KexuanShi commited on
Commit
0541efe
·
verified ·
1 Parent(s): 3539a01

Upload checkpoint-1701

Browse files
checkpoint-1701/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-1701/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-1701/model.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:7b7bddd6307dfebb322c22cb7fcda978a2126f8c624242466007ac58acf4c2bd
3
+ size 5228717512
checkpoint-1701/tokenizer.json ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:487cee8724215dcd2dde8888539e8b1bf844ceb5dbbe27f7845abda69eeb060f
3
+ size 34362872
checkpoint-1701/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-1701/trainer_state.json ADDED
@@ -0,0 +1,1734 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "best_global_step": null,
3
+ "best_metric": null,
4
+ "best_model_checkpoint": null,
5
+ "epoch": 9.0,
6
+ "eval_steps": 500,
7
+ "global_step": 1701,
8
+ "is_hyper_param_search": false,
9
+ "is_local_process_zero": true,
10
+ "is_world_process_zero": true,
11
+ "log_history": [
12
+ {
13
+ "entropy": 1.7671512730419636,
14
+ "epoch": 0.05291005291005291,
15
+ "grad_norm": 14.8125,
16
+ "learning_rate": 3.2142857142857147e-06,
17
+ "loss": 1.706784439086914,
18
+ "mean_token_accuracy": 0.6392517775297165,
19
+ "num_tokens": 710093.0,
20
+ "step": 10
21
+ },
22
+ {
23
+ "entropy": 1.4565640933811665,
24
+ "epoch": 0.10582010582010581,
25
+ "grad_norm": 6.1875,
26
+ "learning_rate": 6.785714285714287e-06,
27
+ "loss": 1.4239604949951172,
28
+ "mean_token_accuracy": 0.6772899225354194,
29
+ "num_tokens": 1429271.0,
30
+ "step": 20
31
+ },
32
+ {
33
+ "entropy": 1.0071680538356305,
34
+ "epoch": 0.15873015873015872,
35
+ "grad_norm": 3.109375,
36
+ "learning_rate": 1.0357142857142859e-05,
37
+ "loss": 1.048217487335205,
38
+ "mean_token_accuracy": 0.7393087036907673,
39
+ "num_tokens": 2149735.0,
40
+ "step": 30
41
+ },
42
+ {
43
+ "entropy": 0.8510661780834198,
44
+ "epoch": 0.21164021164021163,
45
+ "grad_norm": 1.5703125,
46
+ "learning_rate": 1.3928571428571429e-05,
47
+ "loss": 0.8619060516357422,
48
+ "mean_token_accuracy": 0.7765810377895832,
49
+ "num_tokens": 2868867.0,
50
+ "step": 40
51
+ },
52
+ {
53
+ "entropy": 0.7774681895971298,
54
+ "epoch": 0.26455026455026454,
55
+ "grad_norm": 1.578125,
56
+ "learning_rate": 1.7500000000000002e-05,
57
+ "loss": 0.7852191925048828,
58
+ "mean_token_accuracy": 0.7921242833137512,
59
+ "num_tokens": 3582476.0,
60
+ "step": 50
61
+ },
62
+ {
63
+ "entropy": 0.7451388858258724,
64
+ "epoch": 0.31746031746031744,
65
+ "grad_norm": 1.5859375,
66
+ "learning_rate": 1.999986795783372e-05,
67
+ "loss": 0.748009204864502,
68
+ "mean_token_accuracy": 0.7999738156795502,
69
+ "num_tokens": 4287689.0,
70
+ "step": 60
71
+ },
72
+ {
73
+ "entropy": 0.729080930352211,
74
+ "epoch": 0.37037037037037035,
75
+ "grad_norm": 1.4921875,
76
+ "learning_rate": 1.9997520638548385e-05,
77
+ "loss": 0.7400224685668946,
78
+ "mean_token_accuracy": 0.801699859648943,
79
+ "num_tokens": 5011191.0,
80
+ "step": 70
81
+ },
82
+ {
83
+ "entropy": 0.7048044994473457,
84
+ "epoch": 0.42328042328042326,
85
+ "grad_norm": 1.4453125,
86
+ "learning_rate": 1.9992239841688418e-05,
87
+ "loss": 0.7069188117980957,
88
+ "mean_token_accuracy": 0.8093748845160007,
89
+ "num_tokens": 5714842.0,
90
+ "step": 80
91
+ },
92
+ {
93
+ "entropy": 0.6840555652976036,
94
+ "epoch": 0.47619047619047616,
95
+ "grad_norm": 1.3125,
96
+ "learning_rate": 1.9984027116747914e-05,
97
+ "loss": 0.6863717555999755,
98
+ "mean_token_accuracy": 0.8138940587639809,
99
+ "num_tokens": 6424614.0,
100
+ "step": 90
101
+ },
102
+ {
103
+ "entropy": 0.6870591521263123,
104
+ "epoch": 0.5291005291005291,
105
+ "grad_norm": 1.4140625,
106
+ "learning_rate": 1.997288487350878e-05,
107
+ "loss": 0.6901006698608398,
108
+ "mean_token_accuracy": 0.8129441156983376,
109
+ "num_tokens": 7139778.0,
110
+ "step": 100
111
+ },
112
+ {
113
+ "entropy": 0.6756624668836594,
114
+ "epoch": 0.582010582010582,
115
+ "grad_norm": 1.40625,
116
+ "learning_rate": 1.9958816381333684e-05,
117
+ "loss": 0.6771352767944336,
118
+ "mean_token_accuracy": 0.81553890183568,
119
+ "num_tokens": 7852540.0,
120
+ "step": 110
121
+ },
122
+ {
123
+ "entropy": 0.6655688174068928,
124
+ "epoch": 0.6349206349206349,
125
+ "grad_norm": 1.34375,
126
+ "learning_rate": 1.994182576820673e-05,
127
+ "loss": 0.667414665222168,
128
+ "mean_token_accuracy": 0.8177205875515938,
129
+ "num_tokens": 8568056.0,
130
+ "step": 120
131
+ },
132
+ {
133
+ "entropy": 0.6643485642969609,
134
+ "epoch": 0.6878306878306878,
135
+ "grad_norm": 1.296875,
136
+ "learning_rate": 1.9921918019522238e-05,
137
+ "loss": 0.6721675872802735,
138
+ "mean_token_accuracy": 0.816884483397007,
139
+ "num_tokens": 9276742.0,
140
+ "step": 130
141
+ },
142
+ {
143
+ "entropy": 0.6590477801859379,
144
+ "epoch": 0.7407407407407407,
145
+ "grad_norm": 1.3515625,
146
+ "learning_rate": 1.9899098976621922e-05,
147
+ "loss": 0.6590158462524414,
148
+ "mean_token_accuracy": 0.8199199490249157,
149
+ "num_tokens": 9990163.0,
150
+ "step": 140
151
+ },
152
+ {
153
+ "entropy": 0.649939077347517,
154
+ "epoch": 0.7936507936507936,
155
+ "grad_norm": 1.4609375,
156
+ "learning_rate": 1.9873375335080915e-05,
157
+ "loss": 0.6501528263092041,
158
+ "mean_token_accuracy": 0.82164012119174,
159
+ "num_tokens": 10710006.0,
160
+ "step": 150
161
+ },
162
+ {
163
+ "entropy": 0.6468971997499466,
164
+ "epoch": 0.8465608465608465,
165
+ "grad_norm": 1.296875,
166
+ "learning_rate": 1.9844754642743164e-05,
167
+ "loss": 0.6529253959655762,
168
+ "mean_token_accuracy": 0.820851170271635,
169
+ "num_tokens": 11416822.0,
170
+ "step": 160
171
+ },
172
+ {
173
+ "entropy": 0.6556476138532161,
174
+ "epoch": 0.8994708994708994,
175
+ "grad_norm": 1.296875,
176
+ "learning_rate": 1.981324529750672e-05,
177
+ "loss": 0.6604087829589844,
178
+ "mean_token_accuracy": 0.8195641197264194,
179
+ "num_tokens": 12123118.0,
180
+ "step": 170
181
+ },
182
+ {
183
+ "entropy": 0.6358714967966079,
184
+ "epoch": 0.9523809523809523,
185
+ "grad_norm": 1.3515625,
186
+ "learning_rate": 1.9778856544859647e-05,
187
+ "loss": 0.6365831375122071,
188
+ "mean_token_accuracy": 0.8241398617625236,
189
+ "num_tokens": 12831058.0,
190
+ "step": 180
191
+ },
192
+ {
193
+ "entropy": 0.6277156263589859,
194
+ "epoch": 1.0052910052910053,
195
+ "grad_norm": 1.25,
196
+ "learning_rate": 1.9741598475167173e-05,
197
+ "loss": 0.6229054927825928,
198
+ "mean_token_accuracy": 0.827630003541708,
199
+ "num_tokens": 13547943.0,
200
+ "step": 190
201
+ },
202
+ {
203
+ "entropy": 0.5929765947163105,
204
+ "epoch": 1.0582010582010581,
205
+ "grad_norm": 1.421875,
206
+ "learning_rate": 1.9701482020711004e-05,
207
+ "loss": 0.5976824760437012,
208
+ "mean_token_accuracy": 0.8327139303088188,
209
+ "num_tokens": 14264772.0,
210
+ "step": 200
211
+ },
212
+ {
213
+ "entropy": 0.5882044766098261,
214
+ "epoch": 1.1111111111111112,
215
+ "grad_norm": 1.1953125,
216
+ "learning_rate": 1.9658518952481543e-05,
217
+ "loss": 0.5882174968719482,
218
+ "mean_token_accuracy": 0.8350941069424153,
219
+ "num_tokens": 14985565.0,
220
+ "step": 210
221
+ },
222
+ {
223
+ "entropy": 0.5916924901306629,
224
+ "epoch": 1.164021164021164,
225
+ "grad_norm": 1.4140625,
226
+ "learning_rate": 1.9612721876724054e-05,
227
+ "loss": 0.5934844017028809,
228
+ "mean_token_accuracy": 0.8336228899657726,
229
+ "num_tokens": 15701796.0,
230
+ "step": 220
231
+ },
232
+ {
233
+ "entropy": 0.5877517074346542,
234
+ "epoch": 1.216931216931217,
235
+ "grad_norm": 1.34375,
236
+ "learning_rate": 1.9564104231239726e-05,
237
+ "loss": 0.589122200012207,
238
+ "mean_token_accuracy": 0.8334487572312355,
239
+ "num_tokens": 16412522.0,
240
+ "step": 230
241
+ },
242
+ {
243
+ "entropy": 0.5912774313241244,
244
+ "epoch": 1.2698412698412698,
245
+ "grad_norm": 1.3359375,
246
+ "learning_rate": 1.9512680281442745e-05,
247
+ "loss": 0.5940174102783203,
248
+ "mean_token_accuracy": 0.83340530321002,
249
+ "num_tokens": 17126523.0,
250
+ "step": 240
251
+ },
252
+ {
253
+ "entropy": 0.5930401895195245,
254
+ "epoch": 1.3227513227513228,
255
+ "grad_norm": 1.2265625,
256
+ "learning_rate": 1.9458465116174534e-05,
257
+ "loss": 0.5915824890136718,
258
+ "mean_token_accuracy": 0.8344384111464024,
259
+ "num_tokens": 17834934.0,
260
+ "step": 250
261
+ },
262
+ {
263
+ "entropy": 0.5963230866938829,
264
+ "epoch": 1.3756613756613756,
265
+ "grad_norm": 1.296875,
266
+ "learning_rate": 1.940147464327637e-05,
267
+ "loss": 0.5983511924743652,
268
+ "mean_token_accuracy": 0.832802091538906,
269
+ "num_tokens": 18553413.0,
270
+ "step": 260
271
+ },
272
+ {
273
+ "entropy": 0.5813015405088663,
274
+ "epoch": 1.4285714285714286,
275
+ "grad_norm": 1.28125,
276
+ "learning_rate": 1.9341725584921712e-05,
277
+ "loss": 0.5832551956176758,
278
+ "mean_token_accuracy": 0.8360917493700981,
279
+ "num_tokens": 19270191.0,
280
+ "step": 270
281
+ },
282
+ {
283
+ "entropy": 0.5801753208041192,
284
+ "epoch": 1.4814814814814814,
285
+ "grad_norm": 1.3046875,
286
+ "learning_rate": 1.9279235472709554e-05,
287
+ "loss": 0.5805642127990722,
288
+ "mean_token_accuracy": 0.8362542942166329,
289
+ "num_tokens": 19976552.0,
290
+ "step": 280
291
+ },
292
+ {
293
+ "entropy": 0.5879705954343081,
294
+ "epoch": 1.5343915343915344,
295
+ "grad_norm": 1.34375,
296
+ "learning_rate": 1.92140226425203e-05,
297
+ "loss": 0.5905383110046387,
298
+ "mean_token_accuracy": 0.8341313540935517,
299
+ "num_tokens": 20676613.0,
300
+ "step": 290
301
+ },
302
+ {
303
+ "entropy": 0.5889205444604159,
304
+ "epoch": 1.5873015873015874,
305
+ "grad_norm": 1.2578125,
306
+ "learning_rate": 1.914610622913566e-05,
307
+ "loss": 0.5913270950317383,
308
+ "mean_token_accuracy": 0.8349448375403881,
309
+ "num_tokens": 21385978.0,
310
+ "step": 300
311
+ },
312
+ {
313
+ "entropy": 0.5768780928105116,
314
+ "epoch": 1.6402116402116402,
315
+ "grad_norm": 1.234375,
316
+ "learning_rate": 1.9075506160624068e-05,
317
+ "loss": 0.5772612571716309,
318
+ "mean_token_accuracy": 0.8369500823318958,
319
+ "num_tokens": 22094336.0,
320
+ "step": 310
321
+ },
322
+ {
323
+ "entropy": 0.5735550723969937,
324
+ "epoch": 1.693121693121693,
325
+ "grad_norm": 1.3203125,
326
+ "learning_rate": 1.900224315249343e-05,
327
+ "loss": 0.5721980094909668,
328
+ "mean_token_accuracy": 0.8380511656403542,
329
+ "num_tokens": 22798906.0,
330
+ "step": 320
331
+ },
332
+ {
333
+ "entropy": 0.57716367803514,
334
+ "epoch": 1.746031746031746,
335
+ "grad_norm": 1.2109375,
336
+ "learning_rate": 1.8926338701612738e-05,
337
+ "loss": 0.5769223690032959,
338
+ "mean_token_accuracy": 0.8371222764253616,
339
+ "num_tokens": 23511239.0,
340
+ "step": 330
341
+ },
342
+ {
343
+ "entropy": 0.582606166601181,
344
+ "epoch": 1.798941798941799,
345
+ "grad_norm": 1.3671875,
346
+ "learning_rate": 1.884781507990445e-05,
347
+ "loss": 0.5855788230895996,
348
+ "mean_token_accuracy": 0.8356470488011837,
349
+ "num_tokens": 24231185.0,
350
+ "step": 340
351
+ },
352
+ {
353
+ "entropy": 0.5729806918650866,
354
+ "epoch": 1.8518518518518519,
355
+ "grad_norm": 1.21875,
356
+ "learning_rate": 1.8766695327809463e-05,
357
+ "loss": 0.5730391502380371,
358
+ "mean_token_accuracy": 0.8385063141584397,
359
+ "num_tokens": 24944285.0,
360
+ "step": 350
361
+ },
362
+ {
363
+ "entropy": 0.5666072227060794,
364
+ "epoch": 1.9047619047619047,
365
+ "grad_norm": 1.2109375,
366
+ "learning_rate": 1.8683003247526567e-05,
367
+ "loss": 0.5674479007720947,
368
+ "mean_token_accuracy": 0.8392404697835445,
369
+ "num_tokens": 25670441.0,
370
+ "step": 360
371
+ },
372
+ {
373
+ "entropy": 0.5809141755104065,
374
+ "epoch": 1.9576719576719577,
375
+ "grad_norm": 1.2890625,
376
+ "learning_rate": 1.8596763396028384e-05,
377
+ "loss": 0.5828475952148438,
378
+ "mean_token_accuracy": 0.8360860221087932,
379
+ "num_tokens": 26379786.0,
380
+ "step": 370
381
+ },
382
+ {
383
+ "entropy": 0.5695893839001656,
384
+ "epoch": 2.0105820105820107,
385
+ "grad_norm": 1.359375,
386
+ "learning_rate": 1.8508001077855876e-05,
387
+ "loss": 0.5633818626403808,
388
+ "mean_token_accuracy": 0.8408532224595546,
389
+ "num_tokens": 27097166.0,
390
+ "step": 380
391
+ },
392
+ {
393
+ "entropy": 0.5167598329484463,
394
+ "epoch": 2.0634920634920633,
395
+ "grad_norm": 1.2265625,
396
+ "learning_rate": 1.8416742337693436e-05,
397
+ "loss": 0.5171660423278809,
398
+ "mean_token_accuracy": 0.8513922415673733,
399
+ "num_tokens": 27818991.0,
400
+ "step": 390
401
+ },
402
+ {
403
+ "entropy": 0.5240574903786183,
404
+ "epoch": 2.1164021164021163,
405
+ "grad_norm": 1.5546875,
406
+ "learning_rate": 1.8323013952726877e-05,
407
+ "loss": 0.5252029418945312,
408
+ "mean_token_accuracy": 0.8488071605563163,
409
+ "num_tokens": 28524822.0,
410
+ "step": 400
411
+ },
412
+ {
413
+ "entropy": 0.5178623419255018,
414
+ "epoch": 2.1693121693121693,
415
+ "grad_norm": 1.3125,
416
+ "learning_rate": 1.8226843424786434e-05,
417
+ "loss": 0.5203696250915527,
418
+ "mean_token_accuracy": 0.8495552279055119,
419
+ "num_tokens": 29245634.0,
420
+ "step": 410
421
+ },
422
+ {
423
+ "entropy": 0.5194813564419747,
424
+ "epoch": 2.2222222222222223,
425
+ "grad_norm": 1.25,
426
+ "learning_rate": 1.8128258972277176e-05,
427
+ "loss": 0.5169014453887939,
428
+ "mean_token_accuracy": 0.8509399227797985,
429
+ "num_tokens": 29967135.0,
430
+ "step": 420
431
+ },
432
+ {
433
+ "entropy": 0.522173210978508,
434
+ "epoch": 2.2751322751322753,
435
+ "grad_norm": 1.25,
436
+ "learning_rate": 1.8027289521899143e-05,
437
+ "loss": 0.5245429992675781,
438
+ "mean_token_accuracy": 0.848798380792141,
439
+ "num_tokens": 30680274.0,
440
+ "step": 430
441
+ },
442
+ {
443
+ "entropy": 0.5265159368515014,
444
+ "epoch": 2.328042328042328,
445
+ "grad_norm": 1.296875,
446
+ "learning_rate": 1.7923964700159658e-05,
447
+ "loss": 0.5277297019958496,
448
+ "mean_token_accuracy": 0.8482260897755622,
449
+ "num_tokens": 31386859.0,
450
+ "step": 440
451
+ },
452
+ {
453
+ "entropy": 0.5307009011507035,
454
+ "epoch": 2.380952380952381,
455
+ "grad_norm": 1.359375,
456
+ "learning_rate": 1.78183148246803e-05,
457
+ "loss": 0.5284142971038819,
458
+ "mean_token_accuracy": 0.8480213887989521,
459
+ "num_tokens": 32111007.0,
460
+ "step": 450
461
+ },
462
+ {
463
+ "entropy": 0.5215292543172836,
464
+ "epoch": 2.433862433862434,
465
+ "grad_norm": 1.203125,
466
+ "learning_rate": 1.77103708953011e-05,
467
+ "loss": 0.5223949909210205,
468
+ "mean_token_accuracy": 0.8494340799748897,
469
+ "num_tokens": 32819180.0,
470
+ "step": 460
471
+ },
472
+ {
473
+ "entropy": 0.5321523480117321,
474
+ "epoch": 2.4867724867724865,
475
+ "grad_norm": 1.2578125,
476
+ "learning_rate": 1.7600164584984548e-05,
477
+ "loss": 0.5305499076843262,
478
+ "mean_token_accuracy": 0.8471677340567112,
479
+ "num_tokens": 33523113.0,
480
+ "step": 470
481
+ },
482
+ {
483
+ "entropy": 0.5240828461945057,
484
+ "epoch": 2.5396825396825395,
485
+ "grad_norm": 1.3046875,
486
+ "learning_rate": 1.7487728230522085e-05,
487
+ "loss": 0.5236708164215088,
488
+ "mean_token_accuracy": 0.8491978161036968,
489
+ "num_tokens": 34240357.0,
490
+ "step": 480
491
+ },
492
+ {
493
+ "entropy": 0.5249651741236449,
494
+ "epoch": 2.5925925925925926,
495
+ "grad_norm": 1.21875,
496
+ "learning_rate": 1.737309482304585e-05,
497
+ "loss": 0.5244390487670898,
498
+ "mean_token_accuracy": 0.8490321710705757,
499
+ "num_tokens": 34958032.0,
500
+ "step": 490
501
+ },
502
+ {
503
+ "entropy": 0.5262073375284672,
504
+ "epoch": 2.6455026455026456,
505
+ "grad_norm": 1.28125,
506
+ "learning_rate": 1.7256297998348388e-05,
507
+ "loss": 0.5262768268585205,
508
+ "mean_token_accuracy": 0.8484946198761463,
509
+ "num_tokens": 35658658.0,
510
+ "step": 500
511
+ },
512
+ {
513
+ "entropy": 0.5266173977404833,
514
+ "epoch": 2.6984126984126986,
515
+ "grad_norm": 1.34375,
516
+ "learning_rate": 1.7137372027013212e-05,
517
+ "loss": 0.5282751083374023,
518
+ "mean_token_accuracy": 0.8476969927549363,
519
+ "num_tokens": 36364754.0,
520
+ "step": 510
521
+ },
522
+ {
523
+ "entropy": 0.5260804291814566,
524
+ "epoch": 2.751322751322751,
525
+ "grad_norm": 1.25,
526
+ "learning_rate": 1.7016351804359134e-05,
527
+ "loss": 0.5255792140960693,
528
+ "mean_token_accuracy": 0.8494147725403309,
529
+ "num_tokens": 37074678.0,
530
+ "step": 520
531
+ },
532
+ {
533
+ "entropy": 0.5435763075947762,
534
+ "epoch": 2.804232804232804,
535
+ "grad_norm": 1.4296875,
536
+ "learning_rate": 1.6893272840201243e-05,
537
+ "loss": 0.5404428482055664,
538
+ "mean_token_accuracy": 0.8453655727207661,
539
+ "num_tokens": 37787128.0,
540
+ "step": 530
541
+ },
542
+ {
543
+ "entropy": 0.5258786808699369,
544
+ "epoch": 2.857142857142857,
545
+ "grad_norm": 1.265625,
546
+ "learning_rate": 1.6768171248431603e-05,
547
+ "loss": 0.5282983303070068,
548
+ "mean_token_accuracy": 0.8478383302688599,
549
+ "num_tokens": 38504013.0,
550
+ "step": 540
551
+ },
552
+ {
553
+ "entropy": 0.5149568922817707,
554
+ "epoch": 2.91005291005291,
555
+ "grad_norm": 1.2109375,
556
+ "learning_rate": 1.66410837364227e-05,
557
+ "loss": 0.5157745361328125,
558
+ "mean_token_accuracy": 0.8509050741791725,
559
+ "num_tokens": 39212782.0,
560
+ "step": 550
561
+ },
562
+ {
563
+ "entropy": 0.5198382031172514,
564
+ "epoch": 2.962962962962963,
565
+ "grad_norm": 1.296875,
566
+ "learning_rate": 1.651204759425674e-05,
567
+ "loss": 0.5191741943359375,
568
+ "mean_token_accuracy": 0.8499824590981007,
569
+ "num_tokens": 39931947.0,
570
+ "step": 560
571
+ },
572
+ {
573
+ "entropy": 0.5168990429490805,
574
+ "epoch": 3.015873015873016,
575
+ "grad_norm": 1.2265625,
576
+ "learning_rate": 1.6381100683783984e-05,
577
+ "loss": 0.5093559265136719,
578
+ "mean_token_accuracy": 0.8528587318956852,
579
+ "num_tokens": 40643225.0,
580
+ "step": 570
581
+ },
582
+ {
583
+ "entropy": 0.4868734121322632,
584
+ "epoch": 3.068783068783069,
585
+ "grad_norm": 1.2734375,
586
+ "learning_rate": 1.6248281427513293e-05,
587
+ "loss": 0.49291515350341797,
588
+ "mean_token_accuracy": 0.8561817370355129,
589
+ "num_tokens": 41359577.0,
590
+ "step": 580
591
+ },
592
+ {
593
+ "entropy": 0.490956873819232,
594
+ "epoch": 3.121693121693122,
595
+ "grad_norm": 1.34375,
596
+ "learning_rate": 1.6113628797338182e-05,
597
+ "loss": 0.48886570930480955,
598
+ "mean_token_accuracy": 0.85729940533638,
599
+ "num_tokens": 42068599.0,
600
+ "step": 590
601
+ },
602
+ {
603
+ "entropy": 0.47903846949338913,
604
+ "epoch": 3.1746031746031744,
605
+ "grad_norm": 1.265625,
606
+ "learning_rate": 1.597718230310167e-05,
607
+ "loss": 0.4768198013305664,
608
+ "mean_token_accuracy": 0.8604256264865399,
609
+ "num_tokens": 42785175.0,
610
+ "step": 600
611
+ },
612
+ {
613
+ "entropy": 0.4701235756278038,
614
+ "epoch": 3.2275132275132274,
615
+ "grad_norm": 1.21875,
616
+ "learning_rate": 1.5838981981003273e-05,
617
+ "loss": 0.4688236713409424,
618
+ "mean_token_accuracy": 0.8626316115260124,
619
+ "num_tokens": 43504295.0,
620
+ "step": 610
621
+ },
622
+ {
623
+ "entropy": 0.4784398306161165,
624
+ "epoch": 3.2804232804232805,
625
+ "grad_norm": 1.2734375,
626
+ "learning_rate": 1.5699068381851576e-05,
627
+ "loss": 0.47977123260498045,
628
+ "mean_token_accuracy": 0.8589689329266548,
629
+ "num_tokens": 44210262.0,
630
+ "step": 620
631
+ },
632
+ {
633
+ "entropy": 0.48374257422983646,
634
+ "epoch": 3.3333333333333335,
635
+ "grad_norm": 1.28125,
636
+ "learning_rate": 1.5557482559165777e-05,
637
+ "loss": 0.4810905456542969,
638
+ "mean_token_accuracy": 0.8588056646287441,
639
+ "num_tokens": 44921899.0,
640
+ "step": 630
641
+ },
642
+ {
643
+ "entropy": 0.47944297045469286,
644
+ "epoch": 3.386243386243386,
645
+ "grad_norm": 1.234375,
646
+ "learning_rate": 1.541426605712975e-05,
647
+ "loss": 0.480026912689209,
648
+ "mean_token_accuracy": 0.8588721737265587,
649
+ "num_tokens": 45631735.0,
650
+ "step": 640
651
+ },
652
+ {
653
+ "entropy": 0.4891296233981848,
654
+ "epoch": 3.439153439153439,
655
+ "grad_norm": 1.515625,
656
+ "learning_rate": 1.526946089840214e-05,
657
+ "loss": 0.4879168510437012,
658
+ "mean_token_accuracy": 0.8568189851939678,
659
+ "num_tokens": 46339604.0,
660
+ "step": 650
661
+ },
662
+ {
663
+ "entropy": 0.4852400213479996,
664
+ "epoch": 3.492063492063492,
665
+ "grad_norm": 1.2578125,
666
+ "learning_rate": 1.5123109571786034e-05,
667
+ "loss": 0.48280906677246094,
668
+ "mean_token_accuracy": 0.8593833275139332,
669
+ "num_tokens": 47060193.0,
670
+ "step": 660
671
+ },
672
+ {
673
+ "entropy": 0.48768885880708696,
674
+ "epoch": 3.544973544973545,
675
+ "grad_norm": 1.2890625,
676
+ "learning_rate": 1.4975255019761896e-05,
677
+ "loss": 0.4901734352111816,
678
+ "mean_token_accuracy": 0.8567128427326679,
679
+ "num_tokens": 47768090.0,
680
+ "step": 670
681
+ },
682
+ {
683
+ "entropy": 0.4818777460604906,
684
+ "epoch": 3.597883597883598,
685
+ "grad_norm": 1.25,
686
+ "learning_rate": 1.4825940625887343e-05,
687
+ "loss": 0.4794710636138916,
688
+ "mean_token_accuracy": 0.8597230516374111,
689
+ "num_tokens": 48486016.0,
690
+ "step": 680
691
+ },
692
+ {
693
+ "entropy": 0.4782646283507347,
694
+ "epoch": 3.6507936507936507,
695
+ "grad_norm": 1.21875,
696
+ "learning_rate": 1.467521020206753e-05,
697
+ "loss": 0.4785307884216309,
698
+ "mean_token_accuracy": 0.8597849570214748,
699
+ "num_tokens": 49196144.0,
700
+ "step": 690
701
+ },
702
+ {
703
+ "entropy": 0.48901230953633784,
704
+ "epoch": 3.7037037037037037,
705
+ "grad_norm": 1.3046875,
706
+ "learning_rate": 1.4523107975699833e-05,
707
+ "loss": 0.4887538909912109,
708
+ "mean_token_accuracy": 0.8576067104935646,
709
+ "num_tokens": 49912021.0,
710
+ "step": 700
711
+ },
712
+ {
713
+ "entropy": 0.48083703890442847,
714
+ "epoch": 3.7566137566137567,
715
+ "grad_norm": 1.2265625,
716
+ "learning_rate": 1.4369678576696633e-05,
717
+ "loss": 0.47971267700195314,
718
+ "mean_token_accuracy": 0.8592194318771362,
719
+ "num_tokens": 50634262.0,
720
+ "step": 710
721
+ },
722
+ {
723
+ "entropy": 0.47896665148437023,
724
+ "epoch": 3.8095238095238093,
725
+ "grad_norm": 1.265625,
726
+ "learning_rate": 1.4214967024389978e-05,
727
+ "loss": 0.4767903327941895,
728
+ "mean_token_accuracy": 0.8601033940911293,
729
+ "num_tokens": 51343452.0,
730
+ "step": 720
731
+ },
732
+ {
733
+ "entropy": 0.4764186542481184,
734
+ "epoch": 3.8624338624338623,
735
+ "grad_norm": 1.2578125,
736
+ "learning_rate": 1.4059018714321986e-05,
737
+ "loss": 0.47649202346801756,
738
+ "mean_token_accuracy": 0.8595595695078373,
739
+ "num_tokens": 52046363.0,
740
+ "step": 730
741
+ },
742
+ {
743
+ "entropy": 0.476634269207716,
744
+ "epoch": 3.9153439153439153,
745
+ "grad_norm": 1.265625,
746
+ "learning_rate": 1.390187940492486e-05,
747
+ "loss": 0.4764859199523926,
748
+ "mean_token_accuracy": 0.8600306704640388,
749
+ "num_tokens": 52755135.0,
750
+ "step": 740
751
+ },
752
+ {
753
+ "entropy": 0.47171572409570217,
754
+ "epoch": 3.9682539682539684,
755
+ "grad_norm": 1.28125,
756
+ "learning_rate": 1.3743595204094442e-05,
757
+ "loss": 0.47093925476074217,
758
+ "mean_token_accuracy": 0.8609667584300041,
759
+ "num_tokens": 53473566.0,
760
+ "step": 750
761
+ },
762
+ {
763
+ "entropy": 0.464452663064003,
764
+ "epoch": 4.021164021164021,
765
+ "grad_norm": 1.1875,
766
+ "learning_rate": 1.3584212555661187e-05,
767
+ "loss": 0.4580575942993164,
768
+ "mean_token_accuracy": 0.8645686432719231,
769
+ "num_tokens": 54190492.0,
770
+ "step": 760
771
+ },
772
+ {
773
+ "entropy": 0.4455810472369194,
774
+ "epoch": 4.074074074074074,
775
+ "grad_norm": 1.25,
776
+ "learning_rate": 1.3423778225762609e-05,
777
+ "loss": 0.4465456962585449,
778
+ "mean_token_accuracy": 0.8673712603747845,
779
+ "num_tokens": 54897936.0,
780
+ "step": 770
781
+ },
782
+ {
783
+ "entropy": 0.4556675568223,
784
+ "epoch": 4.1269841269841265,
785
+ "grad_norm": 1.3671875,
786
+ "learning_rate": 1.3262339289121129e-05,
787
+ "loss": 0.4551807403564453,
788
+ "mean_token_accuracy": 0.8652156829833985,
789
+ "num_tokens": 55609388.0,
790
+ "step": 780
791
+ },
792
+ {
793
+ "entropy": 0.45422575771808626,
794
+ "epoch": 4.1798941798941796,
795
+ "grad_norm": 1.34375,
796
+ "learning_rate": 1.3099943115231386e-05,
797
+ "loss": 0.45353970527648924,
798
+ "mean_token_accuracy": 0.8658270433545112,
799
+ "num_tokens": 56329299.0,
800
+ "step": 790
801
+ },
802
+ {
803
+ "entropy": 0.45179531499743464,
804
+ "epoch": 4.232804232804233,
805
+ "grad_norm": 1.3046875,
806
+ "learning_rate": 1.2936637354461064e-05,
807
+ "loss": 0.44892282485961915,
808
+ "mean_token_accuracy": 0.8666822284460067,
809
+ "num_tokens": 57042704.0,
810
+ "step": 800
811
+ },
812
+ {
813
+ "entropy": 0.45655727535486224,
814
+ "epoch": 4.285714285714286,
815
+ "grad_norm": 1.3046875,
816
+ "learning_rate": 1.2772469924069285e-05,
817
+ "loss": 0.4565535545349121,
818
+ "mean_token_accuracy": 0.8650884844362736,
819
+ "num_tokens": 57755769.0,
820
+ "step": 810
821
+ },
822
+ {
823
+ "entropy": 0.4504398550838232,
824
+ "epoch": 4.338624338624339,
825
+ "grad_norm": 1.2109375,
826
+ "learning_rate": 1.2607488994146703e-05,
827
+ "loss": 0.44783792495727537,
828
+ "mean_token_accuracy": 0.8668409429490567,
829
+ "num_tokens": 58467967.0,
830
+ "step": 820
831
+ },
832
+ {
833
+ "entropy": 0.45743274949491025,
834
+ "epoch": 4.391534391534392,
835
+ "grad_norm": 1.2734375,
836
+ "learning_rate": 1.2441742973481432e-05,
837
+ "loss": 0.45568046569824217,
838
+ "mean_token_accuracy": 0.8652968890964985,
839
+ "num_tokens": 59188579.0,
840
+ "step": 830
841
+ },
842
+ {
843
+ "entropy": 0.44639028087258337,
844
+ "epoch": 4.444444444444445,
845
+ "grad_norm": 1.3515625,
846
+ "learning_rate": 1.2275280495354889e-05,
847
+ "loss": 0.44577698707580565,
848
+ "mean_token_accuracy": 0.8682797506451607,
849
+ "num_tokens": 59899589.0,
850
+ "step": 840
851
+ },
852
+ {
853
+ "entropy": 0.4485243815928698,
854
+ "epoch": 4.497354497354498,
855
+ "grad_norm": 1.265625,
856
+ "learning_rate": 1.21081504032718e-05,
857
+ "loss": 0.4481363773345947,
858
+ "mean_token_accuracy": 0.8678335674107075,
859
+ "num_tokens": 60620913.0,
860
+ "step": 850
861
+ },
862
+ {
863
+ "entropy": 0.4525593984872103,
864
+ "epoch": 4.550264550264551,
865
+ "grad_norm": 1.3125,
866
+ "learning_rate": 1.1940401736628524e-05,
867
+ "loss": 0.4498861312866211,
868
+ "mean_token_accuracy": 0.8662019960582257,
869
+ "num_tokens": 61330675.0,
870
+ "step": 860
871
+ },
872
+ {
873
+ "entropy": 0.4605423126369715,
874
+ "epoch": 4.603174603174603,
875
+ "grad_norm": 1.4140625,
876
+ "learning_rate": 1.1772083716323878e-05,
877
+ "loss": 0.4589731216430664,
878
+ "mean_token_accuracy": 0.8638311237096786,
879
+ "num_tokens": 62041645.0,
880
+ "step": 870
881
+ },
882
+ {
883
+ "entropy": 0.44991469383239746,
884
+ "epoch": 4.656084656084656,
885
+ "grad_norm": 1.25,
886
+ "learning_rate": 1.1603245730316724e-05,
887
+ "loss": 0.447885799407959,
888
+ "mean_token_accuracy": 0.8659505993127823,
889
+ "num_tokens": 62750446.0,
890
+ "step": 880
891
+ },
892
+ {
893
+ "entropy": 0.4469064295291901,
894
+ "epoch": 4.708994708994709,
895
+ "grad_norm": 1.28125,
896
+ "learning_rate": 1.1433937319134512e-05,
897
+ "loss": 0.44426918029785156,
898
+ "mean_token_accuracy": 0.868197600543499,
899
+ "num_tokens": 63460008.0,
900
+ "step": 890
901
+ },
902
+ {
903
+ "entropy": 0.4514023557305336,
904
+ "epoch": 4.761904761904762,
905
+ "grad_norm": 1.3046875,
906
+ "learning_rate": 1.1264208161337091e-05,
907
+ "loss": 0.45235776901245117,
908
+ "mean_token_accuracy": 0.8655660562217236,
909
+ "num_tokens": 64164803.0,
910
+ "step": 900
911
+ },
912
+ {
913
+ "entropy": 0.4533011969178915,
914
+ "epoch": 4.814814814814815,
915
+ "grad_norm": 1.328125,
916
+ "learning_rate": 1.1094108058940005e-05,
917
+ "loss": 0.4506092548370361,
918
+ "mean_token_accuracy": 0.865882170945406,
919
+ "num_tokens": 64878163.0,
920
+ "step": 910
921
+ },
922
+ {
923
+ "entropy": 0.44160800203680994,
924
+ "epoch": 4.867724867724868,
925
+ "grad_norm": 1.296875,
926
+ "learning_rate": 1.0923686922801537e-05,
927
+ "loss": 0.4400449752807617,
928
+ "mean_token_accuracy": 0.868823654204607,
929
+ "num_tokens": 65599878.0,
930
+ "step": 920
931
+ },
932
+ {
933
+ "entropy": 0.4407290514558554,
934
+ "epoch": 4.920634920634921,
935
+ "grad_norm": 1.2734375,
936
+ "learning_rate": 1.0752994757977862e-05,
937
+ "loss": 0.4377620697021484,
938
+ "mean_token_accuracy": 0.8691893480718136,
939
+ "num_tokens": 66312478.0,
940
+ "step": 930
941
+ },
942
+ {
943
+ "entropy": 0.45381851457059386,
944
+ "epoch": 4.973544973544973,
945
+ "grad_norm": 1.265625,
946
+ "learning_rate": 1.058208164905052e-05,
947
+ "loss": 0.45404396057128904,
948
+ "mean_token_accuracy": 0.8656110934913158,
949
+ "num_tokens": 67026652.0,
950
+ "step": 940
951
+ },
952
+ {
953
+ "entropy": 0.4451509706676006,
954
+ "epoch": 5.026455026455026,
955
+ "grad_norm": 1.265625,
956
+ "learning_rate": 1.041099774543059e-05,
957
+ "loss": 0.43704824447631835,
958
+ "mean_token_accuracy": 0.8695758722722531,
959
+ "num_tokens": 67743837.0,
960
+ "step": 950
961
+ },
962
+ {
963
+ "entropy": 0.4223184697329998,
964
+ "epoch": 5.079365079365079,
965
+ "grad_norm": 1.375,
966
+ "learning_rate": 1.023979324664382e-05,
967
+ "loss": 0.4231288909912109,
968
+ "mean_token_accuracy": 0.8729359813034534,
969
+ "num_tokens": 68459077.0,
970
+ "step": 960
971
+ },
972
+ {
973
+ "entropy": 0.4201000899076462,
974
+ "epoch": 5.132275132275132,
975
+ "grad_norm": 1.3515625,
976
+ "learning_rate": 1.0068518387601038e-05,
977
+ "loss": 0.4157362461090088,
978
+ "mean_token_accuracy": 0.8749549694359302,
979
+ "num_tokens": 69175886.0,
980
+ "step": 970
981
+ },
982
+ {
983
+ "entropy": 0.4203161742538214,
984
+ "epoch": 5.185185185185185,
985
+ "grad_norm": 1.328125,
986
+ "learning_rate": 9.897223423858226e-06,
987
+ "loss": 0.4191565036773682,
988
+ "mean_token_accuracy": 0.8741719849407673,
989
+ "num_tokens": 69888377.0,
990
+ "step": 980
991
+ },
992
+ {
993
+ "entropy": 0.4278976671397686,
994
+ "epoch": 5.238095238095238,
995
+ "grad_norm": 1.3125,
996
+ "learning_rate": 9.725958616870495e-06,
997
+ "loss": 0.42658100128173826,
998
+ "mean_token_accuracy": 0.8726167634129525,
999
+ "num_tokens": 70604584.0,
1000
+ "step": 990
1001
+ },
1002
+ {
1003
+ "entropy": 0.43622247986495494,
1004
+ "epoch": 5.291005291005291,
1005
+ "grad_norm": 1.3359375,
1006
+ "learning_rate": 9.554774219244344e-06,
1007
+ "loss": 0.4340261459350586,
1008
+ "mean_token_accuracy": 0.8702079840004444,
1009
+ "num_tokens": 71321169.0,
1010
+ "step": 1000
1011
+ },
1012
+ {
1013
+ "entropy": 0.42725917547941206,
1014
+ "epoch": 5.343915343915344,
1015
+ "grad_norm": 1.296875,
1016
+ "learning_rate": 9.38372045999252e-06,
1017
+ "loss": 0.4253455638885498,
1018
+ "mean_token_accuracy": 0.8729547396302223,
1019
+ "num_tokens": 72041175.0,
1020
+ "step": 1010
1021
+ },
1022
+ {
1023
+ "entropy": 0.4243607401847839,
1024
+ "epoch": 5.396825396825397,
1025
+ "grad_norm": 1.3828125,
1026
+ "learning_rate": 9.212847529795784e-06,
1027
+ "loss": 0.42325568199157715,
1028
+ "mean_token_accuracy": 0.8731870628893376,
1029
+ "num_tokens": 72749079.0,
1030
+ "step": 1020
1031
+ },
1032
+ {
1033
+ "entropy": 0.4303685359656811,
1034
+ "epoch": 5.449735449735449,
1035
+ "grad_norm": 1.2734375,
1036
+ "learning_rate": 9.042205566275952e-06,
1037
+ "loss": 0.4277714729309082,
1038
+ "mean_token_accuracy": 0.8721092127263546,
1039
+ "num_tokens": 73462893.0,
1040
+ "step": 1030
1041
+ },
1042
+ {
1043
+ "entropy": 0.43162608668208124,
1044
+ "epoch": 5.502645502645502,
1045
+ "grad_norm": 1.28125,
1046
+ "learning_rate": 8.871844639284479e-06,
1047
+ "loss": 0.4301889419555664,
1048
+ "mean_token_accuracy": 0.8712550736963749,
1049
+ "num_tokens": 74167133.0,
1050
+ "step": 1040
1051
+ },
1052
+ {
1053
+ "entropy": 0.43598169907927514,
1054
+ "epoch": 5.555555555555555,
1055
+ "grad_norm": 1.2734375,
1056
+ "learning_rate": 8.701814736210933e-06,
1057
+ "loss": 0.435499382019043,
1058
+ "mean_token_accuracy": 0.8700432546436787,
1059
+ "num_tokens": 74870399.0,
1060
+ "step": 1050
1061
+ },
1062
+ {
1063
+ "entropy": 0.4320717755705118,
1064
+ "epoch": 5.608465608465608,
1065
+ "grad_norm": 1.3203125,
1066
+ "learning_rate": 8.532165747315686e-06,
1067
+ "loss": 0.43087425231933596,
1068
+ "mean_token_accuracy": 0.8712112359702587,
1069
+ "num_tokens": 75590782.0,
1070
+ "step": 1060
1071
+ },
1072
+ {
1073
+ "entropy": 0.43664140328764917,
1074
+ "epoch": 5.661375661375661,
1075
+ "grad_norm": 1.2734375,
1076
+ "learning_rate": 8.362947451091072e-06,
1077
+ "loss": 0.4324526309967041,
1078
+ "mean_token_accuracy": 0.8710537351667881,
1079
+ "num_tokens": 76305788.0,
1080
+ "step": 1070
1081
+ },
1082
+ {
1083
+ "entropy": 0.43034715987741945,
1084
+ "epoch": 5.714285714285714,
1085
+ "grad_norm": 1.2890625,
1086
+ "learning_rate": 8.194209499655364e-06,
1087
+ "loss": 0.4322338581085205,
1088
+ "mean_token_accuracy": 0.8713471658527852,
1089
+ "num_tokens": 77023579.0,
1090
+ "step": 1080
1091
+ },
1092
+ {
1093
+ "entropy": 0.4325398612767458,
1094
+ "epoch": 5.767195767195767,
1095
+ "grad_norm": 1.40625,
1096
+ "learning_rate": 8.026001404183831e-06,
1097
+ "loss": 0.4275611400604248,
1098
+ "mean_token_accuracy": 0.8720052778720856,
1099
+ "num_tokens": 77720821.0,
1100
+ "step": 1090
1101
+ },
1102
+ {
1103
+ "entropy": 0.43604847937822344,
1104
+ "epoch": 5.8201058201058204,
1105
+ "grad_norm": 1.296875,
1106
+ "learning_rate": 7.85837252038112e-06,
1107
+ "loss": 0.43659210205078125,
1108
+ "mean_token_accuracy": 0.8698216550052166,
1109
+ "num_tokens": 78431671.0,
1110
+ "step": 1100
1111
+ },
1112
+ {
1113
+ "entropy": 0.4338367357850075,
1114
+ "epoch": 5.8730158730158735,
1115
+ "grad_norm": 1.3125,
1116
+ "learning_rate": 7.69137203399931e-06,
1117
+ "loss": 0.43023014068603516,
1118
+ "mean_token_accuracy": 0.8713279463350773,
1119
+ "num_tokens": 79141393.0,
1120
+ "step": 1110
1121
+ },
1122
+ {
1123
+ "entropy": 0.42576855383813383,
1124
+ "epoch": 5.925925925925926,
1125
+ "grad_norm": 1.234375,
1126
+ "learning_rate": 7.525048946405791e-06,
1127
+ "loss": 0.4262782096862793,
1128
+ "mean_token_accuracy": 0.8731129266321659,
1129
+ "num_tokens": 79858356.0,
1130
+ "step": 1120
1131
+ },
1132
+ {
1133
+ "entropy": 0.43451742343604566,
1134
+ "epoch": 5.978835978835979,
1135
+ "grad_norm": 1.3046875,
1136
+ "learning_rate": 7.359452060205245e-06,
1137
+ "loss": 0.43216233253479003,
1138
+ "mean_token_accuracy": 0.8712433956563472,
1139
+ "num_tokens": 80566532.0,
1140
+ "step": 1130
1141
+ },
1142
+ {
1143
+ "entropy": 0.42183659635484216,
1144
+ "epoch": 6.031746031746032,
1145
+ "grad_norm": 1.28125,
1146
+ "learning_rate": 7.194629964919989e-06,
1147
+ "loss": 0.4154026985168457,
1148
+ "mean_token_accuracy": 0.8752984322607518,
1149
+ "num_tokens": 81283415.0,
1150
+ "step": 1140
1151
+ },
1152
+ {
1153
+ "entropy": 0.4100382808595896,
1154
+ "epoch": 6.084656084656085,
1155
+ "grad_norm": 1.2890625,
1156
+ "learning_rate": 7.030631022732781e-06,
1157
+ "loss": 0.40886564254760743,
1158
+ "mean_token_accuracy": 0.8769385173916817,
1159
+ "num_tokens": 81995607.0,
1160
+ "step": 1150
1161
+ },
1162
+ {
1163
+ "entropy": 0.4237165842205286,
1164
+ "epoch": 6.137566137566138,
1165
+ "grad_norm": 1.359375,
1166
+ "learning_rate": 6.867503354296392e-06,
1167
+ "loss": 0.42139520645141604,
1168
+ "mean_token_accuracy": 0.8735997013747692,
1169
+ "num_tokens": 82712148.0,
1170
+ "step": 1160
1171
+ },
1172
+ {
1173
+ "entropy": 0.42654485553503035,
1174
+ "epoch": 6.190476190476191,
1175
+ "grad_norm": 1.34375,
1176
+ "learning_rate": 6.705294824614004e-06,
1177
+ "loss": 0.4240893363952637,
1178
+ "mean_token_accuracy": 0.8727411277592182,
1179
+ "num_tokens": 83425506.0,
1180
+ "step": 1170
1181
+ },
1182
+ {
1183
+ "entropy": 0.4187860660254955,
1184
+ "epoch": 6.243386243386244,
1185
+ "grad_norm": 1.328125,
1186
+ "learning_rate": 6.54405302899463e-06,
1187
+ "loss": 0.41619367599487306,
1188
+ "mean_token_accuracy": 0.8750131532549859,
1189
+ "num_tokens": 84137018.0,
1190
+ "step": 1180
1191
+ },
1192
+ {
1193
+ "entropy": 0.4216837890446186,
1194
+ "epoch": 6.296296296296296,
1195
+ "grad_norm": 1.3125,
1196
+ "learning_rate": 6.383825279087699e-06,
1197
+ "loss": 0.41980342864990233,
1198
+ "mean_token_accuracy": 0.8740883760154248,
1199
+ "num_tokens": 84849598.0,
1200
+ "step": 1190
1201
+ },
1202
+ {
1203
+ "entropy": 0.4209607470780611,
1204
+ "epoch": 6.349206349206349,
1205
+ "grad_norm": 1.296875,
1206
+ "learning_rate": 6.2246585890008245e-06,
1207
+ "loss": 0.4199337959289551,
1208
+ "mean_token_accuracy": 0.8746874034404755,
1209
+ "num_tokens": 85561962.0,
1210
+ "step": 1200
1211
+ },
1212
+ {
1213
+ "entropy": 0.41288652531802655,
1214
+ "epoch": 6.402116402116402,
1215
+ "grad_norm": 1.359375,
1216
+ "learning_rate": 6.06659966150491e-06,
1217
+ "loss": 0.4091951370239258,
1218
+ "mean_token_accuracy": 0.8767975211143494,
1219
+ "num_tokens": 86286338.0,
1220
+ "step": 1210
1221
+ },
1222
+ {
1223
+ "entropy": 0.41441997736692426,
1224
+ "epoch": 6.455026455026455,
1225
+ "grad_norm": 1.3125,
1226
+ "learning_rate": 5.909694874330601e-06,
1227
+ "loss": 0.4135852813720703,
1228
+ "mean_token_accuracy": 0.8761776715517045,
1229
+ "num_tokens": 87001911.0,
1230
+ "step": 1220
1231
+ },
1232
+ {
1233
+ "entropy": 0.41416131891310215,
1234
+ "epoch": 6.507936507936508,
1235
+ "grad_norm": 1.28125,
1236
+ "learning_rate": 5.753990266560105e-06,
1237
+ "loss": 0.4115488052368164,
1238
+ "mean_token_accuracy": 0.8769605971872807,
1239
+ "num_tokens": 87714571.0,
1240
+ "step": 1230
1241
+ },
1242
+ {
1243
+ "entropy": 0.4153778996318579,
1244
+ "epoch": 6.560846560846561,
1245
+ "grad_norm": 1.375,
1246
+ "learning_rate": 5.599531525118373e-06,
1247
+ "loss": 0.4159687519073486,
1248
+ "mean_token_accuracy": 0.8752469770610333,
1249
+ "num_tokens": 88417674.0,
1250
+ "step": 1240
1251
+ },
1252
+ {
1253
+ "entropy": 0.4190424919128418,
1254
+ "epoch": 6.613756613756614,
1255
+ "grad_norm": 1.3203125,
1256
+ "learning_rate": 5.446363971367638e-06,
1257
+ "loss": 0.4144894599914551,
1258
+ "mean_token_accuracy": 0.875180634111166,
1259
+ "num_tokens": 89123522.0,
1260
+ "step": 1250
1261
+ },
1262
+ {
1263
+ "entropy": 0.42192798517644403,
1264
+ "epoch": 6.666666666666667,
1265
+ "grad_norm": 1.34375,
1266
+ "learning_rate": 5.29453254780917e-06,
1267
+ "loss": 0.4181344032287598,
1268
+ "mean_token_accuracy": 0.8751389637589455,
1269
+ "num_tokens": 89838831.0,
1270
+ "step": 1260
1271
+ },
1272
+ {
1273
+ "entropy": 0.4147109277546406,
1274
+ "epoch": 6.71957671957672,
1275
+ "grad_norm": 1.2578125,
1276
+ "learning_rate": 5.1440818048962575e-06,
1277
+ "loss": 0.4149214744567871,
1278
+ "mean_token_accuracy": 0.875646834820509,
1279
+ "num_tokens": 90546066.0,
1280
+ "step": 1270
1281
+ },
1282
+ {
1283
+ "entropy": 0.41611550822854043,
1284
+ "epoch": 6.772486772486772,
1285
+ "grad_norm": 1.359375,
1286
+ "learning_rate": 4.995055887962167e-06,
1287
+ "loss": 0.41540231704711916,
1288
+ "mean_token_accuracy": 0.875544361025095,
1289
+ "num_tokens": 91257301.0,
1290
+ "step": 1280
1291
+ },
1292
+ {
1293
+ "entropy": 0.4205140460282564,
1294
+ "epoch": 6.825396825396825,
1295
+ "grad_norm": 1.375,
1296
+ "learning_rate": 4.8474985242669965e-06,
1297
+ "loss": 0.4182106018066406,
1298
+ "mean_token_accuracy": 0.8747602418065071,
1299
+ "num_tokens": 91961075.0,
1300
+ "step": 1290
1301
+ },
1302
+ {
1303
+ "entropy": 0.4291623383760452,
1304
+ "epoch": 6.878306878306878,
1305
+ "grad_norm": 1.3984375,
1306
+ "learning_rate": 4.701453010167217e-06,
1307
+ "loss": 0.42667760848999026,
1308
+ "mean_token_accuracy": 0.8727425791323185,
1309
+ "num_tokens": 92677769.0,
1310
+ "step": 1300
1311
+ },
1312
+ {
1313
+ "entropy": 0.41312588155269625,
1314
+ "epoch": 6.931216931216931,
1315
+ "grad_norm": 1.3359375,
1316
+ "learning_rate": 4.556962198411631e-06,
1317
+ "loss": 0.4099729537963867,
1318
+ "mean_token_accuracy": 0.8769033893942833,
1319
+ "num_tokens": 93390418.0,
1320
+ "step": 1310
1321
+ },
1322
+ {
1323
+ "entropy": 0.4084371831268072,
1324
+ "epoch": 6.984126984126984,
1325
+ "grad_norm": 1.3671875,
1326
+ "learning_rate": 4.414068485567473e-06,
1327
+ "loss": 0.4083920955657959,
1328
+ "mean_token_accuracy": 0.8769270032644272,
1329
+ "num_tokens": 94107638.0,
1330
+ "step": 1320
1331
+ },
1332
+ {
1333
+ "entropy": 0.414901152998209,
1334
+ "epoch": 7.037037037037037,
1335
+ "grad_norm": 1.2890625,
1336
+ "learning_rate": 4.272813799580417e-06,
1337
+ "loss": 0.4091632843017578,
1338
+ "mean_token_accuracy": 0.8766109876334667,
1339
+ "num_tokens": 94829458.0,
1340
+ "step": 1330
1341
+ },
1342
+ {
1343
+ "entropy": 0.4114344149827957,
1344
+ "epoch": 7.08994708994709,
1345
+ "grad_norm": 1.296875,
1346
+ "learning_rate": 4.1332395874720474e-06,
1347
+ "loss": 0.40756850242614745,
1348
+ "mean_token_accuracy": 0.8767555914819241,
1349
+ "num_tokens": 95540140.0,
1350
+ "step": 1340
1351
+ },
1352
+ {
1353
+ "entropy": 0.4090781681239605,
1354
+ "epoch": 7.142857142857143,
1355
+ "grad_norm": 1.265625,
1356
+ "learning_rate": 3.9953868031784415e-06,
1357
+ "loss": 0.4089008331298828,
1358
+ "mean_token_accuracy": 0.876671238988638,
1359
+ "num_tokens": 96254434.0,
1360
+ "step": 1350
1361
+ },
1362
+ {
1363
+ "entropy": 0.4147778756916523,
1364
+ "epoch": 7.195767195767195,
1365
+ "grad_norm": 1.328125,
1366
+ "learning_rate": 3.8592958955334636e-06,
1367
+ "loss": 0.4134193420410156,
1368
+ "mean_token_accuracy": 0.8760911405086518,
1369
+ "num_tokens": 96966014.0,
1370
+ "step": 1360
1371
+ },
1372
+ {
1373
+ "entropy": 0.40922857597470286,
1374
+ "epoch": 7.248677248677248,
1375
+ "grad_norm": 1.34375,
1376
+ "learning_rate": 3.725006796400231e-06,
1377
+ "loss": 0.40600833892822263,
1378
+ "mean_token_accuracy": 0.8775057464838028,
1379
+ "num_tokens": 97677188.0,
1380
+ "step": 1370
1381
+ },
1382
+ {
1383
+ "entropy": 0.41339924447238446,
1384
+ "epoch": 7.301587301587301,
1385
+ "grad_norm": 1.421875,
1386
+ "learning_rate": 3.5925589089542954e-06,
1387
+ "loss": 0.4097587585449219,
1388
+ "mean_token_accuracy": 0.875959849357605,
1389
+ "num_tokens": 98391677.0,
1390
+ "step": 1380
1391
+ },
1392
+ {
1393
+ "entropy": 0.41218244582414626,
1394
+ "epoch": 7.354497354497354,
1395
+ "grad_norm": 1.3046875,
1396
+ "learning_rate": 3.461991096121955e-06,
1397
+ "loss": 0.4099538326263428,
1398
+ "mean_token_accuracy": 0.8761332534253597,
1399
+ "num_tokens": 99105789.0,
1400
+ "step": 1390
1401
+ },
1402
+ {
1403
+ "entropy": 0.410692673176527,
1404
+ "epoch": 7.407407407407407,
1405
+ "grad_norm": 1.3203125,
1406
+ "learning_rate": 3.3333416691770402e-06,
1407
+ "loss": 0.4073766231536865,
1408
+ "mean_token_accuracy": 0.8764613293111324,
1409
+ "num_tokens": 99819774.0,
1410
+ "step": 1400
1411
+ },
1412
+ {
1413
+ "entropy": 0.4076389245688915,
1414
+ "epoch": 7.4603174603174605,
1415
+ "grad_norm": 1.3046875,
1416
+ "learning_rate": 3.206648376499637e-06,
1417
+ "loss": 0.4059453964233398,
1418
+ "mean_token_accuracy": 0.8779646269977093,
1419
+ "num_tokens": 100540345.0,
1420
+ "step": 1410
1421
+ },
1422
+ {
1423
+ "entropy": 0.416654784604907,
1424
+ "epoch": 7.5132275132275135,
1425
+ "grad_norm": 1.3046875,
1426
+ "learning_rate": 3.0819483924999183e-06,
1427
+ "loss": 0.41609792709350585,
1428
+ "mean_token_accuracy": 0.8747843399643898,
1429
+ "num_tokens": 101261007.0,
1430
+ "step": 1420
1431
+ },
1432
+ {
1433
+ "entropy": 0.41376331858336923,
1434
+ "epoch": 7.5661375661375665,
1435
+ "grad_norm": 1.2890625,
1436
+ "learning_rate": 2.959278306710409e-06,
1437
+ "loss": 0.40964488983154296,
1438
+ "mean_token_accuracy": 0.8766733475029469,
1439
+ "num_tokens": 101972692.0,
1440
+ "step": 1430
1441
+ },
1442
+ {
1443
+ "entropy": 0.4122611477971077,
1444
+ "epoch": 7.619047619047619,
1445
+ "grad_norm": 1.3125,
1446
+ "learning_rate": 2.838674113049884e-06,
1447
+ "loss": 0.4077149391174316,
1448
+ "mean_token_accuracy": 0.8768246524035931,
1449
+ "num_tokens": 102684836.0,
1450
+ "step": 1440
1451
+ },
1452
+ {
1453
+ "entropy": 0.42518827989697455,
1454
+ "epoch": 7.671957671957672,
1455
+ "grad_norm": 1.375,
1456
+ "learning_rate": 2.7201711992619873e-06,
1457
+ "loss": 0.42249202728271484,
1458
+ "mean_token_accuracy": 0.8743405237793922,
1459
+ "num_tokens": 103404477.0,
1460
+ "step": 1450
1461
+ },
1462
+ {
1463
+ "entropy": 0.41746283397078515,
1464
+ "epoch": 7.724867724867725,
1465
+ "grad_norm": 1.328125,
1466
+ "learning_rate": 2.6038043365317643e-06,
1467
+ "loss": 0.41622037887573243,
1468
+ "mean_token_accuracy": 0.8755017273128033,
1469
+ "num_tokens": 104111178.0,
1470
+ "step": 1460
1471
+ },
1472
+ {
1473
+ "entropy": 0.40634516142308713,
1474
+ "epoch": 7.777777777777778,
1475
+ "grad_norm": 1.3125,
1476
+ "learning_rate": 2.489607669283088e-06,
1477
+ "loss": 0.4032726287841797,
1478
+ "mean_token_accuracy": 0.8784879930317402,
1479
+ "num_tokens": 104820924.0,
1480
+ "step": 1470
1481
+ },
1482
+ {
1483
+ "entropy": 0.4181860238313675,
1484
+ "epoch": 7.830687830687831,
1485
+ "grad_norm": 1.3046875,
1486
+ "learning_rate": 2.3776147051599518e-06,
1487
+ "loss": 0.4165492057800293,
1488
+ "mean_token_accuracy": 0.8751473076641559,
1489
+ "num_tokens": 105525661.0,
1490
+ "step": 1480
1491
+ },
1492
+ {
1493
+ "entropy": 0.4215067997574806,
1494
+ "epoch": 7.883597883597884,
1495
+ "grad_norm": 1.296875,
1496
+ "learning_rate": 2.2678583051946733e-06,
1497
+ "loss": 0.4215100288391113,
1498
+ "mean_token_accuracy": 0.8738829605281353,
1499
+ "num_tokens": 106244487.0,
1500
+ "step": 1490
1501
+ },
1502
+ {
1503
+ "entropy": 0.422141370177269,
1504
+ "epoch": 7.936507936507937,
1505
+ "grad_norm": 1.4140625,
1506
+ "learning_rate": 2.1603706741657803e-06,
1507
+ "loss": 0.41953001022338865,
1508
+ "mean_token_accuracy": 0.8739961586892605,
1509
+ "num_tokens": 106950025.0,
1510
+ "step": 1500
1511
+ },
1512
+ {
1513
+ "entropy": 0.41841135509312155,
1514
+ "epoch": 7.98941798941799,
1515
+ "grad_norm": 1.3125,
1516
+ "learning_rate": 2.055183351148463e-06,
1517
+ "loss": 0.41535396575927735,
1518
+ "mean_token_accuracy": 0.8752631053328515,
1519
+ "num_tokens": 107659233.0,
1520
+ "step": 1510
1521
+ },
1522
+ {
1523
+ "entropy": 0.4093180373311043,
1524
+ "epoch": 8.042328042328043,
1525
+ "grad_norm": 1.3671875,
1526
+ "learning_rate": 1.9523272002603745e-06,
1527
+ "loss": 0.4061167240142822,
1528
+ "mean_token_accuracy": 0.8781525194644928,
1529
+ "num_tokens": 108374173.0,
1530
+ "step": 1520
1531
+ },
1532
+ {
1533
+ "entropy": 0.3999630343168974,
1534
+ "epoch": 8.095238095238095,
1535
+ "grad_norm": 1.328125,
1536
+ "learning_rate": 1.8518324016054601e-06,
1537
+ "loss": 0.39598617553710935,
1538
+ "mean_token_accuracy": 0.8807546623051167,
1539
+ "num_tokens": 109085583.0,
1540
+ "step": 1530
1541
+ },
1542
+ {
1543
+ "entropy": 0.4098013911396265,
1544
+ "epoch": 8.148148148148149,
1545
+ "grad_norm": 1.3203125,
1546
+ "learning_rate": 1.7537284424184852e-06,
1547
+ "loss": 0.40534296035766604,
1548
+ "mean_token_accuracy": 0.8775141790509224,
1549
+ "num_tokens": 109801232.0,
1550
+ "step": 1540
1551
+ },
1552
+ {
1553
+ "entropy": 0.41676906384527684,
1554
+ "epoch": 8.201058201058201,
1555
+ "grad_norm": 1.34375,
1556
+ "learning_rate": 1.658044108412905e-06,
1557
+ "loss": 0.41460742950439455,
1558
+ "mean_token_accuracy": 0.8753950335085392,
1559
+ "num_tokens": 110516150.0,
1560
+ "step": 1550
1561
+ },
1562
+ {
1563
+ "entropy": 0.4116453532129526,
1564
+ "epoch": 8.253968253968253,
1565
+ "grad_norm": 1.3515625,
1566
+ "learning_rate": 1.564807475334509e-06,
1567
+ "loss": 0.4131956100463867,
1568
+ "mean_token_accuracy": 0.8766112357378006,
1569
+ "num_tokens": 111232078.0,
1570
+ "step": 1560
1571
+ },
1572
+ {
1573
+ "entropy": 0.41225610859692097,
1574
+ "epoch": 8.306878306878307,
1575
+ "grad_norm": 1.3359375,
1576
+ "learning_rate": 1.4740459007234498e-06,
1577
+ "loss": 0.41111059188842775,
1578
+ "mean_token_accuracy": 0.8763026840984821,
1579
+ "num_tokens": 111950711.0,
1580
+ "step": 1570
1581
+ },
1582
+ {
1583
+ "entropy": 0.41501538529992105,
1584
+ "epoch": 8.359788359788359,
1585
+ "grad_norm": 1.3203125,
1586
+ "learning_rate": 1.3857860158869696e-06,
1587
+ "loss": 0.4131760597229004,
1588
+ "mean_token_accuracy": 0.8759730905294418,
1589
+ "num_tokens": 112658277.0,
1590
+ "step": 1580
1591
+ },
1592
+ {
1593
+ "entropy": 0.41387021318078043,
1594
+ "epoch": 8.412698412698413,
1595
+ "grad_norm": 1.2890625,
1596
+ "learning_rate": 1.3000537180852213e-06,
1597
+ "loss": 0.4116035461425781,
1598
+ "mean_token_accuracy": 0.8768503859639167,
1599
+ "num_tokens": 113378303.0,
1600
+ "step": 1590
1601
+ },
1602
+ {
1603
+ "entropy": 0.4176288653165102,
1604
+ "epoch": 8.465608465608465,
1605
+ "grad_norm": 1.328125,
1606
+ "learning_rate": 1.2168741629324944e-06,
1607
+ "loss": 0.4141057014465332,
1608
+ "mean_token_accuracy": 0.8759131275117398,
1609
+ "num_tokens": 114087597.0,
1610
+ "step": 1600
1611
+ },
1612
+ {
1613
+ "entropy": 0.40503756925463674,
1614
+ "epoch": 8.518518518518519,
1615
+ "grad_norm": 1.328125,
1616
+ "learning_rate": 1.1362717570160387e-06,
1617
+ "loss": 0.40210609436035155,
1618
+ "mean_token_accuracy": 0.8788188330829143,
1619
+ "num_tokens": 114803563.0,
1620
+ "step": 1610
1621
+ },
1622
+ {
1623
+ "entropy": 0.4061105038970709,
1624
+ "epoch": 8.571428571428571,
1625
+ "grad_norm": 1.2734375,
1626
+ "learning_rate": 1.0582701507346582e-06,
1627
+ "loss": 0.4021212100982666,
1628
+ "mean_token_accuracy": 0.878644997626543,
1629
+ "num_tokens": 115519221.0,
1630
+ "step": 1620
1631
+ },
1632
+ {
1633
+ "entropy": 0.4147233605384827,
1634
+ "epoch": 8.624338624338625,
1635
+ "grad_norm": 1.25,
1636
+ "learning_rate": 9.828922313592227e-07,
1637
+ "loss": 0.410779333114624,
1638
+ "mean_token_accuracy": 0.876928336918354,
1639
+ "num_tokens": 116238487.0,
1640
+ "step": 1630
1641
+ },
1642
+ {
1643
+ "entropy": 0.4097800888121128,
1644
+ "epoch": 8.677248677248677,
1645
+ "grad_norm": 1.34375,
1646
+ "learning_rate": 9.10160116317047e-07,
1647
+ "loss": 0.40713653564453123,
1648
+ "mean_token_accuracy": 0.8772961221635341,
1649
+ "num_tokens": 116956616.0,
1650
+ "step": 1640
1651
+ },
1652
+ {
1653
+ "entropy": 0.416746861487627,
1654
+ "epoch": 8.73015873015873,
1655
+ "grad_norm": 1.34375,
1656
+ "learning_rate": 8.400951467022067e-07,
1657
+ "loss": 0.41678814888000487,
1658
+ "mean_token_accuracy": 0.8752295158803463,
1659
+ "num_tokens": 117662184.0,
1660
+ "step": 1650
1661
+ },
1662
+ {
1663
+ "entropy": 0.4065455045551062,
1664
+ "epoch": 8.783068783068783,
1665
+ "grad_norm": 1.3203125,
1666
+ "learning_rate": 7.727178810136094e-07,
1667
+ "loss": 0.40470399856567385,
1668
+ "mean_token_accuracy": 0.878332145512104,
1669
+ "num_tokens": 118365440.0,
1670
+ "step": 1660
1671
+ },
1672
+ {
1673
+ "entropy": 0.4117112413048744,
1674
+ "epoch": 8.835978835978835,
1675
+ "grad_norm": 1.2734375,
1676
+ "learning_rate": 7.080480891227126e-07,
1677
+ "loss": 0.40632357597351076,
1678
+ "mean_token_accuracy": 0.8775922633707524,
1679
+ "num_tokens": 119078464.0,
1680
+ "step": 1670
1681
+ },
1682
+ {
1683
+ "entropy": 0.41403592862188815,
1684
+ "epoch": 8.88888888888889,
1685
+ "grad_norm": 1.265625,
1686
+ "learning_rate": 6.461047464726433e-07,
1687
+ "loss": 0.41112241744995115,
1688
+ "mean_token_accuracy": 0.8770215392112732,
1689
+ "num_tokens": 119792226.0,
1690
+ "step": 1680
1691
+ },
1692
+ {
1693
+ "entropy": 0.41094372384250166,
1694
+ "epoch": 8.941798941798941,
1695
+ "grad_norm": 1.3359375,
1696
+ "learning_rate": 5.869060285104089e-07,
1697
+ "loss": 0.4097023010253906,
1698
+ "mean_token_accuracy": 0.8770746536552906,
1699
+ "num_tokens": 120500468.0,
1700
+ "step": 1690
1701
+ },
1702
+ {
1703
+ "entropy": 0.40498077906668184,
1704
+ "epoch": 8.994708994708995,
1705
+ "grad_norm": 1.28125,
1706
+ "learning_rate": 5.304693053538434e-07,
1707
+ "loss": 0.4013853073120117,
1708
+ "mean_token_accuracy": 0.8787798292934894,
1709
+ "num_tokens": 121205368.0,
1710
+ "step": 1700
1711
+ }
1712
+ ],
1713
+ "logging_steps": 10,
1714
+ "max_steps": 1890,
1715
+ "num_input_tokens_seen": 0,
1716
+ "num_train_epochs": 10,
1717
+ "save_steps": 500,
1718
+ "stateful_callbacks": {
1719
+ "TrainerControl": {
1720
+ "args": {
1721
+ "should_epoch_stop": false,
1722
+ "should_evaluate": false,
1723
+ "should_log": false,
1724
+ "should_save": true,
1725
+ "should_training_stop": false
1726
+ },
1727
+ "attributes": {}
1728
+ }
1729
+ },
1730
+ "total_flos": 2.1905957236055736e+18,
1731
+ "train_batch_size": 8,
1732
+ "trial_name": null,
1733
+ "trial_params": null
1734
+ }
checkpoint-1701/training_args.bin ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:12cf8133cc6c754dc73b2cbaecbb0aa0b7e2a28209b2cf3544fd63386978a2bb
3
+ size 5841