KexuanShi commited on
Commit
7ff3dbe
·
verified ·
1 Parent(s): a194742

Upload checkpoint-1323

Browse files
checkpoint-1323/config.json ADDED
@@ -0,0 +1,65 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "architectures": [
3
+ "Gemma2ForCausalLM"
4
+ ],
5
+ "attention_bias": false,
6
+ "attention_dropout": 0.0,
7
+ "attn_logit_softcapping": 50.0,
8
+ "bos_token_id": 2,
9
+ "cache_implementation": "hybrid",
10
+ "dtype": "bfloat16",
11
+ "eos_token_id": 1,
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
+ "layer_types": [
20
+ "sliding_attention",
21
+ "full_attention",
22
+ "sliding_attention",
23
+ "full_attention",
24
+ "sliding_attention",
25
+ "full_attention",
26
+ "sliding_attention",
27
+ "full_attention",
28
+ "sliding_attention",
29
+ "full_attention",
30
+ "sliding_attention",
31
+ "full_attention",
32
+ "sliding_attention",
33
+ "full_attention",
34
+ "sliding_attention",
35
+ "full_attention",
36
+ "sliding_attention",
37
+ "full_attention",
38
+ "sliding_attention",
39
+ "full_attention",
40
+ "sliding_attention",
41
+ "full_attention",
42
+ "sliding_attention",
43
+ "full_attention",
44
+ "sliding_attention",
45
+ "full_attention"
46
+ ],
47
+ "max_position_embeddings": 8192,
48
+ "model_type": "gemma2",
49
+ "num_attention_heads": 8,
50
+ "num_hidden_layers": 26,
51
+ "num_key_value_heads": 4,
52
+ "pad_token_id": 0,
53
+ "query_pre_attn_scalar": 256,
54
+ "rms_norm_eps": 1e-06,
55
+ "rope_parameters": {
56
+ "rope_theta": 10000.0,
57
+ "rope_type": "default"
58
+ },
59
+ "sliding_window": 4096,
60
+ "tie_word_embeddings": true,
61
+ "transformers_version": "5.17.0",
62
+ "use_bidirectional_attention": null,
63
+ "use_cache": false,
64
+ "vocab_size": 256000
65
+ }
checkpoint-1323/generation_config.json ADDED
@@ -0,0 +1,10 @@
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "_from_model_config": true,
3
+ "bos_token_id": 2,
4
+ "cache_implementation": "hybrid",
5
+ "eos_token_id": [
6
+ 1
7
+ ],
8
+ "pad_token_id": 0,
9
+ "transformers_version": "5.17.0"
10
+ }
checkpoint-1323/model.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:79aafaaee9f54ea2446746380e262450130986970658add1d7228997a6849fad
3
+ size 5228717512
checkpoint-1323/tokenizer.json ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:487cee8724215dcd2dde8888539e8b1bf844ceb5dbbe27f7845abda69eeb060f
3
+ size 34362872
checkpoint-1323/tokenizer_config.json ADDED
@@ -0,0 +1,20 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "backend": "tokenizers",
3
+ "bos_token": "<bos>",
4
+ "clean_up_tokenization_spaces": false,
5
+ "eos_token": "<eos>",
6
+ "extra_special_tokens": [
7
+ "<start_of_turn>",
8
+ "<end_of_turn>"
9
+ ],
10
+ "is_local": true,
11
+ "local_files_only": false,
12
+ "mask_token": "<mask>",
13
+ "model_max_length": 1000000000000000019884624838656,
14
+ "pad_token": "<pad>",
15
+ "sp_model_kwargs": {},
16
+ "spaces_between_special_tokens": false,
17
+ "tokenizer_class": "GemmaTokenizer",
18
+ "unk_token": "<unk>",
19
+ "use_default_system_prompt": false
20
+ }
checkpoint-1323/trainer_state.json ADDED
@@ -0,0 +1,1354 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "best_global_step": null,
3
+ "best_metric": null,
4
+ "best_model_checkpoint": null,
5
+ "epoch": 7.0,
6
+ "eval_steps": 500,
7
+ "global_step": 1323,
8
+ "is_hyper_param_search": false,
9
+ "is_local_process_zero": true,
10
+ "is_world_process_zero": true,
11
+ "log_history": [
12
+ {
13
+ "entropy": 1.2536803647875785,
14
+ "epoch": 0.05291005291005291,
15
+ "grad_norm": 8.8125,
16
+ "learning_rate": 3.2142857142857147e-06,
17
+ "loss": 1.1722795486450195,
18
+ "mean_token_accuracy": 0.7258133076131343,
19
+ "num_tokens": 710093.0,
20
+ "step": 10
21
+ },
22
+ {
23
+ "entropy": 0.9760579235851765,
24
+ "epoch": 0.10582010582010581,
25
+ "grad_norm": 2.65625,
26
+ "learning_rate": 6.785714285714287e-06,
27
+ "loss": 0.9763177871704102,
28
+ "mean_token_accuracy": 0.7567757397890091,
29
+ "num_tokens": 1429271.0,
30
+ "step": 20
31
+ },
32
+ {
33
+ "entropy": 0.794147289544344,
34
+ "epoch": 0.15873015873015872,
35
+ "grad_norm": 2.46875,
36
+ "learning_rate": 1.0357142857142859e-05,
37
+ "loss": 0.8046525001525879,
38
+ "mean_token_accuracy": 0.7873034350574016,
39
+ "num_tokens": 2149735.0,
40
+ "step": 30
41
+ },
42
+ {
43
+ "entropy": 0.7088500954210758,
44
+ "epoch": 0.21164021164021163,
45
+ "grad_norm": 1.4609375,
46
+ "learning_rate": 1.3928571428571429e-05,
47
+ "loss": 0.7121071815490723,
48
+ "mean_token_accuracy": 0.8070594437420369,
49
+ "num_tokens": 2868867.0,
50
+ "step": 40
51
+ },
52
+ {
53
+ "entropy": 0.683020292967558,
54
+ "epoch": 0.26455026455026454,
55
+ "grad_norm": 1.46875,
56
+ "learning_rate": 1.7500000000000002e-05,
57
+ "loss": 0.68443603515625,
58
+ "mean_token_accuracy": 0.8138657048344612,
59
+ "num_tokens": 3582476.0,
60
+ "step": 50
61
+ },
62
+ {
63
+ "entropy": 0.6564815938472748,
64
+ "epoch": 0.31746031746031744,
65
+ "grad_norm": 1.4296875,
66
+ "learning_rate": 1.999986795783372e-05,
67
+ "loss": 0.6578450202941895,
68
+ "mean_token_accuracy": 0.8192978233098984,
69
+ "num_tokens": 4287689.0,
70
+ "step": 60
71
+ },
72
+ {
73
+ "entropy": 0.6446133196353913,
74
+ "epoch": 0.37037037037037035,
75
+ "grad_norm": 1.5859375,
76
+ "learning_rate": 1.9997520638548385e-05,
77
+ "loss": 0.651158332824707,
78
+ "mean_token_accuracy": 0.8207244381308556,
79
+ "num_tokens": 5011191.0,
80
+ "step": 70
81
+ },
82
+ {
83
+ "entropy": 0.6360202118754387,
84
+ "epoch": 0.42328042328042326,
85
+ "grad_norm": 1.3984375,
86
+ "learning_rate": 1.9992239841688418e-05,
87
+ "loss": 0.6350794792175293,
88
+ "mean_token_accuracy": 0.8251068785786628,
89
+ "num_tokens": 5714842.0,
90
+ "step": 80
91
+ },
92
+ {
93
+ "entropy": 0.6189912840723991,
94
+ "epoch": 0.47619047619047616,
95
+ "grad_norm": 1.328125,
96
+ "learning_rate": 1.9984027116747914e-05,
97
+ "loss": 0.622182559967041,
98
+ "mean_token_accuracy": 0.8279523886740208,
99
+ "num_tokens": 6424614.0,
100
+ "step": 90
101
+ },
102
+ {
103
+ "entropy": 0.6159785229712724,
104
+ "epoch": 0.5291005291005291,
105
+ "grad_norm": 1.3671875,
106
+ "learning_rate": 1.997288487350878e-05,
107
+ "loss": 0.617555046081543,
108
+ "mean_token_accuracy": 0.8284852005541324,
109
+ "num_tokens": 7139778.0,
110
+ "step": 100
111
+ },
112
+ {
113
+ "entropy": 0.6129939660429955,
114
+ "epoch": 0.582010582010582,
115
+ "grad_norm": 1.328125,
116
+ "learning_rate": 1.9958816381333684e-05,
117
+ "loss": 0.6128347873687744,
118
+ "mean_token_accuracy": 0.8294584318995476,
119
+ "num_tokens": 7852540.0,
120
+ "step": 110
121
+ },
122
+ {
123
+ "entropy": 0.6052542962133884,
124
+ "epoch": 0.6349206349206349,
125
+ "grad_norm": 1.3359375,
126
+ "learning_rate": 1.994182576820673e-05,
127
+ "loss": 0.6046356201171875,
128
+ "mean_token_accuracy": 0.831394524872303,
129
+ "num_tokens": 8568056.0,
130
+ "step": 120
131
+ },
132
+ {
133
+ "entropy": 0.6013224326074124,
134
+ "epoch": 0.6878306878306878,
135
+ "grad_norm": 1.3203125,
136
+ "learning_rate": 1.9921918019522238e-05,
137
+ "loss": 0.6063410758972168,
138
+ "mean_token_accuracy": 0.8313738636672496,
139
+ "num_tokens": 9276742.0,
140
+ "step": 130
141
+ },
142
+ {
143
+ "entropy": 0.5983927462249994,
144
+ "epoch": 0.7407407407407407,
145
+ "grad_norm": 1.3125,
146
+ "learning_rate": 1.9899098976621922e-05,
147
+ "loss": 0.5977112770080566,
148
+ "mean_token_accuracy": 0.8335464030504227,
149
+ "num_tokens": 9990163.0,
150
+ "step": 140
151
+ },
152
+ {
153
+ "entropy": 0.5922725383192301,
154
+ "epoch": 0.7936507936507936,
155
+ "grad_norm": 1.5078125,
156
+ "learning_rate": 1.9873375335080915e-05,
157
+ "loss": 0.5917873382568359,
158
+ "mean_token_accuracy": 0.8347524181008339,
159
+ "num_tokens": 10710006.0,
160
+ "step": 150
161
+ },
162
+ {
163
+ "entropy": 0.5920293003320694,
164
+ "epoch": 0.8465608465608465,
165
+ "grad_norm": 1.3046875,
166
+ "learning_rate": 1.9844754642743164e-05,
167
+ "loss": 0.5962053298950195,
168
+ "mean_token_accuracy": 0.8335559628903866,
169
+ "num_tokens": 11416822.0,
170
+ "step": 160
171
+ },
172
+ {
173
+ "entropy": 0.5966673269867897,
174
+ "epoch": 0.8994708994708994,
175
+ "grad_norm": 1.3828125,
176
+ "learning_rate": 1.981324529750672e-05,
177
+ "loss": 0.6019061088562012,
178
+ "mean_token_accuracy": 0.83258865326643,
179
+ "num_tokens": 12123118.0,
180
+ "step": 170
181
+ },
182
+ {
183
+ "entropy": 0.5854565776884556,
184
+ "epoch": 0.9523809523809523,
185
+ "grad_norm": 1.359375,
186
+ "learning_rate": 1.9778856544859647e-05,
187
+ "loss": 0.5845203399658203,
188
+ "mean_token_accuracy": 0.8360438510775566,
189
+ "num_tokens": 12831058.0,
190
+ "step": 180
191
+ },
192
+ {
193
+ "entropy": 0.5759805988520383,
194
+ "epoch": 1.0052910052910053,
195
+ "grad_norm": 1.3046875,
196
+ "learning_rate": 1.9741598475167173e-05,
197
+ "loss": 0.5696067810058594,
198
+ "mean_token_accuracy": 0.839682175219059,
199
+ "num_tokens": 13547943.0,
200
+ "step": 190
201
+ },
202
+ {
203
+ "entropy": 0.5275494448840619,
204
+ "epoch": 1.0582010582010581,
205
+ "grad_norm": 1.3984375,
206
+ "learning_rate": 1.9701482020711004e-05,
207
+ "loss": 0.5299928665161133,
208
+ "mean_token_accuracy": 0.8484233327209949,
209
+ "num_tokens": 14264772.0,
210
+ "step": 200
211
+ },
212
+ {
213
+ "entropy": 0.5196751751005649,
214
+ "epoch": 1.1111111111111112,
215
+ "grad_norm": 1.2109375,
216
+ "learning_rate": 1.9658518952481543e-05,
217
+ "loss": 0.5208037853240967,
218
+ "mean_token_accuracy": 0.8509362854063511,
219
+ "num_tokens": 14985565.0,
220
+ "step": 210
221
+ },
222
+ {
223
+ "entropy": 0.5303940393030644,
224
+ "epoch": 1.164021164021164,
225
+ "grad_norm": 1.3359375,
226
+ "learning_rate": 1.9612721876724054e-05,
227
+ "loss": 0.5318955421447754,
228
+ "mean_token_accuracy": 0.8480072498321534,
229
+ "num_tokens": 15701796.0,
230
+ "step": 220
231
+ },
232
+ {
233
+ "entropy": 0.5232694145292044,
234
+ "epoch": 1.216931216931217,
235
+ "grad_norm": 1.359375,
236
+ "learning_rate": 1.9564104231239726e-05,
237
+ "loss": 0.5231654167175293,
238
+ "mean_token_accuracy": 0.8490844376385212,
239
+ "num_tokens": 16412522.0,
240
+ "step": 230
241
+ },
242
+ {
243
+ "entropy": 0.529466836899519,
244
+ "epoch": 1.2698412698412698,
245
+ "grad_norm": 1.3046875,
246
+ "learning_rate": 1.9512680281442745e-05,
247
+ "loss": 0.5309067726135254,
248
+ "mean_token_accuracy": 0.8485323786735535,
249
+ "num_tokens": 17126523.0,
250
+ "step": 240
251
+ },
252
+ {
253
+ "entropy": 0.527790617197752,
254
+ "epoch": 1.3227513227513228,
255
+ "grad_norm": 1.3125,
256
+ "learning_rate": 1.9458465116174534e-05,
257
+ "loss": 0.5262090682983398,
258
+ "mean_token_accuracy": 0.8493455089628696,
259
+ "num_tokens": 17834934.0,
260
+ "step": 250
261
+ },
262
+ {
263
+ "entropy": 0.528411227837205,
264
+ "epoch": 1.3756613756613756,
265
+ "grad_norm": 1.265625,
266
+ "learning_rate": 1.940147464327637e-05,
267
+ "loss": 0.529714298248291,
268
+ "mean_token_accuracy": 0.8480914957821369,
269
+ "num_tokens": 18553413.0,
270
+ "step": 260
271
+ },
272
+ {
273
+ "entropy": 0.516933124884963,
274
+ "epoch": 1.4285714285714286,
275
+ "grad_norm": 1.296875,
276
+ "learning_rate": 1.9341725584921712e-05,
277
+ "loss": 0.5172879219055175,
278
+ "mean_token_accuracy": 0.8517736725509166,
279
+ "num_tokens": 19270191.0,
280
+ "step": 270
281
+ },
282
+ {
283
+ "entropy": 0.5223661050200462,
284
+ "epoch": 1.4814814814814814,
285
+ "grad_norm": 1.2421875,
286
+ "learning_rate": 1.9279235472709554e-05,
287
+ "loss": 0.5231949329376221,
288
+ "mean_token_accuracy": 0.8496842503547668,
289
+ "num_tokens": 19976552.0,
290
+ "step": 280
291
+ },
292
+ {
293
+ "entropy": 0.5293067336082459,
294
+ "epoch": 1.5343915343915344,
295
+ "grad_norm": 1.2421875,
296
+ "learning_rate": 1.92140226425203e-05,
297
+ "loss": 0.5295753479003906,
298
+ "mean_token_accuracy": 0.8479710258543491,
299
+ "num_tokens": 20676613.0,
300
+ "step": 290
301
+ },
302
+ {
303
+ "entropy": 0.5302139949053526,
304
+ "epoch": 1.5873015873015874,
305
+ "grad_norm": 1.28125,
306
+ "learning_rate": 1.914610622913566e-05,
307
+ "loss": 0.530374813079834,
308
+ "mean_token_accuracy": 0.8484688423573971,
309
+ "num_tokens": 21385978.0,
310
+ "step": 300
311
+ },
312
+ {
313
+ "entropy": 0.5183596257120371,
314
+ "epoch": 1.6402116402116402,
315
+ "grad_norm": 1.1953125,
316
+ "learning_rate": 1.9075506160624068e-05,
317
+ "loss": 0.519154167175293,
318
+ "mean_token_accuracy": 0.8509341597557067,
319
+ "num_tokens": 22094336.0,
320
+ "step": 310
321
+ },
322
+ {
323
+ "entropy": 0.5201449614018202,
324
+ "epoch": 1.693121693121693,
325
+ "grad_norm": 1.328125,
326
+ "learning_rate": 1.900224315249343e-05,
327
+ "loss": 0.5171724796295166,
328
+ "mean_token_accuracy": 0.8508495308458806,
329
+ "num_tokens": 22798906.0,
330
+ "step": 320
331
+ },
332
+ {
333
+ "entropy": 0.5111908499151468,
334
+ "epoch": 1.746031746031746,
335
+ "grad_norm": 1.2265625,
336
+ "learning_rate": 1.8926338701612738e-05,
337
+ "loss": 0.511273717880249,
338
+ "mean_token_accuracy": 0.8524932019412518,
339
+ "num_tokens": 23511239.0,
340
+ "step": 330
341
+ },
342
+ {
343
+ "entropy": 0.5265866108238697,
344
+ "epoch": 1.798941798941799,
345
+ "grad_norm": 1.265625,
346
+ "learning_rate": 1.884781507990445e-05,
347
+ "loss": 0.5289421558380127,
348
+ "mean_token_accuracy": 0.8486531592905522,
349
+ "num_tokens": 24231185.0,
350
+ "step": 340
351
+ },
352
+ {
353
+ "entropy": 0.5189403794705868,
354
+ "epoch": 1.8518518518518519,
355
+ "grad_norm": 1.2578125,
356
+ "learning_rate": 1.8766695327809463e-05,
357
+ "loss": 0.5158460140228271,
358
+ "mean_token_accuracy": 0.8517816573381424,
359
+ "num_tokens": 24944285.0,
360
+ "step": 350
361
+ },
362
+ {
363
+ "entropy": 0.5055529039353133,
364
+ "epoch": 1.9047619047619047,
365
+ "grad_norm": 1.1953125,
366
+ "learning_rate": 1.8683003247526567e-05,
367
+ "loss": 0.5064782619476318,
368
+ "mean_token_accuracy": 0.8533773042261601,
369
+ "num_tokens": 25670441.0,
370
+ "step": 360
371
+ },
372
+ {
373
+ "entropy": 0.5245815236121416,
374
+ "epoch": 1.9576719576719577,
375
+ "grad_norm": 1.2421875,
376
+ "learning_rate": 1.8596763396028384e-05,
377
+ "loss": 0.5262872219085694,
378
+ "mean_token_accuracy": 0.8492341041564941,
379
+ "num_tokens": 26379786.0,
380
+ "step": 370
381
+ },
382
+ {
383
+ "entropy": 0.5117530789226293,
384
+ "epoch": 2.0105820105820107,
385
+ "grad_norm": 1.390625,
386
+ "learning_rate": 1.8508001077855876e-05,
387
+ "loss": 0.5020773887634278,
388
+ "mean_token_accuracy": 0.854968623816967,
389
+ "num_tokens": 27097166.0,
390
+ "step": 380
391
+ },
392
+ {
393
+ "entropy": 0.44686204604804514,
394
+ "epoch": 2.0634920634920633,
395
+ "grad_norm": 1.3046875,
396
+ "learning_rate": 1.8416742337693436e-05,
397
+ "loss": 0.4472813606262207,
398
+ "mean_token_accuracy": 0.8682501375675201,
399
+ "num_tokens": 27818991.0,
400
+ "step": 390
401
+ },
402
+ {
403
+ "entropy": 0.4516366470605135,
404
+ "epoch": 2.1164021164021163,
405
+ "grad_norm": 1.484375,
406
+ "learning_rate": 1.8323013952726877e-05,
407
+ "loss": 0.45279760360717775,
408
+ "mean_token_accuracy": 0.8667486943304539,
409
+ "num_tokens": 28524822.0,
410
+ "step": 400
411
+ },
412
+ {
413
+ "entropy": 0.44368173032999036,
414
+ "epoch": 2.1693121693121693,
415
+ "grad_norm": 1.328125,
416
+ "learning_rate": 1.8226843424786434e-05,
417
+ "loss": 0.4443831443786621,
418
+ "mean_token_accuracy": 0.8681288480758667,
419
+ "num_tokens": 29245634.0,
420
+ "step": 410
421
+ },
422
+ {
423
+ "entropy": 0.44680118896067145,
424
+ "epoch": 2.2222222222222223,
425
+ "grad_norm": 1.3046875,
426
+ "learning_rate": 1.8128258972277176e-05,
427
+ "loss": 0.4440959930419922,
428
+ "mean_token_accuracy": 0.8689466655254364,
429
+ "num_tokens": 29967135.0,
430
+ "step": 420
431
+ },
432
+ {
433
+ "entropy": 0.4494645904749632,
434
+ "epoch": 2.2751322751322753,
435
+ "grad_norm": 1.328125,
436
+ "learning_rate": 1.8027289521899143e-05,
437
+ "loss": 0.45146980285644533,
438
+ "mean_token_accuracy": 0.8666525632143021,
439
+ "num_tokens": 30680274.0,
440
+ "step": 430
441
+ },
442
+ {
443
+ "entropy": 0.4557589769363403,
444
+ "epoch": 2.328042328042328,
445
+ "grad_norm": 1.2734375,
446
+ "learning_rate": 1.7923964700159658e-05,
447
+ "loss": 0.45585927963256834,
448
+ "mean_token_accuracy": 0.8663338787853718,
449
+ "num_tokens": 31386859.0,
450
+ "step": 440
451
+ },
452
+ {
453
+ "entropy": 0.46223078556358815,
454
+ "epoch": 2.380952380952381,
455
+ "grad_norm": 1.390625,
456
+ "learning_rate": 1.78183148246803e-05,
457
+ "loss": 0.4587599754333496,
458
+ "mean_token_accuracy": 0.86525354757905,
459
+ "num_tokens": 32111007.0,
460
+ "step": 450
461
+ },
462
+ {
463
+ "entropy": 0.4548660650849342,
464
+ "epoch": 2.433862433862434,
465
+ "grad_norm": 1.265625,
466
+ "learning_rate": 1.77103708953011e-05,
467
+ "loss": 0.4531528949737549,
468
+ "mean_token_accuracy": 0.8663033962249755,
469
+ "num_tokens": 32819180.0,
470
+ "step": 460
471
+ },
472
+ {
473
+ "entropy": 0.4630272053182125,
474
+ "epoch": 2.4867724867724865,
475
+ "grad_norm": 1.28125,
476
+ "learning_rate": 1.7600164584984548e-05,
477
+ "loss": 0.4610149383544922,
478
+ "mean_token_accuracy": 0.8639790855348111,
479
+ "num_tokens": 33523113.0,
480
+ "step": 470
481
+ },
482
+ {
483
+ "entropy": 0.4509055320173502,
484
+ "epoch": 2.5396825396825395,
485
+ "grad_norm": 1.34375,
486
+ "learning_rate": 1.7487728230522085e-05,
487
+ "loss": 0.44989833831787107,
488
+ "mean_token_accuracy": 0.8673267938196659,
489
+ "num_tokens": 34240357.0,
490
+ "step": 480
491
+ },
492
+ {
493
+ "entropy": 0.455976203456521,
494
+ "epoch": 2.5925925925925926,
495
+ "grad_norm": 1.2265625,
496
+ "learning_rate": 1.737309482304585e-05,
497
+ "loss": 0.4550386905670166,
498
+ "mean_token_accuracy": 0.8660441562533379,
499
+ "num_tokens": 34958032.0,
500
+ "step": 490
501
+ },
502
+ {
503
+ "entropy": 0.4540476597845554,
504
+ "epoch": 2.6455026455026456,
505
+ "grad_norm": 1.328125,
506
+ "learning_rate": 1.7256297998348388e-05,
507
+ "loss": 0.45427827835083007,
508
+ "mean_token_accuracy": 0.866205920279026,
509
+ "num_tokens": 35658658.0,
510
+ "step": 500
511
+ },
512
+ {
513
+ "entropy": 0.45439887307584287,
514
+ "epoch": 2.6984126984126986,
515
+ "grad_norm": 1.3203125,
516
+ "learning_rate": 1.7137372027013212e-05,
517
+ "loss": 0.45563058853149413,
518
+ "mean_token_accuracy": 0.865429712831974,
519
+ "num_tokens": 36364754.0,
520
+ "step": 510
521
+ },
522
+ {
523
+ "entropy": 0.46197318099439144,
524
+ "epoch": 2.751322751322751,
525
+ "grad_norm": 1.296875,
526
+ "learning_rate": 1.7016351804359134e-05,
527
+ "loss": 0.46036925315856936,
528
+ "mean_token_accuracy": 0.8651507623493672,
529
+ "num_tokens": 37074678.0,
530
+ "step": 520
531
+ },
532
+ {
533
+ "entropy": 0.47125925347208975,
534
+ "epoch": 2.804232804232804,
535
+ "grad_norm": 1.296875,
536
+ "learning_rate": 1.6893272840201243e-05,
537
+ "loss": 0.4696796417236328,
538
+ "mean_token_accuracy": 0.862788101285696,
539
+ "num_tokens": 37787128.0,
540
+ "step": 530
541
+ },
542
+ {
543
+ "entropy": 0.4597961250692606,
544
+ "epoch": 2.857142857142857,
545
+ "grad_norm": 1.3203125,
546
+ "learning_rate": 1.6768171248431603e-05,
547
+ "loss": 0.45906572341918944,
548
+ "mean_token_accuracy": 0.8650128476321697,
549
+ "num_tokens": 38504013.0,
550
+ "step": 540
551
+ },
552
+ {
553
+ "entropy": 0.4489801775664091,
554
+ "epoch": 2.91005291005291,
555
+ "grad_norm": 1.3046875,
556
+ "learning_rate": 1.66410837364227e-05,
557
+ "loss": 0.4477699279785156,
558
+ "mean_token_accuracy": 0.8676867328584195,
559
+ "num_tokens": 39212782.0,
560
+ "step": 550
561
+ },
562
+ {
563
+ "entropy": 0.4513770941644907,
564
+ "epoch": 2.962962962962963,
565
+ "grad_norm": 1.2734375,
566
+ "learning_rate": 1.651204759425674e-05,
567
+ "loss": 0.4503082275390625,
568
+ "mean_token_accuracy": 0.8672237955033779,
569
+ "num_tokens": 39931947.0,
570
+ "step": 560
571
+ },
572
+ {
573
+ "entropy": 0.4483271919190884,
574
+ "epoch": 3.015873015873016,
575
+ "grad_norm": 1.34375,
576
+ "learning_rate": 1.6381100683783984e-05,
577
+ "loss": 0.4379873275756836,
578
+ "mean_token_accuracy": 0.8702801637351513,
579
+ "num_tokens": 40643225.0,
580
+ "step": 570
581
+ },
582
+ {
583
+ "entropy": 0.3994087837636471,
584
+ "epoch": 3.068783068783069,
585
+ "grad_norm": 1.375,
586
+ "learning_rate": 1.6248281427513293e-05,
587
+ "loss": 0.40430545806884766,
588
+ "mean_token_accuracy": 0.8792076490819454,
589
+ "num_tokens": 41359577.0,
590
+ "step": 580
591
+ },
592
+ {
593
+ "entropy": 0.40569392777979374,
594
+ "epoch": 3.121693121693122,
595
+ "grad_norm": 1.46875,
596
+ "learning_rate": 1.6113628797338182e-05,
597
+ "loss": 0.40410671234130857,
598
+ "mean_token_accuracy": 0.8795830741524696,
599
+ "num_tokens": 42068599.0,
600
+ "step": 590
601
+ },
602
+ {
603
+ "entropy": 0.39706668704748155,
604
+ "epoch": 3.1746031746031744,
605
+ "grad_norm": 1.453125,
606
+ "learning_rate": 1.597718230310167e-05,
607
+ "loss": 0.39309277534484866,
608
+ "mean_token_accuracy": 0.8820179499685764,
609
+ "num_tokens": 42785175.0,
610
+ "step": 600
611
+ },
612
+ {
613
+ "entropy": 0.3929748997092247,
614
+ "epoch": 3.2275132275132274,
615
+ "grad_norm": 1.3515625,
616
+ "learning_rate": 1.5838981981003273e-05,
617
+ "loss": 0.391530179977417,
618
+ "mean_token_accuracy": 0.8828775212168694,
619
+ "num_tokens": 43504295.0,
620
+ "step": 610
621
+ },
622
+ {
623
+ "entropy": 0.4006485942751169,
624
+ "epoch": 3.2804232804232805,
625
+ "grad_norm": 1.3359375,
626
+ "learning_rate": 1.5699068381851576e-05,
627
+ "loss": 0.39961411952972414,
628
+ "mean_token_accuracy": 0.8799484111368656,
629
+ "num_tokens": 44210262.0,
630
+ "step": 620
631
+ },
632
+ {
633
+ "entropy": 0.40273183323442935,
634
+ "epoch": 3.3333333333333335,
635
+ "grad_norm": 1.296875,
636
+ "learning_rate": 1.5557482559165777e-05,
637
+ "loss": 0.3995194435119629,
638
+ "mean_token_accuracy": 0.8803919807076455,
639
+ "num_tokens": 44921899.0,
640
+ "step": 630
641
+ },
642
+ {
643
+ "entropy": 0.39858098365366457,
644
+ "epoch": 3.386243386243386,
645
+ "grad_norm": 1.359375,
646
+ "learning_rate": 1.541426605712975e-05,
647
+ "loss": 0.39831721782684326,
648
+ "mean_token_accuracy": 0.8801213204860687,
649
+ "num_tokens": 45631735.0,
650
+ "step": 640
651
+ },
652
+ {
653
+ "entropy": 0.40417286232113836,
654
+ "epoch": 3.439153439153439,
655
+ "grad_norm": 1.390625,
656
+ "learning_rate": 1.526946089840214e-05,
657
+ "loss": 0.4012503147125244,
658
+ "mean_token_accuracy": 0.8794990263879299,
659
+ "num_tokens": 46339604.0,
660
+ "step": 650
661
+ },
662
+ {
663
+ "entropy": 0.4001857254654169,
664
+ "epoch": 3.492063492063492,
665
+ "grad_norm": 1.359375,
666
+ "learning_rate": 1.5123109571786034e-05,
667
+ "loss": 0.3985880374908447,
668
+ "mean_token_accuracy": 0.8811507351696491,
669
+ "num_tokens": 47060193.0,
670
+ "step": 660
671
+ },
672
+ {
673
+ "entropy": 0.40479574501514437,
674
+ "epoch": 3.544973544973545,
675
+ "grad_norm": 1.3515625,
676
+ "learning_rate": 1.4975255019761896e-05,
677
+ "loss": 0.40291218757629393,
678
+ "mean_token_accuracy": 0.8791408479213715,
679
+ "num_tokens": 47768090.0,
680
+ "step": 670
681
+ },
682
+ {
683
+ "entropy": 0.4033591251820326,
684
+ "epoch": 3.597883597883598,
685
+ "grad_norm": 1.3828125,
686
+ "learning_rate": 1.4825940625887343e-05,
687
+ "loss": 0.4013660430908203,
688
+ "mean_token_accuracy": 0.8803223237395287,
689
+ "num_tokens": 48486016.0,
690
+ "step": 680
691
+ },
692
+ {
693
+ "entropy": 0.40111338496208193,
694
+ "epoch": 3.6507936507936507,
695
+ "grad_norm": 1.3203125,
696
+ "learning_rate": 1.467521020206753e-05,
697
+ "loss": 0.4004556655883789,
698
+ "mean_token_accuracy": 0.8803155072033405,
699
+ "num_tokens": 49196144.0,
700
+ "step": 690
701
+ },
702
+ {
703
+ "entropy": 0.4051408216357231,
704
+ "epoch": 3.7037037037037037,
705
+ "grad_norm": 1.4453125,
706
+ "learning_rate": 1.4523107975699833e-05,
707
+ "loss": 0.40234980583190916,
708
+ "mean_token_accuracy": 0.8797906585037708,
709
+ "num_tokens": 49912021.0,
710
+ "step": 700
711
+ },
712
+ {
713
+ "entropy": 0.40589384697377684,
714
+ "epoch": 3.7566137566137567,
715
+ "grad_norm": 1.3828125,
716
+ "learning_rate": 1.4369678576696633e-05,
717
+ "loss": 0.40387759208679197,
718
+ "mean_token_accuracy": 0.8787553466856479,
719
+ "num_tokens": 50634262.0,
720
+ "step": 710
721
+ },
722
+ {
723
+ "entropy": 0.40110721029341223,
724
+ "epoch": 3.8095238095238093,
725
+ "grad_norm": 1.375,
726
+ "learning_rate": 1.4214967024389978e-05,
727
+ "loss": 0.3980360269546509,
728
+ "mean_token_accuracy": 0.8805328562855721,
729
+ "num_tokens": 51343452.0,
730
+ "step": 720
731
+ },
732
+ {
733
+ "entropy": 0.40014385208487513,
734
+ "epoch": 3.8624338624338623,
735
+ "grad_norm": 1.3671875,
736
+ "learning_rate": 1.4059018714321986e-05,
737
+ "loss": 0.39777488708496095,
738
+ "mean_token_accuracy": 0.8803488783538341,
739
+ "num_tokens": 52046363.0,
740
+ "step": 730
741
+ },
742
+ {
743
+ "entropy": 0.39626855254173277,
744
+ "epoch": 3.9153439153439153,
745
+ "grad_norm": 1.3671875,
746
+ "learning_rate": 1.390187940492486e-05,
747
+ "loss": 0.39633784294128416,
748
+ "mean_token_accuracy": 0.8808404296636582,
749
+ "num_tokens": 52755135.0,
750
+ "step": 740
751
+ },
752
+ {
753
+ "entropy": 0.39399982392787936,
754
+ "epoch": 3.9682539682539684,
755
+ "grad_norm": 1.3125,
756
+ "learning_rate": 1.3743595204094442e-05,
757
+ "loss": 0.3921464443206787,
758
+ "mean_token_accuracy": 0.8814935088157654,
759
+ "num_tokens": 53473566.0,
760
+ "step": 750
761
+ },
762
+ {
763
+ "entropy": 0.38536344952881335,
764
+ "epoch": 4.021164021164021,
765
+ "grad_norm": 1.3515625,
766
+ "learning_rate": 1.3584212555661187e-05,
767
+ "loss": 0.37671327590942383,
768
+ "mean_token_accuracy": 0.8863395161926746,
769
+ "num_tokens": 54190492.0,
770
+ "step": 760
771
+ },
772
+ {
773
+ "entropy": 0.353043582290411,
774
+ "epoch": 4.074074074074074,
775
+ "grad_norm": 1.375,
776
+ "learning_rate": 1.3423778225762609e-05,
777
+ "loss": 0.35420970916748046,
778
+ "mean_token_accuracy": 0.8928504519164562,
779
+ "num_tokens": 54897936.0,
780
+ "step": 770
781
+ },
782
+ {
783
+ "entropy": 0.36162243895232676,
784
+ "epoch": 4.1269841269841265,
785
+ "grad_norm": 1.5234375,
786
+ "learning_rate": 1.3262339289121129e-05,
787
+ "loss": 0.3595908164978027,
788
+ "mean_token_accuracy": 0.8911875270307064,
789
+ "num_tokens": 55609388.0,
790
+ "step": 780
791
+ },
792
+ {
793
+ "entropy": 0.35968837030231954,
794
+ "epoch": 4.1798941798941796,
795
+ "grad_norm": 1.5546875,
796
+ "learning_rate": 1.3099943115231386e-05,
797
+ "loss": 0.3570686101913452,
798
+ "mean_token_accuracy": 0.8919634193181991,
799
+ "num_tokens": 56329299.0,
800
+ "step": 790
801
+ },
802
+ {
803
+ "entropy": 0.3573555305600166,
804
+ "epoch": 4.232804232804233,
805
+ "grad_norm": 1.453125,
806
+ "learning_rate": 1.2936637354461064e-05,
807
+ "loss": 0.35538694858551023,
808
+ "mean_token_accuracy": 0.891864238679409,
809
+ "num_tokens": 57042704.0,
810
+ "step": 800
811
+ },
812
+ {
813
+ "entropy": 0.3631866227835417,
814
+ "epoch": 4.285714285714286,
815
+ "grad_norm": 1.453125,
816
+ "learning_rate": 1.2772469924069285e-05,
817
+ "loss": 0.36122586727142336,
818
+ "mean_token_accuracy": 0.8907185509800911,
819
+ "num_tokens": 57755769.0,
820
+ "step": 810
821
+ },
822
+ {
823
+ "entropy": 0.35537443719804285,
824
+ "epoch": 4.338624338624339,
825
+ "grad_norm": 1.3515625,
826
+ "learning_rate": 1.2607488994146703e-05,
827
+ "loss": 0.3507373809814453,
828
+ "mean_token_accuracy": 0.8932205140590668,
829
+ "num_tokens": 58467967.0,
830
+ "step": 820
831
+ },
832
+ {
833
+ "entropy": 0.36577242650091646,
834
+ "epoch": 4.391534391534392,
835
+ "grad_norm": 1.421875,
836
+ "learning_rate": 1.2441742973481432e-05,
837
+ "loss": 0.36283602714538576,
838
+ "mean_token_accuracy": 0.8900898195803165,
839
+ "num_tokens": 59188579.0,
840
+ "step": 830
841
+ },
842
+ {
843
+ "entropy": 0.3536319576203823,
844
+ "epoch": 4.444444444444445,
845
+ "grad_norm": 1.453125,
846
+ "learning_rate": 1.2275280495354889e-05,
847
+ "loss": 0.3507254123687744,
848
+ "mean_token_accuracy": 0.8936197653412818,
849
+ "num_tokens": 59899589.0,
850
+ "step": 840
851
+ },
852
+ {
853
+ "entropy": 0.35393124222755434,
854
+ "epoch": 4.497354497354498,
855
+ "grad_norm": 1.390625,
856
+ "learning_rate": 1.21081504032718e-05,
857
+ "loss": 0.35221099853515625,
858
+ "mean_token_accuracy": 0.8937442168593407,
859
+ "num_tokens": 60620913.0,
860
+ "step": 850
861
+ },
862
+ {
863
+ "entropy": 0.3644785340875387,
864
+ "epoch": 4.550264550264551,
865
+ "grad_norm": 1.4296875,
866
+ "learning_rate": 1.1940401736628524e-05,
867
+ "loss": 0.36148934364318847,
868
+ "mean_token_accuracy": 0.8908700153231621,
869
+ "num_tokens": 61330675.0,
870
+ "step": 860
871
+ },
872
+ {
873
+ "entropy": 0.36559525281190874,
874
+ "epoch": 4.603174603174603,
875
+ "grad_norm": 1.546875,
876
+ "learning_rate": 1.1772083716323878e-05,
877
+ "loss": 0.36388137340545657,
878
+ "mean_token_accuracy": 0.8894597724080086,
879
+ "num_tokens": 62041645.0,
880
+ "step": 870
881
+ },
882
+ {
883
+ "entropy": 0.3636305950582027,
884
+ "epoch": 4.656084656084656,
885
+ "grad_norm": 1.3671875,
886
+ "learning_rate": 1.1603245730316724e-05,
887
+ "loss": 0.35996088981628416,
888
+ "mean_token_accuracy": 0.8900512047111988,
889
+ "num_tokens": 62750446.0,
890
+ "step": 880
891
+ },
892
+ {
893
+ "entropy": 0.35667852535843847,
894
+ "epoch": 4.708994708994709,
895
+ "grad_norm": 1.4296875,
896
+ "learning_rate": 1.1433937319134512e-05,
897
+ "loss": 0.3551082372665405,
898
+ "mean_token_accuracy": 0.8923696964979172,
899
+ "num_tokens": 63460008.0,
900
+ "step": 890
901
+ },
902
+ {
903
+ "entropy": 0.35801570676267147,
904
+ "epoch": 4.761904761904762,
905
+ "grad_norm": 1.4375,
906
+ "learning_rate": 1.1264208161337091e-05,
907
+ "loss": 0.3563329458236694,
908
+ "mean_token_accuracy": 0.8912324517965317,
909
+ "num_tokens": 64164803.0,
910
+ "step": 900
911
+ },
912
+ {
913
+ "entropy": 0.36424682177603246,
914
+ "epoch": 4.814814814814815,
915
+ "grad_norm": 1.4140625,
916
+ "learning_rate": 1.1094108058940005e-05,
917
+ "loss": 0.3606156826019287,
918
+ "mean_token_accuracy": 0.8901903286576272,
919
+ "num_tokens": 64878163.0,
920
+ "step": 910
921
+ },
922
+ {
923
+ "entropy": 0.3541768804192543,
924
+ "epoch": 4.867724867724868,
925
+ "grad_norm": 1.4296875,
926
+ "learning_rate": 1.0923686922801537e-05,
927
+ "loss": 0.34981212615966795,
928
+ "mean_token_accuracy": 0.893631624430418,
929
+ "num_tokens": 65599878.0,
930
+ "step": 920
931
+ },
932
+ {
933
+ "entropy": 0.35314388051629064,
934
+ "epoch": 4.920634920634921,
935
+ "grad_norm": 1.40625,
936
+ "learning_rate": 1.0752994757977862e-05,
937
+ "loss": 0.3507479190826416,
938
+ "mean_token_accuracy": 0.8934470795094966,
939
+ "num_tokens": 66312478.0,
940
+ "step": 930
941
+ },
942
+ {
943
+ "entropy": 0.36552518904209136,
944
+ "epoch": 4.973544973544973,
945
+ "grad_norm": 1.40625,
946
+ "learning_rate": 1.058208164905052e-05,
947
+ "loss": 0.364410400390625,
948
+ "mean_token_accuracy": 0.8898321248590946,
949
+ "num_tokens": 67026652.0,
950
+ "step": 940
951
+ },
952
+ {
953
+ "entropy": 0.34920971691608427,
954
+ "epoch": 5.026455026455026,
955
+ "grad_norm": 1.4921875,
956
+ "learning_rate": 1.041099774543059e-05,
957
+ "loss": 0.33951549530029296,
958
+ "mean_token_accuracy": 0.896391049772501,
959
+ "num_tokens": 67743837.0,
960
+ "step": 950
961
+ },
962
+ {
963
+ "entropy": 0.3251554422080517,
964
+ "epoch": 5.079365079365079,
965
+ "grad_norm": 1.5078125,
966
+ "learning_rate": 1.023979324664382e-05,
967
+ "loss": 0.3226634979248047,
968
+ "mean_token_accuracy": 0.9009568735957145,
969
+ "num_tokens": 68459077.0,
970
+ "step": 960
971
+ },
972
+ {
973
+ "entropy": 0.3228011641651392,
974
+ "epoch": 5.132275132275132,
975
+ "grad_norm": 1.453125,
976
+ "learning_rate": 1.0068518387601038e-05,
977
+ "loss": 0.31957221031188965,
978
+ "mean_token_accuracy": 0.9022335946559906,
979
+ "num_tokens": 69175886.0,
980
+ "step": 970
981
+ },
982
+ {
983
+ "entropy": 0.3227011114358902,
984
+ "epoch": 5.185185185185185,
985
+ "grad_norm": 1.4765625,
986
+ "learning_rate": 9.897223423858226e-06,
987
+ "loss": 0.31845569610595703,
988
+ "mean_token_accuracy": 0.9026259168982506,
989
+ "num_tokens": 69888377.0,
990
+ "step": 980
991
+ },
992
+ {
993
+ "entropy": 0.3288699559867382,
994
+ "epoch": 5.238095238095238,
995
+ "grad_norm": 1.453125,
996
+ "learning_rate": 9.725958616870495e-06,
997
+ "loss": 0.3244814872741699,
998
+ "mean_token_accuracy": 0.9008886761963367,
999
+ "num_tokens": 70604584.0,
1000
+ "step": 990
1001
+ },
1002
+ {
1003
+ "entropy": 0.33571923822164534,
1004
+ "epoch": 5.291005291005291,
1005
+ "grad_norm": 1.546875,
1006
+ "learning_rate": 9.554774219244344e-06,
1007
+ "loss": 0.3316959857940674,
1008
+ "mean_token_accuracy": 0.899080365896225,
1009
+ "num_tokens": 71321169.0,
1010
+ "step": 1000
1011
+ },
1012
+ {
1013
+ "entropy": 0.3336105212569237,
1014
+ "epoch": 5.343915343915344,
1015
+ "grad_norm": 1.4296875,
1016
+ "learning_rate": 9.38372045999252e-06,
1017
+ "loss": 0.32957212924957274,
1018
+ "mean_token_accuracy": 0.8995683066546917,
1019
+ "num_tokens": 72041175.0,
1020
+ "step": 1010
1021
+ },
1022
+ {
1023
+ "entropy": 0.3264105889946222,
1024
+ "epoch": 5.396825396825397,
1025
+ "grad_norm": 1.5390625,
1026
+ "learning_rate": 9.212847529795784e-06,
1027
+ "loss": 0.32186915874481203,
1028
+ "mean_token_accuracy": 0.9013952367007733,
1029
+ "num_tokens": 72749079.0,
1030
+ "step": 1020
1031
+ },
1032
+ {
1033
+ "entropy": 0.3262810740619898,
1034
+ "epoch": 5.449735449735449,
1035
+ "grad_norm": 1.4375,
1036
+ "learning_rate": 9.042205566275952e-06,
1037
+ "loss": 0.32430481910705566,
1038
+ "mean_token_accuracy": 0.9012646913528443,
1039
+ "num_tokens": 73462893.0,
1040
+ "step": 1030
1041
+ },
1042
+ {
1043
+ "entropy": 0.3334294486790895,
1044
+ "epoch": 5.502645502645502,
1045
+ "grad_norm": 1.46875,
1046
+ "learning_rate": 8.871844639284479e-06,
1047
+ "loss": 0.3293310165405273,
1048
+ "mean_token_accuracy": 0.8999485336244106,
1049
+ "num_tokens": 74167133.0,
1050
+ "step": 1040
1051
+ },
1052
+ {
1053
+ "entropy": 0.3317366179078817,
1054
+ "epoch": 5.555555555555555,
1055
+ "grad_norm": 1.4609375,
1056
+ "learning_rate": 8.701814736210933e-06,
1057
+ "loss": 0.3299627065658569,
1058
+ "mean_token_accuracy": 0.8994868867099285,
1059
+ "num_tokens": 74870399.0,
1060
+ "step": 1050
1061
+ },
1062
+ {
1063
+ "entropy": 0.3392947752028704,
1064
+ "epoch": 5.608465608465608,
1065
+ "grad_norm": 1.46875,
1066
+ "learning_rate": 8.532165747315686e-06,
1067
+ "loss": 0.33389506340026853,
1068
+ "mean_token_accuracy": 0.8977861106395721,
1069
+ "num_tokens": 75590782.0,
1070
+ "step": 1060
1071
+ },
1072
+ {
1073
+ "entropy": 0.33543253857642413,
1074
+ "epoch": 5.661375661375661,
1075
+ "grad_norm": 1.4140625,
1076
+ "learning_rate": 8.362947451091072e-06,
1077
+ "loss": 0.33173198699951173,
1078
+ "mean_token_accuracy": 0.8986442036926746,
1079
+ "num_tokens": 76305788.0,
1080
+ "step": 1070
1081
+ },
1082
+ {
1083
+ "entropy": 0.3333395093679428,
1084
+ "epoch": 5.714285714285714,
1085
+ "grad_norm": 1.421875,
1086
+ "learning_rate": 8.194209499655364e-06,
1087
+ "loss": 0.33096561431884763,
1088
+ "mean_token_accuracy": 0.8996573880314827,
1089
+ "num_tokens": 77023579.0,
1090
+ "step": 1080
1091
+ },
1092
+ {
1093
+ "entropy": 0.33497946709394455,
1094
+ "epoch": 5.767195767195767,
1095
+ "grad_norm": 1.5625,
1096
+ "learning_rate": 8.026001404183831e-06,
1097
+ "loss": 0.3307910919189453,
1098
+ "mean_token_accuracy": 0.8994414336979389,
1099
+ "num_tokens": 77720821.0,
1100
+ "step": 1090
1101
+ },
1102
+ {
1103
+ "entropy": 0.3360676342621446,
1104
+ "epoch": 5.8201058201058204,
1105
+ "grad_norm": 1.453125,
1106
+ "learning_rate": 7.85837252038112e-06,
1107
+ "loss": 0.33355400562286375,
1108
+ "mean_token_accuracy": 0.8985835999250412,
1109
+ "num_tokens": 78431671.0,
1110
+ "step": 1100
1111
+ },
1112
+ {
1113
+ "entropy": 0.33601767774671315,
1114
+ "epoch": 5.8730158730158735,
1115
+ "grad_norm": 1.4609375,
1116
+ "learning_rate": 7.69137203399931e-06,
1117
+ "loss": 0.33111412525177003,
1118
+ "mean_token_accuracy": 0.8989483065903187,
1119
+ "num_tokens": 79141393.0,
1120
+ "step": 1110
1121
+ },
1122
+ {
1123
+ "entropy": 0.33282165341079234,
1124
+ "epoch": 5.925925925925926,
1125
+ "grad_norm": 1.4140625,
1126
+ "learning_rate": 7.525048946405791e-06,
1127
+ "loss": 0.33011479377746583,
1128
+ "mean_token_accuracy": 0.8995157510042191,
1129
+ "num_tokens": 79858356.0,
1130
+ "step": 1120
1131
+ },
1132
+ {
1133
+ "entropy": 0.3331317029893398,
1134
+ "epoch": 5.978835978835979,
1135
+ "grad_norm": 1.4296875,
1136
+ "learning_rate": 7.359452060205245e-06,
1137
+ "loss": 0.3283017873764038,
1138
+ "mean_token_accuracy": 0.9000704519450664,
1139
+ "num_tokens": 80566532.0,
1140
+ "step": 1130
1141
+ },
1142
+ {
1143
+ "entropy": 0.3162786137312651,
1144
+ "epoch": 6.031746031746032,
1145
+ "grad_norm": 1.46875,
1146
+ "learning_rate": 7.194629964919989e-06,
1147
+ "loss": 0.30768542289733886,
1148
+ "mean_token_accuracy": 0.9056257985532283,
1149
+ "num_tokens": 81283415.0,
1150
+ "step": 1140
1151
+ },
1152
+ {
1153
+ "entropy": 0.31019794568419456,
1154
+ "epoch": 6.084656084656085,
1155
+ "grad_norm": 1.5078125,
1156
+ "learning_rate": 7.030631022732781e-06,
1157
+ "loss": 0.3101642370223999,
1158
+ "mean_token_accuracy": 0.9054678924381733,
1159
+ "num_tokens": 81995607.0,
1160
+ "step": 1150
1161
+ },
1162
+ {
1163
+ "entropy": 0.3164690284058452,
1164
+ "epoch": 6.137566137566138,
1165
+ "grad_norm": 1.6015625,
1166
+ "learning_rate": 6.867503354296392e-06,
1167
+ "loss": 0.3106247425079346,
1168
+ "mean_token_accuracy": 0.9047419615089893,
1169
+ "num_tokens": 82712148.0,
1170
+ "step": 1160
1171
+ },
1172
+ {
1173
+ "entropy": 0.32531814724206926,
1174
+ "epoch": 6.190476190476191,
1175
+ "grad_norm": 1.5,
1176
+ "learning_rate": 6.705294824614004e-06,
1177
+ "loss": 0.3201704978942871,
1178
+ "mean_token_accuracy": 0.9021763414144516,
1179
+ "num_tokens": 83425506.0,
1180
+ "step": 1170
1181
+ },
1182
+ {
1183
+ "entropy": 0.319172390550375,
1184
+ "epoch": 6.243386243386244,
1185
+ "grad_norm": 1.5,
1186
+ "learning_rate": 6.54405302899463e-06,
1187
+ "loss": 0.31355197429656984,
1188
+ "mean_token_accuracy": 0.9040127523243427,
1189
+ "num_tokens": 84137018.0,
1190
+ "step": 1180
1191
+ },
1192
+ {
1193
+ "entropy": 0.32360128555446865,
1194
+ "epoch": 6.296296296296296,
1195
+ "grad_norm": 1.4765625,
1196
+ "learning_rate": 6.383825279087699e-06,
1197
+ "loss": 0.31948628425598147,
1198
+ "mean_token_accuracy": 0.9021987453103065,
1199
+ "num_tokens": 84849598.0,
1200
+ "step": 1190
1201
+ },
1202
+ {
1203
+ "entropy": 0.3194924917072058,
1204
+ "epoch": 6.349206349206349,
1205
+ "grad_norm": 1.5,
1206
+ "learning_rate": 6.2246585890008245e-06,
1207
+ "loss": 0.31670660972595216,
1208
+ "mean_token_accuracy": 0.9040497578680515,
1209
+ "num_tokens": 85561962.0,
1210
+ "step": 1200
1211
+ },
1212
+ {
1213
+ "entropy": 0.3129683220759034,
1214
+ "epoch": 6.402116402116402,
1215
+ "grad_norm": 1.5859375,
1216
+ "learning_rate": 6.06659966150491e-06,
1217
+ "loss": 0.3079230785369873,
1218
+ "mean_token_accuracy": 0.9053744673728943,
1219
+ "num_tokens": 86286338.0,
1220
+ "step": 1210
1221
+ },
1222
+ {
1223
+ "entropy": 0.3165762163698673,
1224
+ "epoch": 6.455026455026455,
1225
+ "grad_norm": 1.4765625,
1226
+ "learning_rate": 5.909694874330601e-06,
1227
+ "loss": 0.3120215654373169,
1228
+ "mean_token_accuracy": 0.9052867114543914,
1229
+ "num_tokens": 87001911.0,
1230
+ "step": 1220
1231
+ },
1232
+ {
1233
+ "entropy": 0.31230680998414756,
1234
+ "epoch": 6.507936507936508,
1235
+ "grad_norm": 1.5078125,
1236
+ "learning_rate": 5.753990266560105e-06,
1237
+ "loss": 0.30913171768188474,
1238
+ "mean_token_accuracy": 0.9060923457145691,
1239
+ "num_tokens": 87714571.0,
1240
+ "step": 1230
1241
+ },
1242
+ {
1243
+ "entropy": 0.3162264319136739,
1244
+ "epoch": 6.560846560846561,
1245
+ "grad_norm": 1.5546875,
1246
+ "learning_rate": 5.599531525118373e-06,
1247
+ "loss": 0.3127452850341797,
1248
+ "mean_token_accuracy": 0.9043003864586353,
1249
+ "num_tokens": 88417674.0,
1250
+ "step": 1240
1251
+ },
1252
+ {
1253
+ "entropy": 0.3162810893729329,
1254
+ "epoch": 6.613756613756614,
1255
+ "grad_norm": 1.46875,
1256
+ "learning_rate": 5.446363971367638e-06,
1257
+ "loss": 0.30910825729370117,
1258
+ "mean_token_accuracy": 0.9054918497800827,
1259
+ "num_tokens": 89123522.0,
1260
+ "step": 1250
1261
+ },
1262
+ {
1263
+ "entropy": 0.31419141963124275,
1264
+ "epoch": 6.666666666666667,
1265
+ "grad_norm": 1.5,
1266
+ "learning_rate": 5.29453254780917e-06,
1267
+ "loss": 0.31052522659301757,
1268
+ "mean_token_accuracy": 0.9056126736104488,
1269
+ "num_tokens": 89838831.0,
1270
+ "step": 1260
1271
+ },
1272
+ {
1273
+ "entropy": 0.3123618332669139,
1274
+ "epoch": 6.71957671957672,
1275
+ "grad_norm": 1.4453125,
1276
+ "learning_rate": 5.1440818048962575e-06,
1277
+ "loss": 0.3084896087646484,
1278
+ "mean_token_accuracy": 0.905338567495346,
1279
+ "num_tokens": 90546066.0,
1280
+ "step": 1270
1281
+ },
1282
+ {
1283
+ "entropy": 0.31750516016036273,
1284
+ "epoch": 6.772486772486772,
1285
+ "grad_norm": 1.5390625,
1286
+ "learning_rate": 4.995055887962167e-06,
1287
+ "loss": 0.31345534324645996,
1288
+ "mean_token_accuracy": 0.9047295741736889,
1289
+ "num_tokens": 91257301.0,
1290
+ "step": 1280
1291
+ },
1292
+ {
1293
+ "entropy": 0.321336280182004,
1294
+ "epoch": 6.825396825396825,
1295
+ "grad_norm": 1.5703125,
1296
+ "learning_rate": 4.8474985242669965e-06,
1297
+ "loss": 0.31727027893066406,
1298
+ "mean_token_accuracy": 0.9030331708490849,
1299
+ "num_tokens": 91961075.0,
1300
+ "step": 1290
1301
+ },
1302
+ {
1303
+ "entropy": 0.32664828822016717,
1304
+ "epoch": 6.878306878306878,
1305
+ "grad_norm": 1.515625,
1306
+ "learning_rate": 4.701453010167217e-06,
1307
+ "loss": 0.3214175462722778,
1308
+ "mean_token_accuracy": 0.902243759483099,
1309
+ "num_tokens": 92677769.0,
1310
+ "step": 1300
1311
+ },
1312
+ {
1313
+ "entropy": 0.31388139445334673,
1314
+ "epoch": 6.931216931216931,
1315
+ "grad_norm": 1.5078125,
1316
+ "learning_rate": 4.556962198411631e-06,
1317
+ "loss": 0.309248685836792,
1318
+ "mean_token_accuracy": 0.9059560939669609,
1319
+ "num_tokens": 93390418.0,
1320
+ "step": 1310
1321
+ },
1322
+ {
1323
+ "entropy": 0.3076509850099683,
1324
+ "epoch": 6.984126984126984,
1325
+ "grad_norm": 1.515625,
1326
+ "learning_rate": 4.414068485567473e-06,
1327
+ "loss": 0.3037854194641113,
1328
+ "mean_token_accuracy": 0.9067774422466754,
1329
+ "num_tokens": 94107638.0,
1330
+ "step": 1320
1331
+ }
1332
+ ],
1333
+ "logging_steps": 10,
1334
+ "max_steps": 1890,
1335
+ "num_input_tokens_seen": 0,
1336
+ "num_train_epochs": 10,
1337
+ "save_steps": 500,
1338
+ "stateful_callbacks": {
1339
+ "TrainerControl": {
1340
+ "args": {
1341
+ "should_epoch_stop": false,
1342
+ "should_evaluate": false,
1343
+ "should_log": false,
1344
+ "should_save": true,
1345
+ "should_training_stop": false
1346
+ },
1347
+ "attributes": {}
1348
+ }
1349
+ },
1350
+ "total_flos": 1.702936402599084e+18,
1351
+ "train_batch_size": 8,
1352
+ "trial_name": null,
1353
+ "trial_params": null
1354
+ }
checkpoint-1323/training_args.bin ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:b8c85992cf63551e842ab98a6f49581a6e79e2e3e8210ccdf817e4d5ce49c466
3
+ size 5841