": 8.45502218636547e-05,
730
+ "loss": 0.6312,
731
+ "step": 515
732
+ },
733
+ {
734
+ "epoch": 0.598302890838487,
735
+ "grad_norm": 0.06264378875494003,
736
+ "learning_rate": 8.434852763210973e-05,
737
+ "loss": 0.5814,
738
+ "step": 520
739
+ },
740
+ {
741
+ "epoch": 0.6040558032503955,
742
+ "grad_norm": 0.05714103579521179,
743
+ "learning_rate": 8.414683340056475e-05,
744
+ "loss": 0.598,
745
+ "step": 525
746
+ },
747
+ {
748
+ "epoch": 0.609808715662304,
749
+ "grad_norm": 0.06520579010248184,
750
+ "learning_rate": 8.394513916901977e-05,
751
+ "loss": 0.6671,
752
+ "step": 530
753
+ },
754
+ {
755
+ "epoch": 0.6155616280742126,
756
+ "grad_norm": 0.0733422264456749,
757
+ "learning_rate": 8.374344493747479e-05,
758
+ "loss": 0.6382,
759
+ "step": 535
760
+ },
761
+ {
762
+ "epoch": 0.6213145404861211,
763
+ "grad_norm": 0.1022220104932785,
764
+ "learning_rate": 8.354175070592981e-05,
765
+ "loss": 0.608,
766
+ "step": 540
767
+ },
768
+ {
769
+ "epoch": 0.6270674528980297,
770
+ "grad_norm": 0.05892520397901535,
771
+ "learning_rate": 8.334005647438483e-05,
772
+ "loss": 0.5964,
773
+ "step": 545
774
+ },
775
+ {
776
+ "epoch": 0.6328203653099381,
777
+ "grad_norm": 0.06006742641329765,
778
+ "learning_rate": 8.313836224283987e-05,
779
+ "loss": 0.6035,
780
+ "step": 550
781
+ },
782
+ {
783
+ "epoch": 0.6385732777218467,
784
+ "grad_norm": 0.06849201023578644,
785
+ "learning_rate": 8.293666801129489e-05,
786
+ "loss": 0.6267,
787
+ "step": 555
788
+ },
789
+ {
790
+ "epoch": 0.6443261901337552,
791
+ "grad_norm": 0.06012120470404625,
792
+ "learning_rate": 8.273497377974991e-05,
793
+ "loss": 0.6527,
794
+ "step": 560
795
+ },
796
+ {
797
+ "epoch": 0.6500791025456637,
798
+ "grad_norm": 0.05690383538603783,
799
+ "learning_rate": 8.253327954820493e-05,
800
+ "loss": 0.6389,
801
+ "step": 565
802
+ },
803
+ {
804
+ "epoch": 0.6558320149575723,
805
+ "grad_norm": 0.07352601736783981,
806
+ "learning_rate": 8.233158531665995e-05,
807
+ "loss": 0.622,
808
+ "step": 570
809
+ },
810
+ {
811
+ "epoch": 0.6615849273694808,
812
+ "grad_norm": 0.04800200089812279,
813
+ "learning_rate": 8.212989108511497e-05,
814
+ "loss": 0.5976,
815
+ "step": 575
816
+ },
817
+ {
818
+ "epoch": 0.6673378397813893,
819
+ "grad_norm": 0.05528654903173447,
820
+ "learning_rate": 8.192819685356999e-05,
821
+ "loss": 0.6407,
822
+ "step": 580
823
+ },
824
+ {
825
+ "epoch": 0.6730907521932978,
826
+ "grad_norm": 0.056925926357507706,
827
+ "learning_rate": 8.172650262202501e-05,
828
+ "loss": 0.589,
829
+ "step": 585
830
+ },
831
+ {
832
+ "epoch": 0.6788436646052064,
833
+ "grad_norm": 0.052472539246082306,
834
+ "learning_rate": 8.152480839048003e-05,
835
+ "loss": 0.6255,
836
+ "step": 590
837
+ },
838
+ {
839
+ "epoch": 0.684596577017115,
840
+ "grad_norm": 0.09734363853931427,
841
+ "learning_rate": 8.132311415893505e-05,
842
+ "loss": 0.5797,
843
+ "step": 595
844
+ },
845
+ {
846
+ "epoch": 0.6903494894290234,
847
+ "grad_norm": 0.056780166923999786,
848
+ "learning_rate": 8.112141992739009e-05,
849
+ "loss": 0.5805,
850
+ "step": 600
851
+ },
852
+ {
853
+ "epoch": 0.696102401840932,
854
+ "grad_norm": 0.05720958113670349,
855
+ "learning_rate": 8.091972569584511e-05,
856
+ "loss": 0.6521,
857
+ "step": 605
858
+ },
859
+ {
860
+ "epoch": 0.7018553142528405,
861
+ "grad_norm": 0.09339389204978943,
862
+ "learning_rate": 8.071803146430013e-05,
863
+ "loss": 0.6536,
864
+ "step": 610
865
+ },
866
+ {
867
+ "epoch": 0.707608226664749,
868
+ "grad_norm": 0.056131768971681595,
869
+ "learning_rate": 8.051633723275515e-05,
870
+ "loss": 0.6109,
871
+ "step": 615
872
+ },
873
+ {
874
+ "epoch": 0.7133611390766575,
875
+ "grad_norm": 0.07572633028030396,
876
+ "learning_rate": 8.031464300121017e-05,
877
+ "loss": 0.5564,
878
+ "step": 620
879
+ },
880
+ {
881
+ "epoch": 0.7191140514885661,
882
+ "grad_norm": 0.057990267872810364,
883
+ "learning_rate": 8.011294876966519e-05,
884
+ "loss": 0.5882,
885
+ "step": 625
886
+ },
887
+ {
888
+ "epoch": 0.7248669639004747,
889
+ "grad_norm": 0.05926734209060669,
890
+ "learning_rate": 7.991125453812022e-05,
891
+ "loss": 0.6272,
892
+ "step": 630
893
+ },
894
+ {
895
+ "epoch": 0.7306198763123831,
896
+ "grad_norm": 0.05420226603746414,
897
+ "learning_rate": 7.970956030657523e-05,
898
+ "loss": 0.6234,
899
+ "step": 635
900
+ },
901
+ {
902
+ "epoch": 0.7363727887242917,
903
+ "grad_norm": 0.05318234860897064,
904
+ "learning_rate": 7.950786607503025e-05,
905
+ "loss": 0.6191,
906
+ "step": 640
907
+ },
908
+ {
909
+ "epoch": 0.7421257011362002,
910
+ "grad_norm": 0.07230158150196075,
911
+ "learning_rate": 7.930617184348527e-05,
912
+ "loss": 0.5885,
913
+ "step": 645
914
+ },
915
+ {
916
+ "epoch": 0.7478786135481087,
917
+ "grad_norm": 0.0505271852016449,
918
+ "learning_rate": 7.910447761194029e-05,
919
+ "loss": 0.6125,
920
+ "step": 650
921
+ },
922
+ {
923
+ "epoch": 0.7536315259600173,
924
+ "grad_norm": 0.060712337493896484,
925
+ "learning_rate": 7.890278338039533e-05,
926
+ "loss": 0.6147,
927
+ "step": 655
928
+ },
929
+ {
930
+ "epoch": 0.7593844383719258,
931
+ "grad_norm": 0.06349541246891022,
932
+ "learning_rate": 7.870108914885035e-05,
933
+ "loss": 0.6565,
934
+ "step": 660
935
+ },
936
+ {
937
+ "epoch": 0.7651373507838343,
938
+ "grad_norm": 0.05819231644272804,
939
+ "learning_rate": 7.849939491730537e-05,
940
+ "loss": 0.5951,
941
+ "step": 665
942
+ },
943
+ {
944
+ "epoch": 0.7708902631957428,
945
+ "grad_norm": 0.06529917567968369,
946
+ "learning_rate": 7.829770068576039e-05,
947
+ "loss": 0.6018,
948
+ "step": 670
949
+ },
950
+ {
951
+ "epoch": 0.7766431756076514,
952
+ "grad_norm": 0.06111083924770355,
953
+ "learning_rate": 7.809600645421541e-05,
954
+ "loss": 0.5976,
955
+ "step": 675
956
+ },
957
+ {
958
+ "epoch": 0.78239608801956,
959
+ "grad_norm": 0.07779537886381149,
960
+ "learning_rate": 7.789431222267044e-05,
961
+ "loss": 0.6253,
962
+ "step": 680
963
+ },
964
+ {
965
+ "epoch": 0.7881490004314684,
966
+ "grad_norm": 0.058719292283058167,
967
+ "learning_rate": 7.769261799112546e-05,
968
+ "loss": 0.6489,
969
+ "step": 685
970
+ },
971
+ {
972
+ "epoch": 0.793901912843377,
973
+ "grad_norm": 0.0564613863825798,
974
+ "learning_rate": 7.749092375958049e-05,
975
+ "loss": 0.6203,
976
+ "step": 690
977
+ },
978
+ {
979
+ "epoch": 0.7996548252552855,
980
+ "grad_norm": 0.07027581334114075,
981
+ "learning_rate": 7.72892295280355e-05,
982
+ "loss": 0.6316,
983
+ "step": 695
984
+ },
985
+ {
986
+ "epoch": 0.805407737667194,
987
+ "grad_norm": 0.05901184678077698,
988
+ "learning_rate": 7.708753529649053e-05,
989
+ "loss": 0.5849,
990
+ "step": 700
991
+ },
992
+ {
993
+ "epoch": 0.8111606500791025,
994
+ "grad_norm": 0.06016521900892258,
995
+ "learning_rate": 7.688584106494555e-05,
996
+ "loss": 0.6243,
997
+ "step": 705
998
+ },
999
+ {
1000
+ "epoch": 0.8169135624910111,
1001
+ "grad_norm": 0.056609705090522766,
1002
+ "learning_rate": 7.668414683340057e-05,
1003
+ "loss": 0.6255,
1004
+ "step": 710
1005
+ },
1006
+ {
1007
+ "epoch": 0.8226664749029196,
1008
+ "grad_norm": 0.08398044109344482,
1009
+ "learning_rate": 7.648245260185559e-05,
1010
+ "loss": 0.6242,
1011
+ "step": 715
1012
+ },
1013
+ {
1014
+ "epoch": 0.8284193873148281,
1015
+ "grad_norm": 0.06481780856847763,
1016
+ "learning_rate": 7.628075837031061e-05,
1017
+ "loss": 0.6152,
1018
+ "step": 720
1019
+ },
1020
+ {
1021
+ "epoch": 0.8341722997267367,
1022
+ "grad_norm": 0.05826560780405998,
1023
+ "learning_rate": 7.607906413876563e-05,
1024
+ "loss": 0.6091,
1025
+ "step": 725
1026
+ },
1027
+ {
1028
+ "epoch": 0.8399252121386452,
1029
+ "grad_norm": 0.06377377361059189,
1030
+ "learning_rate": 7.587736990722065e-05,
1031
+ "loss": 0.6186,
1032
+ "step": 730
1033
+ },
1034
+ {
1035
+ "epoch": 0.8456781245505537,
1036
+ "grad_norm": 0.061146125197410583,
1037
+ "learning_rate": 7.567567567567568e-05,
1038
+ "loss": 0.6415,
1039
+ "step": 735
1040
+ },
1041
+ {
1042
+ "epoch": 0.8514310369624623,
1043
+ "grad_norm": 0.054572734981775284,
1044
+ "learning_rate": 7.54739814441307e-05,
1045
+ "loss": 0.6347,
1046
+ "step": 740
1047
+ },
1048
+ {
1049
+ "epoch": 0.8571839493743708,
1050
+ "grad_norm": 0.07712185382843018,
1051
+ "learning_rate": 7.527228721258573e-05,
1052
+ "loss": 0.5902,
1053
+ "step": 745
1054
+ },
1055
+ {
1056
+ "epoch": 0.8629368617862793,
1057
+ "grad_norm": 0.057356055825948715,
1058
+ "learning_rate": 7.507059298104075e-05,
1059
+ "loss": 0.5783,
1060
+ "step": 750
1061
+ },
1062
+ {
1063
+ "epoch": 0.8686897741981878,
1064
+ "grad_norm": 0.062310561537742615,
1065
+ "learning_rate": 7.486889874949577e-05,
1066
+ "loss": 0.6559,
1067
+ "step": 755
1068
+ },
1069
+ {
1070
+ "epoch": 0.8744426866100964,
1071
+ "grad_norm": 0.06203293800354004,
1072
+ "learning_rate": 7.46672045179508e-05,
1073
+ "loss": 0.6403,
1074
+ "step": 760
1075
+ },
1076
+ {
1077
+ "epoch": 0.8801955990220048,
1078
+ "grad_norm": 0.0544871985912323,
1079
+ "learning_rate": 7.446551028640582e-05,
1080
+ "loss": 0.6006,
1081
+ "step": 765
1082
+ },
1083
+ {
1084
+ "epoch": 0.8859485114339134,
1085
+ "grad_norm": 0.06846991926431656,
1086
+ "learning_rate": 7.426381605486083e-05,
1087
+ "loss": 0.5854,
1088
+ "step": 770
1089
+ },
1090
+ {
1091
+ "epoch": 0.891701423845822,
1092
+ "grad_norm": 0.061747875064611435,
1093
+ "learning_rate": 7.406212182331585e-05,
1094
+ "loss": 0.5967,
1095
+ "step": 775
1096
+ },
1097
+ {
1098
+ "epoch": 0.8974543362577305,
1099
+ "grad_norm": 0.05422506481409073,
1100
+ "learning_rate": 7.386042759177087e-05,
1101
+ "loss": 0.6202,
1102
+ "step": 780
1103
+ },
1104
+ {
1105
+ "epoch": 0.903207248669639,
1106
+ "grad_norm": 0.06522276252508163,
1107
+ "learning_rate": 7.365873336022589e-05,
1108
+ "loss": 0.6299,
1109
+ "step": 785
1110
+ },
1111
+ {
1112
+ "epoch": 0.9089601610815475,
1113
+ "grad_norm": 0.07020463049411774,
1114
+ "learning_rate": 7.345703912868093e-05,
1115
+ "loss": 0.6253,
1116
+ "step": 790
1117
+ },
1118
+ {
1119
+ "epoch": 0.9147130734934561,
1120
+ "grad_norm": 0.06225111708045006,
1121
+ "learning_rate": 7.325534489713595e-05,
1122
+ "loss": 0.6167,
1123
+ "step": 795
1124
+ },
1125
+ {
1126
+ "epoch": 0.9204659859053645,
1127
+ "grad_norm": 0.05463005602359772,
1128
+ "learning_rate": 7.305365066559097e-05,
1129
+ "loss": 0.5876,
1130
+ "step": 800
1131
+ },
1132
+ {
1133
+ "epoch": 0.9262188983172731,
1134
+ "grad_norm": 0.05861882492899895,
1135
+ "learning_rate": 7.285195643404599e-05,
1136
+ "loss": 0.6313,
1137
+ "step": 805
1138
+ },
1139
+ {
1140
+ "epoch": 0.9319718107291817,
1141
+ "grad_norm": 0.05647750571370125,
1142
+ "learning_rate": 7.265026220250101e-05,
1143
+ "loss": 0.625,
1144
+ "step": 810
1145
+ },
1146
+ {
1147
+ "epoch": 0.9377247231410901,
1148
+ "grad_norm": 0.06673680990934372,
1149
+ "learning_rate": 7.244856797095604e-05,
1150
+ "loss": 0.6002,
1151
+ "step": 815
1152
+ },
1153
+ {
1154
+ "epoch": 0.9434776355529987,
1155
+ "grad_norm": 0.0796564593911171,
1156
+ "learning_rate": 7.224687373941106e-05,
1157
+ "loss": 0.579,
1158
+ "step": 820
1159
+ },
1160
+ {
1161
+ "epoch": 0.9492305479649072,
1162
+ "grad_norm": 0.06967715919017792,
1163
+ "learning_rate": 7.204517950786608e-05,
1164
+ "loss": 0.5999,
1165
+ "step": 825
1166
+ },
1167
+ {
1168
+ "epoch": 0.9549834603768158,
1169
+ "grad_norm": 0.06527382880449295,
1170
+ "learning_rate": 7.18434852763211e-05,
1171
+ "loss": 0.6147,
1172
+ "step": 830
1173
+ },
1174
+ {
1175
+ "epoch": 0.9607363727887243,
1176
+ "grad_norm": 0.06165289878845215,
1177
+ "learning_rate": 7.164179104477612e-05,
1178
+ "loss": 0.6531,
1179
+ "step": 835
1180
+ },
1181
+ {
1182
+ "epoch": 0.9664892852006328,
1183
+ "grad_norm": 0.05502866581082344,
1184
+ "learning_rate": 7.144009681323115e-05,
1185
+ "loss": 0.6056,
1186
+ "step": 840
1187
+ },
1188
+ {
1189
+ "epoch": 0.9722421976125414,
1190
+ "grad_norm": 0.10732096433639526,
1191
+ "learning_rate": 7.123840258168617e-05,
1192
+ "loss": 0.5799,
1193
+ "step": 845
1194
+ },
1195
+ {
1196
+ "epoch": 0.9779951100244498,
1197
+ "grad_norm": 0.054910652339458466,
1198
+ "learning_rate": 7.103670835014119e-05,
1199
+ "loss": 0.6171,
1200
+ "step": 850
1201
+ },
1202
+ {
1203
+ "epoch": 0.9837480224363584,
1204
+ "grad_norm": 0.06319953501224518,
1205
+ "learning_rate": 7.083501411859621e-05,
1206
+ "loss": 0.6174,
1207
+ "step": 855
1208
+ },
1209
+ {
1210
+ "epoch": 0.989500934848267,
1211
+ "grad_norm": 0.05299688130617142,
1212
+ "learning_rate": 7.063331988705123e-05,
1213
+ "loss": 0.6164,
1214
+ "step": 860
1215
+ },
1216
+ {
1217
+ "epoch": 0.9952538472601755,
1218
+ "grad_norm": 0.052180588245391846,
1219
+ "learning_rate": 7.043162565550625e-05,
1220
+ "loss": 0.6281,
1221
+ "step": 865
1222
+ },
1223
+ {
1224
+ "epoch": 1.0,
1225
+ "grad_norm": 0.18281374871730804,
1226
+ "learning_rate": 7.022993142396128e-05,
1227
+ "loss": 0.5448,
1228
+ "step": 870
1229
+ }
1230
+ ],
1231
+ "logging_steps": 5,
1232
+ "max_steps": 2610,
1233
+ "num_input_tokens_seen": 0,
1234
+ "num_train_epochs": 3,
1235
+ "save_steps": 500,
1236
+ "stateful_callbacks": {
1237
+ "TrainerControl": {
1238
+ "args": {
1239
+ "should_epoch_stop": false,
1240
+ "should_evaluate": false,
1241
+ "should_log": false,
1242
+ "should_save": true,
1243
+ "should_training_stop": false
1244
+ },
1245
+ "attributes": {}
1246
+ }
1247
+ },
1248
+ "total_flos": 1.1405546259399626e+20,
1249
+ "train_batch_size": 1,
1250
+ "trial_name": null,
1251
+ "trial_params": null
1252
+ }
Qwen3-30B-A3B-Instruct-2507-sft-fusang-swa=2k_sink=100_fadec/checkpoint-870/vocab.json ADDED
The diff for this file is too large to render. See raw diff
 
Qwen3-30B-A3B-Instruct-2507-sft-fusang-swa=2k_sink=100_falayer=25_from0_step2_fadec/checkpoint-870/README.md ADDED
@@ -0,0 +1,207 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ base_model: /share/models/Qwen3-30B-A3B-Instruct-2507
3
+ library_name: peft
4
+ pipeline_tag: text-generation
5
+ tags:
6
+ - base_model:adapter:/share/models/Qwen3-30B-A3B-Instruct-2507
7
+ - lora
8
+ - transformers
9
+ ---
10
+
11
+ # Model Card for Model ID
12
+
13
+ <!-- Provide a quick summary of what the model is/does. -->
14
+
15
+
16
+
17
+ ## Model Details
18
+
19
+ ### Model Description
20
+
21
+ <!-- Provide a longer summary of what this model is. -->
22
+
23
+
24
+
25
+ - **Developed by:** [More Information Needed]
26
+ - **Funded by [optional]:** [More Information Needed]
27
+ - **Shared by [optional]:** [More Information Needed]
28
+ - **Model type:** [More Information Needed]
29
+ - **Language(s) (NLP):** [More Information Needed]
30
+ - **License:** [More Information Needed]
31
+ - **Finetuned from model [optional]:** [More Information Needed]
32
+
33
+ ### Model Sources [optional]
34
+
35
+ <!-- Provide the basic links for the model. -->
36
+
37
+ - **Repository:** [More Information Needed]
38
+ - **Paper [optional]:** [More Information Needed]
39
+ - **Demo [optional]:** [More Information Needed]
40
+
41
+ ## Uses
42
+
43
+ <!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
44
+
45
+ ### Direct Use
46
+
47
+ <!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
48
+
49
+ [More Information Needed]
50
+
51
+ ### Downstream Use [optional]
52
+
53
+ <!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
54
+
55
+ [More Information Needed]
56
+
57
+ ### Out-of-Scope Use
58
+
59
+ <!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
60
+
61
+ [More Information Needed]
62
+
63
+ ## Bias, Risks, and Limitations
64
+
65
+ <!-- This section is meant to convey both technical and sociotechnical limitations. -->
66
+
67
+ [More Information Needed]
68
+
69
+ ### Recommendations
70
+
71
+ <!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
72
+
73
+ Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
74
+
75
+ ## How to Get Started with the Model
76
+
77
+ Use the code below to get started with the model.
78
+
79
+ [More Information Needed]
80
+
81
+ ## Training Details
82
+
83
+ ### Training Data
84
+
85
+ <!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
86
+
87
+ [More Information Needed]
88
+
89
+ ### Training Procedure
90
+
91
+ <!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
92
+
93
+ #### Preprocessing [optional]
94
+
95
+ [More Information Needed]
96
+
97
+
98
+ #### Training Hyperparameters
99
+
100
+ - **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
101
+
102
+ #### Speeds, Sizes, Times [optional]
103
+
104
+ <!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
105
+
106
+ [More Information Needed]
107
+
108
+ ## Evaluation
109
+
110
+ <!-- This section describes the evaluation protocols and provides the results. -->
111
+
112
+ ### Testing Data, Factors & Metrics
113
+
114
+ #### Testing Data
115
+
116
+ <!-- This should link to a Dataset Card if possible. -->
117
+
118
+ [More Information Needed]
119
+
120
+ #### Factors
121
+
122
+ <!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
123
+
124
+ [More Information Needed]
125
+
126
+ #### Metrics
127
+
128
+ <!-- These are the evaluation metrics being used, ideally with a description of why. -->
129
+
130
+ [More Information Needed]
131
+
132
+ ### Results
133
+
134
+ [More Information Needed]
135
+
136
+ #### Summary
137
+
138
+
139
+
140
+ ## Model Examination [optional]
141
+
142
+ <!-- Relevant interpretability work for the model goes here -->
143
+
144
+ [More Information Needed]
145
+
146
+ ## Environmental Impact
147
+
148
+ <!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
149
+
150
+ Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
151
+
152
+ - **Hardware Type:** [More Information Needed]
153
+ - **Hours used:** [More Information Needed]
154
+ - **Cloud Provider:** [More Information Needed]
155
+ - **Compute Region:** [More Information Needed]
156
+ - **Carbon Emitted:** [More Information Needed]
157
+
158
+ ## Technical Specifications [optional]
159
+
160
+ ### Model Architecture and Objective
161
+
162
+ [More Information Needed]
163
+
164
+ ### Compute Infrastructure
165
+
166
+ [More Information Needed]
167
+
168
+ #### Hardware
169
+
170
+ [More Information Needed]
171
+
172
+ #### Software
173
+
174
+ [More Information Needed]
175
+
176
+ ## Citation [optional]
177
+
178
+ <!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
179
+
180
+ **BibTeX:**
181
+
182
+ [More Information Needed]
183
+
184
+ **APA:**
185
+
186
+ [More Information Needed]
187
+
188
+ ## Glossary [optional]
189
+
190
+ <!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
191
+
192
+ [More Information Needed]
193
+
194
+ ## More Information [optional]
195
+
196
+ [More Information Needed]
197
+
198
+ ## Model Card Authors [optional]
199
+
200
+ [More Information Needed]
201
+
202
+ ## Model Card Contact
203
+
204
+ [More Information Needed]
205
+ ### Framework versions
206
+
207
+ - PEFT 0.17.1
Qwen3-30B-A3B-Instruct-2507-sft-fusang-swa=2k_sink=100_falayer=25_from0_step2_fadec/checkpoint-870/merges.txt ADDED
The diff for this file is too large to render. See raw diff
 
Qwen3-30B-A3B-Instruct-2507-sft-fusang-swa=2k_sink=100_falayer=25_from0_step2_fadec/checkpoint-870/special_tokens_map.json ADDED
@@ -0,0 +1,31 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "additional_special_tokens": [
3
+ "<|im_start|>",
4
+ "<|im_end|>",
5
+ "<|object_ref_start|>",
6
+ "<|object_ref_end|>",
7
+ "<|box_start|>",
8
+ "<|box_end|>",
9
+ "<|quad_start|>",
10
+ "<|quad_end|>",
11
+ "<|vision_start|>",
12
+ "<|vision_end|>",
13
+ "<|vision_pad|>",
14
+ "<|image_pad|>",
15
+ "<|video_pad|>"
16
+ ],
17
+ "eos_token": {
18
+ "content": "<|im_end|>",
19
+ "lstrip": false,
20
+ "normalized": false,
21
+ "rstrip": false,
22
+ "single_word": false
23
+ },
24
+ "pad_token": {
25
+ "content": "<|endoftext|>",
26
+ "lstrip": false,
27
+ "normalized": false,
28
+ "rstrip": false,
29
+ "single_word": false
30
+ }
31
+ }
Qwen3-30B-A3B-Instruct-2507-sft-fusang-swa=2k_sink=100_falayer=25_from0_step2_fadec/checkpoint-870/tokenizer_config.json ADDED
@@ -0,0 +1,239 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "add_bos_token": false,
3
+ "add_prefix_space": false,
4
+ "added_tokens_decoder": {
5
+ "151643": {
6
+ "content": "<|endoftext|>",
7
+ "lstrip": false,
8
+ "normalized": false,
9
+ "rstrip": false,
10
+ "single_word": false,
11
+ "special": true
12
+ },
13
+ "151644": {
14
+ "content": "<|im_start|>",
15
+ "lstrip": false,
16
+ "normalized": false,
17
+ "rstrip": false,
18
+ "single_word": false,
19
+ "special": true
20
+ },
21
+ "151645": {
22
+ "content": "<|im_end|>",
23
+ "lstrip": false,
24
+ "normalized": false,
25
+ "rstrip": false,
26
+ "single_word": false,
27
+ "special": true
28
+ },
29
+ "151646": {
30
+ "content": "<|object_ref_start|>",
31
+ "lstrip": false,
32
+ "normalized": false,
33
+ "rstrip": false,
34
+ "single_word": false,
35
+ "special": true
36
+ },
37
+ "151647": {
38
+ "content": "<|object_ref_end|>",
39
+ "lstrip": false,
40
+ "normalized": false,
41
+ "rstrip": false,
42
+ "single_word": false,
43
+ "special": true
44
+ },
45
+ "151648": {
46
+ "content": "<|box_start|>",
47
+ "lstrip": false,
48
+ "normalized": false,
49
+ "rstrip": false,
50
+ "single_word": false,
51
+ "special": true
52
+ },
53
+ "151649": {
54
+ "content": "<|box_end|>",
55
+ "lstrip": false,
56
+ "normalized": false,
57
+ "rstrip": false,
58
+ "single_word": false,
59
+ "special": true
60
+ },
61
+ "151650": {
62
+ "content": "<|quad_start|>",
63
+ "lstrip": false,
64
+ "normalized": false,
65
+ "rstrip": false,
66
+ "single_word": false,
67
+ "special": true
68
+ },
69
+ "151651": {
70
+ "content": "<|quad_end|>",
71
+ "lstrip": false,
72
+ "normalized": false,
73
+ "rstrip": false,
74
+ "single_word": false,
75
+ "special": true
76
+ },
77
+ "151652": {
78
+ "content": "<|vision_start|>",
79
+ "lstrip": false,
80
+ "normalized": false,
81
+ "rstrip": false,
82
+ "single_word": false,
83
+ "special": true
84
+ },
85
+ "151653": {
86
+ "content": "<|vision_end|>",
87
+ "lstrip": false,
88
+ "normalized": false,
89
+ "rstrip": false,
90
+ "single_word": false,
91
+ "special": true
92
+ },
93
+ "151654": {
94
+ "content": "<|vision_pad|>",
95
+ "lstrip": false,
96
+ "normalized": false,
97
+ "rstrip": false,
98
+ "single_word": false,
99
+ "special": true
100
+ },
101
+ "151655": {
102
+ "content": "<|image_pad|>",
103
+ "lstrip": false,
104
+ "normalized": false,
105
+ "rstrip": false,
106
+ "single_word": false,
107
+ "special": true
108
+ },
109
+ "151656": {
110
+ "content": "<|video_pad|>",
111
+ "lstrip": false,
112
+ "normalized": false,
113
+ "rstrip": false,
114
+ "single_word": false,
115
+ "special": true
116
+ },
117
+ "151657": {
118
+ "content": "<tool_call>",
119
+ "lstrip": false,
120
+ "normalized": false,
121
+ "rstrip": false,
122
+ "single_word": false,
123
+ "special": false
124
+ },
125
+ "151658": {
126
+ "content": "</tool_call>",
127
+ "lstrip": false,
128
+ "normalized": false,
129
+ "rstrip": false,
130
+ "single_word": false,
131
+ "special": false
132
+ },
133
+ "151659": {
134
+ "content": "<|fim_prefix|>",
135
+ "lstrip": false,
136
+ "normalized": false,
137
+ "rstrip": false,
138
+ "single_word": false,
139
+ "special": false
140
+ },
141
+ "151660": {
142
+ "content": "<|fim_middle|>",
143
+ "lstrip": false,
144
+ "normalized": false,
145
+ "rstrip": false,
146
+ "single_word": false,
147
+ "special": false
148
+ },
149
+ "151661": {
150
+ "content": "<|fim_suffix|>",
151
+ "lstrip": false,
152
+ "normalized": false,
153
+ "rstrip": false,
154
+ "single_word": false,
155
+ "special": false
156
+ },
157
+ "151662": {
158
+ "content": "<|fim_pad|>",
159
+ "lstrip": false,
160
+ "normalized": false,
161
+ "rstrip": false,
162
+ "single_word": false,
163
+ "special": false
164
+ },
165
+ "151663": {
166
+ "content": "<|repo_name|>",
167
+ "lstrip": false,
168
+ "normalized": false,
169
+ "rstrip": false,
170
+ "single_word": false,
171
+ "special": false
172
+ },
173
+ "151664": {
174
+ "content": "<|file_sep|>",
175
+ "lstrip": false,
176
+ "normalized": false,
177
+ "rstrip": false,
178
+ "single_word": false,
179
+ "special": false
180
+ },
181
+ "151665": {
182
+ "content": "<tool_response>",
183
+ "lstrip": false,
184
+ "normalized": false,
185
+ "rstrip": false,
186
+ "single_word": false,
187
+ "special": false
188
+ },
189
+ "151666": {
190
+ "content": "</tool_response>",
191
+ "lstrip": false,
192
+ "normalized": false,
193
+ "rstrip": false,
194
+ "single_word": false,
195
+ "special": false
196
+ },
197
+ "151667": {
198
+ "content": "<think>",
199
+ "lstrip": false,
200
+ "normalized": false,
201
+ "rstrip": false,
202
+ "single_word": false,
203
+ "special": false
204
+ },
205
+ "151668": {
206
+ "content": "</think>",
207
+ "lstrip": false,
208
+ "normalized": false,
209
+ "rstrip": false,
210
+ "single_word": false,
211
+ "special": false
212
+ }
213
+ },
214
+ "additional_special_tokens": [
215
+ "<|im_start|>",
216
+ "<|im_end|>",
217
+ "<|object_ref_start|>",
218
+ "<|object_ref_end|>",
219
+ "<|box_start|>",
220
+ "<|box_end|>",
221
+ "<|quad_start|>",
222
+ "<|quad_end|>",
223
+ "<|vision_start|>",
224
+ "<|vision_end|>",
225
+ "<|vision_pad|>",
226
+ "<|image_pad|>",
227
+ "<|video_pad|>"
228
+ ],
229
+ "bos_token": null,
230
+ "clean_up_tokenization_spaces": false,
231
+ "eos_token": "<|im_end|>",
232
+ "errors": "replace",
233
+ "extra_special_tokens": {},
234
+ "model_max_length": 1010000,
235
+ "pad_token": "<|endoftext|>",
236
+ "split_special_tokens": false,
237
+ "tokenizer_class": "Qwen2Tokenizer",
238
+ "unk_token": null
239
+ }
Qwen3-30B-A3B-Instruct-2507-sft-fusang-swa=2k_sink=100_falayer=25_from0_step2_fadec/checkpoint-870/trainer_state.json ADDED
@@ -0,0 +1,1252 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "best_global_step": null,
3
+ "best_metric": null,
4
+ "best_model_checkpoint": null,
5
+ "epoch": 1.0,
6
+ "eval_steps": 500,
7
+ "global_step": 870,
8
+ "is_hyper_param_search": false,
9
+ "is_local_process_zero": true,
10
+ "is_world_process_zero": true,
11
+ "log_history": [
12
+ {
13
+ "epoch": 0.005752912411908529,
14
+ "grad_norm": 0.09720346331596375,
15
+ "learning_rate": 9.090909090909091e-06,
16
+ "loss": 0.6157,
17
+ "step": 5
18
+ },
19
+ {
20
+ "epoch": 0.011505824823817058,
21
+ "grad_norm": 0.08264312148094177,
22
+ "learning_rate": 2.0454545454545457e-05,
23
+ "loss": 0.6159,
24
+ "step": 10
25
+ },
26
+ {
27
+ "epoch": 0.017258737235725585,
28
+ "grad_norm": 0.09239840507507324,
29
+ "learning_rate": 3.181818181818182e-05,
30
+ "loss": 0.5759,
31
+ "step": 15
32
+ },
33
+ {
34
+ "epoch": 0.023011649647634115,
35
+ "grad_norm": 0.08070536702871323,
36
+ "learning_rate": 4.318181818181819e-05,
37
+ "loss": 0.5878,
38
+ "step": 20
39
+ },
40
+ {
41
+ "epoch": 0.028764562059542642,
42
+ "grad_norm": 0.08913712948560715,
43
+ "learning_rate": 5.4545454545454546e-05,
44
+ "loss": 0.6091,
45
+ "step": 25
46
+ },
47
+ {
48
+ "epoch": 0.03451747447145117,
49
+ "grad_norm": 0.08629795163869858,
50
+ "learning_rate": 6.59090909090909e-05,
51
+ "loss": 0.6375,
52
+ "step": 30
53
+ },
54
+ {
55
+ "epoch": 0.0402703868833597,
56
+ "grad_norm": 0.06990034133195877,
57
+ "learning_rate": 7.727272727272727e-05,
58
+ "loss": 0.6347,
59
+ "step": 35
60
+ },
61
+ {
62
+ "epoch": 0.04602329929526823,
63
+ "grad_norm": 0.09788878262042999,
64
+ "learning_rate": 8.863636363636364e-05,
65
+ "loss": 0.6552,
66
+ "step": 40
67
+ },
68
+ {
69
+ "epoch": 0.05177621170717676,
70
+ "grad_norm": 0.0843987911939621,
71
+ "learning_rate": 0.0001,
72
+ "loss": 0.5711,
73
+ "step": 45
74
+ },
75
+ {
76
+ "epoch": 0.057529124119085284,
77
+ "grad_norm": 0.07418220490217209,
78
+ "learning_rate": 9.93946731234867e-05,
79
+ "loss": 0.6091,
80
+ "step": 50
81
+ },
82
+ {
83
+ "epoch": 0.06328203653099382,
84
+ "grad_norm": 0.08357148617506027,
85
+ "learning_rate": 9.878934624697337e-05,
86
+ "loss": 0.5999,
87
+ "step": 55
88
+ },
89
+ {
90
+ "epoch": 0.06903494894290234,
91
+ "grad_norm": 0.06321673095226288,
92
+ "learning_rate": 9.818401937046005e-05,
93
+ "loss": 0.6049,
94
+ "step": 60
95
+ },
96
+ {
97
+ "epoch": 0.07478786135481087,
98
+ "grad_norm": 0.0657491460442543,
99
+ "learning_rate": 9.757869249394673e-05,
100
+ "loss": 0.5894,
101
+ "step": 65
102
+ },
103
+ {
104
+ "epoch": 0.0805407737667194,
105
+ "grad_norm": 0.06759776175022125,
106
+ "learning_rate": 9.697336561743342e-05,
107
+ "loss": 0.5816,
108
+ "step": 70
109
+ },
110
+ {
111
+ "epoch": 0.08629368617862793,
112
+ "grad_norm": 0.0641172006726265,
113
+ "learning_rate": 9.63680387409201e-05,
114
+ "loss": 0.5947,
115
+ "step": 75
116
+ },
117
+ {
118
+ "epoch": 0.09204659859053646,
119
+ "grad_norm": 0.0725429356098175,
120
+ "learning_rate": 9.576271186440679e-05,
121
+ "loss": 0.595,
122
+ "step": 80
123
+ },
124
+ {
125
+ "epoch": 0.097799511002445,
126
+ "grad_norm": 0.06664568185806274,
127
+ "learning_rate": 9.515738498789348e-05,
128
+ "loss": 0.5964,
129
+ "step": 85
130
+ },
131
+ {
132
+ "epoch": 0.10355242341435351,
133
+ "grad_norm": 0.06276047974824905,
134
+ "learning_rate": 9.455205811138015e-05,
135
+ "loss": 0.5885,
136
+ "step": 90
137
+ },
138
+ {
139
+ "epoch": 0.10930533582626205,
140
+ "grad_norm": 0.06737831979990005,
141
+ "learning_rate": 9.394673123486683e-05,
142
+ "loss": 0.5895,
143
+ "step": 95
144
+ },
145
+ {
146
+ "epoch": 0.11505824823817057,
147
+ "grad_norm": 0.05885528400540352,
148
+ "learning_rate": 9.334140435835351e-05,
149
+ "loss": 0.5525,
150
+ "step": 100
151
+ },
152
+ {
153
+ "epoch": 0.1208111606500791,
154
+ "grad_norm": 0.07103564590215683,
155
+ "learning_rate": 9.27360774818402e-05,
156
+ "loss": 0.6206,
157
+ "step": 105
158
+ },
159
+ {
160
+ "epoch": 0.12656407306198764,
161
+ "grad_norm": 0.062482357025146484,
162
+ "learning_rate": 9.213075060532688e-05,
163
+ "loss": 0.6005,
164
+ "step": 110
165
+ },
166
+ {
167
+ "epoch": 0.13231698547389617,
168
+ "grad_norm": 0.056238967925310135,
169
+ "learning_rate": 9.152542372881357e-05,
170
+ "loss": 0.5752,
171
+ "step": 115
172
+ },
173
+ {
174
+ "epoch": 0.13806989788580468,
175
+ "grad_norm": 0.05633089691400528,
176
+ "learning_rate": 9.092009685230026e-05,
177
+ "loss": 0.5817,
178
+ "step": 120
179
+ },
180
+ {
181
+ "epoch": 0.1438228102977132,
182
+ "grad_norm": 0.058726903051137924,
183
+ "learning_rate": 9.031476997578692e-05,
184
+ "loss": 0.5693,
185
+ "step": 125
186
+ },
187
+ {
188
+ "epoch": 0.14957572270962174,
189
+ "grad_norm": 0.05092104524374008,
190
+ "learning_rate": 8.970944309927361e-05,
191
+ "loss": 0.6025,
192
+ "step": 130
193
+ },
194
+ {
195
+ "epoch": 0.15532863512153028,
196
+ "grad_norm": 0.05793561413884163,
197
+ "learning_rate": 8.910411622276029e-05,
198
+ "loss": 0.6117,
199
+ "step": 135
200
+ },
201
+ {
202
+ "epoch": 0.1610815475334388,
203
+ "grad_norm": 0.05897874757647514,
204
+ "learning_rate": 8.849878934624698e-05,
205
+ "loss": 0.6104,
206
+ "step": 140
207
+ },
208
+ {
209
+ "epoch": 0.16683445994534732,
210
+ "grad_norm": 0.054141078144311905,
211
+ "learning_rate": 8.789346246973366e-05,
212
+ "loss": 0.5664,
213
+ "step": 145
214
+ },
215
+ {
216
+ "epoch": 0.17258737235725585,
217
+ "grad_norm": 0.06559433788061142,
218
+ "learning_rate": 8.728813559322035e-05,
219
+ "loss": 0.5655,
220
+ "step": 150
221
+ },
222
+ {
223
+ "epoch": 0.1783402847691644,
224
+ "grad_norm": 0.056958429515361786,
225
+ "learning_rate": 8.668280871670703e-05,
226
+ "loss": 0.6045,
227
+ "step": 155
228
+ },
229
+ {
230
+ "epoch": 0.18409319718107292,
231
+ "grad_norm": 0.054936036467552185,
232
+ "learning_rate": 8.60774818401937e-05,
233
+ "loss": 0.6286,
234
+ "step": 160
235
+ },
236
+ {
237
+ "epoch": 0.18984610959298145,
238
+ "grad_norm": 0.0556117407977581,
239
+ "learning_rate": 8.54721549636804e-05,
240
+ "loss": 0.5918,
241
+ "step": 165
242
+ },
243
+ {
244
+ "epoch": 0.19559902200489,
245
+ "grad_norm": 0.06498974561691284,
246
+ "learning_rate": 8.486682808716707e-05,
247
+ "loss": 0.5793,
248
+ "step": 170
249
+ },
250
+ {
251
+ "epoch": 0.2013519344167985,
252
+ "grad_norm": 0.054659437388181686,
253
+ "learning_rate": 8.426150121065376e-05,
254
+ "loss": 0.5898,
255
+ "step": 175
256
+ },
257
+ {
258
+ "epoch": 0.20710484682870703,
259
+ "grad_norm": 0.05693948268890381,
260
+ "learning_rate": 8.365617433414044e-05,
261
+ "loss": 0.5964,
262
+ "step": 180
263
+ },
264
+ {
265
+ "epoch": 0.21285775924061556,
266
+ "grad_norm": 0.0551651269197464,
267
+ "learning_rate": 8.305084745762712e-05,
268
+ "loss": 0.6269,
269
+ "step": 185
270
+ },
271
+ {
272
+ "epoch": 0.2186106716525241,
273
+ "grad_norm": 0.06025553494691849,
274
+ "learning_rate": 8.244552058111381e-05,
275
+ "loss": 0.6133,
276
+ "step": 190
277
+ },
278
+ {
279
+ "epoch": 0.22436358406443263,
280
+ "grad_norm": 0.061302103102207184,
281
+ "learning_rate": 8.184019370460048e-05,
282
+ "loss": 0.5916,
283
+ "step": 195
284
+ },
285
+ {
286
+ "epoch": 0.23011649647634114,
287
+ "grad_norm": 0.05799054354429245,
288
+ "learning_rate": 8.123486682808718e-05,
289
+ "loss": 0.5664,
290
+ "step": 200
291
+ },
292
+ {
293
+ "epoch": 0.23586940888824967,
294
+ "grad_norm": 0.04856674745678902,
295
+ "learning_rate": 8.062953995157385e-05,
296
+ "loss": 0.6268,
297
+ "step": 205
298
+ },
299
+ {
300
+ "epoch": 0.2416223213001582,
301
+ "grad_norm": 0.053800806403160095,
302
+ "learning_rate": 8.002421307506054e-05,
303
+ "loss": 0.6304,
304
+ "step": 210
305
+ },
306
+ {
307
+ "epoch": 0.24737523371206674,
308
+ "grad_norm": 0.05287880450487137,
309
+ "learning_rate": 7.941888619854722e-05,
310
+ "loss": 0.5846,
311
+ "step": 215
312
+ },
313
+ {
314
+ "epoch": 0.2531281461239753,
315
+ "grad_norm": 0.05466080084443092,
316
+ "learning_rate": 7.88135593220339e-05,
317
+ "loss": 0.6036,
318
+ "step": 220
319
+ },
320
+ {
321
+ "epoch": 0.2588810585358838,
322
+ "grad_norm": 0.05524852126836777,
323
+ "learning_rate": 7.820823244552058e-05,
324
+ "loss": 0.5841,
325
+ "step": 225
326
+ },
327
+ {
328
+ "epoch": 0.26463397094779234,
329
+ "grad_norm": 0.05395347625017166,
330
+ "learning_rate": 7.760290556900727e-05,
331
+ "loss": 0.583,
332
+ "step": 230
333
+ },
334
+ {
335
+ "epoch": 0.27038688335970085,
336
+ "grad_norm": 0.04851017892360687,
337
+ "learning_rate": 7.699757869249396e-05,
338
+ "loss": 0.6182,
339
+ "step": 235
340
+ },
341
+ {
342
+ "epoch": 0.27613979577160935,
343
+ "grad_norm": 0.05496341362595558,
344
+ "learning_rate": 7.639225181598063e-05,
345
+ "loss": 0.6133,
346
+ "step": 240
347
+ },
348
+ {
349
+ "epoch": 0.2818927081835179,
350
+ "grad_norm": 0.06468776613473892,
351
+ "learning_rate": 7.578692493946733e-05,
352
+ "loss": 0.58,
353
+ "step": 245
354
+ },
355
+ {
356
+ "epoch": 0.2876456205954264,
357
+ "grad_norm": 0.04942385479807854,
358
+ "learning_rate": 7.518159806295399e-05,
359
+ "loss": 0.5852,
360
+ "step": 250
361
+ },
362
+ {
363
+ "epoch": 0.293398533007335,
364
+ "grad_norm": 0.056398842483758926,
365
+ "learning_rate": 7.457627118644068e-05,
366
+ "loss": 0.5848,
367
+ "step": 255
368
+ },
369
+ {
370
+ "epoch": 0.2991514454192435,
371
+ "grad_norm": 0.05320358648896217,
372
+ "learning_rate": 7.397094430992736e-05,
373
+ "loss": 0.6006,
374
+ "step": 260
375
+ },
376
+ {
377
+ "epoch": 0.304904357831152,
378
+ "grad_norm": 0.044961217790842056,
379
+ "learning_rate": 7.336561743341405e-05,
380
+ "loss": 0.6032,
381
+ "step": 265
382
+ },
383
+ {
384
+ "epoch": 0.31065727024306056,
385
+ "grad_norm": 0.05962919071316719,
386
+ "learning_rate": 7.276029055690074e-05,
387
+ "loss": 0.5683,
388
+ "step": 270
389
+ },
390
+ {
391
+ "epoch": 0.31641018265496906,
392
+ "grad_norm": 0.05497664213180542,
393
+ "learning_rate": 7.215496368038742e-05,
394
+ "loss": 0.5822,
395
+ "step": 275
396
+ },
397
+ {
398
+ "epoch": 0.3221630950668776,
399
+ "grad_norm": 0.05444960296154022,
400
+ "learning_rate": 7.154963680387409e-05,
401
+ "loss": 0.6252,
402
+ "step": 280
403
+ },
404
+ {
405
+ "epoch": 0.32791600747878613,
406
+ "grad_norm": 0.05389095097780228,
407
+ "learning_rate": 7.094430992736077e-05,
408
+ "loss": 0.6166,
409
+ "step": 285
410
+ },
411
+ {
412
+ "epoch": 0.33366891989069464,
413
+ "grad_norm": 0.057898715138435364,
414
+ "learning_rate": 7.033898305084746e-05,
415
+ "loss": 0.5863,
416
+ "step": 290
417
+ },
418
+ {
419
+ "epoch": 0.3394218323026032,
420
+ "grad_norm": 0.050731223076581955,
421
+ "learning_rate": 6.973365617433414e-05,
422
+ "loss": 0.5842,
423
+ "step": 295
424
+ },
425
+ {
426
+ "epoch": 0.3451747447145117,
427
+ "grad_norm": 0.04853472486138344,
428
+ "learning_rate": 6.912832929782083e-05,
429
+ "loss": 0.5602,
430
+ "step": 300
431
+ },
432
+ {
433
+ "epoch": 0.35092765712642027,
434
+ "grad_norm": 0.05420827865600586,
435
+ "learning_rate": 6.852300242130752e-05,
436
+ "loss": 0.6215,
437
+ "step": 305
438
+ },
439
+ {
440
+ "epoch": 0.3566805695383288,
441
+ "grad_norm": 0.04938114807009697,
442
+ "learning_rate": 6.791767554479418e-05,
443
+ "loss": 0.5975,
444
+ "step": 310
445
+ },
446
+ {
447
+ "epoch": 0.36243348195023734,
448
+ "grad_norm": 0.05102991312742233,
449
+ "learning_rate": 6.731234866828087e-05,
450
+ "loss": 0.5888,
451
+ "step": 315
452
+ },
453
+ {
454
+ "epoch": 0.36818639436214584,
455
+ "grad_norm": 0.058827292174100876,
456
+ "learning_rate": 6.670702179176755e-05,
457
+ "loss": 0.5655,
458
+ "step": 320
459
+ },
460
+ {
461
+ "epoch": 0.37393930677405435,
462
+ "grad_norm": 0.05561719089746475,
463
+ "learning_rate": 6.610169491525424e-05,
464
+ "loss": 0.5866,
465
+ "step": 325
466
+ },
467
+ {
468
+ "epoch": 0.3796922191859629,
469
+ "grad_norm": 0.05027098208665848,
470
+ "learning_rate": 6.549636803874092e-05,
471
+ "loss": 0.5919,
472
+ "step": 330
473
+ },
474
+ {
475
+ "epoch": 0.3854451315978714,
476
+ "grad_norm": 0.059192024171352386,
477
+ "learning_rate": 6.489104116222761e-05,
478
+ "loss": 0.6146,
479
+ "step": 335
480
+ },
481
+ {
482
+ "epoch": 0.39119804400978,
483
+ "grad_norm": 0.05287157744169235,
484
+ "learning_rate": 6.428571428571429e-05,
485
+ "loss": 0.5997,
486
+ "step": 340
487
+ },
488
+ {
489
+ "epoch": 0.3969509564216885,
490
+ "grad_norm": 0.05061454325914383,
491
+ "learning_rate": 6.368038740920097e-05,
492
+ "loss": 0.5697,
493
+ "step": 345
494
+ },
495
+ {
496
+ "epoch": 0.402703868833597,
497
+ "grad_norm": 0.05024266615509987,
498
+ "learning_rate": 6.307506053268766e-05,
499
+ "loss": 0.6183,
500
+ "step": 350
501
+ },
502
+ {
503
+ "epoch": 0.40845678124550555,
504
+ "grad_norm": 0.05443001911044121,
505
+ "learning_rate": 6.246973365617433e-05,
506
+ "loss": 0.6024,
507
+ "step": 355
508
+ },
509
+ {
510
+ "epoch": 0.41420969365741406,
511
+ "grad_norm": 0.05379893258213997,
512
+ "learning_rate": 6.186440677966102e-05,
513
+ "loss": 0.5984,
514
+ "step": 360
515
+ },
516
+ {
517
+ "epoch": 0.4199626060693226,
518
+ "grad_norm": 0.0590442530810833,
519
+ "learning_rate": 6.12590799031477e-05,
520
+ "loss": 0.5771,
521
+ "step": 365
522
+ },
523
+ {
524
+ "epoch": 0.4257155184812311,
525
+ "grad_norm": 0.049811460077762604,
526
+ "learning_rate": 6.0653753026634385e-05,
527
+ "loss": 0.5693,
528
+ "step": 370
529
+ },
530
+ {
531
+ "epoch": 0.43146843089313963,
532
+ "grad_norm": 0.05411714315414429,
533
+ "learning_rate": 6.0048426150121076e-05,
534
+ "loss": 0.5481,
535
+ "step": 375
536
+ },
537
+ {
538
+ "epoch": 0.4372213433050482,
539
+ "grad_norm": 0.05528920516371727,
540
+ "learning_rate": 5.9443099273607746e-05,
541
+ "loss": 0.6075,
542
+ "step": 380
543
+ },
544
+ {
545
+ "epoch": 0.4429742557169567,
546
+ "grad_norm": 0.058539703488349915,
547
+ "learning_rate": 5.883777239709444e-05,
548
+ "loss": 0.6308,
549
+ "step": 385
550
+ },
551
+ {
552
+ "epoch": 0.44872716812886526,
553
+ "grad_norm": 0.0678466260433197,
554
+ "learning_rate": 5.8232445520581115e-05,
555
+ "loss": 0.5817,
556
+ "step": 390
557
+ },
558
+ {
559
+ "epoch": 0.45448008054077377,
560
+ "grad_norm": 0.06085991859436035,
561
+ "learning_rate": 5.76271186440678e-05,
562
+ "loss": 0.5974,
563
+ "step": 395
564
+ },
565
+ {
566
+ "epoch": 0.4602329929526823,
567
+ "grad_norm": 0.05267111957073212,
568
+ "learning_rate": 5.7021791767554476e-05,
569
+ "loss": 0.5671,
570
+ "step": 400
571
+ },
572
+ {
573
+ "epoch": 0.46598590536459084,
574
+ "grad_norm": 0.05617913976311684,
575
+ "learning_rate": 5.641646489104117e-05,
576
+ "loss": 0.5683,
577
+ "step": 405
578
+ },
579
+ {
580
+ "epoch": 0.47173881777649934,
581
+ "grad_norm": 0.05813376232981682,
582
+ "learning_rate": 5.581113801452785e-05,
583
+ "loss": 0.5914,
584
+ "step": 410
585
+ },
586
+ {
587
+ "epoch": 0.4774917301884079,
588
+ "grad_norm": 0.05124270170927048,
589
+ "learning_rate": 5.520581113801453e-05,
590
+ "loss": 0.5915,
591
+ "step": 415
592
+ },
593
+ {
594
+ "epoch": 0.4832446426003164,
595
+ "grad_norm": 0.050044868141412735,
596
+ "learning_rate": 5.460048426150122e-05,
597
+ "loss": 0.5394,
598
+ "step": 420
599
+ },
600
+ {
601
+ "epoch": 0.4889975550122249,
602
+ "grad_norm": 0.050365570932626724,
603
+ "learning_rate": 5.3995157384987896e-05,
604
+ "loss": 0.5817,
605
+ "step": 425
606
+ },
607
+ {
608
+ "epoch": 0.4947504674241335,
609
+ "grad_norm": 0.05484164506196976,
610
+ "learning_rate": 5.338983050847458e-05,
611
+ "loss": 0.612,
612
+ "step": 430
613
+ },
614
+ {
615
+ "epoch": 0.500503379836042,
616
+ "grad_norm": 0.051003798842430115,
617
+ "learning_rate": 5.278450363196126e-05,
618
+ "loss": 0.6136,
619
+ "step": 435
620
+ },
621
+ {
622
+ "epoch": 0.5062562922479505,
623
+ "grad_norm": 0.05378809571266174,
624
+ "learning_rate": 5.217917675544795e-05,
625
+ "loss": 0.5996,
626
+ "step": 440
627
+ },
628
+ {
629
+ "epoch": 0.5120092046598591,
630
+ "grad_norm": 0.19749966263771057,
631
+ "learning_rate": 5.157384987893463e-05,
632
+ "loss": 0.5673,
633
+ "step": 445
634
+ },
635
+ {
636
+ "epoch": 0.5177621170717676,
637
+ "grad_norm": 0.04995080083608627,
638
+ "learning_rate": 5.096852300242131e-05,
639
+ "loss": 0.5965,
640
+ "step": 450
641
+ },
642
+ {
643
+ "epoch": 0.5235150294836761,
644
+ "grad_norm": 0.05096307769417763,
645
+ "learning_rate": 5.036319612590799e-05,
646
+ "loss": 0.6228,
647
+ "step": 455
648
+ },
649
+ {
650
+ "epoch": 0.5292679418955847,
651
+ "grad_norm": 0.057827964425086975,
652
+ "learning_rate": 4.975786924939467e-05,
653
+ "loss": 0.6114,
654
+ "step": 460
655
+ },
656
+ {
657
+ "epoch": 0.5350208543074931,
658
+ "grad_norm": 0.050771113485097885,
659
+ "learning_rate": 4.915254237288136e-05,
660
+ "loss": 0.5919,
661
+ "step": 465
662
+ },
663
+ {
664
+ "epoch": 0.5407737667194017,
665
+ "grad_norm": 0.05562472715973854,
666
+ "learning_rate": 4.8547215496368045e-05,
667
+ "loss": 0.5957,
668
+ "step": 470
669
+ },
670
+ {
671
+ "epoch": 0.5465266791313103,
672
+ "grad_norm": 0.044839441776275635,
673
+ "learning_rate": 4.794188861985472e-05,
674
+ "loss": 0.5733,
675
+ "step": 475
676
+ },
677
+ {
678
+ "epoch": 0.5522795915432187,
679
+ "grad_norm": 0.05540219694375992,
680
+ "learning_rate": 4.733656174334141e-05,
681
+ "loss": 0.6142,
682
+ "step": 480
683
+ },
684
+ {
685
+ "epoch": 0.5580325039551273,
686
+ "grad_norm": 0.05488353222608566,
687
+ "learning_rate": 4.673123486682809e-05,
688
+ "loss": 0.6153,
689
+ "step": 485
690
+ },
691
+ {
692
+ "epoch": 0.5637854163670358,
693
+ "grad_norm": 0.14343105256557465,
694
+ "learning_rate": 4.612590799031477e-05,
695
+ "loss": 0.6032,
696
+ "step": 490
697
+ },
698
+ {
699
+ "epoch": 0.5695383287789444,
700
+ "grad_norm": 0.05371823161840439,
701
+ "learning_rate": 4.552058111380145e-05,
702
+ "loss": 0.5849,
703
+ "step": 495
704
+ },
705
+ {
706
+ "epoch": 0.5752912411908528,
707
+ "grad_norm": 0.058092959225177765,
708
+ "learning_rate": 4.491525423728814e-05,
709
+ "loss": 0.5721,
710
+ "step": 500
711
+ },
712
+ {
713
+ "epoch": 0.5810441536027614,
714
+ "grad_norm": 0.05266360193490982,
715
+ "learning_rate": 4.430992736077482e-05,
716
+ "loss": 0.6214,
717
+ "step": 505
718
+ },
719
+ {
720
+ "epoch": 0.58679706601467,
721
+ "grad_norm": 0.05242326855659485,
722
+ "learning_rate": 4.3704600484261504e-05,
723
+ "loss": 0.635,
724
+ "step": 510
725
+ },
726
+ {
727
+ "epoch": 0.5925499784265784,
728
+ "grad_norm": 0.05718274787068367,
729
+ "learning_rate": 4.309927360774819e-05,
730
+ "loss": 0.6138,
731
+ "step": 515
732
+ },
733
+ {
734
+ "epoch": 0.598302890838487,
735
+ "grad_norm": 0.06271938234567642,
736
+ "learning_rate": 4.2493946731234865e-05,
737
+ "loss": 0.5663,
738
+ "step": 520
739
+ },
740
+ {
741
+ "epoch": 0.6040558032503955,
742
+ "grad_norm": 0.09001675248146057,
743
+ "learning_rate": 4.188861985472155e-05,
744
+ "loss": 0.582,
745
+ "step": 525
746
+ },
747
+ {
748
+ "epoch": 0.609808715662304,
749
+ "grad_norm": 0.054499827325344086,
750
+ "learning_rate": 4.1283292978208233e-05,
751
+ "loss": 0.6487,
752
+ "step": 530
753
+ },
754
+ {
755
+ "epoch": 0.6155616280742126,
756
+ "grad_norm": 0.05621207877993584,
757
+ "learning_rate": 4.067796610169492e-05,
758
+ "loss": 0.6196,
759
+ "step": 535
760
+ },
761
+ {
762
+ "epoch": 0.6213145404861211,
763
+ "grad_norm": 0.0570133738219738,
764
+ "learning_rate": 4.00726392251816e-05,
765
+ "loss": 0.5907,
766
+ "step": 540
767
+ },
768
+ {
769
+ "epoch": 0.6270674528980297,
770
+ "grad_norm": 0.06457239389419556,
771
+ "learning_rate": 3.9467312348668285e-05,
772
+ "loss": 0.5801,
773
+ "step": 545
774
+ },
775
+ {
776
+ "epoch": 0.6328203653099381,
777
+ "grad_norm": 0.052366577088832855,
778
+ "learning_rate": 3.886198547215496e-05,
779
+ "loss": 0.5874,
780
+ "step": 550
781
+ },
782
+ {
783
+ "epoch": 0.6385732777218467,
784
+ "grad_norm": 0.04825791344046593,
785
+ "learning_rate": 3.825665859564165e-05,
786
+ "loss": 0.609,
787
+ "step": 555
788
+ },
789
+ {
790
+ "epoch": 0.6443261901337552,
791
+ "grad_norm": 0.05176164582371712,
792
+ "learning_rate": 3.765133171912833e-05,
793
+ "loss": 0.6364,
794
+ "step": 560
795
+ },
796
+ {
797
+ "epoch": 0.6500791025456637,
798
+ "grad_norm": 0.05070199817419052,
799
+ "learning_rate": 3.7046004842615015e-05,
800
+ "loss": 0.6221,
801
+ "step": 565
802
+ },
803
+ {
804
+ "epoch": 0.6558320149575723,
805
+ "grad_norm": 0.04965753108263016,
806
+ "learning_rate": 3.644067796610169e-05,
807
+ "loss": 0.606,
808
+ "step": 570
809
+ },
810
+ {
811
+ "epoch": 0.6615849273694808,
812
+ "grad_norm": 0.045617908239364624,
813
+ "learning_rate": 3.583535108958838e-05,
814
+ "loss": 0.5832,
815
+ "step": 575
816
+ },
817
+ {
818
+ "epoch": 0.6673378397813893,
819
+ "grad_norm": 0.05255943536758423,
820
+ "learning_rate": 3.523002421307507e-05,
821
+ "loss": 0.6229,
822
+ "step": 580
823
+ },
824
+ {
825
+ "epoch": 0.6730907521932978,
826
+ "grad_norm": 0.054092586040496826,
827
+ "learning_rate": 3.4624697336561744e-05,
828
+ "loss": 0.5717,
829
+ "step": 585
830
+ },
831
+ {
832
+ "epoch": 0.6788436646052064,
833
+ "grad_norm": 0.04618106037378311,
834
+ "learning_rate": 3.401937046004843e-05,
835
+ "loss": 0.61,
836
+ "step": 590
837
+ },
838
+ {
839
+ "epoch": 0.684596577017115,
840
+ "grad_norm": 0.05267634615302086,
841
+ "learning_rate": 3.341404358353511e-05,
842
+ "loss": 0.5638,
843
+ "step": 595
844
+ },
845
+ {
846
+ "epoch": 0.6903494894290234,
847
+ "grad_norm": 0.04927615076303482,
848
+ "learning_rate": 3.280871670702179e-05,
849
+ "loss": 0.5656,
850
+ "step": 600
851
+ },
852
+ {
853
+ "epoch": 0.696102401840932,
854
+ "grad_norm": 0.0523078553378582,
855
+ "learning_rate": 3.2203389830508473e-05,
856
+ "loss": 0.6336,
857
+ "step": 605
858
+ },
859
+ {
860
+ "epoch": 0.7018553142528405,
861
+ "grad_norm": 0.05376661941409111,
862
+ "learning_rate": 3.1598062953995164e-05,
863
+ "loss": 0.6357,
864
+ "step": 610
865
+ },
866
+ {
867
+ "epoch": 0.707608226664749,
868
+ "grad_norm": 0.04982991889119148,
869
+ "learning_rate": 3.099273607748184e-05,
870
+ "loss": 0.5946,
871
+ "step": 615
872
+ },
873
+ {
874
+ "epoch": 0.7133611390766575,
875
+ "grad_norm": 0.054978545755147934,
876
+ "learning_rate": 3.0387409200968526e-05,
877
+ "loss": 0.5403,
878
+ "step": 620
879
+ },
880
+ {
881
+ "epoch": 0.7191140514885661,
882
+ "grad_norm": 0.053609415888786316,
883
+ "learning_rate": 2.9782082324455206e-05,
884
+ "loss": 0.5738,
885
+ "step": 625
886
+ },
887
+ {
888
+ "epoch": 0.7248669639004747,
889
+ "grad_norm": 0.05252372473478317,
890
+ "learning_rate": 2.917675544794189e-05,
891
+ "loss": 0.6081,
892
+ "step": 630
893
+ },
894
+ {
895
+ "epoch": 0.7306198763123831,
896
+ "grad_norm": 0.04830444976687431,
897
+ "learning_rate": 2.857142857142857e-05,
898
+ "loss": 0.6065,
899
+ "step": 635
900
+ },
901
+ {
902
+ "epoch": 0.7363727887242917,
903
+ "grad_norm": 0.04439789429306984,
904
+ "learning_rate": 2.7966101694915255e-05,
905
+ "loss": 0.6029,
906
+ "step": 640
907
+ },
908
+ {
909
+ "epoch": 0.7421257011362002,
910
+ "grad_norm": 0.06717346608638763,
911
+ "learning_rate": 2.7360774818401942e-05,
912
+ "loss": 0.5719,
913
+ "step": 645
914
+ },
915
+ {
916
+ "epoch": 0.7478786135481087,
917
+ "grad_norm": 0.04409700632095337,
918
+ "learning_rate": 2.6755447941888623e-05,
919
+ "loss": 0.5988,
920
+ "step": 650
921
+ },
922
+ {
923
+ "epoch": 0.7536315259600173,
924
+ "grad_norm": 0.05428074672818184,
925
+ "learning_rate": 2.6150121065375304e-05,
926
+ "loss": 0.5967,
927
+ "step": 655
928
+ },
929
+ {
930
+ "epoch": 0.7593844383719258,
931
+ "grad_norm": 0.055107153952121735,
932
+ "learning_rate": 2.5544794188861988e-05,
933
+ "loss": 0.6374,
934
+ "step": 660
935
+ },
936
+ {
937
+ "epoch": 0.7651373507838343,
938
+ "grad_norm": 0.049112468957901,
939
+ "learning_rate": 2.4939467312348668e-05,
940
+ "loss": 0.579,
941
+ "step": 665
942
+ },
943
+ {
944
+ "epoch": 0.7708902631957428,
945
+ "grad_norm": 0.057574160397052765,
946
+ "learning_rate": 2.4334140435835352e-05,
947
+ "loss": 0.5845,
948
+ "step": 670
949
+ },
950
+ {
951
+ "epoch": 0.7766431756076514,
952
+ "grad_norm": 0.04990919306874275,
953
+ "learning_rate": 2.3728813559322036e-05,
954
+ "loss": 0.5828,
955
+ "step": 675
956
+ },
957
+ {
958
+ "epoch": 0.78239608801956,
959
+ "grad_norm": 0.057173240929841995,
960
+ "learning_rate": 2.3123486682808717e-05,
961
+ "loss": 0.6081,
962
+ "step": 680
963
+ },
964
+ {
965
+ "epoch": 0.7881490004314684,
966
+ "grad_norm": 0.05236280709505081,
967
+ "learning_rate": 2.25181598062954e-05,
968
+ "loss": 0.6314,
969
+ "step": 685
970
+ },
971
+ {
972
+ "epoch": 0.793901912843377,
973
+ "grad_norm": 0.04565229266881943,
974
+ "learning_rate": 2.1912832929782085e-05,
975
+ "loss": 0.6043,
976
+ "step": 690
977
+ },
978
+ {
979
+ "epoch": 0.7996548252552855,
980
+ "grad_norm": 0.05781184136867523,
981
+ "learning_rate": 2.1307506053268766e-05,
982
+ "loss": 0.6163,
983
+ "step": 695
984
+ },
985
+ {
986
+ "epoch": 0.805407737667194,
987
+ "grad_norm": 0.04942315071821213,
988
+ "learning_rate": 2.070217917675545e-05,
989
+ "loss": 0.5702,
990
+ "step": 700
991
+ },
992
+ {
993
+ "epoch": 0.8111606500791025,
994
+ "grad_norm": 0.057706985622644424,
995
+ "learning_rate": 2.009685230024213e-05,
996
+ "loss": 0.6071,
997
+ "step": 705
998
+ },
999
+ {
1000
+ "epoch": 0.8169135624910111,
1001
+ "grad_norm": 0.049867674708366394,
1002
+ "learning_rate": 1.9491525423728814e-05,
1003
+ "loss": 0.6072,
1004
+ "step": 710
1005
+ },
1006
+ {
1007
+ "epoch": 0.8226664749029196,
1008
+ "grad_norm": 0.051232825964689255,
1009
+ "learning_rate": 1.88861985472155e-05,
1010
+ "loss": 0.6072,
1011
+ "step": 715
1012
+ },
1013
+ {
1014
+ "epoch": 0.8284193873148281,
1015
+ "grad_norm": 0.05312568321824074,
1016
+ "learning_rate": 1.828087167070218e-05,
1017
+ "loss": 0.5977,
1018
+ "step": 720
1019
+ },
1020
+ {
1021
+ "epoch": 0.8341722997267367,
1022
+ "grad_norm": 0.04792577773332596,
1023
+ "learning_rate": 1.7675544794188863e-05,
1024
+ "loss": 0.595,
1025
+ "step": 725
1026
+ },
1027
+ {
1028
+ "epoch": 0.8399252121386452,
1029
+ "grad_norm": 0.04957397282123566,
1030
+ "learning_rate": 1.7070217917675547e-05,
1031
+ "loss": 0.6016,
1032
+ "step": 730
1033
+ },
1034
+ {
1035
+ "epoch": 0.8456781245505537,
1036
+ "grad_norm": 0.05055704340338707,
1037
+ "learning_rate": 1.6464891041162228e-05,
1038
+ "loss": 0.6239,
1039
+ "step": 735
1040
+ },
1041
+ {
1042
+ "epoch": 0.8514310369624623,
1043
+ "grad_norm": 0.046788472682237625,
1044
+ "learning_rate": 1.5859564164648912e-05,
1045
+ "loss": 0.618,
1046
+ "step": 740
1047
+ },
1048
+ {
1049
+ "epoch": 0.8571839493743708,
1050
+ "grad_norm": 0.06297384202480316,
1051
+ "learning_rate": 1.5254237288135596e-05,
1052
+ "loss": 0.5735,
1053
+ "step": 745
1054
+ },
1055
+ {
1056
+ "epoch": 0.8629368617862793,
1057
+ "grad_norm": 0.05054633319377899,
1058
+ "learning_rate": 1.4648910411622276e-05,
1059
+ "loss": 0.5635,
1060
+ "step": 750
1061
+ },
1062
+ {
1063
+ "epoch": 0.8686897741981878,
1064
+ "grad_norm": 0.04888758435845375,
1065
+ "learning_rate": 1.4043583535108959e-05,
1066
+ "loss": 0.6374,
1067
+ "step": 755
1068
+ },
1069
+ {
1070
+ "epoch": 0.8744426866100964,
1071
+ "grad_norm": 0.05039739981293678,
1072
+ "learning_rate": 1.3438256658595641e-05,
1073
+ "loss": 0.6236,
1074
+ "step": 760
1075
+ },
1076
+ {
1077
+ "epoch": 0.8801955990220048,
1078
+ "grad_norm": 0.04892362654209137,
1079
+ "learning_rate": 1.2832929782082325e-05,
1080
+ "loss": 0.5849,
1081
+ "step": 765
1082
+ },
1083
+ {
1084
+ "epoch": 0.8859485114339134,
1085
+ "grad_norm": 0.05396633222699165,
1086
+ "learning_rate": 1.2227602905569007e-05,
1087
+ "loss": 0.5686,
1088
+ "step": 770
1089
+ },
1090
+ {
1091
+ "epoch": 0.891701423845822,
1092
+ "grad_norm": 0.0467771515250206,
1093
+ "learning_rate": 1.1622276029055691e-05,
1094
+ "loss": 0.5822,
1095
+ "step": 775
1096
+ },
1097
+ {
1098
+ "epoch": 0.8974543362577305,
1099
+ "grad_norm": 0.05482470244169235,
1100
+ "learning_rate": 1.1016949152542374e-05,
1101
+ "loss": 0.6041,
1102
+ "step": 780
1103
+ },
1104
+ {
1105
+ "epoch": 0.903207248669639,
1106
+ "grad_norm": 0.05063646286725998,
1107
+ "learning_rate": 1.0411622276029056e-05,
1108
+ "loss": 0.6134,
1109
+ "step": 785
1110
+ },
1111
+ {
1112
+ "epoch": 0.9089601610815475,
1113
+ "grad_norm": 0.05881096422672272,
1114
+ "learning_rate": 9.806295399515738e-06,
1115
+ "loss": 0.6099,
1116
+ "step": 790
1117
+ },
1118
+ {
1119
+ "epoch": 0.9147130734934561,
1120
+ "grad_norm": 0.054972730576992035,
1121
+ "learning_rate": 9.200968523002422e-06,
1122
+ "loss": 0.6019,
1123
+ "step": 795
1124
+ },
1125
+ {
1126
+ "epoch": 0.9204659859053645,
1127
+ "grad_norm": 0.05072762817144394,
1128
+ "learning_rate": 8.595641646489105e-06,
1129
+ "loss": 0.574,
1130
+ "step": 800
1131
+ },
1132
+ {
1133
+ "epoch": 0.9262188983172731,
1134
+ "grad_norm": 0.06184648349881172,
1135
+ "learning_rate": 7.990314769975787e-06,
1136
+ "loss": 0.6135,
1137
+ "step": 805
1138
+ },
1139
+ {
1140
+ "epoch": 0.9319718107291817,
1141
+ "grad_norm": 0.04959486797451973,
1142
+ "learning_rate": 7.3849878934624694e-06,
1143
+ "loss": 0.6079,
1144
+ "step": 810
1145
+ },
1146
+ {
1147
+ "epoch": 0.9377247231410901,
1148
+ "grad_norm": 0.05266949534416199,
1149
+ "learning_rate": 6.779661016949153e-06,
1150
+ "loss": 0.5847,
1151
+ "step": 815
1152
+ },
1153
+ {
1154
+ "epoch": 0.9434776355529987,
1155
+ "grad_norm": 0.05372299626469612,
1156
+ "learning_rate": 6.174334140435836e-06,
1157
+ "loss": 0.5629,
1158
+ "step": 820
1159
+ },
1160
+ {
1161
+ "epoch": 0.9492305479649072,
1162
+ "grad_norm": 0.056709058582782745,
1163
+ "learning_rate": 5.569007263922518e-06,
1164
+ "loss": 0.5855,
1165
+ "step": 825
1166
+ },
1167
+ {
1168
+ "epoch": 0.9549834603768158,
1169
+ "grad_norm": 0.0584251843392849,
1170
+ "learning_rate": 4.963680387409201e-06,
1171
+ "loss": 0.5966,
1172
+ "step": 830
1173
+ },
1174
+ {
1175
+ "epoch": 0.9607363727887243,
1176
+ "grad_norm": 0.04992934688925743,
1177
+ "learning_rate": 4.358353510895884e-06,
1178
+ "loss": 0.6372,
1179
+ "step": 835
1180
+ },
1181
+ {
1182
+ "epoch": 0.9664892852006328,
1183
+ "grad_norm": 0.05013206973671913,
1184
+ "learning_rate": 3.7530266343825673e-06,
1185
+ "loss": 0.5907,
1186
+ "step": 840
1187
+ },
1188
+ {
1189
+ "epoch": 0.9722421976125414,
1190
+ "grad_norm": 0.05348077416419983,
1191
+ "learning_rate": 3.1476997578692496e-06,
1192
+ "loss": 0.5629,
1193
+ "step": 845
1194
+ },
1195
+ {
1196
+ "epoch": 0.9779951100244498,
1197
+ "grad_norm": 0.04957769066095352,
1198
+ "learning_rate": 2.5423728813559323e-06,
1199
+ "loss": 0.602,
1200
+ "step": 850
1201
+ },
1202
+ {
1203
+ "epoch": 0.9837480224363584,
1204
+ "grad_norm": 0.05082723870873451,
1205
+ "learning_rate": 1.937046004842615e-06,
1206
+ "loss": 0.5999,
1207
+ "step": 855
1208
+ },
1209
+ {
1210
+ "epoch": 0.989500934848267,
1211
+ "grad_norm": 0.04676876962184906,
1212
+ "learning_rate": 1.3317191283292979e-06,
1213
+ "loss": 0.6004,
1214
+ "step": 860
1215
+ },
1216
+ {
1217
+ "epoch": 0.9952538472601755,
1218
+ "grad_norm": 0.05383327603340149,
1219
+ "learning_rate": 7.263922518159807e-07,
1220
+ "loss": 0.6123,
1221
+ "step": 865
1222
+ },
1223
+ {
1224
+ "epoch": 1.0,
1225
+ "grad_norm": 0.1522870808839798,
1226
+ "learning_rate": 1.2106537530266344e-07,
1227
+ "loss": 0.5303,
1228
+ "step": 870
1229
+ }
1230
+ ],
1231
+ "logging_steps": 5,
1232
+ "max_steps": 870,
1233
+ "num_input_tokens_seen": 0,
1234
+ "num_train_epochs": 1,
1235
+ "save_steps": 500,
1236
+ "stateful_callbacks": {
1237
+ "TrainerControl": {
1238
+ "args": {
1239
+ "should_epoch_stop": false,
1240
+ "should_evaluate": false,
1241
+ "should_log": false,
1242
+ "should_save": true,
1243
+ "should_training_stop": true
1244
+ },
1245
+ "attributes": {}
1246
+ }
1247
+ },
1248
+ "total_flos": 1.1405546259399626e+20,
1249
+ "train_batch_size": 1,
1250
+ "trial_name": null,
1251
+ "trial_params": null
1252
+ }
Qwen3-30B-A3B-Instruct-2507-sft-fusang-swa=2k_sink=100_falayer=25_from0_step2_fadec/checkpoint-870/training_args.bin ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:0342d0f1b95d153b2c7ba51d241f4c4c7635df0441899d742902bf967c84b92d
3
+ size 6033
Qwen3-30B-A3B-Instruct-2507-sft-fusang-swa=2k_sink=100_falayer=25_from0_step2_fadec/checkpoint-870/vocab.json ADDED
The diff for this file is too large to render. See raw diff
 
Qwen3-30B-A3B-Thinking-2507-sft-fusang-swa=2k_sink=0/checkpoint-862/tokenizer.json ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:aeb13307a71acd8fe81861d94ad54ab689df773318809eed3cbe794b4492dae4
3
+ size 11422654
Qwen3-30B-A3B-Thinking-2507-sft-fusang-swa=2k_sink=100/checkpoint-862/training_args.bin ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:12c0dc281c95d3429ae3787f8caf681f67719171d325a6e2cae0bdf9309320f7
3
+ size 5969
Qwen3-30B-A3B-Thinking-2507-sft-fusang-swa=2k_sink=100_falayer=25_from0_step2_fadec/checkpoint-862/merges.txt ADDED
The diff for this file is too large to render. See raw diff
 
Qwen3-30B-A3B-Thinking-2507-sft-fusang-swa=2k_sink=100_falayer=25_from0_step2_fadec/checkpoint-862/vocab.json ADDED
The diff for this file is too large to render. See raw diff
 
Qwen3-30B-A3B-Thinking-2507-sft-fusang/checkpoint-862/adapter_config.json ADDED
@@ -0,0 +1,38 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "alpha_pattern": {},
3
+ "auto_mapping": null,
4
+ "base_model_name_or_path": "/share/models/Qwen3-30B-A3B-Thinking-2507",
5
+ "bias": "none",
6
+ "corda_config": null,
7
+ "eva_config": null,
8
+ "exclude_modules": null,
9
+ "fan_in_fan_out": false,
10
+ "inference_mode": true,
11
+ "init_lora_weights": true,
12
+ "layer_replication": null,
13
+ "layers_pattern": null,
14
+ "layers_to_transform": null,
15
+ "loftq_config": {},
16
+ "lora_alpha": 32,
17
+ "lora_bias": false,
18
+ "lora_dropout": 0.1,
19
+ "megatron_config": null,
20
+ "megatron_core": "megatron.core",
21
+ "modules_to_save": null,
22
+ "peft_type": "LORA",
23
+ "qalora_group_size": 16,
24
+ "r": 16,
25
+ "rank_pattern": {},
26
+ "revision": null,
27
+ "target_modules": [
28
+ "q_proj",
29
+ "k_proj",
30
+ "v_proj"
31
+ ],
32
+ "target_parameters": null,
33
+ "task_type": "CAUSAL_LM",
34
+ "trainable_token_indices": null,
35
+ "use_dora": false,
36
+ "use_qalora": false,
37
+ "use_rslora": true
38
+ }
Qwen3-30B-A3B-Thinking-2507-sft-fusang/checkpoint-862/special_tokens_map.json ADDED
@@ -0,0 +1,31 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "additional_special_tokens": [
3
+ "<|im_start|>",
4
+ "<|im_end|>",
5
+ "<|object_ref_start|>",
6
+ "<|object_ref_end|>",
7
+ "<|box_start|>",
8
+ "<|box_end|>",
9
+ "<|quad_start|>",
10
+ "<|quad_end|>",
11
+ "<|vision_start|>",
12
+ "<|vision_end|>",
13
+ "<|vision_pad|>",
14
+ "<|image_pad|>",
15
+ "<|video_pad|>"
16
+ ],
17
+ "eos_token": {
18
+ "content": "<|im_end|>",
19
+ "lstrip": false,
20
+ "normalized": false,
21
+ "rstrip": false,
22
+ "single_word": false
23
+ },
24
+ "pad_token": {
25
+ "content": "<|endoftext|>",
26
+ "lstrip": false,
27
+ "normalized": false,
28
+ "rstrip": false,
29
+ "single_word": false
30
+ }
31
+ }
Qwen3-4B-Instruct-2507-sft-fusang-swa=2k_sink=0/checkpoint-1302/tokenizer.json ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:aeb13307a71acd8fe81861d94ad54ab689df773318809eed3cbe794b4492dae4
3
+ size 11422654
Qwen3-4B-Instruct-2507-sft-fusang-swa=2k_sink=0/checkpoint-1736/tokenizer.json ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:aeb13307a71acd8fe81861d94ad54ab689df773318809eed3cbe794b4492dae4
3
+ size 11422654
Qwen3-4B-Instruct-2507-sft-fusang-swa=2k_sink=0_fadec/checkpoint-1760/tokenizer.json ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:aeb13307a71acd8fe81861d94ad54ab689df773318809eed3cbe794b4492dae4
3
+ size 11422654
Qwen3-4B-Instruct-2507-sft-fusang-swa=2k_sink=0_fadec/checkpoint-1760/trainer_state.json ADDED
@@ -0,0 +1,2498 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "best_global_step": null,
3
+ "best_metric": null,
4
+ "best_model_checkpoint": null,
5
+ "epoch": 1.0,
6
+ "eval_steps": 500,
7
+ "global_step": 1760,
8
+ "is_hyper_param_search": false,
9
+ "is_local_process_zero": true,
10
+ "is_world_process_zero": true,
11
+ "log_history": [
12
+ {
13
+ "epoch": 0.002841716396703609,
14
+ "grad_norm": 1.9044979810714722,
15
+ "learning_rate": 4.5454545454545455e-06,
16
+ "loss": 0.978,
17
+ "step": 5
18
+ },
19
+ {
20
+ "epoch": 0.005683432793407218,
21
+ "grad_norm": 4.298397064208984,
22
+ "learning_rate": 1.0227272727272729e-05,
23
+ "loss": 0.9202,
24
+ "step": 10
25
+ },
26
+ {
27
+ "epoch": 0.008525149190110827,
28
+ "grad_norm": 2.7809829711914062,
29
+ "learning_rate": 1.590909090909091e-05,
30
+ "loss": 0.8541,
31
+ "step": 15
32
+ },
33
+ {
34
+ "epoch": 0.011366865586814436,
35
+ "grad_norm": 1.2808133363723755,
36
+ "learning_rate": 2.1590909090909093e-05,
37
+ "loss": 0.7733,
38
+ "step": 20
39
+ },
40
+ {
41
+ "epoch": 0.014208581983518044,
42
+ "grad_norm": 0.8438812494277954,
43
+ "learning_rate": 2.7272727272727273e-05,
44
+ "loss": 0.7329,
45
+ "step": 25
46
+ },
47
+ {
48
+ "epoch": 0.017050298380221655,
49
+ "grad_norm": 0.8602949380874634,
50
+ "learning_rate": 3.295454545454545e-05,
51
+ "loss": 0.8108,
52
+ "step": 30
53
+ },
54
+ {
55
+ "epoch": 0.019892014776925263,
56
+ "grad_norm": 0.8656124472618103,
57
+ "learning_rate": 3.8636363636363636e-05,
58
+ "loss": 0.7774,
59
+ "step": 35
60
+ },
61
+ {
62
+ "epoch": 0.022733731173628872,
63
+ "grad_norm": 0.7628982663154602,
64
+ "learning_rate": 4.431818181818182e-05,
65
+ "loss": 0.7317,
66
+ "step": 40
67
+ },
68
+ {
69
+ "epoch": 0.02557544757033248,
70
+ "grad_norm": 0.9274219870567322,
71
+ "learning_rate": 5e-05,
72
+ "loss": 0.7412,
73
+ "step": 45
74
+ },
75
+ {
76
+ "epoch": 0.02841716396703609,
77
+ "grad_norm": 1.101454734802246,
78
+ "learning_rate": 5.568181818181818e-05,
79
+ "loss": 0.6865,
80
+ "step": 50
81
+ },
82
+ {
83
+ "epoch": 0.0312588803637397,
84
+ "grad_norm": 0.5100212097167969,
85
+ "learning_rate": 6.136363636363636e-05,
86
+ "loss": 0.7179,
87
+ "step": 55
88
+ },
89
+ {
90
+ "epoch": 0.03410059676044331,
91
+ "grad_norm": 0.6264331340789795,
92
+ "learning_rate": 6.704545454545455e-05,
93
+ "loss": 0.7029,
94
+ "step": 60
95
+ },
96
+ {
97
+ "epoch": 0.03694231315714692,
98
+ "grad_norm": 0.5763853788375854,
99
+ "learning_rate": 7.272727272727273e-05,
100
+ "loss": 0.6812,
101
+ "step": 65
102
+ },
103
+ {
104
+ "epoch": 0.039784029553850526,
105
+ "grad_norm": 0.8327955603599548,
106
+ "learning_rate": 7.840909090909091e-05,
107
+ "loss": 0.6672,
108
+ "step": 70
109
+ },
110
+ {
111
+ "epoch": 0.042625745950554135,
112
+ "grad_norm": 0.533845841884613,
113
+ "learning_rate": 8.40909090909091e-05,
114
+ "loss": 0.6517,
115
+ "step": 75
116
+ },
117
+ {
118
+ "epoch": 0.045467462347257744,
119
+ "grad_norm": 0.46759405732154846,
120
+ "learning_rate": 8.977272727272728e-05,
121
+ "loss": 0.6585,
122
+ "step": 80
123
+ },
124
+ {
125
+ "epoch": 0.04830917874396135,
126
+ "grad_norm": 0.5662851929664612,
127
+ "learning_rate": 9.545454545454546e-05,
128
+ "loss": 0.6698,
129
+ "step": 85
130
+ },
131
+ {
132
+ "epoch": 0.05115089514066496,
133
+ "grad_norm": 0.5162264704704285,
134
+ "learning_rate": 9.994019138755981e-05,
135
+ "loss": 0.6927,
136
+ "step": 90
137
+ },
138
+ {
139
+ "epoch": 0.05399261153736857,
140
+ "grad_norm": 0.5184995532035828,
141
+ "learning_rate": 9.964114832535886e-05,
142
+ "loss": 0.6757,
143
+ "step": 95
144
+ },
145
+ {
146
+ "epoch": 0.05683432793407218,
147
+ "grad_norm": 0.5014625191688538,
148
+ "learning_rate": 9.93421052631579e-05,
149
+ "loss": 0.6551,
150
+ "step": 100
151
+ },
152
+ {
153
+ "epoch": 0.059676044330775786,
154
+ "grad_norm": 0.4743216037750244,
155
+ "learning_rate": 9.904306220095694e-05,
156
+ "loss": 0.6704,
157
+ "step": 105
158
+ },
159
+ {
160
+ "epoch": 0.0625177607274794,
161
+ "grad_norm": 0.4949679672718048,
162
+ "learning_rate": 9.874401913875598e-05,
163
+ "loss": 0.6562,
164
+ "step": 110
165
+ },
166
+ {
167
+ "epoch": 0.06535947712418301,
168
+ "grad_norm": 0.3639834523200989,
169
+ "learning_rate": 9.844497607655503e-05,
170
+ "loss": 0.6461,
171
+ "step": 115
172
+ },
173
+ {
174
+ "epoch": 0.06820119352088662,
175
+ "grad_norm": 0.6026004552841187,
176
+ "learning_rate": 9.814593301435408e-05,
177
+ "loss": 0.6798,
178
+ "step": 120
179
+ },
180
+ {
181
+ "epoch": 0.07104290991759023,
182
+ "grad_norm": 0.4765580892562866,
183
+ "learning_rate": 9.784688995215312e-05,
184
+ "loss": 0.6307,
185
+ "step": 125
186
+ },
187
+ {
188
+ "epoch": 0.07388462631429384,
189
+ "grad_norm": 0.688258171081543,
190
+ "learning_rate": 9.754784688995216e-05,
191
+ "loss": 0.6737,
192
+ "step": 130
193
+ },
194
+ {
195
+ "epoch": 0.07672634271099744,
196
+ "grad_norm": 0.5317380428314209,
197
+ "learning_rate": 9.72488038277512e-05,
198
+ "loss": 0.6539,
199
+ "step": 135
200
+ },
201
+ {
202
+ "epoch": 0.07956805910770105,
203
+ "grad_norm": 0.5078372955322266,
204
+ "learning_rate": 9.694976076555024e-05,
205
+ "loss": 0.6476,
206
+ "step": 140
207
+ },
208
+ {
209
+ "epoch": 0.08240977550440466,
210
+ "grad_norm": 0.5277720093727112,
211
+ "learning_rate": 9.66507177033493e-05,
212
+ "loss": 0.6411,
213
+ "step": 145
214
+ },
215
+ {
216
+ "epoch": 0.08525149190110827,
217
+ "grad_norm": 0.4187789261341095,
218
+ "learning_rate": 9.635167464114832e-05,
219
+ "loss": 0.6299,
220
+ "step": 150
221
+ },
222
+ {
223
+ "epoch": 0.08809320829781188,
224
+ "grad_norm": 0.5423827171325684,
225
+ "learning_rate": 9.605263157894737e-05,
226
+ "loss": 0.6323,
227
+ "step": 155
228
+ },
229
+ {
230
+ "epoch": 0.09093492469451549,
231
+ "grad_norm": 0.4522331655025482,
232
+ "learning_rate": 9.575358851674642e-05,
233
+ "loss": 0.6294,
234
+ "step": 160
235
+ },
236
+ {
237
+ "epoch": 0.0937766410912191,
238
+ "grad_norm": 0.445803701877594,
239
+ "learning_rate": 9.545454545454546e-05,
240
+ "loss": 0.6374,
241
+ "step": 165
242
+ },
243
+ {
244
+ "epoch": 0.0966183574879227,
245
+ "grad_norm": 0.5082485675811768,
246
+ "learning_rate": 9.51555023923445e-05,
247
+ "loss": 0.642,
248
+ "step": 170
249
+ },
250
+ {
251
+ "epoch": 0.09946007388462631,
252
+ "grad_norm": 0.44934386014938354,
253
+ "learning_rate": 9.485645933014354e-05,
254
+ "loss": 0.6306,
255
+ "step": 175
256
+ },
257
+ {
258
+ "epoch": 0.10230179028132992,
259
+ "grad_norm": 0.34924763441085815,
260
+ "learning_rate": 9.455741626794258e-05,
261
+ "loss": 0.6627,
262
+ "step": 180
263
+ },
264
+ {
265
+ "epoch": 0.10514350667803353,
266
+ "grad_norm": 0.37746477127075195,
267
+ "learning_rate": 9.425837320574164e-05,
268
+ "loss": 0.662,
269
+ "step": 185
270
+ },
271
+ {
272
+ "epoch": 0.10798522307473714,
273
+ "grad_norm": 0.46685609221458435,
274
+ "learning_rate": 9.395933014354067e-05,
275
+ "loss": 0.6331,
276
+ "step": 190
277
+ },
278
+ {
279
+ "epoch": 0.11082693947144075,
280
+ "grad_norm": 0.49400794506073,
281
+ "learning_rate": 9.366028708133972e-05,
282
+ "loss": 0.6312,
283
+ "step": 195
284
+ },
285
+ {
286
+ "epoch": 0.11366865586814436,
287
+ "grad_norm": 0.34066227078437805,
288
+ "learning_rate": 9.336124401913876e-05,
289
+ "loss": 0.6466,
290
+ "step": 200
291
+ },
292
+ {
293
+ "epoch": 0.11651037226484796,
294
+ "grad_norm": 0.4472307860851288,
295
+ "learning_rate": 9.30622009569378e-05,
296
+ "loss": 0.6325,
297
+ "step": 205
298
+ },
299
+ {
300
+ "epoch": 0.11935208866155157,
301
+ "grad_norm": 0.43694406747817993,
302
+ "learning_rate": 9.276315789473686e-05,
303
+ "loss": 0.6448,
304
+ "step": 210
305
+ },
306
+ {
307
+ "epoch": 0.12219380505825518,
308
+ "grad_norm": 0.41603100299835205,
309
+ "learning_rate": 9.246411483253588e-05,
310
+ "loss": 0.6372,
311
+ "step": 215
312
+ },
313
+ {
314
+ "epoch": 0.1250355214549588,
315
+ "grad_norm": 0.5098612904548645,
316
+ "learning_rate": 9.216507177033494e-05,
317
+ "loss": 0.5933,
318
+ "step": 220
319
+ },
320
+ {
321
+ "epoch": 0.1278772378516624,
322
+ "grad_norm": 0.42455029487609863,
323
+ "learning_rate": 9.186602870813398e-05,
324
+ "loss": 0.6288,
325
+ "step": 225
326
+ },
327
+ {
328
+ "epoch": 0.13071895424836602,
329
+ "grad_norm": 0.3707619905471802,
330
+ "learning_rate": 9.156698564593302e-05,
331
+ "loss": 0.6052,
332
+ "step": 230
333
+ },
334
+ {
335
+ "epoch": 0.13356067064506963,
336
+ "grad_norm": 0.39465266466140747,
337
+ "learning_rate": 9.126794258373206e-05,
338
+ "loss": 0.6514,
339
+ "step": 235
340
+ },
341
+ {
342
+ "epoch": 0.13640238704177324,
343
+ "grad_norm": 0.41626229882240295,
344
+ "learning_rate": 9.09688995215311e-05,
345
+ "loss": 0.6493,
346
+ "step": 240
347
+ },
348
+ {
349
+ "epoch": 0.13924410343847685,
350
+ "grad_norm": 0.5392955541610718,
351
+ "learning_rate": 9.066985645933016e-05,
352
+ "loss": 0.5882,
353
+ "step": 245
354
+ },
355
+ {
356
+ "epoch": 0.14208581983518045,
357
+ "grad_norm": 0.3979646563529968,
358
+ "learning_rate": 9.03708133971292e-05,
359
+ "loss": 0.6163,
360
+ "step": 250
361
+ },
362
+ {
363
+ "epoch": 0.14492753623188406,
364
+ "grad_norm": 0.40286895632743835,
365
+ "learning_rate": 9.007177033492822e-05,
366
+ "loss": 0.6523,
367
+ "step": 255
368
+ },
369
+ {
370
+ "epoch": 0.14776925262858767,
371
+ "grad_norm": 0.41244709491729736,
372
+ "learning_rate": 8.977272727272728e-05,
373
+ "loss": 0.6307,
374
+ "step": 260
375
+ },
376
+ {
377
+ "epoch": 0.15061096902529128,
378
+ "grad_norm": 0.4113865792751312,
379
+ "learning_rate": 8.947368421052632e-05,
380
+ "loss": 0.6584,
381
+ "step": 265
382
+ },
383
+ {
384
+ "epoch": 0.1534526854219949,
385
+ "grad_norm": 0.42327752709388733,
386
+ "learning_rate": 8.917464114832537e-05,
387
+ "loss": 0.6319,
388
+ "step": 270
389
+ },
390
+ {
391
+ "epoch": 0.1562944018186985,
392
+ "grad_norm": 0.41952893137931824,
393
+ "learning_rate": 8.88755980861244e-05,
394
+ "loss": 0.6391,
395
+ "step": 275
396
+ },
397
+ {
398
+ "epoch": 0.1591361182154021,
399
+ "grad_norm": 0.34094205498695374,
400
+ "learning_rate": 8.857655502392344e-05,
401
+ "loss": 0.6177,
402
+ "step": 280
403
+ },
404
+ {
405
+ "epoch": 0.16197783461210571,
406
+ "grad_norm": 0.39203983545303345,
407
+ "learning_rate": 8.82775119617225e-05,
408
+ "loss": 0.6293,
409
+ "step": 285
410
+ },
411
+ {
412
+ "epoch": 0.16481955100880932,
413
+ "grad_norm": 0.44511377811431885,
414
+ "learning_rate": 8.797846889952154e-05,
415
+ "loss": 0.5795,
416
+ "step": 290
417
+ },
418
+ {
419
+ "epoch": 0.16766126740551293,
420
+ "grad_norm": 0.45011797547340393,
421
+ "learning_rate": 8.767942583732058e-05,
422
+ "loss": 0.6114,
423
+ "step": 295
424
+ },
425
+ {
426
+ "epoch": 0.17050298380221654,
427
+ "grad_norm": 0.4113207757472992,
428
+ "learning_rate": 8.738038277511962e-05,
429
+ "loss": 0.6178,
430
+ "step": 300
431
+ },
432
+ {
433
+ "epoch": 0.17334470019892015,
434
+ "grad_norm": 0.31626060605049133,
435
+ "learning_rate": 8.708133971291866e-05,
436
+ "loss": 0.6449,
437
+ "step": 305
438
+ },
439
+ {
440
+ "epoch": 0.17618641659562376,
441
+ "grad_norm": 0.3195206820964813,
442
+ "learning_rate": 8.678229665071771e-05,
443
+ "loss": 0.6234,
444
+ "step": 310
445
+ },
446
+ {
447
+ "epoch": 0.17902813299232737,
448
+ "grad_norm": 0.3395790755748749,
449
+ "learning_rate": 8.648325358851675e-05,
450
+ "loss": 0.651,
451
+ "step": 315
452
+ },
453
+ {
454
+ "epoch": 0.18186984938903097,
455
+ "grad_norm": 0.4819774031639099,
456
+ "learning_rate": 8.61842105263158e-05,
457
+ "loss": 0.6305,
458
+ "step": 320
459
+ },
460
+ {
461
+ "epoch": 0.18471156578573458,
462
+ "grad_norm": 0.32194578647613525,
463
+ "learning_rate": 8.588516746411484e-05,
464
+ "loss": 0.6121,
465
+ "step": 325
466
+ },
467
+ {
468
+ "epoch": 0.1875532821824382,
469
+ "grad_norm": 0.36362892389297485,
470
+ "learning_rate": 8.558612440191388e-05,
471
+ "loss": 0.6477,
472
+ "step": 330
473
+ },
474
+ {
475
+ "epoch": 0.1903949985791418,
476
+ "grad_norm": 0.37848010659217834,
477
+ "learning_rate": 8.528708133971293e-05,
478
+ "loss": 0.6586,
479
+ "step": 335
480
+ },
481
+ {
482
+ "epoch": 0.1932367149758454,
483
+ "grad_norm": 0.4458913803100586,
484
+ "learning_rate": 8.498803827751196e-05,
485
+ "loss": 0.6425,
486
+ "step": 340
487
+ },
488
+ {
489
+ "epoch": 0.19607843137254902,
490
+ "grad_norm": 0.43683770298957825,
491
+ "learning_rate": 8.468899521531101e-05,
492
+ "loss": 0.6382,
493
+ "step": 345
494
+ },
495
+ {
496
+ "epoch": 0.19892014776925263,
497
+ "grad_norm": 0.3169898986816406,
498
+ "learning_rate": 8.438995215311005e-05,
499
+ "loss": 0.5783,
500
+ "step": 350
501
+ },
502
+ {
503
+ "epoch": 0.20176186416595623,
504
+ "grad_norm": 0.39609768986701965,
505
+ "learning_rate": 8.40909090909091e-05,
506
+ "loss": 0.6062,
507
+ "step": 355
508
+ },
509
+ {
510
+ "epoch": 0.20460358056265984,
511
+ "grad_norm": 0.3926376402378082,
512
+ "learning_rate": 8.379186602870814e-05,
513
+ "loss": 0.6296,
514
+ "step": 360
515
+ },
516
+ {
517
+ "epoch": 0.20744529695936345,
518
+ "grad_norm": 0.3605319857597351,
519
+ "learning_rate": 8.349282296650718e-05,
520
+ "loss": 0.6336,
521
+ "step": 365
522
+ },
523
+ {
524
+ "epoch": 0.21028701335606706,
525
+ "grad_norm": 0.4633958637714386,
526
+ "learning_rate": 8.319377990430622e-05,
527
+ "loss": 0.6268,
528
+ "step": 370
529
+ },
530
+ {
531
+ "epoch": 0.21312872975277067,
532
+ "grad_norm": 0.38720232248306274,
533
+ "learning_rate": 8.289473684210527e-05,
534
+ "loss": 0.6217,
535
+ "step": 375
536
+ },
537
+ {
538
+ "epoch": 0.21597044614947428,
539
+ "grad_norm": 0.36167505383491516,
540
+ "learning_rate": 8.259569377990431e-05,
541
+ "loss": 0.6509,
542
+ "step": 380
543
+ },
544
+ {
545
+ "epoch": 0.21881216254617789,
546
+ "grad_norm": 0.511846661567688,
547
+ "learning_rate": 8.229665071770335e-05,
548
+ "loss": 0.6342,
549
+ "step": 385
550
+ },
551
+ {
552
+ "epoch": 0.2216538789428815,
553
+ "grad_norm": 0.398040771484375,
554
+ "learning_rate": 8.19976076555024e-05,
555
+ "loss": 0.6043,
556
+ "step": 390
557
+ },
558
+ {
559
+ "epoch": 0.2244955953395851,
560
+ "grad_norm": 0.4392949044704437,
561
+ "learning_rate": 8.169856459330143e-05,
562
+ "loss": 0.5964,
563
+ "step": 395
564
+ },
565
+ {
566
+ "epoch": 0.2273373117362887,
567
+ "grad_norm": 0.5045430660247803,
568
+ "learning_rate": 8.139952153110049e-05,
569
+ "loss": 0.6238,
570
+ "step": 400
571
+ },
572
+ {
573
+ "epoch": 0.23017902813299232,
574
+ "grad_norm": 0.3407171070575714,
575
+ "learning_rate": 8.110047846889952e-05,
576
+ "loss": 0.6438,
577
+ "step": 405
578
+ },
579
+ {
580
+ "epoch": 0.23302074452969593,
581
+ "grad_norm": 0.3798493444919586,
582
+ "learning_rate": 8.080143540669857e-05,
583
+ "loss": 0.6466,
584
+ "step": 410
585
+ },
586
+ {
587
+ "epoch": 0.23586246092639954,
588
+ "grad_norm": 1.7736181020736694,
589
+ "learning_rate": 8.050239234449761e-05,
590
+ "loss": 0.6508,
591
+ "step": 415
592
+ },
593
+ {
594
+ "epoch": 0.23870417732310314,
595
+ "grad_norm": 0.433393269777298,
596
+ "learning_rate": 8.020334928229665e-05,
597
+ "loss": 0.6127,
598
+ "step": 420
599
+ },
600
+ {
601
+ "epoch": 0.24154589371980675,
602
+ "grad_norm": 0.3232906460762024,
603
+ "learning_rate": 7.99043062200957e-05,
604
+ "loss": 0.618,
605
+ "step": 425
606
+ },
607
+ {
608
+ "epoch": 0.24438761011651036,
609
+ "grad_norm": 0.504993736743927,
610
+ "learning_rate": 7.960526315789473e-05,
611
+ "loss": 0.6516,
612
+ "step": 430
613
+ },
614
+ {
615
+ "epoch": 0.24722932651321397,
616
+ "grad_norm": 0.36123546957969666,
617
+ "learning_rate": 7.930622009569379e-05,
618
+ "loss": 0.6281,
619
+ "step": 435
620
+ },
621
+ {
622
+ "epoch": 0.2500710429099176,
623
+ "grad_norm": 0.39559683203697205,
624
+ "learning_rate": 7.900717703349283e-05,
625
+ "loss": 0.6399,
626
+ "step": 440
627
+ },
628
+ {
629
+ "epoch": 0.2529127593066212,
630
+ "grad_norm": 0.3769395649433136,
631
+ "learning_rate": 7.870813397129187e-05,
632
+ "loss": 0.6103,
633
+ "step": 445
634
+ },
635
+ {
636
+ "epoch": 0.2557544757033248,
637
+ "grad_norm": 0.3781079649925232,
638
+ "learning_rate": 7.840909090909091e-05,
639
+ "loss": 0.6123,
640
+ "step": 450
641
+ },
642
+ {
643
+ "epoch": 0.25859619210002843,
644
+ "grad_norm": 0.40734273195266724,
645
+ "learning_rate": 7.811004784688995e-05,
646
+ "loss": 0.6361,
647
+ "step": 455
648
+ },
649
+ {
650
+ "epoch": 0.26143790849673204,
651
+ "grad_norm": 0.7569648623466492,
652
+ "learning_rate": 7.7811004784689e-05,
653
+ "loss": 0.6393,
654
+ "step": 460
655
+ },
656
+ {
657
+ "epoch": 0.26427962489343565,
658
+ "grad_norm": 0.4384016990661621,
659
+ "learning_rate": 7.751196172248805e-05,
660
+ "loss": 0.6303,
661
+ "step": 465
662
+ },
663
+ {
664
+ "epoch": 0.26712134129013926,
665
+ "grad_norm": 0.4860619306564331,
666
+ "learning_rate": 7.721291866028707e-05,
667
+ "loss": 0.6457,
668
+ "step": 470
669
+ },
670
+ {
671
+ "epoch": 0.26996305768684287,
672
+ "grad_norm": 0.3775562644004822,
673
+ "learning_rate": 7.691387559808613e-05,
674
+ "loss": 0.5935,
675
+ "step": 475
676
+ },
677
+ {
678
+ "epoch": 0.2728047740835465,
679
+ "grad_norm": 0.3496741056442261,
680
+ "learning_rate": 7.661483253588517e-05,
681
+ "loss": 0.6356,
682
+ "step": 480
683
+ },
684
+ {
685
+ "epoch": 0.2756464904802501,
686
+ "grad_norm": 0.32764124870300293,
687
+ "learning_rate": 7.631578947368422e-05,
688
+ "loss": 0.6228,
689
+ "step": 485
690
+ },
691
+ {
692
+ "epoch": 0.2784882068769537,
693
+ "grad_norm": 0.3466978073120117,
694
+ "learning_rate": 7.601674641148325e-05,
695
+ "loss": 0.6198,
696
+ "step": 490
697
+ },
698
+ {
699
+ "epoch": 0.2813299232736573,
700
+ "grad_norm": 0.4293256103992462,
701
+ "learning_rate": 7.571770334928229e-05,
702
+ "loss": 0.6226,
703
+ "step": 495
704
+ },
705
+ {
706
+ "epoch": 0.2841716396703609,
707
+ "grad_norm": 0.31901758909225464,
708
+ "learning_rate": 7.541866028708135e-05,
709
+ "loss": 0.6207,
710
+ "step": 500
711
+ },
712
+ {
713
+ "epoch": 0.2870133560670645,
714
+ "grad_norm": 0.2964116632938385,
715
+ "learning_rate": 7.511961722488039e-05,
716
+ "loss": 0.6676,
717
+ "step": 505
718
+ },
719
+ {
720
+ "epoch": 0.2898550724637681,
721
+ "grad_norm": 0.3660202920436859,
722
+ "learning_rate": 7.482057416267943e-05,
723
+ "loss": 0.6447,
724
+ "step": 510
725
+ },
726
+ {
727
+ "epoch": 0.29269678886047173,
728
+ "grad_norm": 0.3501967191696167,
729
+ "learning_rate": 7.452153110047847e-05,
730
+ "loss": 0.6375,
731
+ "step": 515
732
+ },
733
+ {
734
+ "epoch": 0.29553850525717534,
735
+ "grad_norm": 0.3197890818119049,
736
+ "learning_rate": 7.422248803827751e-05,
737
+ "loss": 0.6079,
738
+ "step": 520
739
+ },
740
+ {
741
+ "epoch": 0.29838022165387895,
742
+ "grad_norm": 0.29974034428596497,
743
+ "learning_rate": 7.392344497607656e-05,
744
+ "loss": 0.6122,
745
+ "step": 525
746
+ },
747
+ {
748
+ "epoch": 0.30122193805058256,
749
+ "grad_norm": 0.3449777662754059,
750
+ "learning_rate": 7.36244019138756e-05,
751
+ "loss": 0.6421,
752
+ "step": 530
753
+ },
754
+ {
755
+ "epoch": 0.30406365444728617,
756
+ "grad_norm": 0.32783225178718567,
757
+ "learning_rate": 7.332535885167465e-05,
758
+ "loss": 0.6347,
759
+ "step": 535
760
+ },
761
+ {
762
+ "epoch": 0.3069053708439898,
763
+ "grad_norm": 0.3783082365989685,
764
+ "learning_rate": 7.302631578947369e-05,
765
+ "loss": 0.6063,
766
+ "step": 540
767
+ },
768
+ {
769
+ "epoch": 0.3097470872406934,
770
+ "grad_norm": 0.4460213780403137,
771
+ "learning_rate": 7.272727272727273e-05,
772
+ "loss": 0.6173,
773
+ "step": 545
774
+ },
775
+ {
776
+ "epoch": 0.312588803637397,
777
+ "grad_norm": 0.4322983920574188,
778
+ "learning_rate": 7.242822966507178e-05,
779
+ "loss": 0.6018,
780
+ "step": 550
781
+ },
782
+ {
783
+ "epoch": 0.3154305200341006,
784
+ "grad_norm": 0.4428264796733856,
785
+ "learning_rate": 7.212918660287081e-05,
786
+ "loss": 0.6555,
787
+ "step": 555
788
+ },
789
+ {
790
+ "epoch": 0.3182722364308042,
791
+ "grad_norm": 0.37189781665802,
792
+ "learning_rate": 7.183014354066986e-05,
793
+ "loss": 0.6198,
794
+ "step": 560
795
+ },
796
+ {
797
+ "epoch": 0.3211139528275078,
798
+ "grad_norm": 0.3604643642902374,
799
+ "learning_rate": 7.15311004784689e-05,
800
+ "loss": 0.613,
801
+ "step": 565
802
+ },
803
+ {
804
+ "epoch": 0.32395566922421143,
805
+ "grad_norm": 0.5619745850563049,
806
+ "learning_rate": 7.123205741626795e-05,
807
+ "loss": 0.6068,
808
+ "step": 570
809
+ },
810
+ {
811
+ "epoch": 0.32679738562091504,
812
+ "grad_norm": 0.3009982109069824,
813
+ "learning_rate": 7.093301435406699e-05,
814
+ "loss": 0.5959,
815
+ "step": 575
816
+ },
817
+ {
818
+ "epoch": 0.32963910201761865,
819
+ "grad_norm": 0.4559624195098877,
820
+ "learning_rate": 7.063397129186603e-05,
821
+ "loss": 0.6372,
822
+ "step": 580
823
+ },
824
+ {
825
+ "epoch": 0.33248081841432225,
826
+ "grad_norm": 0.35429826378822327,
827
+ "learning_rate": 7.033492822966508e-05,
828
+ "loss": 0.6158,
829
+ "step": 585
830
+ },
831
+ {
832
+ "epoch": 0.33532253481102586,
833
+ "grad_norm": 0.29706600308418274,
834
+ "learning_rate": 7.003588516746412e-05,
835
+ "loss": 0.6209,
836
+ "step": 590
837
+ },
838
+ {
839
+ "epoch": 0.33816425120772947,
840
+ "grad_norm": 0.40374723076820374,
841
+ "learning_rate": 6.973684210526315e-05,
842
+ "loss": 0.6075,
843
+ "step": 595
844
+ },
845
+ {
846
+ "epoch": 0.3410059676044331,
847
+ "grad_norm": 0.30447104573249817,
848
+ "learning_rate": 6.94377990430622e-05,
849
+ "loss": 0.6086,
850
+ "step": 600
851
+ },
852
+ {
853
+ "epoch": 0.3438476840011367,
854
+ "grad_norm": 0.3145429193973541,
855
+ "learning_rate": 6.913875598086125e-05,
856
+ "loss": 0.6463,
857
+ "step": 605
858
+ },
859
+ {
860
+ "epoch": 0.3466894003978403,
861
+ "grad_norm": 0.3488779366016388,
862
+ "learning_rate": 6.883971291866029e-05,
863
+ "loss": 0.632,
864
+ "step": 610
865
+ },
866
+ {
867
+ "epoch": 0.3495311167945439,
868
+ "grad_norm": 0.38298460841178894,
869
+ "learning_rate": 6.854066985645934e-05,
870
+ "loss": 0.6289,
871
+ "step": 615
872
+ },
873
+ {
874
+ "epoch": 0.3523728331912475,
875
+ "grad_norm": 0.33905914425849915,
876
+ "learning_rate": 6.824162679425837e-05,
877
+ "loss": 0.6517,
878
+ "step": 620
879
+ },
880
+ {
881
+ "epoch": 0.3552145495879511,
882
+ "grad_norm": 0.2896723747253418,
883
+ "learning_rate": 6.794258373205742e-05,
884
+ "loss": 0.6327,
885
+ "step": 625
886
+ },
887
+ {
888
+ "epoch": 0.35805626598465473,
889
+ "grad_norm": 0.4509832262992859,
890
+ "learning_rate": 6.764354066985646e-05,
891
+ "loss": 0.6305,
892
+ "step": 630
893
+ },
894
+ {
895
+ "epoch": 0.36089798238135834,
896
+ "grad_norm": 0.4173915386199951,
897
+ "learning_rate": 6.73444976076555e-05,
898
+ "loss": 0.6349,
899
+ "step": 635
900
+ },
901
+ {
902
+ "epoch": 0.36373969877806195,
903
+ "grad_norm": 0.4019744396209717,
904
+ "learning_rate": 6.704545454545455e-05,
905
+ "loss": 0.6271,
906
+ "step": 640
907
+ },
908
+ {
909
+ "epoch": 0.36658141517476556,
910
+ "grad_norm": 0.33760711550712585,
911
+ "learning_rate": 6.674641148325359e-05,
912
+ "loss": 0.6351,
913
+ "step": 645
914
+ },
915
+ {
916
+ "epoch": 0.36942313157146917,
917
+ "grad_norm": 0.4120747745037079,
918
+ "learning_rate": 6.644736842105264e-05,
919
+ "loss": 0.629,
920
+ "step": 650
921
+ },
922
+ {
923
+ "epoch": 0.3722648479681728,
924
+ "grad_norm": 0.3481716215610504,
925
+ "learning_rate": 6.614832535885168e-05,
926
+ "loss": 0.6278,
927
+ "step": 655
928
+ },
929
+ {
930
+ "epoch": 0.3751065643648764,
931
+ "grad_norm": 0.3549588620662689,
932
+ "learning_rate": 6.584928229665072e-05,
933
+ "loss": 0.6412,
934
+ "step": 660
935
+ },
936
+ {
937
+ "epoch": 0.37794828076158,
938
+ "grad_norm": 0.35473644733428955,
939
+ "learning_rate": 6.555023923444976e-05,
940
+ "loss": 0.6274,
941
+ "step": 665
942
+ },
943
+ {
944
+ "epoch": 0.3807899971582836,
945
+ "grad_norm": 0.39416638016700745,
946
+ "learning_rate": 6.52511961722488e-05,
947
+ "loss": 0.6222,
948
+ "step": 670
949
+ },
950
+ {
951
+ "epoch": 0.3836317135549872,
952
+ "grad_norm": 0.40284669399261475,
953
+ "learning_rate": 6.495215311004786e-05,
954
+ "loss": 0.618,
955
+ "step": 675
956
+ },
957
+ {
958
+ "epoch": 0.3864734299516908,
959
+ "grad_norm": 0.3757275938987732,
960
+ "learning_rate": 6.465311004784689e-05,
961
+ "loss": 0.6091,
962
+ "step": 680
963
+ },
964
+ {
965
+ "epoch": 0.3893151463483944,
966
+ "grad_norm": 0.37024828791618347,
967
+ "learning_rate": 6.435406698564594e-05,
968
+ "loss": 0.6651,
969
+ "step": 685
970
+ },
971
+ {
972
+ "epoch": 0.39215686274509803,
973
+ "grad_norm": 0.388102650642395,
974
+ "learning_rate": 6.405502392344498e-05,
975
+ "loss": 0.6431,
976
+ "step": 690
977
+ },
978
+ {
979
+ "epoch": 0.39499857914180164,
980
+ "grad_norm": 0.3869297206401825,
981
+ "learning_rate": 6.375598086124402e-05,
982
+ "loss": 0.6338,
983
+ "step": 695
984
+ },
985
+ {
986
+ "epoch": 0.39784029553850525,
987
+ "grad_norm": 0.35515597462654114,
988
+ "learning_rate": 6.345693779904308e-05,
989
+ "loss": 0.6118,
990
+ "step": 700
991
+ },
992
+ {
993
+ "epoch": 0.40068201193520886,
994
+ "grad_norm": 0.47697246074676514,
995
+ "learning_rate": 6.31578947368421e-05,
996
+ "loss": 0.6341,
997
+ "step": 705
998
+ },
999
+ {
1000
+ "epoch": 0.40352372833191247,
1001
+ "grad_norm": 0.3087664246559143,
1002
+ "learning_rate": 6.285885167464114e-05,
1003
+ "loss": 0.644,
1004
+ "step": 710
1005
+ },
1006
+ {
1007
+ "epoch": 0.4063654447286161,
1008
+ "grad_norm": 0.32958847284317017,
1009
+ "learning_rate": 6.25598086124402e-05,
1010
+ "loss": 0.6274,
1011
+ "step": 715
1012
+ },
1013
+ {
1014
+ "epoch": 0.4092071611253197,
1015
+ "grad_norm": 0.3053686022758484,
1016
+ "learning_rate": 6.226076555023924e-05,
1017
+ "loss": 0.6186,
1018
+ "step": 720
1019
+ },
1020
+ {
1021
+ "epoch": 0.4120488775220233,
1022
+ "grad_norm": 0.3156355619430542,
1023
+ "learning_rate": 6.196172248803828e-05,
1024
+ "loss": 0.5837,
1025
+ "step": 725
1026
+ },
1027
+ {
1028
+ "epoch": 0.4148905939187269,
1029
+ "grad_norm": 0.3579465448856354,
1030
+ "learning_rate": 6.166267942583732e-05,
1031
+ "loss": 0.6296,
1032
+ "step": 730
1033
+ },
1034
+ {
1035
+ "epoch": 0.4177323103154305,
1036
+ "grad_norm": 0.34638217091560364,
1037
+ "learning_rate": 6.136363636363636e-05,
1038
+ "loss": 0.6556,
1039
+ "step": 735
1040
+ },
1041
+ {
1042
+ "epoch": 0.4205740267121341,
1043
+ "grad_norm": 0.3303477168083191,
1044
+ "learning_rate": 6.106459330143542e-05,
1045
+ "loss": 0.6303,
1046
+ "step": 740
1047
+ },
1048
+ {
1049
+ "epoch": 0.4234157431088377,
1050
+ "grad_norm": 0.4013735055923462,
1051
+ "learning_rate": 6.076555023923445e-05,
1052
+ "loss": 0.6111,
1053
+ "step": 745
1054
+ },
1055
+ {
1056
+ "epoch": 0.42625745950554134,
1057
+ "grad_norm": 0.3250595033168793,
1058
+ "learning_rate": 6.046650717703349e-05,
1059
+ "loss": 0.6152,
1060
+ "step": 750
1061
+ },
1062
+ {
1063
+ "epoch": 0.42909917590224494,
1064
+ "grad_norm": 0.44146016240119934,
1065
+ "learning_rate": 6.016746411483254e-05,
1066
+ "loss": 0.6362,
1067
+ "step": 755
1068
+ },
1069
+ {
1070
+ "epoch": 0.43194089229894855,
1071
+ "grad_norm": 0.33652061223983765,
1072
+ "learning_rate": 5.9868421052631587e-05,
1073
+ "loss": 0.6194,
1074
+ "step": 760
1075
+ },
1076
+ {
1077
+ "epoch": 0.43478260869565216,
1078
+ "grad_norm": 0.34849822521209717,
1079
+ "learning_rate": 5.956937799043063e-05,
1080
+ "loss": 0.6048,
1081
+ "step": 765
1082
+ },
1083
+ {
1084
+ "epoch": 0.43762432509235577,
1085
+ "grad_norm": 0.4080073833465576,
1086
+ "learning_rate": 5.927033492822967e-05,
1087
+ "loss": 0.6134,
1088
+ "step": 770
1089
+ },
1090
+ {
1091
+ "epoch": 0.4404660414890594,
1092
+ "grad_norm": 0.3164100646972656,
1093
+ "learning_rate": 5.897129186602871e-05,
1094
+ "loss": 0.621,
1095
+ "step": 775
1096
+ },
1097
+ {
1098
+ "epoch": 0.443307757885763,
1099
+ "grad_norm": 0.3303377330303192,
1100
+ "learning_rate": 5.867224880382776e-05,
1101
+ "loss": 0.644,
1102
+ "step": 780
1103
+ },
1104
+ {
1105
+ "epoch": 0.4461494742824666,
1106
+ "grad_norm": 0.31213077902793884,
1107
+ "learning_rate": 5.8373205741626804e-05,
1108
+ "loss": 0.6079,
1109
+ "step": 785
1110
+ },
1111
+ {
1112
+ "epoch": 0.4489911906791702,
1113
+ "grad_norm": 0.30722832679748535,
1114
+ "learning_rate": 5.807416267942584e-05,
1115
+ "loss": 0.6109,
1116
+ "step": 790
1117
+ },
1118
+ {
1119
+ "epoch": 0.4518329070758738,
1120
+ "grad_norm": 0.4260627031326294,
1121
+ "learning_rate": 5.777511961722488e-05,
1122
+ "loss": 0.6024,
1123
+ "step": 795
1124
+ },
1125
+ {
1126
+ "epoch": 0.4546746234725774,
1127
+ "grad_norm": 0.3097389042377472,
1128
+ "learning_rate": 5.747607655502393e-05,
1129
+ "loss": 0.5984,
1130
+ "step": 800
1131
+ },
1132
+ {
1133
+ "epoch": 0.45751633986928103,
1134
+ "grad_norm": 0.31624141335487366,
1135
+ "learning_rate": 5.7177033492822975e-05,
1136
+ "loss": 0.6036,
1137
+ "step": 805
1138
+ },
1139
+ {
1140
+ "epoch": 0.46035805626598464,
1141
+ "grad_norm": 0.32740023732185364,
1142
+ "learning_rate": 5.687799043062201e-05,
1143
+ "loss": 0.6317,
1144
+ "step": 810
1145
+ },
1146
+ {
1147
+ "epoch": 0.46319977266268825,
1148
+ "grad_norm": 0.338337242603302,
1149
+ "learning_rate": 5.6578947368421056e-05,
1150
+ "loss": 0.6406,
1151
+ "step": 815
1152
+ },
1153
+ {
1154
+ "epoch": 0.46604148905939186,
1155
+ "grad_norm": 0.35633623600006104,
1156
+ "learning_rate": 5.62799043062201e-05,
1157
+ "loss": 0.6294,
1158
+ "step": 820
1159
+ },
1160
+ {
1161
+ "epoch": 0.46888320545609546,
1162
+ "grad_norm": 0.29533228278160095,
1163
+ "learning_rate": 5.5980861244019145e-05,
1164
+ "loss": 0.6397,
1165
+ "step": 825
1166
+ },
1167
+ {
1168
+ "epoch": 0.4717249218527991,
1169
+ "grad_norm": 0.2982120215892792,
1170
+ "learning_rate": 5.568181818181818e-05,
1171
+ "loss": 0.6177,
1172
+ "step": 830
1173
+ },
1174
+ {
1175
+ "epoch": 0.4745666382495027,
1176
+ "grad_norm": 0.31643518805503845,
1177
+ "learning_rate": 5.5382775119617226e-05,
1178
+ "loss": 0.6306,
1179
+ "step": 835
1180
+ },
1181
+ {
1182
+ "epoch": 0.4774083546462063,
1183
+ "grad_norm": 0.3755875527858734,
1184
+ "learning_rate": 5.5083732057416274e-05,
1185
+ "loss": 0.5975,
1186
+ "step": 840
1187
+ },
1188
+ {
1189
+ "epoch": 0.4802500710429099,
1190
+ "grad_norm": 0.3546150326728821,
1191
+ "learning_rate": 5.4784688995215315e-05,
1192
+ "loss": 0.6181,
1193
+ "step": 845
1194
+ },
1195
+ {
1196
+ "epoch": 0.4830917874396135,
1197
+ "grad_norm": 0.334256649017334,
1198
+ "learning_rate": 5.448564593301436e-05,
1199
+ "loss": 0.6029,
1200
+ "step": 850
1201
+ },
1202
+ {
1203
+ "epoch": 0.4859335038363171,
1204
+ "grad_norm": 0.29167649149894714,
1205
+ "learning_rate": 5.4186602870813397e-05,
1206
+ "loss": 0.6158,
1207
+ "step": 855
1208
+ },
1209
+ {
1210
+ "epoch": 0.4887752202330207,
1211
+ "grad_norm": 0.3616998493671417,
1212
+ "learning_rate": 5.3887559808612444e-05,
1213
+ "loss": 0.6634,
1214
+ "step": 860
1215
+ },
1216
+ {
1217
+ "epoch": 0.49161693662972433,
1218
+ "grad_norm": 0.3217279314994812,
1219
+ "learning_rate": 5.3588516746411485e-05,
1220
+ "loss": 0.6273,
1221
+ "step": 865
1222
+ },
1223
+ {
1224
+ "epoch": 0.49445865302642794,
1225
+ "grad_norm": 0.331284761428833,
1226
+ "learning_rate": 5.328947368421053e-05,
1227
+ "loss": 0.5997,
1228
+ "step": 870
1229
+ },
1230
+ {
1231
+ "epoch": 0.49730036942313155,
1232
+ "grad_norm": 0.33052241802215576,
1233
+ "learning_rate": 5.299043062200957e-05,
1234
+ "loss": 0.6166,
1235
+ "step": 875
1236
+ },
1237
+ {
1238
+ "epoch": 0.5001420858198352,
1239
+ "grad_norm": 0.3654235899448395,
1240
+ "learning_rate": 5.2691387559808614e-05,
1241
+ "loss": 0.6235,
1242
+ "step": 880
1243
+ },
1244
+ {
1245
+ "epoch": 0.5029838022165388,
1246
+ "grad_norm": 0.3484792709350586,
1247
+ "learning_rate": 5.239234449760766e-05,
1248
+ "loss": 0.6112,
1249
+ "step": 885
1250
+ },
1251
+ {
1252
+ "epoch": 0.5058255186132424,
1253
+ "grad_norm": 0.34995678067207336,
1254
+ "learning_rate": 5.20933014354067e-05,
1255
+ "loss": 0.6108,
1256
+ "step": 890
1257
+ },
1258
+ {
1259
+ "epoch": 0.508667235009946,
1260
+ "grad_norm": 0.3473326563835144,
1261
+ "learning_rate": 5.179425837320574e-05,
1262
+ "loss": 0.6332,
1263
+ "step": 895
1264
+ },
1265
+ {
1266
+ "epoch": 0.5115089514066496,
1267
+ "grad_norm": 0.3197755515575409,
1268
+ "learning_rate": 5.1495215311004785e-05,
1269
+ "loss": 0.6137,
1270
+ "step": 900
1271
+ },
1272
+ {
1273
+ "epoch": 0.5143506678033533,
1274
+ "grad_norm": 0.3330087959766388,
1275
+ "learning_rate": 5.119617224880383e-05,
1276
+ "loss": 0.6414,
1277
+ "step": 905
1278
+ },
1279
+ {
1280
+ "epoch": 0.5171923842000569,
1281
+ "grad_norm": 0.40548765659332275,
1282
+ "learning_rate": 5.089712918660288e-05,
1283
+ "loss": 0.6472,
1284
+ "step": 910
1285
+ },
1286
+ {
1287
+ "epoch": 0.5200341005967605,
1288
+ "grad_norm": 0.34635302424430847,
1289
+ "learning_rate": 5.0598086124401914e-05,
1290
+ "loss": 0.6019,
1291
+ "step": 915
1292
+ },
1293
+ {
1294
+ "epoch": 0.5228758169934641,
1295
+ "grad_norm": 0.38793084025382996,
1296
+ "learning_rate": 5.0299043062200955e-05,
1297
+ "loss": 0.6166,
1298
+ "step": 920
1299
+ },
1300
+ {
1301
+ "epoch": 0.5257175333901677,
1302
+ "grad_norm": 0.264597088098526,
1303
+ "learning_rate": 5e-05,
1304
+ "loss": 0.6226,
1305
+ "step": 925
1306
+ },
1307
+ {
1308
+ "epoch": 0.5285592497868713,
1309
+ "grad_norm": 0.39633965492248535,
1310
+ "learning_rate": 4.970095693779905e-05,
1311
+ "loss": 0.6373,
1312
+ "step": 930
1313
+ },
1314
+ {
1315
+ "epoch": 0.5314009661835749,
1316
+ "grad_norm": 0.2989448308944702,
1317
+ "learning_rate": 4.940191387559809e-05,
1318
+ "loss": 0.6075,
1319
+ "step": 935
1320
+ },
1321
+ {
1322
+ "epoch": 0.5342426825802785,
1323
+ "grad_norm": 0.29738152027130127,
1324
+ "learning_rate": 4.910287081339713e-05,
1325
+ "loss": 0.6093,
1326
+ "step": 940
1327
+ },
1328
+ {
1329
+ "epoch": 0.5370843989769821,
1330
+ "grad_norm": 0.3517061471939087,
1331
+ "learning_rate": 4.880382775119617e-05,
1332
+ "loss": 0.6066,
1333
+ "step": 945
1334
+ },
1335
+ {
1336
+ "epoch": 0.5399261153736857,
1337
+ "grad_norm": 0.28003114461898804,
1338
+ "learning_rate": 4.850478468899522e-05,
1339
+ "loss": 0.613,
1340
+ "step": 950
1341
+ },
1342
+ {
1343
+ "epoch": 0.5427678317703893,
1344
+ "grad_norm": 0.32199013233184814,
1345
+ "learning_rate": 4.820574162679426e-05,
1346
+ "loss": 0.6376,
1347
+ "step": 955
1348
+ },
1349
+ {
1350
+ "epoch": 0.545609548167093,
1351
+ "grad_norm": 0.3068143129348755,
1352
+ "learning_rate": 4.790669856459331e-05,
1353
+ "loss": 0.6308,
1354
+ "step": 960
1355
+ },
1356
+ {
1357
+ "epoch": 0.5484512645637966,
1358
+ "grad_norm": 0.34744665026664734,
1359
+ "learning_rate": 4.760765550239234e-05,
1360
+ "loss": 0.5983,
1361
+ "step": 965
1362
+ },
1363
+ {
1364
+ "epoch": 0.5512929809605002,
1365
+ "grad_norm": 0.3580724000930786,
1366
+ "learning_rate": 4.730861244019139e-05,
1367
+ "loss": 0.6032,
1368
+ "step": 970
1369
+ },
1370
+ {
1371
+ "epoch": 0.5541346973572038,
1372
+ "grad_norm": 0.32625916600227356,
1373
+ "learning_rate": 4.700956937799043e-05,
1374
+ "loss": 0.604,
1375
+ "step": 975
1376
+ },
1377
+ {
1378
+ "epoch": 0.5569764137539074,
1379
+ "grad_norm": 0.31557053327560425,
1380
+ "learning_rate": 4.671052631578948e-05,
1381
+ "loss": 0.6345,
1382
+ "step": 980
1383
+ },
1384
+ {
1385
+ "epoch": 0.559818130150611,
1386
+ "grad_norm": 0.3296593725681305,
1387
+ "learning_rate": 4.641148325358852e-05,
1388
+ "loss": 0.6327,
1389
+ "step": 985
1390
+ },
1391
+ {
1392
+ "epoch": 0.5626598465473146,
1393
+ "grad_norm": 0.34768185019493103,
1394
+ "learning_rate": 4.611244019138756e-05,
1395
+ "loss": 0.6414,
1396
+ "step": 990
1397
+ },
1398
+ {
1399
+ "epoch": 0.5655015629440182,
1400
+ "grad_norm": 0.3945872485637665,
1401
+ "learning_rate": 4.58133971291866e-05,
1402
+ "loss": 0.6039,
1403
+ "step": 995
1404
+ },
1405
+ {
1406
+ "epoch": 0.5683432793407218,
1407
+ "grad_norm": 0.24539418518543243,
1408
+ "learning_rate": 4.551435406698565e-05,
1409
+ "loss": 0.5948,
1410
+ "step": 1000
1411
+ },
1412
+ {
1413
+ "epoch": 0.5711849957374254,
1414
+ "grad_norm": 0.28328627347946167,
1415
+ "learning_rate": 4.521531100478469e-05,
1416
+ "loss": 0.6401,
1417
+ "step": 1005
1418
+ },
1419
+ {
1420
+ "epoch": 0.574026712134129,
1421
+ "grad_norm": 0.28673475980758667,
1422
+ "learning_rate": 4.491626794258373e-05,
1423
+ "loss": 0.6502,
1424
+ "step": 1010
1425
+ },
1426
+ {
1427
+ "epoch": 0.5768684285308326,
1428
+ "grad_norm": 0.3333253860473633,
1429
+ "learning_rate": 4.461722488038278e-05,
1430
+ "loss": 0.6111,
1431
+ "step": 1015
1432
+ },
1433
+ {
1434
+ "epoch": 0.5797101449275363,
1435
+ "grad_norm": 0.4006788432598114,
1436
+ "learning_rate": 4.431818181818182e-05,
1437
+ "loss": 0.6322,
1438
+ "step": 1020
1439
+ },
1440
+ {
1441
+ "epoch": 0.5825518613242399,
1442
+ "grad_norm": 0.3080488443374634,
1443
+ "learning_rate": 4.401913875598087e-05,
1444
+ "loss": 0.5868,
1445
+ "step": 1025
1446
+ },
1447
+ {
1448
+ "epoch": 0.5853935777209435,
1449
+ "grad_norm": 0.3097079396247864,
1450
+ "learning_rate": 4.372009569377991e-05,
1451
+ "loss": 0.6351,
1452
+ "step": 1030
1453
+ },
1454
+ {
1455
+ "epoch": 0.5882352941176471,
1456
+ "grad_norm": 0.3037853538990021,
1457
+ "learning_rate": 4.342105263157895e-05,
1458
+ "loss": 0.6336,
1459
+ "step": 1035
1460
+ },
1461
+ {
1462
+ "epoch": 0.5910770105143507,
1463
+ "grad_norm": 0.31227898597717285,
1464
+ "learning_rate": 4.312200956937799e-05,
1465
+ "loss": 0.6098,
1466
+ "step": 1040
1467
+ },
1468
+ {
1469
+ "epoch": 0.5939187269110543,
1470
+ "grad_norm": 0.35149067640304565,
1471
+ "learning_rate": 4.282296650717704e-05,
1472
+ "loss": 0.607,
1473
+ "step": 1045
1474
+ },
1475
+ {
1476
+ "epoch": 0.5967604433077579,
1477
+ "grad_norm": 0.29151421785354614,
1478
+ "learning_rate": 4.252392344497608e-05,
1479
+ "loss": 0.5984,
1480
+ "step": 1050
1481
+ },
1482
+ {
1483
+ "epoch": 0.5996021597044615,
1484
+ "grad_norm": 0.3272274434566498,
1485
+ "learning_rate": 4.2224880382775126e-05,
1486
+ "loss": 0.6212,
1487
+ "step": 1055
1488
+ },
1489
+ {
1490
+ "epoch": 0.6024438761011651,
1491
+ "grad_norm": 0.31667497754096985,
1492
+ "learning_rate": 4.192583732057416e-05,
1493
+ "loss": 0.6426,
1494
+ "step": 1060
1495
+ },
1496
+ {
1497
+ "epoch": 0.6052855924978687,
1498
+ "grad_norm": 0.3101164698600769,
1499
+ "learning_rate": 4.162679425837321e-05,
1500
+ "loss": 0.6305,
1501
+ "step": 1065
1502
+ },
1503
+ {
1504
+ "epoch": 0.6081273088945723,
1505
+ "grad_norm": 0.34031999111175537,
1506
+ "learning_rate": 4.132775119617225e-05,
1507
+ "loss": 0.6255,
1508
+ "step": 1070
1509
+ },
1510
+ {
1511
+ "epoch": 0.610969025291276,
1512
+ "grad_norm": 0.27737563848495483,
1513
+ "learning_rate": 4.1028708133971296e-05,
1514
+ "loss": 0.6225,
1515
+ "step": 1075
1516
+ },
1517
+ {
1518
+ "epoch": 0.6138107416879796,
1519
+ "grad_norm": 0.40383121371269226,
1520
+ "learning_rate": 4.0729665071770337e-05,
1521
+ "loss": 0.6025,
1522
+ "step": 1080
1523
+ },
1524
+ {
1525
+ "epoch": 0.6166524580846832,
1526
+ "grad_norm": 0.3496963083744049,
1527
+ "learning_rate": 4.043062200956938e-05,
1528
+ "loss": 0.6315,
1529
+ "step": 1085
1530
+ },
1531
+ {
1532
+ "epoch": 0.6194941744813868,
1533
+ "grad_norm": 0.38513296842575073,
1534
+ "learning_rate": 4.0131578947368425e-05,
1535
+ "loss": 0.6102,
1536
+ "step": 1090
1537
+ },
1538
+ {
1539
+ "epoch": 0.6223358908780904,
1540
+ "grad_norm": 0.34017202258110046,
1541
+ "learning_rate": 3.9832535885167466e-05,
1542
+ "loss": 0.6215,
1543
+ "step": 1095
1544
+ },
1545
+ {
1546
+ "epoch": 0.625177607274794,
1547
+ "grad_norm": 0.554558515548706,
1548
+ "learning_rate": 3.9533492822966514e-05,
1549
+ "loss": 0.6041,
1550
+ "step": 1100
1551
+ },
1552
+ {
1553
+ "epoch": 0.6280193236714976,
1554
+ "grad_norm": 0.32666856050491333,
1555
+ "learning_rate": 3.9234449760765554e-05,
1556
+ "loss": 0.6389,
1557
+ "step": 1105
1558
+ },
1559
+ {
1560
+ "epoch": 0.6308610400682012,
1561
+ "grad_norm": 0.30029165744781494,
1562
+ "learning_rate": 3.8935406698564595e-05,
1563
+ "loss": 0.6032,
1564
+ "step": 1110
1565
+ },
1566
+ {
1567
+ "epoch": 0.6337027564649048,
1568
+ "grad_norm": 0.3267216682434082,
1569
+ "learning_rate": 3.8636363636363636e-05,
1570
+ "loss": 0.5841,
1571
+ "step": 1115
1572
+ },
1573
+ {
1574
+ "epoch": 0.6365444728616084,
1575
+ "grad_norm": 0.36103537678718567,
1576
+ "learning_rate": 3.8337320574162684e-05,
1577
+ "loss": 0.6212,
1578
+ "step": 1120
1579
+ },
1580
+ {
1581
+ "epoch": 0.639386189258312,
1582
+ "grad_norm": 0.3032275438308716,
1583
+ "learning_rate": 3.8038277511961725e-05,
1584
+ "loss": 0.6076,
1585
+ "step": 1125
1586
+ },
1587
+ {
1588
+ "epoch": 0.6422279056550156,
1589
+ "grad_norm": 0.3374808430671692,
1590
+ "learning_rate": 3.7739234449760765e-05,
1591
+ "loss": 0.6367,
1592
+ "step": 1130
1593
+ },
1594
+ {
1595
+ "epoch": 0.6450696220517192,
1596
+ "grad_norm": 0.34244582056999207,
1597
+ "learning_rate": 3.7440191387559806e-05,
1598
+ "loss": 0.6278,
1599
+ "step": 1135
1600
+ },
1601
+ {
1602
+ "epoch": 0.6479113384484229,
1603
+ "grad_norm": 0.3042035698890686,
1604
+ "learning_rate": 3.7141148325358854e-05,
1605
+ "loss": 0.6416,
1606
+ "step": 1140
1607
+ },
1608
+ {
1609
+ "epoch": 0.6507530548451265,
1610
+ "grad_norm": 0.3407208323478699,
1611
+ "learning_rate": 3.6842105263157895e-05,
1612
+ "loss": 0.6308,
1613
+ "step": 1145
1614
+ },
1615
+ {
1616
+ "epoch": 0.6535947712418301,
1617
+ "grad_norm": 0.29242879152297974,
1618
+ "learning_rate": 3.654306220095694e-05,
1619
+ "loss": 0.605,
1620
+ "step": 1150
1621
+ },
1622
+ {
1623
+ "epoch": 0.6564364876385337,
1624
+ "grad_norm": 0.3855131268501282,
1625
+ "learning_rate": 3.624401913875598e-05,
1626
+ "loss": 0.6224,
1627
+ "step": 1155
1628
+ },
1629
+ {
1630
+ "epoch": 0.6592782040352373,
1631
+ "grad_norm": 0.31428593397140503,
1632
+ "learning_rate": 3.5944976076555024e-05,
1633
+ "loss": 0.6192,
1634
+ "step": 1160
1635
+ },
1636
+ {
1637
+ "epoch": 0.6621199204319409,
1638
+ "grad_norm": 0.34910812973976135,
1639
+ "learning_rate": 3.5645933014354065e-05,
1640
+ "loss": 0.6089,
1641
+ "step": 1165
1642
+ },
1643
+ {
1644
+ "epoch": 0.6649616368286445,
1645
+ "grad_norm": 0.477639764547348,
1646
+ "learning_rate": 3.534688995215311e-05,
1647
+ "loss": 0.5893,
1648
+ "step": 1170
1649
+ },
1650
+ {
1651
+ "epoch": 0.6678033532253481,
1652
+ "grad_norm": 0.2696535289287567,
1653
+ "learning_rate": 3.504784688995216e-05,
1654
+ "loss": 0.6152,
1655
+ "step": 1175
1656
+ },
1657
+ {
1658
+ "epoch": 0.6706450696220517,
1659
+ "grad_norm": 0.37156441807746887,
1660
+ "learning_rate": 3.4748803827751194e-05,
1661
+ "loss": 0.6407,
1662
+ "step": 1180
1663
+ },
1664
+ {
1665
+ "epoch": 0.6734867860187553,
1666
+ "grad_norm": 0.41684263944625854,
1667
+ "learning_rate": 3.444976076555024e-05,
1668
+ "loss": 0.6403,
1669
+ "step": 1185
1670
+ },
1671
+ {
1672
+ "epoch": 0.6763285024154589,
1673
+ "grad_norm": 0.33099016547203064,
1674
+ "learning_rate": 3.415071770334928e-05,
1675
+ "loss": 0.6316,
1676
+ "step": 1190
1677
+ },
1678
+ {
1679
+ "epoch": 0.6791702188121626,
1680
+ "grad_norm": 0.3391123116016388,
1681
+ "learning_rate": 3.385167464114833e-05,
1682
+ "loss": 0.6248,
1683
+ "step": 1195
1684
+ },
1685
+ {
1686
+ "epoch": 0.6820119352088662,
1687
+ "grad_norm": 0.300188809633255,
1688
+ "learning_rate": 3.355263157894737e-05,
1689
+ "loss": 0.6126,
1690
+ "step": 1200
1691
+ },
1692
+ {
1693
+ "epoch": 0.6848536516055698,
1694
+ "grad_norm": 0.29901713132858276,
1695
+ "learning_rate": 3.325358851674641e-05,
1696
+ "loss": 0.6323,
1697
+ "step": 1205
1698
+ },
1699
+ {
1700
+ "epoch": 0.6876953680022734,
1701
+ "grad_norm": 0.3513938784599304,
1702
+ "learning_rate": 3.295454545454545e-05,
1703
+ "loss": 0.6217,
1704
+ "step": 1210
1705
+ },
1706
+ {
1707
+ "epoch": 0.690537084398977,
1708
+ "grad_norm": 0.3339061737060547,
1709
+ "learning_rate": 3.26555023923445e-05,
1710
+ "loss": 0.5939,
1711
+ "step": 1215
1712
+ },
1713
+ {
1714
+ "epoch": 0.6933788007956806,
1715
+ "grad_norm": 0.3780442178249359,
1716
+ "learning_rate": 3.235645933014354e-05,
1717
+ "loss": 0.6146,
1718
+ "step": 1220
1719
+ },
1720
+ {
1721
+ "epoch": 0.6962205171923842,
1722
+ "grad_norm": 0.3123363256454468,
1723
+ "learning_rate": 3.205741626794259e-05,
1724
+ "loss": 0.6049,
1725
+ "step": 1225
1726
+ },
1727
+ {
1728
+ "epoch": 0.6990622335890878,
1729
+ "grad_norm": 0.36633139848709106,
1730
+ "learning_rate": 3.175837320574162e-05,
1731
+ "loss": 0.5996,
1732
+ "step": 1230
1733
+ },
1734
+ {
1735
+ "epoch": 0.7019039499857914,
1736
+ "grad_norm": 0.33213770389556885,
1737
+ "learning_rate": 3.145933014354067e-05,
1738
+ "loss": 0.6068,
1739
+ "step": 1235
1740
+ },
1741
+ {
1742
+ "epoch": 0.704745666382495,
1743
+ "grad_norm": 0.34919023513793945,
1744
+ "learning_rate": 3.116028708133971e-05,
1745
+ "loss": 0.6106,
1746
+ "step": 1240
1747
+ },
1748
+ {
1749
+ "epoch": 0.7075873827791986,
1750
+ "grad_norm": 0.36192944645881653,
1751
+ "learning_rate": 3.086124401913876e-05,
1752
+ "loss": 0.6155,
1753
+ "step": 1245
1754
+ },
1755
+ {
1756
+ "epoch": 0.7104290991759022,
1757
+ "grad_norm": 0.27729424834251404,
1758
+ "learning_rate": 3.05622009569378e-05,
1759
+ "loss": 0.5967,
1760
+ "step": 1250
1761
+ },
1762
+ {
1763
+ "epoch": 0.7132708155726059,
1764
+ "grad_norm": 0.3338644802570343,
1765
+ "learning_rate": 3.0263157894736844e-05,
1766
+ "loss": 0.6185,
1767
+ "step": 1255
1768
+ },
1769
+ {
1770
+ "epoch": 0.7161125319693095,
1771
+ "grad_norm": 0.30155307054519653,
1772
+ "learning_rate": 2.996411483253589e-05,
1773
+ "loss": 0.6346,
1774
+ "step": 1260
1775
+ },
1776
+ {
1777
+ "epoch": 0.7189542483660131,
1778
+ "grad_norm": 0.35605382919311523,
1779
+ "learning_rate": 2.966507177033493e-05,
1780
+ "loss": 0.6173,
1781
+ "step": 1265
1782
+ },
1783
+ {
1784
+ "epoch": 0.7217959647627167,
1785
+ "grad_norm": 0.38695403933525085,
1786
+ "learning_rate": 2.9366028708133974e-05,
1787
+ "loss": 0.5802,
1788
+ "step": 1270
1789
+ },
1790
+ {
1791
+ "epoch": 0.7246376811594203,
1792
+ "grad_norm": 0.32234320044517517,
1793
+ "learning_rate": 2.9066985645933014e-05,
1794
+ "loss": 0.5972,
1795
+ "step": 1275
1796
+ },
1797
+ {
1798
+ "epoch": 0.7274793975561239,
1799
+ "grad_norm": 0.3004119396209717,
1800
+ "learning_rate": 2.8767942583732062e-05,
1801
+ "loss": 0.6412,
1802
+ "step": 1280
1803
+ },
1804
+ {
1805
+ "epoch": 0.7303211139528275,
1806
+ "grad_norm": 0.29393115639686584,
1807
+ "learning_rate": 2.84688995215311e-05,
1808
+ "loss": 0.646,
1809
+ "step": 1285
1810
+ },
1811
+ {
1812
+ "epoch": 0.7331628303495311,
1813
+ "grad_norm": 0.36121341586112976,
1814
+ "learning_rate": 2.8169856459330147e-05,
1815
+ "loss": 0.6305,
1816
+ "step": 1290
1817
+ },
1818
+ {
1819
+ "epoch": 0.7360045467462347,
1820
+ "grad_norm": 0.34062379598617554,
1821
+ "learning_rate": 2.7870813397129185e-05,
1822
+ "loss": 0.6302,
1823
+ "step": 1295
1824
+ },
1825
+ {
1826
+ "epoch": 0.7388462631429383,
1827
+ "grad_norm": 0.28281763195991516,
1828
+ "learning_rate": 2.7571770334928232e-05,
1829
+ "loss": 0.6027,
1830
+ "step": 1300
1831
+ },
1832
+ {
1833
+ "epoch": 0.7416879795396419,
1834
+ "grad_norm": 0.384242445230484,
1835
+ "learning_rate": 2.7272727272727273e-05,
1836
+ "loss": 0.6144,
1837
+ "step": 1305
1838
+ },
1839
+ {
1840
+ "epoch": 0.7445296959363455,
1841
+ "grad_norm": 0.32072150707244873,
1842
+ "learning_rate": 2.6973684210526317e-05,
1843
+ "loss": 0.6128,
1844
+ "step": 1310
1845
+ },
1846
+ {
1847
+ "epoch": 0.7473714123330492,
1848
+ "grad_norm": 0.29476672410964966,
1849
+ "learning_rate": 2.6674641148325358e-05,
1850
+ "loss": 0.6026,
1851
+ "step": 1315
1852
+ },
1853
+ {
1854
+ "epoch": 0.7502131287297528,
1855
+ "grad_norm": 0.3681996464729309,
1856
+ "learning_rate": 2.6375598086124402e-05,
1857
+ "loss": 0.6239,
1858
+ "step": 1320
1859
+ },
1860
+ {
1861
+ "epoch": 0.7530548451264564,
1862
+ "grad_norm": 0.26887091994285583,
1863
+ "learning_rate": 2.6076555023923443e-05,
1864
+ "loss": 0.6136,
1865
+ "step": 1325
1866
+ },
1867
+ {
1868
+ "epoch": 0.75589656152316,
1869
+ "grad_norm": 0.3026677072048187,
1870
+ "learning_rate": 2.5777511961722488e-05,
1871
+ "loss": 0.6077,
1872
+ "step": 1330
1873
+ },
1874
+ {
1875
+ "epoch": 0.7587382779198636,
1876
+ "grad_norm": 0.2949201464653015,
1877
+ "learning_rate": 2.5478468899521535e-05,
1878
+ "loss": 0.6348,
1879
+ "step": 1335
1880
+ },
1881
+ {
1882
+ "epoch": 0.7615799943165672,
1883
+ "grad_norm": 0.44444364309310913,
1884
+ "learning_rate": 2.5179425837320576e-05,
1885
+ "loss": 0.6272,
1886
+ "step": 1340
1887
+ },
1888
+ {
1889
+ "epoch": 0.7644217107132708,
1890
+ "grad_norm": 0.3795599341392517,
1891
+ "learning_rate": 2.4880382775119617e-05,
1892
+ "loss": 0.5893,
1893
+ "step": 1345
1894
+ },
1895
+ {
1896
+ "epoch": 0.7672634271099744,
1897
+ "grad_norm": 0.2788754999637604,
1898
+ "learning_rate": 2.458133971291866e-05,
1899
+ "loss": 0.6144,
1900
+ "step": 1350
1901
+ },
1902
+ {
1903
+ "epoch": 0.770105143506678,
1904
+ "grad_norm": 0.2978743016719818,
1905
+ "learning_rate": 2.4282296650717702e-05,
1906
+ "loss": 0.6174,
1907
+ "step": 1355
1908
+ },
1909
+ {
1910
+ "epoch": 0.7729468599033816,
1911
+ "grad_norm": 0.3350945711135864,
1912
+ "learning_rate": 2.3983253588516746e-05,
1913
+ "loss": 0.6386,
1914
+ "step": 1360
1915
+ },
1916
+ {
1917
+ "epoch": 0.7757885763000852,
1918
+ "grad_norm": 0.35026684403419495,
1919
+ "learning_rate": 2.368421052631579e-05,
1920
+ "loss": 0.6056,
1921
+ "step": 1365
1922
+ },
1923
+ {
1924
+ "epoch": 0.7786302926967889,
1925
+ "grad_norm": 0.34024810791015625,
1926
+ "learning_rate": 2.3385167464114835e-05,
1927
+ "loss": 0.5833,
1928
+ "step": 1370
1929
+ },
1930
+ {
1931
+ "epoch": 0.7814720090934925,
1932
+ "grad_norm": 0.24556630849838257,
1933
+ "learning_rate": 2.308612440191388e-05,
1934
+ "loss": 0.6065,
1935
+ "step": 1375
1936
+ },
1937
+ {
1938
+ "epoch": 0.7843137254901961,
1939
+ "grad_norm": 0.28934311866760254,
1940
+ "learning_rate": 2.278708133971292e-05,
1941
+ "loss": 0.6174,
1942
+ "step": 1380
1943
+ },
1944
+ {
1945
+ "epoch": 0.7871554418868997,
1946
+ "grad_norm": 0.3274039626121521,
1947
+ "learning_rate": 2.2488038277511964e-05,
1948
+ "loss": 0.5907,
1949
+ "step": 1385
1950
+ },
1951
+ {
1952
+ "epoch": 0.7899971582836033,
1953
+ "grad_norm": 0.37224775552749634,
1954
+ "learning_rate": 2.2188995215311005e-05,
1955
+ "loss": 0.6162,
1956
+ "step": 1390
1957
+ },
1958
+ {
1959
+ "epoch": 0.7928388746803069,
1960
+ "grad_norm": 0.3983578383922577,
1961
+ "learning_rate": 2.188995215311005e-05,
1962
+ "loss": 0.6076,
1963
+ "step": 1395
1964
+ },
1965
+ {
1966
+ "epoch": 0.7956805910770105,
1967
+ "grad_norm": 0.3312250077724457,
1968
+ "learning_rate": 2.1590909090909093e-05,
1969
+ "loss": 0.5998,
1970
+ "step": 1400
1971
+ },
1972
+ {
1973
+ "epoch": 0.7985223074737141,
1974
+ "grad_norm": 0.30029281973838806,
1975
+ "learning_rate": 2.1291866028708134e-05,
1976
+ "loss": 0.6038,
1977
+ "step": 1405
1978
+ },
1979
+ {
1980
+ "epoch": 0.8013640238704177,
1981
+ "grad_norm": 0.31737014651298523,
1982
+ "learning_rate": 2.099282296650718e-05,
1983
+ "loss": 0.6374,
1984
+ "step": 1410
1985
+ },
1986
+ {
1987
+ "epoch": 0.8042057402671213,
1988
+ "grad_norm": 0.3838863968849182,
1989
+ "learning_rate": 2.069377990430622e-05,
1990
+ "loss": 0.6095,
1991
+ "step": 1415
1992
+ },
1993
+ {
1994
+ "epoch": 0.8070474566638249,
1995
+ "grad_norm": 0.3589194416999817,
1996
+ "learning_rate": 2.0394736842105264e-05,
1997
+ "loss": 0.5981,
1998
+ "step": 1420
1999
+ },
2000
+ {
2001
+ "epoch": 0.8098891730605285,
2002
+ "grad_norm": 0.25485652685165405,
2003
+ "learning_rate": 2.0095693779904308e-05,
2004
+ "loss": 0.613,
2005
+ "step": 1425
2006
+ },
2007
+ {
2008
+ "epoch": 0.8127308894572322,
2009
+ "grad_norm": 0.31829777359962463,
2010
+ "learning_rate": 1.979665071770335e-05,
2011
+ "loss": 0.6162,
2012
+ "step": 1430
2013
+ },
2014
+ {
2015
+ "epoch": 0.8155726058539358,
2016
+ "grad_norm": 0.2884056866168976,
2017
+ "learning_rate": 1.9497607655502393e-05,
2018
+ "loss": 0.6167,
2019
+ "step": 1435
2020
+ },
2021
+ {
2022
+ "epoch": 0.8184143222506394,
2023
+ "grad_norm": 0.32646194100379944,
2024
+ "learning_rate": 1.9198564593301434e-05,
2025
+ "loss": 0.6286,
2026
+ "step": 1440
2027
+ },
2028
+ {
2029
+ "epoch": 0.821256038647343,
2030
+ "grad_norm": 0.3279251754283905,
2031
+ "learning_rate": 1.8899521531100478e-05,
2032
+ "loss": 0.6062,
2033
+ "step": 1445
2034
+ },
2035
+ {
2036
+ "epoch": 0.8240977550440466,
2037
+ "grad_norm": 0.26997557282447815,
2038
+ "learning_rate": 1.8600478468899522e-05,
2039
+ "loss": 0.6047,
2040
+ "step": 1450
2041
+ },
2042
+ {
2043
+ "epoch": 0.8269394714407502,
2044
+ "grad_norm": 0.31314706802368164,
2045
+ "learning_rate": 1.8301435406698566e-05,
2046
+ "loss": 0.6247,
2047
+ "step": 1455
2048
+ },
2049
+ {
2050
+ "epoch": 0.8297811878374538,
2051
+ "grad_norm": 0.35538068413734436,
2052
+ "learning_rate": 1.800239234449761e-05,
2053
+ "loss": 0.6251,
2054
+ "step": 1460
2055
+ },
2056
+ {
2057
+ "epoch": 0.8326229042341574,
2058
+ "grad_norm": 0.2876541018486023,
2059
+ "learning_rate": 1.770334928229665e-05,
2060
+ "loss": 0.6381,
2061
+ "step": 1465
2062
+ },
2063
+ {
2064
+ "epoch": 0.835464620630861,
2065
+ "grad_norm": 0.3482559323310852,
2066
+ "learning_rate": 1.7404306220095696e-05,
2067
+ "loss": 0.6,
2068
+ "step": 1470
2069
+ },
2070
+ {
2071
+ "epoch": 0.8383063370275646,
2072
+ "grad_norm": 0.31553223729133606,
2073
+ "learning_rate": 1.7105263157894737e-05,
2074
+ "loss": 0.6052,
2075
+ "step": 1475
2076
+ },
2077
+ {
2078
+ "epoch": 0.8411480534242682,
2079
+ "grad_norm": 0.3157096803188324,
2080
+ "learning_rate": 1.680622009569378e-05,
2081
+ "loss": 0.6289,
2082
+ "step": 1480
2083
+ },
2084
+ {
2085
+ "epoch": 0.8439897698209718,
2086
+ "grad_norm": 0.8569077849388123,
2087
+ "learning_rate": 1.6507177033492825e-05,
2088
+ "loss": 0.6284,
2089
+ "step": 1485
2090
+ },
2091
+ {
2092
+ "epoch": 0.8468314862176755,
2093
+ "grad_norm": 0.3131512701511383,
2094
+ "learning_rate": 1.6208133971291866e-05,
2095
+ "loss": 0.6119,
2096
+ "step": 1490
2097
+ },
2098
+ {
2099
+ "epoch": 0.8496732026143791,
2100
+ "grad_norm": 0.3499760627746582,
2101
+ "learning_rate": 1.590909090909091e-05,
2102
+ "loss": 0.5917,
2103
+ "step": 1495
2104
+ },
2105
+ {
2106
+ "epoch": 0.8525149190110827,
2107
+ "grad_norm": 0.39078933000564575,
2108
+ "learning_rate": 1.561004784688995e-05,
2109
+ "loss": 0.6052,
2110
+ "step": 1500
2111
+ },
2112
+ {
2113
+ "epoch": 0.8553566354077863,
2114
+ "grad_norm": 0.33687838912010193,
2115
+ "learning_rate": 1.5311004784688995e-05,
2116
+ "loss": 0.5972,
2117
+ "step": 1505
2118
+ },
2119
+ {
2120
+ "epoch": 0.8581983518044899,
2121
+ "grad_norm": 0.3067253530025482,
2122
+ "learning_rate": 1.5011961722488038e-05,
2123
+ "loss": 0.6471,
2124
+ "step": 1510
2125
+ },
2126
+ {
2127
+ "epoch": 0.8610400682011935,
2128
+ "grad_norm": 0.2994327247142792,
2129
+ "learning_rate": 1.471291866028708e-05,
2130
+ "loss": 0.6144,
2131
+ "step": 1515
2132
+ },
2133
+ {
2134
+ "epoch": 0.8638817845978971,
2135
+ "grad_norm": 0.39878565073013306,
2136
+ "learning_rate": 1.4413875598086125e-05,
2137
+ "loss": 0.5926,
2138
+ "step": 1520
2139
+ },
2140
+ {
2141
+ "epoch": 0.8667235009946007,
2142
+ "grad_norm": 0.2688925266265869,
2143
+ "learning_rate": 1.4114832535885167e-05,
2144
+ "loss": 0.6053,
2145
+ "step": 1525
2146
+ },
2147
+ {
2148
+ "epoch": 0.8695652173913043,
2149
+ "grad_norm": 0.32719510793685913,
2150
+ "learning_rate": 1.3815789473684213e-05,
2151
+ "loss": 0.6052,
2152
+ "step": 1530
2153
+ },
2154
+ {
2155
+ "epoch": 0.8724069337880079,
2156
+ "grad_norm": 0.31301820278167725,
2157
+ "learning_rate": 1.3516746411483256e-05,
2158
+ "loss": 0.6229,
2159
+ "step": 1535
2160
+ },
2161
+ {
2162
+ "epoch": 0.8752486501847115,
2163
+ "grad_norm": 0.34368881583213806,
2164
+ "learning_rate": 1.3217703349282298e-05,
2165
+ "loss": 0.6284,
2166
+ "step": 1540
2167
+ },
2168
+ {
2169
+ "epoch": 0.8780903665814151,
2170
+ "grad_norm": 0.39424312114715576,
2171
+ "learning_rate": 1.291866028708134e-05,
2172
+ "loss": 0.6072,
2173
+ "step": 1545
2174
+ },
2175
+ {
2176
+ "epoch": 0.8809320829781188,
2177
+ "grad_norm": 0.29532507061958313,
2178
+ "learning_rate": 1.2619617224880383e-05,
2179
+ "loss": 0.6141,
2180
+ "step": 1550
2181
+ },
2182
+ {
2183
+ "epoch": 0.8837737993748224,
2184
+ "grad_norm": 0.30461254715919495,
2185
+ "learning_rate": 1.2320574162679427e-05,
2186
+ "loss": 0.6188,
2187
+ "step": 1555
2188
+ },
2189
+ {
2190
+ "epoch": 0.886615515771526,
2191
+ "grad_norm": 0.3121693730354309,
2192
+ "learning_rate": 1.202153110047847e-05,
2193
+ "loss": 0.5892,
2194
+ "step": 1560
2195
+ },
2196
+ {
2197
+ "epoch": 0.8894572321682296,
2198
+ "grad_norm": 0.32039588689804077,
2199
+ "learning_rate": 1.1722488038277513e-05,
2200
+ "loss": 0.621,
2201
+ "step": 1565
2202
+ },
2203
+ {
2204
+ "epoch": 0.8922989485649332,
2205
+ "grad_norm": 0.4113796353340149,
2206
+ "learning_rate": 1.1423444976076555e-05,
2207
+ "loss": 0.6084,
2208
+ "step": 1570
2209
+ },
2210
+ {
2211
+ "epoch": 0.8951406649616368,
2212
+ "grad_norm": 0.2877901494503021,
2213
+ "learning_rate": 1.1124401913875598e-05,
2214
+ "loss": 0.5934,
2215
+ "step": 1575
2216
+ },
2217
+ {
2218
+ "epoch": 0.8979823813583404,
2219
+ "grad_norm": 0.3071552515029907,
2220
+ "learning_rate": 1.0825358851674642e-05,
2221
+ "loss": 0.6256,
2222
+ "step": 1580
2223
+ },
2224
+ {
2225
+ "epoch": 0.900824097755044,
2226
+ "grad_norm": 0.33426299691200256,
2227
+ "learning_rate": 1.0526315789473684e-05,
2228
+ "loss": 0.6101,
2229
+ "step": 1585
2230
+ },
2231
+ {
2232
+ "epoch": 0.9036658141517476,
2233
+ "grad_norm": 0.2870624363422394,
2234
+ "learning_rate": 1.0227272727272729e-05,
2235
+ "loss": 0.6125,
2236
+ "step": 1590
2237
+ },
2238
+ {
2239
+ "epoch": 0.9065075305484512,
2240
+ "grad_norm": 0.31790071725845337,
2241
+ "learning_rate": 9.928229665071771e-06,
2242
+ "loss": 0.6092,
2243
+ "step": 1595
2244
+ },
2245
+ {
2246
+ "epoch": 0.9093492469451548,
2247
+ "grad_norm": 0.30653512477874756,
2248
+ "learning_rate": 9.629186602870814e-06,
2249
+ "loss": 0.6341,
2250
+ "step": 1600
2251
+ },
2252
+ {
2253
+ "epoch": 0.9121909633418585,
2254
+ "grad_norm": 0.2956949472427368,
2255
+ "learning_rate": 9.330143540669856e-06,
2256
+ "loss": 0.6311,
2257
+ "step": 1605
2258
+ },
2259
+ {
2260
+ "epoch": 0.9150326797385621,
2261
+ "grad_norm": 0.3035804331302643,
2262
+ "learning_rate": 9.0311004784689e-06,
2263
+ "loss": 0.621,
2264
+ "step": 1610
2265
+ },
2266
+ {
2267
+ "epoch": 0.9178743961352657,
2268
+ "grad_norm": 0.29813966155052185,
2269
+ "learning_rate": 8.732057416267943e-06,
2270
+ "loss": 0.5775,
2271
+ "step": 1615
2272
+ },
2273
+ {
2274
+ "epoch": 0.9207161125319693,
2275
+ "grad_norm": 0.31557461619377136,
2276
+ "learning_rate": 8.433014354066986e-06,
2277
+ "loss": 0.6112,
2278
+ "step": 1620
2279
+ },
2280
+ {
2281
+ "epoch": 0.9235578289286729,
2282
+ "grad_norm": 0.2879723012447357,
2283
+ "learning_rate": 8.133971291866028e-06,
2284
+ "loss": 0.5746,
2285
+ "step": 1625
2286
+ },
2287
+ {
2288
+ "epoch": 0.9263995453253765,
2289
+ "grad_norm": 0.38764870166778564,
2290
+ "learning_rate": 7.834928229665072e-06,
2291
+ "loss": 0.6327,
2292
+ "step": 1630
2293
+ },
2294
+ {
2295
+ "epoch": 0.9292412617220801,
2296
+ "grad_norm": 0.3189394772052765,
2297
+ "learning_rate": 7.535885167464116e-06,
2298
+ "loss": 0.6361,
2299
+ "step": 1635
2300
+ },
2301
+ {
2302
+ "epoch": 0.9320829781187837,
2303
+ "grad_norm": 0.3588159382343292,
2304
+ "learning_rate": 7.236842105263158e-06,
2305
+ "loss": 0.6067,
2306
+ "step": 1640
2307
+ },
2308
+ {
2309
+ "epoch": 0.9349246945154873,
2310
+ "grad_norm": 0.35794174671173096,
2311
+ "learning_rate": 6.937799043062202e-06,
2312
+ "loss": 0.616,
2313
+ "step": 1645
2314
+ },
2315
+ {
2316
+ "epoch": 0.9377664109121909,
2317
+ "grad_norm": 0.2946082651615143,
2318
+ "learning_rate": 6.638755980861244e-06,
2319
+ "loss": 0.6271,
2320
+ "step": 1650
2321
+ },
2322
+ {
2323
+ "epoch": 0.9406081273088945,
2324
+ "grad_norm": 0.3339734375476837,
2325
+ "learning_rate": 6.339712918660287e-06,
2326
+ "loss": 0.6149,
2327
+ "step": 1655
2328
+ },
2329
+ {
2330
+ "epoch": 0.9434498437055981,
2331
+ "grad_norm": 0.3145514726638794,
2332
+ "learning_rate": 6.040669856459331e-06,
2333
+ "loss": 0.6274,
2334
+ "step": 1660
2335
+ },
2336
+ {
2337
+ "epoch": 0.9462915601023018,
2338
+ "grad_norm": 0.3304480016231537,
2339
+ "learning_rate": 5.741626794258374e-06,
2340
+ "loss": 0.6003,
2341
+ "step": 1665
2342
+ },
2343
+ {
2344
+ "epoch": 0.9491332764990054,
2345
+ "grad_norm": 0.37586352229118347,
2346
+ "learning_rate": 5.442583732057416e-06,
2347
+ "loss": 0.618,
2348
+ "step": 1670
2349
+ },
2350
+ {
2351
+ "epoch": 0.951974992895709,
2352
+ "grad_norm": 0.3044584095478058,
2353
+ "learning_rate": 5.14354066985646e-06,
2354
+ "loss": 0.6088,
2355
+ "step": 1675
2356
+ },
2357
+ {
2358
+ "epoch": 0.9548167092924126,
2359
+ "grad_norm": 0.3436095416545868,
2360
+ "learning_rate": 4.844497607655503e-06,
2361
+ "loss": 0.6156,
2362
+ "step": 1680
2363
+ },
2364
+ {
2365
+ "epoch": 0.9576584256891162,
2366
+ "grad_norm": 0.31217479705810547,
2367
+ "learning_rate": 4.5454545454545455e-06,
2368
+ "loss": 0.6412,
2369
+ "step": 1685
2370
+ },
2371
+ {
2372
+ "epoch": 0.9605001420858198,
2373
+ "grad_norm": 0.294657826423645,
2374
+ "learning_rate": 4.246411483253589e-06,
2375
+ "loss": 0.618,
2376
+ "step": 1690
2377
+ },
2378
+ {
2379
+ "epoch": 0.9633418584825234,
2380
+ "grad_norm": 0.4142850637435913,
2381
+ "learning_rate": 3.9473684210526315e-06,
2382
+ "loss": 0.5872,
2383
+ "step": 1695
2384
+ },
2385
+ {
2386
+ "epoch": 0.966183574879227,
2387
+ "grad_norm": 0.26373490691185,
2388
+ "learning_rate": 3.6483253588516753e-06,
2389
+ "loss": 0.585,
2390
+ "step": 1700
2391
+ },
2392
+ {
2393
+ "epoch": 0.9690252912759306,
2394
+ "grad_norm": 0.3273305594921112,
2395
+ "learning_rate": 3.349282296650718e-06,
2396
+ "loss": 0.6098,
2397
+ "step": 1705
2398
+ },
2399
+ {
2400
+ "epoch": 0.9718670076726342,
2401
+ "grad_norm": 0.3144895136356354,
2402
+ "learning_rate": 3.0502392344497608e-06,
2403
+ "loss": 0.6211,
2404
+ "step": 1710
2405
+ },
2406
+ {
2407
+ "epoch": 0.9747087240693378,
2408
+ "grad_norm": 0.3139897882938385,
2409
+ "learning_rate": 2.751196172248804e-06,
2410
+ "loss": 0.6049,
2411
+ "step": 1715
2412
+ },
2413
+ {
2414
+ "epoch": 0.9775504404660414,
2415
+ "grad_norm": 0.31565290689468384,
2416
+ "learning_rate": 2.452153110047847e-06,
2417
+ "loss": 0.609,
2418
+ "step": 1720
2419
+ },
2420
+ {
2421
+ "epoch": 0.9803921568627451,
2422
+ "grad_norm": 0.29738175868988037,
2423
+ "learning_rate": 2.15311004784689e-06,
2424
+ "loss": 0.5981,
2425
+ "step": 1725
2426
+ },
2427
+ {
2428
+ "epoch": 0.9832338732594487,
2429
+ "grad_norm": 0.2919932007789612,
2430
+ "learning_rate": 1.854066985645933e-06,
2431
+ "loss": 0.6005,
2432
+ "step": 1730
2433
+ },
2434
+ {
2435
+ "epoch": 0.9860755896561523,
2436
+ "grad_norm": 0.3323780298233032,
2437
+ "learning_rate": 1.555023923444976e-06,
2438
+ "loss": 0.6207,
2439
+ "step": 1735
2440
+ },
2441
+ {
2442
+ "epoch": 0.9889173060528559,
2443
+ "grad_norm": 0.3152049779891968,
2444
+ "learning_rate": 1.2559808612440192e-06,
2445
+ "loss": 0.6115,
2446
+ "step": 1740
2447
+ },
2448
+ {
2449
+ "epoch": 0.9917590224495595,
2450
+ "grad_norm": 0.3862825036048889,
2451
+ "learning_rate": 9.569377990430622e-07,
2452
+ "loss": 0.6031,
2453
+ "step": 1745
2454
+ },
2455
+ {
2456
+ "epoch": 0.9946007388462631,
2457
+ "grad_norm": 0.2568069100379944,
2458
+ "learning_rate": 6.578947368421053e-07,
2459
+ "loss": 0.6091,
2460
+ "step": 1750
2461
+ },
2462
+ {
2463
+ "epoch": 0.9974424552429667,
2464
+ "grad_norm": 0.3439529836177826,
2465
+ "learning_rate": 3.5885167464114835e-07,
2466
+ "loss": 0.6163,
2467
+ "step": 1755
2468
+ },
2469
+ {
2470
+ "epoch": 1.0,
2471
+ "grad_norm": 0.3953455984592438,
2472
+ "learning_rate": 5.980861244019139e-08,
2473
+ "loss": 0.6129,
2474
+ "step": 1760
2475
+ }
2476
+ ],
2477
+ "logging_steps": 5,
2478
+ "max_steps": 1760,
2479
+ "num_input_tokens_seen": 0,
2480
+ "num_train_epochs": 1,
2481
+ "save_steps": 500,
2482
+ "stateful_callbacks": {
2483
+ "TrainerControl": {
2484
+ "args": {
2485
+ "should_epoch_stop": false,
2486
+ "should_evaluate": false,
2487
+ "should_log": false,
2488
+ "should_save": true,
2489
+ "should_training_stop": true
2490
+ },
2491
+ "attributes": {}
2492
+ }
2493
+ },
2494
+ "total_flos": 2.910811156402248e+19,
2495
+ "train_batch_size": 1,
2496
+ "trial_name": null,
2497
+ "trial_params": null
2498
+ }
Qwen3-4B-Instruct-2507-sft-fusang-swa=2k_sink=0_falayer=13_from1_step4_fadec/checkpoint-921/chat_template.jinja ADDED
@@ -0,0 +1,61 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {%- if tools %}
2
+ {{- '<|im_start|>system\n' }}
3
+ {%- if messages[0].role == 'system' %}
4
+ {{- messages[0].content + '\n\n' }}
5
+ {%- endif %}
6
+ {{- "# Tools\n\nYou may call one or more functions to assist with the user query.\n\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>" }}
7
+ {%- for tool in tools %}
8
+ {{- "\n" }}
9
+ {{- tool | tojson }}
10
+ {%- endfor %}
11
+ {{- "\n</tools>\n\nFor each function call, return a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call><|im_end|>\n" }}
12
+ {%- else %}
13
+ {%- if messages[0].role == 'system' %}
14
+ {{- '<|im_start|>system\n' + messages[0].content + '<|im_end|>\n' }}
15
+ {%- endif %}
16
+ {%- endif %}
17
+ {%- for message in messages %}
18
+ {%- if message.content is string %}
19
+ {%- set content = message.content %}
20
+ {%- else %}
21
+ {%- set content = '' %}
22
+ {%- endif %}
23
+ {%- if (message.role == "user") or (message.role == "system" and not loop.first) %}
24
+ {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
25
+ {%- elif message.role == "assistant" %}
26
+ {{- '<|im_start|>' + message.role + '\n' + content }}
27
+ {%- if message.tool_calls %}
28
+ {%- for tool_call in message.tool_calls %}
29
+ {%- if (loop.first and content) or (not loop.first) %}
30
+ {{- '\n' }}
31
+ {%- endif %}
32
+ {%- if tool_call.function %}
33
+ {%- set tool_call = tool_call.function %}
34
+ {%- endif %}
35
+ {{- '<tool_call>\n{"name": "' }}
36
+ {{- tool_call.name }}
37
+ {{- '", "arguments": ' }}
38
+ {%- if tool_call.arguments is string %}
39
+ {{- tool_call.arguments }}
40
+ {%- else %}
41
+ {{- tool_call.arguments | tojson }}
42
+ {%- endif %}
43
+ {{- '}\n</tool_call>' }}
44
+ {%- endfor %}
45
+ {%- endif %}
46
+ {{- '<|im_end|>\n' }}
47
+ {%- elif message.role == "tool" %}
48
+ {%- if loop.first or (messages[loop.index0 - 1].role != "tool") %}
49
+ {{- '<|im_start|>user' }}
50
+ {%- endif %}
51
+ {{- '\n<tool_response>\n' }}
52
+ {{- content }}
53
+ {{- '\n</tool_response>' }}
54
+ {%- if loop.last or (messages[loop.index0 + 1].role != "tool") %}
55
+ {{- '<|im_end|>\n' }}
56
+ {%- endif %}
57
+ {%- endif %}
58
+ {%- endfor %}
59
+ {%- if add_generation_prompt %}
60
+ {{- '<|im_start|>assistant\n' }}
61
+ {%- endif %}
Qwen3-4B-Instruct-2507-sft-fusang-swa=2k_sink=0_falayer=13_from1_step4_fadec/checkpoint-921/trainer_state.json ADDED
@@ -0,0 +1,1322 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "best_global_step": null,
3
+ "best_metric": null,
4
+ "best_model_checkpoint": null,
5
+ "epoch": 1.0,
6
+ "eval_steps": 500,
7
+ "global_step": 921,
8
+ "is_hyper_param_search": false,
9
+ "is_local_process_zero": true,
10
+ "is_world_process_zero": true,
11
+ "log_history": [
12
+ {
13
+ "epoch": 0.005432568246638598,
14
+ "grad_norm": 1.7039437294006348,
15
+ "learning_rate": 8.510638297872341e-06,
16
+ "loss": 0.8861,
17
+ "step": 5
18
+ },
19
+ {
20
+ "epoch": 0.010865136493277196,
21
+ "grad_norm": 1.4117487668991089,
22
+ "learning_rate": 1.9148936170212766e-05,
23
+ "loss": 0.8896,
24
+ "step": 10
25
+ },
26
+ {
27
+ "epoch": 0.016297704739915795,
28
+ "grad_norm": 1.2073347568511963,
29
+ "learning_rate": 2.9787234042553192e-05,
30
+ "loss": 0.7764,
31
+ "step": 15
32
+ },
33
+ {
34
+ "epoch": 0.021730272986554392,
35
+ "grad_norm": 1.1810095310211182,
36
+ "learning_rate": 4.0425531914893614e-05,
37
+ "loss": 0.7327,
38
+ "step": 20
39
+ },
40
+ {
41
+ "epoch": 0.027162841233192993,
42
+ "grad_norm": 0.9019668698310852,
43
+ "learning_rate": 5.1063829787234044e-05,
44
+ "loss": 0.6974,
45
+ "step": 25
46
+ },
47
+ {
48
+ "epoch": 0.03259540947983159,
49
+ "grad_norm": 1.2329213619232178,
50
+ "learning_rate": 6.170212765957447e-05,
51
+ "loss": 0.7264,
52
+ "step": 30
53
+ },
54
+ {
55
+ "epoch": 0.03802797772647019,
56
+ "grad_norm": 0.42397692799568176,
57
+ "learning_rate": 7.23404255319149e-05,
58
+ "loss": 0.6847,
59
+ "step": 35
60
+ },
61
+ {
62
+ "epoch": 0.043460545973108784,
63
+ "grad_norm": 0.5869530439376831,
64
+ "learning_rate": 8.297872340425533e-05,
65
+ "loss": 0.6424,
66
+ "step": 40
67
+ },
68
+ {
69
+ "epoch": 0.04889311421974739,
70
+ "grad_norm": 0.5472288727760315,
71
+ "learning_rate": 9.361702127659576e-05,
72
+ "loss": 0.6424,
73
+ "step": 45
74
+ },
75
+ {
76
+ "epoch": 0.054325682466385986,
77
+ "grad_norm": 0.4212421774864197,
78
+ "learning_rate": 9.977116704805493e-05,
79
+ "loss": 0.6369,
80
+ "step": 50
81
+ },
82
+ {
83
+ "epoch": 0.05975825071302458,
84
+ "grad_norm": 0.4459717869758606,
85
+ "learning_rate": 9.919908466819223e-05,
86
+ "loss": 0.6401,
87
+ "step": 55
88
+ },
89
+ {
90
+ "epoch": 0.06519081895966318,
91
+ "grad_norm": 0.3814961910247803,
92
+ "learning_rate": 9.862700228832952e-05,
93
+ "loss": 0.6697,
94
+ "step": 60
95
+ },
96
+ {
97
+ "epoch": 0.07062338720630178,
98
+ "grad_norm": 0.4457958936691284,
99
+ "learning_rate": 9.805491990846683e-05,
100
+ "loss": 0.6293,
101
+ "step": 65
102
+ },
103
+ {
104
+ "epoch": 0.07605595545294037,
105
+ "grad_norm": 0.7550793886184692,
106
+ "learning_rate": 9.748283752860413e-05,
107
+ "loss": 0.6309,
108
+ "step": 70
109
+ },
110
+ {
111
+ "epoch": 0.08148852369957897,
112
+ "grad_norm": 0.5747292041778564,
113
+ "learning_rate": 9.691075514874142e-05,
114
+ "loss": 0.6402,
115
+ "step": 75
116
+ },
117
+ {
118
+ "epoch": 0.08692109194621757,
119
+ "grad_norm": 0.4388846158981323,
120
+ "learning_rate": 9.633867276887872e-05,
121
+ "loss": 0.6616,
122
+ "step": 80
123
+ },
124
+ {
125
+ "epoch": 0.09235366019285618,
126
+ "grad_norm": 0.39210548996925354,
127
+ "learning_rate": 9.576659038901601e-05,
128
+ "loss": 0.6387,
129
+ "step": 85
130
+ },
131
+ {
132
+ "epoch": 0.09778622843949478,
133
+ "grad_norm": 0.45312970876693726,
134
+ "learning_rate": 9.519450800915332e-05,
135
+ "loss": 0.6249,
136
+ "step": 90
137
+ },
138
+ {
139
+ "epoch": 0.10321879668613337,
140
+ "grad_norm": 0.7709851264953613,
141
+ "learning_rate": 9.462242562929062e-05,
142
+ "loss": 0.6041,
143
+ "step": 95
144
+ },
145
+ {
146
+ "epoch": 0.10865136493277197,
147
+ "grad_norm": 0.7823439240455627,
148
+ "learning_rate": 9.405034324942791e-05,
149
+ "loss": 0.6106,
150
+ "step": 100
151
+ },
152
+ {
153
+ "epoch": 0.11408393317941057,
154
+ "grad_norm": 2.5020205974578857,
155
+ "learning_rate": 9.347826086956522e-05,
156
+ "loss": 0.6419,
157
+ "step": 105
158
+ },
159
+ {
160
+ "epoch": 0.11951650142604917,
161
+ "grad_norm": 0.6196193099021912,
162
+ "learning_rate": 9.290617848970252e-05,
163
+ "loss": 0.651,
164
+ "step": 110
165
+ },
166
+ {
167
+ "epoch": 0.12494906967268776,
168
+ "grad_norm": 0.4978445768356323,
169
+ "learning_rate": 9.233409610983981e-05,
170
+ "loss": 0.6171,
171
+ "step": 115
172
+ },
173
+ {
174
+ "epoch": 0.13038163791932636,
175
+ "grad_norm": 0.43307456374168396,
176
+ "learning_rate": 9.176201372997712e-05,
177
+ "loss": 0.585,
178
+ "step": 120
179
+ },
180
+ {
181
+ "epoch": 0.13581420616596496,
182
+ "grad_norm": 0.5007528066635132,
183
+ "learning_rate": 9.118993135011442e-05,
184
+ "loss": 0.6041,
185
+ "step": 125
186
+ },
187
+ {
188
+ "epoch": 0.14124677441260355,
189
+ "grad_norm": 0.3539551794528961,
190
+ "learning_rate": 9.061784897025171e-05,
191
+ "loss": 0.6102,
192
+ "step": 130
193
+ },
194
+ {
195
+ "epoch": 0.14667934265924215,
196
+ "grad_norm": 0.7436317801475525,
197
+ "learning_rate": 9.004576659038902e-05,
198
+ "loss": 0.6072,
199
+ "step": 135
200
+ },
201
+ {
202
+ "epoch": 0.15211191090588075,
203
+ "grad_norm": 0.6404476165771484,
204
+ "learning_rate": 8.947368421052632e-05,
205
+ "loss": 0.6112,
206
+ "step": 140
207
+ },
208
+ {
209
+ "epoch": 0.15754447915251935,
210
+ "grad_norm": 0.43317389488220215,
211
+ "learning_rate": 8.890160183066361e-05,
212
+ "loss": 0.6155,
213
+ "step": 145
214
+ },
215
+ {
216
+ "epoch": 0.16297704739915794,
217
+ "grad_norm": 0.4068547785282135,
218
+ "learning_rate": 8.832951945080092e-05,
219
+ "loss": 0.6015,
220
+ "step": 150
221
+ },
222
+ {
223
+ "epoch": 0.16840961564579654,
224
+ "grad_norm": 0.45400139689445496,
225
+ "learning_rate": 8.775743707093822e-05,
226
+ "loss": 0.6214,
227
+ "step": 155
228
+ },
229
+ {
230
+ "epoch": 0.17384218389243514,
231
+ "grad_norm": 0.4221716523170471,
232
+ "learning_rate": 8.718535469107551e-05,
233
+ "loss": 0.6307,
234
+ "step": 160
235
+ },
236
+ {
237
+ "epoch": 0.17927475213907373,
238
+ "grad_norm": 0.9637123942375183,
239
+ "learning_rate": 8.661327231121282e-05,
240
+ "loss": 0.6259,
241
+ "step": 165
242
+ },
243
+ {
244
+ "epoch": 0.18470732038571236,
245
+ "grad_norm": 0.4006732702255249,
246
+ "learning_rate": 8.604118993135012e-05,
247
+ "loss": 0.6105,
248
+ "step": 170
249
+ },
250
+ {
251
+ "epoch": 0.19013988863235096,
252
+ "grad_norm": 0.4035172462463379,
253
+ "learning_rate": 8.546910755148741e-05,
254
+ "loss": 0.5995,
255
+ "step": 175
256
+ },
257
+ {
258
+ "epoch": 0.19557245687898955,
259
+ "grad_norm": 0.371188223361969,
260
+ "learning_rate": 8.489702517162472e-05,
261
+ "loss": 0.6363,
262
+ "step": 180
263
+ },
264
+ {
265
+ "epoch": 0.20100502512562815,
266
+ "grad_norm": 0.3752414286136627,
267
+ "learning_rate": 8.432494279176202e-05,
268
+ "loss": 0.6216,
269
+ "step": 185
270
+ },
271
+ {
272
+ "epoch": 0.20643759337226675,
273
+ "grad_norm": 0.39942365884780884,
274
+ "learning_rate": 8.375286041189931e-05,
275
+ "loss": 0.6178,
276
+ "step": 190
277
+ },
278
+ {
279
+ "epoch": 0.21187016161890534,
280
+ "grad_norm": 0.3931615948677063,
281
+ "learning_rate": 8.318077803203662e-05,
282
+ "loss": 0.625,
283
+ "step": 195
284
+ },
285
+ {
286
+ "epoch": 0.21730272986554394,
287
+ "grad_norm": 0.32698187232017517,
288
+ "learning_rate": 8.260869565217392e-05,
289
+ "loss": 0.6019,
290
+ "step": 200
291
+ },
292
+ {
293
+ "epoch": 0.22273529811218254,
294
+ "grad_norm": 0.3790152966976166,
295
+ "learning_rate": 8.203661327231121e-05,
296
+ "loss": 0.6547,
297
+ "step": 205
298
+ },
299
+ {
300
+ "epoch": 0.22816786635882114,
301
+ "grad_norm": 0.3495067358016968,
302
+ "learning_rate": 8.146453089244852e-05,
303
+ "loss": 0.6217,
304
+ "step": 210
305
+ },
306
+ {
307
+ "epoch": 0.23360043460545973,
308
+ "grad_norm": 0.3497507572174072,
309
+ "learning_rate": 8.089244851258582e-05,
310
+ "loss": 0.6107,
311
+ "step": 215
312
+ },
313
+ {
314
+ "epoch": 0.23903300285209833,
315
+ "grad_norm": 0.43749403953552246,
316
+ "learning_rate": 8.032036613272312e-05,
317
+ "loss": 0.5888,
318
+ "step": 220
319
+ },
320
+ {
321
+ "epoch": 0.24446557109873693,
322
+ "grad_norm": 0.31131184101104736,
323
+ "learning_rate": 7.974828375286042e-05,
324
+ "loss": 0.618,
325
+ "step": 225
326
+ },
327
+ {
328
+ "epoch": 0.24989813934537553,
329
+ "grad_norm": 0.3960503041744232,
330
+ "learning_rate": 7.917620137299772e-05,
331
+ "loss": 0.6137,
332
+ "step": 230
333
+ },
334
+ {
335
+ "epoch": 0.25533070759201415,
336
+ "grad_norm": 0.35304874181747437,
337
+ "learning_rate": 7.860411899313502e-05,
338
+ "loss": 0.6336,
339
+ "step": 235
340
+ },
341
+ {
342
+ "epoch": 0.2607632758386527,
343
+ "grad_norm": 0.3615507483482361,
344
+ "learning_rate": 7.803203661327232e-05,
345
+ "loss": 0.6117,
346
+ "step": 240
347
+ },
348
+ {
349
+ "epoch": 0.26619584408529134,
350
+ "grad_norm": 0.41142377257347107,
351
+ "learning_rate": 7.745995423340962e-05,
352
+ "loss": 0.6091,
353
+ "step": 245
354
+ },
355
+ {
356
+ "epoch": 0.2716284123319299,
357
+ "grad_norm": 0.5120617747306824,
358
+ "learning_rate": 7.688787185354692e-05,
359
+ "loss": 0.5812,
360
+ "step": 250
361
+ },
362
+ {
363
+ "epoch": 0.27706098057856854,
364
+ "grad_norm": 0.3991614580154419,
365
+ "learning_rate": 7.631578947368422e-05,
366
+ "loss": 0.6193,
367
+ "step": 255
368
+ },
369
+ {
370
+ "epoch": 0.2824935488252071,
371
+ "grad_norm": 0.41631630063056946,
372
+ "learning_rate": 7.574370709382152e-05,
373
+ "loss": 0.6385,
374
+ "step": 260
375
+ },
376
+ {
377
+ "epoch": 0.28792611707184573,
378
+ "grad_norm": 0.360675185918808,
379
+ "learning_rate": 7.517162471395882e-05,
380
+ "loss": 0.5948,
381
+ "step": 265
382
+ },
383
+ {
384
+ "epoch": 0.2933586853184843,
385
+ "grad_norm": 0.4160260558128357,
386
+ "learning_rate": 7.459954233409611e-05,
387
+ "loss": 0.6029,
388
+ "step": 270
389
+ },
390
+ {
391
+ "epoch": 0.2987912535651229,
392
+ "grad_norm": 0.3664986789226532,
393
+ "learning_rate": 7.402745995423342e-05,
394
+ "loss": 0.5897,
395
+ "step": 275
396
+ },
397
+ {
398
+ "epoch": 0.3042238218117615,
399
+ "grad_norm": 0.3521420955657959,
400
+ "learning_rate": 7.345537757437072e-05,
401
+ "loss": 0.5827,
402
+ "step": 280
403
+ },
404
+ {
405
+ "epoch": 0.3096563900584001,
406
+ "grad_norm": 0.3810531198978424,
407
+ "learning_rate": 7.288329519450801e-05,
408
+ "loss": 0.6017,
409
+ "step": 285
410
+ },
411
+ {
412
+ "epoch": 0.3150889583050387,
413
+ "grad_norm": 0.38437747955322266,
414
+ "learning_rate": 7.231121281464531e-05,
415
+ "loss": 0.6003,
416
+ "step": 290
417
+ },
418
+ {
419
+ "epoch": 0.3205215265516773,
420
+ "grad_norm": 0.424335777759552,
421
+ "learning_rate": 7.17391304347826e-05,
422
+ "loss": 0.5958,
423
+ "step": 295
424
+ },
425
+ {
426
+ "epoch": 0.3259540947983159,
427
+ "grad_norm": 0.3401501774787903,
428
+ "learning_rate": 7.116704805491991e-05,
429
+ "loss": 0.6095,
430
+ "step": 300
431
+ },
432
+ {
433
+ "epoch": 0.3313866630449545,
434
+ "grad_norm": 0.5081342458724976,
435
+ "learning_rate": 7.059496567505721e-05,
436
+ "loss": 0.5926,
437
+ "step": 305
438
+ },
439
+ {
440
+ "epoch": 0.3368192312915931,
441
+ "grad_norm": 0.41382476687431335,
442
+ "learning_rate": 7.00228832951945e-05,
443
+ "loss": 0.6187,
444
+ "step": 310
445
+ },
446
+ {
447
+ "epoch": 0.3422517995382317,
448
+ "grad_norm": 0.3560533821582794,
449
+ "learning_rate": 6.945080091533181e-05,
450
+ "loss": 0.6137,
451
+ "step": 315
452
+ },
453
+ {
454
+ "epoch": 0.3476843677848703,
455
+ "grad_norm": 0.41233116388320923,
456
+ "learning_rate": 6.887871853546911e-05,
457
+ "loss": 0.6147,
458
+ "step": 320
459
+ },
460
+ {
461
+ "epoch": 0.3531169360315089,
462
+ "grad_norm": 0.29912424087524414,
463
+ "learning_rate": 6.83066361556064e-05,
464
+ "loss": 0.5829,
465
+ "step": 325
466
+ },
467
+ {
468
+ "epoch": 0.35854950427814747,
469
+ "grad_norm": 0.4335904121398926,
470
+ "learning_rate": 6.773455377574371e-05,
471
+ "loss": 0.6105,
472
+ "step": 330
473
+ },
474
+ {
475
+ "epoch": 0.3639820725247861,
476
+ "grad_norm": 0.3471033275127411,
477
+ "learning_rate": 6.716247139588101e-05,
478
+ "loss": 0.5964,
479
+ "step": 335
480
+ },
481
+ {
482
+ "epoch": 0.3694146407714247,
483
+ "grad_norm": 0.3710848093032837,
484
+ "learning_rate": 6.65903890160183e-05,
485
+ "loss": 0.6033,
486
+ "step": 340
487
+ },
488
+ {
489
+ "epoch": 0.3748472090180633,
490
+ "grad_norm": 0.37514394521713257,
491
+ "learning_rate": 6.601830663615561e-05,
492
+ "loss": 0.6221,
493
+ "step": 345
494
+ },
495
+ {
496
+ "epoch": 0.3802797772647019,
497
+ "grad_norm": 0.399243026971817,
498
+ "learning_rate": 6.544622425629291e-05,
499
+ "loss": 0.6057,
500
+ "step": 350
501
+ },
502
+ {
503
+ "epoch": 0.3857123455113405,
504
+ "grad_norm": 0.4189896583557129,
505
+ "learning_rate": 6.48741418764302e-05,
506
+ "loss": 0.623,
507
+ "step": 355
508
+ },
509
+ {
510
+ "epoch": 0.3911449137579791,
511
+ "grad_norm": 0.3610260784626007,
512
+ "learning_rate": 6.430205949656751e-05,
513
+ "loss": 0.6178,
514
+ "step": 360
515
+ },
516
+ {
517
+ "epoch": 0.3965774820046177,
518
+ "grad_norm": 0.3497450351715088,
519
+ "learning_rate": 6.372997711670481e-05,
520
+ "loss": 0.6144,
521
+ "step": 365
522
+ },
523
+ {
524
+ "epoch": 0.4020100502512563,
525
+ "grad_norm": 0.3805111050605774,
526
+ "learning_rate": 6.31578947368421e-05,
527
+ "loss": 0.5682,
528
+ "step": 370
529
+ },
530
+ {
531
+ "epoch": 0.40744261849789487,
532
+ "grad_norm": 0.2750215530395508,
533
+ "learning_rate": 6.258581235697941e-05,
534
+ "loss": 0.5865,
535
+ "step": 375
536
+ },
537
+ {
538
+ "epoch": 0.4128751867445335,
539
+ "grad_norm": 0.2928883731365204,
540
+ "learning_rate": 6.201372997711671e-05,
541
+ "loss": 0.5992,
542
+ "step": 380
543
+ },
544
+ {
545
+ "epoch": 0.41830775499117206,
546
+ "grad_norm": 0.36385443806648254,
547
+ "learning_rate": 6.1441647597254e-05,
548
+ "loss": 0.6091,
549
+ "step": 385
550
+ },
551
+ {
552
+ "epoch": 0.4237403232378107,
553
+ "grad_norm": 0.3043301999568939,
554
+ "learning_rate": 6.086956521739131e-05,
555
+ "loss": 0.5792,
556
+ "step": 390
557
+ },
558
+ {
559
+ "epoch": 0.42917289148444926,
560
+ "grad_norm": 0.37966006994247437,
561
+ "learning_rate": 6.029748283752861e-05,
562
+ "loss": 0.5846,
563
+ "step": 395
564
+ },
565
+ {
566
+ "epoch": 0.4346054597310879,
567
+ "grad_norm": 0.36384427547454834,
568
+ "learning_rate": 5.9725400457665904e-05,
569
+ "loss": 0.5741,
570
+ "step": 400
571
+ },
572
+ {
573
+ "epoch": 0.44003802797772645,
574
+ "grad_norm": 0.3147348463535309,
575
+ "learning_rate": 5.915331807780321e-05,
576
+ "loss": 0.5956,
577
+ "step": 405
578
+ },
579
+ {
580
+ "epoch": 0.4454705962243651,
581
+ "grad_norm": 0.35524097084999084,
582
+ "learning_rate": 5.858123569794051e-05,
583
+ "loss": 0.6223,
584
+ "step": 410
585
+ },
586
+ {
587
+ "epoch": 0.45090316447100365,
588
+ "grad_norm": 0.4252597391605377,
589
+ "learning_rate": 5.8009153318077804e-05,
590
+ "loss": 0.5999,
591
+ "step": 415
592
+ },
593
+ {
594
+ "epoch": 0.4563357327176423,
595
+ "grad_norm": 0.34438687562942505,
596
+ "learning_rate": 5.743707093821511e-05,
597
+ "loss": 0.5938,
598
+ "step": 420
599
+ },
600
+ {
601
+ "epoch": 0.46176830096428084,
602
+ "grad_norm": 0.3218768835067749,
603
+ "learning_rate": 5.686498855835241e-05,
604
+ "loss": 0.5838,
605
+ "step": 425
606
+ },
607
+ {
608
+ "epoch": 0.46720086921091947,
609
+ "grad_norm": 0.40103399753570557,
610
+ "learning_rate": 5.6292906178489704e-05,
611
+ "loss": 0.6078,
612
+ "step": 430
613
+ },
614
+ {
615
+ "epoch": 0.47263343745755804,
616
+ "grad_norm": 0.3864949345588684,
617
+ "learning_rate": 5.5720823798627006e-05,
618
+ "loss": 0.6006,
619
+ "step": 435
620
+ },
621
+ {
622
+ "epoch": 0.47806600570419666,
623
+ "grad_norm": 0.42683881521224976,
624
+ "learning_rate": 5.51487414187643e-05,
625
+ "loss": 0.6136,
626
+ "step": 440
627
+ },
628
+ {
629
+ "epoch": 0.48349857395083523,
630
+ "grad_norm": 0.3744264245033264,
631
+ "learning_rate": 5.4576659038901604e-05,
632
+ "loss": 0.6041,
633
+ "step": 445
634
+ },
635
+ {
636
+ "epoch": 0.48893114219747386,
637
+ "grad_norm": 0.31652387976646423,
638
+ "learning_rate": 5.4004576659038907e-05,
639
+ "loss": 0.6014,
640
+ "step": 450
641
+ },
642
+ {
643
+ "epoch": 0.4943637104441125,
644
+ "grad_norm": 0.3432956337928772,
645
+ "learning_rate": 5.34324942791762e-05,
646
+ "loss": 0.583,
647
+ "step": 455
648
+ },
649
+ {
650
+ "epoch": 0.49979627869075105,
651
+ "grad_norm": 0.3638538718223572,
652
+ "learning_rate": 5.28604118993135e-05,
653
+ "loss": 0.6276,
654
+ "step": 460
655
+ },
656
+ {
657
+ "epoch": 0.5052288469373897,
658
+ "grad_norm": 0.3989753723144531,
659
+ "learning_rate": 5.228832951945081e-05,
660
+ "loss": 0.6157,
661
+ "step": 465
662
+ },
663
+ {
664
+ "epoch": 0.5106614151840283,
665
+ "grad_norm": 0.3906775116920471,
666
+ "learning_rate": 5.17162471395881e-05,
667
+ "loss": 0.582,
668
+ "step": 470
669
+ },
670
+ {
671
+ "epoch": 0.5160939834306668,
672
+ "grad_norm": 0.5225631594657898,
673
+ "learning_rate": 5.11441647597254e-05,
674
+ "loss": 0.6109,
675
+ "step": 475
676
+ },
677
+ {
678
+ "epoch": 0.5215265516773054,
679
+ "grad_norm": 0.31983035802841187,
680
+ "learning_rate": 5.057208237986271e-05,
681
+ "loss": 0.612,
682
+ "step": 480
683
+ },
684
+ {
685
+ "epoch": 0.5269591199239441,
686
+ "grad_norm": 0.3480057418346405,
687
+ "learning_rate": 5e-05,
688
+ "loss": 0.6142,
689
+ "step": 485
690
+ },
691
+ {
692
+ "epoch": 0.5323916881705827,
693
+ "grad_norm": 0.3848414421081543,
694
+ "learning_rate": 4.9427917620137305e-05,
695
+ "loss": 0.59,
696
+ "step": 490
697
+ },
698
+ {
699
+ "epoch": 0.5378242564172212,
700
+ "grad_norm": 0.4043906331062317,
701
+ "learning_rate": 4.88558352402746e-05,
702
+ "loss": 0.5971,
703
+ "step": 495
704
+ },
705
+ {
706
+ "epoch": 0.5432568246638598,
707
+ "grad_norm": 0.4232148230075836,
708
+ "learning_rate": 4.82837528604119e-05,
709
+ "loss": 0.5949,
710
+ "step": 500
711
+ },
712
+ {
713
+ "epoch": 0.5486893929104985,
714
+ "grad_norm": 0.3397533893585205,
715
+ "learning_rate": 4.7711670480549205e-05,
716
+ "loss": 0.6173,
717
+ "step": 505
718
+ },
719
+ {
720
+ "epoch": 0.5541219611571371,
721
+ "grad_norm": 0.35291028022766113,
722
+ "learning_rate": 4.71395881006865e-05,
723
+ "loss": 0.6301,
724
+ "step": 510
725
+ },
726
+ {
727
+ "epoch": 0.5595545294037756,
728
+ "grad_norm": 0.36393406987190247,
729
+ "learning_rate": 4.65675057208238e-05,
730
+ "loss": 0.6102,
731
+ "step": 515
732
+ },
733
+ {
734
+ "epoch": 0.5649870976504142,
735
+ "grad_norm": 0.3933073878288269,
736
+ "learning_rate": 4.5995423340961105e-05,
737
+ "loss": 0.5995,
738
+ "step": 520
739
+ },
740
+ {
741
+ "epoch": 0.5704196658970528,
742
+ "grad_norm": 0.32478901743888855,
743
+ "learning_rate": 4.54233409610984e-05,
744
+ "loss": 0.5576,
745
+ "step": 525
746
+ },
747
+ {
748
+ "epoch": 0.5758522341436915,
749
+ "grad_norm": 0.3688804507255554,
750
+ "learning_rate": 4.48512585812357e-05,
751
+ "loss": 0.607,
752
+ "step": 530
753
+ },
754
+ {
755
+ "epoch": 0.58128480239033,
756
+ "grad_norm": 1.4745358228683472,
757
+ "learning_rate": 4.4279176201373e-05,
758
+ "loss": 0.6067,
759
+ "step": 535
760
+ },
761
+ {
762
+ "epoch": 0.5867173706369686,
763
+ "grad_norm": 0.3768254518508911,
764
+ "learning_rate": 4.37070938215103e-05,
765
+ "loss": 0.5795,
766
+ "step": 540
767
+ },
768
+ {
769
+ "epoch": 0.5921499388836072,
770
+ "grad_norm": 0.38579657673835754,
771
+ "learning_rate": 4.3135011441647596e-05,
772
+ "loss": 0.6103,
773
+ "step": 545
774
+ },
775
+ {
776
+ "epoch": 0.5975825071302459,
777
+ "grad_norm": 0.2949613928794861,
778
+ "learning_rate": 4.25629290617849e-05,
779
+ "loss": 0.5896,
780
+ "step": 550
781
+ },
782
+ {
783
+ "epoch": 0.6030150753768844,
784
+ "grad_norm": 0.35038861632347107,
785
+ "learning_rate": 4.1990846681922194e-05,
786
+ "loss": 0.6099,
787
+ "step": 555
788
+ },
789
+ {
790
+ "epoch": 0.608447643623523,
791
+ "grad_norm": 0.3110353350639343,
792
+ "learning_rate": 4.1418764302059497e-05,
793
+ "loss": 0.606,
794
+ "step": 560
795
+ },
796
+ {
797
+ "epoch": 0.6138802118701616,
798
+ "grad_norm": 0.3179367780685425,
799
+ "learning_rate": 4.08466819221968e-05,
800
+ "loss": 0.6015,
801
+ "step": 565
802
+ },
803
+ {
804
+ "epoch": 0.6193127801168002,
805
+ "grad_norm": 0.35172903537750244,
806
+ "learning_rate": 4.0274599542334094e-05,
807
+ "loss": 0.5965,
808
+ "step": 570
809
+ },
810
+ {
811
+ "epoch": 0.6247453483634389,
812
+ "grad_norm": 0.32751893997192383,
813
+ "learning_rate": 3.97025171624714e-05,
814
+ "loss": 0.5748,
815
+ "step": 575
816
+ },
817
+ {
818
+ "epoch": 0.6301779166100774,
819
+ "grad_norm": 0.3121321201324463,
820
+ "learning_rate": 3.91304347826087e-05,
821
+ "loss": 0.6224,
822
+ "step": 580
823
+ },
824
+ {
825
+ "epoch": 0.635610484856716,
826
+ "grad_norm": 0.35525357723236084,
827
+ "learning_rate": 3.8558352402745995e-05,
828
+ "loss": 0.6179,
829
+ "step": 585
830
+ },
831
+ {
832
+ "epoch": 0.6410430531033546,
833
+ "grad_norm": 0.3244050443172455,
834
+ "learning_rate": 3.79862700228833e-05,
835
+ "loss": 0.6022,
836
+ "step": 590
837
+ },
838
+ {
839
+ "epoch": 0.6464756213499933,
840
+ "grad_norm": 0.44588297605514526,
841
+ "learning_rate": 3.74141876430206e-05,
842
+ "loss": 0.5926,
843
+ "step": 595
844
+ },
845
+ {
846
+ "epoch": 0.6519081895966318,
847
+ "grad_norm": 0.32992833852767944,
848
+ "learning_rate": 3.6842105263157895e-05,
849
+ "loss": 0.6045,
850
+ "step": 600
851
+ },
852
+ {
853
+ "epoch": 0.6573407578432704,
854
+ "grad_norm": 0.33284494280815125,
855
+ "learning_rate": 3.62700228832952e-05,
856
+ "loss": 0.5876,
857
+ "step": 605
858
+ },
859
+ {
860
+ "epoch": 0.662773326089909,
861
+ "grad_norm": 0.32135191559791565,
862
+ "learning_rate": 3.56979405034325e-05,
863
+ "loss": 0.6331,
864
+ "step": 610
865
+ },
866
+ {
867
+ "epoch": 0.6682058943365476,
868
+ "grad_norm": 0.32440677285194397,
869
+ "learning_rate": 3.5125858123569795e-05,
870
+ "loss": 0.5543,
871
+ "step": 615
872
+ },
873
+ {
874
+ "epoch": 0.6736384625831862,
875
+ "grad_norm": 0.39705970883369446,
876
+ "learning_rate": 3.45537757437071e-05,
877
+ "loss": 0.5828,
878
+ "step": 620
879
+ },
880
+ {
881
+ "epoch": 0.6790710308298248,
882
+ "grad_norm": 0.32800349593162537,
883
+ "learning_rate": 3.39816933638444e-05,
884
+ "loss": 0.601,
885
+ "step": 625
886
+ },
887
+ {
888
+ "epoch": 0.6845035990764634,
889
+ "grad_norm": 0.3102910816669464,
890
+ "learning_rate": 3.3409610983981695e-05,
891
+ "loss": 0.6137,
892
+ "step": 630
893
+ },
894
+ {
895
+ "epoch": 0.689936167323102,
896
+ "grad_norm": 0.31854674220085144,
897
+ "learning_rate": 3.2837528604119e-05,
898
+ "loss": 0.5946,
899
+ "step": 635
900
+ },
901
+ {
902
+ "epoch": 0.6953687355697405,
903
+ "grad_norm": 0.308635950088501,
904
+ "learning_rate": 3.226544622425629e-05,
905
+ "loss": 0.573,
906
+ "step": 640
907
+ },
908
+ {
909
+ "epoch": 0.7008013038163792,
910
+ "grad_norm": 0.35990995168685913,
911
+ "learning_rate": 3.1693363844393595e-05,
912
+ "loss": 0.5966,
913
+ "step": 645
914
+ },
915
+ {
916
+ "epoch": 0.7062338720630178,
917
+ "grad_norm": 0.3423658609390259,
918
+ "learning_rate": 3.112128146453089e-05,
919
+ "loss": 0.5892,
920
+ "step": 650
921
+ },
922
+ {
923
+ "epoch": 0.7116664403096564,
924
+ "grad_norm": 0.34332504868507385,
925
+ "learning_rate": 3.054919908466819e-05,
926
+ "loss": 0.6058,
927
+ "step": 655
928
+ },
929
+ {
930
+ "epoch": 0.7170990085562949,
931
+ "grad_norm": 0.3571360111236572,
932
+ "learning_rate": 2.9977116704805492e-05,
933
+ "loss": 0.6073,
934
+ "step": 660
935
+ },
936
+ {
937
+ "epoch": 0.7225315768029336,
938
+ "grad_norm": 0.39828506112098694,
939
+ "learning_rate": 2.9405034324942794e-05,
940
+ "loss": 0.5884,
941
+ "step": 665
942
+ },
943
+ {
944
+ "epoch": 0.7279641450495722,
945
+ "grad_norm": 0.3208368718624115,
946
+ "learning_rate": 2.8832951945080093e-05,
947
+ "loss": 0.5956,
948
+ "step": 670
949
+ },
950
+ {
951
+ "epoch": 0.7333967132962108,
952
+ "grad_norm": 0.3208870589733124,
953
+ "learning_rate": 2.826086956521739e-05,
954
+ "loss": 0.5812,
955
+ "step": 675
956
+ },
957
+ {
958
+ "epoch": 0.7388292815428494,
959
+ "grad_norm": 0.36011451482772827,
960
+ "learning_rate": 2.768878718535469e-05,
961
+ "loss": 0.6119,
962
+ "step": 680
963
+ },
964
+ {
965
+ "epoch": 0.744261849789488,
966
+ "grad_norm": 0.36624807119369507,
967
+ "learning_rate": 2.7116704805491993e-05,
968
+ "loss": 0.5736,
969
+ "step": 685
970
+ },
971
+ {
972
+ "epoch": 0.7496944180361266,
973
+ "grad_norm": 0.3616745173931122,
974
+ "learning_rate": 2.654462242562929e-05,
975
+ "loss": 0.6185,
976
+ "step": 690
977
+ },
978
+ {
979
+ "epoch": 0.7551269862827652,
980
+ "grad_norm": 0.35389941930770874,
981
+ "learning_rate": 2.597254004576659e-05,
982
+ "loss": 0.6022,
983
+ "step": 695
984
+ },
985
+ {
986
+ "epoch": 0.7605595545294038,
987
+ "grad_norm": 0.30083775520324707,
988
+ "learning_rate": 2.5400457665903894e-05,
989
+ "loss": 0.6087,
990
+ "step": 700
991
+ },
992
+ {
993
+ "epoch": 0.7659921227760423,
994
+ "grad_norm": 0.33950841426849365,
995
+ "learning_rate": 2.4828375286041193e-05,
996
+ "loss": 0.6049,
997
+ "step": 705
998
+ },
999
+ {
1000
+ "epoch": 0.771424691022681,
1001
+ "grad_norm": 0.4462699890136719,
1002
+ "learning_rate": 2.425629290617849e-05,
1003
+ "loss": 0.6115,
1004
+ "step": 710
1005
+ },
1006
+ {
1007
+ "epoch": 0.7768572592693196,
1008
+ "grad_norm": 0.3414113223552704,
1009
+ "learning_rate": 2.368421052631579e-05,
1010
+ "loss": 0.6354,
1011
+ "step": 715
1012
+ },
1013
+ {
1014
+ "epoch": 0.7822898275159582,
1015
+ "grad_norm": 0.3125087022781372,
1016
+ "learning_rate": 2.311212814645309e-05,
1017
+ "loss": 0.6071,
1018
+ "step": 720
1019
+ },
1020
+ {
1021
+ "epoch": 0.7877223957625967,
1022
+ "grad_norm": 0.38993561267852783,
1023
+ "learning_rate": 2.2540045766590388e-05,
1024
+ "loss": 0.5923,
1025
+ "step": 725
1026
+ },
1027
+ {
1028
+ "epoch": 0.7931549640092354,
1029
+ "grad_norm": 0.48246586322784424,
1030
+ "learning_rate": 2.1967963386727687e-05,
1031
+ "loss": 0.6048,
1032
+ "step": 730
1033
+ },
1034
+ {
1035
+ "epoch": 0.798587532255874,
1036
+ "grad_norm": 0.37252160906791687,
1037
+ "learning_rate": 2.139588100686499e-05,
1038
+ "loss": 0.5763,
1039
+ "step": 735
1040
+ },
1041
+ {
1042
+ "epoch": 0.8040201005025126,
1043
+ "grad_norm": 0.3334311246871948,
1044
+ "learning_rate": 2.082379862700229e-05,
1045
+ "loss": 0.568,
1046
+ "step": 740
1047
+ },
1048
+ {
1049
+ "epoch": 0.8094526687491511,
1050
+ "grad_norm": 0.6050743460655212,
1051
+ "learning_rate": 2.0251716247139587e-05,
1052
+ "loss": 0.5975,
1053
+ "step": 745
1054
+ },
1055
+ {
1056
+ "epoch": 0.8148852369957897,
1057
+ "grad_norm": 0.2849281132221222,
1058
+ "learning_rate": 1.967963386727689e-05,
1059
+ "loss": 0.5901,
1060
+ "step": 750
1061
+ },
1062
+ {
1063
+ "epoch": 0.8203178052424284,
1064
+ "grad_norm": 0.3431701064109802,
1065
+ "learning_rate": 1.910755148741419e-05,
1066
+ "loss": 0.6199,
1067
+ "step": 755
1068
+ },
1069
+ {
1070
+ "epoch": 0.825750373489067,
1071
+ "grad_norm": 0.4718061089515686,
1072
+ "learning_rate": 1.853546910755149e-05,
1073
+ "loss": 0.599,
1074
+ "step": 760
1075
+ },
1076
+ {
1077
+ "epoch": 0.8311829417357055,
1078
+ "grad_norm": 0.34578534960746765,
1079
+ "learning_rate": 1.796338672768879e-05,
1080
+ "loss": 0.599,
1081
+ "step": 765
1082
+ },
1083
+ {
1084
+ "epoch": 0.8366155099823441,
1085
+ "grad_norm": 0.59501051902771,
1086
+ "learning_rate": 1.739130434782609e-05,
1087
+ "loss": 0.5702,
1088
+ "step": 770
1089
+ },
1090
+ {
1091
+ "epoch": 0.8420480782289828,
1092
+ "grad_norm": 0.3314354121685028,
1093
+ "learning_rate": 1.6819221967963388e-05,
1094
+ "loss": 0.614,
1095
+ "step": 775
1096
+ },
1097
+ {
1098
+ "epoch": 0.8474806464756214,
1099
+ "grad_norm": 0.3568246066570282,
1100
+ "learning_rate": 1.6247139588100687e-05,
1101
+ "loss": 0.628,
1102
+ "step": 780
1103
+ },
1104
+ {
1105
+ "epoch": 0.8529132147222599,
1106
+ "grad_norm": 0.3536950945854187,
1107
+ "learning_rate": 1.5675057208237986e-05,
1108
+ "loss": 0.6322,
1109
+ "step": 785
1110
+ },
1111
+ {
1112
+ "epoch": 0.8583457829688985,
1113
+ "grad_norm": 0.3197844624519348,
1114
+ "learning_rate": 1.5102974828375288e-05,
1115
+ "loss": 0.5998,
1116
+ "step": 790
1117
+ },
1118
+ {
1119
+ "epoch": 0.8637783512155371,
1120
+ "grad_norm": 0.3613077402114868,
1121
+ "learning_rate": 1.4530892448512587e-05,
1122
+ "loss": 0.5944,
1123
+ "step": 795
1124
+ },
1125
+ {
1126
+ "epoch": 0.8692109194621758,
1127
+ "grad_norm": 0.3265378773212433,
1128
+ "learning_rate": 1.3958810068649886e-05,
1129
+ "loss": 0.6016,
1130
+ "step": 800
1131
+ },
1132
+ {
1133
+ "epoch": 0.8746434877088144,
1134
+ "grad_norm": 0.34476912021636963,
1135
+ "learning_rate": 1.3386727688787188e-05,
1136
+ "loss": 0.6227,
1137
+ "step": 805
1138
+ },
1139
+ {
1140
+ "epoch": 0.8800760559554529,
1141
+ "grad_norm": 0.3053128123283386,
1142
+ "learning_rate": 1.2814645308924485e-05,
1143
+ "loss": 0.6063,
1144
+ "step": 810
1145
+ },
1146
+ {
1147
+ "epoch": 0.8855086242020915,
1148
+ "grad_norm": 0.3270540237426758,
1149
+ "learning_rate": 1.2242562929061784e-05,
1150
+ "loss": 0.5835,
1151
+ "step": 815
1152
+ },
1153
+ {
1154
+ "epoch": 0.8909411924487302,
1155
+ "grad_norm": 0.3964345455169678,
1156
+ "learning_rate": 1.1670480549199085e-05,
1157
+ "loss": 0.5934,
1158
+ "step": 820
1159
+ },
1160
+ {
1161
+ "epoch": 0.8963737606953688,
1162
+ "grad_norm": 0.33152106404304504,
1163
+ "learning_rate": 1.1098398169336386e-05,
1164
+ "loss": 0.5935,
1165
+ "step": 825
1166
+ },
1167
+ {
1168
+ "epoch": 0.9018063289420073,
1169
+ "grad_norm": 0.34804388880729675,
1170
+ "learning_rate": 1.0526315789473684e-05,
1171
+ "loss": 0.6137,
1172
+ "step": 830
1173
+ },
1174
+ {
1175
+ "epoch": 0.9072388971886459,
1176
+ "grad_norm": 0.3584604561328888,
1177
+ "learning_rate": 9.954233409610985e-06,
1178
+ "loss": 0.6345,
1179
+ "step": 835
1180
+ },
1181
+ {
1182
+ "epoch": 0.9126714654352845,
1183
+ "grad_norm": 0.3093973994255066,
1184
+ "learning_rate": 9.382151029748284e-06,
1185
+ "loss": 0.6095,
1186
+ "step": 840
1187
+ },
1188
+ {
1189
+ "epoch": 0.9181040336819232,
1190
+ "grad_norm": 0.3735460937023163,
1191
+ "learning_rate": 8.810068649885583e-06,
1192
+ "loss": 0.589,
1193
+ "step": 845
1194
+ },
1195
+ {
1196
+ "epoch": 0.9235366019285617,
1197
+ "grad_norm": 0.27046552300453186,
1198
+ "learning_rate": 8.237986270022884e-06,
1199
+ "loss": 0.58,
1200
+ "step": 850
1201
+ },
1202
+ {
1203
+ "epoch": 0.9289691701752003,
1204
+ "grad_norm": 0.3246326148509979,
1205
+ "learning_rate": 7.665903890160184e-06,
1206
+ "loss": 0.6157,
1207
+ "step": 855
1208
+ },
1209
+ {
1210
+ "epoch": 0.9344017384218389,
1211
+ "grad_norm": 0.41096463799476624,
1212
+ "learning_rate": 7.093821510297482e-06,
1213
+ "loss": 0.6302,
1214
+ "step": 860
1215
+ },
1216
+ {
1217
+ "epoch": 0.9398343066684776,
1218
+ "grad_norm": 0.35395240783691406,
1219
+ "learning_rate": 6.521739130434783e-06,
1220
+ "loss": 0.6155,
1221
+ "step": 865
1222
+ },
1223
+ {
1224
+ "epoch": 0.9452668749151161,
1225
+ "grad_norm": 0.3799790143966675,
1226
+ "learning_rate": 5.949656750572083e-06,
1227
+ "loss": 0.5809,
1228
+ "step": 870
1229
+ },
1230
+ {
1231
+ "epoch": 0.9506994431617547,
1232
+ "grad_norm": 0.3575972318649292,
1233
+ "learning_rate": 5.3775743707093824e-06,
1234
+ "loss": 0.5756,
1235
+ "step": 875
1236
+ },
1237
+ {
1238
+ "epoch": 0.9561320114083933,
1239
+ "grad_norm": 0.41418325901031494,
1240
+ "learning_rate": 4.805491990846682e-06,
1241
+ "loss": 0.628,
1242
+ "step": 880
1243
+ },
1244
+ {
1245
+ "epoch": 0.961564579655032,
1246
+ "grad_norm": 0.3544921576976776,
1247
+ "learning_rate": 4.233409610983982e-06,
1248
+ "loss": 0.6101,
1249
+ "step": 885
1250
+ },
1251
+ {
1252
+ "epoch": 0.9669971479016705,
1253
+ "grad_norm": 0.33101755380630493,
1254
+ "learning_rate": 3.6613272311212818e-06,
1255
+ "loss": 0.5933,
1256
+ "step": 890
1257
+ },
1258
+ {
1259
+ "epoch": 0.9724297161483091,
1260
+ "grad_norm": 0.4163910746574402,
1261
+ "learning_rate": 3.0892448512585815e-06,
1262
+ "loss": 0.5729,
1263
+ "step": 895
1264
+ },
1265
+ {
1266
+ "epoch": 0.9778622843949477,
1267
+ "grad_norm": 0.31177565455436707,
1268
+ "learning_rate": 2.517162471395881e-06,
1269
+ "loss": 0.5886,
1270
+ "step": 900
1271
+ },
1272
+ {
1273
+ "epoch": 0.9832948526415863,
1274
+ "grad_norm": 0.2993391454219818,
1275
+ "learning_rate": 1.9450800915331807e-06,
1276
+ "loss": 0.5701,
1277
+ "step": 905
1278
+ },
1279
+ {
1280
+ "epoch": 0.988727420888225,
1281
+ "grad_norm": 0.4177623987197876,
1282
+ "learning_rate": 1.3729977116704805e-06,
1283
+ "loss": 0.6144,
1284
+ "step": 910
1285
+ },
1286
+ {
1287
+ "epoch": 0.9941599891348635,
1288
+ "grad_norm": 0.366140753030777,
1289
+ "learning_rate": 8.009153318077803e-07,
1290
+ "loss": 0.6029,
1291
+ "step": 915
1292
+ },
1293
+ {
1294
+ "epoch": 0.9995925573815021,
1295
+ "grad_norm": 0.3682883083820343,
1296
+ "learning_rate": 2.288329519450801e-07,
1297
+ "loss": 0.57,
1298
+ "step": 920
1299
+ }
1300
+ ],
1301
+ "logging_steps": 5,
1302
+ "max_steps": 921,
1303
+ "num_input_tokens_seen": 0,
1304
+ "num_train_epochs": 1,
1305
+ "save_steps": 500,
1306
+ "stateful_callbacks": {
1307
+ "TrainerControl": {
1308
+ "args": {
1309
+ "should_epoch_stop": false,
1310
+ "should_evaluate": false,
1311
+ "should_log": false,
1312
+ "should_save": true,
1313
+ "should_training_stop": true
1314
+ },
1315
+ "attributes": {}
1316
+ }
1317
+ },
1318
+ "total_flos": 1.5239517919298716e+19,
1319
+ "train_batch_size": 1,
1320
+ "trial_name": null,
1321
+ "trial_params": null
1322
+ }
Qwen3-4B-Instruct-2507-sft-fusang-swa=2k_sink=0_falayer=25_from1_step2/checkpoint-921/tokenizer.json ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:aeb13307a71acd8fe81861d94ad54ab689df773318809eed3cbe794b4492dae4
3
+ size 11422654
Qwen3-4B-Instruct-2507-sft-fusang-swa=2k_sink=0_falayer=25_from1_step2/checkpoint-921/trainer_state.json ADDED
@@ -0,0 +1,1322 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "best_global_step": null,
3
+ "best_metric": null,
4
+ "best_model_checkpoint": null,
5
+ "epoch": 1.0,
6
+ "eval_steps": 500,
7
+ "global_step": 921,
8
+ "is_hyper_param_search": false,
9
+ "is_local_process_zero": true,
10
+ "is_world_process_zero": true,
11
+ "log_history": [
12
+ {
13
+ "epoch": 0.005432568246638598,
14
+ "grad_norm": 9.459916114807129,
15
+ "learning_rate": 8.510638297872341e-06,
16
+ "loss": 1.4178,
17
+ "step": 5
18
+ },
19
+ {
20
+ "epoch": 0.010865136493277196,
21
+ "grad_norm": 7.1960601806640625,
22
+ "learning_rate": 1.9148936170212766e-05,
23
+ "loss": 1.422,
24
+ "step": 10
25
+ },
26
+ {
27
+ "epoch": 0.016297704739915795,
28
+ "grad_norm": 6.983621597290039,
29
+ "learning_rate": 2.9787234042553192e-05,
30
+ "loss": 1.2892,
31
+ "step": 15
32
+ },
33
+ {
34
+ "epoch": 0.021730272986554392,
35
+ "grad_norm": 13.326706886291504,
36
+ "learning_rate": 4.0425531914893614e-05,
37
+ "loss": 1.1788,
38
+ "step": 20
39
+ },
40
+ {
41
+ "epoch": 0.027162841233192993,
42
+ "grad_norm": 8.636503219604492,
43
+ "learning_rate": 5.1063829787234044e-05,
44
+ "loss": 1.105,
45
+ "step": 25
46
+ },
47
+ {
48
+ "epoch": 0.03259540947983159,
49
+ "grad_norm": 4.728584289550781,
50
+ "learning_rate": 6.170212765957447e-05,
51
+ "loss": 1.0797,
52
+ "step": 30
53
+ },
54
+ {
55
+ "epoch": 0.03802797772647019,
56
+ "grad_norm": 1.2720285654067993,
57
+ "learning_rate": 7.23404255319149e-05,
58
+ "loss": 0.9521,
59
+ "step": 35
60
+ },
61
+ {
62
+ "epoch": 0.043460545973108784,
63
+ "grad_norm": 2.187081813812256,
64
+ "learning_rate": 8.297872340425533e-05,
65
+ "loss": 0.8359,
66
+ "step": 40
67
+ },
68
+ {
69
+ "epoch": 0.04889311421974739,
70
+ "grad_norm": 6.980685234069824,
71
+ "learning_rate": 9.361702127659576e-05,
72
+ "loss": 0.8112,
73
+ "step": 45
74
+ },
75
+ {
76
+ "epoch": 0.054325682466385986,
77
+ "grad_norm": 4.144394397735596,
78
+ "learning_rate": 9.977116704805493e-05,
79
+ "loss": 0.777,
80
+ "step": 50
81
+ },
82
+ {
83
+ "epoch": 0.05975825071302458,
84
+ "grad_norm": 1.1660783290863037,
85
+ "learning_rate": 9.919908466819223e-05,
86
+ "loss": 0.776,
87
+ "step": 55
88
+ },
89
+ {
90
+ "epoch": 0.06519081895966318,
91
+ "grad_norm": 0.829829216003418,
92
+ "learning_rate": 9.862700228832952e-05,
93
+ "loss": 0.7937,
94
+ "step": 60
95
+ },
96
+ {
97
+ "epoch": 0.07062338720630178,
98
+ "grad_norm": 1.0270509719848633,
99
+ "learning_rate": 9.805491990846683e-05,
100
+ "loss": 0.7475,
101
+ "step": 65
102
+ },
103
+ {
104
+ "epoch": 0.07605595545294037,
105
+ "grad_norm": 0.9886857271194458,
106
+ "learning_rate": 9.748283752860413e-05,
107
+ "loss": 0.738,
108
+ "step": 70
109
+ },
110
+ {
111
+ "epoch": 0.08148852369957897,
112
+ "grad_norm": 1.7533588409423828,
113
+ "learning_rate": 9.691075514874142e-05,
114
+ "loss": 0.7351,
115
+ "step": 75
116
+ },
117
+ {
118
+ "epoch": 0.08692109194621757,
119
+ "grad_norm": 1.6888121366500854,
120
+ "learning_rate": 9.633867276887872e-05,
121
+ "loss": 0.7625,
122
+ "step": 80
123
+ },
124
+ {
125
+ "epoch": 0.09235366019285618,
126
+ "grad_norm": 1.1737279891967773,
127
+ "learning_rate": 9.576659038901601e-05,
128
+ "loss": 0.7291,
129
+ "step": 85
130
+ },
131
+ {
132
+ "epoch": 0.09778622843949478,
133
+ "grad_norm": 0.9594925045967102,
134
+ "learning_rate": 9.519450800915332e-05,
135
+ "loss": 0.7158,
136
+ "step": 90
137
+ },
138
+ {
139
+ "epoch": 0.10321879668613337,
140
+ "grad_norm": 0.8971859812736511,
141
+ "learning_rate": 9.462242562929062e-05,
142
+ "loss": 0.6835,
143
+ "step": 95
144
+ },
145
+ {
146
+ "epoch": 0.10865136493277197,
147
+ "grad_norm": 1.7678531408309937,
148
+ "learning_rate": 9.405034324942791e-05,
149
+ "loss": 0.6875,
150
+ "step": 100
151
+ },
152
+ {
153
+ "epoch": 0.11408393317941057,
154
+ "grad_norm": 2.2030041217803955,
155
+ "learning_rate": 9.347826086956522e-05,
156
+ "loss": 0.7258,
157
+ "step": 105
158
+ },
159
+ {
160
+ "epoch": 0.11951650142604917,
161
+ "grad_norm": 0.62217116355896,
162
+ "learning_rate": 9.290617848970252e-05,
163
+ "loss": 0.7267,
164
+ "step": 110
165
+ },
166
+ {
167
+ "epoch": 0.12494906967268776,
168
+ "grad_norm": 3.0713260173797607,
169
+ "learning_rate": 9.233409610983981e-05,
170
+ "loss": 0.693,
171
+ "step": 115
172
+ },
173
+ {
174
+ "epoch": 0.13038163791932636,
175
+ "grad_norm": 0.6462140083312988,
176
+ "learning_rate": 9.176201372997712e-05,
177
+ "loss": 0.6576,
178
+ "step": 120
179
+ },
180
+ {
181
+ "epoch": 0.13581420616596496,
182
+ "grad_norm": 1.0117729902267456,
183
+ "learning_rate": 9.118993135011442e-05,
184
+ "loss": 0.676,
185
+ "step": 125
186
+ },
187
+ {
188
+ "epoch": 0.14124677441260355,
189
+ "grad_norm": 0.6420778632164001,
190
+ "learning_rate": 9.061784897025171e-05,
191
+ "loss": 0.6904,
192
+ "step": 130
193
+ },
194
+ {
195
+ "epoch": 0.14667934265924215,
196
+ "grad_norm": 0.6774643063545227,
197
+ "learning_rate": 9.004576659038902e-05,
198
+ "loss": 0.6811,
199
+ "step": 135
200
+ },
201
+ {
202
+ "epoch": 0.15211191090588075,
203
+ "grad_norm": 0.6019052863121033,
204
+ "learning_rate": 8.947368421052632e-05,
205
+ "loss": 0.6791,
206
+ "step": 140
207
+ },
208
+ {
209
+ "epoch": 0.15754447915251935,
210
+ "grad_norm": 1.2432228326797485,
211
+ "learning_rate": 8.890160183066361e-05,
212
+ "loss": 0.6892,
213
+ "step": 145
214
+ },
215
+ {
216
+ "epoch": 0.16297704739915794,
217
+ "grad_norm": 1.4279394149780273,
218
+ "learning_rate": 8.832951945080092e-05,
219
+ "loss": 0.669,
220
+ "step": 150
221
+ },
222
+ {
223
+ "epoch": 0.16840961564579654,
224
+ "grad_norm": 0.6732116341590881,
225
+ "learning_rate": 8.775743707093822e-05,
226
+ "loss": 0.6941,
227
+ "step": 155
228
+ },
229
+ {
230
+ "epoch": 0.17384218389243514,
231
+ "grad_norm": 0.6370598077774048,
232
+ "learning_rate": 8.718535469107551e-05,
233
+ "loss": 0.7015,
234
+ "step": 160
235
+ },
236
+ {
237
+ "epoch": 0.17927475213907373,
238
+ "grad_norm": 0.6765103340148926,
239
+ "learning_rate": 8.661327231121282e-05,
240
+ "loss": 0.6914,
241
+ "step": 165
242
+ },
243
+ {
244
+ "epoch": 0.18470732038571236,
245
+ "grad_norm": 0.6248242259025574,
246
+ "learning_rate": 8.604118993135012e-05,
247
+ "loss": 0.6767,
248
+ "step": 170
249
+ },
250
+ {
251
+ "epoch": 0.19013988863235096,
252
+ "grad_norm": 1.0242446660995483,
253
+ "learning_rate": 8.546910755148741e-05,
254
+ "loss": 0.6619,
255
+ "step": 175
256
+ },
257
+ {
258
+ "epoch": 0.19557245687898955,
259
+ "grad_norm": 0.9339828491210938,
260
+ "learning_rate": 8.489702517162472e-05,
261
+ "loss": 0.7031,
262
+ "step": 180
263
+ },
264
+ {
265
+ "epoch": 0.20100502512562815,
266
+ "grad_norm": 0.7380356788635254,
267
+ "learning_rate": 8.432494279176202e-05,
268
+ "loss": 0.6834,
269
+ "step": 185
270
+ },
271
+ {
272
+ "epoch": 0.20643759337226675,
273
+ "grad_norm": 0.8154470920562744,
274
+ "learning_rate": 8.375286041189931e-05,
275
+ "loss": 0.6796,
276
+ "step": 190
277
+ },
278
+ {
279
+ "epoch": 0.21187016161890534,
280
+ "grad_norm": 1.1723755598068237,
281
+ "learning_rate": 8.318077803203662e-05,
282
+ "loss": 0.685,
283
+ "step": 195
284
+ },
285
+ {
286
+ "epoch": 0.21730272986554394,
287
+ "grad_norm": 5.695266246795654,
288
+ "learning_rate": 8.260869565217392e-05,
289
+ "loss": 0.6619,
290
+ "step": 200
291
+ },
292
+ {
293
+ "epoch": 0.22273529811218254,
294
+ "grad_norm": 0.6542934775352478,
295
+ "learning_rate": 8.203661327231121e-05,
296
+ "loss": 0.7178,
297
+ "step": 205
298
+ },
299
+ {
300
+ "epoch": 0.22816786635882114,
301
+ "grad_norm": 0.5754424929618835,
302
+ "learning_rate": 8.146453089244852e-05,
303
+ "loss": 0.6905,
304
+ "step": 210
305
+ },
306
+ {
307
+ "epoch": 0.23360043460545973,
308
+ "grad_norm": 0.6104907393455505,
309
+ "learning_rate": 8.089244851258582e-05,
310
+ "loss": 0.6731,
311
+ "step": 215
312
+ },
313
+ {
314
+ "epoch": 0.23903300285209833,
315
+ "grad_norm": 0.6275336742401123,
316
+ "learning_rate": 8.032036613272312e-05,
317
+ "loss": 0.6514,
318
+ "step": 220
319
+ },
320
+ {
321
+ "epoch": 0.24446557109873693,
322
+ "grad_norm": 0.6367227435112,
323
+ "learning_rate": 7.974828375286042e-05,
324
+ "loss": 0.675,
325
+ "step": 225
326
+ },
327
+ {
328
+ "epoch": 0.24989813934537553,
329
+ "grad_norm": 0.5682306289672852,
330
+ "learning_rate": 7.917620137299772e-05,
331
+ "loss": 0.6788,
332
+ "step": 230
333
+ },
334
+ {
335
+ "epoch": 0.25533070759201415,
336
+ "grad_norm": 0.7252913117408752,
337
+ "learning_rate": 7.860411899313502e-05,
338
+ "loss": 0.6908,
339
+ "step": 235
340
+ },
341
+ {
342
+ "epoch": 0.2607632758386527,
343
+ "grad_norm": 0.5313924551010132,
344
+ "learning_rate": 7.803203661327232e-05,
345
+ "loss": 0.6678,
346
+ "step": 240
347
+ },
348
+ {
349
+ "epoch": 0.26619584408529134,
350
+ "grad_norm": 0.5733349919319153,
351
+ "learning_rate": 7.745995423340962e-05,
352
+ "loss": 0.667,
353
+ "step": 245
354
+ },
355
+ {
356
+ "epoch": 0.2716284123319299,
357
+ "grad_norm": 0.5741682052612305,
358
+ "learning_rate": 7.688787185354692e-05,
359
+ "loss": 0.6314,
360
+ "step": 250
361
+ },
362
+ {
363
+ "epoch": 0.27706098057856854,
364
+ "grad_norm": 0.616175651550293,
365
+ "learning_rate": 7.631578947368422e-05,
366
+ "loss": 0.6832,
367
+ "step": 255
368
+ },
369
+ {
370
+ "epoch": 0.2824935488252071,
371
+ "grad_norm": 0.562748908996582,
372
+ "learning_rate": 7.574370709382152e-05,
373
+ "loss": 0.7007,
374
+ "step": 260
375
+ },
376
+ {
377
+ "epoch": 0.28792611707184573,
378
+ "grad_norm": 0.8781464695930481,
379
+ "learning_rate": 7.517162471395882e-05,
380
+ "loss": 0.6527,
381
+ "step": 265
382
+ },
383
+ {
384
+ "epoch": 0.2933586853184843,
385
+ "grad_norm": 1.790806531906128,
386
+ "learning_rate": 7.459954233409611e-05,
387
+ "loss": 0.6589,
388
+ "step": 270
389
+ },
390
+ {
391
+ "epoch": 0.2987912535651229,
392
+ "grad_norm": 0.5946965217590332,
393
+ "learning_rate": 7.402745995423342e-05,
394
+ "loss": 0.6433,
395
+ "step": 275
396
+ },
397
+ {
398
+ "epoch": 0.3042238218117615,
399
+ "grad_norm": 0.7908605337142944,
400
+ "learning_rate": 7.345537757437072e-05,
401
+ "loss": 0.6398,
402
+ "step": 280
403
+ },
404
+ {
405
+ "epoch": 0.3096563900584001,
406
+ "grad_norm": 0.5854300260543823,
407
+ "learning_rate": 7.288329519450801e-05,
408
+ "loss": 0.6637,
409
+ "step": 285
410
+ },
411
+ {
412
+ "epoch": 0.3150889583050387,
413
+ "grad_norm": 0.6470159888267517,
414
+ "learning_rate": 7.231121281464531e-05,
415
+ "loss": 0.6541,
416
+ "step": 290
417
+ },
418
+ {
419
+ "epoch": 0.3205215265516773,
420
+ "grad_norm": 0.8673564195632935,
421
+ "learning_rate": 7.17391304347826e-05,
422
+ "loss": 0.6556,
423
+ "step": 295
424
+ },
425
+ {
426
+ "epoch": 0.3259540947983159,
427
+ "grad_norm": 0.5583031177520752,
428
+ "learning_rate": 7.116704805491991e-05,
429
+ "loss": 0.6626,
430
+ "step": 300
431
+ },
432
+ {
433
+ "epoch": 0.3313866630449545,
434
+ "grad_norm": 0.5254254937171936,
435
+ "learning_rate": 7.059496567505721e-05,
436
+ "loss": 0.6521,
437
+ "step": 305
438
+ },
439
+ {
440
+ "epoch": 0.3368192312915931,
441
+ "grad_norm": 0.593638002872467,
442
+ "learning_rate": 7.00228832951945e-05,
443
+ "loss": 0.6762,
444
+ "step": 310
445
+ },
446
+ {
447
+ "epoch": 0.3422517995382317,
448
+ "grad_norm": 0.6116818189620972,
449
+ "learning_rate": 6.945080091533181e-05,
450
+ "loss": 0.6681,
451
+ "step": 315
452
+ },
453
+ {
454
+ "epoch": 0.3476843677848703,
455
+ "grad_norm": 0.732871949672699,
456
+ "learning_rate": 6.887871853546911e-05,
457
+ "loss": 0.6746,
458
+ "step": 320
459
+ },
460
+ {
461
+ "epoch": 0.3531169360315089,
462
+ "grad_norm": 0.5636444687843323,
463
+ "learning_rate": 6.83066361556064e-05,
464
+ "loss": 0.6342,
465
+ "step": 325
466
+ },
467
+ {
468
+ "epoch": 0.35854950427814747,
469
+ "grad_norm": 0.6112375855445862,
470
+ "learning_rate": 6.773455377574371e-05,
471
+ "loss": 0.6705,
472
+ "step": 330
473
+ },
474
+ {
475
+ "epoch": 0.3639820725247861,
476
+ "grad_norm": 0.7206803560256958,
477
+ "learning_rate": 6.716247139588101e-05,
478
+ "loss": 0.6513,
479
+ "step": 335
480
+ },
481
+ {
482
+ "epoch": 0.3694146407714247,
483
+ "grad_norm": 0.5461758971214294,
484
+ "learning_rate": 6.65903890160183e-05,
485
+ "loss": 0.6604,
486
+ "step": 340
487
+ },
488
+ {
489
+ "epoch": 0.3748472090180633,
490
+ "grad_norm": 0.7067888379096985,
491
+ "learning_rate": 6.601830663615561e-05,
492
+ "loss": 0.6744,
493
+ "step": 345
494
+ },
495
+ {
496
+ "epoch": 0.3802797772647019,
497
+ "grad_norm": 0.5425392389297485,
498
+ "learning_rate": 6.544622425629291e-05,
499
+ "loss": 0.6552,
500
+ "step": 350
501
+ },
502
+ {
503
+ "epoch": 0.3857123455113405,
504
+ "grad_norm": 0.5506808757781982,
505
+ "learning_rate": 6.48741418764302e-05,
506
+ "loss": 0.6812,
507
+ "step": 355
508
+ },
509
+ {
510
+ "epoch": 0.3911449137579791,
511
+ "grad_norm": 0.6555262804031372,
512
+ "learning_rate": 6.430205949656751e-05,
513
+ "loss": 0.6702,
514
+ "step": 360
515
+ },
516
+ {
517
+ "epoch": 0.3965774820046177,
518
+ "grad_norm": 0.5152429342269897,
519
+ "learning_rate": 6.372997711670481e-05,
520
+ "loss": 0.6679,
521
+ "step": 365
522
+ },
523
+ {
524
+ "epoch": 0.4020100502512563,
525
+ "grad_norm": 0.5566269159317017,
526
+ "learning_rate": 6.31578947368421e-05,
527
+ "loss": 0.6221,
528
+ "step": 370
529
+ },
530
+ {
531
+ "epoch": 0.40744261849789487,
532
+ "grad_norm": 0.5410118103027344,
533
+ "learning_rate": 6.258581235697941e-05,
534
+ "loss": 0.6407,
535
+ "step": 375
536
+ },
537
+ {
538
+ "epoch": 0.4128751867445335,
539
+ "grad_norm": 0.4301411807537079,
540
+ "learning_rate": 6.201372997711671e-05,
541
+ "loss": 0.6556,
542
+ "step": 380
543
+ },
544
+ {
545
+ "epoch": 0.41830775499117206,
546
+ "grad_norm": 0.5091935992240906,
547
+ "learning_rate": 6.1441647597254e-05,
548
+ "loss": 0.6627,
549
+ "step": 385
550
+ },
551
+ {
552
+ "epoch": 0.4237403232378107,
553
+ "grad_norm": 0.47411203384399414,
554
+ "learning_rate": 6.086956521739131e-05,
555
+ "loss": 0.6274,
556
+ "step": 390
557
+ },
558
+ {
559
+ "epoch": 0.42917289148444926,
560
+ "grad_norm": 0.5017412900924683,
561
+ "learning_rate": 6.029748283752861e-05,
562
+ "loss": 0.637,
563
+ "step": 395
564
+ },
565
+ {
566
+ "epoch": 0.4346054597310879,
567
+ "grad_norm": 0.7483174800872803,
568
+ "learning_rate": 5.9725400457665904e-05,
569
+ "loss": 0.6255,
570
+ "step": 400
571
+ },
572
+ {
573
+ "epoch": 0.44003802797772645,
574
+ "grad_norm": 0.5736692547798157,
575
+ "learning_rate": 5.915331807780321e-05,
576
+ "loss": 0.6505,
577
+ "step": 405
578
+ },
579
+ {
580
+ "epoch": 0.4454705962243651,
581
+ "grad_norm": 0.6141465306282043,
582
+ "learning_rate": 5.858123569794051e-05,
583
+ "loss": 0.6767,
584
+ "step": 410
585
+ },
586
+ {
587
+ "epoch": 0.45090316447100365,
588
+ "grad_norm": 0.8123056292533875,
589
+ "learning_rate": 5.8009153318077804e-05,
590
+ "loss": 0.6572,
591
+ "step": 415
592
+ },
593
+ {
594
+ "epoch": 0.4563357327176423,
595
+ "grad_norm": 0.4647817611694336,
596
+ "learning_rate": 5.743707093821511e-05,
597
+ "loss": 0.6437,
598
+ "step": 420
599
+ },
600
+ {
601
+ "epoch": 0.46176830096428084,
602
+ "grad_norm": 0.6208211183547974,
603
+ "learning_rate": 5.686498855835241e-05,
604
+ "loss": 0.6322,
605
+ "step": 425
606
+ },
607
+ {
608
+ "epoch": 0.46720086921091947,
609
+ "grad_norm": 0.6200281977653503,
610
+ "learning_rate": 5.6292906178489704e-05,
611
+ "loss": 0.6634,
612
+ "step": 430
613
+ },
614
+ {
615
+ "epoch": 0.47263343745755804,
616
+ "grad_norm": 0.5913947820663452,
617
+ "learning_rate": 5.5720823798627006e-05,
618
+ "loss": 0.6611,
619
+ "step": 435
620
+ },
621
+ {
622
+ "epoch": 0.47806600570419666,
623
+ "grad_norm": 0.5132611393928528,
624
+ "learning_rate": 5.51487414187643e-05,
625
+ "loss": 0.6671,
626
+ "step": 440
627
+ },
628
+ {
629
+ "epoch": 0.48349857395083523,
630
+ "grad_norm": 0.6194720268249512,
631
+ "learning_rate": 5.4576659038901604e-05,
632
+ "loss": 0.654,
633
+ "step": 445
634
+ },
635
+ {
636
+ "epoch": 0.48893114219747386,
637
+ "grad_norm": 0.47601041197776794,
638
+ "learning_rate": 5.4004576659038907e-05,
639
+ "loss": 0.648,
640
+ "step": 450
641
+ },
642
+ {
643
+ "epoch": 0.4943637104441125,
644
+ "grad_norm": 0.5232619047164917,
645
+ "learning_rate": 5.34324942791762e-05,
646
+ "loss": 0.6394,
647
+ "step": 455
648
+ },
649
+ {
650
+ "epoch": 0.49979627869075105,
651
+ "grad_norm": 0.46558257937431335,
652
+ "learning_rate": 5.28604118993135e-05,
653
+ "loss": 0.6784,
654
+ "step": 460
655
+ },
656
+ {
657
+ "epoch": 0.5052288469373897,
658
+ "grad_norm": 0.6142993569374084,
659
+ "learning_rate": 5.228832951945081e-05,
660
+ "loss": 0.6679,
661
+ "step": 465
662
+ },
663
+ {
664
+ "epoch": 0.5106614151840283,
665
+ "grad_norm": 0.5508822202682495,
666
+ "learning_rate": 5.17162471395881e-05,
667
+ "loss": 0.6283,
668
+ "step": 470
669
+ },
670
+ {
671
+ "epoch": 0.5160939834306668,
672
+ "grad_norm": 0.5909676551818848,
673
+ "learning_rate": 5.11441647597254e-05,
674
+ "loss": 0.6604,
675
+ "step": 475
676
+ },
677
+ {
678
+ "epoch": 0.5215265516773054,
679
+ "grad_norm": 0.6020964980125427,
680
+ "learning_rate": 5.057208237986271e-05,
681
+ "loss": 0.666,
682
+ "step": 480
683
+ },
684
+ {
685
+ "epoch": 0.5269591199239441,
686
+ "grad_norm": 0.5518227815628052,
687
+ "learning_rate": 5e-05,
688
+ "loss": 0.6626,
689
+ "step": 485
690
+ },
691
+ {
692
+ "epoch": 0.5323916881705827,
693
+ "grad_norm": 0.5602546334266663,
694
+ "learning_rate": 4.9427917620137305e-05,
695
+ "loss": 0.6426,
696
+ "step": 490
697
+ },
698
+ {
699
+ "epoch": 0.5378242564172212,
700
+ "grad_norm": 0.5417766571044922,
701
+ "learning_rate": 4.88558352402746e-05,
702
+ "loss": 0.6453,
703
+ "step": 495
704
+ },
705
+ {
706
+ "epoch": 0.5432568246638598,
707
+ "grad_norm": 0.5592178106307983,
708
+ "learning_rate": 4.82837528604119e-05,
709
+ "loss": 0.6417,
710
+ "step": 500
711
+ },
712
+ {
713
+ "epoch": 0.5486893929104985,
714
+ "grad_norm": 0.5138264298439026,
715
+ "learning_rate": 4.7711670480549205e-05,
716
+ "loss": 0.6732,
717
+ "step": 505
718
+ },
719
+ {
720
+ "epoch": 0.5541219611571371,
721
+ "grad_norm": 0.4949907958507538,
722
+ "learning_rate": 4.71395881006865e-05,
723
+ "loss": 0.6848,
724
+ "step": 510
725
+ },
726
+ {
727
+ "epoch": 0.5595545294037756,
728
+ "grad_norm": 0.5179665684700012,
729
+ "learning_rate": 4.65675057208238e-05,
730
+ "loss": 0.6601,
731
+ "step": 515
732
+ },
733
+ {
734
+ "epoch": 0.5649870976504142,
735
+ "grad_norm": 0.6265512108802795,
736
+ "learning_rate": 4.5995423340961105e-05,
737
+ "loss": 0.6471,
738
+ "step": 520
739
+ },
740
+ {
741
+ "epoch": 0.5704196658970528,
742
+ "grad_norm": 0.5798054337501526,
743
+ "learning_rate": 4.54233409610984e-05,
744
+ "loss": 0.6117,
745
+ "step": 525
746
+ },
747
+ {
748
+ "epoch": 0.5758522341436915,
749
+ "grad_norm": 0.4561832547187805,
750
+ "learning_rate": 4.48512585812357e-05,
751
+ "loss": 0.661,
752
+ "step": 530
753
+ },
754
+ {
755
+ "epoch": 0.58128480239033,
756
+ "grad_norm": 0.5431075692176819,
757
+ "learning_rate": 4.4279176201373e-05,
758
+ "loss": 0.657,
759
+ "step": 535
760
+ },
761
+ {
762
+ "epoch": 0.5867173706369686,
763
+ "grad_norm": 4.448174953460693,
764
+ "learning_rate": 4.37070938215103e-05,
765
+ "loss": 0.6281,
766
+ "step": 540
767
+ },
768
+ {
769
+ "epoch": 0.5921499388836072,
770
+ "grad_norm": 0.6081518530845642,
771
+ "learning_rate": 4.3135011441647596e-05,
772
+ "loss": 0.6593,
773
+ "step": 545
774
+ },
775
+ {
776
+ "epoch": 0.5975825071302459,
777
+ "grad_norm": 0.4878840744495392,
778
+ "learning_rate": 4.25629290617849e-05,
779
+ "loss": 0.6346,
780
+ "step": 550
781
+ },
782
+ {
783
+ "epoch": 0.6030150753768844,
784
+ "grad_norm": 0.516602098941803,
785
+ "learning_rate": 4.1990846681922194e-05,
786
+ "loss": 0.6672,
787
+ "step": 555
788
+ },
789
+ {
790
+ "epoch": 0.608447643623523,
791
+ "grad_norm": 0.4631401002407074,
792
+ "learning_rate": 4.1418764302059497e-05,
793
+ "loss": 0.6601,
794
+ "step": 560
795
+ },
796
+ {
797
+ "epoch": 0.6138802118701616,
798
+ "grad_norm": 0.5325141549110413,
799
+ "learning_rate": 4.08466819221968e-05,
800
+ "loss": 0.6518,
801
+ "step": 565
802
+ },
803
+ {
804
+ "epoch": 0.6193127801168002,
805
+ "grad_norm": 0.5583243370056152,
806
+ "learning_rate": 4.0274599542334094e-05,
807
+ "loss": 0.6443,
808
+ "step": 570
809
+ },
810
+ {
811
+ "epoch": 0.6247453483634389,
812
+ "grad_norm": 0.5083745718002319,
813
+ "learning_rate": 3.97025171624714e-05,
814
+ "loss": 0.6194,
815
+ "step": 575
816
+ },
817
+ {
818
+ "epoch": 0.6301779166100774,
819
+ "grad_norm": 0.47409185767173767,
820
+ "learning_rate": 3.91304347826087e-05,
821
+ "loss": 0.6769,
822
+ "step": 580
823
+ },
824
+ {
825
+ "epoch": 0.635610484856716,
826
+ "grad_norm": 0.5178033709526062,
827
+ "learning_rate": 3.8558352402745995e-05,
828
+ "loss": 0.6739,
829
+ "step": 585
830
+ },
831
+ {
832
+ "epoch": 0.6410430531033546,
833
+ "grad_norm": 0.5133267045021057,
834
+ "learning_rate": 3.79862700228833e-05,
835
+ "loss": 0.6528,
836
+ "step": 590
837
+ },
838
+ {
839
+ "epoch": 0.6464756213499933,
840
+ "grad_norm": 0.7014407515525818,
841
+ "learning_rate": 3.74141876430206e-05,
842
+ "loss": 0.6448,
843
+ "step": 595
844
+ },
845
+ {
846
+ "epoch": 0.6519081895966318,
847
+ "grad_norm": 0.4593577980995178,
848
+ "learning_rate": 3.6842105263157895e-05,
849
+ "loss": 0.6521,
850
+ "step": 600
851
+ },
852
+ {
853
+ "epoch": 0.6573407578432704,
854
+ "grad_norm": 0.4936089813709259,
855
+ "learning_rate": 3.62700228832952e-05,
856
+ "loss": 0.6395,
857
+ "step": 605
858
+ },
859
+ {
860
+ "epoch": 0.662773326089909,
861
+ "grad_norm": 0.4906123876571655,
862
+ "learning_rate": 3.56979405034325e-05,
863
+ "loss": 0.6861,
864
+ "step": 610
865
+ },
866
+ {
867
+ "epoch": 0.6682058943365476,
868
+ "grad_norm": 0.48499593138694763,
869
+ "learning_rate": 3.5125858123569795e-05,
870
+ "loss": 0.6012,
871
+ "step": 615
872
+ },
873
+ {
874
+ "epoch": 0.6736384625831862,
875
+ "grad_norm": 0.59617018699646,
876
+ "learning_rate": 3.45537757437071e-05,
877
+ "loss": 0.6319,
878
+ "step": 620
879
+ },
880
+ {
881
+ "epoch": 0.6790710308298248,
882
+ "grad_norm": 0.44910475611686707,
883
+ "learning_rate": 3.39816933638444e-05,
884
+ "loss": 0.6452,
885
+ "step": 625
886
+ },
887
+ {
888
+ "epoch": 0.6845035990764634,
889
+ "grad_norm": 0.6694605946540833,
890
+ "learning_rate": 3.3409610983981695e-05,
891
+ "loss": 0.6685,
892
+ "step": 630
893
+ },
894
+ {
895
+ "epoch": 0.689936167323102,
896
+ "grad_norm": 0.4925360381603241,
897
+ "learning_rate": 3.2837528604119e-05,
898
+ "loss": 0.6411,
899
+ "step": 635
900
+ },
901
+ {
902
+ "epoch": 0.6953687355697405,
903
+ "grad_norm": 0.5627233386039734,
904
+ "learning_rate": 3.226544622425629e-05,
905
+ "loss": 0.6199,
906
+ "step": 640
907
+ },
908
+ {
909
+ "epoch": 0.7008013038163792,
910
+ "grad_norm": 0.5734674334526062,
911
+ "learning_rate": 3.1693363844393595e-05,
912
+ "loss": 0.646,
913
+ "step": 645
914
+ },
915
+ {
916
+ "epoch": 0.7062338720630178,
917
+ "grad_norm": 0.5287236571311951,
918
+ "learning_rate": 3.112128146453089e-05,
919
+ "loss": 0.6368,
920
+ "step": 650
921
+ },
922
+ {
923
+ "epoch": 0.7116664403096564,
924
+ "grad_norm": 0.4998726546764374,
925
+ "learning_rate": 3.054919908466819e-05,
926
+ "loss": 0.6572,
927
+ "step": 655
928
+ },
929
+ {
930
+ "epoch": 0.7170990085562949,
931
+ "grad_norm": 0.5494430065155029,
932
+ "learning_rate": 2.9977116704805492e-05,
933
+ "loss": 0.6587,
934
+ "step": 660
935
+ },
936
+ {
937
+ "epoch": 0.7225315768029336,
938
+ "grad_norm": 0.765242338180542,
939
+ "learning_rate": 2.9405034324942794e-05,
940
+ "loss": 0.6385,
941
+ "step": 665
942
+ },
943
+ {
944
+ "epoch": 0.7279641450495722,
945
+ "grad_norm": 0.4946318566799164,
946
+ "learning_rate": 2.8832951945080093e-05,
947
+ "loss": 0.6433,
948
+ "step": 670
949
+ },
950
+ {
951
+ "epoch": 0.7333967132962108,
952
+ "grad_norm": 0.5145821571350098,
953
+ "learning_rate": 2.826086956521739e-05,
954
+ "loss": 0.6257,
955
+ "step": 675
956
+ },
957
+ {
958
+ "epoch": 0.7388292815428494,
959
+ "grad_norm": 0.4741166830062866,
960
+ "learning_rate": 2.768878718535469e-05,
961
+ "loss": 0.662,
962
+ "step": 680
963
+ },
964
+ {
965
+ "epoch": 0.744261849789488,
966
+ "grad_norm": 0.5121444463729858,
967
+ "learning_rate": 2.7116704805491993e-05,
968
+ "loss": 0.6218,
969
+ "step": 685
970
+ },
971
+ {
972
+ "epoch": 0.7496944180361266,
973
+ "grad_norm": 0.5793531537055969,
974
+ "learning_rate": 2.654462242562929e-05,
975
+ "loss": 0.6654,
976
+ "step": 690
977
+ },
978
+ {
979
+ "epoch": 0.7551269862827652,
980
+ "grad_norm": 0.6608290672302246,
981
+ "learning_rate": 2.597254004576659e-05,
982
+ "loss": 0.6495,
983
+ "step": 695
984
+ },
985
+ {
986
+ "epoch": 0.7605595545294038,
987
+ "grad_norm": 0.4930092990398407,
988
+ "learning_rate": 2.5400457665903894e-05,
989
+ "loss": 0.6539,
990
+ "step": 700
991
+ },
992
+ {
993
+ "epoch": 0.7659921227760423,
994
+ "grad_norm": 0.6334433555603027,
995
+ "learning_rate": 2.4828375286041193e-05,
996
+ "loss": 0.6579,
997
+ "step": 705
998
+ },
999
+ {
1000
+ "epoch": 0.771424691022681,
1001
+ "grad_norm": 0.4903906285762787,
1002
+ "learning_rate": 2.425629290617849e-05,
1003
+ "loss": 0.6638,
1004
+ "step": 710
1005
+ },
1006
+ {
1007
+ "epoch": 0.7768572592693196,
1008
+ "grad_norm": 0.553806722164154,
1009
+ "learning_rate": 2.368421052631579e-05,
1010
+ "loss": 0.6857,
1011
+ "step": 715
1012
+ },
1013
+ {
1014
+ "epoch": 0.7822898275159582,
1015
+ "grad_norm": 0.4767685830593109,
1016
+ "learning_rate": 2.311212814645309e-05,
1017
+ "loss": 0.6515,
1018
+ "step": 720
1019
+ },
1020
+ {
1021
+ "epoch": 0.7877223957625967,
1022
+ "grad_norm": 0.49609318375587463,
1023
+ "learning_rate": 2.2540045766590388e-05,
1024
+ "loss": 0.6379,
1025
+ "step": 725
1026
+ },
1027
+ {
1028
+ "epoch": 0.7931549640092354,
1029
+ "grad_norm": 0.5016102194786072,
1030
+ "learning_rate": 2.1967963386727687e-05,
1031
+ "loss": 0.6596,
1032
+ "step": 730
1033
+ },
1034
+ {
1035
+ "epoch": 0.798587532255874,
1036
+ "grad_norm": 0.5809308290481567,
1037
+ "learning_rate": 2.139588100686499e-05,
1038
+ "loss": 0.6282,
1039
+ "step": 735
1040
+ },
1041
+ {
1042
+ "epoch": 0.8040201005025126,
1043
+ "grad_norm": 0.5164983868598938,
1044
+ "learning_rate": 2.082379862700229e-05,
1045
+ "loss": 0.6151,
1046
+ "step": 740
1047
+ },
1048
+ {
1049
+ "epoch": 0.8094526687491511,
1050
+ "grad_norm": 0.5916256308555603,
1051
+ "learning_rate": 2.0251716247139587e-05,
1052
+ "loss": 0.644,
1053
+ "step": 745
1054
+ },
1055
+ {
1056
+ "epoch": 0.8148852369957897,
1057
+ "grad_norm": 0.47339901328086853,
1058
+ "learning_rate": 1.967963386727689e-05,
1059
+ "loss": 0.6373,
1060
+ "step": 750
1061
+ },
1062
+ {
1063
+ "epoch": 0.8203178052424284,
1064
+ "grad_norm": 0.630338728427887,
1065
+ "learning_rate": 1.910755148741419e-05,
1066
+ "loss": 0.6709,
1067
+ "step": 755
1068
+ },
1069
+ {
1070
+ "epoch": 0.825750373489067,
1071
+ "grad_norm": 0.5498789548873901,
1072
+ "learning_rate": 1.853546910755149e-05,
1073
+ "loss": 0.6484,
1074
+ "step": 760
1075
+ },
1076
+ {
1077
+ "epoch": 0.8311829417357055,
1078
+ "grad_norm": 0.7132230997085571,
1079
+ "learning_rate": 1.796338672768879e-05,
1080
+ "loss": 0.6471,
1081
+ "step": 765
1082
+ },
1083
+ {
1084
+ "epoch": 0.8366155099823441,
1085
+ "grad_norm": 0.6490641832351685,
1086
+ "learning_rate": 1.739130434782609e-05,
1087
+ "loss": 0.6154,
1088
+ "step": 770
1089
+ },
1090
+ {
1091
+ "epoch": 0.8420480782289828,
1092
+ "grad_norm": 0.5747671723365784,
1093
+ "learning_rate": 1.6819221967963388e-05,
1094
+ "loss": 0.6564,
1095
+ "step": 775
1096
+ },
1097
+ {
1098
+ "epoch": 0.8474806464756214,
1099
+ "grad_norm": 0.5624738931655884,
1100
+ "learning_rate": 1.6247139588100687e-05,
1101
+ "loss": 0.6756,
1102
+ "step": 780
1103
+ },
1104
+ {
1105
+ "epoch": 0.8529132147222599,
1106
+ "grad_norm": 0.5083646774291992,
1107
+ "learning_rate": 1.5675057208237986e-05,
1108
+ "loss": 0.6839,
1109
+ "step": 785
1110
+ },
1111
+ {
1112
+ "epoch": 0.8583457829688985,
1113
+ "grad_norm": 0.5137049555778503,
1114
+ "learning_rate": 1.5102974828375288e-05,
1115
+ "loss": 0.6459,
1116
+ "step": 790
1117
+ },
1118
+ {
1119
+ "epoch": 0.8637783512155371,
1120
+ "grad_norm": 0.5147213339805603,
1121
+ "learning_rate": 1.4530892448512587e-05,
1122
+ "loss": 0.6412,
1123
+ "step": 795
1124
+ },
1125
+ {
1126
+ "epoch": 0.8692109194621758,
1127
+ "grad_norm": 0.6255154609680176,
1128
+ "learning_rate": 1.3958810068649886e-05,
1129
+ "loss": 0.644,
1130
+ "step": 800
1131
+ },
1132
+ {
1133
+ "epoch": 0.8746434877088144,
1134
+ "grad_norm": 0.5027071237564087,
1135
+ "learning_rate": 1.3386727688787188e-05,
1136
+ "loss": 0.6721,
1137
+ "step": 805
1138
+ },
1139
+ {
1140
+ "epoch": 0.8800760559554529,
1141
+ "grad_norm": 0.5450616478919983,
1142
+ "learning_rate": 1.2814645308924485e-05,
1143
+ "loss": 0.6561,
1144
+ "step": 810
1145
+ },
1146
+ {
1147
+ "epoch": 0.8855086242020915,
1148
+ "grad_norm": 0.6149742603302002,
1149
+ "learning_rate": 1.2242562929061784e-05,
1150
+ "loss": 0.6276,
1151
+ "step": 815
1152
+ },
1153
+ {
1154
+ "epoch": 0.8909411924487302,
1155
+ "grad_norm": 0.6135491728782654,
1156
+ "learning_rate": 1.1670480549199085e-05,
1157
+ "loss": 0.6409,
1158
+ "step": 820
1159
+ },
1160
+ {
1161
+ "epoch": 0.8963737606953688,
1162
+ "grad_norm": 0.4514573812484741,
1163
+ "learning_rate": 1.1098398169336386e-05,
1164
+ "loss": 0.6384,
1165
+ "step": 825
1166
+ },
1167
+ {
1168
+ "epoch": 0.9018063289420073,
1169
+ "grad_norm": 0.5023235082626343,
1170
+ "learning_rate": 1.0526315789473684e-05,
1171
+ "loss": 0.6672,
1172
+ "step": 830
1173
+ },
1174
+ {
1175
+ "epoch": 0.9072388971886459,
1176
+ "grad_norm": 0.4921826422214508,
1177
+ "learning_rate": 9.954233409610985e-06,
1178
+ "loss": 0.6826,
1179
+ "step": 835
1180
+ },
1181
+ {
1182
+ "epoch": 0.9126714654352845,
1183
+ "grad_norm": 0.4660825729370117,
1184
+ "learning_rate": 9.382151029748284e-06,
1185
+ "loss": 0.6534,
1186
+ "step": 840
1187
+ },
1188
+ {
1189
+ "epoch": 0.9181040336819232,
1190
+ "grad_norm": 0.5195198655128479,
1191
+ "learning_rate": 8.810068649885583e-06,
1192
+ "loss": 0.6362,
1193
+ "step": 845
1194
+ },
1195
+ {
1196
+ "epoch": 0.9235366019285617,
1197
+ "grad_norm": 0.4689203202724457,
1198
+ "learning_rate": 8.237986270022884e-06,
1199
+ "loss": 0.6203,
1200
+ "step": 850
1201
+ },
1202
+ {
1203
+ "epoch": 0.9289691701752003,
1204
+ "grad_norm": 0.49104467034339905,
1205
+ "learning_rate": 7.665903890160184e-06,
1206
+ "loss": 0.6599,
1207
+ "step": 855
1208
+ },
1209
+ {
1210
+ "epoch": 0.9344017384218389,
1211
+ "grad_norm": 0.48561811447143555,
1212
+ "learning_rate": 7.093821510297482e-06,
1213
+ "loss": 0.6796,
1214
+ "step": 860
1215
+ },
1216
+ {
1217
+ "epoch": 0.9398343066684776,
1218
+ "grad_norm": 0.47941282391548157,
1219
+ "learning_rate": 6.521739130434783e-06,
1220
+ "loss": 0.6629,
1221
+ "step": 865
1222
+ },
1223
+ {
1224
+ "epoch": 0.9452668749151161,
1225
+ "grad_norm": 0.5367552638053894,
1226
+ "learning_rate": 5.949656750572083e-06,
1227
+ "loss": 0.6243,
1228
+ "step": 870
1229
+ },
1230
+ {
1231
+ "epoch": 0.9506994431617547,
1232
+ "grad_norm": 0.5322713851928711,
1233
+ "learning_rate": 5.3775743707093824e-06,
1234
+ "loss": 0.618,
1235
+ "step": 875
1236
+ },
1237
+ {
1238
+ "epoch": 0.9561320114083933,
1239
+ "grad_norm": 0.5174160003662109,
1240
+ "learning_rate": 4.805491990846682e-06,
1241
+ "loss": 0.6791,
1242
+ "step": 880
1243
+ },
1244
+ {
1245
+ "epoch": 0.961564579655032,
1246
+ "grad_norm": 0.5452148914337158,
1247
+ "learning_rate": 4.233409610983982e-06,
1248
+ "loss": 0.663,
1249
+ "step": 885
1250
+ },
1251
+ {
1252
+ "epoch": 0.9669971479016705,
1253
+ "grad_norm": 0.47627437114715576,
1254
+ "learning_rate": 3.6613272311212818e-06,
1255
+ "loss": 0.6412,
1256
+ "step": 890
1257
+ },
1258
+ {
1259
+ "epoch": 0.9724297161483091,
1260
+ "grad_norm": 0.5921295285224915,
1261
+ "learning_rate": 3.0892448512585815e-06,
1262
+ "loss": 0.6173,
1263
+ "step": 895
1264
+ },
1265
+ {
1266
+ "epoch": 0.9778622843949477,
1267
+ "grad_norm": 0.5013920068740845,
1268
+ "learning_rate": 2.517162471395881e-06,
1269
+ "loss": 0.6338,
1270
+ "step": 900
1271
+ },
1272
+ {
1273
+ "epoch": 0.9832948526415863,
1274
+ "grad_norm": 0.4734630882740021,
1275
+ "learning_rate": 1.9450800915331807e-06,
1276
+ "loss": 0.6189,
1277
+ "step": 905
1278
+ },
1279
+ {
1280
+ "epoch": 0.988727420888225,
1281
+ "grad_norm": 0.6268351078033447,
1282
+ "learning_rate": 1.3729977116704805e-06,
1283
+ "loss": 0.6632,
1284
+ "step": 910
1285
+ },
1286
+ {
1287
+ "epoch": 0.9941599891348635,
1288
+ "grad_norm": 0.5501301288604736,
1289
+ "learning_rate": 8.009153318077803e-07,
1290
+ "loss": 0.6514,
1291
+ "step": 915
1292
+ },
1293
+ {
1294
+ "epoch": 0.9995925573815021,
1295
+ "grad_norm": 0.5812506675720215,
1296
+ "learning_rate": 2.288329519450801e-07,
1297
+ "loss": 0.6137,
1298
+ "step": 920
1299
+ }
1300
+ ],
1301
+ "logging_steps": 5,
1302
+ "max_steps": 921,
1303
+ "num_input_tokens_seen": 0,
1304
+ "num_train_epochs": 1,
1305
+ "save_steps": 500,
1306
+ "stateful_callbacks": {
1307
+ "TrainerControl": {
1308
+ "args": {
1309
+ "should_epoch_stop": false,
1310
+ "should_evaluate": false,
1311
+ "should_log": false,
1312
+ "should_save": true,
1313
+ "should_training_stop": true
1314
+ },
1315
+ "attributes": {}
1316
+ }
1317
+ },
1318
+ "total_flos": 1.5239517919298716e+19,
1319
+ "train_batch_size": 1,
1320
+ "trial_name": null,
1321
+ "trial_params": null
1322
+ }
Qwen3-4B-Instruct-2507-sft-fusang-swa=2k_sink=0_falayer=7_from0_step8_fadec/checkpoint-1085/README.md ADDED
@@ -0,0 +1,207 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ base_model: /share/models/Qwen3-4B-Instruct-2507
3
+ library_name: peft
4
+ pipeline_tag: text-generation
5
+ tags:
6
+ - base_model:adapter:/share/models/Qwen3-4B-Instruct-2507
7
+ - lora
8
+ - transformers
9
+ ---
10
+
11
+ # Model Card for Model ID
12
+
13
+ <!-- Provide a quick summary of what the model is/does. -->
14
+
15
+
16
+
17
+ ## Model Details
18
+
19
+ ### Model Description
20
+
21
+ <!-- Provide a longer summary of what this model is. -->
22
+
23
+
24
+
25
+ - **Developed by:** [More Information Needed]
26
+ - **Funded by [optional]:** [More Information Needed]
27
+ - **Shared by [optional]:** [More Information Needed]
28
+ - **Model type:** [More Information Needed]
29
+ - **Language(s) (NLP):** [More Information Needed]
30
+ - **License:** [More Information Needed]
31
+ - **Finetuned from model [optional]:** [More Information Needed]
32
+
33
+ ### Model Sources [optional]
34
+
35
+ <!-- Provide the basic links for the model. -->
36
+
37
+ - **Repository:** [More Information Needed]
38
+ - **Paper [optional]:** [More Information Needed]
39
+ - **Demo [optional]:** [More Information Needed]
40
+
41
+ ## Uses
42
+
43
+ <!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
44
+
45
+ ### Direct Use
46
+
47
+ <!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
48
+
49
+ [More Information Needed]
50
+
51
+ ### Downstream Use [optional]
52
+
53
+ <!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
54
+
55
+ [More Information Needed]
56
+
57
+ ### Out-of-Scope Use
58
+
59
+ <!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
60
+
61
+ [More Information Needed]
62
+
63
+ ## Bias, Risks, and Limitations
64
+
65
+ <!-- This section is meant to convey both technical and sociotechnical limitations. -->
66
+
67
+ [More Information Needed]
68
+
69
+ ### Recommendations
70
+
71
+ <!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
72
+
73
+ Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
74
+
75
+ ## How to Get Started with the Model
76
+
77
+ Use the code below to get started with the model.
78
+
79
+ [More Information Needed]
80
+
81
+ ## Training Details
82
+
83
+ ### Training Data
84
+
85
+ <!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
86
+
87
+ [More Information Needed]
88
+
89
+ ### Training Procedure
90
+
91
+ <!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
92
+
93
+ #### Preprocessing [optional]
94
+
95
+ [More Information Needed]
96
+
97
+
98
+ #### Training Hyperparameters
99
+
100
+ - **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
101
+
102
+ #### Speeds, Sizes, Times [optional]
103
+
104
+ <!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
105
+
106
+ [More Information Needed]
107
+
108
+ ## Evaluation
109
+
110
+ <!-- This section describes the evaluation protocols and provides the results. -->
111
+
112
+ ### Testing Data, Factors & Metrics
113
+
114
+ #### Testing Data
115
+
116
+ <!-- This should link to a Dataset Card if possible. -->
117
+
118
+ [More Information Needed]
119
+
120
+ #### Factors
121
+
122
+ <!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
123
+
124
+ [More Information Needed]
125
+
126
+ #### Metrics
127
+
128
+ <!-- These are the evaluation metrics being used, ideally with a description of why. -->
129
+
130
+ [More Information Needed]
131
+
132
+ ### Results
133
+
134
+ [More Information Needed]
135
+
136
+ #### Summary
137
+
138
+
139
+
140
+ ## Model Examination [optional]
141
+
142
+ <!-- Relevant interpretability work for the model goes here -->
143
+
144
+ [More Information Needed]
145
+
146
+ ## Environmental Impact
147
+
148
+ <!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
149
+
150
+ Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
151
+
152
+ - **Hardware Type:** [More Information Needed]
153
+ - **Hours used:** [More Information Needed]
154
+ - **Cloud Provider:** [More Information Needed]
155
+ - **Compute Region:** [More Information Needed]
156
+ - **Carbon Emitted:** [More Information Needed]
157
+
158
+ ## Technical Specifications [optional]
159
+
160
+ ### Model Architecture and Objective
161
+
162
+ [More Information Needed]
163
+
164
+ ### Compute Infrastructure
165
+
166
+ [More Information Needed]
167
+
168
+ #### Hardware
169
+
170
+ [More Information Needed]
171
+
172
+ #### Software
173
+
174
+ [More Information Needed]
175
+
176
+ ## Citation [optional]
177
+
178
+ <!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
179
+
180
+ **BibTeX:**
181
+
182
+ [More Information Needed]
183
+
184
+ **APA:**
185
+
186
+ [More Information Needed]
187
+
188
+ ## Glossary [optional]
189
+
190
+ <!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
191
+
192
+ [More Information Needed]
193
+
194
+ ## More Information [optional]
195
+
196
+ [More Information Needed]
197
+
198
+ ## Model Card Authors [optional]
199
+
200
+ [More Information Needed]
201
+
202
+ ## Model Card Contact
203
+
204
+ [More Information Needed]
205
+ ### Framework versions
206
+
207
+ - PEFT 0.17.1
Qwen3-4B-Instruct-2507-sft-fusang-swa=2k_sink=0_falayer=7_from0_step8_fadec/checkpoint-1085/merges.txt ADDED
The diff for this file is too large to render. See raw diff
 
Qwen3-4B-Instruct-2507-sft-fusang-swa=2k_sink=0_falayer=7_from0_step8_fadec/checkpoint-1085/trainer_state.json ADDED
@@ -0,0 +1,1553 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "best_global_step": null,
3
+ "best_metric": null,
4
+ "best_model_checkpoint": null,
5
+ "epoch": 5.0,
6
+ "eval_steps": 500,
7
+ "global_step": 1085,
8
+ "is_hyper_param_search": false,
9
+ "is_local_process_zero": true,
10
+ "is_world_process_zero": true,
11
+ "log_history": [
12
+ {
13
+ "epoch": 0.023108030040439053,
14
+ "grad_norm": 6.191188812255859,
15
+ "learning_rate": 7.272727272727272e-06,
16
+ "loss": 0.7219,
17
+ "step": 5
18
+ },
19
+ {
20
+ "epoch": 0.046216060080878106,
21
+ "grad_norm": 3.22074818611145,
22
+ "learning_rate": 1.6363636363636366e-05,
23
+ "loss": 0.667,
24
+ "step": 10
25
+ },
26
+ {
27
+ "epoch": 0.06932409012131716,
28
+ "grad_norm": 1.6387842893600464,
29
+ "learning_rate": 2.5454545454545454e-05,
30
+ "loss": 0.5676,
31
+ "step": 15
32
+ },
33
+ {
34
+ "epoch": 0.09243212016175621,
35
+ "grad_norm": 1.957845687866211,
36
+ "learning_rate": 3.454545454545455e-05,
37
+ "loss": 0.4833,
38
+ "step": 20
39
+ },
40
+ {
41
+ "epoch": 0.11554015020219527,
42
+ "grad_norm": 1.3626967668533325,
43
+ "learning_rate": 4.3636363636363636e-05,
44
+ "loss": 0.4334,
45
+ "step": 25
46
+ },
47
+ {
48
+ "epoch": 0.1386481802426343,
49
+ "grad_norm": 4.611540794372559,
50
+ "learning_rate": 5.272727272727272e-05,
51
+ "loss": 0.4072,
52
+ "step": 30
53
+ },
54
+ {
55
+ "epoch": 0.16175621028307338,
56
+ "grad_norm": 1.0316890478134155,
57
+ "learning_rate": 6.181818181818182e-05,
58
+ "loss": 0.3794,
59
+ "step": 35
60
+ },
61
+ {
62
+ "epoch": 0.18486424032351242,
63
+ "grad_norm": 0.8542439341545105,
64
+ "learning_rate": 7.090909090909092e-05,
65
+ "loss": 0.3323,
66
+ "step": 40
67
+ },
68
+ {
69
+ "epoch": 0.20797227036395147,
70
+ "grad_norm": 0.8915650844573975,
71
+ "learning_rate": 8e-05,
72
+ "loss": 0.3306,
73
+ "step": 45
74
+ },
75
+ {
76
+ "epoch": 0.23108030040439054,
77
+ "grad_norm": 0.5692363381385803,
78
+ "learning_rate": 8.90909090909091e-05,
79
+ "loss": 0.3092,
80
+ "step": 50
81
+ },
82
+ {
83
+ "epoch": 0.2541883304448296,
84
+ "grad_norm": 0.6329680681228638,
85
+ "learning_rate": 9.818181818181818e-05,
86
+ "loss": 0.3075,
87
+ "step": 55
88
+ },
89
+ {
90
+ "epoch": 0.2772963604852686,
91
+ "grad_norm": 0.6504318118095398,
92
+ "learning_rate": 9.96116504854369e-05,
93
+ "loss": 0.3049,
94
+ "step": 60
95
+ },
96
+ {
97
+ "epoch": 0.30040439052570767,
98
+ "grad_norm": 0.6839262247085571,
99
+ "learning_rate": 9.912621359223302e-05,
100
+ "loss": 0.2694,
101
+ "step": 65
102
+ },
103
+ {
104
+ "epoch": 0.32351242056614676,
105
+ "grad_norm": 0.5372167825698853,
106
+ "learning_rate": 9.864077669902914e-05,
107
+ "loss": 0.2769,
108
+ "step": 70
109
+ },
110
+ {
111
+ "epoch": 0.3466204506065858,
112
+ "grad_norm": 0.6226201057434082,
113
+ "learning_rate": 9.815533980582525e-05,
114
+ "loss": 0.2781,
115
+ "step": 75
116
+ },
117
+ {
118
+ "epoch": 0.36972848064702485,
119
+ "grad_norm": 0.602139949798584,
120
+ "learning_rate": 9.766990291262137e-05,
121
+ "loss": 0.2801,
122
+ "step": 80
123
+ },
124
+ {
125
+ "epoch": 0.3928365106874639,
126
+ "grad_norm": 0.6526163816452026,
127
+ "learning_rate": 9.718446601941747e-05,
128
+ "loss": 0.2731,
129
+ "step": 85
130
+ },
131
+ {
132
+ "epoch": 0.41594454072790293,
133
+ "grad_norm": 0.6916335821151733,
134
+ "learning_rate": 9.66990291262136e-05,
135
+ "loss": 0.2692,
136
+ "step": 90
137
+ },
138
+ {
139
+ "epoch": 0.439052570768342,
140
+ "grad_norm": 0.6062453389167786,
141
+ "learning_rate": 9.621359223300971e-05,
142
+ "loss": 0.2565,
143
+ "step": 95
144
+ },
145
+ {
146
+ "epoch": 0.4621606008087811,
147
+ "grad_norm": 0.5832907557487488,
148
+ "learning_rate": 9.572815533980583e-05,
149
+ "loss": 0.257,
150
+ "step": 100
151
+ },
152
+ {
153
+ "epoch": 0.4852686308492201,
154
+ "grad_norm": 0.49384239315986633,
155
+ "learning_rate": 9.524271844660194e-05,
156
+ "loss": 0.2654,
157
+ "step": 105
158
+ },
159
+ {
160
+ "epoch": 0.5083766608896592,
161
+ "grad_norm": 0.5174527168273926,
162
+ "learning_rate": 9.475728155339806e-05,
163
+ "loss": 0.2677,
164
+ "step": 110
165
+ },
166
+ {
167
+ "epoch": 0.5314846909300982,
168
+ "grad_norm": 0.6395710110664368,
169
+ "learning_rate": 9.427184466019417e-05,
170
+ "loss": 0.2444,
171
+ "step": 115
172
+ },
173
+ {
174
+ "epoch": 0.5545927209705372,
175
+ "grad_norm": 0.7533590793609619,
176
+ "learning_rate": 9.37864077669903e-05,
177
+ "loss": 0.2618,
178
+ "step": 120
179
+ },
180
+ {
181
+ "epoch": 0.5777007510109763,
182
+ "grad_norm": 0.5848538875579834,
183
+ "learning_rate": 9.330097087378641e-05,
184
+ "loss": 0.2525,
185
+ "step": 125
186
+ },
187
+ {
188
+ "epoch": 0.6008087810514153,
189
+ "grad_norm": 0.5242823362350464,
190
+ "learning_rate": 9.281553398058253e-05,
191
+ "loss": 0.2544,
192
+ "step": 130
193
+ },
194
+ {
195
+ "epoch": 0.6239168110918544,
196
+ "grad_norm": 0.5916269421577454,
197
+ "learning_rate": 9.233009708737864e-05,
198
+ "loss": 0.2515,
199
+ "step": 135
200
+ },
201
+ {
202
+ "epoch": 0.6470248411322935,
203
+ "grad_norm": 0.6229275465011597,
204
+ "learning_rate": 9.184466019417476e-05,
205
+ "loss": 0.2555,
206
+ "step": 140
207
+ },
208
+ {
209
+ "epoch": 0.6701328711727326,
210
+ "grad_norm": 0.6378644704818726,
211
+ "learning_rate": 9.135922330097087e-05,
212
+ "loss": 0.2505,
213
+ "step": 145
214
+ },
215
+ {
216
+ "epoch": 0.6932409012131716,
217
+ "grad_norm": 0.5694279074668884,
218
+ "learning_rate": 9.087378640776699e-05,
219
+ "loss": 0.2445,
220
+ "step": 150
221
+ },
222
+ {
223
+ "epoch": 0.7163489312536107,
224
+ "grad_norm": 0.49784737825393677,
225
+ "learning_rate": 9.038834951456311e-05,
226
+ "loss": 0.25,
227
+ "step": 155
228
+ },
229
+ {
230
+ "epoch": 0.7394569612940497,
231
+ "grad_norm": 0.552670419216156,
232
+ "learning_rate": 8.990291262135923e-05,
233
+ "loss": 0.2486,
234
+ "step": 160
235
+ },
236
+ {
237
+ "epoch": 0.7625649913344887,
238
+ "grad_norm": 0.8254408836364746,
239
+ "learning_rate": 8.941747572815534e-05,
240
+ "loss": 0.2409,
241
+ "step": 165
242
+ },
243
+ {
244
+ "epoch": 0.7856730213749278,
245
+ "grad_norm": 0.5778919458389282,
246
+ "learning_rate": 8.893203883495146e-05,
247
+ "loss": 0.2385,
248
+ "step": 170
249
+ },
250
+ {
251
+ "epoch": 0.8087810514153668,
252
+ "grad_norm": 0.508798360824585,
253
+ "learning_rate": 8.844660194174757e-05,
254
+ "loss": 0.2405,
255
+ "step": 175
256
+ },
257
+ {
258
+ "epoch": 0.8318890814558059,
259
+ "grad_norm": 0.4466945230960846,
260
+ "learning_rate": 8.796116504854369e-05,
261
+ "loss": 0.254,
262
+ "step": 180
263
+ },
264
+ {
265
+ "epoch": 0.8549971114962449,
266
+ "grad_norm": 0.4597717225551605,
267
+ "learning_rate": 8.747572815533982e-05,
268
+ "loss": 0.2496,
269
+ "step": 185
270
+ },
271
+ {
272
+ "epoch": 0.878105141536684,
273
+ "grad_norm": 0.5333147644996643,
274
+ "learning_rate": 8.699029126213593e-05,
275
+ "loss": 0.2364,
276
+ "step": 190
277
+ },
278
+ {
279
+ "epoch": 0.901213171577123,
280
+ "grad_norm": 0.5476760268211365,
281
+ "learning_rate": 8.650485436893204e-05,
282
+ "loss": 0.2415,
283
+ "step": 195
284
+ },
285
+ {
286
+ "epoch": 0.9243212016175621,
287
+ "grad_norm": 0.38370969891548157,
288
+ "learning_rate": 8.601941747572816e-05,
289
+ "loss": 0.2395,
290
+ "step": 200
291
+ },
292
+ {
293
+ "epoch": 0.9474292316580012,
294
+ "grad_norm": 0.6363658905029297,
295
+ "learning_rate": 8.553398058252427e-05,
296
+ "loss": 0.2395,
297
+ "step": 205
298
+ },
299
+ {
300
+ "epoch": 0.9705372616984402,
301
+ "grad_norm": 0.4947147071361542,
302
+ "learning_rate": 8.504854368932039e-05,
303
+ "loss": 0.2466,
304
+ "step": 210
305
+ },
306
+ {
307
+ "epoch": 0.9936452917388793,
308
+ "grad_norm": 0.4944188594818115,
309
+ "learning_rate": 8.45631067961165e-05,
310
+ "loss": 0.2416,
311
+ "step": 215
312
+ },
313
+ {
314
+ "epoch": 1.0138648180242635,
315
+ "grad_norm": 0.9084780216217041,
316
+ "learning_rate": 8.407766990291263e-05,
317
+ "loss": 0.2398,
318
+ "step": 220
319
+ },
320
+ {
321
+ "epoch": 1.0369728480647025,
322
+ "grad_norm": 0.49810361862182617,
323
+ "learning_rate": 8.359223300970875e-05,
324
+ "loss": 0.2277,
325
+ "step": 225
326
+ },
327
+ {
328
+ "epoch": 1.0600808781051416,
329
+ "grad_norm": 0.40286511182785034,
330
+ "learning_rate": 8.310679611650486e-05,
331
+ "loss": 0.2211,
332
+ "step": 230
333
+ },
334
+ {
335
+ "epoch": 1.0831889081455806,
336
+ "grad_norm": 0.6430760622024536,
337
+ "learning_rate": 8.262135922330097e-05,
338
+ "loss": 0.2241,
339
+ "step": 235
340
+ },
341
+ {
342
+ "epoch": 1.1062969381860197,
343
+ "grad_norm": 0.49503597617149353,
344
+ "learning_rate": 8.213592233009709e-05,
345
+ "loss": 0.2441,
346
+ "step": 240
347
+ },
348
+ {
349
+ "epoch": 1.1294049682264586,
350
+ "grad_norm": 0.6000072360038757,
351
+ "learning_rate": 8.16504854368932e-05,
352
+ "loss": 0.2178,
353
+ "step": 245
354
+ },
355
+ {
356
+ "epoch": 1.1525129982668978,
357
+ "grad_norm": 0.7129755020141602,
358
+ "learning_rate": 8.116504854368933e-05,
359
+ "loss": 0.2287,
360
+ "step": 250
361
+ },
362
+ {
363
+ "epoch": 1.1756210283073367,
364
+ "grad_norm": 0.5162823796272278,
365
+ "learning_rate": 8.067961165048545e-05,
366
+ "loss": 0.2128,
367
+ "step": 255
368
+ },
369
+ {
370
+ "epoch": 1.1987290583477759,
371
+ "grad_norm": 0.4244346618652344,
372
+ "learning_rate": 8.019417475728156e-05,
373
+ "loss": 0.2233,
374
+ "step": 260
375
+ },
376
+ {
377
+ "epoch": 1.2218370883882148,
378
+ "grad_norm": 0.4976752698421478,
379
+ "learning_rate": 7.970873786407768e-05,
380
+ "loss": 0.2248,
381
+ "step": 265
382
+ },
383
+ {
384
+ "epoch": 1.244945118428654,
385
+ "grad_norm": 0.46501806378364563,
386
+ "learning_rate": 7.922330097087379e-05,
387
+ "loss": 0.2172,
388
+ "step": 270
389
+ },
390
+ {
391
+ "epoch": 1.2680531484690931,
392
+ "grad_norm": 0.4719155430793762,
393
+ "learning_rate": 7.87378640776699e-05,
394
+ "loss": 0.2223,
395
+ "step": 275
396
+ },
397
+ {
398
+ "epoch": 1.291161178509532,
399
+ "grad_norm": 0.6342799067497253,
400
+ "learning_rate": 7.825242718446603e-05,
401
+ "loss": 0.2196,
402
+ "step": 280
403
+ },
404
+ {
405
+ "epoch": 1.314269208549971,
406
+ "grad_norm": 0.3851456046104431,
407
+ "learning_rate": 7.776699029126215e-05,
408
+ "loss": 0.2183,
409
+ "step": 285
410
+ },
411
+ {
412
+ "epoch": 1.3373772385904101,
413
+ "grad_norm": 0.48566851019859314,
414
+ "learning_rate": 7.728155339805826e-05,
415
+ "loss": 0.2229,
416
+ "step": 290
417
+ },
418
+ {
419
+ "epoch": 1.3604852686308493,
420
+ "grad_norm": 0.5373956561088562,
421
+ "learning_rate": 7.679611650485438e-05,
422
+ "loss": 0.2212,
423
+ "step": 295
424
+ },
425
+ {
426
+ "epoch": 1.3835932986712882,
427
+ "grad_norm": 0.49365347623825073,
428
+ "learning_rate": 7.631067961165049e-05,
429
+ "loss": 0.2206,
430
+ "step": 300
431
+ },
432
+ {
433
+ "epoch": 1.4067013287117274,
434
+ "grad_norm": 0.4670807719230652,
435
+ "learning_rate": 7.58252427184466e-05,
436
+ "loss": 0.2086,
437
+ "step": 305
438
+ },
439
+ {
440
+ "epoch": 1.4298093587521663,
441
+ "grad_norm": 0.5047533512115479,
442
+ "learning_rate": 7.533980582524272e-05,
443
+ "loss": 0.2271,
444
+ "step": 310
445
+ },
446
+ {
447
+ "epoch": 1.4529173887926055,
448
+ "grad_norm": 0.5835520029067993,
449
+ "learning_rate": 7.485436893203885e-05,
450
+ "loss": 0.2262,
451
+ "step": 315
452
+ },
453
+ {
454
+ "epoch": 1.4760254188330446,
455
+ "grad_norm": 0.4543643295764923,
456
+ "learning_rate": 7.436893203883496e-05,
457
+ "loss": 0.2144,
458
+ "step": 320
459
+ },
460
+ {
461
+ "epoch": 1.4991334488734835,
462
+ "grad_norm": 0.5098664164543152,
463
+ "learning_rate": 7.388349514563108e-05,
464
+ "loss": 0.2202,
465
+ "step": 325
466
+ },
467
+ {
468
+ "epoch": 1.5222414789139225,
469
+ "grad_norm": 0.4698500335216522,
470
+ "learning_rate": 7.339805825242719e-05,
471
+ "loss": 0.2192,
472
+ "step": 330
473
+ },
474
+ {
475
+ "epoch": 1.5453495089543616,
476
+ "grad_norm": 0.4429137110710144,
477
+ "learning_rate": 7.29126213592233e-05,
478
+ "loss": 0.224,
479
+ "step": 335
480
+ },
481
+ {
482
+ "epoch": 1.5684575389948008,
483
+ "grad_norm": 0.5282397270202637,
484
+ "learning_rate": 7.242718446601942e-05,
485
+ "loss": 0.2326,
486
+ "step": 340
487
+ },
488
+ {
489
+ "epoch": 1.5915655690352397,
490
+ "grad_norm": 0.5188369750976562,
491
+ "learning_rate": 7.194174757281554e-05,
492
+ "loss": 0.2258,
493
+ "step": 345
494
+ },
495
+ {
496
+ "epoch": 1.6146735990756786,
497
+ "grad_norm": 0.47453057765960693,
498
+ "learning_rate": 7.145631067961165e-05,
499
+ "loss": 0.2158,
500
+ "step": 350
501
+ },
502
+ {
503
+ "epoch": 1.6377816291161178,
504
+ "grad_norm": 0.46420833468437195,
505
+ "learning_rate": 7.097087378640776e-05,
506
+ "loss": 0.2161,
507
+ "step": 355
508
+ },
509
+ {
510
+ "epoch": 1.660889659156557,
511
+ "grad_norm": 0.48289036750793457,
512
+ "learning_rate": 7.048543689320388e-05,
513
+ "loss": 0.2227,
514
+ "step": 360
515
+ },
516
+ {
517
+ "epoch": 1.683997689196996,
518
+ "grad_norm": 0.5904167890548706,
519
+ "learning_rate": 7e-05,
520
+ "loss": 0.2269,
521
+ "step": 365
522
+ },
523
+ {
524
+ "epoch": 1.707105719237435,
525
+ "grad_norm": 0.47676077485084534,
526
+ "learning_rate": 6.951456310679612e-05,
527
+ "loss": 0.2094,
528
+ "step": 370
529
+ },
530
+ {
531
+ "epoch": 1.730213749277874,
532
+ "grad_norm": 0.5335319638252258,
533
+ "learning_rate": 6.902912621359224e-05,
534
+ "loss": 0.217,
535
+ "step": 375
536
+ },
537
+ {
538
+ "epoch": 1.7533217793183131,
539
+ "grad_norm": 0.4710823893547058,
540
+ "learning_rate": 6.854368932038835e-05,
541
+ "loss": 0.212,
542
+ "step": 380
543
+ },
544
+ {
545
+ "epoch": 1.7764298093587523,
546
+ "grad_norm": 0.3974303603172302,
547
+ "learning_rate": 6.805825242718447e-05,
548
+ "loss": 0.2292,
549
+ "step": 385
550
+ },
551
+ {
552
+ "epoch": 1.7995378393991912,
553
+ "grad_norm": 1.402204990386963,
554
+ "learning_rate": 6.757281553398058e-05,
555
+ "loss": 0.2329,
556
+ "step": 390
557
+ },
558
+ {
559
+ "epoch": 1.8226458694396301,
560
+ "grad_norm": 0.7323683500289917,
561
+ "learning_rate": 6.70873786407767e-05,
562
+ "loss": 0.223,
563
+ "step": 395
564
+ },
565
+ {
566
+ "epoch": 1.8457538994800693,
567
+ "grad_norm": 0.5015227794647217,
568
+ "learning_rate": 6.660194174757281e-05,
569
+ "loss": 0.2204,
570
+ "step": 400
571
+ },
572
+ {
573
+ "epoch": 1.8688619295205084,
574
+ "grad_norm": 0.5568329095840454,
575
+ "learning_rate": 6.611650485436894e-05,
576
+ "loss": 0.2148,
577
+ "step": 405
578
+ },
579
+ {
580
+ "epoch": 1.8919699595609474,
581
+ "grad_norm": 0.4861031770706177,
582
+ "learning_rate": 6.563106796116505e-05,
583
+ "loss": 0.2185,
584
+ "step": 410
585
+ },
586
+ {
587
+ "epoch": 1.9150779896013865,
588
+ "grad_norm": 0.428276002407074,
589
+ "learning_rate": 6.514563106796117e-05,
590
+ "loss": 0.2124,
591
+ "step": 415
592
+ },
593
+ {
594
+ "epoch": 1.9381860196418255,
595
+ "grad_norm": 0.47187644243240356,
596
+ "learning_rate": 6.466019417475728e-05,
597
+ "loss": 0.217,
598
+ "step": 420
599
+ },
600
+ {
601
+ "epoch": 1.9612940496822646,
602
+ "grad_norm": 0.5285502076148987,
603
+ "learning_rate": 6.41747572815534e-05,
604
+ "loss": 0.2197,
605
+ "step": 425
606
+ },
607
+ {
608
+ "epoch": 1.9844020797227038,
609
+ "grad_norm": 0.4158755838871002,
610
+ "learning_rate": 6.368932038834951e-05,
611
+ "loss": 0.2214,
612
+ "step": 430
613
+ },
614
+ {
615
+ "epoch": 2.004621606008088,
616
+ "grad_norm": 0.5102666616439819,
617
+ "learning_rate": 6.320388349514564e-05,
618
+ "loss": 0.2135,
619
+ "step": 435
620
+ },
621
+ {
622
+ "epoch": 2.027729636048527,
623
+ "grad_norm": 0.34488484263420105,
624
+ "learning_rate": 6.271844660194175e-05,
625
+ "loss": 0.1928,
626
+ "step": 440
627
+ },
628
+ {
629
+ "epoch": 2.050837666088966,
630
+ "grad_norm": 0.5306763052940369,
631
+ "learning_rate": 6.223300970873787e-05,
632
+ "loss": 0.1968,
633
+ "step": 445
634
+ },
635
+ {
636
+ "epoch": 2.073945696129405,
637
+ "grad_norm": 0.44535988569259644,
638
+ "learning_rate": 6.174757281553398e-05,
639
+ "loss": 0.2008,
640
+ "step": 450
641
+ },
642
+ {
643
+ "epoch": 2.097053726169844,
644
+ "grad_norm": 0.9996095895767212,
645
+ "learning_rate": 6.12621359223301e-05,
646
+ "loss": 0.1996,
647
+ "step": 455
648
+ },
649
+ {
650
+ "epoch": 2.1201617562102832,
651
+ "grad_norm": 0.6485170125961304,
652
+ "learning_rate": 6.077669902912622e-05,
653
+ "loss": 0.2017,
654
+ "step": 460
655
+ },
656
+ {
657
+ "epoch": 2.143269786250722,
658
+ "grad_norm": 0.41752228140830994,
659
+ "learning_rate": 6.029126213592233e-05,
660
+ "loss": 0.1982,
661
+ "step": 465
662
+ },
663
+ {
664
+ "epoch": 2.166377816291161,
665
+ "grad_norm": 0.414360374212265,
666
+ "learning_rate": 5.980582524271845e-05,
667
+ "loss": 0.2077,
668
+ "step": 470
669
+ },
670
+ {
671
+ "epoch": 2.1894858463316003,
672
+ "grad_norm": 0.4518888294696808,
673
+ "learning_rate": 5.932038834951457e-05,
674
+ "loss": 0.2079,
675
+ "step": 475
676
+ },
677
+ {
678
+ "epoch": 2.2125938763720394,
679
+ "grad_norm": 0.7040420174598694,
680
+ "learning_rate": 5.883495145631068e-05,
681
+ "loss": 0.1975,
682
+ "step": 480
683
+ },
684
+ {
685
+ "epoch": 2.235701906412478,
686
+ "grad_norm": 0.5127845406532288,
687
+ "learning_rate": 5.83495145631068e-05,
688
+ "loss": 0.2038,
689
+ "step": 485
690
+ },
691
+ {
692
+ "epoch": 2.2588099364529173,
693
+ "grad_norm": 0.5380415320396423,
694
+ "learning_rate": 5.786407766990292e-05,
695
+ "loss": 0.2003,
696
+ "step": 490
697
+ },
698
+ {
699
+ "epoch": 2.2819179664933564,
700
+ "grad_norm": 0.479818195104599,
701
+ "learning_rate": 5.7378640776699034e-05,
702
+ "loss": 0.2066,
703
+ "step": 495
704
+ },
705
+ {
706
+ "epoch": 2.3050259965337956,
707
+ "grad_norm": 0.4582814872264862,
708
+ "learning_rate": 5.689320388349515e-05,
709
+ "loss": 0.1967,
710
+ "step": 500
711
+ },
712
+ {
713
+ "epoch": 2.3281340265742347,
714
+ "grad_norm": 0.4949081838130951,
715
+ "learning_rate": 5.640776699029127e-05,
716
+ "loss": 0.1962,
717
+ "step": 505
718
+ },
719
+ {
720
+ "epoch": 2.3512420566146734,
721
+ "grad_norm": 0.5346571207046509,
722
+ "learning_rate": 5.5922330097087384e-05,
723
+ "loss": 0.2001,
724
+ "step": 510
725
+ },
726
+ {
727
+ "epoch": 2.3743500866551126,
728
+ "grad_norm": 0.8289781212806702,
729
+ "learning_rate": 5.54368932038835e-05,
730
+ "loss": 0.211,
731
+ "step": 515
732
+ },
733
+ {
734
+ "epoch": 2.3974581166955518,
735
+ "grad_norm": 0.44271352887153625,
736
+ "learning_rate": 5.4951456310679613e-05,
737
+ "loss": 0.2034,
738
+ "step": 520
739
+ },
740
+ {
741
+ "epoch": 2.420566146735991,
742
+ "grad_norm": 0.5730252861976624,
743
+ "learning_rate": 5.4466019417475735e-05,
744
+ "loss": 0.2056,
745
+ "step": 525
746
+ },
747
+ {
748
+ "epoch": 2.4436741767764296,
749
+ "grad_norm": 0.4633658230304718,
750
+ "learning_rate": 5.398058252427185e-05,
751
+ "loss": 0.2042,
752
+ "step": 530
753
+ },
754
+ {
755
+ "epoch": 2.4667822068168688,
756
+ "grad_norm": 0.5793203711509705,
757
+ "learning_rate": 5.3495145631067964e-05,
758
+ "loss": 0.1999,
759
+ "step": 535
760
+ },
761
+ {
762
+ "epoch": 2.489890236857308,
763
+ "grad_norm": 0.4566795229911804,
764
+ "learning_rate": 5.3009708737864085e-05,
765
+ "loss": 0.1994,
766
+ "step": 540
767
+ },
768
+ {
769
+ "epoch": 2.512998266897747,
770
+ "grad_norm": 0.6063486933708191,
771
+ "learning_rate": 5.25242718446602e-05,
772
+ "loss": 0.2136,
773
+ "step": 545
774
+ },
775
+ {
776
+ "epoch": 2.5361062969381862,
777
+ "grad_norm": 0.5125712156295776,
778
+ "learning_rate": 5.2038834951456314e-05,
779
+ "loss": 0.2108,
780
+ "step": 550
781
+ },
782
+ {
783
+ "epoch": 2.559214326978625,
784
+ "grad_norm": 0.5539318323135376,
785
+ "learning_rate": 5.1553398058252436e-05,
786
+ "loss": 0.2038,
787
+ "step": 555
788
+ },
789
+ {
790
+ "epoch": 2.582322357019064,
791
+ "grad_norm": 0.5637097358703613,
792
+ "learning_rate": 5.106796116504855e-05,
793
+ "loss": 0.2056,
794
+ "step": 560
795
+ },
796
+ {
797
+ "epoch": 2.6054303870595032,
798
+ "grad_norm": 0.3843493163585663,
799
+ "learning_rate": 5.0582524271844665e-05,
800
+ "loss": 0.1999,
801
+ "step": 565
802
+ },
803
+ {
804
+ "epoch": 2.628538417099942,
805
+ "grad_norm": 0.49531030654907227,
806
+ "learning_rate": 5.0097087378640786e-05,
807
+ "loss": 0.1984,
808
+ "step": 570
809
+ },
810
+ {
811
+ "epoch": 2.651646447140381,
812
+ "grad_norm": 0.467149555683136,
813
+ "learning_rate": 4.9611650485436894e-05,
814
+ "loss": 0.2106,
815
+ "step": 575
816
+ },
817
+ {
818
+ "epoch": 2.6747544771808203,
819
+ "grad_norm": 0.4261501133441925,
820
+ "learning_rate": 4.912621359223301e-05,
821
+ "loss": 0.1996,
822
+ "step": 580
823
+ },
824
+ {
825
+ "epoch": 2.6978625072212594,
826
+ "grad_norm": 0.43489792943000793,
827
+ "learning_rate": 4.864077669902913e-05,
828
+ "loss": 0.1979,
829
+ "step": 585
830
+ },
831
+ {
832
+ "epoch": 2.7209705372616986,
833
+ "grad_norm": 0.3987080454826355,
834
+ "learning_rate": 4.8155339805825245e-05,
835
+ "loss": 0.1932,
836
+ "step": 590
837
+ },
838
+ {
839
+ "epoch": 2.7440785673021377,
840
+ "grad_norm": 0.45573779940605164,
841
+ "learning_rate": 4.766990291262136e-05,
842
+ "loss": 0.2057,
843
+ "step": 595
844
+ },
845
+ {
846
+ "epoch": 2.7671865973425764,
847
+ "grad_norm": 0.46091750264167786,
848
+ "learning_rate": 4.718446601941748e-05,
849
+ "loss": 0.2007,
850
+ "step": 600
851
+ },
852
+ {
853
+ "epoch": 2.7902946273830156,
854
+ "grad_norm": 0.49786409735679626,
855
+ "learning_rate": 4.6699029126213595e-05,
856
+ "loss": 0.2007,
857
+ "step": 605
858
+ },
859
+ {
860
+ "epoch": 2.8134026574234547,
861
+ "grad_norm": 0.6425182819366455,
862
+ "learning_rate": 4.621359223300971e-05,
863
+ "loss": 0.2004,
864
+ "step": 610
865
+ },
866
+ {
867
+ "epoch": 2.8365106874638935,
868
+ "grad_norm": 0.4545029103755951,
869
+ "learning_rate": 4.572815533980583e-05,
870
+ "loss": 0.2048,
871
+ "step": 615
872
+ },
873
+ {
874
+ "epoch": 2.8596187175043326,
875
+ "grad_norm": 0.4330201745033264,
876
+ "learning_rate": 4.5242718446601946e-05,
877
+ "loss": 0.2043,
878
+ "step": 620
879
+ },
880
+ {
881
+ "epoch": 2.8827267475447718,
882
+ "grad_norm": 0.488075852394104,
883
+ "learning_rate": 4.475728155339806e-05,
884
+ "loss": 0.1992,
885
+ "step": 625
886
+ },
887
+ {
888
+ "epoch": 2.905834777585211,
889
+ "grad_norm": 0.43872949481010437,
890
+ "learning_rate": 4.427184466019418e-05,
891
+ "loss": 0.1956,
892
+ "step": 630
893
+ },
894
+ {
895
+ "epoch": 2.92894280762565,
896
+ "grad_norm": 0.6535570621490479,
897
+ "learning_rate": 4.3786407766990296e-05,
898
+ "loss": 0.1984,
899
+ "step": 635
900
+ },
901
+ {
902
+ "epoch": 2.9520508376660892,
903
+ "grad_norm": 0.3532000482082367,
904
+ "learning_rate": 4.330097087378641e-05,
905
+ "loss": 0.1971,
906
+ "step": 640
907
+ },
908
+ {
909
+ "epoch": 2.975158867706528,
910
+ "grad_norm": 0.5041428208351135,
911
+ "learning_rate": 4.2815533980582525e-05,
912
+ "loss": 0.2018,
913
+ "step": 645
914
+ },
915
+ {
916
+ "epoch": 2.998266897746967,
917
+ "grad_norm": 0.43183717131614685,
918
+ "learning_rate": 4.2330097087378647e-05,
919
+ "loss": 0.1998,
920
+ "step": 650
921
+ },
922
+ {
923
+ "epoch": 3.0184864240323512,
924
+ "grad_norm": 0.4431833326816559,
925
+ "learning_rate": 4.184466019417476e-05,
926
+ "loss": 0.2032,
927
+ "step": 655
928
+ },
929
+ {
930
+ "epoch": 3.0415944540727904,
931
+ "grad_norm": 0.4352896213531494,
932
+ "learning_rate": 4.1359223300970876e-05,
933
+ "loss": 0.1936,
934
+ "step": 660
935
+ },
936
+ {
937
+ "epoch": 3.0647024841132295,
938
+ "grad_norm": 0.48912373185157776,
939
+ "learning_rate": 4.087378640776699e-05,
940
+ "loss": 0.1939,
941
+ "step": 665
942
+ },
943
+ {
944
+ "epoch": 3.0878105141536683,
945
+ "grad_norm": 0.4135836660861969,
946
+ "learning_rate": 4.0388349514563105e-05,
947
+ "loss": 0.1831,
948
+ "step": 670
949
+ },
950
+ {
951
+ "epoch": 3.1109185441941074,
952
+ "grad_norm": 0.4050876796245575,
953
+ "learning_rate": 3.9902912621359226e-05,
954
+ "loss": 0.1929,
955
+ "step": 675
956
+ },
957
+ {
958
+ "epoch": 3.1340265742345466,
959
+ "grad_norm": 0.8014656901359558,
960
+ "learning_rate": 3.941747572815534e-05,
961
+ "loss": 0.1875,
962
+ "step": 680
963
+ },
964
+ {
965
+ "epoch": 3.1571346042749857,
966
+ "grad_norm": 0.4781719744205475,
967
+ "learning_rate": 3.8932038834951455e-05,
968
+ "loss": 0.1862,
969
+ "step": 685
970
+ },
971
+ {
972
+ "epoch": 3.1802426343154244,
973
+ "grad_norm": 0.5232430696487427,
974
+ "learning_rate": 3.844660194174757e-05,
975
+ "loss": 0.1806,
976
+ "step": 690
977
+ },
978
+ {
979
+ "epoch": 3.2033506643558636,
980
+ "grad_norm": 0.41624128818511963,
981
+ "learning_rate": 3.796116504854369e-05,
982
+ "loss": 0.1849,
983
+ "step": 695
984
+ },
985
+ {
986
+ "epoch": 3.2264586943963027,
987
+ "grad_norm": 0.45475468039512634,
988
+ "learning_rate": 3.7475728155339806e-05,
989
+ "loss": 0.1927,
990
+ "step": 700
991
+ },
992
+ {
993
+ "epoch": 3.249566724436742,
994
+ "grad_norm": 0.48061472177505493,
995
+ "learning_rate": 3.699029126213592e-05,
996
+ "loss": 0.1781,
997
+ "step": 705
998
+ },
999
+ {
1000
+ "epoch": 3.2726747544771806,
1001
+ "grad_norm": 0.5077313780784607,
1002
+ "learning_rate": 3.650485436893204e-05,
1003
+ "loss": 0.189,
1004
+ "step": 710
1005
+ },
1006
+ {
1007
+ "epoch": 3.2957827845176197,
1008
+ "grad_norm": 0.4867580533027649,
1009
+ "learning_rate": 3.6019417475728156e-05,
1010
+ "loss": 0.1874,
1011
+ "step": 715
1012
+ },
1013
+ {
1014
+ "epoch": 3.318890814558059,
1015
+ "grad_norm": 0.4246606230735779,
1016
+ "learning_rate": 3.553398058252427e-05,
1017
+ "loss": 0.1885,
1018
+ "step": 720
1019
+ },
1020
+ {
1021
+ "epoch": 3.341998844598498,
1022
+ "grad_norm": 0.4367058277130127,
1023
+ "learning_rate": 3.504854368932039e-05,
1024
+ "loss": 0.1895,
1025
+ "step": 725
1026
+ },
1027
+ {
1028
+ "epoch": 3.365106874638937,
1029
+ "grad_norm": 0.41172537207603455,
1030
+ "learning_rate": 3.456310679611651e-05,
1031
+ "loss": 0.191,
1032
+ "step": 730
1033
+ },
1034
+ {
1035
+ "epoch": 3.388214904679376,
1036
+ "grad_norm": 0.44100499153137207,
1037
+ "learning_rate": 3.407766990291262e-05,
1038
+ "loss": 0.187,
1039
+ "step": 735
1040
+ },
1041
+ {
1042
+ "epoch": 3.411322934719815,
1043
+ "grad_norm": 0.5393614768981934,
1044
+ "learning_rate": 3.359223300970874e-05,
1045
+ "loss": 0.1845,
1046
+ "step": 740
1047
+ },
1048
+ {
1049
+ "epoch": 3.4344309647602542,
1050
+ "grad_norm": 0.3750293552875519,
1051
+ "learning_rate": 3.310679611650486e-05,
1052
+ "loss": 0.1829,
1053
+ "step": 745
1054
+ },
1055
+ {
1056
+ "epoch": 3.4575389948006934,
1057
+ "grad_norm": 0.6407081484794617,
1058
+ "learning_rate": 3.262135922330097e-05,
1059
+ "loss": 0.1838,
1060
+ "step": 750
1061
+ },
1062
+ {
1063
+ "epoch": 3.480647024841132,
1064
+ "grad_norm": 0.527770459651947,
1065
+ "learning_rate": 3.213592233009709e-05,
1066
+ "loss": 0.1869,
1067
+ "step": 755
1068
+ },
1069
+ {
1070
+ "epoch": 3.5037550548815712,
1071
+ "grad_norm": 0.4695970118045807,
1072
+ "learning_rate": 3.165048543689321e-05,
1073
+ "loss": 0.1955,
1074
+ "step": 760
1075
+ },
1076
+ {
1077
+ "epoch": 3.5268630849220104,
1078
+ "grad_norm": 0.44379812479019165,
1079
+ "learning_rate": 3.116504854368932e-05,
1080
+ "loss": 0.1851,
1081
+ "step": 765
1082
+ },
1083
+ {
1084
+ "epoch": 3.5499711149624495,
1085
+ "grad_norm": 0.46998265385627747,
1086
+ "learning_rate": 3.067961165048544e-05,
1087
+ "loss": 0.1834,
1088
+ "step": 770
1089
+ },
1090
+ {
1091
+ "epoch": 3.5730791450028887,
1092
+ "grad_norm": 0.4266503155231476,
1093
+ "learning_rate": 3.019417475728156e-05,
1094
+ "loss": 0.1927,
1095
+ "step": 775
1096
+ },
1097
+ {
1098
+ "epoch": 3.5961871750433274,
1099
+ "grad_norm": 0.5007091760635376,
1100
+ "learning_rate": 2.9708737864077673e-05,
1101
+ "loss": 0.1836,
1102
+ "step": 780
1103
+ },
1104
+ {
1105
+ "epoch": 3.6192952050837666,
1106
+ "grad_norm": 0.5991454124450684,
1107
+ "learning_rate": 2.922330097087379e-05,
1108
+ "loss": 0.1938,
1109
+ "step": 785
1110
+ },
1111
+ {
1112
+ "epoch": 3.6424032351242057,
1113
+ "grad_norm": 0.5230900645256042,
1114
+ "learning_rate": 2.8737864077669902e-05,
1115
+ "loss": 0.1882,
1116
+ "step": 790
1117
+ },
1118
+ {
1119
+ "epoch": 3.665511265164645,
1120
+ "grad_norm": 0.5093218684196472,
1121
+ "learning_rate": 2.8252427184466017e-05,
1122
+ "loss": 0.1841,
1123
+ "step": 795
1124
+ },
1125
+ {
1126
+ "epoch": 3.6886192952050836,
1127
+ "grad_norm": 0.4400416910648346,
1128
+ "learning_rate": 2.7766990291262135e-05,
1129
+ "loss": 0.1849,
1130
+ "step": 800
1131
+ },
1132
+ {
1133
+ "epoch": 3.7117273252455227,
1134
+ "grad_norm": 0.4958287477493286,
1135
+ "learning_rate": 2.7281553398058253e-05,
1136
+ "loss": 0.1788,
1137
+ "step": 805
1138
+ },
1139
+ {
1140
+ "epoch": 3.734835355285962,
1141
+ "grad_norm": 0.48447561264038086,
1142
+ "learning_rate": 2.6796116504854367e-05,
1143
+ "loss": 0.1893,
1144
+ "step": 810
1145
+ },
1146
+ {
1147
+ "epoch": 3.757943385326401,
1148
+ "grad_norm": 0.7652764320373535,
1149
+ "learning_rate": 2.6310679611650485e-05,
1150
+ "loss": 0.1891,
1151
+ "step": 815
1152
+ },
1153
+ {
1154
+ "epoch": 3.78105141536684,
1155
+ "grad_norm": 0.39542829990386963,
1156
+ "learning_rate": 2.5825242718446603e-05,
1157
+ "loss": 0.1828,
1158
+ "step": 820
1159
+ },
1160
+ {
1161
+ "epoch": 3.804159445407279,
1162
+ "grad_norm": 0.41204145550727844,
1163
+ "learning_rate": 2.5339805825242718e-05,
1164
+ "loss": 0.1835,
1165
+ "step": 825
1166
+ },
1167
+ {
1168
+ "epoch": 3.827267475447718,
1169
+ "grad_norm": 0.5135506391525269,
1170
+ "learning_rate": 2.4854368932038836e-05,
1171
+ "loss": 0.1941,
1172
+ "step": 830
1173
+ },
1174
+ {
1175
+ "epoch": 3.850375505488157,
1176
+ "grad_norm": 0.47059449553489685,
1177
+ "learning_rate": 2.436893203883495e-05,
1178
+ "loss": 0.1937,
1179
+ "step": 835
1180
+ },
1181
+ {
1182
+ "epoch": 3.873483535528596,
1183
+ "grad_norm": 0.4884239733219147,
1184
+ "learning_rate": 2.3883495145631068e-05,
1185
+ "loss": 0.1886,
1186
+ "step": 840
1187
+ },
1188
+ {
1189
+ "epoch": 3.896591565569035,
1190
+ "grad_norm": 0.41069746017456055,
1191
+ "learning_rate": 2.3398058252427186e-05,
1192
+ "loss": 0.1867,
1193
+ "step": 845
1194
+ },
1195
+ {
1196
+ "epoch": 3.9196995956094742,
1197
+ "grad_norm": 0.4501801133155823,
1198
+ "learning_rate": 2.29126213592233e-05,
1199
+ "loss": 0.1908,
1200
+ "step": 850
1201
+ },
1202
+ {
1203
+ "epoch": 3.9428076256499134,
1204
+ "grad_norm": 0.5859938859939575,
1205
+ "learning_rate": 2.242718446601942e-05,
1206
+ "loss": 0.1987,
1207
+ "step": 855
1208
+ },
1209
+ {
1210
+ "epoch": 3.9659156556903525,
1211
+ "grad_norm": 0.5006229877471924,
1212
+ "learning_rate": 2.1941747572815537e-05,
1213
+ "loss": 0.1873,
1214
+ "step": 860
1215
+ },
1216
+ {
1217
+ "epoch": 3.9890236857307917,
1218
+ "grad_norm": 0.4370214641094208,
1219
+ "learning_rate": 2.145631067961165e-05,
1220
+ "loss": 0.1784,
1221
+ "step": 865
1222
+ },
1223
+ {
1224
+ "epoch": 4.009243212016176,
1225
+ "grad_norm": 0.43814602494239807,
1226
+ "learning_rate": 2.097087378640777e-05,
1227
+ "loss": 0.19,
1228
+ "step": 870
1229
+ },
1230
+ {
1231
+ "epoch": 4.032351242056615,
1232
+ "grad_norm": 0.47679728269577026,
1233
+ "learning_rate": 2.0485436893203887e-05,
1234
+ "loss": 0.1744,
1235
+ "step": 875
1236
+ },
1237
+ {
1238
+ "epoch": 4.055459272097054,
1239
+ "grad_norm": 0.38539212942123413,
1240
+ "learning_rate": 2e-05,
1241
+ "loss": 0.1825,
1242
+ "step": 880
1243
+ },
1244
+ {
1245
+ "epoch": 4.078567302137492,
1246
+ "grad_norm": 0.48871466517448425,
1247
+ "learning_rate": 1.951456310679612e-05,
1248
+ "loss": 0.1886,
1249
+ "step": 885
1250
+ },
1251
+ {
1252
+ "epoch": 4.101675332177932,
1253
+ "grad_norm": 0.4461103677749634,
1254
+ "learning_rate": 1.9029126213592234e-05,
1255
+ "loss": 0.1833,
1256
+ "step": 890
1257
+ },
1258
+ {
1259
+ "epoch": 4.124783362218371,
1260
+ "grad_norm": 0.44559869170188904,
1261
+ "learning_rate": 1.854368932038835e-05,
1262
+ "loss": 0.1813,
1263
+ "step": 895
1264
+ },
1265
+ {
1266
+ "epoch": 4.14789139225881,
1267
+ "grad_norm": 0.4957844614982605,
1268
+ "learning_rate": 1.8058252427184467e-05,
1269
+ "loss": 0.1797,
1270
+ "step": 900
1271
+ },
1272
+ {
1273
+ "epoch": 4.170999422299249,
1274
+ "grad_norm": 0.4256135821342468,
1275
+ "learning_rate": 1.757281553398058e-05,
1276
+ "loss": 0.167,
1277
+ "step": 905
1278
+ },
1279
+ {
1280
+ "epoch": 4.194107452339688,
1281
+ "grad_norm": 0.4208783507347107,
1282
+ "learning_rate": 1.70873786407767e-05,
1283
+ "loss": 0.1828,
1284
+ "step": 910
1285
+ },
1286
+ {
1287
+ "epoch": 4.217215482380127,
1288
+ "grad_norm": 0.4647185206413269,
1289
+ "learning_rate": 1.6601941747572817e-05,
1290
+ "loss": 0.1847,
1291
+ "step": 915
1292
+ },
1293
+ {
1294
+ "epoch": 4.2403235124205665,
1295
+ "grad_norm": 0.5371337532997131,
1296
+ "learning_rate": 1.6116504854368932e-05,
1297
+ "loss": 0.1809,
1298
+ "step": 920
1299
+ },
1300
+ {
1301
+ "epoch": 4.263431542461005,
1302
+ "grad_norm": 0.48796018958091736,
1303
+ "learning_rate": 1.563106796116505e-05,
1304
+ "loss": 0.1731,
1305
+ "step": 925
1306
+ },
1307
+ {
1308
+ "epoch": 4.286539572501444,
1309
+ "grad_norm": 0.4002375304698944,
1310
+ "learning_rate": 1.5145631067961166e-05,
1311
+ "loss": 0.179,
1312
+ "step": 930
1313
+ },
1314
+ {
1315
+ "epoch": 4.309647602541883,
1316
+ "grad_norm": 0.42429423332214355,
1317
+ "learning_rate": 1.4660194174757282e-05,
1318
+ "loss": 0.1727,
1319
+ "step": 935
1320
+ },
1321
+ {
1322
+ "epoch": 4.332755632582322,
1323
+ "grad_norm": 0.484302282333374,
1324
+ "learning_rate": 1.41747572815534e-05,
1325
+ "loss": 0.1809,
1326
+ "step": 940
1327
+ },
1328
+ {
1329
+ "epoch": 4.355863662622761,
1330
+ "grad_norm": 0.49636998772621155,
1331
+ "learning_rate": 1.3689320388349517e-05,
1332
+ "loss": 0.1826,
1333
+ "step": 945
1334
+ },
1335
+ {
1336
+ "epoch": 4.3789716926632005,
1337
+ "grad_norm": 0.5043439269065857,
1338
+ "learning_rate": 1.3203883495145633e-05,
1339
+ "loss": 0.183,
1340
+ "step": 950
1341
+ },
1342
+ {
1343
+ "epoch": 4.40207972270364,
1344
+ "grad_norm": 0.3754549026489258,
1345
+ "learning_rate": 1.2718446601941747e-05,
1346
+ "loss": 0.175,
1347
+ "step": 955
1348
+ },
1349
+ {
1350
+ "epoch": 4.425187752744079,
1351
+ "grad_norm": 0.556167721748352,
1352
+ "learning_rate": 1.2233009708737865e-05,
1353
+ "loss": 0.1774,
1354
+ "step": 960
1355
+ },
1356
+ {
1357
+ "epoch": 4.448295782784518,
1358
+ "grad_norm": 0.41545048356056213,
1359
+ "learning_rate": 1.1747572815533982e-05,
1360
+ "loss": 0.1846,
1361
+ "step": 965
1362
+ },
1363
+ {
1364
+ "epoch": 4.471403812824956,
1365
+ "grad_norm": 0.4662995934486389,
1366
+ "learning_rate": 1.1262135922330098e-05,
1367
+ "loss": 0.1717,
1368
+ "step": 970
1369
+ },
1370
+ {
1371
+ "epoch": 4.494511842865395,
1372
+ "grad_norm": 0.48765814304351807,
1373
+ "learning_rate": 1.0776699029126214e-05,
1374
+ "loss": 0.1815,
1375
+ "step": 975
1376
+ },
1377
+ {
1378
+ "epoch": 4.5176198729058346,
1379
+ "grad_norm": 0.3982609808444977,
1380
+ "learning_rate": 1.029126213592233e-05,
1381
+ "loss": 0.1777,
1382
+ "step": 980
1383
+ },
1384
+ {
1385
+ "epoch": 4.540727902946274,
1386
+ "grad_norm": 0.4367403984069824,
1387
+ "learning_rate": 9.805825242718447e-06,
1388
+ "loss": 0.1833,
1389
+ "step": 985
1390
+ },
1391
+ {
1392
+ "epoch": 4.563835932986713,
1393
+ "grad_norm": 0.5046346783638,
1394
+ "learning_rate": 9.320388349514565e-06,
1395
+ "loss": 0.1751,
1396
+ "step": 990
1397
+ },
1398
+ {
1399
+ "epoch": 4.586943963027152,
1400
+ "grad_norm": 0.44603046774864197,
1401
+ "learning_rate": 8.834951456310681e-06,
1402
+ "loss": 0.1807,
1403
+ "step": 995
1404
+ },
1405
+ {
1406
+ "epoch": 4.610051993067591,
1407
+ "grad_norm": 0.4673083424568176,
1408
+ "learning_rate": 8.349514563106795e-06,
1409
+ "loss": 0.1766,
1410
+ "step": 1000
1411
+ },
1412
+ {
1413
+ "epoch": 4.63316002310803,
1414
+ "grad_norm": 0.3595871925354004,
1415
+ "learning_rate": 7.864077669902912e-06,
1416
+ "loss": 0.1782,
1417
+ "step": 1005
1418
+ },
1419
+ {
1420
+ "epoch": 4.6562680531484695,
1421
+ "grad_norm": 0.45607128739356995,
1422
+ "learning_rate": 7.378640776699029e-06,
1423
+ "loss": 0.179,
1424
+ "step": 1010
1425
+ },
1426
+ {
1427
+ "epoch": 4.679376083188908,
1428
+ "grad_norm": 0.5103349089622498,
1429
+ "learning_rate": 6.893203883495146e-06,
1430
+ "loss": 0.1836,
1431
+ "step": 1015
1432
+ },
1433
+ {
1434
+ "epoch": 4.702484113229347,
1435
+ "grad_norm": 0.43120843172073364,
1436
+ "learning_rate": 6.407766990291263e-06,
1437
+ "loss": 0.175,
1438
+ "step": 1020
1439
+ },
1440
+ {
1441
+ "epoch": 4.725592143269786,
1442
+ "grad_norm": 0.609349250793457,
1443
+ "learning_rate": 5.9223300970873785e-06,
1444
+ "loss": 0.1725,
1445
+ "step": 1025
1446
+ },
1447
+ {
1448
+ "epoch": 4.748700173310225,
1449
+ "grad_norm": 0.3978511393070221,
1450
+ "learning_rate": 5.436893203883496e-06,
1451
+ "loss": 0.1726,
1452
+ "step": 1030
1453
+ },
1454
+ {
1455
+ "epoch": 4.771808203350664,
1456
+ "grad_norm": 0.49570658802986145,
1457
+ "learning_rate": 4.951456310679612e-06,
1458
+ "loss": 0.1772,
1459
+ "step": 1035
1460
+ },
1461
+ {
1462
+ "epoch": 4.7949162333911035,
1463
+ "grad_norm": 0.5324187278747559,
1464
+ "learning_rate": 4.466019417475728e-06,
1465
+ "loss": 0.1864,
1466
+ "step": 1040
1467
+ },
1468
+ {
1469
+ "epoch": 4.818024263431543,
1470
+ "grad_norm": 0.5425381064414978,
1471
+ "learning_rate": 3.980582524271844e-06,
1472
+ "loss": 0.1801,
1473
+ "step": 1045
1474
+ },
1475
+ {
1476
+ "epoch": 4.841132293471982,
1477
+ "grad_norm": 0.5445485711097717,
1478
+ "learning_rate": 3.4951456310679615e-06,
1479
+ "loss": 0.1787,
1480
+ "step": 1050
1481
+ },
1482
+ {
1483
+ "epoch": 4.864240323512421,
1484
+ "grad_norm": 0.4843500852584839,
1485
+ "learning_rate": 3.0097087378640778e-06,
1486
+ "loss": 0.1741,
1487
+ "step": 1055
1488
+ },
1489
+ {
1490
+ "epoch": 4.887348353552859,
1491
+ "grad_norm": 0.5887071490287781,
1492
+ "learning_rate": 2.524271844660194e-06,
1493
+ "loss": 0.1775,
1494
+ "step": 1060
1495
+ },
1496
+ {
1497
+ "epoch": 4.910456383593298,
1498
+ "grad_norm": 0.4869859218597412,
1499
+ "learning_rate": 2.0388349514563107e-06,
1500
+ "loss": 0.18,
1501
+ "step": 1065
1502
+ },
1503
+ {
1504
+ "epoch": 4.9335644136337375,
1505
+ "grad_norm": 0.4527548849582672,
1506
+ "learning_rate": 1.5533980582524272e-06,
1507
+ "loss": 0.1764,
1508
+ "step": 1070
1509
+ },
1510
+ {
1511
+ "epoch": 4.956672443674177,
1512
+ "grad_norm": 0.5315564274787903,
1513
+ "learning_rate": 1.0679611650485437e-06,
1514
+ "loss": 0.1774,
1515
+ "step": 1075
1516
+ },
1517
+ {
1518
+ "epoch": 4.979780473714616,
1519
+ "grad_norm": 0.38287582993507385,
1520
+ "learning_rate": 5.825242718446603e-07,
1521
+ "loss": 0.1833,
1522
+ "step": 1080
1523
+ },
1524
+ {
1525
+ "epoch": 5.0,
1526
+ "grad_norm": 0.5097792744636536,
1527
+ "learning_rate": 9.70873786407767e-08,
1528
+ "loss": 0.1703,
1529
+ "step": 1085
1530
+ }
1531
+ ],
1532
+ "logging_steps": 5,
1533
+ "max_steps": 1085,
1534
+ "num_input_tokens_seen": 0,
1535
+ "num_train_epochs": 5,
1536
+ "save_steps": 500,
1537
+ "stateful_callbacks": {
1538
+ "TrainerControl": {
1539
+ "args": {
1540
+ "should_epoch_stop": false,
1541
+ "should_evaluate": false,
1542
+ "should_log": false,
1543
+ "should_save": true,
1544
+ "should_training_stop": true
1545
+ },
1546
+ "attributes": {}
1547
+ }
1548
+ },
1549
+ "total_flos": 1.7641081104775512e+19,
1550
+ "train_batch_size": 1,
1551
+ "trial_name": null,
1552
+ "trial_params": null
1553
+ }
Qwen3-4B-Instruct-2507-sft-fusang-swa=2k_sink=0_falayer=7_from0_step8_fadec/checkpoint-1085/vocab.json ADDED
The diff for this file is too large to render. See raw diff
 
Qwen3-4B-Instruct-2507-sft-fusang-swa=2k_sink=10/checkpoint-1760/chat_template.jinja ADDED
@@ -0,0 +1,61 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {%- if tools %}
2
+ {{- '<|im_start|>system\n' }}
3
+ {%- if messages[0].role == 'system' %}
4
+ {{- messages[0].content + '\n\n' }}
5
+ {%- endif %}
6
+ {{- "# Tools\n\nYou may call one or more functions to assist with the user query.\n\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>" }}
7
+ {%- for tool in tools %}
8
+ {{- "\n" }}
9
+ {{- tool | tojson }}
10
+ {%- endfor %}
11
+ {{- "\n</tools>\n\nFor each function call, return a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call><|im_end|>\n" }}
12
+ {%- else %}
13
+ {%- if messages[0].role == 'system' %}
14
+ {{- '<|im_start|>system\n' + messages[0].content + '<|im_end|>\n' }}
15
+ {%- endif %}
16
+ {%- endif %}
17
+ {%- for message in messages %}
18
+ {%- if message.content is string %}
19
+ {%- set content = message.content %}
20
+ {%- else %}
21
+ {%- set content = '' %}
22
+ {%- endif %}
23
+ {%- if (message.role == "user") or (message.role == "system" and not loop.first) %}
24
+ {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
25
+ {%- elif message.role == "assistant" %}
26
+ {{- '<|im_start|>' + message.role + '\n' + content }}
27
+ {%- if message.tool_calls %}
28
+ {%- for tool_call in message.tool_calls %}
29
+ {%- if (loop.first and content) or (not loop.first) %}
30
+ {{- '\n' }}
31
+ {%- endif %}
32
+ {%- if tool_call.function %}
33
+ {%- set tool_call = tool_call.function %}
34
+ {%- endif %}
35
+ {{- '<tool_call>\n{"name": "' }}
36
+ {{- tool_call.name }}
37
+ {{- '", "arguments": ' }}
38
+ {%- if tool_call.arguments is string %}
39
+ {{- tool_call.arguments }}
40
+ {%- else %}
41
+ {{- tool_call.arguments | tojson }}
42
+ {%- endif %}
43
+ {{- '}\n</tool_call>' }}
44
+ {%- endfor %}
45
+ {%- endif %}
46
+ {{- '<|im_end|>\n' }}
47
+ {%- elif message.role == "tool" %}
48
+ {%- if loop.first or (messages[loop.index0 - 1].role != "tool") %}
49
+ {{- '<|im_start|>user' }}
50
+ {%- endif %}
51
+ {{- '\n<tool_response>\n' }}
52
+ {{- content }}
53
+ {{- '\n</tool_response>' }}
54
+ {%- if loop.last or (messages[loop.index0 + 1].role != "tool") %}
55
+ {{- '<|im_end|>\n' }}
56
+ {%- endif %}
57
+ {%- endif %}
58
+ {%- endfor %}
59
+ {%- if add_generation_prompt %}
60
+ {{- '<|im_start|>assistant\n' }}
61
+ {%- endif %}
Qwen3-4B-Instruct-2507-sft-fusang-swa=2k_sink=10/checkpoint-1760/tokenizer.json ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:aeb13307a71acd8fe81861d94ad54ab689df773318809eed3cbe794b4492dae4
3
+ size 11422654
Qwen3-4B-Instruct-2507-sft-fusang-swa=2k_sink=100/checkpoint-1760/tokenizer.json ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:aeb13307a71acd8fe81861d94ad54ab689df773318809eed3cbe794b4492dae4
3
+ size 11422654
Qwen3-4B-Instruct-2507-sft-fusang-swa=2k_sink=100_falayer=25_from1_step2_fadec/checkpoint-921/tokenizer.json ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:aeb13307a71acd8fe81861d94ad54ab689df773318809eed3cbe794b4492dae4
3
+ size 11422654
Qwen3-4B-Instruct-2507-sft-fusang-swa=2k_sink=100_falayer=25_from1_step2_fadec/checkpoint-921/training_args.bin ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:f8c942fe75a0e2c41cf1b38165c0d0153276cef7120b6188a3392537833e8ac0
3
+ size 5969
Qwen3-4B-Instruct-2507-sft-fusang/checkpoint-1760/added_tokens.json ADDED
@@ -0,0 +1,28 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "</think>": 151668,
3
+ "</tool_call>": 151658,
4
+ "</tool_response>": 151666,
5
+ "<think>": 151667,
6
+ "<tool_call>": 151657,
7
+ "<tool_response>": 151665,
8
+ "<|box_end|>": 151649,
9
+ "<|box_start|>": 151648,
10
+ "<|endoftext|>": 151643,
11
+ "<|file_sep|>": 151664,
12
+ "<|fim_middle|>": 151660,
13
+ "<|fim_pad|>": 151662,
14
+ "<|fim_prefix|>": 151659,
15
+ "<|fim_suffix|>": 151661,
16
+ "<|im_end|>": 151645,
17
+ "<|im_start|>": 151644,
18
+ "<|image_pad|>": 151655,
19
+ "<|object_ref_end|>": 151647,
20
+ "<|object_ref_start|>": 151646,
21
+ "<|quad_end|>": 151651,
22
+ "<|quad_start|>": 151650,
23
+ "<|repo_name|>": 151663,
24
+ "<|video_pad|>": 151656,
25
+ "<|vision_end|>": 151653,
26
+ "<|vision_pad|>": 151654,
27
+ "<|vision_start|>": 151652
28
+ }
Qwen3-4B-Instruct-2507-sft-fusang/checkpoint-1760/special_tokens_map.json ADDED
@@ -0,0 +1,31 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "additional_special_tokens": [
3
+ "<|im_start|>",
4
+ "<|im_end|>",
5
+ "<|object_ref_start|>",
6
+ "<|object_ref_end|>",
7
+ "<|box_start|>",
8
+ "<|box_end|>",
9
+ "<|quad_start|>",
10
+ "<|quad_end|>",
11
+ "<|vision_start|>",
12
+ "<|vision_end|>",
13
+ "<|vision_pad|>",
14
+ "<|image_pad|>",
15
+ "<|video_pad|>"
16
+ ],
17
+ "eos_token": {
18
+ "content": "<|im_end|>",
19
+ "lstrip": false,
20
+ "normalized": false,
21
+ "rstrip": false,
22
+ "single_word": false
23
+ },
24
+ "pad_token": {
25
+ "content": "<|endoftext|>",
26
+ "lstrip": false,
27
+ "normalized": false,
28
+ "rstrip": false,
29
+ "single_word": false
30
+ }
31
+ }
Qwen3-4B-Instruct-2507-sft-fusang/checkpoint-1760/tokenizer_config.json ADDED
@@ -0,0 +1,239 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "add_bos_token": false,
3
+ "add_prefix_space": false,
4
+ "added_tokens_decoder": {
5
+ "151643": {
6
+ "content": "<|endoftext|>",
7
+ "lstrip": false,
8
+ "normalized": false,
9
+ "rstrip": false,
10
+ "single_word": false,
11
+ "special": true
12
+ },
13
+ "151644": {
14
+ "content": "<|im_start|>",
15
+ "lstrip": false,
16
+ "normalized": false,
17
+ "rstrip": false,
18
+ "single_word": false,
19
+ "special": true
20
+ },
21
+ "151645": {
22
+ "content": "<|im_end|>",
23
+ "lstrip": false,
24
+ "normalized": false,
25
+ "rstrip": false,
26
+ "single_word": false,
27
+ "special": true
28
+ },
29
+ "151646": {
30
+ "content": "<|object_ref_start|>",
31
+ "lstrip": false,
32
+ "normalized": false,
33
+ "rstrip": false,
34
+ "single_word": false,
35
+ "special": true
36
+ },
37
+ "151647": {
38
+ "content": "<|object_ref_end|>",
39
+ "lstrip": false,
40
+ "normalized": false,
41
+ "rstrip": false,
42
+ "single_word": false,
43
+ "special": true
44
+ },
45
+ "151648": {
46
+ "content": "<|box_start|>",
47
+ "lstrip": false,
48
+ "normalized": false,
49
+ "rstrip": false,
50
+ "single_word": false,
51
+ "special": true
52
+ },
53
+ "151649": {
54
+ "content": "<|box_end|>",
55
+ "lstrip": false,
56
+ "normalized": false,
57
+ "rstrip": false,
58
+ "single_word": false,
59
+ "special": true
60
+ },
61
+ "151650": {
62
+ "content": "<|quad_start|>",
63
+ "lstrip": false,
64
+ "normalized": false,
65
+ "rstrip": false,
66
+ "single_word": false,
67
+ "special": true
68
+ },
69
+ "151651": {
70
+ "content": "<|quad_end|>",
71
+ "lstrip": false,
72
+ "normalized": false,
73
+ "rstrip": false,
74
+ "single_word": false,
75
+ "special": true
76
+ },
77
+ "151652": {
78
+ "content": "<|vision_start|>",
79
+ "lstrip": false,
80
+ "normalized": false,
81
+ "rstrip": false,
82
+ "single_word": false,
83
+ "special": true
84
+ },
85
+ "151653": {
86
+ "content": "<|vision_end|>",
87
+ "lstrip": false,
88
+ "normalized": false,
89
+ "rstrip": false,
90
+ "single_word": false,
91
+ "special": true
92
+ },
93
+ "151654": {
94
+ "content": "<|vision_pad|>",
95
+ "lstrip": false,
96
+ "normalized": false,
97
+ "rstrip": false,
98
+ "single_word": false,
99
+ "special": true
100
+ },
101
+ "151655": {
102
+ "content": "<|image_pad|>",
103
+ "lstrip": false,
104
+ "normalized": false,
105
+ "rstrip": false,
106
+ "single_word": false,
107
+ "special": true
108
+ },
109
+ "151656": {
110
+ "content": "<|video_pad|>",
111
+ "lstrip": false,
112
+ "normalized": false,
113
+ "rstrip": false,
114
+ "single_word": false,
115
+ "special": true
116
+ },
117
+ "151657": {
118
+ "content": "<tool_call>",
119
+ "lstrip": false,
120
+ "normalized": false,
121
+ "rstrip": false,
122
+ "single_word": false,
123
+ "special": false
124
+ },
125
+ "151658": {
126
+ "content": "</tool_call>",
127
+ "lstrip": false,
128
+ "normalized": false,
129
+ "rstrip": false,
130
+ "single_word": false,
131
+ "special": false
132
+ },
133
+ "151659": {
134
+ "content": "<|fim_prefix|>",
135
+ "lstrip": false,
136
+ "normalized": false,
137
+ "rstrip": false,
138
+ "single_word": false,
139
+ "special": false
140
+ },
141
+ "151660": {
142
+ "content": "<|fim_middle|>",
143
+ "lstrip": false,
144
+ "normalized": false,
145
+ "rstrip": false,
146
+ "single_word": false,
147
+ "special": false
148
+ },
149
+ "151661": {
150
+ "content": "<|fim_suffix|>",
151
+ "lstrip": false,
152
+ "normalized": false,
153
+ "rstrip": false,
154
+ "single_word": false,
155
+ "special": false
156
+ },
157
+ "151662": {
158
+ "content": "<|fim_pad|>",
159
+ "lstrip": false,
160
+ "normalized": false,
161
+ "rstrip": false,
162
+ "single_word": false,
163
+ "special": false
164
+ },
165
+ "151663": {
166
+ "content": "<|repo_name|>",
167
+ "lstrip": false,
168
+ "normalized": false,
169
+ "rstrip": false,
170
+ "single_word": false,
171
+ "special": false
172
+ },
173
+ "151664": {
174
+ "content": "<|file_sep|>",
175
+ "lstrip": false,
176
+ "normalized": false,
177
+ "rstrip": false,
178
+ "single_word": false,
179
+ "special": false
180
+ },
181
+ "151665": {
182
+ "content": "<tool_response>",
183
+ "lstrip": false,
184
+ "normalized": false,
185
+ "rstrip": false,
186
+ "single_word": false,
187
+ "special": false
188
+ },
189
+ "151666": {
190
+ "content": "</tool_response>",
191
+ "lstrip": false,
192
+ "normalized": false,
193
+ "rstrip": false,
194
+ "single_word": false,
195
+ "special": false
196
+ },
197
+ "151667": {
198
+ "content": "<think>",
199
+ "lstrip": false,
200
+ "normalized": false,
201
+ "rstrip": false,
202
+ "single_word": false,
203
+ "special": false
204
+ },
205
+ "151668": {
206
+ "content": "</think>",
207
+ "lstrip": false,
208
+ "normalized": false,
209
+ "rstrip": false,
210
+ "single_word": false,
211
+ "special": false
212
+ }
213
+ },
214
+ "additional_special_tokens": [
215
+ "<|im_start|>",
216
+ "<|im_end|>",
217
+ "<|object_ref_start|>",
218
+ "<|object_ref_end|>",
219
+ "<|box_start|>",
220
+ "<|box_end|>",
221
+ "<|quad_start|>",
222
+ "<|quad_end|>",
223
+ "<|vision_start|>",
224
+ "<|vision_end|>",
225
+ "<|vision_pad|>",
226
+ "<|image_pad|>",
227
+ "<|video_pad|>"
228
+ ],
229
+ "bos_token": null,
230
+ "clean_up_tokenization_spaces": false,
231
+ "eos_token": "<|im_end|>",
232
+ "errors": "replace",
233
+ "extra_special_tokens": {},
234
+ "model_max_length": 1010000,
235
+ "pad_token": "<|endoftext|>",
236
+ "split_special_tokens": false,
237
+ "tokenizer_class": "Qwen2Tokenizer",
238
+ "unk_token": null
239
+ }