Muqeeth commited on
Commit
38dfe44
·
verified ·
1 Parent(s): df8f866

Training in progress, step 138, checkpoint

Browse files
checkpoint-138/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-138/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-138/model.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:8a0607dda3de045830c01874d13a296ceb9e268007e5dd310d150450772b6112
3
+ size 1621370224
checkpoint-138/optimizer.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:dc0ce270abd6452fe229506ffcbab5f2e59792c3ffc88e3bfa76096de7424a88
3
+ size 3242916026
checkpoint-138/rng_state.pth ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:7bf213abba835be0154fc4d2259de41adf8fad01dced3ee2a9f2a0debbf48179
3
+ size 14180
checkpoint-138/scheduler.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:82b05dc98c33d0c2ca165da5b5356ccf303fa916fb209c460462e7715b3632e0
3
+ size 1064
checkpoint-138/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-138/tokenizer.json ADDED
The diff for this file is too large to render. See raw diff
 
checkpoint-138/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-138/trainer_state.json ADDED
@@ -0,0 +1,2795 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "best_global_step": null,
3
+ "best_metric": null,
4
+ "best_model_checkpoint": null,
5
+ "episode": 70656,
6
+ "epoch": 0.30267306374228925,
7
+ "eval_steps": 500,
8
+ "global_step": 138,
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": 0.15268513560295105,
18
+ "lr": 3e-06,
19
+ "objective/entropy": -45.407432556152344,
20
+ "objective/kl": 0.3935524821281433,
21
+ "objective/non_score_reward": 0.0,
22
+ "objective/rlhf_reward": 4.679054260253906,
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": 1.6687282595739816e-06
32
+ },
33
+ {
34
+ "episode": 1024,
35
+ "epoch": 0.004386566141192597,
36
+ "eps": 6,
37
+ "loss/policy_avg": 0.10589599609375,
38
+ "lr": 2.9882812500000002e-06,
39
+ "objective/entropy": -51.535125732421875,
40
+ "objective/kl": 0.6498137712478638,
41
+ "objective/non_score_reward": 0.0,
42
+ "objective/rlhf_reward": 4.541956901550293,
43
+ "objective/scores": 4.541956901550293,
44
+ "policy/approxkl_avg": 0.10543511807918549,
45
+ "policy/clipfrac_avg": 0.3203125,
46
+ "policy/entropy_avg": 0.8243188858032227,
47
+ "step": 1,
48
+ "val/clipfrac_avg": 0.0,
49
+ "val/num_eos_tokens": 3325,
50
+ "val/ratio": 1.0002458095550537,
51
+ "val/ratio_var": 1.5084623328220914e-06
52
+ },
53
+ {
54
+ "episode": 1536,
55
+ "epoch": 0.006579849211788897,
56
+ "eps": 6,
57
+ "loss/policy_avg": 0.13810233771800995,
58
+ "lr": 2.9765625e-06,
59
+ "objective/entropy": -55.89725112915039,
60
+ "objective/kl": 1.300195574760437,
61
+ "objective/non_score_reward": 0.0,
62
+ "objective/rlhf_reward": 4.745310306549072,
63
+ "objective/scores": 4.745310306549072,
64
+ "policy/approxkl_avg": 0.11424587666988373,
65
+ "policy/clipfrac_avg": 0.302734375,
66
+ "policy/entropy_avg": 0.760991096496582,
67
+ "step": 2,
68
+ "val/clipfrac_avg": 0.0,
69
+ "val/num_eos_tokens": 3801,
70
+ "val/ratio": 0.9999536275863647,
71
+ "val/ratio_var": 1.5804212125658523e-06
72
+ },
73
+ {
74
+ "episode": 2048,
75
+ "epoch": 0.008773132282385195,
76
+ "eps": 6,
77
+ "loss/policy_avg": 0.25361043214797974,
78
+ "lr": 2.96484375e-06,
79
+ "objective/entropy": -54.02920150756836,
80
+ "objective/kl": 1.721588373184204,
81
+ "objective/non_score_reward": 0.0,
82
+ "objective/rlhf_reward": 4.8178606033325195,
83
+ "objective/scores": 4.8178606033325195,
84
+ "policy/approxkl_avg": 0.12672686576843262,
85
+ "policy/clipfrac_avg": 0.310546875,
86
+ "policy/entropy_avg": 0.7554011344909668,
87
+ "step": 3,
88
+ "val/clipfrac_avg": 0.0,
89
+ "val/num_eos_tokens": 4010,
90
+ "val/ratio": 0.9999880790710449,
91
+ "val/ratio_var": 1.6228774484261521e-06
92
+ },
93
+ {
94
+ "episode": 2560,
95
+ "epoch": 0.010966415352981495,
96
+ "eps": 6,
97
+ "loss/policy_avg": 0.17445668578147888,
98
+ "lr": 2.953125e-06,
99
+ "objective/entropy": -60.463539123535156,
100
+ "objective/kl": 2.6447153091430664,
101
+ "objective/non_score_reward": 0.0,
102
+ "objective/rlhf_reward": 4.877408981323242,
103
+ "objective/scores": 4.877408981323242,
104
+ "policy/approxkl_avg": 0.13307543098926544,
105
+ "policy/clipfrac_avg": 0.275390625,
106
+ "policy/entropy_avg": 0.690821647644043,
107
+ "step": 4,
108
+ "val/clipfrac_avg": 0.0,
109
+ "val/num_eos_tokens": 4589,
110
+ "val/ratio": 1.0001537799835205,
111
+ "val/ratio_var": 1.283996084566752e-06
112
+ },
113
+ {
114
+ "episode": 3072,
115
+ "epoch": 0.013159698423577794,
116
+ "eps": 7,
117
+ "loss/policy_avg": 0.18251696228981018,
118
+ "lr": 2.94140625e-06,
119
+ "objective/entropy": -60.38776397705078,
120
+ "objective/kl": 3.3835220336914062,
121
+ "objective/non_score_reward": 0.0,
122
+ "objective/rlhf_reward": 5.029230117797852,
123
+ "objective/scores": 5.029230117797852,
124
+ "policy/approxkl_avg": 0.11075231432914734,
125
+ "policy/clipfrac_avg": 0.30859375,
126
+ "policy/entropy_avg": 0.6673464775085449,
127
+ "step": 5,
128
+ "val/clipfrac_avg": 0.0,
129
+ "val/num_eos_tokens": 4933,
130
+ "val/ratio": 1.0000388622283936,
131
+ "val/ratio_var": 1.886573841147765e-06
132
+ },
133
+ {
134
+ "episode": 3584,
135
+ "epoch": 0.015352981494174092,
136
+ "eps": 7,
137
+ "loss/policy_avg": 0.0862584114074707,
138
+ "lr": 2.9296875e-06,
139
+ "objective/entropy": -64.953857421875,
140
+ "objective/kl": 3.801461696624756,
141
+ "objective/non_score_reward": 0.0,
142
+ "objective/rlhf_reward": 5.2262468338012695,
143
+ "objective/scores": 5.2262468338012695,
144
+ "policy/approxkl_avg": 0.0921158567070961,
145
+ "policy/clipfrac_avg": 0.31640625,
146
+ "policy/entropy_avg": 0.6317387819290161,
147
+ "step": 6,
148
+ "val/clipfrac_avg": 0.0,
149
+ "val/num_eos_tokens": 5224,
150
+ "val/ratio": 1.0000205039978027,
151
+ "val/ratio_var": 1.223428625962697e-06
152
+ },
153
+ {
154
+ "episode": 4096,
155
+ "epoch": 0.01754626456477039,
156
+ "eps": 7,
157
+ "loss/policy_avg": 0.10839146375656128,
158
+ "lr": 2.91796875e-06,
159
+ "objective/entropy": -64.41149139404297,
160
+ "objective/kl": 4.314542770385742,
161
+ "objective/non_score_reward": 0.0,
162
+ "objective/rlhf_reward": 5.495314121246338,
163
+ "objective/scores": 5.495314121246338,
164
+ "policy/approxkl_avg": 0.11755571514368057,
165
+ "policy/clipfrac_avg": 0.28515625,
166
+ "policy/entropy_avg": 0.6287920475006104,
167
+ "step": 7,
168
+ "val/clipfrac_avg": 0.0,
169
+ "val/num_eos_tokens": 5463,
170
+ "val/ratio": 0.9998381733894348,
171
+ "val/ratio_var": 1.6926536545724957e-06
172
+ },
173
+ {
174
+ "episode": 4608,
175
+ "epoch": 0.01973954763536669,
176
+ "eps": 7,
177
+ "loss/policy_avg": 0.16045916080474854,
178
+ "lr": 2.90625e-06,
179
+ "objective/entropy": -65.33601379394531,
180
+ "objective/kl": 4.707500457763672,
181
+ "objective/non_score_reward": 0.0,
182
+ "objective/rlhf_reward": 5.608126640319824,
183
+ "objective/scores": 5.608126640319824,
184
+ "policy/approxkl_avg": 0.10813738405704498,
185
+ "policy/clipfrac_avg": 0.2890625,
186
+ "policy/entropy_avg": 0.6515289545059204,
187
+ "step": 8,
188
+ "val/clipfrac_avg": 0.0,
189
+ "val/num_eos_tokens": 5590,
190
+ "val/ratio": 1.0000832080841064,
191
+ "val/ratio_var": 1.661756641624379e-06
192
+ },
193
+ {
194
+ "episode": 5120,
195
+ "epoch": 0.02193283070596299,
196
+ "eps": 7,
197
+ "loss/policy_avg": 0.16000822186470032,
198
+ "lr": 2.89453125e-06,
199
+ "objective/entropy": -64.82266235351562,
200
+ "objective/kl": 5.513130187988281,
201
+ "objective/non_score_reward": 0.0,
202
+ "objective/rlhf_reward": 5.976546764373779,
203
+ "objective/scores": 5.976546764373779,
204
+ "policy/approxkl_avg": 0.12319613248109818,
205
+ "policy/clipfrac_avg": 0.29296875,
206
+ "policy/entropy_avg": 0.6444021463394165,
207
+ "step": 9,
208
+ "val/clipfrac_avg": 0.0,
209
+ "val/num_eos_tokens": 5330,
210
+ "val/ratio": 0.9997374415397644,
211
+ "val/ratio_var": 1.3334141613086103e-06
212
+ },
213
+ {
214
+ "episode": 5632,
215
+ "epoch": 0.024126113776559287,
216
+ "eps": 7,
217
+ "loss/policy_avg": 0.22769010066986084,
218
+ "lr": 2.8828125e-06,
219
+ "objective/entropy": -62.54362487792969,
220
+ "objective/kl": 6.375988960266113,
221
+ "objective/non_score_reward": 0.0,
222
+ "objective/rlhf_reward": 6.329231262207031,
223
+ "objective/scores": 6.329231262207031,
224
+ "policy/approxkl_avg": 0.133827343583107,
225
+ "policy/clipfrac_avg": 0.30078125,
226
+ "policy/entropy_avg": 0.6720603704452515,
227
+ "step": 10,
228
+ "val/clipfrac_avg": 0.0,
229
+ "val/num_eos_tokens": 4956,
230
+ "val/ratio": 0.9999898672103882,
231
+ "val/ratio_var": 2.208806336057023e-06
232
+ },
233
+ {
234
+ "episode": 6144,
235
+ "epoch": 0.026319396847155587,
236
+ "eps": 7,
237
+ "loss/policy_avg": 0.16269627213478088,
238
+ "lr": 2.87109375e-06,
239
+ "objective/entropy": -56.140655517578125,
240
+ "objective/kl": 7.935493469238281,
241
+ "objective/non_score_reward": 0.0,
242
+ "objective/rlhf_reward": 6.335080146789551,
243
+ "objective/scores": 6.335080146789551,
244
+ "policy/approxkl_avg": 0.13188697397708893,
245
+ "policy/clipfrac_avg": 0.35546875,
246
+ "policy/entropy_avg": 0.6935639381408691,
247
+ "step": 11,
248
+ "val/clipfrac_avg": 0.0,
249
+ "val/num_eos_tokens": 4904,
250
+ "val/ratio": 0.999968409538269,
251
+ "val/ratio_var": 1.8624212998474832e-06
252
+ },
253
+ {
254
+ "episode": 6656,
255
+ "epoch": 0.028512679917751884,
256
+ "eps": 7,
257
+ "loss/policy_avg": 0.1726413369178772,
258
+ "lr": 2.859375e-06,
259
+ "objective/entropy": -56.71982955932617,
260
+ "objective/kl": 10.505476951599121,
261
+ "objective/non_score_reward": 0.0,
262
+ "objective/rlhf_reward": 7.327908515930176,
263
+ "objective/scores": 7.327908515930176,
264
+ "policy/approxkl_avg": 0.13946227729320526,
265
+ "policy/clipfrac_avg": 0.306640625,
266
+ "policy/entropy_avg": 0.6527446508407593,
267
+ "step": 12,
268
+ "val/clipfrac_avg": 0.0,
269
+ "val/num_eos_tokens": 4210,
270
+ "val/ratio": 1.0002007484436035,
271
+ "val/ratio_var": 1.9585920654208167e-06
272
+ },
273
+ {
274
+ "episode": 7168,
275
+ "epoch": 0.030705962988348184,
276
+ "eps": 7,
277
+ "loss/policy_avg": 0.1594904661178589,
278
+ "lr": 2.84765625e-06,
279
+ "objective/entropy": -57.271263122558594,
280
+ "objective/kl": 12.708869934082031,
281
+ "objective/non_score_reward": 0.0,
282
+ "objective/rlhf_reward": 7.788185119628906,
283
+ "objective/scores": 7.788185119628906,
284
+ "policy/approxkl_avg": 0.14631032943725586,
285
+ "policy/clipfrac_avg": 0.33203125,
286
+ "policy/entropy_avg": 0.6090783476829529,
287
+ "step": 13,
288
+ "val/clipfrac_avg": 0.0,
289
+ "val/num_eos_tokens": 3340,
290
+ "val/ratio": 0.9998956918716431,
291
+ "val/ratio_var": 2.45040769186744e-06
292
+ },
293
+ {
294
+ "episode": 7680,
295
+ "epoch": 0.03289924605894448,
296
+ "eps": 7,
297
+ "loss/policy_avg": 0.15880730748176575,
298
+ "lr": 2.8359375e-06,
299
+ "objective/entropy": -53.81092071533203,
300
+ "objective/kl": 15.003689765930176,
301
+ "objective/non_score_reward": 0.0,
302
+ "objective/rlhf_reward": 7.487188339233398,
303
+ "objective/scores": 7.487188339233398,
304
+ "policy/approxkl_avg": 0.17456671595573425,
305
+ "policy/clipfrac_avg": 0.33203125,
306
+ "policy/entropy_avg": 0.5235726237297058,
307
+ "step": 14,
308
+ "val/clipfrac_avg": 0.0,
309
+ "val/num_eos_tokens": 2483,
310
+ "val/ratio": 1.0001764297485352,
311
+ "val/ratio_var": 1.739899175845494e-06
312
+ },
313
+ {
314
+ "episode": 8192,
315
+ "epoch": 0.03509252912954078,
316
+ "eps": 7,
317
+ "loss/policy_avg": 0.16226203739643097,
318
+ "lr": 2.82421875e-06,
319
+ "objective/entropy": -55.564083099365234,
320
+ "objective/kl": 14.753921508789062,
321
+ "objective/non_score_reward": 0.0,
322
+ "objective/rlhf_reward": 7.572874069213867,
323
+ "objective/scores": 7.572874069213867,
324
+ "policy/approxkl_avg": 0.16666629910469055,
325
+ "policy/clipfrac_avg": 0.302734375,
326
+ "policy/entropy_avg": 0.5176196098327637,
327
+ "step": 15,
328
+ "val/clipfrac_avg": 0.0,
329
+ "val/num_eos_tokens": 2405,
330
+ "val/ratio": 0.9999848008155823,
331
+ "val/ratio_var": 2.589519908724469e-06
332
+ },
333
+ {
334
+ "episode": 8704,
335
+ "epoch": 0.03728581220013708,
336
+ "eps": 7,
337
+ "loss/policy_avg": 0.21661071479320526,
338
+ "lr": 2.8125e-06,
339
+ "objective/entropy": -51.922306060791016,
340
+ "objective/kl": 16.144550323486328,
341
+ "objective/non_score_reward": 0.0,
342
+ "objective/rlhf_reward": 7.781338691711426,
343
+ "objective/scores": 7.781338691711426,
344
+ "policy/approxkl_avg": 0.2002604901790619,
345
+ "policy/clipfrac_avg": 0.337890625,
346
+ "policy/entropy_avg": 0.5265668630599976,
347
+ "step": 16,
348
+ "val/clipfrac_avg": 0.0,
349
+ "val/num_eos_tokens": 2439,
350
+ "val/ratio": 0.9997420907020569,
351
+ "val/ratio_var": 2.6209752377326367e-06
352
+ },
353
+ {
354
+ "episode": 9216,
355
+ "epoch": 0.03947909527073338,
356
+ "eps": 7,
357
+ "loss/policy_avg": 0.26599276065826416,
358
+ "lr": 2.80078125e-06,
359
+ "objective/entropy": -50.82372283935547,
360
+ "objective/kl": 17.150707244873047,
361
+ "objective/non_score_reward": 0.0,
362
+ "objective/rlhf_reward": 7.842694282531738,
363
+ "objective/scores": 7.842694282531738,
364
+ "policy/approxkl_avg": 0.19617530703544617,
365
+ "policy/clipfrac_avg": 0.345703125,
366
+ "policy/entropy_avg": 0.5609679222106934,
367
+ "step": 17,
368
+ "val/clipfrac_avg": 0.0,
369
+ "val/num_eos_tokens": 2719,
370
+ "val/ratio": 1.000065803527832,
371
+ "val/ratio_var": 2.653461933732615e-06
372
+ },
373
+ {
374
+ "episode": 9728,
375
+ "epoch": 0.041672378341329676,
376
+ "eps": 7,
377
+ "loss/policy_avg": 0.18615023791790009,
378
+ "lr": 2.7890625e-06,
379
+ "objective/entropy": -50.224365234375,
380
+ "objective/kl": 18.064285278320312,
381
+ "objective/non_score_reward": 0.0,
382
+ "objective/rlhf_reward": 8.09566879272461,
383
+ "objective/scores": 8.09566879272461,
384
+ "policy/approxkl_avg": 0.2271636724472046,
385
+ "policy/clipfrac_avg": 0.37890625,
386
+ "policy/entropy_avg": 0.5856865048408508,
387
+ "step": 18,
388
+ "val/clipfrac_avg": 0.0,
389
+ "val/num_eos_tokens": 2943,
390
+ "val/ratio": 1.0001342296600342,
391
+ "val/ratio_var": 3.7938905279588653e-06
392
+ },
393
+ {
394
+ "episode": 10240,
395
+ "epoch": 0.04386566141192598,
396
+ "eps": 7,
397
+ "loss/policy_avg": 0.20427988469600677,
398
+ "lr": 2.77734375e-06,
399
+ "objective/entropy": -50.77484893798828,
400
+ "objective/kl": 17.688308715820312,
401
+ "objective/non_score_reward": 0.0,
402
+ "objective/rlhf_reward": 7.970284461975098,
403
+ "objective/scores": 7.970284461975098,
404
+ "policy/approxkl_avg": 0.23018702864646912,
405
+ "policy/clipfrac_avg": 0.341796875,
406
+ "policy/entropy_avg": 0.5944638252258301,
407
+ "step": 19,
408
+ "val/clipfrac_avg": 0.0,
409
+ "val/num_eos_tokens": 3183,
410
+ "val/ratio": 1.000162124633789,
411
+ "val/ratio_var": 3.043409833480837e-06
412
+ },
413
+ {
414
+ "episode": 10752,
415
+ "epoch": 0.04605894448252228,
416
+ "eps": 7,
417
+ "loss/policy_avg": 0.282001793384552,
418
+ "lr": 2.765625e-06,
419
+ "objective/entropy": -44.80044937133789,
420
+ "objective/kl": 19.763790130615234,
421
+ "objective/non_score_reward": 0.0,
422
+ "objective/rlhf_reward": 8.012289047241211,
423
+ "objective/scores": 8.012289047241211,
424
+ "policy/approxkl_avg": 0.24563264846801758,
425
+ "policy/clipfrac_avg": 0.3359375,
426
+ "policy/entropy_avg": 0.6678606271743774,
427
+ "step": 20,
428
+ "val/clipfrac_avg": 0.0,
429
+ "val/num_eos_tokens": 3309,
430
+ "val/ratio": 0.9997632503509521,
431
+ "val/ratio_var": 3.618298023866373e-06
432
+ },
433
+ {
434
+ "episode": 11264,
435
+ "epoch": 0.048252227553118573,
436
+ "eps": 7,
437
+ "loss/policy_avg": 0.15888847410678864,
438
+ "lr": 2.75390625e-06,
439
+ "objective/entropy": -45.08258056640625,
440
+ "objective/kl": 19.886899948120117,
441
+ "objective/non_score_reward": 0.0,
442
+ "objective/rlhf_reward": 8.057466506958008,
443
+ "objective/scores": 8.057466506958008,
444
+ "policy/approxkl_avg": 0.2164287269115448,
445
+ "policy/clipfrac_avg": 0.3671875,
446
+ "policy/entropy_avg": 0.6539000272750854,
447
+ "step": 21,
448
+ "val/clipfrac_avg": 0.0,
449
+ "val/num_eos_tokens": 3399,
450
+ "val/ratio": 0.9997671842575073,
451
+ "val/ratio_var": 3.0791625249548815e-06
452
+ },
453
+ {
454
+ "episode": 11776,
455
+ "epoch": 0.05044551062371487,
456
+ "eps": 7,
457
+ "loss/policy_avg": 0.276736855506897,
458
+ "lr": 2.7421875e-06,
459
+ "objective/entropy": -40.99183654785156,
460
+ "objective/kl": 22.14881134033203,
461
+ "objective/non_score_reward": 0.0,
462
+ "objective/rlhf_reward": 8.060696601867676,
463
+ "objective/scores": 8.060696601867676,
464
+ "policy/approxkl_avg": 0.31746962666511536,
465
+ "policy/clipfrac_avg": 0.349609375,
466
+ "policy/entropy_avg": 0.7243172526359558,
467
+ "step": 22,
468
+ "val/clipfrac_avg": 0.0,
469
+ "val/num_eos_tokens": 3907,
470
+ "val/ratio": 0.9996672868728638,
471
+ "val/ratio_var": 4.220366463414393e-06
472
+ },
473
+ {
474
+ "episode": 12288,
475
+ "epoch": 0.052638793694311174,
476
+ "eps": 7,
477
+ "loss/policy_avg": 0.4665297865867615,
478
+ "lr": 2.73046875e-06,
479
+ "objective/entropy": -35.2369384765625,
480
+ "objective/kl": 23.011898040771484,
481
+ "objective/non_score_reward": 0.0,
482
+ "objective/rlhf_reward": 7.889699935913086,
483
+ "objective/scores": 7.889699935913086,
484
+ "policy/approxkl_avg": 0.3784940838813782,
485
+ "policy/clipfrac_avg": 0.3515625,
486
+ "policy/entropy_avg": 0.7655227184295654,
487
+ "step": 23,
488
+ "val/clipfrac_avg": 0.0,
489
+ "val/num_eos_tokens": 3972,
490
+ "val/ratio": 0.9998047351837158,
491
+ "val/ratio_var": 4.525894382823026e-06
492
+ },
493
+ {
494
+ "episode": 12800,
495
+ "epoch": 0.05483207676490747,
496
+ "eps": 7,
497
+ "loss/policy_avg": 0.43642693758010864,
498
+ "lr": 2.71875e-06,
499
+ "objective/entropy": -36.217777252197266,
500
+ "objective/kl": 22.847183227539062,
501
+ "objective/non_score_reward": 0.0,
502
+ "objective/rlhf_reward": 8.134607315063477,
503
+ "objective/scores": 8.134607315063477,
504
+ "policy/approxkl_avg": 0.330811470746994,
505
+ "policy/clipfrac_avg": 0.39453125,
506
+ "policy/entropy_avg": 0.7883225679397583,
507
+ "step": 24,
508
+ "val/clipfrac_avg": 0.0,
509
+ "val/num_eos_tokens": 4373,
510
+ "val/ratio": 0.9999172687530518,
511
+ "val/ratio_var": 5.261089881969383e-06
512
+ },
513
+ {
514
+ "episode": 13312,
515
+ "epoch": 0.05702535983550377,
516
+ "eps": 7,
517
+ "loss/policy_avg": 0.3709805905818939,
518
+ "lr": 2.70703125e-06,
519
+ "objective/entropy": -28.68521499633789,
520
+ "objective/kl": 24.381669998168945,
521
+ "objective/non_score_reward": 0.0,
522
+ "objective/rlhf_reward": 8.424071311950684,
523
+ "objective/scores": 8.424071311950684,
524
+ "policy/approxkl_avg": 0.26343879103660583,
525
+ "policy/clipfrac_avg": 0.359375,
526
+ "policy/entropy_avg": 0.7675451040267944,
527
+ "step": 25,
528
+ "val/clipfrac_avg": 0.0,
529
+ "val/num_eos_tokens": 4221,
530
+ "val/ratio": 1.0003796815872192,
531
+ "val/ratio_var": 4.443480065674521e-06
532
+ },
533
+ {
534
+ "episode": 13824,
535
+ "epoch": 0.05921864290610007,
536
+ "eps": 7,
537
+ "loss/policy_avg": 0.32363858819007874,
538
+ "lr": 2.6953125e-06,
539
+ "objective/entropy": -28.342884063720703,
540
+ "objective/kl": 24.456398010253906,
541
+ "objective/non_score_reward": 0.0,
542
+ "objective/rlhf_reward": 8.066158294677734,
543
+ "objective/scores": 8.066158294677734,
544
+ "policy/approxkl_avg": 0.2860070466995239,
545
+ "policy/clipfrac_avg": 0.40234375,
546
+ "policy/entropy_avg": 0.7560139298439026,
547
+ "step": 26,
548
+ "val/clipfrac_avg": 0.0,
549
+ "val/num_eos_tokens": 3967,
550
+ "val/ratio": 0.999995231628418,
551
+ "val/ratio_var": 5.418739419837948e-06
552
+ },
553
+ {
554
+ "episode": 14336,
555
+ "epoch": 0.06141192597669637,
556
+ "eps": 7,
557
+ "loss/policy_avg": 0.27364417910575867,
558
+ "lr": 2.68359375e-06,
559
+ "objective/entropy": -21.591156005859375,
560
+ "objective/kl": 26.56311798095703,
561
+ "objective/non_score_reward": 0.0,
562
+ "objective/rlhf_reward": 8.133939743041992,
563
+ "objective/scores": 8.133939743041992,
564
+ "policy/approxkl_avg": 0.34295791387557983,
565
+ "policy/clipfrac_avg": 0.3984375,
566
+ "policy/entropy_avg": 0.7555699944496155,
567
+ "step": 27,
568
+ "val/clipfrac_avg": 0.0,
569
+ "val/num_eos_tokens": 3701,
570
+ "val/ratio": 0.9998049139976501,
571
+ "val/ratio_var": 5.146652256371453e-06
572
+ },
573
+ {
574
+ "episode": 14848,
575
+ "epoch": 0.06360520904729267,
576
+ "eps": 7,
577
+ "loss/policy_avg": 0.3071547746658325,
578
+ "lr": 2.671875e-06,
579
+ "objective/entropy": -23.85720443725586,
580
+ "objective/kl": 26.94635009765625,
581
+ "objective/non_score_reward": 0.0,
582
+ "objective/rlhf_reward": 8.451851844787598,
583
+ "objective/scores": 8.451851844787598,
584
+ "policy/approxkl_avg": 0.301169216632843,
585
+ "policy/clipfrac_avg": 0.375,
586
+ "policy/entropy_avg": 0.7123795747756958,
587
+ "step": 28,
588
+ "val/clipfrac_avg": 0.0,
589
+ "val/num_eos_tokens": 3446,
590
+ "val/ratio": 1.0001068115234375,
591
+ "val/ratio_var": 8.123261977743823e-06
592
+ },
593
+ {
594
+ "episode": 15360,
595
+ "epoch": 0.06579849211788896,
596
+ "eps": 7,
597
+ "loss/policy_avg": 0.24686866998672485,
598
+ "lr": 2.66015625e-06,
599
+ "objective/entropy": -23.746164321899414,
600
+ "objective/kl": 26.223037719726562,
601
+ "objective/non_score_reward": 0.0,
602
+ "objective/rlhf_reward": 8.474257469177246,
603
+ "objective/scores": 8.474257469177246,
604
+ "policy/approxkl_avg": 0.2871531844139099,
605
+ "policy/clipfrac_avg": 0.384765625,
606
+ "policy/entropy_avg": 0.7285399436950684,
607
+ "step": 29,
608
+ "val/clipfrac_avg": 0.0,
609
+ "val/num_eos_tokens": 3073,
610
+ "val/ratio": 0.9999381899833679,
611
+ "val/ratio_var": 4.340675786806969e-06
612
+ },
613
+ {
614
+ "episode": 15872,
615
+ "epoch": 0.06799177518848526,
616
+ "eps": 7,
617
+ "loss/policy_avg": 0.3226906359195709,
618
+ "lr": 2.6484375e-06,
619
+ "objective/entropy": -21.334320068359375,
620
+ "objective/kl": 27.262338638305664,
621
+ "objective/non_score_reward": 0.0,
622
+ "objective/rlhf_reward": 8.353540420532227,
623
+ "objective/scores": 8.353540420532227,
624
+ "policy/approxkl_avg": 0.26032739877700806,
625
+ "policy/clipfrac_avg": 0.3671875,
626
+ "policy/entropy_avg": 0.7068374156951904,
627
+ "step": 30,
628
+ "val/clipfrac_avg": 0.0,
629
+ "val/num_eos_tokens": 2797,
630
+ "val/ratio": 1.0000503063201904,
631
+ "val/ratio_var": 3.849579115922097e-06
632
+ },
633
+ {
634
+ "episode": 16384,
635
+ "epoch": 0.07018505825908156,
636
+ "eps": 7,
637
+ "loss/policy_avg": 0.8861868977546692,
638
+ "lr": 2.63671875e-06,
639
+ "objective/entropy": -16.9752197265625,
640
+ "objective/kl": 27.735870361328125,
641
+ "objective/non_score_reward": 0.0,
642
+ "objective/rlhf_reward": 8.225103378295898,
643
+ "objective/scores": 8.225103378295898,
644
+ "policy/approxkl_avg": 0.3011099100112915,
645
+ "policy/clipfrac_avg": 0.373046875,
646
+ "policy/entropy_avg": 0.6992452144622803,
647
+ "step": 31,
648
+ "val/clipfrac_avg": 0.0,
649
+ "val/num_eos_tokens": 2656,
650
+ "val/ratio": 1.0000051259994507,
651
+ "val/ratio_var": 3.2618629575154046e-06
652
+ },
653
+ {
654
+ "episode": 16896,
655
+ "epoch": 0.07237834132967787,
656
+ "eps": 7,
657
+ "loss/policy_avg": 0.42881882190704346,
658
+ "lr": 2.6250000000000003e-06,
659
+ "objective/entropy": -25.429481506347656,
660
+ "objective/kl": 26.258373260498047,
661
+ "objective/non_score_reward": 0.0,
662
+ "objective/rlhf_reward": 8.536768913269043,
663
+ "objective/scores": 8.536768913269043,
664
+ "policy/approxkl_avg": 0.2783983647823334,
665
+ "policy/clipfrac_avg": 0.357421875,
666
+ "policy/entropy_avg": 0.6586651802062988,
667
+ "step": 32,
668
+ "val/clipfrac_avg": 0.0,
669
+ "val/num_eos_tokens": 2739,
670
+ "val/ratio": 1.0000145435333252,
671
+ "val/ratio_var": 5.060162948211655e-06
672
+ },
673
+ {
674
+ "episode": 17408,
675
+ "epoch": 0.07457162440027416,
676
+ "eps": 7,
677
+ "loss/policy_avg": 0.3053734004497528,
678
+ "lr": 2.61328125e-06,
679
+ "objective/entropy": -27.58635711669922,
680
+ "objective/kl": 26.437305450439453,
681
+ "objective/non_score_reward": 0.0,
682
+ "objective/rlhf_reward": 8.641759872436523,
683
+ "objective/scores": 8.641759872436523,
684
+ "policy/approxkl_avg": 0.27716416120529175,
685
+ "policy/clipfrac_avg": 0.384765625,
686
+ "policy/entropy_avg": 0.6354501843452454,
687
+ "step": 33,
688
+ "val/clipfrac_avg": 0.0,
689
+ "val/num_eos_tokens": 2508,
690
+ "val/ratio": 1.0002658367156982,
691
+ "val/ratio_var": 5.2127916205790825e-06
692
+ },
693
+ {
694
+ "episode": 17920,
695
+ "epoch": 0.07676490747087046,
696
+ "eps": 7,
697
+ "loss/policy_avg": 0.29732874035835266,
698
+ "lr": 2.6015625e-06,
699
+ "objective/entropy": -21.862457275390625,
700
+ "objective/kl": 26.802410125732422,
701
+ "objective/non_score_reward": 0.0,
702
+ "objective/rlhf_reward": 8.533287048339844,
703
+ "objective/scores": 8.533287048339844,
704
+ "policy/approxkl_avg": 0.2853066325187683,
705
+ "policy/clipfrac_avg": 0.34765625,
706
+ "policy/entropy_avg": 0.6434289813041687,
707
+ "step": 34,
708
+ "val/clipfrac_avg": 0.0,
709
+ "val/num_eos_tokens": 2256,
710
+ "val/ratio": 1.0001918077468872,
711
+ "val/ratio_var": 3.2171374186873436e-06
712
+ },
713
+ {
714
+ "episode": 18432,
715
+ "epoch": 0.07895819054146676,
716
+ "eps": 7,
717
+ "loss/policy_avg": 0.29931700229644775,
718
+ "lr": 2.5898437500000003e-06,
719
+ "objective/entropy": -26.582080841064453,
720
+ "objective/kl": 26.952320098876953,
721
+ "objective/non_score_reward": 0.0,
722
+ "objective/rlhf_reward": 8.649660110473633,
723
+ "objective/scores": 8.649660110473633,
724
+ "policy/approxkl_avg": 0.29774773120880127,
725
+ "policy/clipfrac_avg": 0.369140625,
726
+ "policy/entropy_avg": 0.6092206835746765,
727
+ "step": 35,
728
+ "val/clipfrac_avg": 0.0,
729
+ "val/num_eos_tokens": 2189,
730
+ "val/ratio": 0.9998046159744263,
731
+ "val/ratio_var": 4.721144250652287e-06
732
+ },
733
+ {
734
+ "episode": 18944,
735
+ "epoch": 0.08115147361206305,
736
+ "eps": 7,
737
+ "loss/policy_avg": 0.29533323645591736,
738
+ "lr": 2.578125e-06,
739
+ "objective/entropy": -28.472558975219727,
740
+ "objective/kl": 26.34886932373047,
741
+ "objective/non_score_reward": 0.0,
742
+ "objective/rlhf_reward": 8.683137893676758,
743
+ "objective/scores": 8.683137893676758,
744
+ "policy/approxkl_avg": 0.2845500707626343,
745
+ "policy/clipfrac_avg": 0.3515625,
746
+ "policy/entropy_avg": 0.6110947132110596,
747
+ "step": 36,
748
+ "val/clipfrac_avg": 0.0,
749
+ "val/num_eos_tokens": 2113,
750
+ "val/ratio": 1.0004630088806152,
751
+ "val/ratio_var": 3.1590116122970358e-06
752
+ },
753
+ {
754
+ "episode": 19456,
755
+ "epoch": 0.08334475668265935,
756
+ "eps": 7,
757
+ "loss/policy_avg": 0.1853514462709427,
758
+ "lr": 2.56640625e-06,
759
+ "objective/entropy": -28.34747314453125,
760
+ "objective/kl": 26.15137481689453,
761
+ "objective/non_score_reward": 0.0,
762
+ "objective/rlhf_reward": 8.642951965332031,
763
+ "objective/scores": 8.642951965332031,
764
+ "policy/approxkl_avg": 0.26775410771369934,
765
+ "policy/clipfrac_avg": 0.369140625,
766
+ "policy/entropy_avg": 0.5745651721954346,
767
+ "step": 37,
768
+ "val/clipfrac_avg": 0.0,
769
+ "val/num_eos_tokens": 2008,
770
+ "val/ratio": 0.9999517202377319,
771
+ "val/ratio_var": 3.727831654032343e-06
772
+ },
773
+ {
774
+ "episode": 19968,
775
+ "epoch": 0.08553803975325565,
776
+ "eps": 7,
777
+ "loss/policy_avg": 0.43564116954803467,
778
+ "lr": 2.5546875000000003e-06,
779
+ "objective/entropy": -30.9276065826416,
780
+ "objective/kl": 26.764188766479492,
781
+ "objective/non_score_reward": 0.0,
782
+ "objective/rlhf_reward": 8.747605323791504,
783
+ "objective/scores": 8.747605323791504,
784
+ "policy/approxkl_avg": 0.2727493345737457,
785
+ "policy/clipfrac_avg": 0.408203125,
786
+ "policy/entropy_avg": 0.5555662512779236,
787
+ "step": 38,
788
+ "val/clipfrac_avg": 0.0,
789
+ "val/num_eos_tokens": 1949,
790
+ "val/ratio": 1.000280737876892,
791
+ "val/ratio_var": 4.035889560327632e-06
792
+ },
793
+ {
794
+ "episode": 20480,
795
+ "epoch": 0.08773132282385196,
796
+ "eps": 7,
797
+ "loss/policy_avg": 0.33555662631988525,
798
+ "lr": 2.54296875e-06,
799
+ "objective/entropy": -30.725311279296875,
800
+ "objective/kl": 26.652061462402344,
801
+ "objective/non_score_reward": 0.0,
802
+ "objective/rlhf_reward": 8.656892776489258,
803
+ "objective/scores": 8.656892776489258,
804
+ "policy/approxkl_avg": 0.3008157014846802,
805
+ "policy/clipfrac_avg": 0.3515625,
806
+ "policy/entropy_avg": 0.5807157754898071,
807
+ "step": 39,
808
+ "val/clipfrac_avg": 0.0,
809
+ "val/num_eos_tokens": 1964,
810
+ "val/ratio": 1.0001399517059326,
811
+ "val/ratio_var": 5.213231816014741e-06
812
+ },
813
+ {
814
+ "episode": 20992,
815
+ "epoch": 0.08992460589444826,
816
+ "eps": 7,
817
+ "loss/policy_avg": 0.251528263092041,
818
+ "lr": 2.53125e-06,
819
+ "objective/entropy": -31.469741821289062,
820
+ "objective/kl": 27.247440338134766,
821
+ "objective/non_score_reward": 0.0,
822
+ "objective/rlhf_reward": 8.747458457946777,
823
+ "objective/scores": 8.747458457946777,
824
+ "policy/approxkl_avg": 0.31611520051956177,
825
+ "policy/clipfrac_avg": 0.39453125,
826
+ "policy/entropy_avg": 0.5639370679855347,
827
+ "step": 40,
828
+ "val/clipfrac_avg": 0.0,
829
+ "val/num_eos_tokens": 1938,
830
+ "val/ratio": 0.9993410110473633,
831
+ "val/ratio_var": 3.2128821203514235e-06
832
+ },
833
+ {
834
+ "episode": 21504,
835
+ "epoch": 0.09211788896504455,
836
+ "eps": 7,
837
+ "loss/policy_avg": 0.2754300832748413,
838
+ "lr": 2.5195312500000003e-06,
839
+ "objective/entropy": -32.1821403503418,
840
+ "objective/kl": 26.554561614990234,
841
+ "objective/non_score_reward": 0.0,
842
+ "objective/rlhf_reward": 8.68823528289795,
843
+ "objective/scores": 8.68823528289795,
844
+ "policy/approxkl_avg": 0.2688625454902649,
845
+ "policy/clipfrac_avg": 0.3984375,
846
+ "policy/entropy_avg": 0.5564634203910828,
847
+ "step": 41,
848
+ "val/clipfrac_avg": 0.0,
849
+ "val/num_eos_tokens": 1861,
850
+ "val/ratio": 1.0001417398452759,
851
+ "val/ratio_var": 3.7918011912552174e-06
852
+ },
853
+ {
854
+ "episode": 22016,
855
+ "epoch": 0.09431117203564085,
856
+ "eps": 7,
857
+ "loss/policy_avg": 0.32364964485168457,
858
+ "lr": 2.5078125e-06,
859
+ "objective/entropy": -31.65809440612793,
860
+ "objective/kl": 26.914566040039062,
861
+ "objective/non_score_reward": 0.0,
862
+ "objective/rlhf_reward": 8.63361930847168,
863
+ "objective/scores": 8.63361930847168,
864
+ "policy/approxkl_avg": 0.319877952337265,
865
+ "policy/clipfrac_avg": 0.357421875,
866
+ "policy/entropy_avg": 0.5548701882362366,
867
+ "step": 42,
868
+ "val/clipfrac_avg": 0.0,
869
+ "val/num_eos_tokens": 1894,
870
+ "val/ratio": 0.9996398687362671,
871
+ "val/ratio_var": 3.310229885755689e-06
872
+ },
873
+ {
874
+ "episode": 22528,
875
+ "epoch": 0.09650445510623715,
876
+ "eps": 7,
877
+ "loss/policy_avg": 0.2156452238559723,
878
+ "lr": 2.49609375e-06,
879
+ "objective/entropy": -33.42183303833008,
880
+ "objective/kl": 25.768882751464844,
881
+ "objective/non_score_reward": 0.0,
882
+ "objective/rlhf_reward": 8.491684913635254,
883
+ "objective/scores": 8.491684913635254,
884
+ "policy/approxkl_avg": 0.37833964824676514,
885
+ "policy/clipfrac_avg": 0.345703125,
886
+ "policy/entropy_avg": 0.5088037252426147,
887
+ "step": 43,
888
+ "val/clipfrac_avg": 0.0,
889
+ "val/num_eos_tokens": 1638,
890
+ "val/ratio": 0.9993675947189331,
891
+ "val/ratio_var": 4.396941676532151e-06
892
+ },
893
+ {
894
+ "episode": 23040,
895
+ "epoch": 0.09869773817683344,
896
+ "eps": 7,
897
+ "loss/policy_avg": 0.283500611782074,
898
+ "lr": 2.4843750000000002e-06,
899
+ "objective/entropy": -38.62915802001953,
900
+ "objective/kl": 24.288204193115234,
901
+ "objective/non_score_reward": 0.0,
902
+ "objective/rlhf_reward": 8.610681533813477,
903
+ "objective/scores": 8.610681533813477,
904
+ "policy/approxkl_avg": 0.26353907585144043,
905
+ "policy/clipfrac_avg": 0.341796875,
906
+ "policy/entropy_avg": 0.524039626121521,
907
+ "step": 44,
908
+ "val/clipfrac_avg": 0.0,
909
+ "val/num_eos_tokens": 1721,
910
+ "val/ratio": 1.000080943107605,
911
+ "val/ratio_var": 5.167307790543418e-06
912
+ },
913
+ {
914
+ "episode": 23552,
915
+ "epoch": 0.10089102124742974,
916
+ "eps": 7,
917
+ "loss/policy_avg": 0.26535850763320923,
918
+ "lr": 2.47265625e-06,
919
+ "objective/entropy": -38.197269439697266,
920
+ "objective/kl": 23.877710342407227,
921
+ "objective/non_score_reward": 0.0,
922
+ "objective/rlhf_reward": 8.435002326965332,
923
+ "objective/scores": 8.435002326965332,
924
+ "policy/approxkl_avg": 0.3536168932914734,
925
+ "policy/clipfrac_avg": 0.353515625,
926
+ "policy/entropy_avg": 0.5166954398155212,
927
+ "step": 45,
928
+ "val/clipfrac_avg": 0.0,
929
+ "val/num_eos_tokens": 1756,
930
+ "val/ratio": 1.0001449584960938,
931
+ "val/ratio_var": 3.216674940631492e-06
932
+ },
933
+ {
934
+ "episode": 24064,
935
+ "epoch": 0.10308430431802605,
936
+ "eps": 7,
937
+ "loss/policy_avg": 0.31962326169013977,
938
+ "lr": 2.4609375e-06,
939
+ "objective/entropy": -42.57143020629883,
940
+ "objective/kl": 23.369226455688477,
941
+ "objective/non_score_reward": 0.0,
942
+ "objective/rlhf_reward": 8.742530822753906,
943
+ "objective/scores": 8.742530822753906,
944
+ "policy/approxkl_avg": 0.2840489149093628,
945
+ "policy/clipfrac_avg": 0.34375,
946
+ "policy/entropy_avg": 0.5359203815460205,
947
+ "step": 46,
948
+ "val/clipfrac_avg": 0.0,
949
+ "val/num_eos_tokens": 1840,
950
+ "val/ratio": 1.0000276565551758,
951
+ "val/ratio_var": 3.076736902585253e-06
952
+ },
953
+ {
954
+ "episode": 24576,
955
+ "epoch": 0.10527758738862235,
956
+ "eps": 7,
957
+ "loss/policy_avg": 0.1421811878681183,
958
+ "lr": 2.4492187500000002e-06,
959
+ "objective/entropy": -45.13118362426758,
960
+ "objective/kl": 22.429969787597656,
961
+ "objective/non_score_reward": 0.0,
962
+ "objective/rlhf_reward": 8.711127281188965,
963
+ "objective/scores": 8.711127281188965,
964
+ "policy/approxkl_avg": 0.2315264642238617,
965
+ "policy/clipfrac_avg": 0.353515625,
966
+ "policy/entropy_avg": 0.5265815854072571,
967
+ "step": 47,
968
+ "val/clipfrac_avg": 0.0,
969
+ "val/num_eos_tokens": 1829,
970
+ "val/ratio": 1.000090479850769,
971
+ "val/ratio_var": 2.3445722945325542e-06
972
+ },
973
+ {
974
+ "episode": 25088,
975
+ "epoch": 0.10747087045921865,
976
+ "eps": 7,
977
+ "loss/policy_avg": 0.29607272148132324,
978
+ "lr": 2.4375e-06,
979
+ "objective/entropy": -47.1614875793457,
980
+ "objective/kl": 21.755367279052734,
981
+ "objective/non_score_reward": 0.0,
982
+ "objective/rlhf_reward": 8.467025756835938,
983
+ "objective/scores": 8.467025756835938,
984
+ "policy/approxkl_avg": 0.23295345902442932,
985
+ "policy/clipfrac_avg": 0.310546875,
986
+ "policy/entropy_avg": 0.5038943290710449,
987
+ "step": 48,
988
+ "val/clipfrac_avg": 0.0,
989
+ "val/num_eos_tokens": 1841,
990
+ "val/ratio": 1.0003342628479004,
991
+ "val/ratio_var": 2.7536118523130426e-06
992
+ },
993
+ {
994
+ "episode": 25600,
995
+ "epoch": 0.10966415352981494,
996
+ "eps": 7,
997
+ "loss/policy_avg": 0.26943719387054443,
998
+ "lr": 2.42578125e-06,
999
+ "objective/entropy": -47.87194061279297,
1000
+ "objective/kl": 22.272235870361328,
1001
+ "objective/non_score_reward": 0.0,
1002
+ "objective/rlhf_reward": 8.529675483703613,
1003
+ "objective/scores": 8.529675483703613,
1004
+ "policy/approxkl_avg": 0.23642447590827942,
1005
+ "policy/clipfrac_avg": 0.30078125,
1006
+ "policy/entropy_avg": 0.48028820753097534,
1007
+ "step": 49,
1008
+ "val/clipfrac_avg": 0.0,
1009
+ "val/num_eos_tokens": 1737,
1010
+ "val/ratio": 1.0001472234725952,
1011
+ "val/ratio_var": 3.836444648186443e-06
1012
+ },
1013
+ {
1014
+ "episode": 26112,
1015
+ "epoch": 0.11185743660041124,
1016
+ "eps": 7,
1017
+ "loss/policy_avg": 0.19396540522575378,
1018
+ "lr": 2.4140625000000002e-06,
1019
+ "objective/entropy": -46.96308135986328,
1020
+ "objective/kl": 21.81145477294922,
1021
+ "objective/non_score_reward": 0.0,
1022
+ "objective/rlhf_reward": 8.142440795898438,
1023
+ "objective/scores": 8.142440795898438,
1024
+ "policy/approxkl_avg": 0.23152337968349457,
1025
+ "policy/clipfrac_avg": 0.375,
1026
+ "policy/entropy_avg": 0.546012282371521,
1027
+ "step": 50,
1028
+ "val/clipfrac_avg": 0.0,
1029
+ "val/num_eos_tokens": 1915,
1030
+ "val/ratio": 0.9999135136604309,
1031
+ "val/ratio_var": 4.054582859680522e-06
1032
+ },
1033
+ {
1034
+ "episode": 26624,
1035
+ "epoch": 0.11405071967100754,
1036
+ "eps": 7,
1037
+ "loss/policy_avg": 0.17617464065551758,
1038
+ "lr": 2.40234375e-06,
1039
+ "objective/entropy": -50.88390350341797,
1040
+ "objective/kl": 21.53017807006836,
1041
+ "objective/non_score_reward": 0.0,
1042
+ "objective/rlhf_reward": 8.507946014404297,
1043
+ "objective/scores": 8.507946014404297,
1044
+ "policy/approxkl_avg": 0.23789095878601074,
1045
+ "policy/clipfrac_avg": 0.345703125,
1046
+ "policy/entropy_avg": 0.5126174092292786,
1047
+ "step": 51,
1048
+ "val/clipfrac_avg": 0.0,
1049
+ "val/num_eos_tokens": 1825,
1050
+ "val/ratio": 0.9999390840530396,
1051
+ "val/ratio_var": 3.831887170235859e-06
1052
+ },
1053
+ {
1054
+ "episode": 27136,
1055
+ "epoch": 0.11624400274160383,
1056
+ "eps": 7,
1057
+ "loss/policy_avg": 0.2936260402202606,
1058
+ "lr": 2.390625e-06,
1059
+ "objective/entropy": -49.55311584472656,
1060
+ "objective/kl": 21.252099990844727,
1061
+ "objective/non_score_reward": 0.0,
1062
+ "objective/rlhf_reward": 8.27835464477539,
1063
+ "objective/scores": 8.27835464477539,
1064
+ "policy/approxkl_avg": 0.26326990127563477,
1065
+ "policy/clipfrac_avg": 0.310546875,
1066
+ "policy/entropy_avg": 0.5240695476531982,
1067
+ "step": 52,
1068
+ "val/clipfrac_avg": 0.0,
1069
+ "val/num_eos_tokens": 1999,
1070
+ "val/ratio": 1.00055992603302,
1071
+ "val/ratio_var": 3.646538061730098e-06
1072
+ },
1073
+ {
1074
+ "episode": 27648,
1075
+ "epoch": 0.11843728581220014,
1076
+ "eps": 7,
1077
+ "loss/policy_avg": 0.28315332531929016,
1078
+ "lr": 2.3789062500000002e-06,
1079
+ "objective/entropy": -50.70973205566406,
1080
+ "objective/kl": 20.522171020507812,
1081
+ "objective/non_score_reward": 0.0,
1082
+ "objective/rlhf_reward": 8.508468627929688,
1083
+ "objective/scores": 8.508468627929688,
1084
+ "policy/approxkl_avg": 0.27963656187057495,
1085
+ "policy/clipfrac_avg": 0.32421875,
1086
+ "policy/entropy_avg": 0.5086656808853149,
1087
+ "step": 53,
1088
+ "val/clipfrac_avg": 0.0,
1089
+ "val/num_eos_tokens": 1877,
1090
+ "val/ratio": 0.9998332262039185,
1091
+ "val/ratio_var": 3.3264000194321852e-06
1092
+ },
1093
+ {
1094
+ "episode": 28160,
1095
+ "epoch": 0.12063056888279644,
1096
+ "eps": 7,
1097
+ "loss/policy_avg": 0.1823330521583557,
1098
+ "lr": 2.3671875e-06,
1099
+ "objective/entropy": -51.118812561035156,
1100
+ "objective/kl": 21.896831512451172,
1101
+ "objective/non_score_reward": 0.0,
1102
+ "objective/rlhf_reward": 8.411741256713867,
1103
+ "objective/scores": 8.411741256713867,
1104
+ "policy/approxkl_avg": 0.22382813692092896,
1105
+ "policy/clipfrac_avg": 0.357421875,
1106
+ "policy/entropy_avg": 0.5047107934951782,
1107
+ "step": 54,
1108
+ "val/clipfrac_avg": 0.0,
1109
+ "val/num_eos_tokens": 1885,
1110
+ "val/ratio": 0.999740481376648,
1111
+ "val/ratio_var": 2.7290527668810682e-06
1112
+ },
1113
+ {
1114
+ "episode": 28672,
1115
+ "epoch": 0.12282385195339274,
1116
+ "eps": 7,
1117
+ "loss/policy_avg": 0.25143423676490784,
1118
+ "lr": 2.35546875e-06,
1119
+ "objective/entropy": -50.577789306640625,
1120
+ "objective/kl": 21.04738998413086,
1121
+ "objective/non_score_reward": 0.0,
1122
+ "objective/rlhf_reward": 8.411981582641602,
1123
+ "objective/scores": 8.411981582641602,
1124
+ "policy/approxkl_avg": 0.22654137015342712,
1125
+ "policy/clipfrac_avg": 0.32421875,
1126
+ "policy/entropy_avg": 0.5130290985107422,
1127
+ "step": 55,
1128
+ "val/clipfrac_avg": 0.0,
1129
+ "val/num_eos_tokens": 1979,
1130
+ "val/ratio": 1.0000386238098145,
1131
+ "val/ratio_var": 2.045640940195881e-06
1132
+ },
1133
+ {
1134
+ "episode": 29184,
1135
+ "epoch": 0.12501713502398903,
1136
+ "eps": 7,
1137
+ "loss/policy_avg": 0.14284127950668335,
1138
+ "lr": 2.3437500000000002e-06,
1139
+ "objective/entropy": -51.72842788696289,
1140
+ "objective/kl": 21.067766189575195,
1141
+ "objective/non_score_reward": 0.0,
1142
+ "objective/rlhf_reward": 8.488981246948242,
1143
+ "objective/scores": 8.488981246948242,
1144
+ "policy/approxkl_avg": 0.2212221920490265,
1145
+ "policy/clipfrac_avg": 0.34375,
1146
+ "policy/entropy_avg": 0.4959627687931061,
1147
+ "step": 56,
1148
+ "val/clipfrac_avg": 0.0,
1149
+ "val/num_eos_tokens": 2032,
1150
+ "val/ratio": 1.0002572536468506,
1151
+ "val/ratio_var": 3.1830661555432016e-06
1152
+ },
1153
+ {
1154
+ "episode": 29696,
1155
+ "epoch": 0.12721041809458533,
1156
+ "eps": 7,
1157
+ "loss/policy_avg": 0.2772967517375946,
1158
+ "lr": 2.33203125e-06,
1159
+ "objective/entropy": -50.611061096191406,
1160
+ "objective/kl": 20.819595336914062,
1161
+ "objective/non_score_reward": 0.0,
1162
+ "objective/rlhf_reward": 8.417530059814453,
1163
+ "objective/scores": 8.417530059814453,
1164
+ "policy/approxkl_avg": 0.20757423341274261,
1165
+ "policy/clipfrac_avg": 0.345703125,
1166
+ "policy/entropy_avg": 0.5225619077682495,
1167
+ "step": 57,
1168
+ "val/clipfrac_avg": 0.0,
1169
+ "val/num_eos_tokens": 2159,
1170
+ "val/ratio": 1.0001440048217773,
1171
+ "val/ratio_var": 3.0312132821563864e-06
1172
+ },
1173
+ {
1174
+ "episode": 30208,
1175
+ "epoch": 0.12940370116518163,
1176
+ "eps": 7,
1177
+ "loss/policy_avg": 0.24927854537963867,
1178
+ "lr": 2.3203125e-06,
1179
+ "objective/entropy": -52.76747512817383,
1180
+ "objective/kl": 20.694204330444336,
1181
+ "objective/non_score_reward": 0.0,
1182
+ "objective/rlhf_reward": 8.423263549804688,
1183
+ "objective/scores": 8.423263549804688,
1184
+ "policy/approxkl_avg": 0.2698250412940979,
1185
+ "policy/clipfrac_avg": 0.318359375,
1186
+ "policy/entropy_avg": 0.5205322504043579,
1187
+ "step": 58,
1188
+ "val/clipfrac_avg": 0.0,
1189
+ "val/num_eos_tokens": 2188,
1190
+ "val/ratio": 1.0001823902130127,
1191
+ "val/ratio_var": 3.7059455735288793e-06
1192
+ },
1193
+ {
1194
+ "episode": 30720,
1195
+ "epoch": 0.13159698423577793,
1196
+ "eps": 7,
1197
+ "loss/policy_avg": 0.32080554962158203,
1198
+ "lr": 2.30859375e-06,
1199
+ "objective/entropy": -47.49811553955078,
1200
+ "objective/kl": 21.551021575927734,
1201
+ "objective/non_score_reward": 0.0,
1202
+ "objective/rlhf_reward": 8.264215469360352,
1203
+ "objective/scores": 8.264215469360352,
1204
+ "policy/approxkl_avg": 0.25660836696624756,
1205
+ "policy/clipfrac_avg": 0.322265625,
1206
+ "policy/entropy_avg": 0.5375617146492004,
1207
+ "step": 59,
1208
+ "val/clipfrac_avg": 0.0,
1209
+ "val/num_eos_tokens": 2128,
1210
+ "val/ratio": 1.0002331733703613,
1211
+ "val/ratio_var": 2.6993873234459898e-06
1212
+ },
1213
+ {
1214
+ "episode": 31232,
1215
+ "epoch": 0.13379026730637422,
1216
+ "eps": 7,
1217
+ "loss/policy_avg": 0.14086592197418213,
1218
+ "lr": 2.296875e-06,
1219
+ "objective/entropy": -51.65747833251953,
1220
+ "objective/kl": 20.77255630493164,
1221
+ "objective/non_score_reward": 0.0,
1222
+ "objective/rlhf_reward": 8.408650398254395,
1223
+ "objective/scores": 8.408650398254395,
1224
+ "policy/approxkl_avg": 0.23100361227989197,
1225
+ "policy/clipfrac_avg": 0.3515625,
1226
+ "policy/entropy_avg": 0.5363867878913879,
1227
+ "step": 60,
1228
+ "val/clipfrac_avg": 0.0,
1229
+ "val/num_eos_tokens": 2336,
1230
+ "val/ratio": 1.000304102897644,
1231
+ "val/ratio_var": 2.394616331002908e-06
1232
+ },
1233
+ {
1234
+ "episode": 31744,
1235
+ "epoch": 0.13598355037697052,
1236
+ "eps": 7,
1237
+ "loss/policy_avg": 0.2954641282558441,
1238
+ "lr": 2.28515625e-06,
1239
+ "objective/entropy": -50.37486267089844,
1240
+ "objective/kl": 21.061641693115234,
1241
+ "objective/non_score_reward": 0.0,
1242
+ "objective/rlhf_reward": 8.459826469421387,
1243
+ "objective/scores": 8.459826469421387,
1244
+ "policy/approxkl_avg": 0.24728071689605713,
1245
+ "policy/clipfrac_avg": 0.3203125,
1246
+ "policy/entropy_avg": 0.5351447463035583,
1247
+ "step": 61,
1248
+ "val/clipfrac_avg": 0.0,
1249
+ "val/num_eos_tokens": 2212,
1250
+ "val/ratio": 1.000131607055664,
1251
+ "val/ratio_var": 2.8336792183836224e-06
1252
+ },
1253
+ {
1254
+ "episode": 32256,
1255
+ "epoch": 0.13817683344756682,
1256
+ "eps": 7,
1257
+ "loss/policy_avg": 0.2441791594028473,
1258
+ "lr": 2.2734375e-06,
1259
+ "objective/entropy": -51.75057601928711,
1260
+ "objective/kl": 21.086687088012695,
1261
+ "objective/non_score_reward": 0.0,
1262
+ "objective/rlhf_reward": 8.422032356262207,
1263
+ "objective/scores": 8.422032356262207,
1264
+ "policy/approxkl_avg": 0.23406898975372314,
1265
+ "policy/clipfrac_avg": 0.310546875,
1266
+ "policy/entropy_avg": 0.5176072120666504,
1267
+ "step": 62,
1268
+ "val/clipfrac_avg": 0.0,
1269
+ "val/num_eos_tokens": 2191,
1270
+ "val/ratio": 1.0005054473876953,
1271
+ "val/ratio_var": 3.65329015039606e-06
1272
+ },
1273
+ {
1274
+ "episode": 32768,
1275
+ "epoch": 0.1403701165181631,
1276
+ "eps": 7,
1277
+ "loss/policy_avg": 0.22965137660503387,
1278
+ "lr": 2.26171875e-06,
1279
+ "objective/entropy": -51.704444885253906,
1280
+ "objective/kl": 21.60610008239746,
1281
+ "objective/non_score_reward": 0.0,
1282
+ "objective/rlhf_reward": 8.439580917358398,
1283
+ "objective/scores": 8.439580917358398,
1284
+ "policy/approxkl_avg": 0.21967627108097076,
1285
+ "policy/clipfrac_avg": 0.333984375,
1286
+ "policy/entropy_avg": 0.524364173412323,
1287
+ "step": 63,
1288
+ "val/clipfrac_avg": 0.0,
1289
+ "val/num_eos_tokens": 2332,
1290
+ "val/ratio": 0.9997615814208984,
1291
+ "val/ratio_var": 3.444454250711715e-06
1292
+ },
1293
+ {
1294
+ "episode": 33280,
1295
+ "epoch": 0.14256339958875944,
1296
+ "eps": 7,
1297
+ "loss/policy_avg": 0.11451836675405502,
1298
+ "lr": 2.25e-06,
1299
+ "objective/entropy": -52.000946044921875,
1300
+ "objective/kl": 21.42214012145996,
1301
+ "objective/non_score_reward": 0.0,
1302
+ "objective/rlhf_reward": 8.531951904296875,
1303
+ "objective/scores": 8.531951904296875,
1304
+ "policy/approxkl_avg": 0.21284940838813782,
1305
+ "policy/clipfrac_avg": 0.3359375,
1306
+ "policy/entropy_avg": 0.551526665687561,
1307
+ "step": 64,
1308
+ "val/clipfrac_avg": 0.0,
1309
+ "val/num_eos_tokens": 2463,
1310
+ "val/ratio": 0.9998893141746521,
1311
+ "val/ratio_var": 2.1311489035724662e-06
1312
+ },
1313
+ {
1314
+ "episode": 33792,
1315
+ "epoch": 0.14475668265935573,
1316
+ "eps": 7,
1317
+ "loss/policy_avg": 0.1996244490146637,
1318
+ "lr": 2.23828125e-06,
1319
+ "objective/entropy": -49.46868896484375,
1320
+ "objective/kl": 22.10938262939453,
1321
+ "objective/non_score_reward": 0.0,
1322
+ "objective/rlhf_reward": 8.431674003601074,
1323
+ "objective/scores": 8.431674003601074,
1324
+ "policy/approxkl_avg": 0.2345934510231018,
1325
+ "policy/clipfrac_avg": 0.3671875,
1326
+ "policy/entropy_avg": 0.5430195331573486,
1327
+ "step": 65,
1328
+ "val/clipfrac_avg": 0.0,
1329
+ "val/num_eos_tokens": 2527,
1330
+ "val/ratio": 1.000130534172058,
1331
+ "val/ratio_var": 2.9720413294853643e-06
1332
+ },
1333
+ {
1334
+ "episode": 34304,
1335
+ "epoch": 0.14694996572995203,
1336
+ "eps": 7,
1337
+ "loss/policy_avg": 0.24282732605934143,
1338
+ "lr": 2.2265625e-06,
1339
+ "objective/entropy": -49.502906799316406,
1340
+ "objective/kl": 21.639503479003906,
1341
+ "objective/non_score_reward": 0.0,
1342
+ "objective/rlhf_reward": 8.574268341064453,
1343
+ "objective/scores": 8.574268341064453,
1344
+ "policy/approxkl_avg": 0.27672481536865234,
1345
+ "policy/clipfrac_avg": 0.373046875,
1346
+ "policy/entropy_avg": 0.5653039216995239,
1347
+ "step": 66,
1348
+ "val/clipfrac_avg": 0.0,
1349
+ "val/num_eos_tokens": 2667,
1350
+ "val/ratio": 1.000170111656189,
1351
+ "val/ratio_var": 3.4806193980330136e-06
1352
+ },
1353
+ {
1354
+ "episode": 34816,
1355
+ "epoch": 0.14914324880054833,
1356
+ "eps": 7,
1357
+ "loss/policy_avg": 0.2232193946838379,
1358
+ "lr": 2.21484375e-06,
1359
+ "objective/entropy": -49.555030822753906,
1360
+ "objective/kl": 22.374631881713867,
1361
+ "objective/non_score_reward": 0.0,
1362
+ "objective/rlhf_reward": 8.583580017089844,
1363
+ "objective/scores": 8.583580017089844,
1364
+ "policy/approxkl_avg": 0.3105822801589966,
1365
+ "policy/clipfrac_avg": 0.365234375,
1366
+ "policy/entropy_avg": 0.5452811121940613,
1367
+ "step": 67,
1368
+ "val/clipfrac_avg": 0.0,
1369
+ "val/num_eos_tokens": 2650,
1370
+ "val/ratio": 1.0001208782196045,
1371
+ "val/ratio_var": 8.155854629876558e-06
1372
+ },
1373
+ {
1374
+ "episode": 35328,
1375
+ "epoch": 0.15133653187114463,
1376
+ "eps": 7,
1377
+ "loss/policy_avg": 0.23417699337005615,
1378
+ "lr": 2.203125e-06,
1379
+ "objective/entropy": -46.87506866455078,
1380
+ "objective/kl": 23.203367233276367,
1381
+ "objective/non_score_reward": 0.0,
1382
+ "objective/rlhf_reward": 8.507232666015625,
1383
+ "objective/scores": 8.507232666015625,
1384
+ "policy/approxkl_avg": 0.3590773344039917,
1385
+ "policy/clipfrac_avg": 0.349609375,
1386
+ "policy/entropy_avg": 0.5592520833015442,
1387
+ "step": 68,
1388
+ "val/clipfrac_avg": 0.0,
1389
+ "val/num_eos_tokens": 2704,
1390
+ "val/ratio": 0.9996556043624878,
1391
+ "val/ratio_var": 3.690880021167686e-06
1392
+ },
1393
+ {
1394
+ "episode": 35840,
1395
+ "epoch": 0.15352981494174092,
1396
+ "eps": 7,
1397
+ "loss/policy_avg": 0.4602639079093933,
1398
+ "lr": 2.19140625e-06,
1399
+ "objective/entropy": -46.802467346191406,
1400
+ "objective/kl": 23.537273406982422,
1401
+ "objective/non_score_reward": 0.0,
1402
+ "objective/rlhf_reward": 8.526100158691406,
1403
+ "objective/scores": 8.526100158691406,
1404
+ "policy/approxkl_avg": 0.2938518822193146,
1405
+ "policy/clipfrac_avg": 0.32421875,
1406
+ "policy/entropy_avg": 0.562473714351654,
1407
+ "step": 69,
1408
+ "val/clipfrac_avg": 0.0,
1409
+ "val/num_eos_tokens": 2694,
1410
+ "val/ratio": 1.0002529621124268,
1411
+ "val/ratio_var": 4.539975634543225e-06
1412
+ },
1413
+ {
1414
+ "episode": 36352,
1415
+ "epoch": 0.15572309801233722,
1416
+ "eps": 7,
1417
+ "loss/policy_avg": 0.2672228515148163,
1418
+ "lr": 2.1796875e-06,
1419
+ "objective/entropy": -45.053016662597656,
1420
+ "objective/kl": 23.575054168701172,
1421
+ "objective/non_score_reward": 0.0,
1422
+ "objective/rlhf_reward": 8.515979766845703,
1423
+ "objective/scores": 8.515979766845703,
1424
+ "policy/approxkl_avg": 0.30573177337646484,
1425
+ "policy/clipfrac_avg": 0.322265625,
1426
+ "policy/entropy_avg": 0.5777955055236816,
1427
+ "step": 70,
1428
+ "val/clipfrac_avg": 0.0,
1429
+ "val/num_eos_tokens": 2750,
1430
+ "val/ratio": 1.0002713203430176,
1431
+ "val/ratio_var": 5.217795205680886e-06
1432
+ },
1433
+ {
1434
+ "episode": 36864,
1435
+ "epoch": 0.15791638108293352,
1436
+ "eps": 7,
1437
+ "loss/policy_avg": 0.3402022123336792,
1438
+ "lr": 2.16796875e-06,
1439
+ "objective/entropy": -42.40455627441406,
1440
+ "objective/kl": 25.648136138916016,
1441
+ "objective/non_score_reward": 0.0,
1442
+ "objective/rlhf_reward": 8.605457305908203,
1443
+ "objective/scores": 8.605457305908203,
1444
+ "policy/approxkl_avg": 0.35311511158943176,
1445
+ "policy/clipfrac_avg": 0.375,
1446
+ "policy/entropy_avg": 0.5561670064926147,
1447
+ "step": 71,
1448
+ "val/clipfrac_avg": 0.0,
1449
+ "val/num_eos_tokens": 2687,
1450
+ "val/ratio": 0.9998939037322998,
1451
+ "val/ratio_var": 4.396093117975397e-06
1452
+ },
1453
+ {
1454
+ "episode": 37376,
1455
+ "epoch": 0.1601096641535298,
1456
+ "eps": 7,
1457
+ "loss/policy_avg": 0.3977116346359253,
1458
+ "lr": 2.15625e-06,
1459
+ "objective/entropy": -42.49177169799805,
1460
+ "objective/kl": 26.084217071533203,
1461
+ "objective/non_score_reward": 0.0,
1462
+ "objective/rlhf_reward": 8.741865158081055,
1463
+ "objective/scores": 8.741865158081055,
1464
+ "policy/approxkl_avg": 0.43490105867385864,
1465
+ "policy/clipfrac_avg": 0.353515625,
1466
+ "policy/entropy_avg": 0.5506061315536499,
1467
+ "step": 72,
1468
+ "val/clipfrac_avg": 0.0,
1469
+ "val/num_eos_tokens": 2755,
1470
+ "val/ratio": 0.999513566493988,
1471
+ "val/ratio_var": 5.121092272020178e-06
1472
+ },
1473
+ {
1474
+ "episode": 37888,
1475
+ "epoch": 0.1623029472241261,
1476
+ "eps": 7,
1477
+ "loss/policy_avg": 0.26749086380004883,
1478
+ "lr": 2.14453125e-06,
1479
+ "objective/entropy": -43.96216583251953,
1480
+ "objective/kl": 25.2254638671875,
1481
+ "objective/non_score_reward": 0.0,
1482
+ "objective/rlhf_reward": 8.43868637084961,
1483
+ "objective/scores": 8.43868637084961,
1484
+ "policy/approxkl_avg": 0.37626683712005615,
1485
+ "policy/clipfrac_avg": 0.353515625,
1486
+ "policy/entropy_avg": 0.5635879039764404,
1487
+ "step": 73,
1488
+ "val/clipfrac_avg": 0.0,
1489
+ "val/num_eos_tokens": 2632,
1490
+ "val/ratio": 1.0003223419189453,
1491
+ "val/ratio_var": 4.566423740470782e-06
1492
+ },
1493
+ {
1494
+ "episode": 38400,
1495
+ "epoch": 0.1644962302947224,
1496
+ "eps": 7,
1497
+ "loss/policy_avg": 0.2991521656513214,
1498
+ "lr": 2.1328125e-06,
1499
+ "objective/entropy": -43.968116760253906,
1500
+ "objective/kl": 26.172449111938477,
1501
+ "objective/non_score_reward": 0.0,
1502
+ "objective/rlhf_reward": 8.619085311889648,
1503
+ "objective/scores": 8.619085311889648,
1504
+ "policy/approxkl_avg": 0.36614692211151123,
1505
+ "policy/clipfrac_avg": 0.361328125,
1506
+ "policy/entropy_avg": 0.5751455426216125,
1507
+ "step": 74,
1508
+ "val/clipfrac_avg": 0.0,
1509
+ "val/num_eos_tokens": 2762,
1510
+ "val/ratio": 0.9997708797454834,
1511
+ "val/ratio_var": 5.6084491006913595e-06
1512
+ },
1513
+ {
1514
+ "episode": 38912,
1515
+ "epoch": 0.1666895133653187,
1516
+ "eps": 7,
1517
+ "loss/policy_avg": 0.31739309430122375,
1518
+ "lr": 2.12109375e-06,
1519
+ "objective/entropy": -44.440555572509766,
1520
+ "objective/kl": 27.502099990844727,
1521
+ "objective/non_score_reward": 0.0,
1522
+ "objective/rlhf_reward": 8.551536560058594,
1523
+ "objective/scores": 8.551536560058594,
1524
+ "policy/approxkl_avg": 0.4447302520275116,
1525
+ "policy/clipfrac_avg": 0.37890625,
1526
+ "policy/entropy_avg": 0.5544260740280151,
1527
+ "step": 75,
1528
+ "val/clipfrac_avg": 0.0,
1529
+ "val/num_eos_tokens": 2669,
1530
+ "val/ratio": 0.9993975162506104,
1531
+ "val/ratio_var": 4.979194272891618e-06
1532
+ },
1533
+ {
1534
+ "episode": 39424,
1535
+ "epoch": 0.168882796435915,
1536
+ "eps": 7,
1537
+ "loss/policy_avg": 0.2567436695098877,
1538
+ "lr": 2.109375e-06,
1539
+ "objective/entropy": -43.823219299316406,
1540
+ "objective/kl": 26.728805541992188,
1541
+ "objective/non_score_reward": 0.0,
1542
+ "objective/rlhf_reward": 8.652971267700195,
1543
+ "objective/scores": 8.652971267700195,
1544
+ "policy/approxkl_avg": 0.3301650285720825,
1545
+ "policy/clipfrac_avg": 0.36328125,
1546
+ "policy/entropy_avg": 0.5270683765411377,
1547
+ "step": 76,
1548
+ "val/clipfrac_avg": 0.0,
1549
+ "val/num_eos_tokens": 2688,
1550
+ "val/ratio": 1.0002834796905518,
1551
+ "val/ratio_var": 4.673919193010079e-06
1552
+ },
1553
+ {
1554
+ "episode": 39936,
1555
+ "epoch": 0.1710760795065113,
1556
+ "eps": 7,
1557
+ "loss/policy_avg": 0.4975939393043518,
1558
+ "lr": 2.09765625e-06,
1559
+ "objective/entropy": -40.26116943359375,
1560
+ "objective/kl": 28.93381690979004,
1561
+ "objective/non_score_reward": 0.0,
1562
+ "objective/rlhf_reward": 8.432289123535156,
1563
+ "objective/scores": 8.432289123535156,
1564
+ "policy/approxkl_avg": 0.5036437511444092,
1565
+ "policy/clipfrac_avg": 0.380859375,
1566
+ "policy/entropy_avg": 0.5689441561698914,
1567
+ "step": 77,
1568
+ "val/clipfrac_avg": 0.0,
1569
+ "val/num_eos_tokens": 2716,
1570
+ "val/ratio": 0.9997487664222717,
1571
+ "val/ratio_var": 5.711378889827756e-06
1572
+ },
1573
+ {
1574
+ "episode": 40448,
1575
+ "epoch": 0.17326936257710762,
1576
+ "eps": 7,
1577
+ "loss/policy_avg": 0.2763368487358093,
1578
+ "lr": 2.0859375e-06,
1579
+ "objective/entropy": -45.21563720703125,
1580
+ "objective/kl": 27.39434242248535,
1581
+ "objective/non_score_reward": 0.0,
1582
+ "objective/rlhf_reward": 8.653337478637695,
1583
+ "objective/scores": 8.653337478637695,
1584
+ "policy/approxkl_avg": 0.2975656986236572,
1585
+ "policy/clipfrac_avg": 0.37890625,
1586
+ "policy/entropy_avg": 0.5368204116821289,
1587
+ "step": 78,
1588
+ "val/clipfrac_avg": 0.0,
1589
+ "val/num_eos_tokens": 2757,
1590
+ "val/ratio": 1.0003888607025146,
1591
+ "val/ratio_var": 3.7864729165448807e-06
1592
+ },
1593
+ {
1594
+ "episode": 40960,
1595
+ "epoch": 0.17546264564770392,
1596
+ "eps": 7,
1597
+ "loss/policy_avg": 0.4642253518104553,
1598
+ "lr": 2.07421875e-06,
1599
+ "objective/entropy": -41.11317825317383,
1600
+ "objective/kl": 27.240459442138672,
1601
+ "objective/non_score_reward": 0.0,
1602
+ "objective/rlhf_reward": 8.210575103759766,
1603
+ "objective/scores": 8.210575103759766,
1604
+ "policy/approxkl_avg": 0.4027579426765442,
1605
+ "policy/clipfrac_avg": 0.375,
1606
+ "policy/entropy_avg": 0.5595495700836182,
1607
+ "step": 79,
1608
+ "val/clipfrac_avg": 0.0,
1609
+ "val/num_eos_tokens": 2787,
1610
+ "val/ratio": 1.0000183582305908,
1611
+ "val/ratio_var": 3.993025529780425e-06
1612
+ },
1613
+ {
1614
+ "episode": 41472,
1615
+ "epoch": 0.17765592871830022,
1616
+ "eps": 7,
1617
+ "loss/policy_avg": 0.24956947565078735,
1618
+ "lr": 2.0625e-06,
1619
+ "objective/entropy": -40.720481872558594,
1620
+ "objective/kl": 28.056228637695312,
1621
+ "objective/non_score_reward": 0.0,
1622
+ "objective/rlhf_reward": 8.479925155639648,
1623
+ "objective/scores": 8.479925155639648,
1624
+ "policy/approxkl_avg": 0.31501132249832153,
1625
+ "policy/clipfrac_avg": 0.36328125,
1626
+ "policy/entropy_avg": 0.5648977756500244,
1627
+ "step": 80,
1628
+ "val/clipfrac_avg": 0.0,
1629
+ "val/num_eos_tokens": 2724,
1630
+ "val/ratio": 0.9998843669891357,
1631
+ "val/ratio_var": 4.445482773007825e-06
1632
+ },
1633
+ {
1634
+ "episode": 41984,
1635
+ "epoch": 0.1798492117888965,
1636
+ "eps": 7,
1637
+ "loss/policy_avg": 0.29195988178253174,
1638
+ "lr": 2.05078125e-06,
1639
+ "objective/entropy": -43.005149841308594,
1640
+ "objective/kl": 26.689315795898438,
1641
+ "objective/non_score_reward": 0.0,
1642
+ "objective/rlhf_reward": 8.50869083404541,
1643
+ "objective/scores": 8.50869083404541,
1644
+ "policy/approxkl_avg": 0.3312462866306305,
1645
+ "policy/clipfrac_avg": 0.34375,
1646
+ "policy/entropy_avg": 0.5605291724205017,
1647
+ "step": 81,
1648
+ "val/clipfrac_avg": 0.0,
1649
+ "val/num_eos_tokens": 2891,
1650
+ "val/ratio": 0.9999593496322632,
1651
+ "val/ratio_var": 3.4600634535308927e-06
1652
+ },
1653
+ {
1654
+ "episode": 42496,
1655
+ "epoch": 0.1820424948594928,
1656
+ "eps": 7,
1657
+ "loss/policy_avg": 0.22648774087429047,
1658
+ "lr": 2.0390625e-06,
1659
+ "objective/entropy": -44.053279876708984,
1660
+ "objective/kl": 26.523632049560547,
1661
+ "objective/non_score_reward": 0.0,
1662
+ "objective/rlhf_reward": 8.40163516998291,
1663
+ "objective/scores": 8.40163516998291,
1664
+ "policy/approxkl_avg": 0.33716368675231934,
1665
+ "policy/clipfrac_avg": 0.33984375,
1666
+ "policy/entropy_avg": 0.5656260251998901,
1667
+ "step": 82,
1668
+ "val/clipfrac_avg": 0.0,
1669
+ "val/num_eos_tokens": 2995,
1670
+ "val/ratio": 0.9996054172515869,
1671
+ "val/ratio_var": 4.090476068085991e-06
1672
+ },
1673
+ {
1674
+ "episode": 43008,
1675
+ "epoch": 0.1842357779300891,
1676
+ "eps": 7,
1677
+ "loss/policy_avg": 0.4572581648826599,
1678
+ "lr": 2.02734375e-06,
1679
+ "objective/entropy": -40.501346588134766,
1680
+ "objective/kl": 26.860363006591797,
1681
+ "objective/non_score_reward": 0.0,
1682
+ "objective/rlhf_reward": 8.303882598876953,
1683
+ "objective/scores": 8.303882598876953,
1684
+ "policy/approxkl_avg": 0.350925087928772,
1685
+ "policy/clipfrac_avg": 0.373046875,
1686
+ "policy/entropy_avg": 0.557375431060791,
1687
+ "step": 83,
1688
+ "val/clipfrac_avg": 0.0,
1689
+ "val/num_eos_tokens": 3036,
1690
+ "val/ratio": 0.9999963641166687,
1691
+ "val/ratio_var": 4.392126811580965e-06
1692
+ },
1693
+ {
1694
+ "episode": 43520,
1695
+ "epoch": 0.1864290610006854,
1696
+ "eps": 7,
1697
+ "loss/policy_avg": 0.2703511714935303,
1698
+ "lr": 2.015625e-06,
1699
+ "objective/entropy": -38.586124420166016,
1700
+ "objective/kl": 27.219919204711914,
1701
+ "objective/non_score_reward": 0.0,
1702
+ "objective/rlhf_reward": 8.605122566223145,
1703
+ "objective/scores": 8.605122566223145,
1704
+ "policy/approxkl_avg": 0.32834964990615845,
1705
+ "policy/clipfrac_avg": 0.390625,
1706
+ "policy/entropy_avg": 0.5444257259368896,
1707
+ "step": 84,
1708
+ "val/clipfrac_avg": 0.0,
1709
+ "val/num_eos_tokens": 2940,
1710
+ "val/ratio": 1.0001357793807983,
1711
+ "val/ratio_var": 4.64254026155686e-06
1712
+ },
1713
+ {
1714
+ "episode": 44032,
1715
+ "epoch": 0.1886223440712817,
1716
+ "eps": 7,
1717
+ "loss/policy_avg": 0.2800593972206116,
1718
+ "lr": 2.00390625e-06,
1719
+ "objective/entropy": -46.36981964111328,
1720
+ "objective/kl": 25.693588256835938,
1721
+ "objective/non_score_reward": 0.0,
1722
+ "objective/rlhf_reward": 8.409098625183105,
1723
+ "objective/scores": 8.409098625183105,
1724
+ "policy/approxkl_avg": 0.3142682909965515,
1725
+ "policy/clipfrac_avg": 0.37890625,
1726
+ "policy/entropy_avg": 0.5546380281448364,
1727
+ "step": 85,
1728
+ "val/clipfrac_avg": 0.0,
1729
+ "val/num_eos_tokens": 3223,
1730
+ "val/ratio": 1.0001696348190308,
1731
+ "val/ratio_var": 4.388642537378473e-06
1732
+ },
1733
+ {
1734
+ "episode": 44544,
1735
+ "epoch": 0.190815627141878,
1736
+ "eps": 7,
1737
+ "loss/policy_avg": 0.2657047212123871,
1738
+ "lr": 1.9921875e-06,
1739
+ "objective/entropy": -45.28251647949219,
1740
+ "objective/kl": 25.583412170410156,
1741
+ "objective/non_score_reward": 0.0,
1742
+ "objective/rlhf_reward": 8.682401657104492,
1743
+ "objective/scores": 8.682401657104492,
1744
+ "policy/approxkl_avg": 0.2970484495162964,
1745
+ "policy/clipfrac_avg": 0.341796875,
1746
+ "policy/entropy_avg": 0.5628992319107056,
1747
+ "step": 86,
1748
+ "val/clipfrac_avg": 0.0,
1749
+ "val/num_eos_tokens": 3049,
1750
+ "val/ratio": 1.0001099109649658,
1751
+ "val/ratio_var": 3.3909404919540975e-06
1752
+ },
1753
+ {
1754
+ "episode": 45056,
1755
+ "epoch": 0.1930089102124743,
1756
+ "eps": 7,
1757
+ "loss/policy_avg": 0.31868281960487366,
1758
+ "lr": 1.98046875e-06,
1759
+ "objective/entropy": -44.419532775878906,
1760
+ "objective/kl": 25.884258270263672,
1761
+ "objective/non_score_reward": 0.0,
1762
+ "objective/rlhf_reward": 8.452953338623047,
1763
+ "objective/scores": 8.452953338623047,
1764
+ "policy/approxkl_avg": 0.2683827877044678,
1765
+ "policy/clipfrac_avg": 0.375,
1766
+ "policy/entropy_avg": 0.5685138702392578,
1767
+ "step": 87,
1768
+ "val/clipfrac_avg": 0.0,
1769
+ "val/num_eos_tokens": 3113,
1770
+ "val/ratio": 1.0001598596572876,
1771
+ "val/ratio_var": 3.914261469617486e-06
1772
+ },
1773
+ {
1774
+ "episode": 45568,
1775
+ "epoch": 0.1952021932830706,
1776
+ "eps": 7,
1777
+ "loss/policy_avg": 0.224508136510849,
1778
+ "lr": 1.96875e-06,
1779
+ "objective/entropy": -48.27018737792969,
1780
+ "objective/kl": 24.89946174621582,
1781
+ "objective/non_score_reward": 0.0,
1782
+ "objective/rlhf_reward": 8.871210098266602,
1783
+ "objective/scores": 8.871210098266602,
1784
+ "policy/approxkl_avg": 0.20233646035194397,
1785
+ "policy/clipfrac_avg": 0.361328125,
1786
+ "policy/entropy_avg": 0.5470612645149231,
1787
+ "step": 88,
1788
+ "val/clipfrac_avg": 0.0,
1789
+ "val/num_eos_tokens": 3254,
1790
+ "val/ratio": 1.0006496906280518,
1791
+ "val/ratio_var": 3.585116246540565e-06
1792
+ },
1793
+ {
1794
+ "episode": 46080,
1795
+ "epoch": 0.1973954763536669,
1796
+ "eps": 7,
1797
+ "loss/policy_avg": 0.17881432175636292,
1798
+ "lr": 1.95703125e-06,
1799
+ "objective/entropy": -39.60390853881836,
1800
+ "objective/kl": 26.777809143066406,
1801
+ "objective/non_score_reward": 0.0,
1802
+ "objective/rlhf_reward": 8.489433288574219,
1803
+ "objective/scores": 8.489433288574219,
1804
+ "policy/approxkl_avg": 0.30483895540237427,
1805
+ "policy/clipfrac_avg": 0.3828125,
1806
+ "policy/entropy_avg": 0.5878047943115234,
1807
+ "step": 89,
1808
+ "val/clipfrac_avg": 0.0,
1809
+ "val/num_eos_tokens": 3105,
1810
+ "val/ratio": 1.0000858306884766,
1811
+ "val/ratio_var": 4.914289092994295e-06
1812
+ },
1813
+ {
1814
+ "episode": 46592,
1815
+ "epoch": 0.19958875942426318,
1816
+ "eps": 7,
1817
+ "loss/policy_avg": 0.33269819617271423,
1818
+ "lr": 1.9453125e-06,
1819
+ "objective/entropy": -42.723388671875,
1820
+ "objective/kl": 26.804061889648438,
1821
+ "objective/non_score_reward": 0.0,
1822
+ "objective/rlhf_reward": 8.605770111083984,
1823
+ "objective/scores": 8.605770111083984,
1824
+ "policy/approxkl_avg": 0.30788183212280273,
1825
+ "policy/clipfrac_avg": 0.34765625,
1826
+ "policy/entropy_avg": 0.557858407497406,
1827
+ "step": 90,
1828
+ "val/clipfrac_avg": 0.0,
1829
+ "val/num_eos_tokens": 3024,
1830
+ "val/ratio": 1.0000172853469849,
1831
+ "val/ratio_var": 3.867803116008872e-06
1832
+ },
1833
+ {
1834
+ "episode": 47104,
1835
+ "epoch": 0.20178204249485948,
1836
+ "eps": 7,
1837
+ "loss/policy_avg": 0.2373349964618683,
1838
+ "lr": 1.93359375e-06,
1839
+ "objective/entropy": -41.628440856933594,
1840
+ "objective/kl": 26.94710350036621,
1841
+ "objective/non_score_reward": 0.0,
1842
+ "objective/rlhf_reward": 8.717164039611816,
1843
+ "objective/scores": 8.717164039611816,
1844
+ "policy/approxkl_avg": 0.28812965750694275,
1845
+ "policy/clipfrac_avg": 0.38671875,
1846
+ "policy/entropy_avg": 0.5584580302238464,
1847
+ "step": 91,
1848
+ "val/clipfrac_avg": 0.0,
1849
+ "val/num_eos_tokens": 3233,
1850
+ "val/ratio": 0.9999223351478577,
1851
+ "val/ratio_var": 3.552548605512129e-06
1852
+ },
1853
+ {
1854
+ "episode": 47616,
1855
+ "epoch": 0.20397532556545578,
1856
+ "eps": 7,
1857
+ "loss/policy_avg": 0.46636855602264404,
1858
+ "lr": 1.921875e-06,
1859
+ "objective/entropy": -42.61951446533203,
1860
+ "objective/kl": 27.169050216674805,
1861
+ "objective/non_score_reward": 0.0,
1862
+ "objective/rlhf_reward": 8.583017349243164,
1863
+ "objective/scores": 8.583017349243164,
1864
+ "policy/approxkl_avg": 0.3408641815185547,
1865
+ "policy/clipfrac_avg": 0.33984375,
1866
+ "policy/entropy_avg": 0.56282639503479,
1867
+ "step": 92,
1868
+ "val/clipfrac_avg": 0.0,
1869
+ "val/num_eos_tokens": 3154,
1870
+ "val/ratio": 0.9996733069419861,
1871
+ "val/ratio_var": 2.5412805371161085e-06
1872
+ },
1873
+ {
1874
+ "episode": 48128,
1875
+ "epoch": 0.2061686086360521,
1876
+ "eps": 7,
1877
+ "loss/policy_avg": 0.33910614252090454,
1878
+ "lr": 1.91015625e-06,
1879
+ "objective/entropy": -43.72690200805664,
1880
+ "objective/kl": 27.0749568939209,
1881
+ "objective/non_score_reward": 0.0,
1882
+ "objective/rlhf_reward": 8.784333229064941,
1883
+ "objective/scores": 8.784333229064941,
1884
+ "policy/approxkl_avg": 0.3085598945617676,
1885
+ "policy/clipfrac_avg": 0.35546875,
1886
+ "policy/entropy_avg": 0.5553750991821289,
1887
+ "step": 93,
1888
+ "val/clipfrac_avg": 0.0,
1889
+ "val/num_eos_tokens": 3186,
1890
+ "val/ratio": 0.9999817609786987,
1891
+ "val/ratio_var": 3.2739924336055992e-06
1892
+ },
1893
+ {
1894
+ "episode": 48640,
1895
+ "epoch": 0.2083618917066484,
1896
+ "eps": 7,
1897
+ "loss/policy_avg": 0.3540273904800415,
1898
+ "lr": 1.8984375e-06,
1899
+ "objective/entropy": -41.448604583740234,
1900
+ "objective/kl": 27.778303146362305,
1901
+ "objective/non_score_reward": 0.0,
1902
+ "objective/rlhf_reward": 8.456501007080078,
1903
+ "objective/scores": 8.456501007080078,
1904
+ "policy/approxkl_avg": 0.33497947454452515,
1905
+ "policy/clipfrac_avg": 0.345703125,
1906
+ "policy/entropy_avg": 0.5644186735153198,
1907
+ "step": 94,
1908
+ "val/clipfrac_avg": 0.0,
1909
+ "val/num_eos_tokens": 3182,
1910
+ "val/ratio": 0.9999628663063049,
1911
+ "val/ratio_var": 4.411289864947321e-06
1912
+ },
1913
+ {
1914
+ "episode": 49152,
1915
+ "epoch": 0.2105551747772447,
1916
+ "eps": 7,
1917
+ "loss/policy_avg": 0.4336338937282562,
1918
+ "lr": 1.8867187500000001e-06,
1919
+ "objective/entropy": -41.28203582763672,
1920
+ "objective/kl": 27.56344985961914,
1921
+ "objective/non_score_reward": 0.0,
1922
+ "objective/rlhf_reward": 8.71638011932373,
1923
+ "objective/scores": 8.71638011932373,
1924
+ "policy/approxkl_avg": 0.32576072216033936,
1925
+ "policy/clipfrac_avg": 0.365234375,
1926
+ "policy/entropy_avg": 0.5579572916030884,
1927
+ "step": 95,
1928
+ "val/clipfrac_avg": 0.0,
1929
+ "val/num_eos_tokens": 3167,
1930
+ "val/ratio": 0.9997182488441467,
1931
+ "val/ratio_var": 3.5568716612033313e-06
1932
+ },
1933
+ {
1934
+ "episode": 49664,
1935
+ "epoch": 0.212748457847841,
1936
+ "eps": 7,
1937
+ "loss/policy_avg": 0.37640368938446045,
1938
+ "lr": 1.875e-06,
1939
+ "objective/entropy": -39.93994140625,
1940
+ "objective/kl": 28.659408569335938,
1941
+ "objective/non_score_reward": 0.0,
1942
+ "objective/rlhf_reward": 8.587747573852539,
1943
+ "objective/scores": 8.587747573852539,
1944
+ "policy/approxkl_avg": 0.2930169403553009,
1945
+ "policy/clipfrac_avg": 0.38671875,
1946
+ "policy/entropy_avg": 0.5753656625747681,
1947
+ "step": 96,
1948
+ "val/clipfrac_avg": 0.0,
1949
+ "val/num_eos_tokens": 3320,
1950
+ "val/ratio": 0.9999222755432129,
1951
+ "val/ratio_var": 2.8392282729328144e-06
1952
+ },
1953
+ {
1954
+ "episode": 50176,
1955
+ "epoch": 0.2149417409184373,
1956
+ "eps": 7,
1957
+ "loss/policy_avg": 0.3704265356063843,
1958
+ "lr": 1.86328125e-06,
1959
+ "objective/entropy": -39.154541015625,
1960
+ "objective/kl": 28.530101776123047,
1961
+ "objective/non_score_reward": 0.0,
1962
+ "objective/rlhf_reward": 8.800564765930176,
1963
+ "objective/scores": 8.800564765930176,
1964
+ "policy/approxkl_avg": 0.30916929244995117,
1965
+ "policy/clipfrac_avg": 0.333984375,
1966
+ "policy/entropy_avg": 0.5570034384727478,
1967
+ "step": 97,
1968
+ "val/clipfrac_avg": 0.0,
1969
+ "val/num_eos_tokens": 3072,
1970
+ "val/ratio": 1.0001652240753174,
1971
+ "val/ratio_var": 4.666152108256938e-06
1972
+ },
1973
+ {
1974
+ "episode": 50688,
1975
+ "epoch": 0.2171350239890336,
1976
+ "eps": 7,
1977
+ "loss/policy_avg": 0.33340394496917725,
1978
+ "lr": 1.8515625000000001e-06,
1979
+ "objective/entropy": -42.39177703857422,
1980
+ "objective/kl": 26.74437141418457,
1981
+ "objective/non_score_reward": 0.0,
1982
+ "objective/rlhf_reward": 8.953100204467773,
1983
+ "objective/scores": 8.953100204467773,
1984
+ "policy/approxkl_avg": 0.25182923674583435,
1985
+ "policy/clipfrac_avg": 0.34765625,
1986
+ "policy/entropy_avg": 0.5791052579879761,
1987
+ "step": 98,
1988
+ "val/clipfrac_avg": 0.0,
1989
+ "val/num_eos_tokens": 3253,
1990
+ "val/ratio": 1.0006077289581299,
1991
+ "val/ratio_var": 4.077707671967801e-06
1992
+ },
1993
+ {
1994
+ "episode": 51200,
1995
+ "epoch": 0.21932830705962988,
1996
+ "eps": 7,
1997
+ "loss/policy_avg": 0.3429536521434784,
1998
+ "lr": 1.83984375e-06,
1999
+ "objective/entropy": -42.56659698486328,
2000
+ "objective/kl": 27.317535400390625,
2001
+ "objective/non_score_reward": 0.0,
2002
+ "objective/rlhf_reward": 8.745502471923828,
2003
+ "objective/scores": 8.745502471923828,
2004
+ "policy/approxkl_avg": 0.3268476724624634,
2005
+ "policy/clipfrac_avg": 0.34375,
2006
+ "policy/entropy_avg": 0.5513091087341309,
2007
+ "step": 99,
2008
+ "val/clipfrac_avg": 0.0,
2009
+ "val/num_eos_tokens": 3308,
2010
+ "val/ratio": 1.0001206398010254,
2011
+ "val/ratio_var": 6.498310995084466e-06
2012
+ },
2013
+ {
2014
+ "episode": 51712,
2015
+ "epoch": 0.22152159013022618,
2016
+ "eps": 7,
2017
+ "loss/policy_avg": 0.18512360751628876,
2018
+ "lr": 1.828125e-06,
2019
+ "objective/entropy": -39.8660888671875,
2020
+ "objective/kl": 28.532289505004883,
2021
+ "objective/non_score_reward": 0.0,
2022
+ "objective/rlhf_reward": 8.849493026733398,
2023
+ "objective/scores": 8.849493026733398,
2024
+ "policy/approxkl_avg": 0.28483685851097107,
2025
+ "policy/clipfrac_avg": 0.40625,
2026
+ "policy/entropy_avg": 0.567420244216919,
2027
+ "step": 100,
2028
+ "val/clipfrac_avg": 0.0,
2029
+ "val/num_eos_tokens": 3124,
2030
+ "val/ratio": 1.0002485513687134,
2031
+ "val/ratio_var": 3.6835951959801605e-06
2032
+ },
2033
+ {
2034
+ "episode": 52224,
2035
+ "epoch": 0.22371487320082248,
2036
+ "eps": 7,
2037
+ "loss/policy_avg": 0.28863823413848877,
2038
+ "lr": 1.81640625e-06,
2039
+ "objective/entropy": -43.047508239746094,
2040
+ "objective/kl": 27.43829345703125,
2041
+ "objective/non_score_reward": 0.0,
2042
+ "objective/rlhf_reward": 9.060672760009766,
2043
+ "objective/scores": 9.060672760009766,
2044
+ "policy/approxkl_avg": 0.3050522208213806,
2045
+ "policy/clipfrac_avg": 0.33984375,
2046
+ "policy/entropy_avg": 0.5633636713027954,
2047
+ "step": 101,
2048
+ "val/clipfrac_avg": 0.0,
2049
+ "val/num_eos_tokens": 3185,
2050
+ "val/ratio": 1.0002562999725342,
2051
+ "val/ratio_var": 4.66149003841565e-06
2052
+ },
2053
+ {
2054
+ "episode": 52736,
2055
+ "epoch": 0.22590815627141878,
2056
+ "eps": 7,
2057
+ "loss/policy_avg": 0.27026695013046265,
2058
+ "lr": 1.8046875e-06,
2059
+ "objective/entropy": -44.638450622558594,
2060
+ "objective/kl": 26.97273826599121,
2061
+ "objective/non_score_reward": 0.0,
2062
+ "objective/rlhf_reward": 8.926921844482422,
2063
+ "objective/scores": 8.926921844482422,
2064
+ "policy/approxkl_avg": 0.26449355483055115,
2065
+ "policy/clipfrac_avg": 0.337890625,
2066
+ "policy/entropy_avg": 0.5377113223075867,
2067
+ "step": 102,
2068
+ "val/clipfrac_avg": 0.0,
2069
+ "val/num_eos_tokens": 2916,
2070
+ "val/ratio": 0.9999263882637024,
2071
+ "val/ratio_var": 2.791050519590499e-06
2072
+ },
2073
+ {
2074
+ "episode": 53248,
2075
+ "epoch": 0.22810143934201507,
2076
+ "eps": 7,
2077
+ "loss/policy_avg": 0.31279855966567993,
2078
+ "lr": 1.79296875e-06,
2079
+ "objective/entropy": -39.981178283691406,
2080
+ "objective/kl": 28.416162490844727,
2081
+ "objective/non_score_reward": 0.0,
2082
+ "objective/rlhf_reward": 8.903759956359863,
2083
+ "objective/scores": 8.903759956359863,
2084
+ "policy/approxkl_avg": 0.3121040463447571,
2085
+ "policy/clipfrac_avg": 0.35546875,
2086
+ "policy/entropy_avg": 0.5868556499481201,
2087
+ "step": 103,
2088
+ "val/clipfrac_avg": 0.0,
2089
+ "val/num_eos_tokens": 3269,
2090
+ "val/ratio": 0.9997861385345459,
2091
+ "val/ratio_var": 4.62020989289158e-06
2092
+ },
2093
+ {
2094
+ "episode": 53760,
2095
+ "epoch": 0.23029472241261137,
2096
+ "eps": 7,
2097
+ "loss/policy_avg": 0.20185165107250214,
2098
+ "lr": 1.78125e-06,
2099
+ "objective/entropy": -38.68260955810547,
2100
+ "objective/kl": 28.745805740356445,
2101
+ "objective/non_score_reward": 0.0,
2102
+ "objective/rlhf_reward": 8.793964385986328,
2103
+ "objective/scores": 8.793964385986328,
2104
+ "policy/approxkl_avg": 0.3357839286327362,
2105
+ "policy/clipfrac_avg": 0.37890625,
2106
+ "policy/entropy_avg": 0.5680782198905945,
2107
+ "step": 104,
2108
+ "val/clipfrac_avg": 0.0,
2109
+ "val/num_eos_tokens": 3203,
2110
+ "val/ratio": 0.9997376203536987,
2111
+ "val/ratio_var": 3.0809435429546284e-06
2112
+ },
2113
+ {
2114
+ "episode": 54272,
2115
+ "epoch": 0.23248800548320767,
2116
+ "eps": 7,
2117
+ "loss/policy_avg": 0.27374643087387085,
2118
+ "lr": 1.76953125e-06,
2119
+ "objective/entropy": -41.97970199584961,
2120
+ "objective/kl": 27.869659423828125,
2121
+ "objective/non_score_reward": 0.0,
2122
+ "objective/rlhf_reward": 8.844221115112305,
2123
+ "objective/scores": 8.844221115112305,
2124
+ "policy/approxkl_avg": 0.2983897030353546,
2125
+ "policy/clipfrac_avg": 0.388671875,
2126
+ "policy/entropy_avg": 0.569934606552124,
2127
+ "step": 105,
2128
+ "val/clipfrac_avg": 0.0,
2129
+ "val/num_eos_tokens": 3125,
2130
+ "val/ratio": 0.9998430013656616,
2131
+ "val/ratio_var": 4.563687980407849e-06
2132
+ },
2133
+ {
2134
+ "episode": 54784,
2135
+ "epoch": 0.23468128855380396,
2136
+ "eps": 7,
2137
+ "loss/policy_avg": 0.28290292620658875,
2138
+ "lr": 1.7578125e-06,
2139
+ "objective/entropy": -41.56986618041992,
2140
+ "objective/kl": 27.95354652404785,
2141
+ "objective/non_score_reward": 0.0,
2142
+ "objective/rlhf_reward": 8.726874351501465,
2143
+ "objective/scores": 8.726874351501465,
2144
+ "policy/approxkl_avg": 0.35987648367881775,
2145
+ "policy/clipfrac_avg": 0.37109375,
2146
+ "policy/entropy_avg": 0.5299930572509766,
2147
+ "step": 106,
2148
+ "val/clipfrac_avg": 0.0,
2149
+ "val/num_eos_tokens": 2941,
2150
+ "val/ratio": 0.9999336004257202,
2151
+ "val/ratio_var": 5.3527737691183574e-06
2152
+ },
2153
+ {
2154
+ "episode": 55296,
2155
+ "epoch": 0.2368745716244003,
2156
+ "eps": 7,
2157
+ "loss/policy_avg": 0.2681512236595154,
2158
+ "lr": 1.74609375e-06,
2159
+ "objective/entropy": -43.27360916137695,
2160
+ "objective/kl": 27.097171783447266,
2161
+ "objective/non_score_reward": 0.0,
2162
+ "objective/rlhf_reward": 9.131484031677246,
2163
+ "objective/scores": 9.131484031677246,
2164
+ "policy/approxkl_avg": 0.27127525210380554,
2165
+ "policy/clipfrac_avg": 0.34765625,
2166
+ "policy/entropy_avg": 0.5349740982055664,
2167
+ "step": 107,
2168
+ "val/clipfrac_avg": 0.0,
2169
+ "val/num_eos_tokens": 3110,
2170
+ "val/ratio": 1.0000276565551758,
2171
+ "val/ratio_var": 3.7708189211116405e-06
2172
+ },
2173
+ {
2174
+ "episode": 55808,
2175
+ "epoch": 0.23906785469499658,
2176
+ "eps": 7,
2177
+ "loss/policy_avg": 0.3031000792980194,
2178
+ "lr": 1.734375e-06,
2179
+ "objective/entropy": -41.32918930053711,
2180
+ "objective/kl": 28.28663444519043,
2181
+ "objective/non_score_reward": 0.0,
2182
+ "objective/rlhf_reward": 8.978072166442871,
2183
+ "objective/scores": 8.978072166442871,
2184
+ "policy/approxkl_avg": 0.28883644938468933,
2185
+ "policy/clipfrac_avg": 0.384765625,
2186
+ "policy/entropy_avg": 0.5501591563224792,
2187
+ "step": 108,
2188
+ "val/clipfrac_avg": 0.0,
2189
+ "val/num_eos_tokens": 3257,
2190
+ "val/ratio": 1.0002058744430542,
2191
+ "val/ratio_var": 2.6081083888129797e-06
2192
+ },
2193
+ {
2194
+ "episode": 56320,
2195
+ "epoch": 0.24126113776559288,
2196
+ "eps": 7,
2197
+ "loss/policy_avg": 0.22930988669395447,
2198
+ "lr": 1.72265625e-06,
2199
+ "objective/entropy": -44.260066986083984,
2200
+ "objective/kl": 27.051563262939453,
2201
+ "objective/non_score_reward": 0.0,
2202
+ "objective/rlhf_reward": 9.064173698425293,
2203
+ "objective/scores": 9.064173698425293,
2204
+ "policy/approxkl_avg": 0.25021883845329285,
2205
+ "policy/clipfrac_avg": 0.353515625,
2206
+ "policy/entropy_avg": 0.5275313854217529,
2207
+ "step": 109,
2208
+ "val/clipfrac_avg": 0.0,
2209
+ "val/num_eos_tokens": 3274,
2210
+ "val/ratio": 1.0002585649490356,
2211
+ "val/ratio_var": 4.10879465562175e-06
2212
+ },
2213
+ {
2214
+ "episode": 56832,
2215
+ "epoch": 0.24345442083618918,
2216
+ "eps": 7,
2217
+ "loss/policy_avg": 0.32286208868026733,
2218
+ "lr": 1.7109375e-06,
2219
+ "objective/entropy": -44.09670639038086,
2220
+ "objective/kl": 27.300418853759766,
2221
+ "objective/non_score_reward": 0.0,
2222
+ "objective/rlhf_reward": 8.928081512451172,
2223
+ "objective/scores": 8.928081512451172,
2224
+ "policy/approxkl_avg": 0.2881784737110138,
2225
+ "policy/clipfrac_avg": 0.337890625,
2226
+ "policy/entropy_avg": 0.5374830961227417,
2227
+ "step": 110,
2228
+ "val/clipfrac_avg": 0.0,
2229
+ "val/num_eos_tokens": 3310,
2230
+ "val/ratio": 1.000059962272644,
2231
+ "val/ratio_var": 4.859982709604083e-06
2232
+ },
2233
+ {
2234
+ "episode": 57344,
2235
+ "epoch": 0.24564770390678548,
2236
+ "eps": 7,
2237
+ "loss/policy_avg": 0.1695113182067871,
2238
+ "lr": 1.69921875e-06,
2239
+ "objective/entropy": -42.695960998535156,
2240
+ "objective/kl": 26.841167449951172,
2241
+ "objective/non_score_reward": 0.0,
2242
+ "objective/rlhf_reward": 9.010541915893555,
2243
+ "objective/scores": 9.010541915893555,
2244
+ "policy/approxkl_avg": 0.3248569369316101,
2245
+ "policy/clipfrac_avg": 0.3828125,
2246
+ "policy/entropy_avg": 0.557858943939209,
2247
+ "step": 111,
2248
+ "val/clipfrac_avg": 0.0,
2249
+ "val/num_eos_tokens": 3280,
2250
+ "val/ratio": 0.9998301267623901,
2251
+ "val/ratio_var": 3.137585736112669e-06
2252
+ },
2253
+ {
2254
+ "episode": 57856,
2255
+ "epoch": 0.24784098697738177,
2256
+ "eps": 7,
2257
+ "loss/policy_avg": 0.23961672186851501,
2258
+ "lr": 1.6875e-06,
2259
+ "objective/entropy": -41.781715393066406,
2260
+ "objective/kl": 27.928836822509766,
2261
+ "objective/non_score_reward": 0.0,
2262
+ "objective/rlhf_reward": 8.928047180175781,
2263
+ "objective/scores": 8.928047180175781,
2264
+ "policy/approxkl_avg": 0.2984079420566559,
2265
+ "policy/clipfrac_avg": 0.37109375,
2266
+ "policy/entropy_avg": 0.5693182945251465,
2267
+ "step": 112,
2268
+ "val/clipfrac_avg": 0.0,
2269
+ "val/num_eos_tokens": 3228,
2270
+ "val/ratio": 0.9999939203262329,
2271
+ "val/ratio_var": 3.22428149956977e-06
2272
+ },
2273
+ {
2274
+ "episode": 58368,
2275
+ "epoch": 0.25003427004797807,
2276
+ "eps": 7,
2277
+ "loss/policy_avg": 0.3170594573020935,
2278
+ "lr": 1.67578125e-06,
2279
+ "objective/entropy": -44.25893783569336,
2280
+ "objective/kl": 27.417301177978516,
2281
+ "objective/non_score_reward": 0.0,
2282
+ "objective/rlhf_reward": 9.042917251586914,
2283
+ "objective/scores": 9.042917251586914,
2284
+ "policy/approxkl_avg": 0.28602951765060425,
2285
+ "policy/clipfrac_avg": 0.33984375,
2286
+ "policy/entropy_avg": 0.5440371632575989,
2287
+ "step": 113,
2288
+ "val/clipfrac_avg": 0.0,
2289
+ "val/num_eos_tokens": 3082,
2290
+ "val/ratio": 1.0005700588226318,
2291
+ "val/ratio_var": 1.94922395166941e-05
2292
+ },
2293
+ {
2294
+ "episode": 58880,
2295
+ "epoch": 0.25222755311857437,
2296
+ "eps": 7,
2297
+ "loss/policy_avg": 0.40132516622543335,
2298
+ "lr": 1.6640625e-06,
2299
+ "objective/entropy": -40.10181427001953,
2300
+ "objective/kl": 27.927541732788086,
2301
+ "objective/non_score_reward": 0.0,
2302
+ "objective/rlhf_reward": 8.806680679321289,
2303
+ "objective/scores": 8.806680679321289,
2304
+ "policy/approxkl_avg": 0.3599626421928406,
2305
+ "policy/clipfrac_avg": 0.353515625,
2306
+ "policy/entropy_avg": 0.5521677732467651,
2307
+ "step": 114,
2308
+ "val/clipfrac_avg": 0.0,
2309
+ "val/num_eos_tokens": 3332,
2310
+ "val/ratio": 0.9998911023139954,
2311
+ "val/ratio_var": 4.160545358899981e-06
2312
+ },
2313
+ {
2314
+ "episode": 59392,
2315
+ "epoch": 0.25442083618917066,
2316
+ "eps": 7,
2317
+ "loss/policy_avg": 0.3721383512020111,
2318
+ "lr": 1.6523437500000001e-06,
2319
+ "objective/entropy": -40.903629302978516,
2320
+ "objective/kl": 28.625869750976562,
2321
+ "objective/non_score_reward": 0.0,
2322
+ "objective/rlhf_reward": 8.888092994689941,
2323
+ "objective/scores": 8.888092994689941,
2324
+ "policy/approxkl_avg": 0.3267340660095215,
2325
+ "policy/clipfrac_avg": 0.34765625,
2326
+ "policy/entropy_avg": 0.5643830299377441,
2327
+ "step": 115,
2328
+ "val/clipfrac_avg": 0.0,
2329
+ "val/num_eos_tokens": 3251,
2330
+ "val/ratio": 1.0002787113189697,
2331
+ "val/ratio_var": 4.806289325642865e-06
2332
+ },
2333
+ {
2334
+ "episode": 59904,
2335
+ "epoch": 0.25661411925976696,
2336
+ "eps": 7,
2337
+ "loss/policy_avg": 0.27686673402786255,
2338
+ "lr": 1.640625e-06,
2339
+ "objective/entropy": -44.158592224121094,
2340
+ "objective/kl": 26.43977928161621,
2341
+ "objective/non_score_reward": 0.0,
2342
+ "objective/rlhf_reward": 8.87618637084961,
2343
+ "objective/scores": 8.87618637084961,
2344
+ "policy/approxkl_avg": 0.2663460671901703,
2345
+ "policy/clipfrac_avg": 0.318359375,
2346
+ "policy/entropy_avg": 0.5564490556716919,
2347
+ "step": 116,
2348
+ "val/clipfrac_avg": 0.0,
2349
+ "val/num_eos_tokens": 3377,
2350
+ "val/ratio": 0.9997222423553467,
2351
+ "val/ratio_var": 4.367329438537126e-06
2352
+ },
2353
+ {
2354
+ "episode": 60416,
2355
+ "epoch": 0.25880740233036326,
2356
+ "eps": 7,
2357
+ "loss/policy_avg": 0.25908106565475464,
2358
+ "lr": 1.62890625e-06,
2359
+ "objective/entropy": -41.13191223144531,
2360
+ "objective/kl": 27.067249298095703,
2361
+ "objective/non_score_reward": 0.0,
2362
+ "objective/rlhf_reward": 8.76617431640625,
2363
+ "objective/scores": 8.76617431640625,
2364
+ "policy/approxkl_avg": 0.2690734267234802,
2365
+ "policy/clipfrac_avg": 0.3359375,
2366
+ "policy/entropy_avg": 0.5597757697105408,
2367
+ "step": 117,
2368
+ "val/clipfrac_avg": 0.0,
2369
+ "val/num_eos_tokens": 3447,
2370
+ "val/ratio": 1.0001375675201416,
2371
+ "val/ratio_var": 3.1488286822423106e-06
2372
+ },
2373
+ {
2374
+ "episode": 60928,
2375
+ "epoch": 0.26100068540095955,
2376
+ "eps": 7,
2377
+ "loss/policy_avg": 0.3060078024864197,
2378
+ "lr": 1.6171875000000001e-06,
2379
+ "objective/entropy": -43.39402770996094,
2380
+ "objective/kl": 26.568906784057617,
2381
+ "objective/non_score_reward": 0.0,
2382
+ "objective/rlhf_reward": 9.188919067382812,
2383
+ "objective/scores": 9.188919067382812,
2384
+ "policy/approxkl_avg": 0.30087438225746155,
2385
+ "policy/clipfrac_avg": 0.345703125,
2386
+ "policy/entropy_avg": 0.543894350528717,
2387
+ "step": 118,
2388
+ "val/clipfrac_avg": 0.0,
2389
+ "val/num_eos_tokens": 3462,
2390
+ "val/ratio": 1.0000953674316406,
2391
+ "val/ratio_var": 3.770735020225402e-06
2392
+ },
2393
+ {
2394
+ "episode": 61440,
2395
+ "epoch": 0.26319396847155585,
2396
+ "eps": 7,
2397
+ "loss/policy_avg": 0.22503231465816498,
2398
+ "lr": 1.60546875e-06,
2399
+ "objective/entropy": -41.05299758911133,
2400
+ "objective/kl": 27.94589614868164,
2401
+ "objective/non_score_reward": 0.0,
2402
+ "objective/rlhf_reward": 8.894856452941895,
2403
+ "objective/scores": 8.894856452941895,
2404
+ "policy/approxkl_avg": 0.2811037003993988,
2405
+ "policy/clipfrac_avg": 0.359375,
2406
+ "policy/entropy_avg": 0.5578025579452515,
2407
+ "step": 119,
2408
+ "val/clipfrac_avg": 0.0,
2409
+ "val/num_eos_tokens": 3312,
2410
+ "val/ratio": 0.9998151063919067,
2411
+ "val/ratio_var": 3.168901230310439e-06
2412
+ },
2413
+ {
2414
+ "episode": 61952,
2415
+ "epoch": 0.26538725154215215,
2416
+ "eps": 7,
2417
+ "loss/policy_avg": 0.18785294890403748,
2418
+ "lr": 1.59375e-06,
2419
+ "objective/entropy": -44.082115173339844,
2420
+ "objective/kl": 26.886157989501953,
2421
+ "objective/non_score_reward": 0.0,
2422
+ "objective/rlhf_reward": 9.022890090942383,
2423
+ "objective/scores": 9.022890090942383,
2424
+ "policy/approxkl_avg": 0.23531067371368408,
2425
+ "policy/clipfrac_avg": 0.330078125,
2426
+ "policy/entropy_avg": 0.5334745049476624,
2427
+ "step": 120,
2428
+ "val/clipfrac_avg": 0.0,
2429
+ "val/num_eos_tokens": 3293,
2430
+ "val/ratio": 1.0002555847167969,
2431
+ "val/ratio_var": 3.1270933504856657e-06
2432
+ },
2433
+ {
2434
+ "episode": 62464,
2435
+ "epoch": 0.26758053461274844,
2436
+ "eps": 7,
2437
+ "loss/policy_avg": 0.2825835943222046,
2438
+ "lr": 1.5820312500000001e-06,
2439
+ "objective/entropy": -41.7612190246582,
2440
+ "objective/kl": 28.106294631958008,
2441
+ "objective/non_score_reward": 0.0,
2442
+ "objective/rlhf_reward": 8.90147590637207,
2443
+ "objective/scores": 8.90147590637207,
2444
+ "policy/approxkl_avg": 0.28883862495422363,
2445
+ "policy/clipfrac_avg": 0.353515625,
2446
+ "policy/entropy_avg": 0.5609524846076965,
2447
+ "step": 121,
2448
+ "val/clipfrac_avg": 0.0,
2449
+ "val/num_eos_tokens": 3501,
2450
+ "val/ratio": 0.9996764659881592,
2451
+ "val/ratio_var": 3.0721889743290376e-06
2452
+ },
2453
+ {
2454
+ "episode": 62976,
2455
+ "epoch": 0.26977381768334474,
2456
+ "eps": 7,
2457
+ "loss/policy_avg": 0.42201805114746094,
2458
+ "lr": 1.5703125e-06,
2459
+ "objective/entropy": -40.16608428955078,
2460
+ "objective/kl": 29.224674224853516,
2461
+ "objective/non_score_reward": 0.0,
2462
+ "objective/rlhf_reward": 9.230374336242676,
2463
+ "objective/scores": 9.230374336242676,
2464
+ "policy/approxkl_avg": 0.3132435083389282,
2465
+ "policy/clipfrac_avg": 0.369140625,
2466
+ "policy/entropy_avg": 0.545167863368988,
2467
+ "step": 122,
2468
+ "val/clipfrac_avg": 0.0,
2469
+ "val/num_eos_tokens": 3192,
2470
+ "val/ratio": 1.000016689300537,
2471
+ "val/ratio_var": 5.478911134559894e-06
2472
+ },
2473
+ {
2474
+ "episode": 63488,
2475
+ "epoch": 0.27196710075394104,
2476
+ "eps": 7,
2477
+ "loss/policy_avg": 0.13765352964401245,
2478
+ "lr": 1.55859375e-06,
2479
+ "objective/entropy": -38.40263366699219,
2480
+ "objective/kl": 28.912315368652344,
2481
+ "objective/non_score_reward": 0.0,
2482
+ "objective/rlhf_reward": 8.914356231689453,
2483
+ "objective/scores": 8.914356231689453,
2484
+ "policy/approxkl_avg": 0.26425743103027344,
2485
+ "policy/clipfrac_avg": 0.37890625,
2486
+ "policy/entropy_avg": 0.5654212236404419,
2487
+ "step": 123,
2488
+ "val/clipfrac_avg": 0.0,
2489
+ "val/num_eos_tokens": 3085,
2490
+ "val/ratio": 0.9999741315841675,
2491
+ "val/ratio_var": 3.0958819934312487e-06
2492
+ },
2493
+ {
2494
+ "episode": 64000,
2495
+ "epoch": 0.27416038382453733,
2496
+ "eps": 7,
2497
+ "loss/policy_avg": 0.281406968832016,
2498
+ "lr": 1.5468750000000001e-06,
2499
+ "objective/entropy": -41.14397430419922,
2500
+ "objective/kl": 27.943584442138672,
2501
+ "objective/non_score_reward": 0.0,
2502
+ "objective/rlhf_reward": 8.850857734680176,
2503
+ "objective/scores": 8.850857734680176,
2504
+ "policy/approxkl_avg": 0.26425665616989136,
2505
+ "policy/clipfrac_avg": 0.341796875,
2506
+ "policy/entropy_avg": 0.5333235263824463,
2507
+ "step": 124,
2508
+ "val/clipfrac_avg": 0.0,
2509
+ "val/num_eos_tokens": 2966,
2510
+ "val/ratio": 1.0001806020736694,
2511
+ "val/ratio_var": 4.1297735151601955e-06
2512
+ },
2513
+ {
2514
+ "episode": 64512,
2515
+ "epoch": 0.27635366689513363,
2516
+ "eps": 7,
2517
+ "loss/policy_avg": 0.3354412317276001,
2518
+ "lr": 1.53515625e-06,
2519
+ "objective/entropy": -40.96007537841797,
2520
+ "objective/kl": 28.996366500854492,
2521
+ "objective/non_score_reward": 0.0,
2522
+ "objective/rlhf_reward": 8.931587219238281,
2523
+ "objective/scores": 8.931587219238281,
2524
+ "policy/approxkl_avg": 0.2954847514629364,
2525
+ "policy/clipfrac_avg": 0.359375,
2526
+ "policy/entropy_avg": 0.5366266965866089,
2527
+ "step": 125,
2528
+ "val/clipfrac_avg": 0.0,
2529
+ "val/num_eos_tokens": 2979,
2530
+ "val/ratio": 1.000289797782898,
2531
+ "val/ratio_var": 3.790634536926518e-06
2532
+ },
2533
+ {
2534
+ "episode": 65024,
2535
+ "epoch": 0.27854694996572993,
2536
+ "eps": 7,
2537
+ "loss/policy_avg": 0.2738310396671295,
2538
+ "lr": 1.5234375e-06,
2539
+ "objective/entropy": -40.16575622558594,
2540
+ "objective/kl": 29.01610565185547,
2541
+ "objective/non_score_reward": 0.0,
2542
+ "objective/rlhf_reward": 9.160589218139648,
2543
+ "objective/scores": 9.160589218139648,
2544
+ "policy/approxkl_avg": 0.30494195222854614,
2545
+ "policy/clipfrac_avg": 0.33984375,
2546
+ "policy/entropy_avg": 0.5441122055053711,
2547
+ "step": 126,
2548
+ "val/clipfrac_avg": 0.0,
2549
+ "val/num_eos_tokens": 3075,
2550
+ "val/ratio": 0.9997348785400391,
2551
+ "val/ratio_var": 3.715499815371004e-06
2552
+ },
2553
+ {
2554
+ "episode": 65536,
2555
+ "epoch": 0.2807402330363262,
2556
+ "eps": 7,
2557
+ "loss/policy_avg": 0.3204410672187805,
2558
+ "lr": 1.5117187500000001e-06,
2559
+ "objective/entropy": -41.28175354003906,
2560
+ "objective/kl": 28.728578567504883,
2561
+ "objective/non_score_reward": 0.0,
2562
+ "objective/rlhf_reward": 9.17586612701416,
2563
+ "objective/scores": 9.17586612701416,
2564
+ "policy/approxkl_avg": 0.25520649552345276,
2565
+ "policy/clipfrac_avg": 0.375,
2566
+ "policy/entropy_avg": 0.5370270013809204,
2567
+ "step": 127,
2568
+ "val/clipfrac_avg": 0.0,
2569
+ "val/num_eos_tokens": 2918,
2570
+ "val/ratio": 1.0001736879348755,
2571
+ "val/ratio_var": 3.415733999645454e-06
2572
+ },
2573
+ {
2574
+ "episode": 66048,
2575
+ "epoch": 0.2829335161069225,
2576
+ "eps": 7,
2577
+ "loss/policy_avg": 0.23167520761489868,
2578
+ "lr": 1.5e-06,
2579
+ "objective/entropy": -40.13172912597656,
2580
+ "objective/kl": 29.34070587158203,
2581
+ "objective/non_score_reward": 0.0,
2582
+ "objective/rlhf_reward": 8.950373649597168,
2583
+ "objective/scores": 8.950373649597168,
2584
+ "policy/approxkl_avg": 0.2691437304019928,
2585
+ "policy/clipfrac_avg": 0.36328125,
2586
+ "policy/entropy_avg": 0.534875750541687,
2587
+ "step": 128,
2588
+ "val/clipfrac_avg": 0.0,
2589
+ "val/num_eos_tokens": 2974,
2590
+ "val/ratio": 0.9997478723526001,
2591
+ "val/ratio_var": 3.2285590805258835e-06
2592
+ },
2593
+ {
2594
+ "episode": 66560,
2595
+ "epoch": 0.2851267991775189,
2596
+ "eps": 7,
2597
+ "loss/policy_avg": 0.2088831514120102,
2598
+ "lr": 1.48828125e-06,
2599
+ "objective/entropy": -38.12837600708008,
2600
+ "objective/kl": 28.746782302856445,
2601
+ "objective/non_score_reward": 0.0,
2602
+ "objective/rlhf_reward": 9.036067962646484,
2603
+ "objective/scores": 9.036067962646484,
2604
+ "policy/approxkl_avg": 0.3211039900779724,
2605
+ "policy/clipfrac_avg": 0.361328125,
2606
+ "policy/entropy_avg": 0.5321791172027588,
2607
+ "step": 129,
2608
+ "val/clipfrac_avg": 0.0,
2609
+ "val/num_eos_tokens": 3088,
2610
+ "val/ratio": 1.0001753568649292,
2611
+ "val/ratio_var": 5.1817132771248e-06
2612
+ },
2613
+ {
2614
+ "episode": 67072,
2615
+ "epoch": 0.28732008224811517,
2616
+ "eps": 7,
2617
+ "loss/policy_avg": 0.2225157618522644,
2618
+ "lr": 1.4765625e-06,
2619
+ "objective/entropy": -38.43605041503906,
2620
+ "objective/kl": 30.095699310302734,
2621
+ "objective/non_score_reward": 0.0,
2622
+ "objective/rlhf_reward": 9.020343780517578,
2623
+ "objective/scores": 9.020343780517578,
2624
+ "policy/approxkl_avg": 0.29494479298591614,
2625
+ "policy/clipfrac_avg": 0.365234375,
2626
+ "policy/entropy_avg": 0.528764009475708,
2627
+ "step": 130,
2628
+ "val/clipfrac_avg": 0.0,
2629
+ "val/num_eos_tokens": 2872,
2630
+ "val/ratio": 0.9999305009841919,
2631
+ "val/ratio_var": 4.7066237129911315e-06
2632
+ },
2633
+ {
2634
+ "episode": 67584,
2635
+ "epoch": 0.28951336531871147,
2636
+ "eps": 7,
2637
+ "loss/policy_avg": 0.18626652657985687,
2638
+ "lr": 1.46484375e-06,
2639
+ "objective/entropy": -38.2808837890625,
2640
+ "objective/kl": 29.611225128173828,
2641
+ "objective/non_score_reward": 0.0,
2642
+ "objective/rlhf_reward": 9.177016258239746,
2643
+ "objective/scores": 9.177016258239746,
2644
+ "policy/approxkl_avg": 0.24960750341415405,
2645
+ "policy/clipfrac_avg": 0.37890625,
2646
+ "policy/entropy_avg": 0.5560537576675415,
2647
+ "step": 131,
2648
+ "val/clipfrac_avg": 0.0,
2649
+ "val/num_eos_tokens": 3039,
2650
+ "val/ratio": 0.9996904134750366,
2651
+ "val/ratio_var": 2.8264887532714056e-06
2652
+ },
2653
+ {
2654
+ "episode": 68096,
2655
+ "epoch": 0.29170664838930777,
2656
+ "eps": 7,
2657
+ "loss/policy_avg": 0.34811753034591675,
2658
+ "lr": 1.453125e-06,
2659
+ "objective/entropy": -37.813743591308594,
2660
+ "objective/kl": 30.11647605895996,
2661
+ "objective/non_score_reward": 0.0,
2662
+ "objective/rlhf_reward": 9.143380165100098,
2663
+ "objective/scores": 9.143380165100098,
2664
+ "policy/approxkl_avg": 0.24777096509933472,
2665
+ "policy/clipfrac_avg": 0.322265625,
2666
+ "policy/entropy_avg": 0.5264837741851807,
2667
+ "step": 132,
2668
+ "val/clipfrac_avg": 0.0,
2669
+ "val/num_eos_tokens": 2881,
2670
+ "val/ratio": 1.0001630783081055,
2671
+ "val/ratio_var": 2.7501182557898574e-06
2672
+ },
2673
+ {
2674
+ "episode": 68608,
2675
+ "epoch": 0.29389993145990406,
2676
+ "eps": 7,
2677
+ "loss/policy_avg": 1.7297974824905396,
2678
+ "lr": 1.44140625e-06,
2679
+ "objective/entropy": -38.04670715332031,
2680
+ "objective/kl": 29.923572540283203,
2681
+ "objective/non_score_reward": 0.0,
2682
+ "objective/rlhf_reward": 9.23299789428711,
2683
+ "objective/scores": 9.23299789428711,
2684
+ "policy/approxkl_avg": 0.3208100199699402,
2685
+ "policy/clipfrac_avg": 0.349609375,
2686
+ "policy/entropy_avg": 0.5299798250198364,
2687
+ "step": 133,
2688
+ "val/clipfrac_avg": 0.0,
2689
+ "val/num_eos_tokens": 2948,
2690
+ "val/ratio": 1.0002143383026123,
2691
+ "val/ratio_var": 6.060246505512623e-06
2692
+ },
2693
+ {
2694
+ "episode": 69120,
2695
+ "epoch": 0.29609321453050036,
2696
+ "eps": 7,
2697
+ "loss/policy_avg": 0.2908250093460083,
2698
+ "lr": 1.4296875e-06,
2699
+ "objective/entropy": -39.67091751098633,
2700
+ "objective/kl": 29.891416549682617,
2701
+ "objective/non_score_reward": 0.0,
2702
+ "objective/rlhf_reward": 9.22230339050293,
2703
+ "objective/scores": 9.22230339050293,
2704
+ "policy/approxkl_avg": 0.2994619607925415,
2705
+ "policy/clipfrac_avg": 0.361328125,
2706
+ "policy/entropy_avg": 0.5484848618507385,
2707
+ "step": 134,
2708
+ "val/clipfrac_avg": 0.0,
2709
+ "val/num_eos_tokens": 2992,
2710
+ "val/ratio": 0.9998666048049927,
2711
+ "val/ratio_var": 4.691268713941099e-06
2712
+ },
2713
+ {
2714
+ "episode": 69632,
2715
+ "epoch": 0.29828649760109666,
2716
+ "eps": 6,
2717
+ "loss/policy_avg": 0.16967931389808655,
2718
+ "lr": 1.41796875e-06,
2719
+ "objective/entropy": -40.47759246826172,
2720
+ "objective/kl": 28.60110855102539,
2721
+ "objective/non_score_reward": 0.0,
2722
+ "objective/rlhf_reward": 9.249778747558594,
2723
+ "objective/scores": 9.249778747558594,
2724
+ "policy/approxkl_avg": 0.30396944284439087,
2725
+ "policy/clipfrac_avg": 0.369140625,
2726
+ "policy/entropy_avg": 0.5294628739356995,
2727
+ "step": 135,
2728
+ "val/clipfrac_avg": 0.0,
2729
+ "val/num_eos_tokens": 3158,
2730
+ "val/ratio": 0.9999102354049683,
2731
+ "val/ratio_var": 4.847398031415651e-06
2732
+ },
2733
+ {
2734
+ "episode": 70144,
2735
+ "epoch": 0.30047978067169295,
2736
+ "eps": 6,
2737
+ "loss/policy_avg": 0.30670228600502014,
2738
+ "lr": 1.40625e-06,
2739
+ "objective/entropy": -41.60497283935547,
2740
+ "objective/kl": 28.722915649414062,
2741
+ "objective/non_score_reward": 0.0,
2742
+ "objective/rlhf_reward": 9.269282341003418,
2743
+ "objective/scores": 9.269282341003418,
2744
+ "policy/approxkl_avg": 0.27743324637413025,
2745
+ "policy/clipfrac_avg": 0.35546875,
2746
+ "policy/entropy_avg": 0.5208898782730103,
2747
+ "step": 136,
2748
+ "val/clipfrac_avg": 0.0,
2749
+ "val/num_eos_tokens": 3202,
2750
+ "val/ratio": 1.0001318454742432,
2751
+ "val/ratio_var": 4.43136786998366e-06
2752
+ },
2753
+ {
2754
+ "episode": 70656,
2755
+ "epoch": 0.30267306374228925,
2756
+ "eps": 6,
2757
+ "loss/policy_avg": 0.24920354783535004,
2758
+ "lr": 1.39453125e-06,
2759
+ "objective/entropy": -42.12297439575195,
2760
+ "objective/kl": 28.017452239990234,
2761
+ "objective/non_score_reward": 0.0,
2762
+ "objective/rlhf_reward": 8.922027587890625,
2763
+ "objective/scores": 8.922027587890625,
2764
+ "policy/approxkl_avg": 0.23385344445705414,
2765
+ "policy/clipfrac_avg": 0.365234375,
2766
+ "policy/entropy_avg": 0.5616132616996765,
2767
+ "step": 137,
2768
+ "val/clipfrac_avg": 0.0,
2769
+ "val/num_eos_tokens": 3543,
2770
+ "val/ratio": 0.9999379515647888,
2771
+ "val/ratio_var": 3.0522689939971315e-06
2772
+ }
2773
+ ],
2774
+ "logging_steps": 1,
2775
+ "max_steps": 128,
2776
+ "num_input_tokens_seen": 0,
2777
+ "num_train_epochs": 1.122960932145305,
2778
+ "save_steps": 26,
2779
+ "stateful_callbacks": {
2780
+ "TrainerControl": {
2781
+ "args": {
2782
+ "should_epoch_stop": false,
2783
+ "should_evaluate": false,
2784
+ "should_log": true,
2785
+ "should_save": true,
2786
+ "should_training_stop": true
2787
+ },
2788
+ "attributes": {}
2789
+ }
2790
+ },
2791
+ "total_flos": 0,
2792
+ "train_batch_size": null,
2793
+ "trial_name": null,
2794
+ "trial_params": null
2795
+ }
checkpoint-138/training_args.bin ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:a70ca9efe45d40957298b73ba73bfdfa16ec22c723e687576204531d74270deb
3
+ size 6584