Hugging Face's logo

flair
/
bueble-lm-2b

Text Generation
Transformers
Safetensors
German
gemma2
german
causal-lm
text-generation-inference
Inference Endpoints
Model card Files Community
1
pdelobelle commited on
Commit
200bad4
·
verified ·
1 Parent(s): e819242

Training in progress, step 400, checkpoint

Browse files
Files changed (14) hide show
  1. checkpoint-400/config.json +33 -0
  2. checkpoint-400/generation_config.json +11 -0
  3. checkpoint-400/model.safetensors +3 -0
  4. checkpoint-400/optimizer.pt +3 -0
  5. checkpoint-400/rng_state_0.pth +3 -0
  6. checkpoint-400/rng_state_1.pth +3 -0
  7. checkpoint-400/rng_state_2.pth +3 -0
  8. checkpoint-400/rng_state_3.pth +3 -0
  9. checkpoint-400/scheduler.pt +3 -0
  10. checkpoint-400/special_tokens_map.json +24 -0
  11. checkpoint-400/tokenizer.model +3 -0
  12. checkpoint-400/tokenizer_config.json +322 -0
  13. checkpoint-400/trainer_state.json +3269 -0
  14. checkpoint-400/training_args.bin +3 -0
checkpoint-400/config.json ADDED
@@ -0,0 +1,33 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "_name_or_path": "pdelobelle/gemma-2-2b-de",
3
+ "architectures": [
4
+ "Gemma2ForCausalLM"
5
+ ],
6
+ "attention_bias": false,
7
+ "attention_dropout": 0.0,
8
+ "attn_logit_softcapping": 50.0,
9
+ "bos_token_id": 2,
10
+ "cache_implementation": "hybrid",
11
+ "eos_token_id": 2,
12
+ "final_logit_softcapping": 30.0,
13
+ "head_dim": 256,
14
+ "hidden_act": "gelu_pytorch_tanh",
15
+ "hidden_activation": "gelu_pytorch_tanh",
16
+ "hidden_size": 2304,
17
+ "initializer_range": 0.02,
18
+ "intermediate_size": 9216,
19
+ "max_position_embeddings": 8192,
20
+ "model_type": "gemma2",
21
+ "num_attention_heads": 8,
22
+ "num_hidden_layers": 26,
23
+ "num_key_value_heads": 4,
24
+ "pad_token_id": 3,
25
+ "query_pre_attn_scalar": 256,
26
+ "rms_norm_eps": 1e-06,
27
+ "rope_theta": 10000.0,
28
+ "sliding_window": 4096,
29
+ "torch_dtype": "bfloat16",
30
+ "transformers_version": "4.44.1",
31
+ "use_cache": true,
32
+ "vocab_size": 20000
33
+ }
checkpoint-400/generation_config.json ADDED
@@ -0,0 +1,11 @@
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "_from_model_config": true,
3
+ "bos_token_id": 2,
4
+ "cache_implementation": "hybrid",
5
+ "eos_token_id": [
6
+ 1,
7
+ 107
8
+ ],
9
+ "pad_token_id": 0,
10
+ "transformers_version": "4.44.1"
11
+ }
checkpoint-400/model.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:ddac9f6ef270ea099a6eae929a99e6b01fd2fcb85c764b9fcbd9e2819e56974a
3
+ size 4141229384
checkpoint-400/optimizer.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:8df97e9bc59e6e5508abc758cb14f429cbe1750bfa40ff6aca822d98a027952c
3
+ size 4151012602
checkpoint-400/rng_state_0.pth ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:b17fafdd35d8ad0eb80e4ccba707f48c96039f1c88132431337868342514a0fd
3
+ size 14960
checkpoint-400/rng_state_1.pth ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:132a8be8ab8be442151d175a7ce8a2a006745bb79e10296155f54f454806b382
3
+ size 14960
checkpoint-400/rng_state_2.pth ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:acbf026b7fe6ec5094c1315c0c97d9c1cf14fe060ebebca6a2edc535ebf601ee
3
+ size 14960
checkpoint-400/rng_state_3.pth ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:6340f8982774627ee32b8f6d4076ae5b2fbd62885b292f0b7eb26b63fd57bd06
3
+ size 14960
checkpoint-400/scheduler.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:ddd98d2e423d714b02679d03ad4c6b2f18ee3aabf6f5e702c65a106e036e1eea
3
+ size 1064
checkpoint-400/special_tokens_map.json ADDED
@@ -0,0 +1,24 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "bos_token": {
3
+ "content": "<s>",
4
+ "lstrip": false,
5
+ "normalized": false,
6
+ "rstrip": false,
7
+ "single_word": false
8
+ },
9
+ "eos_token": {
10
+ "content": "</s>",
11
+ "lstrip": false,
12
+ "normalized": false,
13
+ "rstrip": false,
14
+ "single_word": false
15
+ },
16
+ "pad_token": "</s>",
17
+ "unk_token": {
18
+ "content": "<unk>",
19
+ "lstrip": false,
20
+ "normalized": false,
21
+ "rstrip": false,
22
+ "single_word": false
23
+ }
24
+ }
checkpoint-400/tokenizer.model ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:16e8773affbd03448ffb79173feed1884514012160d3074641ee402dcad4f481
3
+ size 579378
checkpoint-400/tokenizer_config.json ADDED
@@ -0,0 +1,322 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "add_bos_token": true,
3
+ "add_eos_token": false,
4
+ "add_prefix_space": null,
5
+ "added_tokens_decoder": {
6
+ "0": {
7
+ "content": "<unk>",
8
+ "lstrip": false,
9
+ "normalized": false,
10
+ "rstrip": false,
11
+ "single_word": false,
12
+ "special": true
13
+ },
14
+ "1": {
15
+ "content": "<s>",
16
+ "lstrip": false,
17
+ "normalized": false,
18
+ "rstrip": false,
19
+ "single_word": false,
20
+ "special": true
21
+ },
22
+ "2": {
23
+ "content": "</s>",
24
+ "lstrip": false,
25
+ "normalized": false,
26
+ "rstrip": false,
27
+ "single_word": false,
28
+ "special": true
29
+ },
30
+ "3": {
31
+ "content": "<pad>",
32
+ "lstrip": false,
33
+ "normalized": false,
34
+ "rstrip": false,
35
+ "single_word": false,
36
+ "special": true
37
+ },
38
+ "4": {
39
+ "content": "▁▁",
40
+ "lstrip": false,
41
+ "normalized": false,
42
+ "rstrip": false,
43
+ "single_word": false,
44
+ "special": false
45
+ },
46
+ "5": {
47
+ "content": "▁▁▁",
48
+ "lstrip": false,
49
+ "normalized": false,
50
+ "rstrip": false,
51
+ "single_word": false,
52
+ "special": false
53
+ },
54
+ "6": {
55
+ "content": "▁▁▁▁",
56
+ "lstrip": false,
57
+ "normalized": false,
58
+ "rstrip": false,
59
+ "single_word": false,
60
+ "special": false
61
+ },
62
+ "7": {
63
+ "content": "▁▁▁▁▁▁▁▁",
64
+ "lstrip": false,
65
+ "normalized": false,
66
+ "rstrip": false,
67
+ "single_word": false,
68
+ "special": false
69
+ },
70
+ "8": {
71
+ "content": "▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁",
72
+ "lstrip": false,
73
+ "normalized": false,
74
+ "rstrip": false,
75
+ "single_word": false,
76
+ "special": false
77
+ },
78
+ "9": {
79
+ "content": "--",
80
+ "lstrip": false,
81
+ "normalized": false,
82
+ "rstrip": false,
83
+ "single_word": false,
84
+ "special": false
85
+ },
86
+ "10": {
87
+ "content": "----",
88
+ "lstrip": false,
89
+ "normalized": false,
90
+ "rstrip": false,
91
+ "single_word": false,
92
+ "special": false
93
+ },
94
+ "11": {
95
+ "content": "-----",
96
+ "lstrip": false,
97
+ "normalized": false,
98
+ "rstrip": false,
99
+ "single_word": false,
100
+ "special": false
101
+ },
102
+ "12": {
103
+ "content": "--------",
104
+ "lstrip": false,
105
+ "normalized": false,
106
+ "rstrip": false,
107
+ "single_word": false,
108
+ "special": false
109
+ },
110
+ "13": {
111
+ "content": "----------------",
112
+ "lstrip": false,
113
+ "normalized": false,
114
+ "rstrip": false,
115
+ "single_word": false,
116
+ "special": false
117
+ },
118
+ "14": {
119
+ "content": "++",
120
+ "lstrip": false,
121
+ "normalized": false,
122
+ "rstrip": false,
123
+ "single_word": false,
124
+ "special": false
125
+ },
126
+ "15": {
127
+ "content": "/**",
128
+ "lstrip": false,
129
+ "normalized": false,
130
+ "rstrip": false,
131
+ "single_word": false,
132
+ "special": false
133
+ },
134
+ "16": {
135
+ "content": "***",
136
+ "lstrip": false,
137
+ "normalized": false,
138
+ "rstrip": false,
139
+ "single_word": false,
140
+ "special": false
141
+ },
142
+ "17": {
143
+ "content": "****",
144
+ "lstrip": false,
145
+ "normalized": false,
146
+ "rstrip": false,
147
+ "single_word": false,
148
+ "special": false
149
+ },
150
+ "18": {
151
+ "content": "******",
152
+ "lstrip": false,
153
+ "normalized": false,
154
+ "rstrip": false,
155
+ "single_word": false,
156
+ "special": false
157
+ },
158
+ "19": {
159
+ "content": "********",
160
+ "lstrip": false,
161
+ "normalized": false,
162
+ "rstrip": false,
163
+ "single_word": false,
164
+ "special": false
165
+ },
166
+ "20": {
167
+ "content": "**/",
168
+ "lstrip": false,
169
+ "normalized": false,
170
+ "rstrip": false,
171
+ "single_word": false,
172
+ "special": false
173
+ },
174
+ "21": {
175
+ "content": "##",
176
+ "lstrip": false,
177
+ "normalized": false,
178
+ "rstrip": false,
179
+ "single_word": false,
180
+ "special": false
181
+ },
182
+ "22": {
183
+ "content": "###",
184
+ "lstrip": false,
185
+ "normalized": false,
186
+ "rstrip": false,
187
+ "single_word": false,
188
+ "special": false
189
+ },
190
+ "23": {
191
+ "content": "<|im_start|>",
192
+ "lstrip": false,
193
+ "normalized": false,
194
+ "rstrip": false,
195
+ "single_word": false,
196
+ "special": false
197
+ },
198
+ "24": {
199
+ "content": "<|im_end|>",
200
+ "lstrip": false,
201
+ "normalized": false,
202
+ "rstrip": false,
203
+ "single_word": false,
204
+ "special": false
205
+ },
206
+ "25": {
207
+ "content": "<|system|>",
208
+ "lstrip": false,
209
+ "normalized": false,
210
+ "rstrip": false,
211
+ "single_word": false,
212
+ "special": false
213
+ },
214
+ "26": {
215
+ "content": "<|user|>",
216
+ "lstrip": false,
217
+ "normalized": false,
218
+ "rstrip": false,
219
+ "single_word": false,
220
+ "special": false
221
+ },
222
+ "27": {
223
+ "content": "<|assistant|>",
224
+ "lstrip": false,
225
+ "normalized": false,
226
+ "rstrip": false,
227
+ "single_word": false,
228
+ "special": false
229
+ },
230
+ "28": {
231
+ "content": "▁—",
232
+ "lstrip": false,
233
+ "normalized": false,
234
+ "rstrip": false,
235
+ "single_word": false,
236
+ "special": false
237
+ },
238
+ "29": {
239
+ "content": "▁“",
240
+ "lstrip": false,
241
+ "normalized": false,
242
+ "rstrip": false,
243
+ "single_word": false,
244
+ "special": false
245
+ },
246
+ "30": {
247
+ "content": "“",
248
+ "lstrip": false,
249
+ "normalized": false,
250
+ "rstrip": false,
251
+ "single_word": false,
252
+ "special": false
253
+ },
254
+ "31": {
255
+ "content": "”",
256
+ "lstrip": false,
257
+ "normalized": false,
258
+ "rstrip": false,
259
+ "single_word": false,
260
+ "special": false
261
+ },
262
+ "32": {
263
+ "content": "’",
264
+ "lstrip": false,
265
+ "normalized": false,
266
+ "rstrip": false,
267
+ "single_word": false,
268
+ "special": false
269
+ },
270
+ "33": {
271
+ "content": "—",
272
+ "lstrip": false,
273
+ "normalized": false,
274
+ "rstrip": false,
275
+ "single_word": false,
276
+ "special": false
277
+ },
278
+ "34": {
279
+ "content": "{",
280
+ "lstrip": false,
281
+ "normalized": false,
282
+ "rstrip": false,
283
+ "single_word": false,
284
+ "special": false
285
+ },
286
+ "35": {
287
+ "content": "}\"",
288
+ "lstrip": false,
289
+ "normalized": false,
290
+ "rstrip": false,
291
+ "single_word": false,
292
+ "special": false
293
+ },
294
+ "36": {
295
+ "content": "{\"",
296
+ "lstrip": false,
297
+ "normalized": false,
298
+ "rstrip": false,
299
+ "single_word": false,
300
+ "special": false
301
+ },
302
+ "37": {
303
+ "content": "}",
304
+ "lstrip": false,
305
+ "normalized": false,
306
+ "rstrip": false,
307
+ "single_word": false,
308
+ "special": false
309
+ }
310
+ },
311
+ "bos_token": "<s>",
312
+ "clean_up_tokenization_spaces": false,
313
+ "eos_token": "</s>",
314
+ "legacy": true,
315
+ "model_max_length": 1000000000000000019884624838656,
316
+ "pad_token": "</s>",
317
+ "sp_model_kwargs": {},
318
+ "spaces_between_special_tokens": false,
319
+ "tokenizer_class": "LlamaTokenizer",
320
+ "unk_token": "<unk>",
321
+ "use_default_system_prompt": false
322
+ }
checkpoint-400/trainer_state.json ADDED
@@ -0,0 +1,3269 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "best_metric": null,
3
+ "best_model_checkpoint": null,
4
+ "epoch": 0.04718650465966733,
5
+ "eval_steps": 200,
6
+ "global_step": 400,
7
+ "is_hyper_param_search": false,
8
+ "is_local_process_zero": true,
9
+ "is_world_process_zero": true,
10
+ "log_history": [
11
+ {
12
+ "epoch": 0.00011796626164916834,
13
+ "grad_norm": 93.0,
14
+ "learning_rate": 1.8912529550827425e-07,
15
+ "loss": 7.9641,
16
+ "num_input_tokens_seen": 1179648,
17
+ "step": 1
18
+ },
19
+ {
20
+ "epoch": 0.0002359325232983367,
21
+ "grad_norm": 95.0,
22
+ "learning_rate": 3.782505910165485e-07,
23
+ "loss": 7.9866,
24
+ "num_input_tokens_seen": 2359296,
25
+ "step": 2
26
+ },
27
+ {
28
+ "epoch": 0.000353898784947505,
29
+ "grad_norm": 97.5,
30
+ "learning_rate": 5.673758865248227e-07,
31
+ "loss": 7.983,
32
+ "num_input_tokens_seen": 3538944,
33
+ "step": 3
34
+ },
35
+ {
36
+ "epoch": 0.0004718650465966734,
37
+ "grad_norm": 92.0,
38
+ "learning_rate": 7.56501182033097e-07,
39
+ "loss": 7.9018,
40
+ "num_input_tokens_seen": 4718592,
41
+ "step": 4
42
+ },
43
+ {
44
+ "epoch": 0.0005898313082458417,
45
+ "grad_norm": 102.5,
46
+ "learning_rate": 9.456264775413712e-07,
47
+ "loss": 8.0784,
48
+ "num_input_tokens_seen": 5898240,
49
+ "step": 5
50
+ },
51
+ {
52
+ "epoch": 0.00070779756989501,
53
+ "grad_norm": 95.0,
54
+ "learning_rate": 1.1347517730496454e-06,
55
+ "loss": 7.9578,
56
+ "num_input_tokens_seen": 7077888,
57
+ "step": 6
58
+ },
59
+ {
60
+ "epoch": 0.0008257638315441783,
61
+ "grad_norm": 93.5,
62
+ "learning_rate": 1.3238770685579196e-06,
63
+ "loss": 7.975,
64
+ "num_input_tokens_seen": 8257536,
65
+ "step": 7
66
+ },
67
+ {
68
+ "epoch": 0.0009437300931933467,
69
+ "grad_norm": 85.0,
70
+ "learning_rate": 1.513002364066194e-06,
71
+ "loss": 7.8405,
72
+ "num_input_tokens_seen": 9437184,
73
+ "step": 8
74
+ },
75
+ {
76
+ "epoch": 0.001061696354842515,
77
+ "grad_norm": 83.0,
78
+ "learning_rate": 1.7021276595744682e-06,
79
+ "loss": 7.841,
80
+ "num_input_tokens_seen": 10616832,
81
+ "step": 9
82
+ },
83
+ {
84
+ "epoch": 0.0011796626164916834,
85
+ "grad_norm": 82.5,
86
+ "learning_rate": 1.8912529550827423e-06,
87
+ "loss": 7.9491,
88
+ "num_input_tokens_seen": 11796480,
89
+ "step": 10
90
+ },
91
+ {
92
+ "epoch": 0.0012976288781408518,
93
+ "grad_norm": 68.0,
94
+ "learning_rate": 2.0803782505910165e-06,
95
+ "loss": 7.5612,
96
+ "num_input_tokens_seen": 12976128,
97
+ "step": 11
98
+ },
99
+ {
100
+ "epoch": 0.00141559513979002,
101
+ "grad_norm": 66.5,
102
+ "learning_rate": 2.269503546099291e-06,
103
+ "loss": 7.5808,
104
+ "num_input_tokens_seen": 14155776,
105
+ "step": 12
106
+ },
107
+ {
108
+ "epoch": 0.0015335614014391884,
109
+ "grad_norm": 57.5,
110
+ "learning_rate": 2.4586288416075653e-06,
111
+ "loss": 7.4494,
112
+ "num_input_tokens_seen": 15335424,
113
+ "step": 13
114
+ },
115
+ {
116
+ "epoch": 0.0016515276630883566,
117
+ "grad_norm": 51.25,
118
+ "learning_rate": 2.6477541371158392e-06,
119
+ "loss": 7.4606,
120
+ "num_input_tokens_seen": 16515072,
121
+ "step": 14
122
+ },
123
+ {
124
+ "epoch": 0.001769493924737525,
125
+ "grad_norm": 42.75,
126
+ "learning_rate": 2.836879432624114e-06,
127
+ "loss": 7.2804,
128
+ "num_input_tokens_seen": 17694720,
129
+ "step": 15
130
+ },
131
+ {
132
+ "epoch": 0.0018874601863866935,
133
+ "grad_norm": 38.0,
134
+ "learning_rate": 3.026004728132388e-06,
135
+ "loss": 7.2173,
136
+ "num_input_tokens_seen": 18874368,
137
+ "step": 16
138
+ },
139
+ {
140
+ "epoch": 0.0020054264480358617,
141
+ "grad_norm": 33.75,
142
+ "learning_rate": 3.2151300236406624e-06,
143
+ "loss": 7.0934,
144
+ "num_input_tokens_seen": 20054016,
145
+ "step": 17
146
+ },
147
+ {
148
+ "epoch": 0.00212339270968503,
149
+ "grad_norm": 30.5,
150
+ "learning_rate": 3.4042553191489363e-06,
151
+ "loss": 7.1284,
152
+ "num_input_tokens_seen": 21233664,
153
+ "step": 18
154
+ },
155
+ {
156
+ "epoch": 0.0022413589713341986,
157
+ "grad_norm": 26.5,
158
+ "learning_rate": 3.5933806146572107e-06,
159
+ "loss": 6.932,
160
+ "num_input_tokens_seen": 22413312,
161
+ "step": 19
162
+ },
163
+ {
164
+ "epoch": 0.0023593252329833668,
165
+ "grad_norm": 23.875,
166
+ "learning_rate": 3.7825059101654847e-06,
167
+ "loss": 6.9163,
168
+ "num_input_tokens_seen": 23592960,
169
+ "step": 20
170
+ },
171
+ {
172
+ "epoch": 0.002477291494632535,
173
+ "grad_norm": 21.625,
174
+ "learning_rate": 3.9716312056737595e-06,
175
+ "loss": 6.8335,
176
+ "num_input_tokens_seen": 24772608,
177
+ "step": 21
178
+ },
179
+ {
180
+ "epoch": 0.0025952577562817036,
181
+ "grad_norm": 18.625,
182
+ "learning_rate": 4.160756501182033e-06,
183
+ "loss": 6.6761,
184
+ "num_input_tokens_seen": 25952256,
185
+ "step": 22
186
+ },
187
+ {
188
+ "epoch": 0.002713224017930872,
189
+ "grad_norm": 16.0,
190
+ "learning_rate": 4.349881796690308e-06,
191
+ "loss": 6.6474,
192
+ "num_input_tokens_seen": 27131904,
193
+ "step": 23
194
+ },
195
+ {
196
+ "epoch": 0.00283119027958004,
197
+ "grad_norm": 18.0,
198
+ "learning_rate": 4.539007092198582e-06,
199
+ "loss": 6.6062,
200
+ "num_input_tokens_seen": 28311552,
201
+ "step": 24
202
+ },
203
+ {
204
+ "epoch": 0.0029491565412292082,
205
+ "grad_norm": 17.5,
206
+ "learning_rate": 4.728132387706856e-06,
207
+ "loss": 6.6435,
208
+ "num_input_tokens_seen": 29491200,
209
+ "step": 25
210
+ },
211
+ {
212
+ "epoch": 0.003067122802878377,
213
+ "grad_norm": 14.25,
214
+ "learning_rate": 4.9172576832151305e-06,
215
+ "loss": 6.57,
216
+ "num_input_tokens_seen": 30670848,
217
+ "step": 26
218
+ },
219
+ {
220
+ "epoch": 0.003185089064527545,
221
+ "grad_norm": 13.3125,
222
+ "learning_rate": 5.106382978723404e-06,
223
+ "loss": 6.4037,
224
+ "num_input_tokens_seen": 31850496,
225
+ "step": 27
226
+ },
227
+ {
228
+ "epoch": 0.0033030553261767133,
229
+ "grad_norm": 10.8125,
230
+ "learning_rate": 5.2955082742316784e-06,
231
+ "loss": 6.2927,
232
+ "num_input_tokens_seen": 33030144,
233
+ "step": 28
234
+ },
235
+ {
236
+ "epoch": 0.003421021587825882,
237
+ "grad_norm": 9.75,
238
+ "learning_rate": 5.484633569739954e-06,
239
+ "loss": 6.2084,
240
+ "num_input_tokens_seen": 34209792,
241
+ "step": 29
242
+ },
243
+ {
244
+ "epoch": 0.00353898784947505,
245
+ "grad_norm": 10.25,
246
+ "learning_rate": 5.673758865248228e-06,
247
+ "loss": 6.2349,
248
+ "num_input_tokens_seen": 35389440,
249
+ "step": 30
250
+ },
251
+ {
252
+ "epoch": 0.0036569541111242184,
253
+ "grad_norm": 10.8125,
254
+ "learning_rate": 5.862884160756502e-06,
255
+ "loss": 6.155,
256
+ "num_input_tokens_seen": 36569088,
257
+ "step": 31
258
+ },
259
+ {
260
+ "epoch": 0.003774920372773387,
261
+ "grad_norm": 10.0,
262
+ "learning_rate": 6.052009456264776e-06,
263
+ "loss": 6.1846,
264
+ "num_input_tokens_seen": 37748736,
265
+ "step": 32
266
+ },
267
+ {
268
+ "epoch": 0.003892886634422555,
269
+ "grad_norm": 8.0,
270
+ "learning_rate": 6.24113475177305e-06,
271
+ "loss": 6.0656,
272
+ "num_input_tokens_seen": 38928384,
273
+ "step": 33
274
+ },
275
+ {
276
+ "epoch": 0.004010852896071723,
277
+ "grad_norm": 8.0625,
278
+ "learning_rate": 6.430260047281325e-06,
279
+ "loss": 5.9873,
280
+ "num_input_tokens_seen": 40108032,
281
+ "step": 34
282
+ },
283
+ {
284
+ "epoch": 0.004128819157720892,
285
+ "grad_norm": 8.875,
286
+ "learning_rate": 6.619385342789598e-06,
287
+ "loss": 6.0008,
288
+ "num_input_tokens_seen": 41287680,
289
+ "step": 35
290
+ },
291
+ {
292
+ "epoch": 0.00424678541937006,
293
+ "grad_norm": 7.34375,
294
+ "learning_rate": 6.808510638297873e-06,
295
+ "loss": 6.0852,
296
+ "num_input_tokens_seen": 42467328,
297
+ "step": 36
298
+ },
299
+ {
300
+ "epoch": 0.004364751681019229,
301
+ "grad_norm": 6.28125,
302
+ "learning_rate": 6.997635933806147e-06,
303
+ "loss": 5.8264,
304
+ "num_input_tokens_seen": 43646976,
305
+ "step": 37
306
+ },
307
+ {
308
+ "epoch": 0.004482717942668397,
309
+ "grad_norm": 6.75,
310
+ "learning_rate": 7.186761229314421e-06,
311
+ "loss": 5.9511,
312
+ "num_input_tokens_seen": 44826624,
313
+ "step": 38
314
+ },
315
+ {
316
+ "epoch": 0.004600684204317565,
317
+ "grad_norm": 6.96875,
318
+ "learning_rate": 7.375886524822695e-06,
319
+ "loss": 5.8042,
320
+ "num_input_tokens_seen": 46006272,
321
+ "step": 39
322
+ },
323
+ {
324
+ "epoch": 0.0047186504659667335,
325
+ "grad_norm": 5.59375,
326
+ "learning_rate": 7.565011820330969e-06,
327
+ "loss": 5.8955,
328
+ "num_input_tokens_seen": 47185920,
329
+ "step": 40
330
+ },
331
+ {
332
+ "epoch": 0.004836616727615902,
333
+ "grad_norm": 5.3125,
334
+ "learning_rate": 7.754137115839244e-06,
335
+ "loss": 5.7764,
336
+ "num_input_tokens_seen": 48365568,
337
+ "step": 41
338
+ },
339
+ {
340
+ "epoch": 0.00495458298926507,
341
+ "grad_norm": 5.40625,
342
+ "learning_rate": 7.943262411347519e-06,
343
+ "loss": 5.7784,
344
+ "num_input_tokens_seen": 49545216,
345
+ "step": 42
346
+ },
347
+ {
348
+ "epoch": 0.005072549250914238,
349
+ "grad_norm": 5.03125,
350
+ "learning_rate": 8.132387706855792e-06,
351
+ "loss": 5.7388,
352
+ "num_input_tokens_seen": 50724864,
353
+ "step": 43
354
+ },
355
+ {
356
+ "epoch": 0.005190515512563407,
357
+ "grad_norm": 4.84375,
358
+ "learning_rate": 8.321513002364066e-06,
359
+ "loss": 5.7821,
360
+ "num_input_tokens_seen": 51904512,
361
+ "step": 44
362
+ },
363
+ {
364
+ "epoch": 0.0053084817742125754,
365
+ "grad_norm": 4.78125,
366
+ "learning_rate": 8.510638297872341e-06,
367
+ "loss": 5.6466,
368
+ "num_input_tokens_seen": 53084160,
369
+ "step": 45
370
+ },
371
+ {
372
+ "epoch": 0.005426448035861744,
373
+ "grad_norm": 4.125,
374
+ "learning_rate": 8.699763593380616e-06,
375
+ "loss": 5.6224,
376
+ "num_input_tokens_seen": 54263808,
377
+ "step": 46
378
+ },
379
+ {
380
+ "epoch": 0.005544414297510912,
381
+ "grad_norm": 5.0,
382
+ "learning_rate": 8.888888888888888e-06,
383
+ "loss": 5.9019,
384
+ "num_input_tokens_seen": 55443456,
385
+ "step": 47
386
+ },
387
+ {
388
+ "epoch": 0.00566238055916008,
389
+ "grad_norm": 4.0,
390
+ "learning_rate": 9.078014184397164e-06,
391
+ "loss": 5.5812,
392
+ "num_input_tokens_seen": 56623104,
393
+ "step": 48
394
+ },
395
+ {
396
+ "epoch": 0.005780346820809248,
397
+ "grad_norm": 3.796875,
398
+ "learning_rate": 9.267139479905439e-06,
399
+ "loss": 5.633,
400
+ "num_input_tokens_seen": 57802752,
401
+ "step": 49
402
+ },
403
+ {
404
+ "epoch": 0.0058983130824584165,
405
+ "grad_norm": 3.84375,
406
+ "learning_rate": 9.456264775413712e-06,
407
+ "loss": 5.5031,
408
+ "num_input_tokens_seen": 58982400,
409
+ "step": 50
410
+ },
411
+ {
412
+ "epoch": 0.0060162793441075856,
413
+ "grad_norm": 3.4375,
414
+ "learning_rate": 9.645390070921986e-06,
415
+ "loss": 5.5146,
416
+ "num_input_tokens_seen": 60162048,
417
+ "step": 51
418
+ },
419
+ {
420
+ "epoch": 0.006134245605756754,
421
+ "grad_norm": 3.609375,
422
+ "learning_rate": 9.834515366430261e-06,
423
+ "loss": 5.3805,
424
+ "num_input_tokens_seen": 61341696,
425
+ "step": 52
426
+ },
427
+ {
428
+ "epoch": 0.006252211867405922,
429
+ "grad_norm": 3.234375,
430
+ "learning_rate": 1.0023640661938535e-05,
431
+ "loss": 5.4098,
432
+ "num_input_tokens_seen": 62521344,
433
+ "step": 53
434
+ },
435
+ {
436
+ "epoch": 0.00637017812905509,
437
+ "grad_norm": 3.40625,
438
+ "learning_rate": 1.0212765957446808e-05,
439
+ "loss": 5.3773,
440
+ "num_input_tokens_seen": 63700992,
441
+ "step": 54
442
+ },
443
+ {
444
+ "epoch": 0.006488144390704258,
445
+ "grad_norm": 3.1875,
446
+ "learning_rate": 1.0401891252955083e-05,
447
+ "loss": 5.3287,
448
+ "num_input_tokens_seen": 64880640,
449
+ "step": 55
450
+ },
451
+ {
452
+ "epoch": 0.006606110652353427,
453
+ "grad_norm": 3.046875,
454
+ "learning_rate": 1.0591016548463357e-05,
455
+ "loss": 5.2282,
456
+ "num_input_tokens_seen": 66060288,
457
+ "step": 56
458
+ },
459
+ {
460
+ "epoch": 0.006724076914002596,
461
+ "grad_norm": 3.0625,
462
+ "learning_rate": 1.0780141843971632e-05,
463
+ "loss": 5.2967,
464
+ "num_input_tokens_seen": 67239936,
465
+ "step": 57
466
+ },
467
+ {
468
+ "epoch": 0.006842043175651764,
469
+ "grad_norm": 2.859375,
470
+ "learning_rate": 1.0969267139479907e-05,
471
+ "loss": 5.1126,
472
+ "num_input_tokens_seen": 68419584,
473
+ "step": 58
474
+ },
475
+ {
476
+ "epoch": 0.006960009437300932,
477
+ "grad_norm": 3.0625,
478
+ "learning_rate": 1.1158392434988181e-05,
479
+ "loss": 5.3309,
480
+ "num_input_tokens_seen": 69599232,
481
+ "step": 59
482
+ },
483
+ {
484
+ "epoch": 0.0070779756989501,
485
+ "grad_norm": 3.0625,
486
+ "learning_rate": 1.1347517730496456e-05,
487
+ "loss": 5.2134,
488
+ "num_input_tokens_seen": 70778880,
489
+ "step": 60
490
+ },
491
+ {
492
+ "epoch": 0.0071959419605992685,
493
+ "grad_norm": 3.546875,
494
+ "learning_rate": 1.153664302600473e-05,
495
+ "loss": 5.1972,
496
+ "num_input_tokens_seen": 71958528,
497
+ "step": 61
498
+ },
499
+ {
500
+ "epoch": 0.007313908222248437,
501
+ "grad_norm": 2.828125,
502
+ "learning_rate": 1.1725768321513003e-05,
503
+ "loss": 5.0123,
504
+ "num_input_tokens_seen": 73138176,
505
+ "step": 62
506
+ },
507
+ {
508
+ "epoch": 0.007431874483897605,
509
+ "grad_norm": 3.203125,
510
+ "learning_rate": 1.1914893617021277e-05,
511
+ "loss": 5.1108,
512
+ "num_input_tokens_seen": 74317824,
513
+ "step": 63
514
+ },
515
+ {
516
+ "epoch": 0.007549840745546774,
517
+ "grad_norm": 3.609375,
518
+ "learning_rate": 1.2104018912529552e-05,
519
+ "loss": 5.2287,
520
+ "num_input_tokens_seen": 75497472,
521
+ "step": 64
522
+ },
523
+ {
524
+ "epoch": 0.007667807007195942,
525
+ "grad_norm": 2.703125,
526
+ "learning_rate": 1.2293144208037825e-05,
527
+ "loss": 5.0902,
528
+ "num_input_tokens_seen": 76677120,
529
+ "step": 65
530
+ },
531
+ {
532
+ "epoch": 0.00778577326884511,
533
+ "grad_norm": 3.1875,
534
+ "learning_rate": 1.24822695035461e-05,
535
+ "loss": 5.1526,
536
+ "num_input_tokens_seen": 77856768,
537
+ "step": 66
538
+ },
539
+ {
540
+ "epoch": 0.007903739530494279,
541
+ "grad_norm": 3.359375,
542
+ "learning_rate": 1.2671394799054376e-05,
543
+ "loss": 4.8855,
544
+ "num_input_tokens_seen": 79036416,
545
+ "step": 67
546
+ },
547
+ {
548
+ "epoch": 0.008021705792143447,
549
+ "grad_norm": 2.765625,
550
+ "learning_rate": 1.286052009456265e-05,
551
+ "loss": 5.0226,
552
+ "num_input_tokens_seen": 80216064,
553
+ "step": 68
554
+ },
555
+ {
556
+ "epoch": 0.008139672053792615,
557
+ "grad_norm": 3.078125,
558
+ "learning_rate": 1.3049645390070925e-05,
559
+ "loss": 5.1366,
560
+ "num_input_tokens_seen": 81395712,
561
+ "step": 69
562
+ },
563
+ {
564
+ "epoch": 0.008257638315441783,
565
+ "grad_norm": 3.71875,
566
+ "learning_rate": 1.3238770685579197e-05,
567
+ "loss": 5.0749,
568
+ "num_input_tokens_seen": 82575360,
569
+ "step": 70
570
+ },
571
+ {
572
+ "epoch": 0.008375604577090951,
573
+ "grad_norm": 2.765625,
574
+ "learning_rate": 1.3427895981087472e-05,
575
+ "loss": 4.8899,
576
+ "num_input_tokens_seen": 83755008,
577
+ "step": 71
578
+ },
579
+ {
580
+ "epoch": 0.00849357083874012,
581
+ "grad_norm": 3.140625,
582
+ "learning_rate": 1.3617021276595745e-05,
583
+ "loss": 4.8809,
584
+ "num_input_tokens_seen": 84934656,
585
+ "step": 72
586
+ },
587
+ {
588
+ "epoch": 0.008611537100389288,
589
+ "grad_norm": 3.296875,
590
+ "learning_rate": 1.380614657210402e-05,
591
+ "loss": 4.8839,
592
+ "num_input_tokens_seen": 86114304,
593
+ "step": 73
594
+ },
595
+ {
596
+ "epoch": 0.008729503362038458,
597
+ "grad_norm": 2.953125,
598
+ "learning_rate": 1.3995271867612294e-05,
599
+ "loss": 4.9931,
600
+ "num_input_tokens_seen": 87293952,
601
+ "step": 74
602
+ },
603
+ {
604
+ "epoch": 0.008847469623687626,
605
+ "grad_norm": 2.9375,
606
+ "learning_rate": 1.418439716312057e-05,
607
+ "loss": 4.8041,
608
+ "num_input_tokens_seen": 88473600,
609
+ "step": 75
610
+ },
611
+ {
612
+ "epoch": 0.008965435885336794,
613
+ "grad_norm": 3.5625,
614
+ "learning_rate": 1.4373522458628843e-05,
615
+ "loss": 4.8611,
616
+ "num_input_tokens_seen": 89653248,
617
+ "step": 76
618
+ },
619
+ {
620
+ "epoch": 0.009083402146985962,
621
+ "grad_norm": 3.734375,
622
+ "learning_rate": 1.4562647754137118e-05,
623
+ "loss": 4.7747,
624
+ "num_input_tokens_seen": 90832896,
625
+ "step": 77
626
+ },
627
+ {
628
+ "epoch": 0.00920136840863513,
629
+ "grad_norm": 2.984375,
630
+ "learning_rate": 1.475177304964539e-05,
631
+ "loss": 4.6884,
632
+ "num_input_tokens_seen": 92012544,
633
+ "step": 78
634
+ },
635
+ {
636
+ "epoch": 0.009319334670284299,
637
+ "grad_norm": 2.671875,
638
+ "learning_rate": 1.4940898345153665e-05,
639
+ "loss": 4.6617,
640
+ "num_input_tokens_seen": 93192192,
641
+ "step": 79
642
+ },
643
+ {
644
+ "epoch": 0.009437300931933467,
645
+ "grad_norm": 2.84375,
646
+ "learning_rate": 1.5130023640661939e-05,
647
+ "loss": 4.6174,
648
+ "num_input_tokens_seen": 94371840,
649
+ "step": 80
650
+ },
651
+ {
652
+ "epoch": 0.009555267193582635,
653
+ "grad_norm": 3.71875,
654
+ "learning_rate": 1.5319148936170214e-05,
655
+ "loss": 4.5483,
656
+ "num_input_tokens_seen": 95551488,
657
+ "step": 81
658
+ },
659
+ {
660
+ "epoch": 0.009673233455231803,
661
+ "grad_norm": 4.34375,
662
+ "learning_rate": 1.5508274231678487e-05,
663
+ "loss": 4.604,
664
+ "num_input_tokens_seen": 96731136,
665
+ "step": 82
666
+ },
667
+ {
668
+ "epoch": 0.009791199716880972,
669
+ "grad_norm": 3.09375,
670
+ "learning_rate": 1.5697399527186764e-05,
671
+ "loss": 4.5806,
672
+ "num_input_tokens_seen": 97910784,
673
+ "step": 83
674
+ },
675
+ {
676
+ "epoch": 0.00990916597853014,
677
+ "grad_norm": 3.203125,
678
+ "learning_rate": 1.5886524822695038e-05,
679
+ "loss": 4.4723,
680
+ "num_input_tokens_seen": 99090432,
681
+ "step": 84
682
+ },
683
+ {
684
+ "epoch": 0.010027132240179308,
685
+ "grad_norm": 3.8125,
686
+ "learning_rate": 1.607565011820331e-05,
687
+ "loss": 4.4723,
688
+ "num_input_tokens_seen": 100270080,
689
+ "step": 85
690
+ },
691
+ {
692
+ "epoch": 0.010145098501828476,
693
+ "grad_norm": 3.03125,
694
+ "learning_rate": 1.6264775413711585e-05,
695
+ "loss": 4.4519,
696
+ "num_input_tokens_seen": 101449728,
697
+ "step": 86
698
+ },
699
+ {
700
+ "epoch": 0.010263064763477646,
701
+ "grad_norm": 5.53125,
702
+ "learning_rate": 1.645390070921986e-05,
703
+ "loss": 4.4077,
704
+ "num_input_tokens_seen": 102629376,
705
+ "step": 87
706
+ },
707
+ {
708
+ "epoch": 0.010381031025126814,
709
+ "grad_norm": 4.03125,
710
+ "learning_rate": 1.6643026004728132e-05,
711
+ "loss": 4.5295,
712
+ "num_input_tokens_seen": 103809024,
713
+ "step": 88
714
+ },
715
+ {
716
+ "epoch": 0.010498997286775983,
717
+ "grad_norm": 3.484375,
718
+ "learning_rate": 1.683215130023641e-05,
719
+ "loss": 4.3235,
720
+ "num_input_tokens_seen": 104988672,
721
+ "step": 89
722
+ },
723
+ {
724
+ "epoch": 0.010616963548425151,
725
+ "grad_norm": 5.78125,
726
+ "learning_rate": 1.7021276595744682e-05,
727
+ "loss": 4.3245,
728
+ "num_input_tokens_seen": 106168320,
729
+ "step": 90
730
+ },
731
+ {
732
+ "epoch": 0.010734929810074319,
733
+ "grad_norm": 3.0625,
734
+ "learning_rate": 1.7210401891252956e-05,
735
+ "loss": 4.2554,
736
+ "num_input_tokens_seen": 107347968,
737
+ "step": 91
738
+ },
739
+ {
740
+ "epoch": 0.010852896071723487,
741
+ "grad_norm": 3.671875,
742
+ "learning_rate": 1.7399527186761233e-05,
743
+ "loss": 4.2572,
744
+ "num_input_tokens_seen": 108527616,
745
+ "step": 92
746
+ },
747
+ {
748
+ "epoch": 0.010970862333372655,
749
+ "grad_norm": 5.0625,
750
+ "learning_rate": 1.7588652482269506e-05,
751
+ "loss": 4.2442,
752
+ "num_input_tokens_seen": 109707264,
753
+ "step": 93
754
+ },
755
+ {
756
+ "epoch": 0.011088828595021824,
757
+ "grad_norm": 2.6875,
758
+ "learning_rate": 1.7777777777777777e-05,
759
+ "loss": 4.3006,
760
+ "num_input_tokens_seen": 110886912,
761
+ "step": 94
762
+ },
763
+ {
764
+ "epoch": 0.011206794856670992,
765
+ "grad_norm": 3.171875,
766
+ "learning_rate": 1.7966903073286054e-05,
767
+ "loss": 4.2813,
768
+ "num_input_tokens_seen": 112066560,
769
+ "step": 95
770
+ },
771
+ {
772
+ "epoch": 0.01132476111832016,
773
+ "grad_norm": 2.765625,
774
+ "learning_rate": 1.8156028368794327e-05,
775
+ "loss": 4.1186,
776
+ "num_input_tokens_seen": 113246208,
777
+ "step": 96
778
+ },
779
+ {
780
+ "epoch": 0.011442727379969328,
781
+ "grad_norm": 3.1875,
782
+ "learning_rate": 1.83451536643026e-05,
783
+ "loss": 4.1652,
784
+ "num_input_tokens_seen": 114425856,
785
+ "step": 97
786
+ },
787
+ {
788
+ "epoch": 0.011560693641618497,
789
+ "grad_norm": 4.71875,
790
+ "learning_rate": 1.8534278959810878e-05,
791
+ "loss": 4.1923,
792
+ "num_input_tokens_seen": 115605504,
793
+ "step": 98
794
+ },
795
+ {
796
+ "epoch": 0.011678659903267665,
797
+ "grad_norm": 3.359375,
798
+ "learning_rate": 1.872340425531915e-05,
799
+ "loss": 4.0728,
800
+ "num_input_tokens_seen": 116785152,
801
+ "step": 99
802
+ },
803
+ {
804
+ "epoch": 0.011796626164916833,
805
+ "grad_norm": 2.484375,
806
+ "learning_rate": 1.8912529550827425e-05,
807
+ "loss": 4.2189,
808
+ "num_input_tokens_seen": 117964800,
809
+ "step": 100
810
+ },
811
+ {
812
+ "epoch": 0.011914592426566003,
813
+ "grad_norm": 3.109375,
814
+ "learning_rate": 1.91016548463357e-05,
815
+ "loss": 4.0632,
816
+ "num_input_tokens_seen": 119144448,
817
+ "step": 101
818
+ },
819
+ {
820
+ "epoch": 0.012032558688215171,
821
+ "grad_norm": 4.40625,
822
+ "learning_rate": 1.929078014184397e-05,
823
+ "loss": 3.8775,
824
+ "num_input_tokens_seen": 120324096,
825
+ "step": 102
826
+ },
827
+ {
828
+ "epoch": 0.01215052494986434,
829
+ "grad_norm": 2.5625,
830
+ "learning_rate": 1.9479905437352245e-05,
831
+ "loss": 3.8658,
832
+ "num_input_tokens_seen": 121503744,
833
+ "step": 103
834
+ },
835
+ {
836
+ "epoch": 0.012268491211513508,
837
+ "grad_norm": 3.703125,
838
+ "learning_rate": 1.9669030732860522e-05,
839
+ "loss": 3.9888,
840
+ "num_input_tokens_seen": 122683392,
841
+ "step": 104
842
+ },
843
+ {
844
+ "epoch": 0.012386457473162676,
845
+ "grad_norm": 2.78125,
846
+ "learning_rate": 1.9858156028368796e-05,
847
+ "loss": 3.9439,
848
+ "num_input_tokens_seen": 123863040,
849
+ "step": 105
850
+ },
851
+ {
852
+ "epoch": 0.012504423734811844,
853
+ "grad_norm": 3.4375,
854
+ "learning_rate": 2.004728132387707e-05,
855
+ "loss": 4.0098,
856
+ "num_input_tokens_seen": 125042688,
857
+ "step": 106
858
+ },
859
+ {
860
+ "epoch": 0.012622389996461012,
861
+ "grad_norm": 2.71875,
862
+ "learning_rate": 2.0236406619385343e-05,
863
+ "loss": 3.9454,
864
+ "num_input_tokens_seen": 126222336,
865
+ "step": 107
866
+ },
867
+ {
868
+ "epoch": 0.01274035625811018,
869
+ "grad_norm": 3.03125,
870
+ "learning_rate": 2.0425531914893616e-05,
871
+ "loss": 3.8491,
872
+ "num_input_tokens_seen": 127401984,
873
+ "step": 108
874
+ },
875
+ {
876
+ "epoch": 0.012858322519759349,
877
+ "grad_norm": 3.59375,
878
+ "learning_rate": 2.0614657210401893e-05,
879
+ "loss": 3.9602,
880
+ "num_input_tokens_seen": 128581632,
881
+ "step": 109
882
+ },
883
+ {
884
+ "epoch": 0.012976288781408517,
885
+ "grad_norm": 5.25,
886
+ "learning_rate": 2.0803782505910167e-05,
887
+ "loss": 3.9542,
888
+ "num_input_tokens_seen": 129761280,
889
+ "step": 110
890
+ },
891
+ {
892
+ "epoch": 0.013094255043057685,
893
+ "grad_norm": 2.34375,
894
+ "learning_rate": 2.099290780141844e-05,
895
+ "loss": 3.7846,
896
+ "num_input_tokens_seen": 130940928,
897
+ "step": 111
898
+ },
899
+ {
900
+ "epoch": 0.013212221304706853,
901
+ "grad_norm": 3.484375,
902
+ "learning_rate": 2.1182033096926714e-05,
903
+ "loss": 3.9128,
904
+ "num_input_tokens_seen": 132120576,
905
+ "step": 112
906
+ },
907
+ {
908
+ "epoch": 0.013330187566356021,
909
+ "grad_norm": 7.4375,
910
+ "learning_rate": 2.137115839243499e-05,
911
+ "loss": 3.7365,
912
+ "num_input_tokens_seen": 133300224,
913
+ "step": 113
914
+ },
915
+ {
916
+ "epoch": 0.013448153828005191,
917
+ "grad_norm": 3.6875,
918
+ "learning_rate": 2.1560283687943264e-05,
919
+ "loss": 3.8298,
920
+ "num_input_tokens_seen": 134479872,
921
+ "step": 114
922
+ },
923
+ {
924
+ "epoch": 0.01356612008965436,
925
+ "grad_norm": 11.25,
926
+ "learning_rate": 2.1749408983451538e-05,
927
+ "loss": 3.9192,
928
+ "num_input_tokens_seen": 135659520,
929
+ "step": 115
930
+ },
931
+ {
932
+ "epoch": 0.013684086351303528,
933
+ "grad_norm": 9.875,
934
+ "learning_rate": 2.1938534278959815e-05,
935
+ "loss": 3.7628,
936
+ "num_input_tokens_seen": 136839168,
937
+ "step": 116
938
+ },
939
+ {
940
+ "epoch": 0.013802052612952696,
941
+ "grad_norm": 4.90625,
942
+ "learning_rate": 2.2127659574468088e-05,
943
+ "loss": 3.7212,
944
+ "num_input_tokens_seen": 138018816,
945
+ "step": 117
946
+ },
947
+ {
948
+ "epoch": 0.013920018874601864,
949
+ "grad_norm": 6.875,
950
+ "learning_rate": 2.2316784869976362e-05,
951
+ "loss": 3.8291,
952
+ "num_input_tokens_seen": 139198464,
953
+ "step": 118
954
+ },
955
+ {
956
+ "epoch": 0.014037985136251032,
957
+ "grad_norm": 3.984375,
958
+ "learning_rate": 2.2505910165484635e-05,
959
+ "loss": 3.7104,
960
+ "num_input_tokens_seen": 140378112,
961
+ "step": 119
962
+ },
963
+ {
964
+ "epoch": 0.0141559513979002,
965
+ "grad_norm": 5.0625,
966
+ "learning_rate": 2.2695035460992912e-05,
967
+ "loss": 3.6898,
968
+ "num_input_tokens_seen": 141557760,
969
+ "step": 120
970
+ },
971
+ {
972
+ "epoch": 0.014273917659549369,
973
+ "grad_norm": 3.828125,
974
+ "learning_rate": 2.2884160756501186e-05,
975
+ "loss": 3.61,
976
+ "num_input_tokens_seen": 142737408,
977
+ "step": 121
978
+ },
979
+ {
980
+ "epoch": 0.014391883921198537,
981
+ "grad_norm": 4.875,
982
+ "learning_rate": 2.307328605200946e-05,
983
+ "loss": 3.6886,
984
+ "num_input_tokens_seen": 143917056,
985
+ "step": 122
986
+ },
987
+ {
988
+ "epoch": 0.014509850182847705,
989
+ "grad_norm": 4.75,
990
+ "learning_rate": 2.326241134751773e-05,
991
+ "loss": 3.6435,
992
+ "num_input_tokens_seen": 145096704,
993
+ "step": 123
994
+ },
995
+ {
996
+ "epoch": 0.014627816444496873,
997
+ "grad_norm": 3.75,
998
+ "learning_rate": 2.3451536643026006e-05,
999
+ "loss": 3.6619,
1000
+ "num_input_tokens_seen": 146276352,
1001
+ "step": 124
1002
+ },
1003
+ {
1004
+ "epoch": 0.014745782706146042,
1005
+ "grad_norm": 5.375,
1006
+ "learning_rate": 2.364066193853428e-05,
1007
+ "loss": 3.631,
1008
+ "num_input_tokens_seen": 147456000,
1009
+ "step": 125
1010
+ },
1011
+ {
1012
+ "epoch": 0.01486374896779521,
1013
+ "grad_norm": 4.125,
1014
+ "learning_rate": 2.3829787234042553e-05,
1015
+ "loss": 3.6482,
1016
+ "num_input_tokens_seen": 148635648,
1017
+ "step": 126
1018
+ },
1019
+ {
1020
+ "epoch": 0.014981715229444378,
1021
+ "grad_norm": 7.78125,
1022
+ "learning_rate": 2.4018912529550827e-05,
1023
+ "loss": 3.6785,
1024
+ "num_input_tokens_seen": 149815296,
1025
+ "step": 127
1026
+ },
1027
+ {
1028
+ "epoch": 0.015099681491093548,
1029
+ "grad_norm": 7.0,
1030
+ "learning_rate": 2.4208037825059104e-05,
1031
+ "loss": 3.6113,
1032
+ "num_input_tokens_seen": 150994944,
1033
+ "step": 128
1034
+ },
1035
+ {
1036
+ "epoch": 0.015217647752742716,
1037
+ "grad_norm": 4.46875,
1038
+ "learning_rate": 2.4397163120567377e-05,
1039
+ "loss": 3.6015,
1040
+ "num_input_tokens_seen": 152174592,
1041
+ "step": 129
1042
+ },
1043
+ {
1044
+ "epoch": 0.015335614014391884,
1045
+ "grad_norm": 4.1875,
1046
+ "learning_rate": 2.458628841607565e-05,
1047
+ "loss": 3.5241,
1048
+ "num_input_tokens_seen": 153354240,
1049
+ "step": 130
1050
+ },
1051
+ {
1052
+ "epoch": 0.015453580276041053,
1053
+ "grad_norm": 4.0625,
1054
+ "learning_rate": 2.4775413711583928e-05,
1055
+ "loss": 3.6007,
1056
+ "num_input_tokens_seen": 154533888,
1057
+ "step": 131
1058
+ },
1059
+ {
1060
+ "epoch": 0.01557154653769022,
1061
+ "grad_norm": 3.046875,
1062
+ "learning_rate": 2.49645390070922e-05,
1063
+ "loss": 3.5949,
1064
+ "num_input_tokens_seen": 155713536,
1065
+ "step": 132
1066
+ },
1067
+ {
1068
+ "epoch": 0.01568951279933939,
1069
+ "grad_norm": 3.265625,
1070
+ "learning_rate": 2.5153664302600475e-05,
1071
+ "loss": 3.5403,
1072
+ "num_input_tokens_seen": 156893184,
1073
+ "step": 133
1074
+ },
1075
+ {
1076
+ "epoch": 0.015807479060988557,
1077
+ "grad_norm": 2.859375,
1078
+ "learning_rate": 2.5342789598108752e-05,
1079
+ "loss": 3.5399,
1080
+ "num_input_tokens_seen": 158072832,
1081
+ "step": 134
1082
+ },
1083
+ {
1084
+ "epoch": 0.015925445322637725,
1085
+ "grad_norm": 2.6875,
1086
+ "learning_rate": 2.5531914893617025e-05,
1087
+ "loss": 3.5408,
1088
+ "num_input_tokens_seen": 159252480,
1089
+ "step": 135
1090
+ },
1091
+ {
1092
+ "epoch": 0.016043411584286894,
1093
+ "grad_norm": 2.03125,
1094
+ "learning_rate": 2.57210401891253e-05,
1095
+ "loss": 3.4678,
1096
+ "num_input_tokens_seen": 160432128,
1097
+ "step": 136
1098
+ },
1099
+ {
1100
+ "epoch": 0.016161377845936062,
1101
+ "grad_norm": 2.796875,
1102
+ "learning_rate": 2.5910165484633572e-05,
1103
+ "loss": 3.6019,
1104
+ "num_input_tokens_seen": 161611776,
1105
+ "step": 137
1106
+ },
1107
+ {
1108
+ "epoch": 0.01627934410758523,
1109
+ "grad_norm": 1.96875,
1110
+ "learning_rate": 2.609929078014185e-05,
1111
+ "loss": 3.4525,
1112
+ "num_input_tokens_seen": 162791424,
1113
+ "step": 138
1114
+ },
1115
+ {
1116
+ "epoch": 0.0163973103692344,
1117
+ "grad_norm": 2.390625,
1118
+ "learning_rate": 2.628841607565012e-05,
1119
+ "loss": 3.557,
1120
+ "num_input_tokens_seen": 163971072,
1121
+ "step": 139
1122
+ },
1123
+ {
1124
+ "epoch": 0.016515276630883566,
1125
+ "grad_norm": 2.734375,
1126
+ "learning_rate": 2.6477541371158393e-05,
1127
+ "loss": 3.5324,
1128
+ "num_input_tokens_seen": 165150720,
1129
+ "step": 140
1130
+ },
1131
+ {
1132
+ "epoch": 0.016633242892532735,
1133
+ "grad_norm": 3.703125,
1134
+ "learning_rate": 2.6666666666666667e-05,
1135
+ "loss": 3.4707,
1136
+ "num_input_tokens_seen": 166330368,
1137
+ "step": 141
1138
+ },
1139
+ {
1140
+ "epoch": 0.016751209154181903,
1141
+ "grad_norm": 3.53125,
1142
+ "learning_rate": 2.6855791962174944e-05,
1143
+ "loss": 3.3545,
1144
+ "num_input_tokens_seen": 167510016,
1145
+ "step": 142
1146
+ },
1147
+ {
1148
+ "epoch": 0.01686917541583107,
1149
+ "grad_norm": 1.7890625,
1150
+ "learning_rate": 2.7044917257683217e-05,
1151
+ "loss": 3.4051,
1152
+ "num_input_tokens_seen": 168689664,
1153
+ "step": 143
1154
+ },
1155
+ {
1156
+ "epoch": 0.01698714167748024,
1157
+ "grad_norm": 4.78125,
1158
+ "learning_rate": 2.723404255319149e-05,
1159
+ "loss": 3.4119,
1160
+ "num_input_tokens_seen": 169869312,
1161
+ "step": 144
1162
+ },
1163
+ {
1164
+ "epoch": 0.017105107939129408,
1165
+ "grad_norm": 3.0625,
1166
+ "learning_rate": 2.7423167848699764e-05,
1167
+ "loss": 3.5149,
1168
+ "num_input_tokens_seen": 171048960,
1169
+ "step": 145
1170
+ },
1171
+ {
1172
+ "epoch": 0.017223074200778576,
1173
+ "grad_norm": 3.671875,
1174
+ "learning_rate": 2.761229314420804e-05,
1175
+ "loss": 3.3442,
1176
+ "num_input_tokens_seen": 172228608,
1177
+ "step": 146
1178
+ },
1179
+ {
1180
+ "epoch": 0.017341040462427744,
1181
+ "grad_norm": 3.453125,
1182
+ "learning_rate": 2.7801418439716315e-05,
1183
+ "loss": 3.3277,
1184
+ "num_input_tokens_seen": 173408256,
1185
+ "step": 147
1186
+ },
1187
+ {
1188
+ "epoch": 0.017459006724076916,
1189
+ "grad_norm": 3.4375,
1190
+ "learning_rate": 2.7990543735224588e-05,
1191
+ "loss": 3.3905,
1192
+ "num_input_tokens_seen": 174587904,
1193
+ "step": 148
1194
+ },
1195
+ {
1196
+ "epoch": 0.017576972985726084,
1197
+ "grad_norm": 2.0,
1198
+ "learning_rate": 2.8179669030732865e-05,
1199
+ "loss": 3.276,
1200
+ "num_input_tokens_seen": 175767552,
1201
+ "step": 149
1202
+ },
1203
+ {
1204
+ "epoch": 0.017694939247375252,
1205
+ "grad_norm": 4.0,
1206
+ "learning_rate": 2.836879432624114e-05,
1207
+ "loss": 3.3211,
1208
+ "num_input_tokens_seen": 176947200,
1209
+ "step": 150
1210
+ },
1211
+ {
1212
+ "epoch": 0.01781290550902442,
1213
+ "grad_norm": 4.03125,
1214
+ "learning_rate": 2.8557919621749412e-05,
1215
+ "loss": 3.3483,
1216
+ "num_input_tokens_seen": 178126848,
1217
+ "step": 151
1218
+ },
1219
+ {
1220
+ "epoch": 0.01793087177067359,
1221
+ "grad_norm": 2.53125,
1222
+ "learning_rate": 2.8747044917257686e-05,
1223
+ "loss": 3.3391,
1224
+ "num_input_tokens_seen": 179306496,
1225
+ "step": 152
1226
+ },
1227
+ {
1228
+ "epoch": 0.018048838032322757,
1229
+ "grad_norm": 5.5625,
1230
+ "learning_rate": 2.8936170212765963e-05,
1231
+ "loss": 3.4719,
1232
+ "num_input_tokens_seen": 180486144,
1233
+ "step": 153
1234
+ },
1235
+ {
1236
+ "epoch": 0.018166804293971925,
1237
+ "grad_norm": 3.171875,
1238
+ "learning_rate": 2.9125295508274236e-05,
1239
+ "loss": 3.2727,
1240
+ "num_input_tokens_seen": 181665792,
1241
+ "step": 154
1242
+ },
1243
+ {
1244
+ "epoch": 0.018284770555621093,
1245
+ "grad_norm": 5.09375,
1246
+ "learning_rate": 2.9314420803782506e-05,
1247
+ "loss": 3.284,
1248
+ "num_input_tokens_seen": 182845440,
1249
+ "step": 155
1250
+ },
1251
+ {
1252
+ "epoch": 0.01840273681727026,
1253
+ "grad_norm": 3.25,
1254
+ "learning_rate": 2.950354609929078e-05,
1255
+ "loss": 3.2279,
1256
+ "num_input_tokens_seen": 184025088,
1257
+ "step": 156
1258
+ },
1259
+ {
1260
+ "epoch": 0.01852070307891943,
1261
+ "grad_norm": 3.71875,
1262
+ "learning_rate": 2.9692671394799057e-05,
1263
+ "loss": 3.2438,
1264
+ "num_input_tokens_seen": 185204736,
1265
+ "step": 157
1266
+ },
1267
+ {
1268
+ "epoch": 0.018638669340568598,
1269
+ "grad_norm": 3.875,
1270
+ "learning_rate": 2.988179669030733e-05,
1271
+ "loss": 3.3257,
1272
+ "num_input_tokens_seen": 186384384,
1273
+ "step": 158
1274
+ },
1275
+ {
1276
+ "epoch": 0.018756635602217766,
1277
+ "grad_norm": 2.21875,
1278
+ "learning_rate": 3.0070921985815604e-05,
1279
+ "loss": 3.2727,
1280
+ "num_input_tokens_seen": 187564032,
1281
+ "step": 159
1282
+ },
1283
+ {
1284
+ "epoch": 0.018874601863866934,
1285
+ "grad_norm": 4.125,
1286
+ "learning_rate": 3.0260047281323877e-05,
1287
+ "loss": 3.245,
1288
+ "num_input_tokens_seen": 188743680,
1289
+ "step": 160
1290
+ },
1291
+ {
1292
+ "epoch": 0.018992568125516102,
1293
+ "grad_norm": 3.640625,
1294
+ "learning_rate": 3.0449172576832154e-05,
1295
+ "loss": 3.1904,
1296
+ "num_input_tokens_seen": 189923328,
1297
+ "step": 161
1298
+ },
1299
+ {
1300
+ "epoch": 0.01911053438716527,
1301
+ "grad_norm": 2.625,
1302
+ "learning_rate": 3.063829787234043e-05,
1303
+ "loss": 3.2754,
1304
+ "num_input_tokens_seen": 191102976,
1305
+ "step": 162
1306
+ },
1307
+ {
1308
+ "epoch": 0.01922850064881444,
1309
+ "grad_norm": 3.578125,
1310
+ "learning_rate": 3.0827423167848705e-05,
1311
+ "loss": 3.1889,
1312
+ "num_input_tokens_seen": 192282624,
1313
+ "step": 163
1314
+ },
1315
+ {
1316
+ "epoch": 0.019346466910463607,
1317
+ "grad_norm": 2.953125,
1318
+ "learning_rate": 3.1016548463356975e-05,
1319
+ "loss": 3.2809,
1320
+ "num_input_tokens_seen": 193462272,
1321
+ "step": 164
1322
+ },
1323
+ {
1324
+ "epoch": 0.019464433172112775,
1325
+ "grad_norm": 2.234375,
1326
+ "learning_rate": 3.120567375886525e-05,
1327
+ "loss": 3.1929,
1328
+ "num_input_tokens_seen": 194641920,
1329
+ "step": 165
1330
+ },
1331
+ {
1332
+ "epoch": 0.019582399433761943,
1333
+ "grad_norm": 2.515625,
1334
+ "learning_rate": 3.139479905437353e-05,
1335
+ "loss": 3.1474,
1336
+ "num_input_tokens_seen": 195821568,
1337
+ "step": 166
1338
+ },
1339
+ {
1340
+ "epoch": 0.01970036569541111,
1341
+ "grad_norm": 4.09375,
1342
+ "learning_rate": 3.15839243498818e-05,
1343
+ "loss": 3.0433,
1344
+ "num_input_tokens_seen": 197001216,
1345
+ "step": 167
1346
+ },
1347
+ {
1348
+ "epoch": 0.01981833195706028,
1349
+ "grad_norm": 2.6875,
1350
+ "learning_rate": 3.1773049645390076e-05,
1351
+ "loss": 3.1527,
1352
+ "num_input_tokens_seen": 198180864,
1353
+ "step": 168
1354
+ },
1355
+ {
1356
+ "epoch": 0.019936298218709448,
1357
+ "grad_norm": 3.5,
1358
+ "learning_rate": 3.196217494089835e-05,
1359
+ "loss": 3.2145,
1360
+ "num_input_tokens_seen": 199360512,
1361
+ "step": 169
1362
+ },
1363
+ {
1364
+ "epoch": 0.020054264480358616,
1365
+ "grad_norm": 4.9375,
1366
+ "learning_rate": 3.215130023640662e-05,
1367
+ "loss": 3.2642,
1368
+ "num_input_tokens_seen": 200540160,
1369
+ "step": 170
1370
+ },
1371
+ {
1372
+ "epoch": 0.020172230742007784,
1373
+ "grad_norm": 2.484375,
1374
+ "learning_rate": 3.234042553191489e-05,
1375
+ "loss": 3.1286,
1376
+ "num_input_tokens_seen": 201719808,
1377
+ "step": 171
1378
+ },
1379
+ {
1380
+ "epoch": 0.020290197003656953,
1381
+ "grad_norm": 8.1875,
1382
+ "learning_rate": 3.252955082742317e-05,
1383
+ "loss": 3.1274,
1384
+ "num_input_tokens_seen": 202899456,
1385
+ "step": 172
1386
+ },
1387
+ {
1388
+ "epoch": 0.02040816326530612,
1389
+ "grad_norm": 6.375,
1390
+ "learning_rate": 3.271867612293144e-05,
1391
+ "loss": 3.096,
1392
+ "num_input_tokens_seen": 204079104,
1393
+ "step": 173
1394
+ },
1395
+ {
1396
+ "epoch": 0.020526129526955292,
1397
+ "grad_norm": 5.8125,
1398
+ "learning_rate": 3.290780141843972e-05,
1399
+ "loss": 3.1647,
1400
+ "num_input_tokens_seen": 205258752,
1401
+ "step": 174
1402
+ },
1403
+ {
1404
+ "epoch": 0.02064409578860446,
1405
+ "grad_norm": 5.6875,
1406
+ "learning_rate": 3.3096926713947994e-05,
1407
+ "loss": 3.269,
1408
+ "num_input_tokens_seen": 206438400,
1409
+ "step": 175
1410
+ },
1411
+ {
1412
+ "epoch": 0.02076206205025363,
1413
+ "grad_norm": 4.4375,
1414
+ "learning_rate": 3.3286052009456264e-05,
1415
+ "loss": 3.2106,
1416
+ "num_input_tokens_seen": 207618048,
1417
+ "step": 176
1418
+ },
1419
+ {
1420
+ "epoch": 0.020880028311902797,
1421
+ "grad_norm": 3.109375,
1422
+ "learning_rate": 3.347517730496454e-05,
1423
+ "loss": 3.1545,
1424
+ "num_input_tokens_seen": 208797696,
1425
+ "step": 177
1426
+ },
1427
+ {
1428
+ "epoch": 0.020997994573551965,
1429
+ "grad_norm": 6.6875,
1430
+ "learning_rate": 3.366430260047282e-05,
1431
+ "loss": 3.1045,
1432
+ "num_input_tokens_seen": 209977344,
1433
+ "step": 178
1434
+ },
1435
+ {
1436
+ "epoch": 0.021115960835201134,
1437
+ "grad_norm": 6.0625,
1438
+ "learning_rate": 3.385342789598109e-05,
1439
+ "loss": 3.0496,
1440
+ "num_input_tokens_seen": 211156992,
1441
+ "step": 179
1442
+ },
1443
+ {
1444
+ "epoch": 0.021233927096850302,
1445
+ "grad_norm": 4.03125,
1446
+ "learning_rate": 3.4042553191489365e-05,
1447
+ "loss": 3.081,
1448
+ "num_input_tokens_seen": 212336640,
1449
+ "step": 180
1450
+ },
1451
+ {
1452
+ "epoch": 0.02135189335849947,
1453
+ "grad_norm": 3.921875,
1454
+ "learning_rate": 3.423167848699764e-05,
1455
+ "loss": 3.0552,
1456
+ "num_input_tokens_seen": 213516288,
1457
+ "step": 181
1458
+ },
1459
+ {
1460
+ "epoch": 0.021469859620148638,
1461
+ "grad_norm": 4.5,
1462
+ "learning_rate": 3.442080378250591e-05,
1463
+ "loss": 3.0172,
1464
+ "num_input_tokens_seen": 214695936,
1465
+ "step": 182
1466
+ },
1467
+ {
1468
+ "epoch": 0.021587825881797806,
1469
+ "grad_norm": 4.0625,
1470
+ "learning_rate": 3.460992907801419e-05,
1471
+ "loss": 3.1379,
1472
+ "num_input_tokens_seen": 215875584,
1473
+ "step": 183
1474
+ },
1475
+ {
1476
+ "epoch": 0.021705792143446975,
1477
+ "grad_norm": 4.9375,
1478
+ "learning_rate": 3.4799054373522466e-05,
1479
+ "loss": 3.1235,
1480
+ "num_input_tokens_seen": 217055232,
1481
+ "step": 184
1482
+ },
1483
+ {
1484
+ "epoch": 0.021823758405096143,
1485
+ "grad_norm": 3.78125,
1486
+ "learning_rate": 3.4988179669030736e-05,
1487
+ "loss": 3.1189,
1488
+ "num_input_tokens_seen": 218234880,
1489
+ "step": 185
1490
+ },
1491
+ {
1492
+ "epoch": 0.02194172466674531,
1493
+ "grad_norm": 5.0625,
1494
+ "learning_rate": 3.517730496453901e-05,
1495
+ "loss": 3.0035,
1496
+ "num_input_tokens_seen": 219414528,
1497
+ "step": 186
1498
+ },
1499
+ {
1500
+ "epoch": 0.02205969092839448,
1501
+ "grad_norm": 4.28125,
1502
+ "learning_rate": 3.536643026004728e-05,
1503
+ "loss": 3.0478,
1504
+ "num_input_tokens_seen": 220594176,
1505
+ "step": 187
1506
+ },
1507
+ {
1508
+ "epoch": 0.022177657190043647,
1509
+ "grad_norm": 5.8125,
1510
+ "learning_rate": 3.555555555555555e-05,
1511
+ "loss": 3.0777,
1512
+ "num_input_tokens_seen": 221773824,
1513
+ "step": 188
1514
+ },
1515
+ {
1516
+ "epoch": 0.022295623451692816,
1517
+ "grad_norm": 5.0625,
1518
+ "learning_rate": 3.574468085106383e-05,
1519
+ "loss": 3.0665,
1520
+ "num_input_tokens_seen": 222953472,
1521
+ "step": 189
1522
+ },
1523
+ {
1524
+ "epoch": 0.022413589713341984,
1525
+ "grad_norm": 4.25,
1526
+ "learning_rate": 3.593380614657211e-05,
1527
+ "loss": 3.0271,
1528
+ "num_input_tokens_seen": 224133120,
1529
+ "step": 190
1530
+ },
1531
+ {
1532
+ "epoch": 0.022531555974991152,
1533
+ "grad_norm": 4.03125,
1534
+ "learning_rate": 3.612293144208038e-05,
1535
+ "loss": 3.033,
1536
+ "num_input_tokens_seen": 225312768,
1537
+ "step": 191
1538
+ },
1539
+ {
1540
+ "epoch": 0.02264952223664032,
1541
+ "grad_norm": 5.09375,
1542
+ "learning_rate": 3.6312056737588654e-05,
1543
+ "loss": 3.143,
1544
+ "num_input_tokens_seen": 226492416,
1545
+ "step": 192
1546
+ },
1547
+ {
1548
+ "epoch": 0.02276748849828949,
1549
+ "grad_norm": 4.09375,
1550
+ "learning_rate": 3.650118203309693e-05,
1551
+ "loss": 3.0347,
1552
+ "num_input_tokens_seen": 227672064,
1553
+ "step": 193
1554
+ },
1555
+ {
1556
+ "epoch": 0.022885454759938657,
1557
+ "grad_norm": 4.65625,
1558
+ "learning_rate": 3.66903073286052e-05,
1559
+ "loss": 3.07,
1560
+ "num_input_tokens_seen": 228851712,
1561
+ "step": 194
1562
+ },
1563
+ {
1564
+ "epoch": 0.023003421021587825,
1565
+ "grad_norm": 3.90625,
1566
+ "learning_rate": 3.687943262411348e-05,
1567
+ "loss": 3.0225,
1568
+ "num_input_tokens_seen": 230031360,
1569
+ "step": 195
1570
+ },
1571
+ {
1572
+ "epoch": 0.023121387283236993,
1573
+ "grad_norm": 4.96875,
1574
+ "learning_rate": 3.7068557919621755e-05,
1575
+ "loss": 3.0222,
1576
+ "num_input_tokens_seen": 231211008,
1577
+ "step": 196
1578
+ },
1579
+ {
1580
+ "epoch": 0.02323935354488616,
1581
+ "grad_norm": 3.84375,
1582
+ "learning_rate": 3.7257683215130025e-05,
1583
+ "loss": 2.9755,
1584
+ "num_input_tokens_seen": 232390656,
1585
+ "step": 197
1586
+ },
1587
+ {
1588
+ "epoch": 0.02335731980653533,
1589
+ "grad_norm": 5.0,
1590
+ "learning_rate": 3.74468085106383e-05,
1591
+ "loss": 3.0163,
1592
+ "num_input_tokens_seen": 233570304,
1593
+ "step": 198
1594
+ },
1595
+ {
1596
+ "epoch": 0.023475286068184498,
1597
+ "grad_norm": 4.53125,
1598
+ "learning_rate": 3.763593380614658e-05,
1599
+ "loss": 2.9553,
1600
+ "num_input_tokens_seen": 234749952,
1601
+ "step": 199
1602
+ },
1603
+ {
1604
+ "epoch": 0.023593252329833666,
1605
+ "grad_norm": 4.6875,
1606
+ "learning_rate": 3.782505910165485e-05,
1607
+ "loss": 3.015,
1608
+ "num_input_tokens_seen": 235929600,
1609
+ "step": 200
1610
+ },
1611
+ {
1612
+ "epoch": 0.023593252329833666,
1613
+ "eval_wikipedia_loss": 2.9615590572357178,
1614
+ "eval_wikipedia_runtime": 172.3085,
1615
+ "eval_wikipedia_samples_per_second": 4.074,
1616
+ "eval_wikipedia_steps_per_second": 0.174,
1617
+ "num_input_tokens_seen": 235929600,
1618
+ "step": 200
1619
+ },
1620
+ {
1621
+ "epoch": 0.023593252329833666,
1622
+ "eval_toxicity_loss": 4.73836088180542,
1623
+ "eval_toxicity_runtime": 0.999,
1624
+ "eval_toxicity_samples_per_second": 2.002,
1625
+ "eval_toxicity_steps_per_second": 1.001,
1626
+ "num_input_tokens_seen": 235929600,
1627
+ "step": 200
1628
+ },
1629
+ {
1630
+ "epoch": 0.023711218591482838,
1631
+ "grad_norm": 4.0,
1632
+ "learning_rate": 3.8014184397163126e-05,
1633
+ "loss": 2.9448,
1634
+ "num_input_tokens_seen": 237109248,
1635
+ "step": 201
1636
+ },
1637
+ {
1638
+ "epoch": 0.023829184853132006,
1639
+ "grad_norm": 4.78125,
1640
+ "learning_rate": 3.82033096926714e-05,
1641
+ "loss": 3.1131,
1642
+ "num_input_tokens_seen": 238288896,
1643
+ "step": 202
1644
+ },
1645
+ {
1646
+ "epoch": 0.023947151114781174,
1647
+ "grad_norm": 3.828125,
1648
+ "learning_rate": 3.839243498817967e-05,
1649
+ "loss": 2.9217,
1650
+ "num_input_tokens_seen": 239468544,
1651
+ "step": 203
1652
+ },
1653
+ {
1654
+ "epoch": 0.024065117376430342,
1655
+ "grad_norm": 4.46875,
1656
+ "learning_rate": 3.858156028368794e-05,
1657
+ "loss": 2.9658,
1658
+ "num_input_tokens_seen": 240648192,
1659
+ "step": 204
1660
+ },
1661
+ {
1662
+ "epoch": 0.02418308363807951,
1663
+ "grad_norm": 3.703125,
1664
+ "learning_rate": 3.877068557919622e-05,
1665
+ "loss": 2.9857,
1666
+ "num_input_tokens_seen": 241827840,
1667
+ "step": 205
1668
+ },
1669
+ {
1670
+ "epoch": 0.02430104989972868,
1671
+ "grad_norm": 4.9375,
1672
+ "learning_rate": 3.895981087470449e-05,
1673
+ "loss": 2.9761,
1674
+ "num_input_tokens_seen": 243007488,
1675
+ "step": 206
1676
+ },
1677
+ {
1678
+ "epoch": 0.024419016161377847,
1679
+ "grad_norm": 4.28125,
1680
+ "learning_rate": 3.914893617021277e-05,
1681
+ "loss": 2.981,
1682
+ "num_input_tokens_seen": 244187136,
1683
+ "step": 207
1684
+ },
1685
+ {
1686
+ "epoch": 0.024536982423027015,
1687
+ "grad_norm": 4.3125,
1688
+ "learning_rate": 3.9338061465721044e-05,
1689
+ "loss": 3.0308,
1690
+ "num_input_tokens_seen": 245366784,
1691
+ "step": 208
1692
+ },
1693
+ {
1694
+ "epoch": 0.024654948684676183,
1695
+ "grad_norm": 4.0,
1696
+ "learning_rate": 3.9527186761229314e-05,
1697
+ "loss": 2.9529,
1698
+ "num_input_tokens_seen": 246546432,
1699
+ "step": 209
1700
+ },
1701
+ {
1702
+ "epoch": 0.02477291494632535,
1703
+ "grad_norm": 4.28125,
1704
+ "learning_rate": 3.971631205673759e-05,
1705
+ "loss": 2.9297,
1706
+ "num_input_tokens_seen": 247726080,
1707
+ "step": 210
1708
+ },
1709
+ {
1710
+ "epoch": 0.02489088120797452,
1711
+ "grad_norm": 3.328125,
1712
+ "learning_rate": 3.990543735224587e-05,
1713
+ "loss": 2.8328,
1714
+ "num_input_tokens_seen": 248905728,
1715
+ "step": 211
1716
+ },
1717
+ {
1718
+ "epoch": 0.025008847469623688,
1719
+ "grad_norm": 4.6875,
1720
+ "learning_rate": 4.009456264775414e-05,
1721
+ "loss": 2.9565,
1722
+ "num_input_tokens_seen": 250085376,
1723
+ "step": 212
1724
+ },
1725
+ {
1726
+ "epoch": 0.025126813731272856,
1727
+ "grad_norm": 3.734375,
1728
+ "learning_rate": 4.028368794326241e-05,
1729
+ "loss": 2.9276,
1730
+ "num_input_tokens_seen": 251265024,
1731
+ "step": 213
1732
+ },
1733
+ {
1734
+ "epoch": 0.025244779992922024,
1735
+ "grad_norm": 4.90625,
1736
+ "learning_rate": 4.0472813238770685e-05,
1737
+ "loss": 2.8872,
1738
+ "num_input_tokens_seen": 252444672,
1739
+ "step": 214
1740
+ },
1741
+ {
1742
+ "epoch": 0.025362746254571193,
1743
+ "grad_norm": 3.984375,
1744
+ "learning_rate": 4.066193853427896e-05,
1745
+ "loss": 2.9423,
1746
+ "num_input_tokens_seen": 253624320,
1747
+ "step": 215
1748
+ },
1749
+ {
1750
+ "epoch": 0.02548071251622036,
1751
+ "grad_norm": 4.28125,
1752
+ "learning_rate": 4.085106382978723e-05,
1753
+ "loss": 2.9417,
1754
+ "num_input_tokens_seen": 254803968,
1755
+ "step": 216
1756
+ },
1757
+ {
1758
+ "epoch": 0.02559867877786953,
1759
+ "grad_norm": 3.53125,
1760
+ "learning_rate": 4.104018912529551e-05,
1761
+ "loss": 2.878,
1762
+ "num_input_tokens_seen": 255983616,
1763
+ "step": 217
1764
+ },
1765
+ {
1766
+ "epoch": 0.025716645039518697,
1767
+ "grad_norm": 3.71875,
1768
+ "learning_rate": 4.1229314420803786e-05,
1769
+ "loss": 2.8925,
1770
+ "num_input_tokens_seen": 257163264,
1771
+ "step": 218
1772
+ },
1773
+ {
1774
+ "epoch": 0.025834611301167865,
1775
+ "grad_norm": 3.109375,
1776
+ "learning_rate": 4.1418439716312056e-05,
1777
+ "loss": 2.9432,
1778
+ "num_input_tokens_seen": 258342912,
1779
+ "step": 219
1780
+ },
1781
+ {
1782
+ "epoch": 0.025952577562817034,
1783
+ "grad_norm": 4.875,
1784
+ "learning_rate": 4.1607565011820333e-05,
1785
+ "loss": 2.8552,
1786
+ "num_input_tokens_seen": 259522560,
1787
+ "step": 220
1788
+ },
1789
+ {
1790
+ "epoch": 0.026070543824466202,
1791
+ "grad_norm": 3.09375,
1792
+ "learning_rate": 4.1796690307328604e-05,
1793
+ "loss": 2.8685,
1794
+ "num_input_tokens_seen": 260702208,
1795
+ "step": 221
1796
+ },
1797
+ {
1798
+ "epoch": 0.02618851008611537,
1799
+ "grad_norm": 4.6875,
1800
+ "learning_rate": 4.198581560283688e-05,
1801
+ "loss": 2.9284,
1802
+ "num_input_tokens_seen": 261881856,
1803
+ "step": 222
1804
+ },
1805
+ {
1806
+ "epoch": 0.026306476347764538,
1807
+ "grad_norm": 3.265625,
1808
+ "learning_rate": 4.217494089834516e-05,
1809
+ "loss": 2.8862,
1810
+ "num_input_tokens_seen": 263061504,
1811
+ "step": 223
1812
+ },
1813
+ {
1814
+ "epoch": 0.026424442609413706,
1815
+ "grad_norm": 4.5625,
1816
+ "learning_rate": 4.236406619385343e-05,
1817
+ "loss": 2.8105,
1818
+ "num_input_tokens_seen": 264241152,
1819
+ "step": 224
1820
+ },
1821
+ {
1822
+ "epoch": 0.026542408871062875,
1823
+ "grad_norm": 3.546875,
1824
+ "learning_rate": 4.2553191489361704e-05,
1825
+ "loss": 2.9406,
1826
+ "num_input_tokens_seen": 265420800,
1827
+ "step": 225
1828
+ },
1829
+ {
1830
+ "epoch": 0.026660375132712043,
1831
+ "grad_norm": 3.640625,
1832
+ "learning_rate": 4.274231678486998e-05,
1833
+ "loss": 2.907,
1834
+ "num_input_tokens_seen": 266600448,
1835
+ "step": 226
1836
+ },
1837
+ {
1838
+ "epoch": 0.02677834139436121,
1839
+ "grad_norm": 3.015625,
1840
+ "learning_rate": 4.293144208037825e-05,
1841
+ "loss": 3.0143,
1842
+ "num_input_tokens_seen": 267780096,
1843
+ "step": 227
1844
+ },
1845
+ {
1846
+ "epoch": 0.026896307656010383,
1847
+ "grad_norm": 4.25,
1848
+ "learning_rate": 4.312056737588653e-05,
1849
+ "loss": 2.9126,
1850
+ "num_input_tokens_seen": 268959744,
1851
+ "step": 228
1852
+ },
1853
+ {
1854
+ "epoch": 0.02701427391765955,
1855
+ "grad_norm": 2.5,
1856
+ "learning_rate": 4.3309692671394805e-05,
1857
+ "loss": 2.8879,
1858
+ "num_input_tokens_seen": 270139392,
1859
+ "step": 229
1860
+ },
1861
+ {
1862
+ "epoch": 0.02713224017930872,
1863
+ "grad_norm": 3.34375,
1864
+ "learning_rate": 4.3498817966903076e-05,
1865
+ "loss": 2.8174,
1866
+ "num_input_tokens_seen": 271319040,
1867
+ "step": 230
1868
+ },
1869
+ {
1870
+ "epoch": 0.027250206440957887,
1871
+ "grad_norm": 4.40625,
1872
+ "learning_rate": 4.368794326241135e-05,
1873
+ "loss": 2.895,
1874
+ "num_input_tokens_seen": 272498688,
1875
+ "step": 231
1876
+ },
1877
+ {
1878
+ "epoch": 0.027368172702607056,
1879
+ "grad_norm": 2.140625,
1880
+ "learning_rate": 4.387706855791963e-05,
1881
+ "loss": 2.8573,
1882
+ "num_input_tokens_seen": 273678336,
1883
+ "step": 232
1884
+ },
1885
+ {
1886
+ "epoch": 0.027486138964256224,
1887
+ "grad_norm": 3.609375,
1888
+ "learning_rate": 4.40661938534279e-05,
1889
+ "loss": 2.7662,
1890
+ "num_input_tokens_seen": 274857984,
1891
+ "step": 233
1892
+ },
1893
+ {
1894
+ "epoch": 0.027604105225905392,
1895
+ "grad_norm": 3.484375,
1896
+ "learning_rate": 4.4255319148936176e-05,
1897
+ "loss": 2.8103,
1898
+ "num_input_tokens_seen": 276037632,
1899
+ "step": 234
1900
+ },
1901
+ {
1902
+ "epoch": 0.02772207148755456,
1903
+ "grad_norm": 1.9765625,
1904
+ "learning_rate": 4.444444444444445e-05,
1905
+ "loss": 2.8271,
1906
+ "num_input_tokens_seen": 277217280,
1907
+ "step": 235
1908
+ },
1909
+ {
1910
+ "epoch": 0.02784003774920373,
1911
+ "grad_norm": 3.328125,
1912
+ "learning_rate": 4.4633569739952723e-05,
1913
+ "loss": 2.8881,
1914
+ "num_input_tokens_seen": 278396928,
1915
+ "step": 236
1916
+ },
1917
+ {
1918
+ "epoch": 0.027958004010852897,
1919
+ "grad_norm": 3.9375,
1920
+ "learning_rate": 4.4822695035461e-05,
1921
+ "loss": 2.9018,
1922
+ "num_input_tokens_seen": 279576576,
1923
+ "step": 237
1924
+ },
1925
+ {
1926
+ "epoch": 0.028075970272502065,
1927
+ "grad_norm": 5.25,
1928
+ "learning_rate": 4.501182033096927e-05,
1929
+ "loss": 2.8584,
1930
+ "num_input_tokens_seen": 280756224,
1931
+ "step": 238
1932
+ },
1933
+ {
1934
+ "epoch": 0.028193936534151233,
1935
+ "grad_norm": 2.328125,
1936
+ "learning_rate": 4.520094562647755e-05,
1937
+ "loss": 2.8435,
1938
+ "num_input_tokens_seen": 281935872,
1939
+ "step": 239
1940
+ },
1941
+ {
1942
+ "epoch": 0.0283119027958004,
1943
+ "grad_norm": 3.375,
1944
+ "learning_rate": 4.5390070921985824e-05,
1945
+ "loss": 2.8827,
1946
+ "num_input_tokens_seen": 283115520,
1947
+ "step": 240
1948
+ },
1949
+ {
1950
+ "epoch": 0.02842986905744957,
1951
+ "grad_norm": 2.3125,
1952
+ "learning_rate": 4.5579196217494095e-05,
1953
+ "loss": 2.8415,
1954
+ "num_input_tokens_seen": 284295168,
1955
+ "step": 241
1956
+ },
1957
+ {
1958
+ "epoch": 0.028547835319098738,
1959
+ "grad_norm": 3.828125,
1960
+ "learning_rate": 4.576832151300237e-05,
1961
+ "loss": 2.8305,
1962
+ "num_input_tokens_seen": 285474816,
1963
+ "step": 242
1964
+ },
1965
+ {
1966
+ "epoch": 0.028665801580747906,
1967
+ "grad_norm": 2.5625,
1968
+ "learning_rate": 4.595744680851065e-05,
1969
+ "loss": 2.7947,
1970
+ "num_input_tokens_seen": 286654464,
1971
+ "step": 243
1972
+ },
1973
+ {
1974
+ "epoch": 0.028783767842397074,
1975
+ "grad_norm": 2.375,
1976
+ "learning_rate": 4.614657210401892e-05,
1977
+ "loss": 2.812,
1978
+ "num_input_tokens_seen": 287834112,
1979
+ "step": 244
1980
+ },
1981
+ {
1982
+ "epoch": 0.028901734104046242,
1983
+ "grad_norm": 2.578125,
1984
+ "learning_rate": 4.633569739952719e-05,
1985
+ "loss": 2.8157,
1986
+ "num_input_tokens_seen": 289013760,
1987
+ "step": 245
1988
+ },
1989
+ {
1990
+ "epoch": 0.02901970036569541,
1991
+ "grad_norm": 3.625,
1992
+ "learning_rate": 4.652482269503546e-05,
1993
+ "loss": 2.8768,
1994
+ "num_input_tokens_seen": 290193408,
1995
+ "step": 246
1996
+ },
1997
+ {
1998
+ "epoch": 0.02913766662734458,
1999
+ "grad_norm": 2.328125,
2000
+ "learning_rate": 4.6713947990543736e-05,
2001
+ "loss": 2.834,
2002
+ "num_input_tokens_seen": 291373056,
2003
+ "step": 247
2004
+ },
2005
+ {
2006
+ "epoch": 0.029255632888993747,
2007
+ "grad_norm": 2.1875,
2008
+ "learning_rate": 4.690307328605201e-05,
2009
+ "loss": 2.8125,
2010
+ "num_input_tokens_seen": 292552704,
2011
+ "step": 248
2012
+ },
2013
+ {
2014
+ "epoch": 0.029373599150642915,
2015
+ "grad_norm": 2.921875,
2016
+ "learning_rate": 4.709219858156028e-05,
2017
+ "loss": 2.7906,
2018
+ "num_input_tokens_seen": 293732352,
2019
+ "step": 249
2020
+ },
2021
+ {
2022
+ "epoch": 0.029491565412292083,
2023
+ "grad_norm": 1.921875,
2024
+ "learning_rate": 4.728132387706856e-05,
2025
+ "loss": 2.8163,
2026
+ "num_input_tokens_seen": 294912000,
2027
+ "step": 250
2028
+ },
2029
+ {
2030
+ "epoch": 0.02960953167394125,
2031
+ "grad_norm": 2.75,
2032
+ "learning_rate": 4.747044917257684e-05,
2033
+ "loss": 2.7826,
2034
+ "num_input_tokens_seen": 296091648,
2035
+ "step": 251
2036
+ },
2037
+ {
2038
+ "epoch": 0.02972749793559042,
2039
+ "grad_norm": 2.671875,
2040
+ "learning_rate": 4.765957446808511e-05,
2041
+ "loss": 2.8368,
2042
+ "num_input_tokens_seen": 297271296,
2043
+ "step": 252
2044
+ },
2045
+ {
2046
+ "epoch": 0.029845464197239588,
2047
+ "grad_norm": 2.796875,
2048
+ "learning_rate": 4.7848699763593384e-05,
2049
+ "loss": 2.7188,
2050
+ "num_input_tokens_seen": 298450944,
2051
+ "step": 253
2052
+ },
2053
+ {
2054
+ "epoch": 0.029963430458888756,
2055
+ "grad_norm": 1.984375,
2056
+ "learning_rate": 4.8037825059101654e-05,
2057
+ "loss": 2.7609,
2058
+ "num_input_tokens_seen": 299630592,
2059
+ "step": 254
2060
+ },
2061
+ {
2062
+ "epoch": 0.030081396720537928,
2063
+ "grad_norm": 2.484375,
2064
+ "learning_rate": 4.822695035460993e-05,
2065
+ "loss": 2.742,
2066
+ "num_input_tokens_seen": 300810240,
2067
+ "step": 255
2068
+ },
2069
+ {
2070
+ "epoch": 0.030199362982187096,
2071
+ "grad_norm": 1.84375,
2072
+ "learning_rate": 4.841607565011821e-05,
2073
+ "loss": 2.7751,
2074
+ "num_input_tokens_seen": 301989888,
2075
+ "step": 256
2076
+ },
2077
+ {
2078
+ "epoch": 0.030317329243836264,
2079
+ "grad_norm": 3.078125,
2080
+ "learning_rate": 4.860520094562648e-05,
2081
+ "loss": 2.6949,
2082
+ "num_input_tokens_seen": 303169536,
2083
+ "step": 257
2084
+ },
2085
+ {
2086
+ "epoch": 0.030435295505485432,
2087
+ "grad_norm": 3.25,
2088
+ "learning_rate": 4.8794326241134755e-05,
2089
+ "loss": 2.8174,
2090
+ "num_input_tokens_seen": 304349184,
2091
+ "step": 258
2092
+ },
2093
+ {
2094
+ "epoch": 0.0305532617671346,
2095
+ "grad_norm": 1.4921875,
2096
+ "learning_rate": 4.898345153664303e-05,
2097
+ "loss": 2.6962,
2098
+ "num_input_tokens_seen": 305528832,
2099
+ "step": 259
2100
+ },
2101
+ {
2102
+ "epoch": 0.03067122802878377,
2103
+ "grad_norm": 3.046875,
2104
+ "learning_rate": 4.91725768321513e-05,
2105
+ "loss": 2.8484,
2106
+ "num_input_tokens_seen": 306708480,
2107
+ "step": 260
2108
+ },
2109
+ {
2110
+ "epoch": 0.030789194290432937,
2111
+ "grad_norm": 2.78125,
2112
+ "learning_rate": 4.936170212765958e-05,
2113
+ "loss": 2.632,
2114
+ "num_input_tokens_seen": 307888128,
2115
+ "step": 261
2116
+ },
2117
+ {
2118
+ "epoch": 0.030907160552082105,
2119
+ "grad_norm": 2.546875,
2120
+ "learning_rate": 4.9550827423167856e-05,
2121
+ "loss": 2.8744,
2122
+ "num_input_tokens_seen": 309067776,
2123
+ "step": 262
2124
+ },
2125
+ {
2126
+ "epoch": 0.031025126813731273,
2127
+ "grad_norm": 2.359375,
2128
+ "learning_rate": 4.9739952718676126e-05,
2129
+ "loss": 2.7714,
2130
+ "num_input_tokens_seen": 310247424,
2131
+ "step": 263
2132
+ },
2133
+ {
2134
+ "epoch": 0.03114309307538044,
2135
+ "grad_norm": 1.8515625,
2136
+ "learning_rate": 4.99290780141844e-05,
2137
+ "loss": 2.8042,
2138
+ "num_input_tokens_seen": 311427072,
2139
+ "step": 264
2140
+ },
2141
+ {
2142
+ "epoch": 0.031261059337029606,
2143
+ "grad_norm": 2.265625,
2144
+ "learning_rate": 5.011820330969268e-05,
2145
+ "loss": 2.7448,
2146
+ "num_input_tokens_seen": 312606720,
2147
+ "step": 265
2148
+ },
2149
+ {
2150
+ "epoch": 0.03137902559867878,
2151
+ "grad_norm": 1.78125,
2152
+ "learning_rate": 5.030732860520095e-05,
2153
+ "loss": 2.678,
2154
+ "num_input_tokens_seen": 313786368,
2155
+ "step": 266
2156
+ },
2157
+ {
2158
+ "epoch": 0.03149699186032794,
2159
+ "grad_norm": 2.203125,
2160
+ "learning_rate": 5.049645390070923e-05,
2161
+ "loss": 2.676,
2162
+ "num_input_tokens_seen": 314966016,
2163
+ "step": 267
2164
+ },
2165
+ {
2166
+ "epoch": 0.031614958121977114,
2167
+ "grad_norm": 1.90625,
2168
+ "learning_rate": 5.0685579196217504e-05,
2169
+ "loss": 2.6778,
2170
+ "num_input_tokens_seen": 316145664,
2171
+ "step": 268
2172
+ },
2173
+ {
2174
+ "epoch": 0.031732924383626286,
2175
+ "grad_norm": 2.203125,
2176
+ "learning_rate": 5.0874704491725774e-05,
2177
+ "loss": 2.7624,
2178
+ "num_input_tokens_seen": 317325312,
2179
+ "step": 269
2180
+ },
2181
+ {
2182
+ "epoch": 0.03185089064527545,
2183
+ "grad_norm": 3.53125,
2184
+ "learning_rate": 5.106382978723405e-05,
2185
+ "loss": 2.6462,
2186
+ "num_input_tokens_seen": 318504960,
2187
+ "step": 270
2188
+ },
2189
+ {
2190
+ "epoch": 0.03196885690692462,
2191
+ "grad_norm": 2.1875,
2192
+ "learning_rate": 5.125295508274232e-05,
2193
+ "loss": 2.6475,
2194
+ "num_input_tokens_seen": 319684608,
2195
+ "step": 271
2196
+ },
2197
+ {
2198
+ "epoch": 0.03208682316857379,
2199
+ "grad_norm": 1.703125,
2200
+ "learning_rate": 5.14420803782506e-05,
2201
+ "loss": 2.7214,
2202
+ "num_input_tokens_seen": 320864256,
2203
+ "step": 272
2204
+ },
2205
+ {
2206
+ "epoch": 0.03220478943022296,
2207
+ "grad_norm": 2.25,
2208
+ "learning_rate": 5.1631205673758875e-05,
2209
+ "loss": 2.6597,
2210
+ "num_input_tokens_seen": 322043904,
2211
+ "step": 273
2212
+ },
2213
+ {
2214
+ "epoch": 0.032322755691872124,
2215
+ "grad_norm": 2.171875,
2216
+ "learning_rate": 5.1820330969267145e-05,
2217
+ "loss": 2.6412,
2218
+ "num_input_tokens_seen": 323223552,
2219
+ "step": 274
2220
+ },
2221
+ {
2222
+ "epoch": 0.032440721953521295,
2223
+ "grad_norm": 2.25,
2224
+ "learning_rate": 5.200945626477542e-05,
2225
+ "loss": 2.7664,
2226
+ "num_input_tokens_seen": 324403200,
2227
+ "step": 275
2228
+ },
2229
+ {
2230
+ "epoch": 0.03255868821517046,
2231
+ "grad_norm": 3.296875,
2232
+ "learning_rate": 5.21985815602837e-05,
2233
+ "loss": 2.7168,
2234
+ "num_input_tokens_seen": 325582848,
2235
+ "step": 276
2236
+ },
2237
+ {
2238
+ "epoch": 0.03267665447681963,
2239
+ "grad_norm": 1.515625,
2240
+ "learning_rate": 5.238770685579196e-05,
2241
+ "loss": 2.698,
2242
+ "num_input_tokens_seen": 326762496,
2243
+ "step": 277
2244
+ },
2245
+ {
2246
+ "epoch": 0.0327946207384688,
2247
+ "grad_norm": 1.75,
2248
+ "learning_rate": 5.257683215130024e-05,
2249
+ "loss": 2.7068,
2250
+ "num_input_tokens_seen": 327942144,
2251
+ "step": 278
2252
+ },
2253
+ {
2254
+ "epoch": 0.03291258700011797,
2255
+ "grad_norm": 2.859375,
2256
+ "learning_rate": 5.276595744680851e-05,
2257
+ "loss": 2.6487,
2258
+ "num_input_tokens_seen": 329121792,
2259
+ "step": 279
2260
+ },
2261
+ {
2262
+ "epoch": 0.03303055326176713,
2263
+ "grad_norm": 2.515625,
2264
+ "learning_rate": 5.2955082742316786e-05,
2265
+ "loss": 2.5925,
2266
+ "num_input_tokens_seen": 330301440,
2267
+ "step": 280
2268
+ },
2269
+ {
2270
+ "epoch": 0.033148519523416305,
2271
+ "grad_norm": 2.40625,
2272
+ "learning_rate": 5.314420803782506e-05,
2273
+ "loss": 2.6826,
2274
+ "num_input_tokens_seen": 331481088,
2275
+ "step": 281
2276
+ },
2277
+ {
2278
+ "epoch": 0.03326648578506547,
2279
+ "grad_norm": 2.015625,
2280
+ "learning_rate": 5.333333333333333e-05,
2281
+ "loss": 2.6997,
2282
+ "num_input_tokens_seen": 332660736,
2283
+ "step": 282
2284
+ },
2285
+ {
2286
+ "epoch": 0.03338445204671464,
2287
+ "grad_norm": 2.234375,
2288
+ "learning_rate": 5.352245862884161e-05,
2289
+ "loss": 2.6378,
2290
+ "num_input_tokens_seen": 333840384,
2291
+ "step": 283
2292
+ },
2293
+ {
2294
+ "epoch": 0.033502418308363806,
2295
+ "grad_norm": 1.984375,
2296
+ "learning_rate": 5.371158392434989e-05,
2297
+ "loss": 2.588,
2298
+ "num_input_tokens_seen": 335020032,
2299
+ "step": 284
2300
+ },
2301
+ {
2302
+ "epoch": 0.03362038457001298,
2303
+ "grad_norm": 2.546875,
2304
+ "learning_rate": 5.390070921985816e-05,
2305
+ "loss": 2.6051,
2306
+ "num_input_tokens_seen": 336199680,
2307
+ "step": 285
2308
+ },
2309
+ {
2310
+ "epoch": 0.03373835083166214,
2311
+ "grad_norm": 1.7421875,
2312
+ "learning_rate": 5.4089834515366434e-05,
2313
+ "loss": 2.6874,
2314
+ "num_input_tokens_seen": 337379328,
2315
+ "step": 286
2316
+ },
2317
+ {
2318
+ "epoch": 0.033856317093311314,
2319
+ "grad_norm": 2.09375,
2320
+ "learning_rate": 5.4278959810874704e-05,
2321
+ "loss": 2.7186,
2322
+ "num_input_tokens_seen": 338558976,
2323
+ "step": 287
2324
+ },
2325
+ {
2326
+ "epoch": 0.03397428335496048,
2327
+ "grad_norm": 1.796875,
2328
+ "learning_rate": 5.446808510638298e-05,
2329
+ "loss": 2.6698,
2330
+ "num_input_tokens_seen": 339738624,
2331
+ "step": 288
2332
+ },
2333
+ {
2334
+ "epoch": 0.03409224961660965,
2335
+ "grad_norm": 2.3125,
2336
+ "learning_rate": 5.465721040189126e-05,
2337
+ "loss": 2.603,
2338
+ "num_input_tokens_seen": 340918272,
2339
+ "step": 289
2340
+ },
2341
+ {
2342
+ "epoch": 0.034210215878258815,
2343
+ "grad_norm": 2.140625,
2344
+ "learning_rate": 5.484633569739953e-05,
2345
+ "loss": 2.6479,
2346
+ "num_input_tokens_seen": 342097920,
2347
+ "step": 290
2348
+ },
2349
+ {
2350
+ "epoch": 0.03432818213990799,
2351
+ "grad_norm": 2.140625,
2352
+ "learning_rate": 5.5035460992907805e-05,
2353
+ "loss": 2.7037,
2354
+ "num_input_tokens_seen": 343277568,
2355
+ "step": 291
2356
+ },
2357
+ {
2358
+ "epoch": 0.03444614840155715,
2359
+ "grad_norm": 1.90625,
2360
+ "learning_rate": 5.522458628841608e-05,
2361
+ "loss": 2.681,
2362
+ "num_input_tokens_seen": 344457216,
2363
+ "step": 292
2364
+ },
2365
+ {
2366
+ "epoch": 0.03456411466320632,
2367
+ "grad_norm": 2.046875,
2368
+ "learning_rate": 5.541371158392435e-05,
2369
+ "loss": 2.5947,
2370
+ "num_input_tokens_seen": 345636864,
2371
+ "step": 293
2372
+ },
2373
+ {
2374
+ "epoch": 0.03468208092485549,
2375
+ "grad_norm": 1.734375,
2376
+ "learning_rate": 5.560283687943263e-05,
2377
+ "loss": 2.6439,
2378
+ "num_input_tokens_seen": 346816512,
2379
+ "step": 294
2380
+ },
2381
+ {
2382
+ "epoch": 0.03480004718650466,
2383
+ "grad_norm": 2.28125,
2384
+ "learning_rate": 5.5791962174940906e-05,
2385
+ "loss": 2.6404,
2386
+ "num_input_tokens_seen": 347996160,
2387
+ "step": 295
2388
+ },
2389
+ {
2390
+ "epoch": 0.03491801344815383,
2391
+ "grad_norm": 2.03125,
2392
+ "learning_rate": 5.5981087470449176e-05,
2393
+ "loss": 2.6686,
2394
+ "num_input_tokens_seen": 349175808,
2395
+ "step": 296
2396
+ },
2397
+ {
2398
+ "epoch": 0.035035979709802996,
2399
+ "grad_norm": 1.7265625,
2400
+ "learning_rate": 5.617021276595745e-05,
2401
+ "loss": 2.666,
2402
+ "num_input_tokens_seen": 350355456,
2403
+ "step": 297
2404
+ },
2405
+ {
2406
+ "epoch": 0.03515394597145217,
2407
+ "grad_norm": 2.109375,
2408
+ "learning_rate": 5.635933806146573e-05,
2409
+ "loss": 2.7266,
2410
+ "num_input_tokens_seen": 351535104,
2411
+ "step": 298
2412
+ },
2413
+ {
2414
+ "epoch": 0.03527191223310133,
2415
+ "grad_norm": 2.03125,
2416
+ "learning_rate": 5.6548463356974e-05,
2417
+ "loss": 2.6655,
2418
+ "num_input_tokens_seen": 352714752,
2419
+ "step": 299
2420
+ },
2421
+ {
2422
+ "epoch": 0.035389878494750504,
2423
+ "grad_norm": 1.9296875,
2424
+ "learning_rate": 5.673758865248228e-05,
2425
+ "loss": 2.6975,
2426
+ "num_input_tokens_seen": 353894400,
2427
+ "step": 300
2428
+ },
2429
+ {
2430
+ "epoch": 0.03550784475639967,
2431
+ "grad_norm": 2.15625,
2432
+ "learning_rate": 5.692671394799055e-05,
2433
+ "loss": 2.6232,
2434
+ "num_input_tokens_seen": 355074048,
2435
+ "step": 301
2436
+ },
2437
+ {
2438
+ "epoch": 0.03562581101804884,
2439
+ "grad_norm": 2.140625,
2440
+ "learning_rate": 5.7115839243498824e-05,
2441
+ "loss": 2.6344,
2442
+ "num_input_tokens_seen": 356253696,
2443
+ "step": 302
2444
+ },
2445
+ {
2446
+ "epoch": 0.035743777279698005,
2447
+ "grad_norm": 1.7109375,
2448
+ "learning_rate": 5.73049645390071e-05,
2449
+ "loss": 2.5703,
2450
+ "num_input_tokens_seen": 357433344,
2451
+ "step": 303
2452
+ },
2453
+ {
2454
+ "epoch": 0.03586174354134718,
2455
+ "grad_norm": 2.125,
2456
+ "learning_rate": 5.749408983451537e-05,
2457
+ "loss": 2.5405,
2458
+ "num_input_tokens_seen": 358612992,
2459
+ "step": 304
2460
+ },
2461
+ {
2462
+ "epoch": 0.03597970980299634,
2463
+ "grad_norm": 2.546875,
2464
+ "learning_rate": 5.768321513002365e-05,
2465
+ "loss": 2.6079,
2466
+ "num_input_tokens_seen": 359792640,
2467
+ "step": 305
2468
+ },
2469
+ {
2470
+ "epoch": 0.03609767606464551,
2471
+ "grad_norm": 2.015625,
2472
+ "learning_rate": 5.7872340425531925e-05,
2473
+ "loss": 2.6958,
2474
+ "num_input_tokens_seen": 360972288,
2475
+ "step": 306
2476
+ },
2477
+ {
2478
+ "epoch": 0.03621564232629468,
2479
+ "grad_norm": 1.46875,
2480
+ "learning_rate": 5.8061465721040195e-05,
2481
+ "loss": 2.5541,
2482
+ "num_input_tokens_seen": 362151936,
2483
+ "step": 307
2484
+ },
2485
+ {
2486
+ "epoch": 0.03633360858794385,
2487
+ "grad_norm": 1.796875,
2488
+ "learning_rate": 5.825059101654847e-05,
2489
+ "loss": 2.6427,
2490
+ "num_input_tokens_seen": 363331584,
2491
+ "step": 308
2492
+ },
2493
+ {
2494
+ "epoch": 0.036451574849593014,
2495
+ "grad_norm": 2.171875,
2496
+ "learning_rate": 5.843971631205675e-05,
2497
+ "loss": 2.5853,
2498
+ "num_input_tokens_seen": 364511232,
2499
+ "step": 309
2500
+ },
2501
+ {
2502
+ "epoch": 0.036569541111242186,
2503
+ "grad_norm": 1.8046875,
2504
+ "learning_rate": 5.862884160756501e-05,
2505
+ "loss": 2.5426,
2506
+ "num_input_tokens_seen": 365690880,
2507
+ "step": 310
2508
+ },
2509
+ {
2510
+ "epoch": 0.03668750737289135,
2511
+ "grad_norm": 1.625,
2512
+ "learning_rate": 5.881796690307329e-05,
2513
+ "loss": 2.6835,
2514
+ "num_input_tokens_seen": 366870528,
2515
+ "step": 311
2516
+ },
2517
+ {
2518
+ "epoch": 0.03680547363454052,
2519
+ "grad_norm": 2.640625,
2520
+ "learning_rate": 5.900709219858156e-05,
2521
+ "loss": 2.626,
2522
+ "num_input_tokens_seen": 368050176,
2523
+ "step": 312
2524
+ },
2525
+ {
2526
+ "epoch": 0.03692343989618969,
2527
+ "grad_norm": 1.4765625,
2528
+ "learning_rate": 5.9196217494089836e-05,
2529
+ "loss": 2.5361,
2530
+ "num_input_tokens_seen": 369229824,
2531
+ "step": 313
2532
+ },
2533
+ {
2534
+ "epoch": 0.03704140615783886,
2535
+ "grad_norm": 3.03125,
2536
+ "learning_rate": 5.938534278959811e-05,
2537
+ "loss": 2.5762,
2538
+ "num_input_tokens_seen": 370409472,
2539
+ "step": 314
2540
+ },
2541
+ {
2542
+ "epoch": 0.037159372419488024,
2543
+ "grad_norm": 1.890625,
2544
+ "learning_rate": 5.9574468085106384e-05,
2545
+ "loss": 2.5874,
2546
+ "num_input_tokens_seen": 371589120,
2547
+ "step": 315
2548
+ },
2549
+ {
2550
+ "epoch": 0.037277338681137195,
2551
+ "grad_norm": 2.6875,
2552
+ "learning_rate": 5.976359338061466e-05,
2553
+ "loss": 2.5874,
2554
+ "num_input_tokens_seen": 372768768,
2555
+ "step": 316
2556
+ },
2557
+ {
2558
+ "epoch": 0.03739530494278636,
2559
+ "grad_norm": 1.8359375,
2560
+ "learning_rate": 5.995271867612294e-05,
2561
+ "loss": 2.4888,
2562
+ "num_input_tokens_seen": 373948416,
2563
+ "step": 317
2564
+ },
2565
+ {
2566
+ "epoch": 0.03751327120443553,
2567
+ "grad_norm": 2.25,
2568
+ "learning_rate": 6.014184397163121e-05,
2569
+ "loss": 2.5391,
2570
+ "num_input_tokens_seen": 375128064,
2571
+ "step": 318
2572
+ },
2573
+ {
2574
+ "epoch": 0.0376312374660847,
2575
+ "grad_norm": 1.890625,
2576
+ "learning_rate": 6.0330969267139484e-05,
2577
+ "loss": 2.6462,
2578
+ "num_input_tokens_seen": 376307712,
2579
+ "step": 319
2580
+ },
2581
+ {
2582
+ "epoch": 0.03774920372773387,
2583
+ "grad_norm": 1.875,
2584
+ "learning_rate": 6.0520094562647755e-05,
2585
+ "loss": 2.4726,
2586
+ "num_input_tokens_seen": 377487360,
2587
+ "step": 320
2588
+ },
2589
+ {
2590
+ "epoch": 0.03786716998938304,
2591
+ "grad_norm": 2.296875,
2592
+ "learning_rate": 6.070921985815603e-05,
2593
+ "loss": 2.5759,
2594
+ "num_input_tokens_seen": 378667008,
2595
+ "step": 321
2596
+ },
2597
+ {
2598
+ "epoch": 0.037985136251032205,
2599
+ "grad_norm": 2.703125,
2600
+ "learning_rate": 6.089834515366431e-05,
2601
+ "loss": 2.5199,
2602
+ "num_input_tokens_seen": 379846656,
2603
+ "step": 322
2604
+ },
2605
+ {
2606
+ "epoch": 0.038103102512681376,
2607
+ "grad_norm": 2.390625,
2608
+ "learning_rate": 6.108747044917259e-05,
2609
+ "loss": 2.6243,
2610
+ "num_input_tokens_seen": 381026304,
2611
+ "step": 323
2612
+ },
2613
+ {
2614
+ "epoch": 0.03822106877433054,
2615
+ "grad_norm": 1.65625,
2616
+ "learning_rate": 6.127659574468086e-05,
2617
+ "loss": 2.5922,
2618
+ "num_input_tokens_seen": 382205952,
2619
+ "step": 324
2620
+ },
2621
+ {
2622
+ "epoch": 0.03833903503597971,
2623
+ "grad_norm": 1.640625,
2624
+ "learning_rate": 6.146572104018913e-05,
2625
+ "loss": 2.5376,
2626
+ "num_input_tokens_seen": 383385600,
2627
+ "step": 325
2628
+ },
2629
+ {
2630
+ "epoch": 0.03845700129762888,
2631
+ "grad_norm": 2.03125,
2632
+ "learning_rate": 6.165484633569741e-05,
2633
+ "loss": 2.547,
2634
+ "num_input_tokens_seen": 384565248,
2635
+ "step": 326
2636
+ },
2637
+ {
2638
+ "epoch": 0.03857496755927805,
2639
+ "grad_norm": 2.765625,
2640
+ "learning_rate": 6.184397163120568e-05,
2641
+ "loss": 2.5095,
2642
+ "num_input_tokens_seen": 385744896,
2643
+ "step": 327
2644
+ },
2645
+ {
2646
+ "epoch": 0.038692933820927214,
2647
+ "grad_norm": 1.40625,
2648
+ "learning_rate": 6.203309692671395e-05,
2649
+ "loss": 2.5431,
2650
+ "num_input_tokens_seen": 386924544,
2651
+ "step": 328
2652
+ },
2653
+ {
2654
+ "epoch": 0.038810900082576386,
2655
+ "grad_norm": 2.203125,
2656
+ "learning_rate": 6.222222222222223e-05,
2657
+ "loss": 2.5744,
2658
+ "num_input_tokens_seen": 388104192,
2659
+ "step": 329
2660
+ },
2661
+ {
2662
+ "epoch": 0.03892886634422555,
2663
+ "grad_norm": 1.9921875,
2664
+ "learning_rate": 6.24113475177305e-05,
2665
+ "loss": 2.6511,
2666
+ "num_input_tokens_seen": 389283840,
2667
+ "step": 330
2668
+ },
2669
+ {
2670
+ "epoch": 0.03904683260587472,
2671
+ "grad_norm": 3.15625,
2672
+ "learning_rate": 6.260047281323877e-05,
2673
+ "loss": 2.5556,
2674
+ "num_input_tokens_seen": 390463488,
2675
+ "step": 331
2676
+ },
2677
+ {
2678
+ "epoch": 0.03916479886752389,
2679
+ "grad_norm": 1.953125,
2680
+ "learning_rate": 6.278959810874706e-05,
2681
+ "loss": 2.5616,
2682
+ "num_input_tokens_seen": 391643136,
2683
+ "step": 332
2684
+ },
2685
+ {
2686
+ "epoch": 0.03928276512917306,
2687
+ "grad_norm": 1.6484375,
2688
+ "learning_rate": 6.297872340425533e-05,
2689
+ "loss": 2.5747,
2690
+ "num_input_tokens_seen": 392822784,
2691
+ "step": 333
2692
+ },
2693
+ {
2694
+ "epoch": 0.03940073139082222,
2695
+ "grad_norm": 3.15625,
2696
+ "learning_rate": 6.31678486997636e-05,
2697
+ "loss": 2.4747,
2698
+ "num_input_tokens_seen": 394002432,
2699
+ "step": 334
2700
+ },
2701
+ {
2702
+ "epoch": 0.039518697652471395,
2703
+ "grad_norm": 1.765625,
2704
+ "learning_rate": 6.335697399527188e-05,
2705
+ "loss": 2.5124,
2706
+ "num_input_tokens_seen": 395182080,
2707
+ "step": 335
2708
+ },
2709
+ {
2710
+ "epoch": 0.03963666391412056,
2711
+ "grad_norm": 2.15625,
2712
+ "learning_rate": 6.354609929078015e-05,
2713
+ "loss": 2.552,
2714
+ "num_input_tokens_seen": 396361728,
2715
+ "step": 336
2716
+ },
2717
+ {
2718
+ "epoch": 0.03975463017576973,
2719
+ "grad_norm": 1.8125,
2720
+ "learning_rate": 6.373522458628842e-05,
2721
+ "loss": 2.4541,
2722
+ "num_input_tokens_seen": 397541376,
2723
+ "step": 337
2724
+ },
2725
+ {
2726
+ "epoch": 0.039872596437418896,
2727
+ "grad_norm": 3.15625,
2728
+ "learning_rate": 6.39243498817967e-05,
2729
+ "loss": 2.4496,
2730
+ "num_input_tokens_seen": 398721024,
2731
+ "step": 338
2732
+ },
2733
+ {
2734
+ "epoch": 0.03999056269906807,
2735
+ "grad_norm": 2.015625,
2736
+ "learning_rate": 6.411347517730498e-05,
2737
+ "loss": 2.5182,
2738
+ "num_input_tokens_seen": 399900672,
2739
+ "step": 339
2740
+ },
2741
+ {
2742
+ "epoch": 0.04010852896071723,
2743
+ "grad_norm": 1.8671875,
2744
+ "learning_rate": 6.430260047281325e-05,
2745
+ "loss": 2.5118,
2746
+ "num_input_tokens_seen": 401080320,
2747
+ "step": 340
2748
+ },
2749
+ {
2750
+ "epoch": 0.040226495222366404,
2751
+ "grad_norm": 4.46875,
2752
+ "learning_rate": 6.449172576832153e-05,
2753
+ "loss": 2.5161,
2754
+ "num_input_tokens_seen": 402259968,
2755
+ "step": 341
2756
+ },
2757
+ {
2758
+ "epoch": 0.04034446148401557,
2759
+ "grad_norm": 1.8828125,
2760
+ "learning_rate": 6.468085106382979e-05,
2761
+ "loss": 2.5411,
2762
+ "num_input_tokens_seen": 403439616,
2763
+ "step": 342
2764
+ },
2765
+ {
2766
+ "epoch": 0.04046242774566474,
2767
+ "grad_norm": 6.53125,
2768
+ "learning_rate": 6.486997635933806e-05,
2769
+ "loss": 2.5341,
2770
+ "num_input_tokens_seen": 404619264,
2771
+ "step": 343
2772
+ },
2773
+ {
2774
+ "epoch": 0.040580394007313905,
2775
+ "grad_norm": 5.0,
2776
+ "learning_rate": 6.505910165484634e-05,
2777
+ "loss": 2.5139,
2778
+ "num_input_tokens_seen": 405798912,
2779
+ "step": 344
2780
+ },
2781
+ {
2782
+ "epoch": 0.04069836026896308,
2783
+ "grad_norm": 6.9375,
2784
+ "learning_rate": 6.524822695035461e-05,
2785
+ "loss": 2.542,
2786
+ "num_input_tokens_seen": 406978560,
2787
+ "step": 345
2788
+ },
2789
+ {
2790
+ "epoch": 0.04081632653061224,
2791
+ "grad_norm": 6.625,
2792
+ "learning_rate": 6.543735224586288e-05,
2793
+ "loss": 2.6481,
2794
+ "num_input_tokens_seen": 408158208,
2795
+ "step": 346
2796
+ },
2797
+ {
2798
+ "epoch": 0.04093429279226141,
2799
+ "grad_norm": 1.9375,
2800
+ "learning_rate": 6.562647754137116e-05,
2801
+ "loss": 2.5413,
2802
+ "num_input_tokens_seen": 409337856,
2803
+ "step": 347
2804
+ },
2805
+ {
2806
+ "epoch": 0.041052259053910585,
2807
+ "grad_norm": 3.25,
2808
+ "learning_rate": 6.581560283687943e-05,
2809
+ "loss": 2.5869,
2810
+ "num_input_tokens_seen": 410517504,
2811
+ "step": 348
2812
+ },
2813
+ {
2814
+ "epoch": 0.04117022531555975,
2815
+ "grad_norm": 2.171875,
2816
+ "learning_rate": 6.60047281323877e-05,
2817
+ "loss": 2.5717,
2818
+ "num_input_tokens_seen": 411697152,
2819
+ "step": 349
2820
+ },
2821
+ {
2822
+ "epoch": 0.04128819157720892,
2823
+ "grad_norm": 2.296875,
2824
+ "learning_rate": 6.619385342789599e-05,
2825
+ "loss": 2.569,
2826
+ "num_input_tokens_seen": 412876800,
2827
+ "step": 350
2828
+ },
2829
+ {
2830
+ "epoch": 0.041406157838858086,
2831
+ "grad_norm": 2.53125,
2832
+ "learning_rate": 6.638297872340426e-05,
2833
+ "loss": 2.5656,
2834
+ "num_input_tokens_seen": 414056448,
2835
+ "step": 351
2836
+ },
2837
+ {
2838
+ "epoch": 0.04152412410050726,
2839
+ "grad_norm": 1.5625,
2840
+ "learning_rate": 6.657210401891253e-05,
2841
+ "loss": 2.4558,
2842
+ "num_input_tokens_seen": 415236096,
2843
+ "step": 352
2844
+ },
2845
+ {
2846
+ "epoch": 0.04164209036215642,
2847
+ "grad_norm": 3.046875,
2848
+ "learning_rate": 6.676122931442081e-05,
2849
+ "loss": 2.5107,
2850
+ "num_input_tokens_seen": 416415744,
2851
+ "step": 353
2852
+ },
2853
+ {
2854
+ "epoch": 0.041760056623805594,
2855
+ "grad_norm": 2.03125,
2856
+ "learning_rate": 6.695035460992908e-05,
2857
+ "loss": 2.5576,
2858
+ "num_input_tokens_seen": 417595392,
2859
+ "step": 354
2860
+ },
2861
+ {
2862
+ "epoch": 0.04187802288545476,
2863
+ "grad_norm": 2.015625,
2864
+ "learning_rate": 6.713947990543735e-05,
2865
+ "loss": 2.5029,
2866
+ "num_input_tokens_seen": 418775040,
2867
+ "step": 355
2868
+ },
2869
+ {
2870
+ "epoch": 0.04199598914710393,
2871
+ "grad_norm": 1.84375,
2872
+ "learning_rate": 6.732860520094564e-05,
2873
+ "loss": 2.547,
2874
+ "num_input_tokens_seen": 419954688,
2875
+ "step": 356
2876
+ },
2877
+ {
2878
+ "epoch": 0.042113955408753095,
2879
+ "grad_norm": 1.9453125,
2880
+ "learning_rate": 6.75177304964539e-05,
2881
+ "loss": 2.4942,
2882
+ "num_input_tokens_seen": 421134336,
2883
+ "step": 357
2884
+ },
2885
+ {
2886
+ "epoch": 0.04223192167040227,
2887
+ "grad_norm": 2.109375,
2888
+ "learning_rate": 6.770685579196218e-05,
2889
+ "loss": 2.4546,
2890
+ "num_input_tokens_seen": 422313984,
2891
+ "step": 358
2892
+ },
2893
+ {
2894
+ "epoch": 0.04234988793205143,
2895
+ "grad_norm": 1.875,
2896
+ "learning_rate": 6.789598108747046e-05,
2897
+ "loss": 2.6115,
2898
+ "num_input_tokens_seen": 423493632,
2899
+ "step": 359
2900
+ },
2901
+ {
2902
+ "epoch": 0.042467854193700603,
2903
+ "grad_norm": 1.5,
2904
+ "learning_rate": 6.808510638297873e-05,
2905
+ "loss": 2.4977,
2906
+ "num_input_tokens_seen": 424673280,
2907
+ "step": 360
2908
+ },
2909
+ {
2910
+ "epoch": 0.04258582045534977,
2911
+ "grad_norm": 1.84375,
2912
+ "learning_rate": 6.8274231678487e-05,
2913
+ "loss": 2.4208,
2914
+ "num_input_tokens_seen": 425852928,
2915
+ "step": 361
2916
+ },
2917
+ {
2918
+ "epoch": 0.04270378671699894,
2919
+ "grad_norm": 1.421875,
2920
+ "learning_rate": 6.846335697399528e-05,
2921
+ "loss": 2.5416,
2922
+ "num_input_tokens_seen": 427032576,
2923
+ "step": 362
2924
+ },
2925
+ {
2926
+ "epoch": 0.042821752978648105,
2927
+ "grad_norm": 1.78125,
2928
+ "learning_rate": 6.865248226950355e-05,
2929
+ "loss": 2.4938,
2930
+ "num_input_tokens_seen": 428212224,
2931
+ "step": 363
2932
+ },
2933
+ {
2934
+ "epoch": 0.042939719240297276,
2935
+ "grad_norm": 1.5859375,
2936
+ "learning_rate": 6.884160756501182e-05,
2937
+ "loss": 2.5203,
2938
+ "num_input_tokens_seen": 429391872,
2939
+ "step": 364
2940
+ },
2941
+ {
2942
+ "epoch": 0.04305768550194644,
2943
+ "grad_norm": 1.9296875,
2944
+ "learning_rate": 6.903073286052011e-05,
2945
+ "loss": 2.515,
2946
+ "num_input_tokens_seen": 430571520,
2947
+ "step": 365
2948
+ },
2949
+ {
2950
+ "epoch": 0.04317565176359561,
2951
+ "grad_norm": 2.25,
2952
+ "learning_rate": 6.921985815602838e-05,
2953
+ "loss": 2.5211,
2954
+ "num_input_tokens_seen": 431751168,
2955
+ "step": 366
2956
+ },
2957
+ {
2958
+ "epoch": 0.04329361802524478,
2959
+ "grad_norm": 1.796875,
2960
+ "learning_rate": 6.940898345153665e-05,
2961
+ "loss": 2.4887,
2962
+ "num_input_tokens_seen": 432930816,
2963
+ "step": 367
2964
+ },
2965
+ {
2966
+ "epoch": 0.04341158428689395,
2967
+ "grad_norm": 1.5234375,
2968
+ "learning_rate": 6.959810874704493e-05,
2969
+ "loss": 2.4896,
2970
+ "num_input_tokens_seen": 434110464,
2971
+ "step": 368
2972
+ },
2973
+ {
2974
+ "epoch": 0.043529550548543114,
2975
+ "grad_norm": 1.6171875,
2976
+ "learning_rate": 6.97872340425532e-05,
2977
+ "loss": 2.4172,
2978
+ "num_input_tokens_seen": 435290112,
2979
+ "step": 369
2980
+ },
2981
+ {
2982
+ "epoch": 0.043647516810192286,
2983
+ "grad_norm": 2.1875,
2984
+ "learning_rate": 6.997635933806147e-05,
2985
+ "loss": 2.4538,
2986
+ "num_input_tokens_seen": 436469760,
2987
+ "step": 370
2988
+ },
2989
+ {
2990
+ "epoch": 0.04376548307184145,
2991
+ "grad_norm": 2.015625,
2992
+ "learning_rate": 7.016548463356976e-05,
2993
+ "loss": 2.4439,
2994
+ "num_input_tokens_seen": 437649408,
2995
+ "step": 371
2996
+ },
2997
+ {
2998
+ "epoch": 0.04388344933349062,
2999
+ "grad_norm": 1.828125,
3000
+ "learning_rate": 7.035460992907803e-05,
3001
+ "loss": 2.4898,
3002
+ "num_input_tokens_seen": 438829056,
3003
+ "step": 372
3004
+ },
3005
+ {
3006
+ "epoch": 0.04400141559513979,
3007
+ "grad_norm": 1.6171875,
3008
+ "learning_rate": 7.05437352245863e-05,
3009
+ "loss": 2.4337,
3010
+ "num_input_tokens_seen": 440008704,
3011
+ "step": 373
3012
+ },
3013
+ {
3014
+ "epoch": 0.04411938185678896,
3015
+ "grad_norm": 1.53125,
3016
+ "learning_rate": 7.073286052009457e-05,
3017
+ "loss": 2.4729,
3018
+ "num_input_tokens_seen": 441188352,
3019
+ "step": 374
3020
+ },
3021
+ {
3022
+ "epoch": 0.04423734811843813,
3023
+ "grad_norm": 1.9765625,
3024
+ "learning_rate": 7.092198581560284e-05,
3025
+ "loss": 2.4801,
3026
+ "num_input_tokens_seen": 442368000,
3027
+ "step": 375
3028
+ },
3029
+ {
3030
+ "epoch": 0.044355314380087295,
3031
+ "grad_norm": 2.0,
3032
+ "learning_rate": 7.11111111111111e-05,
3033
+ "loss": 2.5573,
3034
+ "num_input_tokens_seen": 443547648,
3035
+ "step": 376
3036
+ },
3037
+ {
3038
+ "epoch": 0.044473280641736467,
3039
+ "grad_norm": 1.640625,
3040
+ "learning_rate": 7.130023640661939e-05,
3041
+ "loss": 2.4675,
3042
+ "num_input_tokens_seen": 444727296,
3043
+ "step": 377
3044
+ },
3045
+ {
3046
+ "epoch": 0.04459124690338563,
3047
+ "grad_norm": 1.8984375,
3048
+ "learning_rate": 7.148936170212766e-05,
3049
+ "loss": 2.4965,
3050
+ "num_input_tokens_seen": 445906944,
3051
+ "step": 378
3052
+ },
3053
+ {
3054
+ "epoch": 0.0447092131650348,
3055
+ "grad_norm": 1.2578125,
3056
+ "learning_rate": 7.167848699763593e-05,
3057
+ "loss": 2.4161,
3058
+ "num_input_tokens_seen": 447086592,
3059
+ "step": 379
3060
+ },
3061
+ {
3062
+ "epoch": 0.04482717942668397,
3063
+ "grad_norm": 1.8359375,
3064
+ "learning_rate": 7.186761229314421e-05,
3065
+ "loss": 2.4437,
3066
+ "num_input_tokens_seen": 448266240,
3067
+ "step": 380
3068
+ },
3069
+ {
3070
+ "epoch": 0.04494514568833314,
3071
+ "grad_norm": 1.8046875,
3072
+ "learning_rate": 7.205673758865248e-05,
3073
+ "loss": 2.4978,
3074
+ "num_input_tokens_seen": 449445888,
3075
+ "step": 381
3076
+ },
3077
+ {
3078
+ "epoch": 0.045063111949982304,
3079
+ "grad_norm": 1.53125,
3080
+ "learning_rate": 7.224586288416075e-05,
3081
+ "loss": 2.4316,
3082
+ "num_input_tokens_seen": 450625536,
3083
+ "step": 382
3084
+ },
3085
+ {
3086
+ "epoch": 0.045181078211631476,
3087
+ "grad_norm": 2.0625,
3088
+ "learning_rate": 7.243498817966904e-05,
3089
+ "loss": 2.5152,
3090
+ "num_input_tokens_seen": 451805184,
3091
+ "step": 383
3092
+ },
3093
+ {
3094
+ "epoch": 0.04529904447328064,
3095
+ "grad_norm": 1.65625,
3096
+ "learning_rate": 7.262411347517731e-05,
3097
+ "loss": 2.4161,
3098
+ "num_input_tokens_seen": 452984832,
3099
+ "step": 384
3100
+ },
3101
+ {
3102
+ "epoch": 0.04541701073492981,
3103
+ "grad_norm": 1.8515625,
3104
+ "learning_rate": 7.281323877068558e-05,
3105
+ "loss": 2.4274,
3106
+ "num_input_tokens_seen": 454164480,
3107
+ "step": 385
3108
+ },
3109
+ {
3110
+ "epoch": 0.04553497699657898,
3111
+ "grad_norm": 2.0625,
3112
+ "learning_rate": 7.300236406619386e-05,
3113
+ "loss": 2.4693,
3114
+ "num_input_tokens_seen": 455344128,
3115
+ "step": 386
3116
+ },
3117
+ {
3118
+ "epoch": 0.04565294325822815,
3119
+ "grad_norm": 1.765625,
3120
+ "learning_rate": 7.319148936170213e-05,
3121
+ "loss": 2.4907,
3122
+ "num_input_tokens_seen": 456523776,
3123
+ "step": 387
3124
+ },
3125
+ {
3126
+ "epoch": 0.04577090951987731,
3127
+ "grad_norm": 1.71875,
3128
+ "learning_rate": 7.33806146572104e-05,
3129
+ "loss": 2.4603,
3130
+ "num_input_tokens_seen": 457703424,
3131
+ "step": 388
3132
+ },
3133
+ {
3134
+ "epoch": 0.045888875781526485,
3135
+ "grad_norm": 1.7734375,
3136
+ "learning_rate": 7.356973995271869e-05,
3137
+ "loss": 2.4651,
3138
+ "num_input_tokens_seen": 458883072,
3139
+ "step": 389
3140
+ },
3141
+ {
3142
+ "epoch": 0.04600684204317565,
3143
+ "grad_norm": 1.9765625,
3144
+ "learning_rate": 7.375886524822696e-05,
3145
+ "loss": 2.4041,
3146
+ "num_input_tokens_seen": 460062720,
3147
+ "step": 390
3148
+ },
3149
+ {
3150
+ "epoch": 0.04612480830482482,
3151
+ "grad_norm": 2.90625,
3152
+ "learning_rate": 7.394799054373523e-05,
3153
+ "loss": 2.3896,
3154
+ "num_input_tokens_seen": 461242368,
3155
+ "step": 391
3156
+ },
3157
+ {
3158
+ "epoch": 0.046242774566473986,
3159
+ "grad_norm": 1.515625,
3160
+ "learning_rate": 7.413711583924351e-05,
3161
+ "loss": 2.3966,
3162
+ "num_input_tokens_seen": 462422016,
3163
+ "step": 392
3164
+ },
3165
+ {
3166
+ "epoch": 0.04636074082812316,
3167
+ "grad_norm": 1.984375,
3168
+ "learning_rate": 7.432624113475178e-05,
3169
+ "loss": 2.5009,
3170
+ "num_input_tokens_seen": 463601664,
3171
+ "step": 393
3172
+ },
3173
+ {
3174
+ "epoch": 0.04647870708977232,
3175
+ "grad_norm": 1.8984375,
3176
+ "learning_rate": 7.451536643026005e-05,
3177
+ "loss": 2.4524,
3178
+ "num_input_tokens_seen": 464781312,
3179
+ "step": 394
3180
+ },
3181
+ {
3182
+ "epoch": 0.046596673351421494,
3183
+ "grad_norm": 2.46875,
3184
+ "learning_rate": 7.470449172576833e-05,
3185
+ "loss": 2.4169,
3186
+ "num_input_tokens_seen": 465960960,
3187
+ "step": 395
3188
+ },
3189
+ {
3190
+ "epoch": 0.04671463961307066,
3191
+ "grad_norm": 1.609375,
3192
+ "learning_rate": 7.48936170212766e-05,
3193
+ "loss": 2.5313,
3194
+ "num_input_tokens_seen": 467140608,
3195
+ "step": 396
3196
+ },
3197
+ {
3198
+ "epoch": 0.04683260587471983,
3199
+ "grad_norm": 1.7890625,
3200
+ "learning_rate": 7.508274231678487e-05,
3201
+ "loss": 2.5893,
3202
+ "num_input_tokens_seen": 468320256,
3203
+ "step": 397
3204
+ },
3205
+ {
3206
+ "epoch": 0.046950572136368995,
3207
+ "grad_norm": 1.9453125,
3208
+ "learning_rate": 7.527186761229316e-05,
3209
+ "loss": 2.453,
3210
+ "num_input_tokens_seen": 469499904,
3211
+ "step": 398
3212
+ },
3213
+ {
3214
+ "epoch": 0.04706853839801817,
3215
+ "grad_norm": 1.78125,
3216
+ "learning_rate": 7.546099290780143e-05,
3217
+ "loss": 2.4462,
3218
+ "num_input_tokens_seen": 470679552,
3219
+ "step": 399
3220
+ },
3221
+ {
3222
+ "epoch": 0.04718650465966733,
3223
+ "grad_norm": 2.140625,
3224
+ "learning_rate": 7.56501182033097e-05,
3225
+ "loss": 2.5393,
3226
+ "num_input_tokens_seen": 471859200,
3227
+ "step": 400
3228
+ },
3229
+ {
3230
+ "epoch": 0.04718650465966733,
3231
+ "eval_wikipedia_loss": 2.4841043949127197,
3232
+ "eval_wikipedia_runtime": 173.8643,
3233
+ "eval_wikipedia_samples_per_second": 4.038,
3234
+ "eval_wikipedia_steps_per_second": 0.173,
3235
+ "num_input_tokens_seen": 471859200,
3236
+ "step": 400
3237
+ },
3238
+ {
3239
+ "epoch": 0.04718650465966733,
3240
+ "eval_toxicity_loss": 4.2398834228515625,
3241
+ "eval_toxicity_runtime": 0.9943,
3242
+ "eval_toxicity_samples_per_second": 2.011,
3243
+ "eval_toxicity_steps_per_second": 1.006,
3244
+ "num_input_tokens_seen": 471859200,
3245
+ "step": 400
3246
+ }
3247
+ ],
3248
+ "logging_steps": 1,
3249
+ "max_steps": 8477,
3250
+ "num_input_tokens_seen": 471859200,
3251
+ "num_train_epochs": 9223372036854775807,
3252
+ "save_steps": 200,
3253
+ "stateful_callbacks": {
3254
+ "TrainerControl": {
3255
+ "args": {
3256
+ "should_epoch_stop": false,
3257
+ "should_evaluate": false,
3258
+ "should_log": false,
3259
+ "should_save": true,
3260
+ "should_training_stop": false
3261
+ },
3262
+ "attributes": {}
3263
+ }
3264
+ },
3265
+ "total_flos": 5.731724480321946e+18,
3266
+ "train_batch_size": 6,
3267
+ "trial_name": null,
3268
+ "trial_params": null
3269
+ }
checkpoint-400/training_args.bin ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:17ddf3fb8fdb9baa07c29ebd5a499cb1523c190e3c7c6707752f236dce6f27d2
3
+ size 5304