Muqeeth commited on
Commit
49e6034
·
verified ·
1 Parent(s): 814fe3e

Training in progress, step 128, checkpoint

Browse files
checkpoint-128/config.json ADDED
@@ -0,0 +1,31 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "architectures": [
3
+ "GPTNeoXForCausalLM"
4
+ ],
5
+ "attention_bias": true,
6
+ "attention_dropout": 0.0,
7
+ "bos_token_id": 0,
8
+ "classifier_dropout": 0.1,
9
+ "eos_token_id": 0,
10
+ "hidden_act": "gelu",
11
+ "hidden_dropout": 0.0,
12
+ "hidden_size": 1024,
13
+ "initializer_range": 0.02,
14
+ "intermediate_size": 4096,
15
+ "layer_norm_eps": 1e-05,
16
+ "max_position_embeddings": 2048,
17
+ "model_type": "gpt_neox",
18
+ "num_attention_heads": 16,
19
+ "num_hidden_layers": 24,
20
+ "partial_rotary_factor": 0.25,
21
+ "rope_scaling": null,
22
+ "rope_theta": 10000,
23
+ "rotary_emb_base": 10000,
24
+ "rotary_pct": 0.25,
25
+ "tie_word_embeddings": false,
26
+ "torch_dtype": "float32",
27
+ "transformers_version": "4.50.3",
28
+ "use_cache": true,
29
+ "use_parallel_residual": true,
30
+ "vocab_size": 50304
31
+ }
checkpoint-128/generation_config.json ADDED
@@ -0,0 +1,5 @@
 
 
 
 
 
 
1
+ {
2
+ "_from_model_config": true,
3
+ "bos_token_id": 0,
4
+ "transformers_version": "4.50.3"
5
+ }
checkpoint-128/model.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:ca299c33662b2f44dbd37fb60e2b84b1d6e399303e5ab59488f0218a64172815
3
+ size 1621370224
checkpoint-128/optimizer.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:2e1c43ff2daabb7b124bad38541080b36c0d18f3c9541deaf328a9cd85248d72
3
+ size 3242916026
checkpoint-128/rng_state.pth ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:9692cf5feaaa8d2743bbff939530c4618410016328517797b44cbef85cb24ab8
3
+ size 14244
checkpoint-128/scheduler.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:b94e984393329de60dc7fa902fd68a2fd74f248d210b82c75caa9d0d693010e7
3
+ size 1064
checkpoint-128/special_tokens_map.json ADDED
@@ -0,0 +1,30 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "bos_token": {
3
+ "content": "<|endoftext|>",
4
+ "lstrip": false,
5
+ "normalized": false,
6
+ "rstrip": false,
7
+ "single_word": false
8
+ },
9
+ "eos_token": {
10
+ "content": "<|endoftext|>",
11
+ "lstrip": false,
12
+ "normalized": false,
13
+ "rstrip": false,
14
+ "single_word": false
15
+ },
16
+ "pad_token": {
17
+ "content": "<|padding|>",
18
+ "lstrip": false,
19
+ "normalized": false,
20
+ "rstrip": false,
21
+ "single_word": false
22
+ },
23
+ "unk_token": {
24
+ "content": "<|endoftext|>",
25
+ "lstrip": false,
26
+ "normalized": false,
27
+ "rstrip": false,
28
+ "single_word": false
29
+ }
30
+ }
checkpoint-128/tokenizer.json ADDED
The diff for this file is too large to render. See raw diff
 
checkpoint-128/tokenizer_config.json ADDED
@@ -0,0 +1,220 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "add_bos_token": false,
3
+ "add_eos_token": false,
4
+ "add_prefix_space": false,
5
+ "added_tokens_decoder": {
6
+ "0": {
7
+ "content": "<|endoftext|>",
8
+ "lstrip": false,
9
+ "normalized": false,
10
+ "rstrip": false,
11
+ "single_word": false,
12
+ "special": true
13
+ },
14
+ "1": {
15
+ "content": "<|padding|>",
16
+ "lstrip": false,
17
+ "normalized": false,
18
+ "rstrip": false,
19
+ "single_word": false,
20
+ "special": true
21
+ },
22
+ "50254": {
23
+ "content": " ",
24
+ "lstrip": false,
25
+ "normalized": true,
26
+ "rstrip": false,
27
+ "single_word": false,
28
+ "special": false
29
+ },
30
+ "50255": {
31
+ "content": " ",
32
+ "lstrip": false,
33
+ "normalized": true,
34
+ "rstrip": false,
35
+ "single_word": false,
36
+ "special": false
37
+ },
38
+ "50256": {
39
+ "content": " ",
40
+ "lstrip": false,
41
+ "normalized": true,
42
+ "rstrip": false,
43
+ "single_word": false,
44
+ "special": false
45
+ },
46
+ "50257": {
47
+ "content": " ",
48
+ "lstrip": false,
49
+ "normalized": true,
50
+ "rstrip": false,
51
+ "single_word": false,
52
+ "special": false
53
+ },
54
+ "50258": {
55
+ "content": " ",
56
+ "lstrip": false,
57
+ "normalized": true,
58
+ "rstrip": false,
59
+ "single_word": false,
60
+ "special": false
61
+ },
62
+ "50259": {
63
+ "content": " ",
64
+ "lstrip": false,
65
+ "normalized": true,
66
+ "rstrip": false,
67
+ "single_word": false,
68
+ "special": false
69
+ },
70
+ "50260": {
71
+ "content": " ",
72
+ "lstrip": false,
73
+ "normalized": true,
74
+ "rstrip": false,
75
+ "single_word": false,
76
+ "special": false
77
+ },
78
+ "50261": {
79
+ "content": " ",
80
+ "lstrip": false,
81
+ "normalized": true,
82
+ "rstrip": false,
83
+ "single_word": false,
84
+ "special": false
85
+ },
86
+ "50262": {
87
+ "content": " ",
88
+ "lstrip": false,
89
+ "normalized": true,
90
+ "rstrip": false,
91
+ "single_word": false,
92
+ "special": false
93
+ },
94
+ "50263": {
95
+ "content": " ",
96
+ "lstrip": false,
97
+ "normalized": true,
98
+ "rstrip": false,
99
+ "single_word": false,
100
+ "special": false
101
+ },
102
+ "50264": {
103
+ "content": " ",
104
+ "lstrip": false,
105
+ "normalized": true,
106
+ "rstrip": false,
107
+ "single_word": false,
108
+ "special": false
109
+ },
110
+ "50265": {
111
+ "content": " ",
112
+ "lstrip": false,
113
+ "normalized": true,
114
+ "rstrip": false,
115
+ "single_word": false,
116
+ "special": false
117
+ },
118
+ "50266": {
119
+ "content": " ",
120
+ "lstrip": false,
121
+ "normalized": true,
122
+ "rstrip": false,
123
+ "single_word": false,
124
+ "special": false
125
+ },
126
+ "50267": {
127
+ "content": " ",
128
+ "lstrip": false,
129
+ "normalized": true,
130
+ "rstrip": false,
131
+ "single_word": false,
132
+ "special": false
133
+ },
134
+ "50268": {
135
+ "content": " ",
136
+ "lstrip": false,
137
+ "normalized": true,
138
+ "rstrip": false,
139
+ "single_word": false,
140
+ "special": false
141
+ },
142
+ "50269": {
143
+ "content": " ",
144
+ "lstrip": false,
145
+ "normalized": true,
146
+ "rstrip": false,
147
+ "single_word": false,
148
+ "special": false
149
+ },
150
+ "50270": {
151
+ "content": " ",
152
+ "lstrip": false,
153
+ "normalized": true,
154
+ "rstrip": false,
155
+ "single_word": false,
156
+ "special": false
157
+ },
158
+ "50271": {
159
+ "content": " ",
160
+ "lstrip": false,
161
+ "normalized": true,
162
+ "rstrip": false,
163
+ "single_word": false,
164
+ "special": false
165
+ },
166
+ "50272": {
167
+ "content": " ",
168
+ "lstrip": false,
169
+ "normalized": true,
170
+ "rstrip": false,
171
+ "single_word": false,
172
+ "special": false
173
+ },
174
+ "50273": {
175
+ "content": " ",
176
+ "lstrip": false,
177
+ "normalized": true,
178
+ "rstrip": false,
179
+ "single_word": false,
180
+ "special": false
181
+ },
182
+ "50274": {
183
+ "content": " ",
184
+ "lstrip": false,
185
+ "normalized": true,
186
+ "rstrip": false,
187
+ "single_word": false,
188
+ "special": false
189
+ },
190
+ "50275": {
191
+ "content": " ",
192
+ "lstrip": false,
193
+ "normalized": true,
194
+ "rstrip": false,
195
+ "single_word": false,
196
+ "special": false
197
+ },
198
+ "50276": {
199
+ "content": " ",
200
+ "lstrip": false,
201
+ "normalized": true,
202
+ "rstrip": false,
203
+ "single_word": false,
204
+ "special": false
205
+ }
206
+ },
207
+ "bos_token": "<|endoftext|>",
208
+ "clean_up_tokenization_spaces": true,
209
+ "eos_token": "<|endoftext|>",
210
+ "extra_special_tokens": {},
211
+ "max_length": 580,
212
+ "model_max_length": 1000000000000000019884624838656,
213
+ "pad_token": "<|padding|>",
214
+ "padding_side": "left",
215
+ "stride": 0,
216
+ "tokenizer_class": "GPTNeoXTokenizer",
217
+ "truncation_side": "right",
218
+ "truncation_strategy": "longest_first",
219
+ "unk_token": "<|endoftext|>"
220
+ }
checkpoint-128/trainer_state.json ADDED
@@ -0,0 +1,2595 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "best_global_step": null,
3
+ "best_metric": null,
4
+ "best_model_checkpoint": null,
5
+ "episode": 65536,
6
+ "epoch": 0.2807402330363262,
7
+ "eval_steps": 500,
8
+ "global_step": 128,
9
+ "is_hyper_param_search": false,
10
+ "is_local_process_zero": true,
11
+ "is_world_process_zero": true,
12
+ "log_history": [
13
+ {
14
+ "episode": 512,
15
+ "epoch": 0.0021932830705962986,
16
+ "eps": 6,
17
+ "loss/policy_avg": 0.15234875679016113,
18
+ "lr": 3e-06,
19
+ "objective/entropy": -45.407432556152344,
20
+ "objective/kl": 0.3935524821281433,
21
+ "objective/non_score_reward": -0.001967762364074588,
22
+ "objective/rlhf_reward": 4.677086353302002,
23
+ "objective/scores": 4.679054260253906,
24
+ "policy/approxkl_avg": 0.11690396815538406,
25
+ "policy/clipfrac_avg": 0.34375,
26
+ "policy/entropy_avg": 0.8878304958343506,
27
+ "step": 0,
28
+ "val/clipfrac_avg": 0.0,
29
+ "val/num_eos_tokens": 2729,
30
+ "val/ratio": 1.000101089477539,
31
+ "val/ratio_var": 2.341874278499745e-06
32
+ },
33
+ {
34
+ "episode": 1024,
35
+ "epoch": 0.004386566141192597,
36
+ "eps": 6,
37
+ "loss/policy_avg": 0.10656304657459259,
38
+ "lr": 2.9882812500000002e-06,
39
+ "objective/entropy": -49.59076690673828,
40
+ "objective/kl": 0.6760815978050232,
41
+ "objective/non_score_reward": -0.0033804080449044704,
42
+ "objective/rlhf_reward": 4.551654815673828,
43
+ "objective/scores": 4.55503511428833,
44
+ "policy/approxkl_avg": 0.12034796178340912,
45
+ "policy/clipfrac_avg": 0.32421875,
46
+ "policy/entropy_avg": 0.8024331331253052,
47
+ "step": 1,
48
+ "val/clipfrac_avg": 0.0,
49
+ "val/num_eos_tokens": 3517,
50
+ "val/ratio": 0.9998391270637512,
51
+ "val/ratio_var": 2.0723566649394343e-06
52
+ },
53
+ {
54
+ "episode": 1536,
55
+ "epoch": 0.006579849211788897,
56
+ "eps": 6,
57
+ "loss/policy_avg": 0.17722828686237335,
58
+ "lr": 2.9765625e-06,
59
+ "objective/entropy": -51.09684371948242,
60
+ "objective/kl": 1.214200735092163,
61
+ "objective/non_score_reward": -0.006071004085242748,
62
+ "objective/rlhf_reward": 4.974597930908203,
63
+ "objective/scores": 4.980669021606445,
64
+ "policy/approxkl_avg": 0.12037836015224457,
65
+ "policy/clipfrac_avg": 0.318359375,
66
+ "policy/entropy_avg": 0.8043380379676819,
67
+ "step": 2,
68
+ "val/clipfrac_avg": 0.0,
69
+ "val/num_eos_tokens": 3252,
70
+ "val/ratio": 1.0001596212387085,
71
+ "val/ratio_var": 2.1233340703474823e-06
72
+ },
73
+ {
74
+ "episode": 2048,
75
+ "epoch": 0.008773132282385195,
76
+ "eps": 6,
77
+ "loss/policy_avg": 0.11922580748796463,
78
+ "lr": 2.96484375e-06,
79
+ "objective/entropy": -44.18528366088867,
80
+ "objective/kl": 1.4338339567184448,
81
+ "objective/non_score_reward": -0.007169169839471579,
82
+ "objective/rlhf_reward": 4.915553569793701,
83
+ "objective/scores": 4.922722339630127,
84
+ "policy/approxkl_avg": 0.16550302505493164,
85
+ "policy/clipfrac_avg": 0.375,
86
+ "policy/entropy_avg": 0.7921671271324158,
87
+ "step": 3,
88
+ "val/clipfrac_avg": 0.0,
89
+ "val/num_eos_tokens": 2435,
90
+ "val/ratio": 0.9996773600578308,
91
+ "val/ratio_var": 2.4303119516844163e-06
92
+ },
93
+ {
94
+ "episode": 2560,
95
+ "epoch": 0.010966415352981495,
96
+ "eps": 6,
97
+ "loss/policy_avg": 0.15014484524726868,
98
+ "lr": 2.953125e-06,
99
+ "objective/entropy": -42.067405700683594,
100
+ "objective/kl": 2.0778450965881348,
101
+ "objective/non_score_reward": -0.01038922369480133,
102
+ "objective/rlhf_reward": 5.2342209815979,
103
+ "objective/scores": 5.244609832763672,
104
+ "policy/approxkl_avg": 0.1276472508907318,
105
+ "policy/clipfrac_avg": 0.31640625,
106
+ "policy/entropy_avg": 0.7496817111968994,
107
+ "step": 4,
108
+ "val/clipfrac_avg": 0.0,
109
+ "val/num_eos_tokens": 1877,
110
+ "val/ratio": 1.0003182888031006,
111
+ "val/ratio_var": 2.2540482405020157e-06
112
+ },
113
+ {
114
+ "episode": 3072,
115
+ "epoch": 0.013159698423577794,
116
+ "eps": 6,
117
+ "loss/policy_avg": 0.1916506588459015,
118
+ "lr": 2.94140625e-06,
119
+ "objective/entropy": -39.82306671142578,
120
+ "objective/kl": 3.079103708267212,
121
+ "objective/non_score_reward": -0.015395518392324448,
122
+ "objective/rlhf_reward": 5.611381530761719,
123
+ "objective/scores": 5.626776695251465,
124
+ "policy/approxkl_avg": 0.13991527259349823,
125
+ "policy/clipfrac_avg": 0.33203125,
126
+ "policy/entropy_avg": 0.7127301692962646,
127
+ "step": 5,
128
+ "val/clipfrac_avg": 0.0,
129
+ "val/num_eos_tokens": 1565,
130
+ "val/ratio": 0.9999192357063293,
131
+ "val/ratio_var": 1.5017312762211077e-06
132
+ },
133
+ {
134
+ "episode": 3584,
135
+ "epoch": 0.015352981494174092,
136
+ "eps": 6,
137
+ "loss/policy_avg": 0.1968970149755478,
138
+ "lr": 2.9296875e-06,
139
+ "objective/entropy": -32.000404357910156,
140
+ "objective/kl": 4.284768581390381,
141
+ "objective/non_score_reward": -0.021423842757940292,
142
+ "objective/rlhf_reward": 5.7136030197143555,
143
+ "objective/scores": 5.7350263595581055,
144
+ "policy/approxkl_avg": 0.1647772341966629,
145
+ "policy/clipfrac_avg": 0.3515625,
146
+ "policy/entropy_avg": 0.6934345364570618,
147
+ "step": 6,
148
+ "val/clipfrac_avg": 0.0,
149
+ "val/num_eos_tokens": 1364,
150
+ "val/ratio": 0.9998640418052673,
151
+ "val/ratio_var": 2.735638872763957e-06
152
+ },
153
+ {
154
+ "episode": 4096,
155
+ "epoch": 0.01754626456477039,
156
+ "eps": 6,
157
+ "loss/policy_avg": 0.20194123685359955,
158
+ "lr": 2.91796875e-06,
159
+ "objective/entropy": -35.15373992919922,
160
+ "objective/kl": 4.795171737670898,
161
+ "objective/non_score_reward": -0.023975860327482224,
162
+ "objective/rlhf_reward": 6.039769649505615,
163
+ "objective/scores": 6.063745498657227,
164
+ "policy/approxkl_avg": 0.19040407240390778,
165
+ "policy/clipfrac_avg": 0.365234375,
166
+ "policy/entropy_avg": 0.6659716367721558,
167
+ "step": 7,
168
+ "val/clipfrac_avg": 0.0,
169
+ "val/num_eos_tokens": 1595,
170
+ "val/ratio": 0.9995955228805542,
171
+ "val/ratio_var": 2.328864411538234e-06
172
+ },
173
+ {
174
+ "episode": 4608,
175
+ "epoch": 0.01973954763536669,
176
+ "eps": 6,
177
+ "loss/policy_avg": 0.2287895679473877,
178
+ "lr": 2.90625e-06,
179
+ "objective/entropy": -33.49115753173828,
180
+ "objective/kl": 6.028232097625732,
181
+ "objective/non_score_reward": -0.030141159892082214,
182
+ "objective/rlhf_reward": 6.286189556121826,
183
+ "objective/scores": 6.316330909729004,
184
+ "policy/approxkl_avg": 0.16508515179157257,
185
+ "policy/clipfrac_avg": 0.35546875,
186
+ "policy/entropy_avg": 0.6881623268127441,
187
+ "step": 8,
188
+ "val/clipfrac_avg": 0.0,
189
+ "val/num_eos_tokens": 1875,
190
+ "val/ratio": 0.9999189376831055,
191
+ "val/ratio_var": 2.560210077717784e-06
192
+ },
193
+ {
194
+ "episode": 5120,
195
+ "epoch": 0.02193283070596299,
196
+ "eps": 6,
197
+ "loss/policy_avg": 0.23305176198482513,
198
+ "lr": 2.89453125e-06,
199
+ "objective/entropy": -31.895954132080078,
200
+ "objective/kl": 7.27902889251709,
201
+ "objective/non_score_reward": -0.03639514371752739,
202
+ "objective/rlhf_reward": 6.584256172180176,
203
+ "objective/scores": 6.6206512451171875,
204
+ "policy/approxkl_avg": 0.15825235843658447,
205
+ "policy/clipfrac_avg": 0.333984375,
206
+ "policy/entropy_avg": 0.6986380219459534,
207
+ "step": 9,
208
+ "val/clipfrac_avg": 0.0,
209
+ "val/num_eos_tokens": 2233,
210
+ "val/ratio": 1.000351905822754,
211
+ "val/ratio_var": 2.54610608863004e-06
212
+ },
213
+ {
214
+ "episode": 5632,
215
+ "epoch": 0.024126113776559287,
216
+ "eps": 6,
217
+ "loss/policy_avg": 0.2743307948112488,
218
+ "lr": 2.8828125e-06,
219
+ "objective/entropy": -30.13780975341797,
220
+ "objective/kl": 8.618773460388184,
221
+ "objective/non_score_reward": -0.04309386759996414,
222
+ "objective/rlhf_reward": 6.671762466430664,
223
+ "objective/scores": 6.714856147766113,
224
+ "policy/approxkl_avg": 0.2310631275177002,
225
+ "policy/clipfrac_avg": 0.376953125,
226
+ "policy/entropy_avg": 0.6997334957122803,
227
+ "step": 10,
228
+ "val/clipfrac_avg": 0.0,
229
+ "val/num_eos_tokens": 2288,
230
+ "val/ratio": 0.9999397397041321,
231
+ "val/ratio_var": 2.7119385777041316e-06
232
+ },
233
+ {
234
+ "episode": 6144,
235
+ "epoch": 0.026319396847155587,
236
+ "eps": 6,
237
+ "loss/policy_avg": 0.47530221939086914,
238
+ "lr": 2.87109375e-06,
239
+ "objective/entropy": -25.868932723999023,
240
+ "objective/kl": 10.715995788574219,
241
+ "objective/non_score_reward": -0.05357997864484787,
242
+ "objective/rlhf_reward": 6.660526275634766,
243
+ "objective/scores": 6.714106559753418,
244
+ "policy/approxkl_avg": 0.23832854628562927,
245
+ "policy/clipfrac_avg": 0.33984375,
246
+ "policy/entropy_avg": 0.6960214376449585,
247
+ "step": 11,
248
+ "val/clipfrac_avg": 0.0,
249
+ "val/num_eos_tokens": 2465,
250
+ "val/ratio": 1.0001269578933716,
251
+ "val/ratio_var": 4.06193794333376e-06
252
+ },
253
+ {
254
+ "episode": 6656,
255
+ "epoch": 0.028512679917751884,
256
+ "eps": 6,
257
+ "loss/policy_avg": 0.2478591650724411,
258
+ "lr": 2.859375e-06,
259
+ "objective/entropy": -28.481605529785156,
260
+ "objective/kl": 11.356575012207031,
261
+ "objective/non_score_reward": -0.05678287148475647,
262
+ "objective/rlhf_reward": 7.32498025894165,
263
+ "objective/scores": 7.381762981414795,
264
+ "policy/approxkl_avg": 0.23808607459068298,
265
+ "policy/clipfrac_avg": 0.345703125,
266
+ "policy/entropy_avg": 0.6703023910522461,
267
+ "step": 12,
268
+ "val/clipfrac_avg": 0.0,
269
+ "val/num_eos_tokens": 2564,
270
+ "val/ratio": 0.999792754650116,
271
+ "val/ratio_var": 2.613600599943311e-06
272
+ },
273
+ {
274
+ "episode": 7168,
275
+ "epoch": 0.030705962988348184,
276
+ "eps": 6,
277
+ "loss/policy_avg": 0.18863295018672943,
278
+ "lr": 2.84765625e-06,
279
+ "objective/entropy": -31.87090301513672,
280
+ "objective/kl": 12.704984664916992,
281
+ "objective/non_score_reward": -0.06352491676807404,
282
+ "objective/rlhf_reward": 7.6809868812561035,
283
+ "objective/scores": 7.744511604309082,
284
+ "policy/approxkl_avg": 0.20181894302368164,
285
+ "policy/clipfrac_avg": 0.3828125,
286
+ "policy/entropy_avg": 0.6681352853775024,
287
+ "step": 13,
288
+ "val/clipfrac_avg": 0.0,
289
+ "val/num_eos_tokens": 2607,
290
+ "val/ratio": 1.0001722574234009,
291
+ "val/ratio_var": 3.1368097097583814e-06
292
+ },
293
+ {
294
+ "episode": 7680,
295
+ "epoch": 0.03289924605894448,
296
+ "eps": 6,
297
+ "loss/policy_avg": 0.23149724304676056,
298
+ "lr": 2.8359375e-06,
299
+ "objective/entropy": -31.151302337646484,
300
+ "objective/kl": 14.230264663696289,
301
+ "objective/non_score_reward": -0.07115132361650467,
302
+ "objective/rlhf_reward": 7.844120025634766,
303
+ "objective/scores": 7.915271282196045,
304
+ "policy/approxkl_avg": 0.23936033248901367,
305
+ "policy/clipfrac_avg": 0.40625,
306
+ "policy/entropy_avg": 0.6505059003829956,
307
+ "step": 14,
308
+ "val/clipfrac_avg": 0.0,
309
+ "val/num_eos_tokens": 2523,
310
+ "val/ratio": 0.999779224395752,
311
+ "val/ratio_var": 3.855011527775787e-06
312
+ },
313
+ {
314
+ "episode": 8192,
315
+ "epoch": 0.03509252912954078,
316
+ "eps": 6,
317
+ "loss/policy_avg": 0.25306057929992676,
318
+ "lr": 2.82421875e-06,
319
+ "objective/entropy": -33.95362091064453,
320
+ "objective/kl": 15.211322784423828,
321
+ "objective/non_score_reward": -0.07605661451816559,
322
+ "objective/rlhf_reward": 8.07334041595459,
323
+ "objective/scores": 8.149396896362305,
324
+ "policy/approxkl_avg": 0.2219444364309311,
325
+ "policy/clipfrac_avg": 0.376953125,
326
+ "policy/entropy_avg": 0.6534073352813721,
327
+ "step": 15,
328
+ "val/clipfrac_avg": 0.0,
329
+ "val/num_eos_tokens": 2392,
330
+ "val/ratio": 1.000091791152954,
331
+ "val/ratio_var": 3.7099030123499688e-06
332
+ },
333
+ {
334
+ "episode": 8704,
335
+ "epoch": 0.03728581220013708,
336
+ "eps": 6,
337
+ "loss/policy_avg": 0.3671952486038208,
338
+ "lr": 2.8125e-06,
339
+ "objective/entropy": -36.56987762451172,
340
+ "objective/kl": 15.896276473999023,
341
+ "objective/non_score_reward": -0.07948137074708939,
342
+ "objective/rlhf_reward": 8.115487098693848,
343
+ "objective/scores": 8.194968223571777,
344
+ "policy/approxkl_avg": 0.2001374065876007,
345
+ "policy/clipfrac_avg": 0.376953125,
346
+ "policy/entropy_avg": 0.5611248016357422,
347
+ "step": 16,
348
+ "val/clipfrac_avg": 0.0,
349
+ "val/num_eos_tokens": 2099,
350
+ "val/ratio": 1.0001428127288818,
351
+ "val/ratio_var": 2.9763129987259163e-06
352
+ },
353
+ {
354
+ "episode": 9216,
355
+ "epoch": 0.03947909527073338,
356
+ "eps": 6,
357
+ "loss/policy_avg": 0.2377510964870453,
358
+ "lr": 2.80078125e-06,
359
+ "objective/entropy": -38.7183837890625,
360
+ "objective/kl": 16.948450088500977,
361
+ "objective/non_score_reward": -0.08474224805831909,
362
+ "objective/rlhf_reward": 8.287287712097168,
363
+ "objective/scores": 8.372029304504395,
364
+ "policy/approxkl_avg": 0.19774162769317627,
365
+ "policy/clipfrac_avg": 0.33984375,
366
+ "policy/entropy_avg": 0.5588769912719727,
367
+ "step": 17,
368
+ "val/clipfrac_avg": 0.0,
369
+ "val/num_eos_tokens": 2173,
370
+ "val/ratio": 0.9999226331710815,
371
+ "val/ratio_var": 2.539955858082976e-06
372
+ },
373
+ {
374
+ "episode": 9728,
375
+ "epoch": 0.041672378341329676,
376
+ "eps": 6,
377
+ "loss/policy_avg": 0.30208033323287964,
378
+ "lr": 2.7890625e-06,
379
+ "objective/entropy": -37.54444122314453,
380
+ "objective/kl": 17.52410316467285,
381
+ "objective/non_score_reward": -0.08762051165103912,
382
+ "objective/rlhf_reward": 8.367655754089355,
383
+ "objective/scores": 8.455276489257812,
384
+ "policy/approxkl_avg": 0.20374038815498352,
385
+ "policy/clipfrac_avg": 0.33984375,
386
+ "policy/entropy_avg": 0.5387827157974243,
387
+ "step": 18,
388
+ "val/clipfrac_avg": 0.0,
389
+ "val/num_eos_tokens": 2168,
390
+ "val/ratio": 0.99991774559021,
391
+ "val/ratio_var": 3.0660403353977017e-06
392
+ },
393
+ {
394
+ "episode": 10240,
395
+ "epoch": 0.04386566141192598,
396
+ "eps": 6,
397
+ "loss/policy_avg": 0.22399874031543732,
398
+ "lr": 2.77734375e-06,
399
+ "objective/entropy": -39.96161651611328,
400
+ "objective/kl": 17.749210357666016,
401
+ "objective/non_score_reward": -0.08874605596065521,
402
+ "objective/rlhf_reward": 8.257219314575195,
403
+ "objective/scores": 8.345966339111328,
404
+ "policy/approxkl_avg": 0.2131994068622589,
405
+ "policy/clipfrac_avg": 0.359375,
406
+ "policy/entropy_avg": 0.5233398079872131,
407
+ "step": 19,
408
+ "val/clipfrac_avg": 0.0,
409
+ "val/num_eos_tokens": 2201,
410
+ "val/ratio": 1.0001959800720215,
411
+ "val/ratio_var": 3.099843979725847e-06
412
+ },
413
+ {
414
+ "episode": 10752,
415
+ "epoch": 0.04605894448252228,
416
+ "eps": 6,
417
+ "loss/policy_avg": 0.16000297665596008,
418
+ "lr": 2.765625e-06,
419
+ "objective/entropy": -40.21980285644531,
420
+ "objective/kl": 18.693737030029297,
421
+ "objective/non_score_reward": -0.09346868097782135,
422
+ "objective/rlhf_reward": 8.49026107788086,
423
+ "objective/scores": 8.58372974395752,
424
+ "policy/approxkl_avg": 0.2675846219062805,
425
+ "policy/clipfrac_avg": 0.38671875,
426
+ "policy/entropy_avg": 0.524081826210022,
427
+ "step": 20,
428
+ "val/clipfrac_avg": 0.0,
429
+ "val/num_eos_tokens": 2342,
430
+ "val/ratio": 0.9995419979095459,
431
+ "val/ratio_var": 3.2623988772684243e-06
432
+ },
433
+ {
434
+ "episode": 11264,
435
+ "epoch": 0.048252227553118573,
436
+ "eps": 6,
437
+ "loss/policy_avg": 0.23491168022155762,
438
+ "lr": 2.75390625e-06,
439
+ "objective/entropy": -38.962947845458984,
440
+ "objective/kl": 19.98727035522461,
441
+ "objective/non_score_reward": -0.099936343729496,
442
+ "objective/rlhf_reward": 8.68085765838623,
443
+ "objective/scores": 8.780793190002441,
444
+ "policy/approxkl_avg": 0.19925493001937866,
445
+ "policy/clipfrac_avg": 0.3359375,
446
+ "policy/entropy_avg": 0.5088595151901245,
447
+ "step": 21,
448
+ "val/clipfrac_avg": 0.0,
449
+ "val/num_eos_tokens": 2385,
450
+ "val/ratio": 0.9999128580093384,
451
+ "val/ratio_var": 3.195402769051725e-06
452
+ },
453
+ {
454
+ "episode": 11776,
455
+ "epoch": 0.05044551062371487,
456
+ "eps": 6,
457
+ "loss/policy_avg": 0.20272764563560486,
458
+ "lr": 2.7421875e-06,
459
+ "objective/entropy": -39.06739807128906,
460
+ "objective/kl": 20.73778533935547,
461
+ "objective/non_score_reward": -0.10368892550468445,
462
+ "objective/rlhf_reward": 8.66939640045166,
463
+ "objective/scores": 8.77308464050293,
464
+ "policy/approxkl_avg": 0.24808868765830994,
465
+ "policy/clipfrac_avg": 0.3671875,
466
+ "policy/entropy_avg": 0.5206277370452881,
467
+ "step": 22,
468
+ "val/clipfrac_avg": 0.0,
469
+ "val/num_eos_tokens": 2328,
470
+ "val/ratio": 0.9997999668121338,
471
+ "val/ratio_var": 2.6692378014558926e-06
472
+ },
473
+ {
474
+ "episode": 12288,
475
+ "epoch": 0.052638793694311174,
476
+ "eps": 6,
477
+ "loss/policy_avg": 0.1657373309135437,
478
+ "lr": 2.73046875e-06,
479
+ "objective/entropy": -41.76697540283203,
480
+ "objective/kl": 21.265926361083984,
481
+ "objective/non_score_reward": -0.10632962733507156,
482
+ "objective/rlhf_reward": 8.899429321289062,
483
+ "objective/scores": 9.005760192871094,
484
+ "policy/approxkl_avg": 0.21000100672245026,
485
+ "policy/clipfrac_avg": 0.330078125,
486
+ "policy/entropy_avg": 0.4891219437122345,
487
+ "step": 23,
488
+ "val/clipfrac_avg": 0.0,
489
+ "val/num_eos_tokens": 2460,
490
+ "val/ratio": 1.000035285949707,
491
+ "val/ratio_var": 3.269958597229561e-06
492
+ },
493
+ {
494
+ "episode": 12800,
495
+ "epoch": 0.05483207676490747,
496
+ "eps": 6,
497
+ "loss/policy_avg": 0.17451098561286926,
498
+ "lr": 2.71875e-06,
499
+ "objective/entropy": -40.41400146484375,
500
+ "objective/kl": 22.69107437133789,
501
+ "objective/non_score_reward": -0.11345535516738892,
502
+ "objective/rlhf_reward": 8.97903060913086,
503
+ "objective/scores": 9.092485427856445,
504
+ "policy/approxkl_avg": 0.2612704336643219,
505
+ "policy/clipfrac_avg": 0.37109375,
506
+ "policy/entropy_avg": 0.4792313873767853,
507
+ "step": 24,
508
+ "val/clipfrac_avg": 0.0,
509
+ "val/num_eos_tokens": 2361,
510
+ "val/ratio": 0.9997962713241577,
511
+ "val/ratio_var": 4.033691766380798e-06
512
+ },
513
+ {
514
+ "episode": 13312,
515
+ "epoch": 0.05702535983550377,
516
+ "eps": 6,
517
+ "loss/policy_avg": 0.3060297966003418,
518
+ "lr": 2.70703125e-06,
519
+ "objective/entropy": -39.72850036621094,
520
+ "objective/kl": 24.020402908325195,
521
+ "objective/non_score_reward": -0.12010201811790466,
522
+ "objective/rlhf_reward": 8.757214546203613,
523
+ "objective/scores": 8.877317428588867,
524
+ "policy/approxkl_avg": 0.20275026559829712,
525
+ "policy/clipfrac_avg": 0.333984375,
526
+ "policy/entropy_avg": 0.47031739354133606,
527
+ "step": 25,
528
+ "val/clipfrac_avg": 0.0,
529
+ "val/num_eos_tokens": 2365,
530
+ "val/ratio": 1.0004143714904785,
531
+ "val/ratio_var": 4.571053977997508e-06
532
+ },
533
+ {
534
+ "episode": 13824,
535
+ "epoch": 0.05921864290610007,
536
+ "eps": 6,
537
+ "loss/policy_avg": 0.2459714114665985,
538
+ "lr": 2.6953125e-06,
539
+ "objective/entropy": -42.67210388183594,
540
+ "objective/kl": 24.816062927246094,
541
+ "objective/non_score_reward": -0.12408032268285751,
542
+ "objective/rlhf_reward": 8.767488479614258,
543
+ "objective/scores": 8.891569137573242,
544
+ "policy/approxkl_avg": 0.16891685128211975,
545
+ "policy/clipfrac_avg": 0.314453125,
546
+ "policy/entropy_avg": 0.4268378019332886,
547
+ "step": 26,
548
+ "val/clipfrac_avg": 0.0,
549
+ "val/num_eos_tokens": 2338,
550
+ "val/ratio": 1.0001232624053955,
551
+ "val/ratio_var": 2.055706318060402e-06
552
+ },
553
+ {
554
+ "episode": 14336,
555
+ "epoch": 0.06141192597669637,
556
+ "eps": 6,
557
+ "loss/policy_avg": 0.2477143555879593,
558
+ "lr": 2.68359375e-06,
559
+ "objective/entropy": -40.98384094238281,
560
+ "objective/kl": 26.158355712890625,
561
+ "objective/non_score_reward": -0.1307917833328247,
562
+ "objective/rlhf_reward": 8.869270324707031,
563
+ "objective/scores": 9.000061988830566,
564
+ "policy/approxkl_avg": 0.21648374199867249,
565
+ "policy/clipfrac_avg": 0.35546875,
566
+ "policy/entropy_avg": 0.43942803144454956,
567
+ "step": 27,
568
+ "val/clipfrac_avg": 0.0,
569
+ "val/num_eos_tokens": 2325,
570
+ "val/ratio": 1.00002920627594,
571
+ "val/ratio_var": 3.6563492358254734e-06
572
+ },
573
+ {
574
+ "episode": 14848,
575
+ "epoch": 0.06360520904729267,
576
+ "eps": 6,
577
+ "loss/policy_avg": 0.18666958808898926,
578
+ "lr": 2.671875e-06,
579
+ "objective/entropy": -39.86817169189453,
580
+ "objective/kl": 26.283226013183594,
581
+ "objective/non_score_reward": -0.13141612708568573,
582
+ "objective/rlhf_reward": 9.001875877380371,
583
+ "objective/scores": 9.133292198181152,
584
+ "policy/approxkl_avg": 0.22604086995124817,
585
+ "policy/clipfrac_avg": 0.37109375,
586
+ "policy/entropy_avg": 0.4275825023651123,
587
+ "step": 28,
588
+ "val/clipfrac_avg": 0.0,
589
+ "val/num_eos_tokens": 2388,
590
+ "val/ratio": 0.999720573425293,
591
+ "val/ratio_var": 2.5127403660007985e-06
592
+ },
593
+ {
594
+ "episode": 15360,
595
+ "epoch": 0.06579849211788896,
596
+ "eps": 6,
597
+ "loss/policy_avg": 0.21501849591732025,
598
+ "lr": 2.66015625e-06,
599
+ "objective/entropy": -34.56945037841797,
600
+ "objective/kl": 28.54375457763672,
601
+ "objective/non_score_reward": -0.14271876215934753,
602
+ "objective/rlhf_reward": 8.915486335754395,
603
+ "objective/scores": 9.058204650878906,
604
+ "policy/approxkl_avg": 0.23621705174446106,
605
+ "policy/clipfrac_avg": 0.361328125,
606
+ "policy/entropy_avg": 0.42039746046066284,
607
+ "step": 29,
608
+ "val/clipfrac_avg": 0.0,
609
+ "val/num_eos_tokens": 2389,
610
+ "val/ratio": 0.9995044469833374,
611
+ "val/ratio_var": 4.271460511517944e-06
612
+ },
613
+ {
614
+ "episode": 15872,
615
+ "epoch": 0.06799177518848526,
616
+ "eps": 6,
617
+ "loss/policy_avg": 0.343304842710495,
618
+ "lr": 2.6484375e-06,
619
+ "objective/entropy": -35.42715072631836,
620
+ "objective/kl": 28.722532272338867,
621
+ "objective/non_score_reward": -0.14361265301704407,
622
+ "objective/rlhf_reward": 9.159289360046387,
623
+ "objective/scores": 9.302902221679688,
624
+ "policy/approxkl_avg": 0.221934974193573,
625
+ "policy/clipfrac_avg": 0.3046875,
626
+ "policy/entropy_avg": 0.3945683240890503,
627
+ "step": 30,
628
+ "val/clipfrac_avg": 0.0,
629
+ "val/num_eos_tokens": 2236,
630
+ "val/ratio": 1.0000040531158447,
631
+ "val/ratio_var": 3.5196928820369067e-06
632
+ },
633
+ {
634
+ "episode": 16384,
635
+ "epoch": 0.07018505825908156,
636
+ "eps": 6,
637
+ "loss/policy_avg": 0.4800252318382263,
638
+ "lr": 2.63671875e-06,
639
+ "objective/entropy": -33.33995056152344,
640
+ "objective/kl": 29.83435821533203,
641
+ "objective/non_score_reward": -0.14917179942131042,
642
+ "objective/rlhf_reward": 9.096776962280273,
643
+ "objective/scores": 9.245948791503906,
644
+ "policy/approxkl_avg": 0.267780601978302,
645
+ "policy/clipfrac_avg": 0.337890625,
646
+ "policy/entropy_avg": 0.401678204536438,
647
+ "step": 31,
648
+ "val/clipfrac_avg": 0.0,
649
+ "val/num_eos_tokens": 2179,
650
+ "val/ratio": 1.0002527236938477,
651
+ "val/ratio_var": 3.2502473459317116e-06
652
+ },
653
+ {
654
+ "episode": 16896,
655
+ "epoch": 0.07237834132967787,
656
+ "eps": 6,
657
+ "loss/policy_avg": 0.2809556722640991,
658
+ "lr": 2.6250000000000003e-06,
659
+ "objective/entropy": -34.06169509887695,
660
+ "objective/kl": 31.27130889892578,
661
+ "objective/non_score_reward": -0.156356543302536,
662
+ "objective/rlhf_reward": 9.190299034118652,
663
+ "objective/scores": 9.346654891967773,
664
+ "policy/approxkl_avg": 0.2590065896511078,
665
+ "policy/clipfrac_avg": 0.333984375,
666
+ "policy/entropy_avg": 0.39545533061027527,
667
+ "step": 32,
668
+ "val/clipfrac_avg": 0.0,
669
+ "val/num_eos_tokens": 2267,
670
+ "val/ratio": 1.0000057220458984,
671
+ "val/ratio_var": 3.7364695799624315e-06
672
+ },
673
+ {
674
+ "episode": 17408,
675
+ "epoch": 0.07457162440027416,
676
+ "eps": 6,
677
+ "loss/policy_avg": 0.22726668417453766,
678
+ "lr": 2.61328125e-06,
679
+ "objective/entropy": -29.238605499267578,
680
+ "objective/kl": 32.63374710083008,
681
+ "objective/non_score_reward": -0.16316872835159302,
682
+ "objective/rlhf_reward": 9.312760353088379,
683
+ "objective/scores": 9.475929260253906,
684
+ "policy/approxkl_avg": 0.25571638345718384,
685
+ "policy/clipfrac_avg": 0.328125,
686
+ "policy/entropy_avg": 0.38992634415626526,
687
+ "step": 33,
688
+ "val/clipfrac_avg": 0.0,
689
+ "val/num_eos_tokens": 2153,
690
+ "val/ratio": 0.9997001886367798,
691
+ "val/ratio_var": 2.696072442631703e-06
692
+ },
693
+ {
694
+ "episode": 17920,
695
+ "epoch": 0.07676490747087046,
696
+ "eps": 6,
697
+ "loss/policy_avg": 0.31760329008102417,
698
+ "lr": 2.6015625e-06,
699
+ "objective/entropy": -30.36035919189453,
700
+ "objective/kl": 33.105316162109375,
701
+ "objective/non_score_reward": -0.16552656888961792,
702
+ "objective/rlhf_reward": 9.344242095947266,
703
+ "objective/scores": 9.509769439697266,
704
+ "policy/approxkl_avg": 0.26468268036842346,
705
+ "policy/clipfrac_avg": 0.31640625,
706
+ "policy/entropy_avg": 0.3962373435497284,
707
+ "step": 34,
708
+ "val/clipfrac_avg": 0.0,
709
+ "val/num_eos_tokens": 2063,
710
+ "val/ratio": 0.9997377395629883,
711
+ "val/ratio_var": 3.3684978006931487e-06
712
+ },
713
+ {
714
+ "episode": 18432,
715
+ "epoch": 0.07895819054146676,
716
+ "eps": 6,
717
+ "loss/policy_avg": 0.29205620288848877,
718
+ "lr": 2.5898437500000003e-06,
719
+ "objective/entropy": -31.297311782836914,
720
+ "objective/kl": 34.06657409667969,
721
+ "objective/non_score_reward": -0.1703328639268875,
722
+ "objective/rlhf_reward": 9.36227035522461,
723
+ "objective/scores": 9.532604217529297,
724
+ "policy/approxkl_avg": 0.23790021240711212,
725
+ "policy/clipfrac_avg": 0.359375,
726
+ "policy/entropy_avg": 0.39074409008026123,
727
+ "step": 35,
728
+ "val/clipfrac_avg": 0.0,
729
+ "val/num_eos_tokens": 2030,
730
+ "val/ratio": 1.0001909732818604,
731
+ "val/ratio_var": 4.307842573325615e-06
732
+ },
733
+ {
734
+ "episode": 18944,
735
+ "epoch": 0.08115147361206305,
736
+ "eps": 6,
737
+ "loss/policy_avg": 0.27079224586486816,
738
+ "lr": 2.578125e-06,
739
+ "objective/entropy": -31.059741973876953,
740
+ "objective/kl": 34.23755645751953,
741
+ "objective/non_score_reward": -0.17118775844573975,
742
+ "objective/rlhf_reward": 9.575616836547852,
743
+ "objective/scores": 9.746803283691406,
744
+ "policy/approxkl_avg": 0.2780650556087494,
745
+ "policy/clipfrac_avg": 0.359375,
746
+ "policy/entropy_avg": 0.36144694685935974,
747
+ "step": 36,
748
+ "val/clipfrac_avg": 0.0,
749
+ "val/num_eos_tokens": 1946,
750
+ "val/ratio": 1.0000677108764648,
751
+ "val/ratio_var": 5.890160991839366e-06
752
+ },
753
+ {
754
+ "episode": 19456,
755
+ "epoch": 0.08334475668265935,
756
+ "eps": 6,
757
+ "loss/policy_avg": 0.35481345653533936,
758
+ "lr": 2.56640625e-06,
759
+ "objective/entropy": -29.504077911376953,
760
+ "objective/kl": 35.572723388671875,
761
+ "objective/non_score_reward": -0.17786362767219543,
762
+ "objective/rlhf_reward": 9.650212287902832,
763
+ "objective/scores": 9.828075408935547,
764
+ "policy/approxkl_avg": 0.27989301085472107,
765
+ "policy/clipfrac_avg": 0.341796875,
766
+ "policy/entropy_avg": 0.36728498339653015,
767
+ "step": 37,
768
+ "val/clipfrac_avg": 0.0,
769
+ "val/num_eos_tokens": 1913,
770
+ "val/ratio": 1.000013828277588,
771
+ "val/ratio_var": 3.834660674328916e-06
772
+ },
773
+ {
774
+ "episode": 19968,
775
+ "epoch": 0.08553803975325565,
776
+ "eps": 6,
777
+ "loss/policy_avg": 0.32023417949676514,
778
+ "lr": 2.5546875000000003e-06,
779
+ "objective/entropy": -32.94467544555664,
780
+ "objective/kl": 35.05192565917969,
781
+ "objective/non_score_reward": -0.17525961995124817,
782
+ "objective/rlhf_reward": 9.715032577514648,
783
+ "objective/scores": 9.89029312133789,
784
+ "policy/approxkl_avg": 0.28202196955680847,
785
+ "policy/clipfrac_avg": 0.32421875,
786
+ "policy/entropy_avg": 0.3595014214515686,
787
+ "step": 38,
788
+ "val/clipfrac_avg": 0.0,
789
+ "val/num_eos_tokens": 1895,
790
+ "val/ratio": 1.0004128217697144,
791
+ "val/ratio_var": 3.976154857809888e-06
792
+ },
793
+ {
794
+ "episode": 20480,
795
+ "epoch": 0.08773132282385196,
796
+ "eps": 6,
797
+ "loss/policy_avg": 0.3284844160079956,
798
+ "lr": 2.54296875e-06,
799
+ "objective/entropy": -33.950965881347656,
800
+ "objective/kl": 35.06300354003906,
801
+ "objective/non_score_reward": -0.1753150224685669,
802
+ "objective/rlhf_reward": 9.614059448242188,
803
+ "objective/scores": 9.789375305175781,
804
+ "policy/approxkl_avg": 0.2808968126773834,
805
+ "policy/clipfrac_avg": 0.3359375,
806
+ "policy/entropy_avg": 0.3299378454685211,
807
+ "step": 39,
808
+ "val/clipfrac_avg": 0.0,
809
+ "val/num_eos_tokens": 1822,
810
+ "val/ratio": 1.0001529455184937,
811
+ "val/ratio_var": 3.939061116398079e-06
812
+ },
813
+ {
814
+ "episode": 20992,
815
+ "epoch": 0.08992460589444826,
816
+ "eps": 6,
817
+ "loss/policy_avg": 0.25570446252822876,
818
+ "lr": 2.53125e-06,
819
+ "objective/entropy": -33.10383605957031,
820
+ "objective/kl": 37.40079116821289,
821
+ "objective/non_score_reward": -0.1870039403438568,
822
+ "objective/rlhf_reward": 9.649069786071777,
823
+ "objective/scores": 9.836073875427246,
824
+ "policy/approxkl_avg": 0.28119906783103943,
825
+ "policy/clipfrac_avg": 0.349609375,
826
+ "policy/entropy_avg": 0.33145490288734436,
827
+ "step": 40,
828
+ "val/clipfrac_avg": 0.0,
829
+ "val/num_eos_tokens": 1757,
830
+ "val/ratio": 0.999976396560669,
831
+ "val/ratio_var": 3.935067525162594e-06
832
+ },
833
+ {
834
+ "episode": 21504,
835
+ "epoch": 0.09211788896504455,
836
+ "eps": 6,
837
+ "loss/policy_avg": 0.3803982436656952,
838
+ "lr": 2.5195312500000003e-06,
839
+ "objective/entropy": -35.517391204833984,
840
+ "objective/kl": 35.55995178222656,
841
+ "objective/non_score_reward": -0.1777997612953186,
842
+ "objective/rlhf_reward": 9.623468399047852,
843
+ "objective/scores": 9.801267623901367,
844
+ "policy/approxkl_avg": 0.28167036175727844,
845
+ "policy/clipfrac_avg": 0.3828125,
846
+ "policy/entropy_avg": 0.33138006925582886,
847
+ "step": 41,
848
+ "val/clipfrac_avg": 0.0,
849
+ "val/num_eos_tokens": 1765,
850
+ "val/ratio": 1.000349760055542,
851
+ "val/ratio_var": 4.55302324553486e-06
852
+ },
853
+ {
854
+ "episode": 22016,
855
+ "epoch": 0.09431117203564085,
856
+ "eps": 6,
857
+ "loss/policy_avg": 0.2066161185503006,
858
+ "lr": 2.5078125e-06,
859
+ "objective/entropy": -34.937923431396484,
860
+ "objective/kl": 34.828224182128906,
861
+ "objective/non_score_reward": -0.17414110898971558,
862
+ "objective/rlhf_reward": 10.007875442504883,
863
+ "objective/scores": 10.18201732635498,
864
+ "policy/approxkl_avg": 0.25226256251335144,
865
+ "policy/clipfrac_avg": 0.349609375,
866
+ "policy/entropy_avg": 0.31548449397087097,
867
+ "step": 42,
868
+ "val/clipfrac_avg": 0.0,
869
+ "val/num_eos_tokens": 1692,
870
+ "val/ratio": 1.0003182888031006,
871
+ "val/ratio_var": 3.0859027901897207e-06
872
+ },
873
+ {
874
+ "episode": 22528,
875
+ "epoch": 0.09650445510623715,
876
+ "eps": 6,
877
+ "loss/policy_avg": 0.21718214452266693,
878
+ "lr": 2.49609375e-06,
879
+ "objective/entropy": -36.30521774291992,
880
+ "objective/kl": 35.64817810058594,
881
+ "objective/non_score_reward": -0.17824086546897888,
882
+ "objective/rlhf_reward": 9.524388313293457,
883
+ "objective/scores": 9.702629089355469,
884
+ "policy/approxkl_avg": 0.3034236431121826,
885
+ "policy/clipfrac_avg": 0.359375,
886
+ "policy/entropy_avg": 0.333159863948822,
887
+ "step": 43,
888
+ "val/clipfrac_avg": 0.0,
889
+ "val/num_eos_tokens": 1736,
890
+ "val/ratio": 0.9996454119682312,
891
+ "val/ratio_var": 3.5439479688648134e-06
892
+ },
893
+ {
894
+ "episode": 23040,
895
+ "epoch": 0.09869773817683344,
896
+ "eps": 6,
897
+ "loss/policy_avg": 0.20844724774360657,
898
+ "lr": 2.4843750000000002e-06,
899
+ "objective/entropy": -37.294647216796875,
900
+ "objective/kl": 34.449188232421875,
901
+ "objective/non_score_reward": -0.17224593460559845,
902
+ "objective/rlhf_reward": 9.562028884887695,
903
+ "objective/scores": 9.734274864196777,
904
+ "policy/approxkl_avg": 0.2691035270690918,
905
+ "policy/clipfrac_avg": 0.369140625,
906
+ "policy/entropy_avg": 0.3254753351211548,
907
+ "step": 44,
908
+ "val/clipfrac_avg": 0.0,
909
+ "val/num_eos_tokens": 1712,
910
+ "val/ratio": 0.9999078512191772,
911
+ "val/ratio_var": 3.873953573929612e-06
912
+ },
913
+ {
914
+ "episode": 23552,
915
+ "epoch": 0.10089102124742974,
916
+ "eps": 6,
917
+ "loss/policy_avg": 0.26617997884750366,
918
+ "lr": 2.47265625e-06,
919
+ "objective/entropy": -38.45973205566406,
920
+ "objective/kl": 33.36463165283203,
921
+ "objective/non_score_reward": -0.166823148727417,
922
+ "objective/rlhf_reward": 9.529053688049316,
923
+ "objective/scores": 9.695877075195312,
924
+ "policy/approxkl_avg": 0.2848384976387024,
925
+ "policy/clipfrac_avg": 0.32421875,
926
+ "policy/entropy_avg": 0.33362650871276855,
927
+ "step": 45,
928
+ "val/clipfrac_avg": 0.0,
929
+ "val/num_eos_tokens": 1776,
930
+ "val/ratio": 0.9998953938484192,
931
+ "val/ratio_var": 4.734669346362352e-06
932
+ },
933
+ {
934
+ "episode": 24064,
935
+ "epoch": 0.10308430431802605,
936
+ "eps": 6,
937
+ "loss/policy_avg": 0.14936941862106323,
938
+ "lr": 2.4609375e-06,
939
+ "objective/entropy": -41.61138153076172,
940
+ "objective/kl": 32.80962371826172,
941
+ "objective/non_score_reward": -0.16404810547828674,
942
+ "objective/rlhf_reward": 9.842905044555664,
943
+ "objective/scores": 10.006953239440918,
944
+ "policy/approxkl_avg": 0.237547367811203,
945
+ "policy/clipfrac_avg": 0.376953125,
946
+ "policy/entropy_avg": 0.311695396900177,
947
+ "step": 46,
948
+ "val/clipfrac_avg": 0.0,
949
+ "val/num_eos_tokens": 1758,
950
+ "val/ratio": 1.0000953674316406,
951
+ "val/ratio_var": 2.808139925036812e-06
952
+ },
953
+ {
954
+ "episode": 24576,
955
+ "epoch": 0.10527758738862235,
956
+ "eps": 6,
957
+ "loss/policy_avg": 0.30603936314582825,
958
+ "lr": 2.4492187500000002e-06,
959
+ "objective/entropy": -41.17787551879883,
960
+ "objective/kl": 32.624759674072266,
961
+ "objective/non_score_reward": -0.16312378644943237,
962
+ "objective/rlhf_reward": 9.724346160888672,
963
+ "objective/scores": 9.887470245361328,
964
+ "policy/approxkl_avg": 0.29364651441574097,
965
+ "policy/clipfrac_avg": 0.353515625,
966
+ "policy/entropy_avg": 0.3342776894569397,
967
+ "step": 47,
968
+ "val/clipfrac_avg": 0.0,
969
+ "val/num_eos_tokens": 1791,
970
+ "val/ratio": 1.000326156616211,
971
+ "val/ratio_var": 4.5322440200834535e-06
972
+ },
973
+ {
974
+ "episode": 25088,
975
+ "epoch": 0.10747087045921865,
976
+ "eps": 6,
977
+ "loss/policy_avg": 0.2642097771167755,
978
+ "lr": 2.4375e-06,
979
+ "objective/entropy": -39.200443267822266,
980
+ "objective/kl": 32.45161437988281,
981
+ "objective/non_score_reward": -0.1622580885887146,
982
+ "objective/rlhf_reward": 9.750046730041504,
983
+ "objective/scores": 9.912304878234863,
984
+ "policy/approxkl_avg": 0.30972546339035034,
985
+ "policy/clipfrac_avg": 0.345703125,
986
+ "policy/entropy_avg": 0.33516496419906616,
987
+ "step": 48,
988
+ "val/clipfrac_avg": 0.0,
989
+ "val/num_eos_tokens": 1872,
990
+ "val/ratio": 1.0000823736190796,
991
+ "val/ratio_var": 3.669098759928602e-06
992
+ },
993
+ {
994
+ "episode": 25600,
995
+ "epoch": 0.10966415352981494,
996
+ "eps": 6,
997
+ "loss/policy_avg": 0.2650497555732727,
998
+ "lr": 2.42578125e-06,
999
+ "objective/entropy": -42.08147430419922,
1000
+ "objective/kl": 32.27870178222656,
1001
+ "objective/non_score_reward": -0.16139349341392517,
1002
+ "objective/rlhf_reward": 9.627498626708984,
1003
+ "objective/scores": 9.788891792297363,
1004
+ "policy/approxkl_avg": 0.24272559583187103,
1005
+ "policy/clipfrac_avg": 0.34375,
1006
+ "policy/entropy_avg": 0.3419151306152344,
1007
+ "step": 49,
1008
+ "val/clipfrac_avg": 0.0,
1009
+ "val/num_eos_tokens": 1885,
1010
+ "val/ratio": 1.0000895261764526,
1011
+ "val/ratio_var": 3.1425495308212703e-06
1012
+ },
1013
+ {
1014
+ "episode": 26112,
1015
+ "epoch": 0.11185743660041124,
1016
+ "eps": 6,
1017
+ "loss/policy_avg": 0.19408683478832245,
1018
+ "lr": 2.4140625000000002e-06,
1019
+ "objective/entropy": -40.94256591796875,
1020
+ "objective/kl": 31.59925651550293,
1021
+ "objective/non_score_reward": -0.1579962968826294,
1022
+ "objective/rlhf_reward": 9.357413291931152,
1023
+ "objective/scores": 9.515409469604492,
1024
+ "policy/approxkl_avg": 0.2539810538291931,
1025
+ "policy/clipfrac_avg": 0.345703125,
1026
+ "policy/entropy_avg": 0.33615610003471375,
1027
+ "step": 50,
1028
+ "val/clipfrac_avg": 0.0,
1029
+ "val/num_eos_tokens": 1861,
1030
+ "val/ratio": 1.0000097751617432,
1031
+ "val/ratio_var": 4.071615421707975e-06
1032
+ },
1033
+ {
1034
+ "episode": 26624,
1035
+ "epoch": 0.11405071967100754,
1036
+ "eps": 6,
1037
+ "loss/policy_avg": 0.12506979703903198,
1038
+ "lr": 2.40234375e-06,
1039
+ "objective/entropy": -43.869224548339844,
1040
+ "objective/kl": 30.23246955871582,
1041
+ "objective/non_score_reward": -0.15116234123706818,
1042
+ "objective/rlhf_reward": 9.635049819946289,
1043
+ "objective/scores": 9.786212921142578,
1044
+ "policy/approxkl_avg": 0.22058835625648499,
1045
+ "policy/clipfrac_avg": 0.3828125,
1046
+ "policy/entropy_avg": 0.35689815878868103,
1047
+ "step": 51,
1048
+ "val/clipfrac_avg": 0.0,
1049
+ "val/num_eos_tokens": 1916,
1050
+ "val/ratio": 1.0003091096878052,
1051
+ "val/ratio_var": 3.508681174935191e-06
1052
+ },
1053
+ {
1054
+ "episode": 27136,
1055
+ "epoch": 0.11624400274160383,
1056
+ "eps": 6,
1057
+ "loss/policy_avg": 0.2060900628566742,
1058
+ "lr": 2.390625e-06,
1059
+ "objective/entropy": -45.10155487060547,
1060
+ "objective/kl": 30.40532684326172,
1061
+ "objective/non_score_reward": -0.15202662348747253,
1062
+ "objective/rlhf_reward": 9.444595336914062,
1063
+ "objective/scores": 9.596622467041016,
1064
+ "policy/approxkl_avg": 0.2469383329153061,
1065
+ "policy/clipfrac_avg": 0.337890625,
1066
+ "policy/entropy_avg": 0.33809155225753784,
1067
+ "step": 52,
1068
+ "val/clipfrac_avg": 0.0,
1069
+ "val/num_eos_tokens": 1910,
1070
+ "val/ratio": 0.9997259974479675,
1071
+ "val/ratio_var": 3.441498392930953e-06
1072
+ },
1073
+ {
1074
+ "episode": 27648,
1075
+ "epoch": 0.11843728581220014,
1076
+ "eps": 6,
1077
+ "loss/policy_avg": 0.19970384240150452,
1078
+ "lr": 2.3789062500000002e-06,
1079
+ "objective/entropy": -43.50267028808594,
1080
+ "objective/kl": 29.95575714111328,
1081
+ "objective/non_score_reward": -0.14977876842021942,
1082
+ "objective/rlhf_reward": 9.758223533630371,
1083
+ "objective/scores": 9.908000946044922,
1084
+ "policy/approxkl_avg": 0.23129594326019287,
1085
+ "policy/clipfrac_avg": 0.34765625,
1086
+ "policy/entropy_avg": 0.3461906909942627,
1087
+ "step": 53,
1088
+ "val/clipfrac_avg": 0.0,
1089
+ "val/num_eos_tokens": 1897,
1090
+ "val/ratio": 1.00007164478302,
1091
+ "val/ratio_var": 4.8727483772381674e-06
1092
+ },
1093
+ {
1094
+ "episode": 28160,
1095
+ "epoch": 0.12063056888279644,
1096
+ "eps": 6,
1097
+ "loss/policy_avg": 0.17737600207328796,
1098
+ "lr": 2.3671875e-06,
1099
+ "objective/entropy": -43.82114791870117,
1100
+ "objective/kl": 30.448320388793945,
1101
+ "objective/non_score_reward": -0.15224160254001617,
1102
+ "objective/rlhf_reward": 9.586687088012695,
1103
+ "objective/scores": 9.738929748535156,
1104
+ "policy/approxkl_avg": 0.2304307222366333,
1105
+ "policy/clipfrac_avg": 0.38671875,
1106
+ "policy/entropy_avg": 0.35419148206710815,
1107
+ "step": 54,
1108
+ "val/clipfrac_avg": 0.0,
1109
+ "val/num_eos_tokens": 1930,
1110
+ "val/ratio": 1.000131368637085,
1111
+ "val/ratio_var": 2.835533905454213e-06
1112
+ },
1113
+ {
1114
+ "episode": 28672,
1115
+ "epoch": 0.12282385195339274,
1116
+ "eps": 6,
1117
+ "loss/policy_avg": 0.20781438052654266,
1118
+ "lr": 2.35546875e-06,
1119
+ "objective/entropy": -42.808937072753906,
1120
+ "objective/kl": 30.716829299926758,
1121
+ "objective/non_score_reward": -0.15358413755893707,
1122
+ "objective/rlhf_reward": 9.48070240020752,
1123
+ "objective/scores": 9.634286880493164,
1124
+ "policy/approxkl_avg": 0.2257956713438034,
1125
+ "policy/clipfrac_avg": 0.34375,
1126
+ "policy/entropy_avg": 0.35503190755844116,
1127
+ "step": 55,
1128
+ "val/clipfrac_avg": 0.0,
1129
+ "val/num_eos_tokens": 1936,
1130
+ "val/ratio": 0.9999227523803711,
1131
+ "val/ratio_var": 3.047591235372238e-06
1132
+ },
1133
+ {
1134
+ "episode": 29184,
1135
+ "epoch": 0.12501713502398903,
1136
+ "eps": 6,
1137
+ "loss/policy_avg": 0.17236250638961792,
1138
+ "lr": 2.3437500000000002e-06,
1139
+ "objective/entropy": -42.00170135498047,
1140
+ "objective/kl": 30.85394859313965,
1141
+ "objective/non_score_reward": -0.154269739985466,
1142
+ "objective/rlhf_reward": 9.769851684570312,
1143
+ "objective/scores": 9.924121856689453,
1144
+ "policy/approxkl_avg": 0.25827232003211975,
1145
+ "policy/clipfrac_avg": 0.365234375,
1146
+ "policy/entropy_avg": 0.35941898822784424,
1147
+ "step": 56,
1148
+ "val/clipfrac_avg": 0.0,
1149
+ "val/num_eos_tokens": 1899,
1150
+ "val/ratio": 0.9998503923416138,
1151
+ "val/ratio_var": 3.5402435969444923e-06
1152
+ },
1153
+ {
1154
+ "episode": 29696,
1155
+ "epoch": 0.12721041809458533,
1156
+ "eps": 6,
1157
+ "loss/policy_avg": 0.1874779909849167,
1158
+ "lr": 2.33203125e-06,
1159
+ "objective/entropy": -41.37017822265625,
1160
+ "objective/kl": 30.405405044555664,
1161
+ "objective/non_score_reward": -0.15202701091766357,
1162
+ "objective/rlhf_reward": 9.552864074707031,
1163
+ "objective/scores": 9.704890251159668,
1164
+ "policy/approxkl_avg": 0.2579526901245117,
1165
+ "policy/clipfrac_avg": 0.3671875,
1166
+ "policy/entropy_avg": 0.36883309483528137,
1167
+ "step": 57,
1168
+ "val/clipfrac_avg": 0.0,
1169
+ "val/num_eos_tokens": 1996,
1170
+ "val/ratio": 1.0000630617141724,
1171
+ "val/ratio_var": 3.2543882753088837e-06
1172
+ },
1173
+ {
1174
+ "episode": 30208,
1175
+ "epoch": 0.12940370116518163,
1176
+ "eps": 6,
1177
+ "loss/policy_avg": 0.24861405789852142,
1178
+ "lr": 2.3203125e-06,
1179
+ "objective/entropy": -39.66204833984375,
1180
+ "objective/kl": 32.23133850097656,
1181
+ "objective/non_score_reward": -0.16115668416023254,
1182
+ "objective/rlhf_reward": 9.291080474853516,
1183
+ "objective/scores": 9.452238082885742,
1184
+ "policy/approxkl_avg": 0.24864479899406433,
1185
+ "policy/clipfrac_avg": 0.34765625,
1186
+ "policy/entropy_avg": 0.39884617924690247,
1187
+ "step": 58,
1188
+ "val/clipfrac_avg": 0.0,
1189
+ "val/num_eos_tokens": 1956,
1190
+ "val/ratio": 0.9995733499526978,
1191
+ "val/ratio_var": 2.4565586045355303e-06
1192
+ },
1193
+ {
1194
+ "episode": 30720,
1195
+ "epoch": 0.13159698423577793,
1196
+ "eps": 6,
1197
+ "loss/policy_avg": 0.15614402294158936,
1198
+ "lr": 2.30859375e-06,
1199
+ "objective/entropy": -39.843353271484375,
1200
+ "objective/kl": 31.747220993041992,
1201
+ "objective/non_score_reward": -0.15873610973358154,
1202
+ "objective/rlhf_reward": 9.32120418548584,
1203
+ "objective/scores": 9.479940414428711,
1204
+ "policy/approxkl_avg": 0.2175564169883728,
1205
+ "policy/clipfrac_avg": 0.375,
1206
+ "policy/entropy_avg": 0.40173786878585815,
1207
+ "step": 59,
1208
+ "val/clipfrac_avg": 0.0,
1209
+ "val/num_eos_tokens": 1870,
1210
+ "val/ratio": 1.0000216960906982,
1211
+ "val/ratio_var": 2.7907508410862647e-06
1212
+ },
1213
+ {
1214
+ "episode": 31232,
1215
+ "epoch": 0.13379026730637422,
1216
+ "eps": 6,
1217
+ "loss/policy_avg": 0.1781693994998932,
1218
+ "lr": 2.296875e-06,
1219
+ "objective/entropy": -39.122676849365234,
1220
+ "objective/kl": 32.07661437988281,
1221
+ "objective/non_score_reward": -0.16038307547569275,
1222
+ "objective/rlhf_reward": 9.458475112915039,
1223
+ "objective/scores": 9.618858337402344,
1224
+ "policy/approxkl_avg": 0.2337334156036377,
1225
+ "policy/clipfrac_avg": 0.3828125,
1226
+ "policy/entropy_avg": 0.40583640336990356,
1227
+ "step": 60,
1228
+ "val/clipfrac_avg": 0.0,
1229
+ "val/num_eos_tokens": 1988,
1230
+ "val/ratio": 1.000226378440857,
1231
+ "val/ratio_var": 2.6703253297455376e-06
1232
+ },
1233
+ {
1234
+ "episode": 31744,
1235
+ "epoch": 0.13598355037697052,
1236
+ "eps": 6,
1237
+ "loss/policy_avg": 0.22470299899578094,
1238
+ "lr": 2.28515625e-06,
1239
+ "objective/entropy": -35.38658905029297,
1240
+ "objective/kl": 33.76303482055664,
1241
+ "objective/non_score_reward": -0.16881518065929413,
1242
+ "objective/rlhf_reward": 9.489850044250488,
1243
+ "objective/scores": 9.658665657043457,
1244
+ "policy/approxkl_avg": 0.2829976975917816,
1245
+ "policy/clipfrac_avg": 0.359375,
1246
+ "policy/entropy_avg": 0.40766507387161255,
1247
+ "step": 61,
1248
+ "val/clipfrac_avg": 0.0,
1249
+ "val/num_eos_tokens": 1899,
1250
+ "val/ratio": 1.0002920627593994,
1251
+ "val/ratio_var": 4.702184469351778e-06
1252
+ },
1253
+ {
1254
+ "episode": 32256,
1255
+ "epoch": 0.13817683344756682,
1256
+ "eps": 6,
1257
+ "loss/policy_avg": 0.18476197123527527,
1258
+ "lr": 2.2734375e-06,
1259
+ "objective/entropy": -35.610679626464844,
1260
+ "objective/kl": 32.4565544128418,
1261
+ "objective/non_score_reward": -0.1622827649116516,
1262
+ "objective/rlhf_reward": 9.31367015838623,
1263
+ "objective/scores": 9.475953102111816,
1264
+ "policy/approxkl_avg": 0.2603687644004822,
1265
+ "policy/clipfrac_avg": 0.375,
1266
+ "policy/entropy_avg": 0.4185265302658081,
1267
+ "step": 62,
1268
+ "val/clipfrac_avg": 0.0,
1269
+ "val/num_eos_tokens": 1909,
1270
+ "val/ratio": 0.9998178482055664,
1271
+ "val/ratio_var": 3.281563522250508e-06
1272
+ },
1273
+ {
1274
+ "episode": 32768,
1275
+ "epoch": 0.1403701165181631,
1276
+ "eps": 6,
1277
+ "loss/policy_avg": 0.26672354340553284,
1278
+ "lr": 2.26171875e-06,
1279
+ "objective/entropy": -34.83911895751953,
1280
+ "objective/kl": 33.32848358154297,
1281
+ "objective/non_score_reward": -0.166642427444458,
1282
+ "objective/rlhf_reward": 9.365612030029297,
1283
+ "objective/scores": 9.532255172729492,
1284
+ "policy/approxkl_avg": 0.21640878915786743,
1285
+ "policy/clipfrac_avg": 0.35546875,
1286
+ "policy/entropy_avg": 0.4103761911392212,
1287
+ "step": 63,
1288
+ "val/clipfrac_avg": 0.0,
1289
+ "val/num_eos_tokens": 1924,
1290
+ "val/ratio": 1.000182867050171,
1291
+ "val/ratio_var": 2.308673174411524e-06
1292
+ },
1293
+ {
1294
+ "episode": 33280,
1295
+ "epoch": 0.14256339958875944,
1296
+ "eps": 6,
1297
+ "loss/policy_avg": 0.22516316175460815,
1298
+ "lr": 2.25e-06,
1299
+ "objective/entropy": -35.21662902832031,
1300
+ "objective/kl": 33.290374755859375,
1301
+ "objective/non_score_reward": -0.16645187139511108,
1302
+ "objective/rlhf_reward": 9.381125450134277,
1303
+ "objective/scores": 9.547576904296875,
1304
+ "policy/approxkl_avg": 0.3286188542842865,
1305
+ "policy/clipfrac_avg": 0.357421875,
1306
+ "policy/entropy_avg": 0.41928941011428833,
1307
+ "step": 64,
1308
+ "val/clipfrac_avg": 0.0,
1309
+ "val/num_eos_tokens": 1964,
1310
+ "val/ratio": 1.000014066696167,
1311
+ "val/ratio_var": 3.4851202599384123e-06
1312
+ },
1313
+ {
1314
+ "episode": 33792,
1315
+ "epoch": 0.14475668265935573,
1316
+ "eps": 6,
1317
+ "loss/policy_avg": 0.18076206743717194,
1318
+ "lr": 2.23828125e-06,
1319
+ "objective/entropy": -37.27294921875,
1320
+ "objective/kl": 32.67399978637695,
1321
+ "objective/non_score_reward": -0.16336998343467712,
1322
+ "objective/rlhf_reward": 9.350773811340332,
1323
+ "objective/scores": 9.514143943786621,
1324
+ "policy/approxkl_avg": 0.23823928833007812,
1325
+ "policy/clipfrac_avg": 0.388671875,
1326
+ "policy/entropy_avg": 0.4244874119758606,
1327
+ "step": 65,
1328
+ "val/clipfrac_avg": 0.0,
1329
+ "val/num_eos_tokens": 1935,
1330
+ "val/ratio": 0.9997513294219971,
1331
+ "val/ratio_var": 3.028096216439735e-06
1332
+ },
1333
+ {
1334
+ "episode": 34304,
1335
+ "epoch": 0.14694996572995203,
1336
+ "eps": 6,
1337
+ "loss/policy_avg": 0.17251071333885193,
1338
+ "lr": 2.2265625e-06,
1339
+ "objective/entropy": -33.87648010253906,
1340
+ "objective/kl": 33.30876159667969,
1341
+ "objective/non_score_reward": -0.16654381155967712,
1342
+ "objective/rlhf_reward": 9.384592056274414,
1343
+ "objective/scores": 9.551136016845703,
1344
+ "policy/approxkl_avg": 0.25553739070892334,
1345
+ "policy/clipfrac_avg": 0.369140625,
1346
+ "policy/entropy_avg": 0.450764536857605,
1347
+ "step": 66,
1348
+ "val/clipfrac_avg": 0.0,
1349
+ "val/num_eos_tokens": 1977,
1350
+ "val/ratio": 0.9999786019325256,
1351
+ "val/ratio_var": 3.189375547663076e-06
1352
+ },
1353
+ {
1354
+ "episode": 34816,
1355
+ "epoch": 0.14914324880054833,
1356
+ "eps": 6,
1357
+ "loss/policy_avg": 0.2089308649301529,
1358
+ "lr": 2.21484375e-06,
1359
+ "objective/entropy": -34.71873092651367,
1360
+ "objective/kl": 32.921295166015625,
1361
+ "objective/non_score_reward": -0.1646064817905426,
1362
+ "objective/rlhf_reward": 9.405241966247559,
1363
+ "objective/scores": 9.56984806060791,
1364
+ "policy/approxkl_avg": 0.24852168560028076,
1365
+ "policy/clipfrac_avg": 0.369140625,
1366
+ "policy/entropy_avg": 0.4264836013317108,
1367
+ "step": 67,
1368
+ "val/clipfrac_avg": 0.0,
1369
+ "val/num_eos_tokens": 2023,
1370
+ "val/ratio": 0.9999498724937439,
1371
+ "val/ratio_var": 3.718509105965495e-06
1372
+ },
1373
+ {
1374
+ "episode": 35328,
1375
+ "epoch": 0.15133653187114463,
1376
+ "eps": 6,
1377
+ "loss/policy_avg": 0.30707770586013794,
1378
+ "lr": 2.203125e-06,
1379
+ "objective/entropy": -33.18562698364258,
1380
+ "objective/kl": 34.398094177246094,
1381
+ "objective/non_score_reward": -0.17199048399925232,
1382
+ "objective/rlhf_reward": 9.318955421447754,
1383
+ "objective/scores": 9.490945816040039,
1384
+ "policy/approxkl_avg": 0.2854606509208679,
1385
+ "policy/clipfrac_avg": 0.35546875,
1386
+ "policy/entropy_avg": 0.43861570954322815,
1387
+ "step": 68,
1388
+ "val/clipfrac_avg": 0.0,
1389
+ "val/num_eos_tokens": 2134,
1390
+ "val/ratio": 1.0002244710922241,
1391
+ "val/ratio_var": 3.3594094475120073e-06
1392
+ },
1393
+ {
1394
+ "episode": 35840,
1395
+ "epoch": 0.15352981494174092,
1396
+ "eps": 6,
1397
+ "loss/policy_avg": 0.21776500344276428,
1398
+ "lr": 2.19140625e-06,
1399
+ "objective/entropy": -33.90901565551758,
1400
+ "objective/kl": 33.42001724243164,
1401
+ "objective/non_score_reward": -0.16710007190704346,
1402
+ "objective/rlhf_reward": 9.314750671386719,
1403
+ "objective/scores": 9.481850624084473,
1404
+ "policy/approxkl_avg": 0.27862101793289185,
1405
+ "policy/clipfrac_avg": 0.318359375,
1406
+ "policy/entropy_avg": 0.4583716094493866,
1407
+ "step": 69,
1408
+ "val/clipfrac_avg": 0.0,
1409
+ "val/num_eos_tokens": 2177,
1410
+ "val/ratio": 1.0001673698425293,
1411
+ "val/ratio_var": 3.758811544685159e-06
1412
+ },
1413
+ {
1414
+ "episode": 36352,
1415
+ "epoch": 0.15572309801233722,
1416
+ "eps": 6,
1417
+ "loss/policy_avg": 0.2049793154001236,
1418
+ "lr": 2.1796875e-06,
1419
+ "objective/entropy": -33.58552169799805,
1420
+ "objective/kl": 33.38475799560547,
1421
+ "objective/non_score_reward": -0.16692380607128143,
1422
+ "objective/rlhf_reward": 9.471214294433594,
1423
+ "objective/scores": 9.638137817382812,
1424
+ "policy/approxkl_avg": 0.23455402255058289,
1425
+ "policy/clipfrac_avg": 0.353515625,
1426
+ "policy/entropy_avg": 0.4269852638244629,
1427
+ "step": 70,
1428
+ "val/clipfrac_avg": 0.0,
1429
+ "val/num_eos_tokens": 2168,
1430
+ "val/ratio": 1.0000548362731934,
1431
+ "val/ratio_var": 3.26197096001124e-06
1432
+ },
1433
+ {
1434
+ "episode": 36864,
1435
+ "epoch": 0.15791638108293352,
1436
+ "eps": 6,
1437
+ "loss/policy_avg": 0.2173652946949005,
1438
+ "lr": 2.16796875e-06,
1439
+ "objective/entropy": -32.59899139404297,
1440
+ "objective/kl": 34.50146484375,
1441
+ "objective/non_score_reward": -0.17250731587409973,
1442
+ "objective/rlhf_reward": 9.358166694641113,
1443
+ "objective/scores": 9.53067398071289,
1444
+ "policy/approxkl_avg": 0.26884737610816956,
1445
+ "policy/clipfrac_avg": 0.3828125,
1446
+ "policy/entropy_avg": 0.4478238523006439,
1447
+ "step": 71,
1448
+ "val/clipfrac_avg": 0.0,
1449
+ "val/num_eos_tokens": 2152,
1450
+ "val/ratio": 0.9999924302101135,
1451
+ "val/ratio_var": 3.7096308460604632e-06
1452
+ },
1453
+ {
1454
+ "episode": 37376,
1455
+ "epoch": 0.1601096641535298,
1456
+ "eps": 6,
1457
+ "loss/policy_avg": 0.1738312691450119,
1458
+ "lr": 2.15625e-06,
1459
+ "objective/entropy": -33.73279571533203,
1460
+ "objective/kl": 34.007720947265625,
1461
+ "objective/non_score_reward": -0.1700385957956314,
1462
+ "objective/rlhf_reward": 9.712380409240723,
1463
+ "objective/scores": 9.88241958618164,
1464
+ "policy/approxkl_avg": 0.27387821674346924,
1465
+ "policy/clipfrac_avg": 0.365234375,
1466
+ "policy/entropy_avg": 0.44403284788131714,
1467
+ "step": 72,
1468
+ "val/clipfrac_avg": 0.0,
1469
+ "val/num_eos_tokens": 2140,
1470
+ "val/ratio": 0.9999028444290161,
1471
+ "val/ratio_var": 4.58646536571905e-06
1472
+ },
1473
+ {
1474
+ "episode": 37888,
1475
+ "epoch": 0.1623029472241261,
1476
+ "eps": 6,
1477
+ "loss/policy_avg": 0.16620802879333496,
1478
+ "lr": 2.14453125e-06,
1479
+ "objective/entropy": -35.81139373779297,
1480
+ "objective/kl": 32.9052734375,
1481
+ "objective/non_score_reward": -0.1645263433456421,
1482
+ "objective/rlhf_reward": 9.479964256286621,
1483
+ "objective/scores": 9.644491195678711,
1484
+ "policy/approxkl_avg": 0.25569581985473633,
1485
+ "policy/clipfrac_avg": 0.3984375,
1486
+ "policy/entropy_avg": 0.42350655794143677,
1487
+ "step": 73,
1488
+ "val/clipfrac_avg": 0.0,
1489
+ "val/num_eos_tokens": 2139,
1490
+ "val/ratio": 0.9998481273651123,
1491
+ "val/ratio_var": 3.5795840176433558e-06
1492
+ },
1493
+ {
1494
+ "episode": 38400,
1495
+ "epoch": 0.1644962302947224,
1496
+ "eps": 6,
1497
+ "loss/policy_avg": 0.22138690948486328,
1498
+ "lr": 2.1328125e-06,
1499
+ "objective/entropy": -34.88341522216797,
1500
+ "objective/kl": 33.08055877685547,
1501
+ "objective/non_score_reward": -0.16540278494358063,
1502
+ "objective/rlhf_reward": 9.436692237854004,
1503
+ "objective/scores": 9.602094650268555,
1504
+ "policy/approxkl_avg": 0.21768277883529663,
1505
+ "policy/clipfrac_avg": 0.33203125,
1506
+ "policy/entropy_avg": 0.43899866938591003,
1507
+ "step": 74,
1508
+ "val/clipfrac_avg": 0.0,
1509
+ "val/num_eos_tokens": 2290,
1510
+ "val/ratio": 1.000012993812561,
1511
+ "val/ratio_var": 2.8327724521659547e-06
1512
+ },
1513
+ {
1514
+ "episode": 38912,
1515
+ "epoch": 0.1666895133653187,
1516
+ "eps": 6,
1517
+ "loss/policy_avg": 0.17552754282951355,
1518
+ "lr": 2.12109375e-06,
1519
+ "objective/entropy": -32.78399658203125,
1520
+ "objective/kl": 33.330780029296875,
1521
+ "objective/non_score_reward": -0.16665390133857727,
1522
+ "objective/rlhf_reward": 9.520395278930664,
1523
+ "objective/scores": 9.687049865722656,
1524
+ "policy/approxkl_avg": 0.24098150432109833,
1525
+ "policy/clipfrac_avg": 0.396484375,
1526
+ "policy/entropy_avg": 0.43451428413391113,
1527
+ "step": 75,
1528
+ "val/clipfrac_avg": 0.0,
1529
+ "val/num_eos_tokens": 2224,
1530
+ "val/ratio": 1.0003948211669922,
1531
+ "val/ratio_var": 3.863194706354989e-06
1532
+ },
1533
+ {
1534
+ "episode": 39424,
1535
+ "epoch": 0.168882796435915,
1536
+ "eps": 6,
1537
+ "loss/policy_avg": 0.16361400485038757,
1538
+ "lr": 2.109375e-06,
1539
+ "objective/entropy": -36.86976623535156,
1540
+ "objective/kl": 32.36768341064453,
1541
+ "objective/non_score_reward": -0.16183842718601227,
1542
+ "objective/rlhf_reward": 9.400718688964844,
1543
+ "objective/scores": 9.562556266784668,
1544
+ "policy/approxkl_avg": 0.2723867893218994,
1545
+ "policy/clipfrac_avg": 0.353515625,
1546
+ "policy/entropy_avg": 0.4423583149909973,
1547
+ "step": 76,
1548
+ "val/clipfrac_avg": 0.0,
1549
+ "val/num_eos_tokens": 2366,
1550
+ "val/ratio": 0.9998538494110107,
1551
+ "val/ratio_var": 3.881733846355928e-06
1552
+ },
1553
+ {
1554
+ "episode": 39936,
1555
+ "epoch": 0.1710760795065113,
1556
+ "eps": 6,
1557
+ "loss/policy_avg": 0.20113158226013184,
1558
+ "lr": 2.09765625e-06,
1559
+ "objective/entropy": -36.3262939453125,
1560
+ "objective/kl": 31.993881225585938,
1561
+ "objective/non_score_reward": -0.15996940433979034,
1562
+ "objective/rlhf_reward": 9.591435432434082,
1563
+ "objective/scores": 9.751405715942383,
1564
+ "policy/approxkl_avg": 0.21415743231773376,
1565
+ "policy/clipfrac_avg": 0.330078125,
1566
+ "policy/entropy_avg": 0.417125403881073,
1567
+ "step": 77,
1568
+ "val/clipfrac_avg": 0.0,
1569
+ "val/num_eos_tokens": 2304,
1570
+ "val/ratio": 1.0003247261047363,
1571
+ "val/ratio_var": 3.1106928872759454e-06
1572
+ },
1573
+ {
1574
+ "episode": 40448,
1575
+ "epoch": 0.17326936257710762,
1576
+ "eps": 6,
1577
+ "loss/policy_avg": 0.19219069182872772,
1578
+ "lr": 2.0859375e-06,
1579
+ "objective/entropy": -35.67451858520508,
1580
+ "objective/kl": 32.77976989746094,
1581
+ "objective/non_score_reward": -0.16389885544776917,
1582
+ "objective/rlhf_reward": 9.648540496826172,
1583
+ "objective/scores": 9.81243896484375,
1584
+ "policy/approxkl_avg": 0.24051856994628906,
1585
+ "policy/clipfrac_avg": 0.357421875,
1586
+ "policy/entropy_avg": 0.39866721630096436,
1587
+ "step": 78,
1588
+ "val/clipfrac_avg": 0.0,
1589
+ "val/num_eos_tokens": 2242,
1590
+ "val/ratio": 0.9996334910392761,
1591
+ "val/ratio_var": 2.991866267620935e-06
1592
+ },
1593
+ {
1594
+ "episode": 40960,
1595
+ "epoch": 0.17546264564770392,
1596
+ "eps": 6,
1597
+ "loss/policy_avg": 0.22769129276275635,
1598
+ "lr": 2.07421875e-06,
1599
+ "objective/entropy": -33.94448471069336,
1600
+ "objective/kl": 33.3216552734375,
1601
+ "objective/non_score_reward": -0.1666082888841629,
1602
+ "objective/rlhf_reward": 9.335718154907227,
1603
+ "objective/scores": 9.502326965332031,
1604
+ "policy/approxkl_avg": 0.2864202558994293,
1605
+ "policy/clipfrac_avg": 0.3515625,
1606
+ "policy/entropy_avg": 0.4178268313407898,
1607
+ "step": 79,
1608
+ "val/clipfrac_avg": 0.0,
1609
+ "val/num_eos_tokens": 2216,
1610
+ "val/ratio": 1.0000176429748535,
1611
+ "val/ratio_var": 2.602319455036195e-06
1612
+ },
1613
+ {
1614
+ "episode": 41472,
1615
+ "epoch": 0.17765592871830022,
1616
+ "eps": 6,
1617
+ "loss/policy_avg": 0.1986619234085083,
1618
+ "lr": 2.0625e-06,
1619
+ "objective/entropy": -36.479896545410156,
1620
+ "objective/kl": 33.37843322753906,
1621
+ "objective/non_score_reward": -0.1668921709060669,
1622
+ "objective/rlhf_reward": 9.689064025878906,
1623
+ "objective/scores": 9.855956077575684,
1624
+ "policy/approxkl_avg": 0.2828975319862366,
1625
+ "policy/clipfrac_avg": 0.369140625,
1626
+ "policy/entropy_avg": 0.39649999141693115,
1627
+ "step": 80,
1628
+ "val/clipfrac_avg": 0.0,
1629
+ "val/num_eos_tokens": 2166,
1630
+ "val/ratio": 1.0000145435333252,
1631
+ "val/ratio_var": 3.323544660815969e-06
1632
+ },
1633
+ {
1634
+ "episode": 41984,
1635
+ "epoch": 0.1798492117888965,
1636
+ "eps": 6,
1637
+ "loss/policy_avg": 0.36803221702575684,
1638
+ "lr": 2.05078125e-06,
1639
+ "objective/entropy": -34.6856803894043,
1640
+ "objective/kl": 32.34107971191406,
1641
+ "objective/non_score_reward": -0.1617053896188736,
1642
+ "objective/rlhf_reward": 9.39497184753418,
1643
+ "objective/scores": 9.556676864624023,
1644
+ "policy/approxkl_avg": 0.30969473719596863,
1645
+ "policy/clipfrac_avg": 0.337890625,
1646
+ "policy/entropy_avg": 0.40510374307632446,
1647
+ "step": 81,
1648
+ "val/clipfrac_avg": 0.0,
1649
+ "val/num_eos_tokens": 2162,
1650
+ "val/ratio": 0.9998123049736023,
1651
+ "val/ratio_var": 5.09162737216684e-06
1652
+ },
1653
+ {
1654
+ "episode": 42496,
1655
+ "epoch": 0.1820424948594928,
1656
+ "eps": 6,
1657
+ "loss/policy_avg": 0.16523130238056183,
1658
+ "lr": 2.0390625e-06,
1659
+ "objective/entropy": -36.272003173828125,
1660
+ "objective/kl": 34.003143310546875,
1661
+ "objective/non_score_reward": -0.17001570761203766,
1662
+ "objective/rlhf_reward": 9.380304336547852,
1663
+ "objective/scores": 9.55031967163086,
1664
+ "policy/approxkl_avg": 0.24730241298675537,
1665
+ "policy/clipfrac_avg": 0.345703125,
1666
+ "policy/entropy_avg": 0.40937918424606323,
1667
+ "step": 82,
1668
+ "val/clipfrac_avg": 0.0,
1669
+ "val/num_eos_tokens": 2186,
1670
+ "val/ratio": 0.999603807926178,
1671
+ "val/ratio_var": 2.4847686290740967e-06
1672
+ },
1673
+ {
1674
+ "episode": 43008,
1675
+ "epoch": 0.1842357779300891,
1676
+ "eps": 6,
1677
+ "loss/policy_avg": 0.2093135118484497,
1678
+ "lr": 2.02734375e-06,
1679
+ "objective/entropy": -35.804931640625,
1680
+ "objective/kl": 33.306732177734375,
1681
+ "objective/non_score_reward": -0.1665336787700653,
1682
+ "objective/rlhf_reward": 9.440333366394043,
1683
+ "objective/scores": 9.606866836547852,
1684
+ "policy/approxkl_avg": 0.22019661962985992,
1685
+ "policy/clipfrac_avg": 0.3359375,
1686
+ "policy/entropy_avg": 0.404262512922287,
1687
+ "step": 83,
1688
+ "val/clipfrac_avg": 0.0,
1689
+ "val/num_eos_tokens": 2208,
1690
+ "val/ratio": 1.000349521636963,
1691
+ "val/ratio_var": 2.718377345445333e-06
1692
+ },
1693
+ {
1694
+ "episode": 43520,
1695
+ "epoch": 0.1864290610006854,
1696
+ "eps": 6,
1697
+ "loss/policy_avg": 0.24589769542217255,
1698
+ "lr": 2.015625e-06,
1699
+ "objective/entropy": -35.195167541503906,
1700
+ "objective/kl": 32.886680603027344,
1701
+ "objective/non_score_reward": -0.16443338990211487,
1702
+ "objective/rlhf_reward": 9.737478256225586,
1703
+ "objective/scores": 9.901910781860352,
1704
+ "policy/approxkl_avg": 0.224727600812912,
1705
+ "policy/clipfrac_avg": 0.357421875,
1706
+ "policy/entropy_avg": 0.3911252021789551,
1707
+ "step": 84,
1708
+ "val/clipfrac_avg": 0.0,
1709
+ "val/num_eos_tokens": 2169,
1710
+ "val/ratio": 1.0001827478408813,
1711
+ "val/ratio_var": 2.8814699817303335e-06
1712
+ },
1713
+ {
1714
+ "episode": 44032,
1715
+ "epoch": 0.1886223440712817,
1716
+ "eps": 6,
1717
+ "loss/policy_avg": 0.2011430561542511,
1718
+ "lr": 2.00390625e-06,
1719
+ "objective/entropy": -36.270198822021484,
1720
+ "objective/kl": 32.86201095581055,
1721
+ "objective/non_score_reward": -0.1643100529909134,
1722
+ "objective/rlhf_reward": 9.618223190307617,
1723
+ "objective/scores": 9.782532691955566,
1724
+ "policy/approxkl_avg": 0.29179248213768005,
1725
+ "policy/clipfrac_avg": 0.33203125,
1726
+ "policy/entropy_avg": 0.3870548605918884,
1727
+ "step": 85,
1728
+ "val/clipfrac_avg": 0.0,
1729
+ "val/num_eos_tokens": 2167,
1730
+ "val/ratio": 1.0001106262207031,
1731
+ "val/ratio_var": 3.691868414534838e-06
1732
+ },
1733
+ {
1734
+ "episode": 44544,
1735
+ "epoch": 0.190815627141878,
1736
+ "eps": 6,
1737
+ "loss/policy_avg": 0.1226603090763092,
1738
+ "lr": 1.9921875e-06,
1739
+ "objective/entropy": -35.539337158203125,
1740
+ "objective/kl": 33.36035919189453,
1741
+ "objective/non_score_reward": -0.1668018102645874,
1742
+ "objective/rlhf_reward": 9.765243530273438,
1743
+ "objective/scores": 9.932045936584473,
1744
+ "policy/approxkl_avg": 0.23382121324539185,
1745
+ "policy/clipfrac_avg": 0.365234375,
1746
+ "policy/entropy_avg": 0.39480116963386536,
1747
+ "step": 86,
1748
+ "val/clipfrac_avg": 0.0,
1749
+ "val/num_eos_tokens": 2204,
1750
+ "val/ratio": 0.9996500015258789,
1751
+ "val/ratio_var": 3.449556515988661e-06
1752
+ },
1753
+ {
1754
+ "episode": 45056,
1755
+ "epoch": 0.1930089102124743,
1756
+ "eps": 6,
1757
+ "loss/policy_avg": 0.22705024480819702,
1758
+ "lr": 1.98046875e-06,
1759
+ "objective/entropy": -34.78813171386719,
1760
+ "objective/kl": 33.85824203491211,
1761
+ "objective/non_score_reward": -0.1692911982536316,
1762
+ "objective/rlhf_reward": 9.508092880249023,
1763
+ "objective/scores": 9.677383422851562,
1764
+ "policy/approxkl_avg": 0.2564248740673065,
1765
+ "policy/clipfrac_avg": 0.341796875,
1766
+ "policy/entropy_avg": 0.39244288206100464,
1767
+ "step": 87,
1768
+ "val/clipfrac_avg": 0.0,
1769
+ "val/num_eos_tokens": 2230,
1770
+ "val/ratio": 0.9998552799224854,
1771
+ "val/ratio_var": 2.9433140298351645e-06
1772
+ },
1773
+ {
1774
+ "episode": 45568,
1775
+ "epoch": 0.1952021932830706,
1776
+ "eps": 6,
1777
+ "loss/policy_avg": 0.25029683113098145,
1778
+ "lr": 1.96875e-06,
1779
+ "objective/entropy": -38.32817077636719,
1780
+ "objective/kl": 32.53593826293945,
1781
+ "objective/non_score_reward": -0.16267967224121094,
1782
+ "objective/rlhf_reward": 9.676530838012695,
1783
+ "objective/scores": 9.839210510253906,
1784
+ "policy/approxkl_avg": 0.24715110659599304,
1785
+ "policy/clipfrac_avg": 0.345703125,
1786
+ "policy/entropy_avg": 0.36855989694595337,
1787
+ "step": 88,
1788
+ "val/clipfrac_avg": 0.0,
1789
+ "val/num_eos_tokens": 2186,
1790
+ "val/ratio": 1.0001134872436523,
1791
+ "val/ratio_var": 3.477802692941623e-06
1792
+ },
1793
+ {
1794
+ "episode": 46080,
1795
+ "epoch": 0.1973954763536669,
1796
+ "eps": 6,
1797
+ "loss/policy_avg": 0.2676100730895996,
1798
+ "lr": 1.95703125e-06,
1799
+ "objective/entropy": -34.89455795288086,
1800
+ "objective/kl": 34.006202697753906,
1801
+ "objective/non_score_reward": -0.17003101110458374,
1802
+ "objective/rlhf_reward": 9.682475090026855,
1803
+ "objective/scores": 9.852506637573242,
1804
+ "policy/approxkl_avg": 0.2565373182296753,
1805
+ "policy/clipfrac_avg": 0.337890625,
1806
+ "policy/entropy_avg": 0.38856256008148193,
1807
+ "step": 89,
1808
+ "val/clipfrac_avg": 0.0,
1809
+ "val/num_eos_tokens": 2148,
1810
+ "val/ratio": 0.9999434947967529,
1811
+ "val/ratio_var": 3.6854912650596816e-06
1812
+ },
1813
+ {
1814
+ "episode": 46592,
1815
+ "epoch": 0.19958875942426318,
1816
+ "eps": 6,
1817
+ "loss/policy_avg": 0.17851054668426514,
1818
+ "lr": 1.9453125e-06,
1819
+ "objective/entropy": -37.629817962646484,
1820
+ "objective/kl": 33.07708740234375,
1821
+ "objective/non_score_reward": -0.1653854250907898,
1822
+ "objective/rlhf_reward": 9.640422821044922,
1823
+ "objective/scores": 9.805809020996094,
1824
+ "policy/approxkl_avg": 0.2017795443534851,
1825
+ "policy/clipfrac_avg": 0.34765625,
1826
+ "policy/entropy_avg": 0.37295690178871155,
1827
+ "step": 90,
1828
+ "val/clipfrac_avg": 0.0,
1829
+ "val/num_eos_tokens": 2112,
1830
+ "val/ratio": 1.0001370906829834,
1831
+ "val/ratio_var": 2.355475317017408e-06
1832
+ },
1833
+ {
1834
+ "episode": 47104,
1835
+ "epoch": 0.20178204249485948,
1836
+ "eps": 6,
1837
+ "loss/policy_avg": 0.1686045229434967,
1838
+ "lr": 1.93359375e-06,
1839
+ "objective/entropy": -38.061744689941406,
1840
+ "objective/kl": 32.129512786865234,
1841
+ "objective/non_score_reward": -0.16064755618572235,
1842
+ "objective/rlhf_reward": 9.790443420410156,
1843
+ "objective/scores": 9.951091766357422,
1844
+ "policy/approxkl_avg": 0.21093934774398804,
1845
+ "policy/clipfrac_avg": 0.36328125,
1846
+ "policy/entropy_avg": 0.3660760521888733,
1847
+ "step": 91,
1848
+ "val/clipfrac_avg": 0.0,
1849
+ "val/num_eos_tokens": 2086,
1850
+ "val/ratio": 0.9999697208404541,
1851
+ "val/ratio_var": 3.8051555293350248e-06
1852
+ },
1853
+ {
1854
+ "episode": 47616,
1855
+ "epoch": 0.20397532556545578,
1856
+ "eps": 6,
1857
+ "loss/policy_avg": 0.20496009290218353,
1858
+ "lr": 1.921875e-06,
1859
+ "objective/entropy": -38.42815399169922,
1860
+ "objective/kl": 33.15034484863281,
1861
+ "objective/non_score_reward": -0.16575172543525696,
1862
+ "objective/rlhf_reward": 9.645820617675781,
1863
+ "objective/scores": 9.811572074890137,
1864
+ "policy/approxkl_avg": 0.2742815613746643,
1865
+ "policy/clipfrac_avg": 0.353515625,
1866
+ "policy/entropy_avg": 0.35775893926620483,
1867
+ "step": 92,
1868
+ "val/clipfrac_avg": 0.0,
1869
+ "val/num_eos_tokens": 2077,
1870
+ "val/ratio": 1.0003482103347778,
1871
+ "val/ratio_var": 3.3726596484484617e-06
1872
+ },
1873
+ {
1874
+ "episode": 48128,
1875
+ "epoch": 0.2061686086360521,
1876
+ "eps": 6,
1877
+ "loss/policy_avg": 0.1944173276424408,
1878
+ "lr": 1.91015625e-06,
1879
+ "objective/entropy": -37.453712463378906,
1880
+ "objective/kl": 33.70136260986328,
1881
+ "objective/non_score_reward": -0.16850680112838745,
1882
+ "objective/rlhf_reward": 9.822844505310059,
1883
+ "objective/scores": 9.991350173950195,
1884
+ "policy/approxkl_avg": 0.264553964138031,
1885
+ "policy/clipfrac_avg": 0.337890625,
1886
+ "policy/entropy_avg": 0.3743836581707001,
1887
+ "step": 93,
1888
+ "val/clipfrac_avg": 0.0,
1889
+ "val/num_eos_tokens": 2082,
1890
+ "val/ratio": 0.9996594786643982,
1891
+ "val/ratio_var": 2.970580226246966e-06
1892
+ },
1893
+ {
1894
+ "episode": 48640,
1895
+ "epoch": 0.2083618917066484,
1896
+ "eps": 6,
1897
+ "loss/policy_avg": 0.20501361787319183,
1898
+ "lr": 1.8984375e-06,
1899
+ "objective/entropy": -36.34868621826172,
1900
+ "objective/kl": 33.67938995361328,
1901
+ "objective/non_score_reward": -0.1683969646692276,
1902
+ "objective/rlhf_reward": 9.771129608154297,
1903
+ "objective/scores": 9.939525604248047,
1904
+ "policy/approxkl_avg": 0.29675331711769104,
1905
+ "policy/clipfrac_avg": 0.31640625,
1906
+ "policy/entropy_avg": 0.35725563764572144,
1907
+ "step": 94,
1908
+ "val/clipfrac_avg": 0.0,
1909
+ "val/num_eos_tokens": 2083,
1910
+ "val/ratio": 1.0001224279403687,
1911
+ "val/ratio_var": 3.695943178172456e-06
1912
+ },
1913
+ {
1914
+ "episode": 49152,
1915
+ "epoch": 0.2105551747772447,
1916
+ "eps": 6,
1917
+ "loss/policy_avg": 0.19600501656532288,
1918
+ "lr": 1.8867187500000001e-06,
1919
+ "objective/entropy": -34.53193664550781,
1920
+ "objective/kl": 34.90612030029297,
1921
+ "objective/non_score_reward": -0.17453059554100037,
1922
+ "objective/rlhf_reward": 9.953956604003906,
1923
+ "objective/scores": 10.128486633300781,
1924
+ "policy/approxkl_avg": 0.21787747740745544,
1925
+ "policy/clipfrac_avg": 0.318359375,
1926
+ "policy/entropy_avg": 0.339312881231308,
1927
+ "step": 95,
1928
+ "val/clipfrac_avg": 0.0,
1929
+ "val/num_eos_tokens": 1995,
1930
+ "val/ratio": 0.999893844127655,
1931
+ "val/ratio_var": 2.4975431642815238e-06
1932
+ },
1933
+ {
1934
+ "episode": 49664,
1935
+ "epoch": 0.212748457847841,
1936
+ "eps": 6,
1937
+ "loss/policy_avg": 0.22088220715522766,
1938
+ "lr": 1.875e-06,
1939
+ "objective/entropy": -34.256370544433594,
1940
+ "objective/kl": 35.238216400146484,
1941
+ "objective/non_score_reward": -0.1761910766363144,
1942
+ "objective/rlhf_reward": 9.564779281616211,
1943
+ "objective/scores": 9.74096965789795,
1944
+ "policy/approxkl_avg": 0.245886892080307,
1945
+ "policy/clipfrac_avg": 0.35546875,
1946
+ "policy/entropy_avg": 0.35175642371177673,
1947
+ "step": 96,
1948
+ "val/clipfrac_avg": 0.0,
1949
+ "val/num_eos_tokens": 1972,
1950
+ "val/ratio": 0.9998222589492798,
1951
+ "val/ratio_var": 3.3520850593049545e-06
1952
+ },
1953
+ {
1954
+ "episode": 50176,
1955
+ "epoch": 0.2149417409184373,
1956
+ "eps": 6,
1957
+ "loss/policy_avg": 0.1739550083875656,
1958
+ "lr": 1.86328125e-06,
1959
+ "objective/entropy": -36.02185821533203,
1960
+ "objective/kl": 35.261474609375,
1961
+ "objective/non_score_reward": -0.17630735039710999,
1962
+ "objective/rlhf_reward": 9.913810729980469,
1963
+ "objective/scores": 10.090118408203125,
1964
+ "policy/approxkl_avg": 0.19460904598236084,
1965
+ "policy/clipfrac_avg": 0.337890625,
1966
+ "policy/entropy_avg": 0.3500151038169861,
1967
+ "step": 97,
1968
+ "val/clipfrac_avg": 0.0,
1969
+ "val/num_eos_tokens": 2014,
1970
+ "val/ratio": 1.0000877380371094,
1971
+ "val/ratio_var": 2.4969581318146084e-06
1972
+ },
1973
+ {
1974
+ "episode": 50688,
1975
+ "epoch": 0.2171350239890336,
1976
+ "eps": 6,
1977
+ "loss/policy_avg": 0.27473074197769165,
1978
+ "lr": 1.8515625000000001e-06,
1979
+ "objective/entropy": -35.280155181884766,
1980
+ "objective/kl": 36.09398651123047,
1981
+ "objective/non_score_reward": -0.18046993017196655,
1982
+ "objective/rlhf_reward": 9.731412887573242,
1983
+ "objective/scores": 9.911883354187012,
1984
+ "policy/approxkl_avg": 0.23806479573249817,
1985
+ "policy/clipfrac_avg": 0.34765625,
1986
+ "policy/entropy_avg": 0.36351966857910156,
1987
+ "step": 98,
1988
+ "val/clipfrac_avg": 0.0,
1989
+ "val/num_eos_tokens": 2009,
1990
+ "val/ratio": 1.0001752376556396,
1991
+ "val/ratio_var": 4.085913587914547e-06
1992
+ },
1993
+ {
1994
+ "episode": 51200,
1995
+ "epoch": 0.21932830705962988,
1996
+ "eps": 6,
1997
+ "loss/policy_avg": 0.28736796975135803,
1998
+ "lr": 1.83984375e-06,
1999
+ "objective/entropy": -35.07633972167969,
2000
+ "objective/kl": 35.50556182861328,
2001
+ "objective/non_score_reward": -0.1775278002023697,
2002
+ "objective/rlhf_reward": 9.971885681152344,
2003
+ "objective/scores": 10.1494140625,
2004
+ "policy/approxkl_avg": 0.22533981502056122,
2005
+ "policy/clipfrac_avg": 0.345703125,
2006
+ "policy/entropy_avg": 0.33290359377861023,
2007
+ "step": 99,
2008
+ "val/clipfrac_avg": 0.0,
2009
+ "val/num_eos_tokens": 1977,
2010
+ "val/ratio": 0.9998661279678345,
2011
+ "val/ratio_var": 5.175881597097032e-06
2012
+ },
2013
+ {
2014
+ "episode": 51712,
2015
+ "epoch": 0.22152159013022618,
2016
+ "eps": 6,
2017
+ "loss/policy_avg": 0.24332170188426971,
2018
+ "lr": 1.828125e-06,
2019
+ "objective/entropy": -31.962793350219727,
2020
+ "objective/kl": 36.79883575439453,
2021
+ "objective/non_score_reward": -0.18399417400360107,
2022
+ "objective/rlhf_reward": 9.824310302734375,
2023
+ "objective/scores": 10.008304595947266,
2024
+ "policy/approxkl_avg": 0.2577008008956909,
2025
+ "policy/clipfrac_avg": 0.40234375,
2026
+ "policy/entropy_avg": 0.3309575021266937,
2027
+ "step": 100,
2028
+ "val/clipfrac_avg": 0.0,
2029
+ "val/num_eos_tokens": 1903,
2030
+ "val/ratio": 1.000428318977356,
2031
+ "val/ratio_var": 2.6899542717728764e-06
2032
+ },
2033
+ {
2034
+ "episode": 52224,
2035
+ "epoch": 0.22371487320082248,
2036
+ "eps": 6,
2037
+ "loss/policy_avg": 0.17923694849014282,
2038
+ "lr": 1.81640625e-06,
2039
+ "objective/entropy": -34.664024353027344,
2040
+ "objective/kl": 35.354034423828125,
2041
+ "objective/non_score_reward": -0.1767701804637909,
2042
+ "objective/rlhf_reward": 10.139245986938477,
2043
+ "objective/scores": 10.316017150878906,
2044
+ "policy/approxkl_avg": 0.19910496473312378,
2045
+ "policy/clipfrac_avg": 0.31640625,
2046
+ "policy/entropy_avg": 0.3107799291610718,
2047
+ "step": 101,
2048
+ "val/clipfrac_avg": 0.0,
2049
+ "val/num_eos_tokens": 1891,
2050
+ "val/ratio": 0.9996460676193237,
2051
+ "val/ratio_var": 3.313231900392566e-06
2052
+ },
2053
+ {
2054
+ "episode": 52736,
2055
+ "epoch": 0.22590815627141878,
2056
+ "eps": 6,
2057
+ "loss/policy_avg": 0.17553430795669556,
2058
+ "lr": 1.8046875e-06,
2059
+ "objective/entropy": -34.35253143310547,
2060
+ "objective/kl": 35.79362869262695,
2061
+ "objective/non_score_reward": -0.1789681315422058,
2062
+ "objective/rlhf_reward": 10.026729583740234,
2063
+ "objective/scores": 10.205698013305664,
2064
+ "policy/approxkl_avg": 0.21303945779800415,
2065
+ "policy/clipfrac_avg": 0.328125,
2066
+ "policy/entropy_avg": 0.3087007999420166,
2067
+ "step": 102,
2068
+ "val/clipfrac_avg": 0.0,
2069
+ "val/num_eos_tokens": 1905,
2070
+ "val/ratio": 0.9998683929443359,
2071
+ "val/ratio_var": 3.267524562033941e-06
2072
+ },
2073
+ {
2074
+ "episode": 53248,
2075
+ "epoch": 0.22810143934201507,
2076
+ "eps": 6,
2077
+ "loss/policy_avg": 0.20496255159378052,
2078
+ "lr": 1.79296875e-06,
2079
+ "objective/entropy": -34.044639587402344,
2080
+ "objective/kl": 35.621826171875,
2081
+ "objective/non_score_reward": -0.17810912430286407,
2082
+ "objective/rlhf_reward": 10.048910140991211,
2083
+ "objective/scores": 10.227019309997559,
2084
+ "policy/approxkl_avg": 0.2181515246629715,
2085
+ "policy/clipfrac_avg": 0.357421875,
2086
+ "policy/entropy_avg": 0.3069969415664673,
2087
+ "step": 103,
2088
+ "val/clipfrac_avg": 0.0,
2089
+ "val/num_eos_tokens": 1940,
2090
+ "val/ratio": 0.9999847412109375,
2091
+ "val/ratio_var": 3.6619542242988246e-06
2092
+ },
2093
+ {
2094
+ "episode": 53760,
2095
+ "epoch": 0.23029472241261137,
2096
+ "eps": 6,
2097
+ "loss/policy_avg": 0.22224432229995728,
2098
+ "lr": 1.78125e-06,
2099
+ "objective/entropy": -32.43903732299805,
2100
+ "objective/kl": 36.86625289916992,
2101
+ "objective/non_score_reward": -0.1843312531709671,
2102
+ "objective/rlhf_reward": 9.84052562713623,
2103
+ "objective/scores": 10.024856567382812,
2104
+ "policy/approxkl_avg": 0.1902885138988495,
2105
+ "policy/clipfrac_avg": 0.328125,
2106
+ "policy/entropy_avg": 0.30074501037597656,
2107
+ "step": 104,
2108
+ "val/clipfrac_avg": 0.0,
2109
+ "val/num_eos_tokens": 1854,
2110
+ "val/ratio": 1.0004634857177734,
2111
+ "val/ratio_var": 2.30164823733503e-06
2112
+ },
2113
+ {
2114
+ "episode": 54272,
2115
+ "epoch": 0.23248800548320767,
2116
+ "eps": 6,
2117
+ "loss/policy_avg": 0.13901078701019287,
2118
+ "lr": 1.76953125e-06,
2119
+ "objective/entropy": -33.599708557128906,
2120
+ "objective/kl": 35.801998138427734,
2121
+ "objective/non_score_reward": -0.17900997400283813,
2122
+ "objective/rlhf_reward": 9.889688491821289,
2123
+ "objective/scores": 10.06869888305664,
2124
+ "policy/approxkl_avg": 0.16544947028160095,
2125
+ "policy/clipfrac_avg": 0.337890625,
2126
+ "policy/entropy_avg": 0.30036184191703796,
2127
+ "step": 105,
2128
+ "val/clipfrac_avg": 0.0,
2129
+ "val/num_eos_tokens": 1859,
2130
+ "val/ratio": 1.0002105236053467,
2131
+ "val/ratio_var": 2.6306556719646323e-06
2132
+ },
2133
+ {
2134
+ "episode": 54784,
2135
+ "epoch": 0.23468128855380396,
2136
+ "eps": 6,
2137
+ "loss/policy_avg": 0.14493289589881897,
2138
+ "lr": 1.7578125e-06,
2139
+ "objective/entropy": -33.55315017700195,
2140
+ "objective/kl": 35.825531005859375,
2141
+ "objective/non_score_reward": -0.17912766337394714,
2142
+ "objective/rlhf_reward": 9.871797561645508,
2143
+ "objective/scores": 10.050926208496094,
2144
+ "policy/approxkl_avg": 0.1833547204732895,
2145
+ "policy/clipfrac_avg": 0.328125,
2146
+ "policy/entropy_avg": 0.2901462912559509,
2147
+ "step": 106,
2148
+ "val/clipfrac_avg": 0.0,
2149
+ "val/num_eos_tokens": 1835,
2150
+ "val/ratio": 0.9998902678489685,
2151
+ "val/ratio_var": 2.489499593139044e-06
2152
+ },
2153
+ {
2154
+ "episode": 55296,
2155
+ "epoch": 0.2368745716244003,
2156
+ "eps": 6,
2157
+ "loss/policy_avg": 0.17336603999137878,
2158
+ "lr": 1.74609375e-06,
2159
+ "objective/entropy": -34.462337493896484,
2160
+ "objective/kl": 37.12619400024414,
2161
+ "objective/non_score_reward": -0.18563096225261688,
2162
+ "objective/rlhf_reward": 10.155291557312012,
2163
+ "objective/scores": 10.340921401977539,
2164
+ "policy/approxkl_avg": 0.18350008130073547,
2165
+ "policy/clipfrac_avg": 0.3359375,
2166
+ "policy/entropy_avg": 0.29074928164482117,
2167
+ "step": 107,
2168
+ "val/clipfrac_avg": 0.0,
2169
+ "val/num_eos_tokens": 1954,
2170
+ "val/ratio": 1.0000966787338257,
2171
+ "val/ratio_var": 2.0360935195640195e-06
2172
+ },
2173
+ {
2174
+ "episode": 55808,
2175
+ "epoch": 0.23906785469499658,
2176
+ "eps": 6,
2177
+ "loss/policy_avg": 0.20065046846866608,
2178
+ "lr": 1.734375e-06,
2179
+ "objective/entropy": -35.58911895751953,
2180
+ "objective/kl": 35.46983337402344,
2181
+ "objective/non_score_reward": -0.17734915018081665,
2182
+ "objective/rlhf_reward": 9.807424545288086,
2183
+ "objective/scores": 9.984773635864258,
2184
+ "policy/approxkl_avg": 0.21259769797325134,
2185
+ "policy/clipfrac_avg": 0.3359375,
2186
+ "policy/entropy_avg": 0.2804480195045471,
2187
+ "step": 108,
2188
+ "val/clipfrac_avg": 0.0,
2189
+ "val/num_eos_tokens": 1868,
2190
+ "val/ratio": 0.9998682737350464,
2191
+ "val/ratio_var": 4.149359028815525e-06
2192
+ },
2193
+ {
2194
+ "episode": 56320,
2195
+ "epoch": 0.24126113776559288,
2196
+ "eps": 6,
2197
+ "loss/policy_avg": 0.31327635049819946,
2198
+ "lr": 1.72265625e-06,
2199
+ "objective/entropy": -35.60870361328125,
2200
+ "objective/kl": 34.600460052490234,
2201
+ "objective/non_score_reward": -0.17300228774547577,
2202
+ "objective/rlhf_reward": 9.828847885131836,
2203
+ "objective/scores": 10.001850128173828,
2204
+ "policy/approxkl_avg": 0.2347819209098816,
2205
+ "policy/clipfrac_avg": 0.326171875,
2206
+ "policy/entropy_avg": 0.2796494960784912,
2207
+ "step": 109,
2208
+ "val/clipfrac_avg": 0.0,
2209
+ "val/num_eos_tokens": 1903,
2210
+ "val/ratio": 1.000025749206543,
2211
+ "val/ratio_var": 3.987162926932797e-06
2212
+ },
2213
+ {
2214
+ "episode": 56832,
2215
+ "epoch": 0.24345442083618918,
2216
+ "eps": 6,
2217
+ "loss/policy_avg": 0.19177474081516266,
2218
+ "lr": 1.7109375e-06,
2219
+ "objective/entropy": -35.206573486328125,
2220
+ "objective/kl": 35.755706787109375,
2221
+ "objective/non_score_reward": -0.1787785291671753,
2222
+ "objective/rlhf_reward": 10.03399658203125,
2223
+ "objective/scores": 10.212775230407715,
2224
+ "policy/approxkl_avg": 0.2251434177160263,
2225
+ "policy/clipfrac_avg": 0.388671875,
2226
+ "policy/entropy_avg": 0.2768189311027527,
2227
+ "step": 110,
2228
+ "val/clipfrac_avg": 0.0,
2229
+ "val/num_eos_tokens": 1854,
2230
+ "val/ratio": 0.9997500777244568,
2231
+ "val/ratio_var": 2.821056341417716e-06
2232
+ },
2233
+ {
2234
+ "episode": 57344,
2235
+ "epoch": 0.24564770390678548,
2236
+ "eps": 6,
2237
+ "loss/policy_avg": 0.17819970846176147,
2238
+ "lr": 1.69921875e-06,
2239
+ "objective/entropy": -35.49714279174805,
2240
+ "objective/kl": 35.21353530883789,
2241
+ "objective/non_score_reward": -0.17606768012046814,
2242
+ "objective/rlhf_reward": 9.934215545654297,
2243
+ "objective/scores": 10.110282897949219,
2244
+ "policy/approxkl_avg": 0.2090197205543518,
2245
+ "policy/clipfrac_avg": 0.30859375,
2246
+ "policy/entropy_avg": 0.2791983485221863,
2247
+ "step": 111,
2248
+ "val/clipfrac_avg": 0.0,
2249
+ "val/num_eos_tokens": 1825,
2250
+ "val/ratio": 1.0002068281173706,
2251
+ "val/ratio_var": 3.191226596754859e-06
2252
+ },
2253
+ {
2254
+ "episode": 57856,
2255
+ "epoch": 0.24784098697738177,
2256
+ "eps": 6,
2257
+ "loss/policy_avg": 0.13477346301078796,
2258
+ "lr": 1.6875e-06,
2259
+ "objective/entropy": -35.38134002685547,
2260
+ "objective/kl": 35.48783874511719,
2261
+ "objective/non_score_reward": -0.17743918299674988,
2262
+ "objective/rlhf_reward": 10.00422191619873,
2263
+ "objective/scores": 10.181660652160645,
2264
+ "policy/approxkl_avg": 0.21763411164283752,
2265
+ "policy/clipfrac_avg": 0.37109375,
2266
+ "policy/entropy_avg": 0.27607789635658264,
2267
+ "step": 112,
2268
+ "val/clipfrac_avg": 0.0,
2269
+ "val/num_eos_tokens": 1811,
2270
+ "val/ratio": 1.0001035928726196,
2271
+ "val/ratio_var": 3.5984410260425648e-06
2272
+ },
2273
+ {
2274
+ "episode": 58368,
2275
+ "epoch": 0.25003427004797807,
2276
+ "eps": 6,
2277
+ "loss/policy_avg": 0.16193845868110657,
2278
+ "lr": 1.67578125e-06,
2279
+ "objective/entropy": -35.136634826660156,
2280
+ "objective/kl": 35.53947830200195,
2281
+ "objective/non_score_reward": -0.17769737541675568,
2282
+ "objective/rlhf_reward": 10.041696548461914,
2283
+ "objective/scores": 10.219392776489258,
2284
+ "policy/approxkl_avg": 0.18851657211780548,
2285
+ "policy/clipfrac_avg": 0.322265625,
2286
+ "policy/entropy_avg": 0.2754863500595093,
2287
+ "step": 113,
2288
+ "val/clipfrac_avg": 0.0,
2289
+ "val/num_eos_tokens": 1822,
2290
+ "val/ratio": 0.9998536109924316,
2291
+ "val/ratio_var": 2.3457650968339294e-06
2292
+ },
2293
+ {
2294
+ "episode": 58880,
2295
+ "epoch": 0.25222755311857437,
2296
+ "eps": 6,
2297
+ "loss/policy_avg": 0.2662353217601776,
2298
+ "lr": 1.6640625e-06,
2299
+ "objective/entropy": -36.19813537597656,
2300
+ "objective/kl": 35.548973083496094,
2301
+ "objective/non_score_reward": -0.17774485051631927,
2302
+ "objective/rlhf_reward": 9.807806015014648,
2303
+ "objective/scores": 9.985549926757812,
2304
+ "policy/approxkl_avg": 0.22413387894630432,
2305
+ "policy/clipfrac_avg": 0.318359375,
2306
+ "policy/entropy_avg": 0.28221410512924194,
2307
+ "step": 114,
2308
+ "val/clipfrac_avg": 0.0,
2309
+ "val/num_eos_tokens": 1945,
2310
+ "val/ratio": 1.0000545978546143,
2311
+ "val/ratio_var": 3.614477236624225e-06
2312
+ },
2313
+ {
2314
+ "episode": 59392,
2315
+ "epoch": 0.25442083618917066,
2316
+ "eps": 6,
2317
+ "loss/policy_avg": 0.1269383728504181,
2318
+ "lr": 1.6523437500000001e-06,
2319
+ "objective/entropy": -37.98196792602539,
2320
+ "objective/kl": 34.821964263916016,
2321
+ "objective/non_score_reward": -0.1741098165512085,
2322
+ "objective/rlhf_reward": 9.85723876953125,
2323
+ "objective/scores": 10.031347274780273,
2324
+ "policy/approxkl_avg": 0.21144835650920868,
2325
+ "policy/clipfrac_avg": 0.31640625,
2326
+ "policy/entropy_avg": 0.27860069274902344,
2327
+ "step": 115,
2328
+ "val/clipfrac_avg": 0.0,
2329
+ "val/num_eos_tokens": 1918,
2330
+ "val/ratio": 1.0000715255737305,
2331
+ "val/ratio_var": 4.044930847157957e-06
2332
+ },
2333
+ {
2334
+ "episode": 59904,
2335
+ "epoch": 0.25661411925976696,
2336
+ "eps": 6,
2337
+ "loss/policy_avg": 0.18478526175022125,
2338
+ "lr": 1.640625e-06,
2339
+ "objective/entropy": -39.03168869018555,
2340
+ "objective/kl": 34.09989929199219,
2341
+ "objective/non_score_reward": -0.1704995036125183,
2342
+ "objective/rlhf_reward": 10.072548866271973,
2343
+ "objective/scores": 10.243047714233398,
2344
+ "policy/approxkl_avg": 0.18763652443885803,
2345
+ "policy/clipfrac_avg": 0.32421875,
2346
+ "policy/entropy_avg": 0.276102751493454,
2347
+ "step": 116,
2348
+ "val/clipfrac_avg": 0.0,
2349
+ "val/num_eos_tokens": 1964,
2350
+ "val/ratio": 1.0003482103347778,
2351
+ "val/ratio_var": 3.092050519626355e-06
2352
+ },
2353
+ {
2354
+ "episode": 60416,
2355
+ "epoch": 0.25880740233036326,
2356
+ "eps": 6,
2357
+ "loss/policy_avg": 0.2445678561925888,
2358
+ "lr": 1.62890625e-06,
2359
+ "objective/entropy": -37.85968017578125,
2360
+ "objective/kl": 34.59711456298828,
2361
+ "objective/non_score_reward": -0.17298556864261627,
2362
+ "objective/rlhf_reward": 9.838810920715332,
2363
+ "objective/scores": 10.011796951293945,
2364
+ "policy/approxkl_avg": 0.2142867147922516,
2365
+ "policy/clipfrac_avg": 0.287109375,
2366
+ "policy/entropy_avg": 0.2769862413406372,
2367
+ "step": 117,
2368
+ "val/clipfrac_avg": 0.0,
2369
+ "val/num_eos_tokens": 1913,
2370
+ "val/ratio": 1.0000524520874023,
2371
+ "val/ratio_var": 2.518341943869018e-06
2372
+ },
2373
+ {
2374
+ "episode": 60928,
2375
+ "epoch": 0.26100068540095955,
2376
+ "eps": 6,
2377
+ "loss/policy_avg": 0.14503635466098785,
2378
+ "lr": 1.6171875000000001e-06,
2379
+ "objective/entropy": -42.66570281982422,
2380
+ "objective/kl": 33.559688568115234,
2381
+ "objective/non_score_reward": -0.1677984595298767,
2382
+ "objective/rlhf_reward": 10.2135591506958,
2383
+ "objective/scores": 10.381357192993164,
2384
+ "policy/approxkl_avg": 0.17557722330093384,
2385
+ "policy/clipfrac_avg": 0.267578125,
2386
+ "policy/entropy_avg": 0.27216678857803345,
2387
+ "step": 118,
2388
+ "val/clipfrac_avg": 0.0,
2389
+ "val/num_eos_tokens": 1978,
2390
+ "val/ratio": 0.9997793436050415,
2391
+ "val/ratio_var": 2.1884602574573364e-06
2392
+ },
2393
+ {
2394
+ "episode": 61440,
2395
+ "epoch": 0.26319396847155585,
2396
+ "eps": 6,
2397
+ "loss/policy_avg": 0.12860137224197388,
2398
+ "lr": 1.60546875e-06,
2399
+ "objective/entropy": -40.882850646972656,
2400
+ "objective/kl": 35.01697540283203,
2401
+ "objective/non_score_reward": -0.17508485913276672,
2402
+ "objective/rlhf_reward": 9.812532424926758,
2403
+ "objective/scores": 9.987617492675781,
2404
+ "policy/approxkl_avg": 0.21601755917072296,
2405
+ "policy/clipfrac_avg": 0.333984375,
2406
+ "policy/entropy_avg": 0.2853110432624817,
2407
+ "step": 119,
2408
+ "val/clipfrac_avg": 0.0,
2409
+ "val/num_eos_tokens": 1991,
2410
+ "val/ratio": 1.0000758171081543,
2411
+ "val/ratio_var": 3.1630045214114944e-06
2412
+ },
2413
+ {
2414
+ "episode": 61952,
2415
+ "epoch": 0.26538725154215215,
2416
+ "eps": 6,
2417
+ "loss/policy_avg": 0.13918112218379974,
2418
+ "lr": 1.59375e-06,
2419
+ "objective/entropy": -42.673091888427734,
2420
+ "objective/kl": 33.7802734375,
2421
+ "objective/non_score_reward": -0.16890135407447815,
2422
+ "objective/rlhf_reward": 9.991598129272461,
2423
+ "objective/scores": 10.160500526428223,
2424
+ "policy/approxkl_avg": 0.2096889615058899,
2425
+ "policy/clipfrac_avg": 0.3125,
2426
+ "policy/entropy_avg": 0.2698075771331787,
2427
+ "step": 120,
2428
+ "val/clipfrac_avg": 0.0,
2429
+ "val/num_eos_tokens": 1962,
2430
+ "val/ratio": 0.9996597766876221,
2431
+ "val/ratio_var": 2.5640338208177127e-06
2432
+ },
2433
+ {
2434
+ "episode": 62464,
2435
+ "epoch": 0.26758053461274844,
2436
+ "eps": 6,
2437
+ "loss/policy_avg": 0.14262011647224426,
2438
+ "lr": 1.5820312500000001e-06,
2439
+ "objective/entropy": -40.981964111328125,
2440
+ "objective/kl": 33.93827819824219,
2441
+ "objective/non_score_reward": -0.16969136893749237,
2442
+ "objective/rlhf_reward": 9.967924118041992,
2443
+ "objective/scores": 10.137615203857422,
2444
+ "policy/approxkl_avg": 0.17477768659591675,
2445
+ "policy/clipfrac_avg": 0.326171875,
2446
+ "policy/entropy_avg": 0.27177703380584717,
2447
+ "step": 121,
2448
+ "val/clipfrac_avg": 0.0,
2449
+ "val/num_eos_tokens": 1948,
2450
+ "val/ratio": 1.0003581047058105,
2451
+ "val/ratio_var": 2.333908696527942e-06
2452
+ },
2453
+ {
2454
+ "episode": 62976,
2455
+ "epoch": 0.26977381768334474,
2456
+ "eps": 6,
2457
+ "loss/policy_avg": 0.1601879894733429,
2458
+ "lr": 1.5703125e-06,
2459
+ "objective/entropy": -39.27589797973633,
2460
+ "objective/kl": 34.60676574707031,
2461
+ "objective/non_score_reward": -0.17303383350372314,
2462
+ "objective/rlhf_reward": 9.912843704223633,
2463
+ "objective/scores": 10.085877418518066,
2464
+ "policy/approxkl_avg": 0.20070144534111023,
2465
+ "policy/clipfrac_avg": 0.337890625,
2466
+ "policy/entropy_avg": 0.2859034836292267,
2467
+ "step": 122,
2468
+ "val/clipfrac_avg": 0.0,
2469
+ "val/num_eos_tokens": 1959,
2470
+ "val/ratio": 0.9996767044067383,
2471
+ "val/ratio_var": 2.508757233954384e-06
2472
+ },
2473
+ {
2474
+ "episode": 63488,
2475
+ "epoch": 0.27196710075394104,
2476
+ "eps": 6,
2477
+ "loss/policy_avg": 0.15391629934310913,
2478
+ "lr": 1.55859375e-06,
2479
+ "objective/entropy": -38.832088470458984,
2480
+ "objective/kl": 35.0777473449707,
2481
+ "objective/non_score_reward": -0.17538872361183167,
2482
+ "objective/rlhf_reward": 9.996062278747559,
2483
+ "objective/scores": 10.1714506149292,
2484
+ "policy/approxkl_avg": 0.22946523129940033,
2485
+ "policy/clipfrac_avg": 0.326171875,
2486
+ "policy/entropy_avg": 0.27930349111557007,
2487
+ "step": 123,
2488
+ "val/clipfrac_avg": 0.0,
2489
+ "val/num_eos_tokens": 1959,
2490
+ "val/ratio": 0.9997979402542114,
2491
+ "val/ratio_var": 3.5869422845280496e-06
2492
+ },
2493
+ {
2494
+ "episode": 64000,
2495
+ "epoch": 0.27416038382453733,
2496
+ "eps": 6,
2497
+ "loss/policy_avg": 0.16380949318408966,
2498
+ "lr": 1.5468750000000001e-06,
2499
+ "objective/entropy": -39.75455856323242,
2500
+ "objective/kl": 35.5320930480957,
2501
+ "objective/non_score_reward": -0.17766046524047852,
2502
+ "objective/rlhf_reward": 9.862564086914062,
2503
+ "objective/scores": 10.040224075317383,
2504
+ "policy/approxkl_avg": 0.1650054007768631,
2505
+ "policy/clipfrac_avg": 0.318359375,
2506
+ "policy/entropy_avg": 0.28795722126960754,
2507
+ "step": 124,
2508
+ "val/clipfrac_avg": 0.0,
2509
+ "val/num_eos_tokens": 1916,
2510
+ "val/ratio": 0.9999822378158569,
2511
+ "val/ratio_var": 2.5005276711453917e-06
2512
+ },
2513
+ {
2514
+ "episode": 64512,
2515
+ "epoch": 0.27635366689513363,
2516
+ "eps": 6,
2517
+ "loss/policy_avg": 0.14548206329345703,
2518
+ "lr": 1.53515625e-06,
2519
+ "objective/entropy": -40.37278747558594,
2520
+ "objective/kl": 34.826271057128906,
2521
+ "objective/non_score_reward": -0.1741313487291336,
2522
+ "objective/rlhf_reward": 10.072320938110352,
2523
+ "objective/scores": 10.246453285217285,
2524
+ "policy/approxkl_avg": 0.1895635575056076,
2525
+ "policy/clipfrac_avg": 0.328125,
2526
+ "policy/entropy_avg": 0.28454479575157166,
2527
+ "step": 125,
2528
+ "val/clipfrac_avg": 0.0,
2529
+ "val/num_eos_tokens": 1992,
2530
+ "val/ratio": 0.999925971031189,
2531
+ "val/ratio_var": 2.291703822265845e-06
2532
+ },
2533
+ {
2534
+ "episode": 65024,
2535
+ "epoch": 0.27854694996572993,
2536
+ "eps": 6,
2537
+ "loss/policy_avg": 0.17899231612682343,
2538
+ "lr": 1.5234375e-06,
2539
+ "objective/entropy": -39.21260070800781,
2540
+ "objective/kl": 36.2283935546875,
2541
+ "objective/non_score_reward": -0.18114197254180908,
2542
+ "objective/rlhf_reward": 9.876224517822266,
2543
+ "objective/scores": 10.057367324829102,
2544
+ "policy/approxkl_avg": 0.17429962754249573,
2545
+ "policy/clipfrac_avg": 0.328125,
2546
+ "policy/entropy_avg": 0.27740728855133057,
2547
+ "step": 126,
2548
+ "val/clipfrac_avg": 0.0,
2549
+ "val/num_eos_tokens": 1933,
2550
+ "val/ratio": 1.000385046005249,
2551
+ "val/ratio_var": 2.276981604154571e-06
2552
+ },
2553
+ {
2554
+ "episode": 65536,
2555
+ "epoch": 0.2807402330363262,
2556
+ "eps": 6,
2557
+ "loss/policy_avg": 0.2138238549232483,
2558
+ "lr": 1.5117187500000001e-06,
2559
+ "objective/entropy": -40.66987991333008,
2560
+ "objective/kl": 34.04345703125,
2561
+ "objective/non_score_reward": -0.17021729052066803,
2562
+ "objective/rlhf_reward": 10.10722541809082,
2563
+ "objective/scores": 10.277442932128906,
2564
+ "policy/approxkl_avg": 0.184827983379364,
2565
+ "policy/clipfrac_avg": 0.306640625,
2566
+ "policy/entropy_avg": 0.2644522190093994,
2567
+ "step": 127,
2568
+ "val/clipfrac_avg": 0.0,
2569
+ "val/num_eos_tokens": 1960,
2570
+ "val/ratio": 0.9999203681945801,
2571
+ "val/ratio_var": 2.535048452045885e-06
2572
+ }
2573
+ ],
2574
+ "logging_steps": 1,
2575
+ "max_steps": 128,
2576
+ "num_input_tokens_seen": 0,
2577
+ "num_train_epochs": 1.122960932145305,
2578
+ "save_steps": 26,
2579
+ "stateful_callbacks": {
2580
+ "TrainerControl": {
2581
+ "args": {
2582
+ "should_epoch_stop": false,
2583
+ "should_evaluate": false,
2584
+ "should_log": true,
2585
+ "should_save": true,
2586
+ "should_training_stop": true
2587
+ },
2588
+ "attributes": {}
2589
+ }
2590
+ },
2591
+ "total_flos": 0,
2592
+ "train_batch_size": null,
2593
+ "trial_name": null,
2594
+ "trial_params": null
2595
+ }
checkpoint-128/training_args.bin ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:30edc5fdb14278c841efedd8fdc4865c8766890f13483d7f2957be7405e19d4d
3
+ size 6584