Muqeeth commited on
Commit
2215525
·
verified ·
1 Parent(s): 4356624

Training in progress, step 128, checkpoint

Browse files
checkpoint-128/config.json ADDED
@@ -0,0 +1,31 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "architectures": [
3
+ "GPTNeoXForCausalLM"
4
+ ],
5
+ "attention_bias": true,
6
+ "attention_dropout": 0.0,
7
+ "bos_token_id": 0,
8
+ "classifier_dropout": 0.1,
9
+ "eos_token_id": 0,
10
+ "hidden_act": "gelu",
11
+ "hidden_dropout": 0.0,
12
+ "hidden_size": 1024,
13
+ "initializer_range": 0.02,
14
+ "intermediate_size": 4096,
15
+ "layer_norm_eps": 1e-05,
16
+ "max_position_embeddings": 2048,
17
+ "model_type": "gpt_neox",
18
+ "num_attention_heads": 16,
19
+ "num_hidden_layers": 24,
20
+ "partial_rotary_factor": 0.25,
21
+ "rope_scaling": null,
22
+ "rope_theta": 10000,
23
+ "rotary_emb_base": 10000,
24
+ "rotary_pct": 0.25,
25
+ "tie_word_embeddings": false,
26
+ "torch_dtype": "float32",
27
+ "transformers_version": "4.50.3",
28
+ "use_cache": true,
29
+ "use_parallel_residual": true,
30
+ "vocab_size": 50304
31
+ }
checkpoint-128/generation_config.json ADDED
@@ -0,0 +1,5 @@
 
 
 
 
 
 
1
+ {
2
+ "_from_model_config": true,
3
+ "bos_token_id": 0,
4
+ "transformers_version": "4.50.3"
5
+ }
checkpoint-128/model.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:5364d11167515e47173b7a12e2022fe6448b030239e624443bc4c334c049ab48
3
+ size 1621370224
checkpoint-128/optimizer.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:b97d49c1f401edb09f8cb62033d80ee0f6a3d67416ee4e25d7fd7c6b248abf69
3
+ size 3242916026
checkpoint-128/rng_state.pth ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:e50bab34303e0e8c6bdbfd672e56940a40386a6dbcc8e58f48465bab1a04184d
3
+ size 14180
checkpoint-128/scheduler.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:b94e984393329de60dc7fa902fd68a2fd74f248d210b82c75caa9d0d693010e7
3
+ size 1064
checkpoint-128/special_tokens_map.json ADDED
@@ -0,0 +1,30 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "bos_token": {
3
+ "content": "<|endoftext|>",
4
+ "lstrip": false,
5
+ "normalized": false,
6
+ "rstrip": false,
7
+ "single_word": false
8
+ },
9
+ "eos_token": {
10
+ "content": "<|endoftext|>",
11
+ "lstrip": false,
12
+ "normalized": false,
13
+ "rstrip": false,
14
+ "single_word": false
15
+ },
16
+ "pad_token": {
17
+ "content": "<|padding|>",
18
+ "lstrip": false,
19
+ "normalized": false,
20
+ "rstrip": false,
21
+ "single_word": false
22
+ },
23
+ "unk_token": {
24
+ "content": "<|endoftext|>",
25
+ "lstrip": false,
26
+ "normalized": false,
27
+ "rstrip": false,
28
+ "single_word": false
29
+ }
30
+ }
checkpoint-128/tokenizer.json ADDED
The diff for this file is too large to render. See raw diff
 
checkpoint-128/tokenizer_config.json ADDED
@@ -0,0 +1,216 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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": false,
209
+ "eos_token": "<|endoftext|>",
210
+ "extra_special_tokens": {},
211
+ "model_max_length": 1000000000000000019884624838656,
212
+ "pad_token": "<|padding|>",
213
+ "padding_side": "left",
214
+ "tokenizer_class": "GPTNeoXTokenizer",
215
+ "unk_token": "<|endoftext|>"
216
+ }
checkpoint-128/trainer_state.json ADDED
@@ -0,0 +1,2595 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "best_global_step": null,
3
+ "best_metric": null,
4
+ "best_model_checkpoint": null,
5
+ "episode": 65536,
6
+ "epoch": 0.2807402330363262,
7
+ "eval_steps": 500,
8
+ "global_step": 128,
9
+ "is_hyper_param_search": false,
10
+ "is_local_process_zero": true,
11
+ "is_world_process_zero": true,
12
+ "log_history": [
13
+ {
14
+ "episode": 512,
15
+ "epoch": 0.0021932830705962986,
16
+ "eps": 6,
17
+ "loss/policy_avg": 0.16531690955162048,
18
+ "lr": 3e-06,
19
+ "objective/entropy": -55.32527542114258,
20
+ "objective/kl": 0.2650028467178345,
21
+ "objective/non_score_reward": -0.013250143267214298,
22
+ "objective/rlhf_reward": 6.5853118896484375,
23
+ "objective/scores": 6.598561763763428,
24
+ "policy/approxkl_avg": 0.09512920677661896,
25
+ "policy/clipfrac_avg": 0.294921875,
26
+ "policy/entropy_avg": 0.610973596572876,
27
+ "step": 0,
28
+ "val/clipfrac_avg": 0.0,
29
+ "val/num_eos_tokens": 6004,
30
+ "val/ratio": 0.9999569058418274,
31
+ "val/ratio_var": 1.1622008742051548e-06
32
+ },
33
+ {
34
+ "episode": 1024,
35
+ "epoch": 0.004386566141192597,
36
+ "eps": 6,
37
+ "loss/policy_avg": 0.22872287034988403,
38
+ "lr": 2.9882812500000002e-06,
39
+ "objective/entropy": -50.97515106201172,
40
+ "objective/kl": 0.687268853187561,
41
+ "objective/non_score_reward": -0.03436344116926193,
42
+ "objective/rlhf_reward": 6.763760089874268,
43
+ "objective/scores": 6.798122882843018,
44
+ "policy/approxkl_avg": 0.11450275778770447,
45
+ "policy/clipfrac_avg": 0.28125,
46
+ "policy/entropy_avg": 0.6294225454330444,
47
+ "step": 1,
48
+ "val/clipfrac_avg": 0.0,
49
+ "val/num_eos_tokens": 5686,
50
+ "val/ratio": 0.9997917413711548,
51
+ "val/ratio_var": 1.6605762311883154e-06
52
+ },
53
+ {
54
+ "episode": 1536,
55
+ "epoch": 0.006579849211788897,
56
+ "eps": 6,
57
+ "loss/policy_avg": 0.1424816995859146,
58
+ "lr": 2.9765625e-06,
59
+ "objective/entropy": -57.432891845703125,
60
+ "objective/kl": 1.1181037425994873,
61
+ "objective/non_score_reward": -0.05590519309043884,
62
+ "objective/rlhf_reward": 7.33503532409668,
63
+ "objective/scores": 7.390939712524414,
64
+ "policy/approxkl_avg": 0.0936519205570221,
65
+ "policy/clipfrac_avg": 0.322265625,
66
+ "policy/entropy_avg": 0.5745395421981812,
67
+ "step": 2,
68
+ "val/clipfrac_avg": 0.0,
69
+ "val/num_eos_tokens": 5664,
70
+ "val/ratio": 1.0001490116119385,
71
+ "val/ratio_var": 1.4631596059189178e-06
72
+ },
73
+ {
74
+ "episode": 2048,
75
+ "epoch": 0.008773132282385195,
76
+ "eps": 6,
77
+ "loss/policy_avg": 0.17385584115982056,
78
+ "lr": 2.96484375e-06,
79
+ "objective/entropy": -54.846282958984375,
80
+ "objective/kl": 1.592078447341919,
81
+ "objective/non_score_reward": -0.07960392534732819,
82
+ "objective/rlhf_reward": 7.236883163452148,
83
+ "objective/scores": 7.316486835479736,
84
+ "policy/approxkl_avg": 0.11385183781385422,
85
+ "policy/clipfrac_avg": 0.283203125,
86
+ "policy/entropy_avg": 0.5554953813552856,
87
+ "step": 3,
88
+ "val/clipfrac_avg": 0.0,
89
+ "val/num_eos_tokens": 5669,
90
+ "val/ratio": 1.0001744031906128,
91
+ "val/ratio_var": 2.2094427549745888e-06
92
+ },
93
+ {
94
+ "episode": 2560,
95
+ "epoch": 0.010966415352981495,
96
+ "eps": 6,
97
+ "loss/policy_avg": 0.13433611392974854,
98
+ "lr": 2.953125e-06,
99
+ "objective/entropy": -57.38214111328125,
100
+ "objective/kl": 2.255763292312622,
101
+ "objective/non_score_reward": -0.11278817057609558,
102
+ "objective/rlhf_reward": 7.51954460144043,
103
+ "objective/scores": 7.6323323249816895,
104
+ "policy/approxkl_avg": 0.11168310046195984,
105
+ "policy/clipfrac_avg": 0.328125,
106
+ "policy/entropy_avg": 0.5522703528404236,
107
+ "step": 4,
108
+ "val/clipfrac_avg": 0.0,
109
+ "val/num_eos_tokens": 5788,
110
+ "val/ratio": 1.0000145435333252,
111
+ "val/ratio_var": 1.634987597753934e-06
112
+ },
113
+ {
114
+ "episode": 3072,
115
+ "epoch": 0.013159698423577794,
116
+ "eps": 7,
117
+ "loss/policy_avg": 0.13945499062538147,
118
+ "lr": 2.94140625e-06,
119
+ "objective/entropy": -56.70085144042969,
120
+ "objective/kl": 3.472144603729248,
121
+ "objective/non_score_reward": -0.1736072301864624,
122
+ "objective/rlhf_reward": 7.586057662963867,
123
+ "objective/scores": 7.759665012359619,
124
+ "policy/approxkl_avg": 0.10526105761528015,
125
+ "policy/clipfrac_avg": 0.31640625,
126
+ "policy/entropy_avg": 0.5675480365753174,
127
+ "step": 5,
128
+ "val/clipfrac_avg": 0.0,
129
+ "val/num_eos_tokens": 5734,
130
+ "val/ratio": 1.0001214742660522,
131
+ "val/ratio_var": 1.3759735111307236e-06
132
+ },
133
+ {
134
+ "episode": 3584,
135
+ "epoch": 0.015352981494174092,
136
+ "eps": 7,
137
+ "loss/policy_avg": 0.09546089917421341,
138
+ "lr": 2.9296875e-06,
139
+ "objective/entropy": -57.61265563964844,
140
+ "objective/kl": 4.814455986022949,
141
+ "objective/non_score_reward": -0.24072279036045074,
142
+ "objective/rlhf_reward": 8.14423942565918,
143
+ "objective/scores": 8.38496208190918,
144
+ "policy/approxkl_avg": 0.11149222403764725,
145
+ "policy/clipfrac_avg": 0.34765625,
146
+ "policy/entropy_avg": 0.5545904040336609,
147
+ "step": 6,
148
+ "val/clipfrac_avg": 0.0,
149
+ "val/num_eos_tokens": 5490,
150
+ "val/ratio": 1.0001678466796875,
151
+ "val/ratio_var": 1.6421691952928086e-06
152
+ },
153
+ {
154
+ "episode": 4096,
155
+ "epoch": 0.01754626456477039,
156
+ "eps": 7,
157
+ "loss/policy_avg": 0.235539510846138,
158
+ "lr": 2.91796875e-06,
159
+ "objective/entropy": -52.81565856933594,
160
+ "objective/kl": 7.018827438354492,
161
+ "objective/non_score_reward": -0.35094135999679565,
162
+ "objective/rlhf_reward": 8.318111419677734,
163
+ "objective/scores": 8.669052124023438,
164
+ "policy/approxkl_avg": 0.14312206208705902,
165
+ "policy/clipfrac_avg": 0.2890625,
166
+ "policy/entropy_avg": 0.5730656385421753,
167
+ "step": 7,
168
+ "val/clipfrac_avg": 0.0,
169
+ "val/num_eos_tokens": 5627,
170
+ "val/ratio": 0.9999744892120361,
171
+ "val/ratio_var": 2.2067135887482436e-06
172
+ },
173
+ {
174
+ "episode": 4608,
175
+ "epoch": 0.01973954763536669,
176
+ "eps": 7,
177
+ "loss/policy_avg": 0.18809324502944946,
178
+ "lr": 2.90625e-06,
179
+ "objective/entropy": -51.69276428222656,
180
+ "objective/kl": 8.87761116027832,
181
+ "objective/non_score_reward": -0.4438805878162384,
182
+ "objective/rlhf_reward": 8.72425651550293,
183
+ "objective/scores": 9.168136596679688,
184
+ "policy/approxkl_avg": 0.13404184579849243,
185
+ "policy/clipfrac_avg": 0.31640625,
186
+ "policy/entropy_avg": 0.5807090997695923,
187
+ "step": 8,
188
+ "val/clipfrac_avg": 0.0,
189
+ "val/num_eos_tokens": 5006,
190
+ "val/ratio": 0.9999812841415405,
191
+ "val/ratio_var": 2.421862973278621e-06
192
+ },
193
+ {
194
+ "episode": 5120,
195
+ "epoch": 0.02193283070596299,
196
+ "eps": 7,
197
+ "loss/policy_avg": 0.13251258432865143,
198
+ "lr": 2.89453125e-06,
199
+ "objective/entropy": -52.892486572265625,
200
+ "objective/kl": 11.408520698547363,
201
+ "objective/non_score_reward": -0.5704259872436523,
202
+ "objective/rlhf_reward": 9.237460136413574,
203
+ "objective/scores": 9.807886123657227,
204
+ "policy/approxkl_avg": 0.12708792090415955,
205
+ "policy/clipfrac_avg": 0.345703125,
206
+ "policy/entropy_avg": 0.5779529213905334,
207
+ "step": 9,
208
+ "val/clipfrac_avg": 0.0,
209
+ "val/num_eos_tokens": 4455,
210
+ "val/ratio": 1.0001035928726196,
211
+ "val/ratio_var": 1.636367528590199e-06
212
+ },
213
+ {
214
+ "episode": 5632,
215
+ "epoch": 0.024126113776559287,
216
+ "eps": 7,
217
+ "loss/policy_avg": 0.18584033846855164,
218
+ "lr": 2.8828125e-06,
219
+ "objective/entropy": -52.1838264465332,
220
+ "objective/kl": 12.914525032043457,
221
+ "objective/non_score_reward": -0.645726203918457,
222
+ "objective/rlhf_reward": 9.471303939819336,
223
+ "objective/scores": 10.11703109741211,
224
+ "policy/approxkl_avg": 0.14649230241775513,
225
+ "policy/clipfrac_avg": 0.310546875,
226
+ "policy/entropy_avg": 0.5547796487808228,
227
+ "step": 10,
228
+ "val/clipfrac_avg": 0.0,
229
+ "val/num_eos_tokens": 3205,
230
+ "val/ratio": 1.0001304149627686,
231
+ "val/ratio_var": 1.861714054030017e-06
232
+ },
233
+ {
234
+ "episode": 6144,
235
+ "epoch": 0.026319396847155587,
236
+ "eps": 7,
237
+ "loss/policy_avg": 0.29413461685180664,
238
+ "lr": 2.87109375e-06,
239
+ "objective/entropy": -50.931846618652344,
240
+ "objective/kl": 14.186361312866211,
241
+ "objective/non_score_reward": -0.7093180418014526,
242
+ "objective/rlhf_reward": 9.125755310058594,
243
+ "objective/scores": 9.835073471069336,
244
+ "policy/approxkl_avg": 0.10939589142799377,
245
+ "policy/clipfrac_avg": 0.265625,
246
+ "policy/entropy_avg": 0.5228562355041504,
247
+ "step": 11,
248
+ "val/clipfrac_avg": 0.0,
249
+ "val/num_eos_tokens": 2346,
250
+ "val/ratio": 1.0003429651260376,
251
+ "val/ratio_var": 2.3797676931280876e-06
252
+ },
253
+ {
254
+ "episode": 6656,
255
+ "epoch": 0.028512679917751884,
256
+ "eps": 7,
257
+ "loss/policy_avg": 0.2645286023616791,
258
+ "lr": 2.859375e-06,
259
+ "objective/entropy": -43.677490234375,
260
+ "objective/kl": 15.47945785522461,
261
+ "objective/non_score_reward": -0.7739728689193726,
262
+ "objective/rlhf_reward": 9.23073959350586,
263
+ "objective/scores": 10.00471305847168,
264
+ "policy/approxkl_avg": 0.12562136352062225,
265
+ "policy/clipfrac_avg": 0.318359375,
266
+ "policy/entropy_avg": 0.5210846066474915,
267
+ "step": 12,
268
+ "val/clipfrac_avg": 0.0,
269
+ "val/num_eos_tokens": 2033,
270
+ "val/ratio": 0.9997715950012207,
271
+ "val/ratio_var": 2.0162181044725003e-06
272
+ },
273
+ {
274
+ "episode": 7168,
275
+ "epoch": 0.030705962988348184,
276
+ "eps": 7,
277
+ "loss/policy_avg": 0.11199663579463959,
278
+ "lr": 2.84765625e-06,
279
+ "objective/entropy": -48.57129669189453,
280
+ "objective/kl": 15.832266807556152,
281
+ "objective/non_score_reward": -0.7916133403778076,
282
+ "objective/rlhf_reward": 9.829407691955566,
283
+ "objective/scores": 10.621020317077637,
284
+ "policy/approxkl_avg": 0.11009921133518219,
285
+ "policy/clipfrac_avg": 0.33203125,
286
+ "policy/entropy_avg": 0.5088958740234375,
287
+ "step": 13,
288
+ "val/clipfrac_avg": 0.0,
289
+ "val/num_eos_tokens": 2090,
290
+ "val/ratio": 0.9999546408653259,
291
+ "val/ratio_var": 1.2889486242784187e-06
292
+ },
293
+ {
294
+ "episode": 7680,
295
+ "epoch": 0.03289924605894448,
296
+ "eps": 7,
297
+ "loss/policy_avg": 0.2023422122001648,
298
+ "lr": 2.8359375e-06,
299
+ "objective/entropy": -44.59040451049805,
300
+ "objective/kl": 16.71160888671875,
301
+ "objective/non_score_reward": -0.8355805277824402,
302
+ "objective/rlhf_reward": 9.781827926635742,
303
+ "objective/scores": 10.61740779876709,
304
+ "policy/approxkl_avg": 0.1277025043964386,
305
+ "policy/clipfrac_avg": 0.3203125,
306
+ "policy/entropy_avg": 0.5323245525360107,
307
+ "step": 14,
308
+ "val/clipfrac_avg": 0.0,
309
+ "val/num_eos_tokens": 2203,
310
+ "val/ratio": 1.0000615119934082,
311
+ "val/ratio_var": 1.983828269658261e-06
312
+ },
313
+ {
314
+ "episode": 8192,
315
+ "epoch": 0.03509252912954078,
316
+ "eps": 7,
317
+ "loss/policy_avg": 0.11938837170600891,
318
+ "lr": 2.82421875e-06,
319
+ "objective/entropy": -42.89122772216797,
320
+ "objective/kl": 17.852779388427734,
321
+ "objective/non_score_reward": -0.8926390409469604,
322
+ "objective/rlhf_reward": 9.708269119262695,
323
+ "objective/scores": 10.600908279418945,
324
+ "policy/approxkl_avg": 0.13053694367408752,
325
+ "policy/clipfrac_avg": 0.326171875,
326
+ "policy/entropy_avg": 0.5566774606704712,
327
+ "step": 15,
328
+ "val/clipfrac_avg": 0.0,
329
+ "val/num_eos_tokens": 2344,
330
+ "val/ratio": 0.9997885823249817,
331
+ "val/ratio_var": 1.6801355968709686e-06
332
+ },
333
+ {
334
+ "episode": 8704,
335
+ "epoch": 0.03728581220013708,
336
+ "eps": 7,
337
+ "loss/policy_avg": 0.13911566138267517,
338
+ "lr": 2.8125e-06,
339
+ "objective/entropy": -41.679447174072266,
340
+ "objective/kl": 19.221725463867188,
341
+ "objective/non_score_reward": -0.9610862731933594,
342
+ "objective/rlhf_reward": 10.175975799560547,
343
+ "objective/scores": 11.137063026428223,
344
+ "policy/approxkl_avg": 0.13960514962673187,
345
+ "policy/clipfrac_avg": 0.341796875,
346
+ "policy/entropy_avg": 0.5391991138458252,
347
+ "step": 16,
348
+ "val/clipfrac_avg": 0.0,
349
+ "val/num_eos_tokens": 2362,
350
+ "val/ratio": 1.0002028942108154,
351
+ "val/ratio_var": 1.7096700730689918e-06
352
+ },
353
+ {
354
+ "episode": 9216,
355
+ "epoch": 0.03947909527073338,
356
+ "eps": 7,
357
+ "loss/policy_avg": 0.1141306459903717,
358
+ "lr": 2.80078125e-06,
359
+ "objective/entropy": -38.35896682739258,
360
+ "objective/kl": 21.127605438232422,
361
+ "objective/non_score_reward": -1.0563803911209106,
362
+ "objective/rlhf_reward": 10.05744743347168,
363
+ "objective/scores": 11.113826751708984,
364
+ "policy/approxkl_avg": 0.16503070294857025,
365
+ "policy/clipfrac_avg": 0.357421875,
366
+ "policy/entropy_avg": 0.5556284189224243,
367
+ "step": 17,
368
+ "val/clipfrac_avg": 0.0,
369
+ "val/num_eos_tokens": 2512,
370
+ "val/ratio": 0.9997155070304871,
371
+ "val/ratio_var": 2.5818217181949876e-06
372
+ },
373
+ {
374
+ "episode": 9728,
375
+ "epoch": 0.041672378341329676,
376
+ "eps": 7,
377
+ "loss/policy_avg": 0.1537306159734726,
378
+ "lr": 2.7890625e-06,
379
+ "objective/entropy": -34.704803466796875,
380
+ "objective/kl": 21.983102798461914,
381
+ "objective/non_score_reward": -1.0991551876068115,
382
+ "objective/rlhf_reward": 9.975513458251953,
383
+ "objective/scores": 11.074668884277344,
384
+ "policy/approxkl_avg": 0.16006678342819214,
385
+ "policy/clipfrac_avg": 0.369140625,
386
+ "policy/entropy_avg": 0.5572025775909424,
387
+ "step": 18,
388
+ "val/clipfrac_avg": 0.0,
389
+ "val/num_eos_tokens": 2321,
390
+ "val/ratio": 0.9999374747276306,
391
+ "val/ratio_var": 2.024344212259166e-06
392
+ },
393
+ {
394
+ "episode": 10240,
395
+ "epoch": 0.04386566141192598,
396
+ "eps": 7,
397
+ "loss/policy_avg": 0.18893875181674957,
398
+ "lr": 2.77734375e-06,
399
+ "objective/entropy": -33.0050163269043,
400
+ "objective/kl": 23.7343807220459,
401
+ "objective/non_score_reward": -1.1867190599441528,
402
+ "objective/rlhf_reward": 9.882341384887695,
403
+ "objective/scores": 11.069059371948242,
404
+ "policy/approxkl_avg": 0.16320574283599854,
405
+ "policy/clipfrac_avg": 0.3203125,
406
+ "policy/entropy_avg": 0.5377587676048279,
407
+ "step": 19,
408
+ "val/clipfrac_avg": 0.0,
409
+ "val/num_eos_tokens": 2204,
410
+ "val/ratio": 0.9999250173568726,
411
+ "val/ratio_var": 1.6979042811726686e-06
412
+ },
413
+ {
414
+ "episode": 10752,
415
+ "epoch": 0.04605894448252228,
416
+ "eps": 7,
417
+ "loss/policy_avg": 0.1287260800600052,
418
+ "lr": 2.765625e-06,
419
+ "objective/entropy": -30.667198181152344,
420
+ "objective/kl": 25.033987045288086,
421
+ "objective/non_score_reward": -1.251699447631836,
422
+ "objective/rlhf_reward": 10.062000274658203,
423
+ "objective/scores": 11.313698768615723,
424
+ "policy/approxkl_avg": 0.1715945601463318,
425
+ "policy/clipfrac_avg": 0.357421875,
426
+ "policy/entropy_avg": 0.5366133451461792,
427
+ "step": 20,
428
+ "val/clipfrac_avg": 0.0,
429
+ "val/num_eos_tokens": 2182,
430
+ "val/ratio": 0.9997314214706421,
431
+ "val/ratio_var": 2.0495133412623545e-06
432
+ },
433
+ {
434
+ "episode": 11264,
435
+ "epoch": 0.048252227553118573,
436
+ "eps": 7,
437
+ "loss/policy_avg": 0.19849294424057007,
438
+ "lr": 2.75390625e-06,
439
+ "objective/entropy": -33.668846130371094,
440
+ "objective/kl": 24.064254760742188,
441
+ "objective/non_score_reward": -1.2032127380371094,
442
+ "objective/rlhf_reward": 10.199529647827148,
443
+ "objective/scores": 11.402742385864258,
444
+ "policy/approxkl_avg": 0.16391384601593018,
445
+ "policy/clipfrac_avg": 0.310546875,
446
+ "policy/entropy_avg": 0.5062953233718872,
447
+ "step": 21,
448
+ "val/clipfrac_avg": 0.0,
449
+ "val/num_eos_tokens": 2180,
450
+ "val/ratio": 0.9998014569282532,
451
+ "val/ratio_var": 2.044889470198541e-06
452
+ },
453
+ {
454
+ "episode": 11776,
455
+ "epoch": 0.05044551062371487,
456
+ "eps": 7,
457
+ "loss/policy_avg": 0.10195901989936829,
458
+ "lr": 2.7421875e-06,
459
+ "objective/entropy": -39.06690979003906,
460
+ "objective/kl": 22.586650848388672,
461
+ "objective/non_score_reward": -1.1293325424194336,
462
+ "objective/rlhf_reward": 10.262170791625977,
463
+ "objective/scores": 11.391502380371094,
464
+ "policy/approxkl_avg": 0.1463184356689453,
465
+ "policy/clipfrac_avg": 0.3671875,
466
+ "policy/entropy_avg": 0.4969005286693573,
467
+ "step": 22,
468
+ "val/clipfrac_avg": 0.0,
469
+ "val/num_eos_tokens": 2233,
470
+ "val/ratio": 0.999759316444397,
471
+ "val/ratio_var": 2.41930774791399e-06
472
+ },
473
+ {
474
+ "episode": 12288,
475
+ "epoch": 0.052638793694311174,
476
+ "eps": 7,
477
+ "loss/policy_avg": 0.10310226678848267,
478
+ "lr": 2.73046875e-06,
479
+ "objective/entropy": -46.28221893310547,
480
+ "objective/kl": 22.036273956298828,
481
+ "objective/non_score_reward": -1.101813793182373,
482
+ "objective/rlhf_reward": 10.483089447021484,
483
+ "objective/scores": 11.5849027633667,
484
+ "policy/approxkl_avg": 0.1329042762517929,
485
+ "policy/clipfrac_avg": 0.326171875,
486
+ "policy/entropy_avg": 0.4535849094390869,
487
+ "step": 23,
488
+ "val/clipfrac_avg": 0.0,
489
+ "val/num_eos_tokens": 2392,
490
+ "val/ratio": 0.9999419450759888,
491
+ "val/ratio_var": 1.8799750023390516e-06
492
+ },
493
+ {
494
+ "episode": 12800,
495
+ "epoch": 0.05483207676490747,
496
+ "eps": 7,
497
+ "loss/policy_avg": 0.11269126832485199,
498
+ "lr": 2.71875e-06,
499
+ "objective/entropy": -43.433353424072266,
500
+ "objective/kl": 22.60041046142578,
501
+ "objective/non_score_reward": -1.1300204992294312,
502
+ "objective/rlhf_reward": 10.324277877807617,
503
+ "objective/scores": 11.45429801940918,
504
+ "policy/approxkl_avg": 0.12673968076705933,
505
+ "policy/clipfrac_avg": 0.287109375,
506
+ "policy/entropy_avg": 0.4514559209346771,
507
+ "step": 24,
508
+ "val/clipfrac_avg": 0.0,
509
+ "val/num_eos_tokens": 2316,
510
+ "val/ratio": 0.9997400045394897,
511
+ "val/ratio_var": 1.5890127542661503e-06
512
+ },
513
+ {
514
+ "episode": 13312,
515
+ "epoch": 0.05702535983550377,
516
+ "eps": 7,
517
+ "loss/policy_avg": 0.2015887349843979,
518
+ "lr": 2.70703125e-06,
519
+ "objective/entropy": -46.94512176513672,
520
+ "objective/kl": 21.664478302001953,
521
+ "objective/non_score_reward": -1.0832240581512451,
522
+ "objective/rlhf_reward": 10.445198059082031,
523
+ "objective/scores": 11.528423309326172,
524
+ "policy/approxkl_avg": 0.12853267788887024,
525
+ "policy/clipfrac_avg": 0.3203125,
526
+ "policy/entropy_avg": 0.39970842003822327,
527
+ "step": 25,
528
+ "val/clipfrac_avg": 0.0,
529
+ "val/num_eos_tokens": 2327,
530
+ "val/ratio": 1.000229835510254,
531
+ "val/ratio_var": 1.6951470342974062e-06
532
+ },
533
+ {
534
+ "episode": 13824,
535
+ "epoch": 0.05921864290610007,
536
+ "eps": 7,
537
+ "loss/policy_avg": 0.07959149777889252,
538
+ "lr": 2.6953125e-06,
539
+ "objective/entropy": -52.64186096191406,
540
+ "objective/kl": 21.013965606689453,
541
+ "objective/non_score_reward": -1.0506982803344727,
542
+ "objective/rlhf_reward": 10.545175552368164,
543
+ "objective/scores": 11.595874786376953,
544
+ "policy/approxkl_avg": 0.08630259335041046,
545
+ "policy/clipfrac_avg": 0.26953125,
546
+ "policy/entropy_avg": 0.3702251613140106,
547
+ "step": 26,
548
+ "val/clipfrac_avg": 0.0,
549
+ "val/num_eos_tokens": 2337,
550
+ "val/ratio": 0.999913215637207,
551
+ "val/ratio_var": 9.910673952617799e-07
552
+ },
553
+ {
554
+ "episode": 14336,
555
+ "epoch": 0.06141192597669637,
556
+ "eps": 7,
557
+ "loss/policy_avg": 0.133376345038414,
558
+ "lr": 2.68359375e-06,
559
+ "objective/entropy": -52.62764358520508,
560
+ "objective/kl": 21.873783111572266,
561
+ "objective/non_score_reward": -1.0936890840530396,
562
+ "objective/rlhf_reward": 10.531662940979004,
563
+ "objective/scores": 11.62535285949707,
564
+ "policy/approxkl_avg": 0.09903587400913239,
565
+ "policy/clipfrac_avg": 0.259765625,
566
+ "policy/entropy_avg": 0.3591485619544983,
567
+ "step": 27,
568
+ "val/clipfrac_avg": 0.0,
569
+ "val/num_eos_tokens": 2342,
570
+ "val/ratio": 1.0000910758972168,
571
+ "val/ratio_var": 1.4223566040527658e-06
572
+ },
573
+ {
574
+ "episode": 14848,
575
+ "epoch": 0.06360520904729267,
576
+ "eps": 7,
577
+ "loss/policy_avg": 0.09103123843669891,
578
+ "lr": 2.671875e-06,
579
+ "objective/entropy": -53.78799057006836,
580
+ "objective/kl": 22.003097534179688,
581
+ "objective/non_score_reward": -1.1001547574996948,
582
+ "objective/rlhf_reward": 10.68588924407959,
583
+ "objective/scores": 11.786044120788574,
584
+ "policy/approxkl_avg": 0.09524992108345032,
585
+ "policy/clipfrac_avg": 0.2734375,
586
+ "policy/entropy_avg": 0.34278032183647156,
587
+ "step": 28,
588
+ "val/clipfrac_avg": 0.0,
589
+ "val/num_eos_tokens": 2247,
590
+ "val/ratio": 0.9998263120651245,
591
+ "val/ratio_var": 1.1149676311106305e-06
592
+ },
593
+ {
594
+ "episode": 15360,
595
+ "epoch": 0.06579849211788896,
596
+ "eps": 7,
597
+ "loss/policy_avg": 0.11379580199718475,
598
+ "lr": 2.66015625e-06,
599
+ "objective/entropy": -51.830589294433594,
600
+ "objective/kl": 22.123140335083008,
601
+ "objective/non_score_reward": -1.1061570644378662,
602
+ "objective/rlhf_reward": 10.770975112915039,
603
+ "objective/scores": 11.877132415771484,
604
+ "policy/approxkl_avg": 0.0886557325720787,
605
+ "policy/clipfrac_avg": 0.275390625,
606
+ "policy/entropy_avg": 0.32301610708236694,
607
+ "step": 29,
608
+ "val/clipfrac_avg": 0.0,
609
+ "val/num_eos_tokens": 2108,
610
+ "val/ratio": 0.9998676776885986,
611
+ "val/ratio_var": 9.721951528263162e-07
612
+ },
613
+ {
614
+ "episode": 15872,
615
+ "epoch": 0.06799177518848526,
616
+ "eps": 7,
617
+ "loss/policy_avg": 0.1263660490512848,
618
+ "lr": 2.6484375e-06,
619
+ "objective/entropy": -52.30486297607422,
620
+ "objective/kl": 22.81517219543457,
621
+ "objective/non_score_reward": -1.1407586336135864,
622
+ "objective/rlhf_reward": 10.94261360168457,
623
+ "objective/scores": 12.083372116088867,
624
+ "policy/approxkl_avg": 0.08135354518890381,
625
+ "policy/clipfrac_avg": 0.2578125,
626
+ "policy/entropy_avg": 0.3432938754558563,
627
+ "step": 30,
628
+ "val/clipfrac_avg": 0.0,
629
+ "val/num_eos_tokens": 2136,
630
+ "val/ratio": 0.9998652935028076,
631
+ "val/ratio_var": 1.0441638096381212e-06
632
+ },
633
+ {
634
+ "episode": 16384,
635
+ "epoch": 0.07018505825908156,
636
+ "eps": 7,
637
+ "loss/policy_avg": 0.11862729489803314,
638
+ "lr": 2.63671875e-06,
639
+ "objective/entropy": -52.75566101074219,
640
+ "objective/kl": 23.3867130279541,
641
+ "objective/non_score_reward": -1.1693357229232788,
642
+ "objective/rlhf_reward": 10.69216251373291,
643
+ "objective/scores": 11.86149787902832,
644
+ "policy/approxkl_avg": 0.08456984162330627,
645
+ "policy/clipfrac_avg": 0.25390625,
646
+ "policy/entropy_avg": 0.28824326395988464,
647
+ "step": 31,
648
+ "val/clipfrac_avg": 0.0,
649
+ "val/num_eos_tokens": 2113,
650
+ "val/ratio": 0.9998213052749634,
651
+ "val/ratio_var": 1.4351767276821192e-06
652
+ },
653
+ {
654
+ "episode": 16896,
655
+ "epoch": 0.07237834132967787,
656
+ "eps": 7,
657
+ "loss/policy_avg": 0.1567326784133911,
658
+ "lr": 2.6250000000000003e-06,
659
+ "objective/entropy": -54.1901741027832,
660
+ "objective/kl": 22.347511291503906,
661
+ "objective/non_score_reward": -1.1173756122589111,
662
+ "objective/rlhf_reward": 10.90484619140625,
663
+ "objective/scores": 12.022221565246582,
664
+ "policy/approxkl_avg": 0.07176832109689713,
665
+ "policy/clipfrac_avg": 0.240234375,
666
+ "policy/entropy_avg": 0.29196977615356445,
667
+ "step": 32,
668
+ "val/clipfrac_avg": 0.0,
669
+ "val/num_eos_tokens": 2180,
670
+ "val/ratio": 1.0000624656677246,
671
+ "val/ratio_var": 1.1582534398257849e-06
672
+ },
673
+ {
674
+ "episode": 17408,
675
+ "epoch": 0.07457162440027416,
676
+ "eps": 7,
677
+ "loss/policy_avg": 0.08574818074703217,
678
+ "lr": 2.61328125e-06,
679
+ "objective/entropy": -53.509254455566406,
680
+ "objective/kl": 23.09099578857422,
681
+ "objective/non_score_reward": -1.1545497179031372,
682
+ "objective/rlhf_reward": 10.845325469970703,
683
+ "objective/scores": 11.99987506866455,
684
+ "policy/approxkl_avg": 0.07496441900730133,
685
+ "policy/clipfrac_avg": 0.265625,
686
+ "policy/entropy_avg": 0.2839745581150055,
687
+ "step": 33,
688
+ "val/clipfrac_avg": 0.0,
689
+ "val/num_eos_tokens": 2102,
690
+ "val/ratio": 0.999615490436554,
691
+ "val/ratio_var": 1.1524506362547982e-06
692
+ },
693
+ {
694
+ "episode": 17920,
695
+ "epoch": 0.07676490747087046,
696
+ "eps": 7,
697
+ "loss/policy_avg": 0.0927099734544754,
698
+ "lr": 2.6015625e-06,
699
+ "objective/entropy": -52.29348373413086,
700
+ "objective/kl": 23.082698822021484,
701
+ "objective/non_score_reward": -1.15413498878479,
702
+ "objective/rlhf_reward": 10.935807228088379,
703
+ "objective/scores": 12.08994197845459,
704
+ "policy/approxkl_avg": 0.15793153643608093,
705
+ "policy/clipfrac_avg": 0.265625,
706
+ "policy/entropy_avg": 0.2997421622276306,
707
+ "step": 34,
708
+ "val/clipfrac_avg": 0.0,
709
+ "val/num_eos_tokens": 2219,
710
+ "val/ratio": 1.0001107454299927,
711
+ "val/ratio_var": 9.742823294800473e-07
712
+ },
713
+ {
714
+ "episode": 18432,
715
+ "epoch": 0.07895819054146676,
716
+ "eps": 7,
717
+ "loss/policy_avg": 0.033941492438316345,
718
+ "lr": 2.5898437500000003e-06,
719
+ "objective/entropy": -53.88836669921875,
720
+ "objective/kl": 22.903392791748047,
721
+ "objective/non_score_reward": -1.1451694965362549,
722
+ "objective/rlhf_reward": 11.03926944732666,
723
+ "objective/scores": 12.184438705444336,
724
+ "policy/approxkl_avg": 0.09045101702213287,
725
+ "policy/clipfrac_avg": 0.298828125,
726
+ "policy/entropy_avg": 0.31827011704444885,
727
+ "step": 35,
728
+ "val/clipfrac_avg": 0.0,
729
+ "val/num_eos_tokens": 2304,
730
+ "val/ratio": 0.9999386072158813,
731
+ "val/ratio_var": 1.0199778444075491e-06
732
+ },
733
+ {
734
+ "episode": 18944,
735
+ "epoch": 0.08115147361206305,
736
+ "eps": 7,
737
+ "loss/policy_avg": 0.08208812773227692,
738
+ "lr": 2.578125e-06,
739
+ "objective/entropy": -53.41987609863281,
740
+ "objective/kl": 23.544723510742188,
741
+ "objective/non_score_reward": -1.1772363185882568,
742
+ "objective/rlhf_reward": 11.115821838378906,
743
+ "objective/scores": 12.293057441711426,
744
+ "policy/approxkl_avg": 0.08877880871295929,
745
+ "policy/clipfrac_avg": 0.296875,
746
+ "policy/entropy_avg": 0.31178680062294006,
747
+ "step": 36,
748
+ "val/clipfrac_avg": 0.0,
749
+ "val/num_eos_tokens": 2276,
750
+ "val/ratio": 1.000001072883606,
751
+ "val/ratio_var": 1.1726771163012017e-06
752
+ },
753
+ {
754
+ "episode": 19456,
755
+ "epoch": 0.08334475668265935,
756
+ "eps": 7,
757
+ "loss/policy_avg": 0.11828020215034485,
758
+ "lr": 2.56640625e-06,
759
+ "objective/entropy": -54.274024963378906,
760
+ "objective/kl": 23.336807250976562,
761
+ "objective/non_score_reward": -1.1668403148651123,
762
+ "objective/rlhf_reward": 11.09507942199707,
763
+ "objective/scores": 12.261919975280762,
764
+ "policy/approxkl_avg": 0.09653887152671814,
765
+ "policy/clipfrac_avg": 0.294921875,
766
+ "policy/entropy_avg": 0.32997897267341614,
767
+ "step": 37,
768
+ "val/clipfrac_avg": 0.0,
769
+ "val/num_eos_tokens": 2241,
770
+ "val/ratio": 1.000211477279663,
771
+ "val/ratio_var": 1.642234337850823e-06
772
+ },
773
+ {
774
+ "episode": 19968,
775
+ "epoch": 0.08553803975325565,
776
+ "eps": 7,
777
+ "loss/policy_avg": 0.11273995041847229,
778
+ "lr": 2.5546875000000003e-06,
779
+ "objective/entropy": -53.54825210571289,
780
+ "objective/kl": 23.62361717224121,
781
+ "objective/non_score_reward": -1.1811809539794922,
782
+ "objective/rlhf_reward": 11.013565063476562,
783
+ "objective/scores": 12.194746017456055,
784
+ "policy/approxkl_avg": 0.100711889564991,
785
+ "policy/clipfrac_avg": 0.291015625,
786
+ "policy/entropy_avg": 0.34581631422042847,
787
+ "step": 38,
788
+ "val/clipfrac_avg": 0.0,
789
+ "val/num_eos_tokens": 2229,
790
+ "val/ratio": 0.9998159408569336,
791
+ "val/ratio_var": 1.3141981298758765e-06
792
+ },
793
+ {
794
+ "episode": 20480,
795
+ "epoch": 0.08773132282385196,
796
+ "eps": 7,
797
+ "loss/policy_avg": 0.05108005180954933,
798
+ "lr": 2.54296875e-06,
799
+ "objective/entropy": -51.924232482910156,
800
+ "objective/kl": 23.902057647705078,
801
+ "objective/non_score_reward": -1.1951029300689697,
802
+ "objective/rlhf_reward": 11.001468658447266,
803
+ "objective/scores": 12.196571350097656,
804
+ "policy/approxkl_avg": 0.0894564539194107,
805
+ "policy/clipfrac_avg": 0.28515625,
806
+ "policy/entropy_avg": 0.3434683382511139,
807
+ "step": 39,
808
+ "val/clipfrac_avg": 0.0,
809
+ "val/num_eos_tokens": 2254,
810
+ "val/ratio": 1.0003117322921753,
811
+ "val/ratio_var": 1.1836557405331405e-06
812
+ },
813
+ {
814
+ "episode": 20992,
815
+ "epoch": 0.08992460589444826,
816
+ "eps": 7,
817
+ "loss/policy_avg": 0.09136471152305603,
818
+ "lr": 2.53125e-06,
819
+ "objective/entropy": -52.43147277832031,
820
+ "objective/kl": 25.310583114624023,
821
+ "objective/non_score_reward": -1.2655291557312012,
822
+ "objective/rlhf_reward": 10.961225509643555,
823
+ "objective/scores": 12.226754188537598,
824
+ "policy/approxkl_avg": 0.10527636855840683,
825
+ "policy/clipfrac_avg": 0.296875,
826
+ "policy/entropy_avg": 0.32169109582901,
827
+ "step": 40,
828
+ "val/clipfrac_avg": 0.0,
829
+ "val/num_eos_tokens": 2151,
830
+ "val/ratio": 0.9999606013298035,
831
+ "val/ratio_var": 1.430940642421774e-06
832
+ },
833
+ {
834
+ "episode": 21504,
835
+ "epoch": 0.09211788896504455,
836
+ "eps": 7,
837
+ "loss/policy_avg": 0.11661476641893387,
838
+ "lr": 2.5195312500000003e-06,
839
+ "objective/entropy": -46.344642639160156,
840
+ "objective/kl": 27.143455505371094,
841
+ "objective/non_score_reward": -1.3571728467941284,
842
+ "objective/rlhf_reward": 10.886428833007812,
843
+ "objective/scores": 12.243600845336914,
844
+ "policy/approxkl_avg": 0.11585399508476257,
845
+ "policy/clipfrac_avg": 0.28515625,
846
+ "policy/entropy_avg": 0.3539811074733734,
847
+ "step": 41,
848
+ "val/clipfrac_avg": 0.0,
849
+ "val/num_eos_tokens": 2093,
850
+ "val/ratio": 0.999966025352478,
851
+ "val/ratio_var": 2.110434706992237e-06
852
+ },
853
+ {
854
+ "episode": 22016,
855
+ "epoch": 0.09431117203564085,
856
+ "eps": 7,
857
+ "loss/policy_avg": 0.1412871778011322,
858
+ "lr": 2.5078125e-06,
859
+ "objective/entropy": -48.18741989135742,
860
+ "objective/kl": 26.137479782104492,
861
+ "objective/non_score_reward": -1.3068740367889404,
862
+ "objective/rlhf_reward": 11.087350845336914,
863
+ "objective/scores": 12.394224166870117,
864
+ "policy/approxkl_avg": 0.10230039805173874,
865
+ "policy/clipfrac_avg": 0.30859375,
866
+ "policy/entropy_avg": 0.35482996702194214,
867
+ "step": 42,
868
+ "val/clipfrac_avg": 0.0,
869
+ "val/num_eos_tokens": 1995,
870
+ "val/ratio": 1.0001914501190186,
871
+ "val/ratio_var": 1.8418863874103408e-06
872
+ },
873
+ {
874
+ "episode": 22528,
875
+ "epoch": 0.09650445510623715,
876
+ "eps": 7,
877
+ "loss/policy_avg": 0.14972682297229767,
878
+ "lr": 2.49609375e-06,
879
+ "objective/entropy": -47.18873977661133,
880
+ "objective/kl": 27.248594284057617,
881
+ "objective/non_score_reward": -1.3624296188354492,
882
+ "objective/rlhf_reward": 10.733315467834473,
883
+ "objective/scores": 12.095744132995605,
884
+ "policy/approxkl_avg": 0.11158671975135803,
885
+ "policy/clipfrac_avg": 0.2890625,
886
+ "policy/entropy_avg": 0.3710693418979645,
887
+ "step": 43,
888
+ "val/clipfrac_avg": 0.0,
889
+ "val/num_eos_tokens": 2105,
890
+ "val/ratio": 1.000127911567688,
891
+ "val/ratio_var": 1.7979554058911162e-06
892
+ },
893
+ {
894
+ "episode": 23040,
895
+ "epoch": 0.09869773817683344,
896
+ "eps": 7,
897
+ "loss/policy_avg": 0.1013510599732399,
898
+ "lr": 2.4843750000000002e-06,
899
+ "objective/entropy": -48.7869758605957,
900
+ "objective/kl": 26.826696395874023,
901
+ "objective/non_score_reward": -1.3413348197937012,
902
+ "objective/rlhf_reward": 10.954305648803711,
903
+ "objective/scores": 12.295639038085938,
904
+ "policy/approxkl_avg": 0.09727650135755539,
905
+ "policy/clipfrac_avg": 0.267578125,
906
+ "policy/entropy_avg": 0.341769278049469,
907
+ "step": 44,
908
+ "val/clipfrac_avg": 0.0,
909
+ "val/num_eos_tokens": 1957,
910
+ "val/ratio": 0.9999566078186035,
911
+ "val/ratio_var": 9.611487712390954e-07
912
+ },
913
+ {
914
+ "episode": 23552,
915
+ "epoch": 0.10089102124742974,
916
+ "eps": 7,
917
+ "loss/policy_avg": 0.09324927628040314,
918
+ "lr": 2.47265625e-06,
919
+ "objective/entropy": -48.091217041015625,
920
+ "objective/kl": 27.1876163482666,
921
+ "objective/non_score_reward": -1.3593807220458984,
922
+ "objective/rlhf_reward": 11.09292984008789,
923
+ "objective/scores": 12.452309608459473,
924
+ "policy/approxkl_avg": 0.09258678555488586,
925
+ "policy/clipfrac_avg": 0.294921875,
926
+ "policy/entropy_avg": 0.3322508931159973,
927
+ "step": 45,
928
+ "val/clipfrac_avg": 0.0,
929
+ "val/num_eos_tokens": 1981,
930
+ "val/ratio": 0.9999479055404663,
931
+ "val/ratio_var": 1.1157736707900767e-06
932
+ },
933
+ {
934
+ "episode": 24064,
935
+ "epoch": 0.10308430431802605,
936
+ "eps": 7,
937
+ "loss/policy_avg": 0.05657796561717987,
938
+ "lr": 2.4609375e-06,
939
+ "objective/entropy": -50.28386688232422,
940
+ "objective/kl": 26.767160415649414,
941
+ "objective/non_score_reward": -1.338357925415039,
942
+ "objective/rlhf_reward": 11.471373558044434,
943
+ "objective/scores": 12.809732437133789,
944
+ "policy/approxkl_avg": 0.08614060282707214,
945
+ "policy/clipfrac_avg": 0.28125,
946
+ "policy/entropy_avg": 0.32660749554634094,
947
+ "step": 46,
948
+ "val/clipfrac_avg": 0.0,
949
+ "val/num_eos_tokens": 1974,
950
+ "val/ratio": 1.000014305114746,
951
+ "val/ratio_var": 1.3160678236090462e-06
952
+ },
953
+ {
954
+ "episode": 24576,
955
+ "epoch": 0.10527758738862235,
956
+ "eps": 7,
957
+ "loss/policy_avg": 0.11685071885585785,
958
+ "lr": 2.4492187500000002e-06,
959
+ "objective/entropy": -47.901275634765625,
960
+ "objective/kl": 28.288379669189453,
961
+ "objective/non_score_reward": -1.414419174194336,
962
+ "objective/rlhf_reward": 11.333412170410156,
963
+ "objective/scores": 12.747832298278809,
964
+ "policy/approxkl_avg": 0.10291337966918945,
965
+ "policy/clipfrac_avg": 0.25390625,
966
+ "policy/entropy_avg": 0.31352245807647705,
967
+ "step": 47,
968
+ "val/clipfrac_avg": 0.0,
969
+ "val/num_eos_tokens": 1895,
970
+ "val/ratio": 1.0002663135528564,
971
+ "val/ratio_var": 1.1478540500320378e-06
972
+ },
973
+ {
974
+ "episode": 25088,
975
+ "epoch": 0.10747087045921865,
976
+ "eps": 7,
977
+ "loss/policy_avg": 0.09044268727302551,
978
+ "lr": 2.4375e-06,
979
+ "objective/entropy": -46.25014877319336,
980
+ "objective/kl": 29.569900512695312,
981
+ "objective/non_score_reward": -1.4784950017929077,
982
+ "objective/rlhf_reward": 11.207091331481934,
983
+ "objective/scores": 12.685586929321289,
984
+ "policy/approxkl_avg": 0.08932553231716156,
985
+ "policy/clipfrac_avg": 0.2578125,
986
+ "policy/entropy_avg": 0.30406129360198975,
987
+ "step": 48,
988
+ "val/clipfrac_avg": 0.0,
989
+ "val/num_eos_tokens": 1779,
990
+ "val/ratio": 0.9998706579208374,
991
+ "val/ratio_var": 9.449911431147484e-07
992
+ },
993
+ {
994
+ "episode": 25600,
995
+ "epoch": 0.10966415352981494,
996
+ "eps": 7,
997
+ "loss/policy_avg": 0.0890183225274086,
998
+ "lr": 2.42578125e-06,
999
+ "objective/entropy": -48.26085662841797,
1000
+ "objective/kl": 30.082365036010742,
1001
+ "objective/non_score_reward": -1.5041183233261108,
1002
+ "objective/rlhf_reward": 11.227607727050781,
1003
+ "objective/scores": 12.731725692749023,
1004
+ "policy/approxkl_avg": 0.08349761366844177,
1005
+ "policy/clipfrac_avg": 0.302734375,
1006
+ "policy/entropy_avg": 0.2885536849498749,
1007
+ "step": 49,
1008
+ "val/clipfrac_avg": 0.0,
1009
+ "val/num_eos_tokens": 1885,
1010
+ "val/ratio": 1.000079870223999,
1011
+ "val/ratio_var": 1.2881323527835775e-06
1012
+ },
1013
+ {
1014
+ "episode": 26112,
1015
+ "epoch": 0.11185743660041124,
1016
+ "eps": 7,
1017
+ "loss/policy_avg": 0.1166183277964592,
1018
+ "lr": 2.4140625000000002e-06,
1019
+ "objective/entropy": -45.06092071533203,
1020
+ "objective/kl": 30.306961059570312,
1021
+ "objective/non_score_reward": -1.515347957611084,
1022
+ "objective/rlhf_reward": 11.126758575439453,
1023
+ "objective/scores": 12.642105102539062,
1024
+ "policy/approxkl_avg": 0.08703693002462387,
1025
+ "policy/clipfrac_avg": 0.279296875,
1026
+ "policy/entropy_avg": 0.281158983707428,
1027
+ "step": 50,
1028
+ "val/clipfrac_avg": 0.0,
1029
+ "val/num_eos_tokens": 1857,
1030
+ "val/ratio": 1.000382423400879,
1031
+ "val/ratio_var": 1.3761283526037005e-06
1032
+ },
1033
+ {
1034
+ "episode": 26624,
1035
+ "epoch": 0.11405071967100754,
1036
+ "eps": 7,
1037
+ "loss/policy_avg": 0.09397272765636444,
1038
+ "lr": 2.40234375e-06,
1039
+ "objective/entropy": -47.06073760986328,
1040
+ "objective/kl": 30.048595428466797,
1041
+ "objective/non_score_reward": -1.502429723739624,
1042
+ "objective/rlhf_reward": 11.373811721801758,
1043
+ "objective/scores": 12.876239776611328,
1044
+ "policy/approxkl_avg": 0.07360626757144928,
1045
+ "policy/clipfrac_avg": 0.23828125,
1046
+ "policy/entropy_avg": 0.25442755222320557,
1047
+ "step": 51,
1048
+ "val/clipfrac_avg": 0.0,
1049
+ "val/num_eos_tokens": 1919,
1050
+ "val/ratio": 1.0000107288360596,
1051
+ "val/ratio_var": 1.3316806644070311e-06
1052
+ },
1053
+ {
1054
+ "episode": 27136,
1055
+ "epoch": 0.11624400274160383,
1056
+ "eps": 7,
1057
+ "loss/policy_avg": 0.05725743621587753,
1058
+ "lr": 2.390625e-06,
1059
+ "objective/entropy": -50.653358459472656,
1060
+ "objective/kl": 28.8651123046875,
1061
+ "objective/non_score_reward": -1.4432556629180908,
1062
+ "objective/rlhf_reward": 11.409313201904297,
1063
+ "objective/scores": 12.852569580078125,
1064
+ "policy/approxkl_avg": 0.0559498593211174,
1065
+ "policy/clipfrac_avg": 0.23828125,
1066
+ "policy/entropy_avg": 0.2538541257381439,
1067
+ "step": 52,
1068
+ "val/clipfrac_avg": 0.0,
1069
+ "val/num_eos_tokens": 2038,
1070
+ "val/ratio": 1.0001037120819092,
1071
+ "val/ratio_var": 1.2117743608541787e-06
1072
+ },
1073
+ {
1074
+ "episode": 27648,
1075
+ "epoch": 0.11843728581220014,
1076
+ "eps": 7,
1077
+ "loss/policy_avg": 0.04747297987341881,
1078
+ "lr": 2.3789062500000002e-06,
1079
+ "objective/entropy": -49.83374786376953,
1080
+ "objective/kl": 29.47016716003418,
1081
+ "objective/non_score_reward": -1.473508358001709,
1082
+ "objective/rlhf_reward": 11.507312774658203,
1083
+ "objective/scores": 12.98082160949707,
1084
+ "policy/approxkl_avg": 0.07053554803133011,
1085
+ "policy/clipfrac_avg": 0.244140625,
1086
+ "policy/entropy_avg": 0.23207688331604004,
1087
+ "step": 53,
1088
+ "val/clipfrac_avg": 0.0,
1089
+ "val/num_eos_tokens": 2084,
1090
+ "val/ratio": 0.9998562335968018,
1091
+ "val/ratio_var": 7.929276648610539e-07
1092
+ },
1093
+ {
1094
+ "episode": 28160,
1095
+ "epoch": 0.12063056888279644,
1096
+ "eps": 7,
1097
+ "loss/policy_avg": 0.11368705332279205,
1098
+ "lr": 2.3671875e-06,
1099
+ "objective/entropy": -51.63511657714844,
1100
+ "objective/kl": 28.794090270996094,
1101
+ "objective/non_score_reward": -1.4397046566009521,
1102
+ "objective/rlhf_reward": 11.756919860839844,
1103
+ "objective/scores": 13.196623802185059,
1104
+ "policy/approxkl_avg": 0.06422245502471924,
1105
+ "policy/clipfrac_avg": 0.203125,
1106
+ "policy/entropy_avg": 0.21511417627334595,
1107
+ "step": 54,
1108
+ "val/clipfrac_avg": 0.0,
1109
+ "val/num_eos_tokens": 2029,
1110
+ "val/ratio": 0.9999565482139587,
1111
+ "val/ratio_var": 9.939893743649009e-07
1112
+ },
1113
+ {
1114
+ "episode": 28672,
1115
+ "epoch": 0.12282385195339274,
1116
+ "eps": 7,
1117
+ "loss/policy_avg": 0.0857730507850647,
1118
+ "lr": 2.35546875e-06,
1119
+ "objective/entropy": -48.501441955566406,
1120
+ "objective/kl": 29.051284790039062,
1121
+ "objective/non_score_reward": -1.4525642395019531,
1122
+ "objective/rlhf_reward": 11.569185256958008,
1123
+ "objective/scores": 13.021749496459961,
1124
+ "policy/approxkl_avg": 0.05940084159374237,
1125
+ "policy/clipfrac_avg": 0.193359375,
1126
+ "policy/entropy_avg": 0.20681166648864746,
1127
+ "step": 55,
1128
+ "val/clipfrac_avg": 0.0,
1129
+ "val/num_eos_tokens": 2137,
1130
+ "val/ratio": 0.9998863935470581,
1131
+ "val/ratio_var": 7.905163101895596e-07
1132
+ },
1133
+ {
1134
+ "episode": 29184,
1135
+ "epoch": 0.12501713502398903,
1136
+ "eps": 7,
1137
+ "loss/policy_avg": 0.06318055093288422,
1138
+ "lr": 2.3437500000000002e-06,
1139
+ "objective/entropy": -52.91703414916992,
1140
+ "objective/kl": 28.871070861816406,
1141
+ "objective/non_score_reward": -1.4435535669326782,
1142
+ "objective/rlhf_reward": 11.726110458374023,
1143
+ "objective/scores": 13.169663429260254,
1144
+ "policy/approxkl_avg": 0.0693005621433258,
1145
+ "policy/clipfrac_avg": 0.236328125,
1146
+ "policy/entropy_avg": 0.2014627903699875,
1147
+ "step": 56,
1148
+ "val/clipfrac_avg": 0.0,
1149
+ "val/num_eos_tokens": 2150,
1150
+ "val/ratio": 0.9999423027038574,
1151
+ "val/ratio_var": 9.004759249364724e-07
1152
+ },
1153
+ {
1154
+ "episode": 29696,
1155
+ "epoch": 0.12721041809458533,
1156
+ "eps": 7,
1157
+ "loss/policy_avg": 0.025315172970294952,
1158
+ "lr": 2.33203125e-06,
1159
+ "objective/entropy": -51.94420623779297,
1160
+ "objective/kl": 28.651988983154297,
1161
+ "objective/non_score_reward": -1.432599425315857,
1162
+ "objective/rlhf_reward": 11.433904647827148,
1163
+ "objective/scores": 12.86650276184082,
1164
+ "policy/approxkl_avg": 0.055910706520080566,
1165
+ "policy/clipfrac_avg": 0.203125,
1166
+ "policy/entropy_avg": 0.21284107863903046,
1167
+ "step": 57,
1168
+ "val/clipfrac_avg": 0.0,
1169
+ "val/num_eos_tokens": 2193,
1170
+ "val/ratio": 1.000023603439331,
1171
+ "val/ratio_var": 7.501536742893222e-07
1172
+ },
1173
+ {
1174
+ "episode": 30208,
1175
+ "epoch": 0.12940370116518163,
1176
+ "eps": 7,
1177
+ "loss/policy_avg": 0.04747464135289192,
1178
+ "lr": 2.3203125e-06,
1179
+ "objective/entropy": -54.25482940673828,
1180
+ "objective/kl": 28.553482055664062,
1181
+ "objective/non_score_reward": -1.4276741743087769,
1182
+ "objective/rlhf_reward": 11.46591567993164,
1183
+ "objective/scores": 12.893589973449707,
1184
+ "policy/approxkl_avg": 0.04430193454027176,
1185
+ "policy/clipfrac_avg": 0.181640625,
1186
+ "policy/entropy_avg": 0.197004497051239,
1187
+ "step": 58,
1188
+ "val/clipfrac_avg": 0.0,
1189
+ "val/num_eos_tokens": 2182,
1190
+ "val/ratio": 1.0000498294830322,
1191
+ "val/ratio_var": 5.298446126289491e-07
1192
+ },
1193
+ {
1194
+ "episode": 30720,
1195
+ "epoch": 0.13159698423577793,
1196
+ "eps": 7,
1197
+ "loss/policy_avg": 0.07828865945339203,
1198
+ "lr": 2.30859375e-06,
1199
+ "objective/entropy": -55.69086837768555,
1200
+ "objective/kl": 27.18333625793457,
1201
+ "objective/non_score_reward": -1.3591668605804443,
1202
+ "objective/rlhf_reward": 11.24705696105957,
1203
+ "objective/scores": 12.606224060058594,
1204
+ "policy/approxkl_avg": 0.050338346511125565,
1205
+ "policy/clipfrac_avg": 0.177734375,
1206
+ "policy/entropy_avg": 0.19780844449996948,
1207
+ "step": 59,
1208
+ "val/clipfrac_avg": 0.0,
1209
+ "val/num_eos_tokens": 2173,
1210
+ "val/ratio": 0.9999244213104248,
1211
+ "val/ratio_var": 8.30796864192962e-07
1212
+ },
1213
+ {
1214
+ "episode": 31232,
1215
+ "epoch": 0.13379026730637422,
1216
+ "eps": 7,
1217
+ "loss/policy_avg": 0.06892979145050049,
1218
+ "lr": 2.296875e-06,
1219
+ "objective/entropy": -55.50807189941406,
1220
+ "objective/kl": 27.259765625,
1221
+ "objective/non_score_reward": -1.3629883527755737,
1222
+ "objective/rlhf_reward": 11.379904747009277,
1223
+ "objective/scores": 12.74289321899414,
1224
+ "policy/approxkl_avg": 0.05562983453273773,
1225
+ "policy/clipfrac_avg": 0.1953125,
1226
+ "policy/entropy_avg": 0.2033534049987793,
1227
+ "step": 60,
1228
+ "val/clipfrac_avg": 0.0,
1229
+ "val/num_eos_tokens": 2124,
1230
+ "val/ratio": 0.9998317956924438,
1231
+ "val/ratio_var": 7.12839153038658e-07
1232
+ },
1233
+ {
1234
+ "episode": 31744,
1235
+ "epoch": 0.13598355037697052,
1236
+ "eps": 7,
1237
+ "loss/policy_avg": 0.06427884846925735,
1238
+ "lr": 2.28515625e-06,
1239
+ "objective/entropy": -53.272911071777344,
1240
+ "objective/kl": 28.615087509155273,
1241
+ "objective/non_score_reward": -1.4307544231414795,
1242
+ "objective/rlhf_reward": 11.401360511779785,
1243
+ "objective/scores": 12.832115173339844,
1244
+ "policy/approxkl_avg": 0.05360804870724678,
1245
+ "policy/clipfrac_avg": 0.181640625,
1246
+ "policy/entropy_avg": 0.20671643316745758,
1247
+ "step": 61,
1248
+ "val/clipfrac_avg": 0.0,
1249
+ "val/num_eos_tokens": 2243,
1250
+ "val/ratio": 0.9998288750648499,
1251
+ "val/ratio_var": 7.739292300357192e-07
1252
+ },
1253
+ {
1254
+ "episode": 32256,
1255
+ "epoch": 0.13817683344756682,
1256
+ "eps": 7,
1257
+ "loss/policy_avg": 0.036871928721666336,
1258
+ "lr": 2.2734375e-06,
1259
+ "objective/entropy": -56.196807861328125,
1260
+ "objective/kl": 26.758060455322266,
1261
+ "objective/non_score_reward": -1.3379030227661133,
1262
+ "objective/rlhf_reward": 11.421424865722656,
1263
+ "objective/scores": 12.759328842163086,
1264
+ "policy/approxkl_avg": 0.06217839941382408,
1265
+ "policy/clipfrac_avg": 0.224609375,
1266
+ "policy/entropy_avg": 0.20198120176792145,
1267
+ "step": 62,
1268
+ "val/clipfrac_avg": 0.0,
1269
+ "val/num_eos_tokens": 2232,
1270
+ "val/ratio": 0.9998797178268433,
1271
+ "val/ratio_var": 9.791126558411634e-07
1272
+ },
1273
+ {
1274
+ "episode": 32768,
1275
+ "epoch": 0.1403701165181631,
1276
+ "eps": 7,
1277
+ "loss/policy_avg": 0.05493534356355667,
1278
+ "lr": 2.26171875e-06,
1279
+ "objective/entropy": -56.74695587158203,
1280
+ "objective/kl": 27.563007354736328,
1281
+ "objective/non_score_reward": -1.378150463104248,
1282
+ "objective/rlhf_reward": 11.451597213745117,
1283
+ "objective/scores": 12.829747200012207,
1284
+ "policy/approxkl_avg": 0.0605655238032341,
1285
+ "policy/clipfrac_avg": 0.205078125,
1286
+ "policy/entropy_avg": 0.21349653601646423,
1287
+ "step": 63,
1288
+ "val/clipfrac_avg": 0.0,
1289
+ "val/num_eos_tokens": 2429,
1290
+ "val/ratio": 1.0000948905944824,
1291
+ "val/ratio_var": 8.029837204048818e-07
1292
+ },
1293
+ {
1294
+ "episode": 33280,
1295
+ "epoch": 0.14256339958875944,
1296
+ "eps": 7,
1297
+ "loss/policy_avg": 0.06948225200176239,
1298
+ "lr": 2.25e-06,
1299
+ "objective/entropy": -55.68669128417969,
1300
+ "objective/kl": 27.24061393737793,
1301
+ "objective/non_score_reward": -1.3620307445526123,
1302
+ "objective/rlhf_reward": 11.590509414672852,
1303
+ "objective/scores": 12.95254135131836,
1304
+ "policy/approxkl_avg": 0.058622099459171295,
1305
+ "policy/clipfrac_avg": 0.2109375,
1306
+ "policy/entropy_avg": 0.20613785088062286,
1307
+ "step": 64,
1308
+ "val/clipfrac_avg": 0.0,
1309
+ "val/num_eos_tokens": 2500,
1310
+ "val/ratio": 1.0000636577606201,
1311
+ "val/ratio_var": 1.0106417676070123e-06
1312
+ },
1313
+ {
1314
+ "episode": 33792,
1315
+ "epoch": 0.14475668265935573,
1316
+ "eps": 7,
1317
+ "loss/policy_avg": 0.0674080029129982,
1318
+ "lr": 2.23828125e-06,
1319
+ "objective/entropy": -56.052215576171875,
1320
+ "objective/kl": 27.84102439880371,
1321
+ "objective/non_score_reward": -1.3920512199401855,
1322
+ "objective/rlhf_reward": 11.380613327026367,
1323
+ "objective/scores": 12.772663116455078,
1324
+ "policy/approxkl_avg": 0.05588747560977936,
1325
+ "policy/clipfrac_avg": 0.1875,
1326
+ "policy/entropy_avg": 0.21730531752109528,
1327
+ "step": 65,
1328
+ "val/clipfrac_avg": 0.0,
1329
+ "val/num_eos_tokens": 2533,
1330
+ "val/ratio": 1.000044822692871,
1331
+ "val/ratio_var": 1.0719528518166044e-06
1332
+ },
1333
+ {
1334
+ "episode": 34304,
1335
+ "epoch": 0.14694996572995203,
1336
+ "eps": 7,
1337
+ "loss/policy_avg": 0.1223275363445282,
1338
+ "lr": 2.2265625e-06,
1339
+ "objective/entropy": -55.07699966430664,
1340
+ "objective/kl": 28.410842895507812,
1341
+ "objective/non_score_reward": -1.4205421209335327,
1342
+ "objective/rlhf_reward": 11.466856002807617,
1343
+ "objective/scores": 12.887397766113281,
1344
+ "policy/approxkl_avg": 0.07928226888179779,
1345
+ "policy/clipfrac_avg": 0.189453125,
1346
+ "policy/entropy_avg": 0.22597652673721313,
1347
+ "step": 66,
1348
+ "val/clipfrac_avg": 0.0,
1349
+ "val/num_eos_tokens": 2586,
1350
+ "val/ratio": 1.000163197517395,
1351
+ "val/ratio_var": 1.0074684269056888e-06
1352
+ },
1353
+ {
1354
+ "episode": 34816,
1355
+ "epoch": 0.14914324880054833,
1356
+ "eps": 7,
1357
+ "loss/policy_avg": 0.06255296617746353,
1358
+ "lr": 2.21484375e-06,
1359
+ "objective/entropy": -56.23228454589844,
1360
+ "objective/kl": 27.560832977294922,
1361
+ "objective/non_score_reward": -1.3780417442321777,
1362
+ "objective/rlhf_reward": 11.549421310424805,
1363
+ "objective/scores": 12.927461624145508,
1364
+ "policy/approxkl_avg": 0.07416369020938873,
1365
+ "policy/clipfrac_avg": 0.2109375,
1366
+ "policy/entropy_avg": 0.22490039467811584,
1367
+ "step": 67,
1368
+ "val/clipfrac_avg": 0.0,
1369
+ "val/num_eos_tokens": 2561,
1370
+ "val/ratio": 0.9998071193695068,
1371
+ "val/ratio_var": 7.868264901844668e-07
1372
+ },
1373
+ {
1374
+ "episode": 35328,
1375
+ "epoch": 0.15133653187114463,
1376
+ "eps": 7,
1377
+ "loss/policy_avg": 0.07749070227146149,
1378
+ "lr": 2.203125e-06,
1379
+ "objective/entropy": -58.02006912231445,
1380
+ "objective/kl": 28.338050842285156,
1381
+ "objective/non_score_reward": -1.4169025421142578,
1382
+ "objective/rlhf_reward": 11.502689361572266,
1383
+ "objective/scores": 12.919590950012207,
1384
+ "policy/approxkl_avg": 0.050264619290828705,
1385
+ "policy/clipfrac_avg": 0.17578125,
1386
+ "policy/entropy_avg": 0.22556829452514648,
1387
+ "step": 68,
1388
+ "val/clipfrac_avg": 0.0,
1389
+ "val/num_eos_tokens": 2776,
1390
+ "val/ratio": 0.9998827576637268,
1391
+ "val/ratio_var": 6.7324697283766e-07
1392
+ },
1393
+ {
1394
+ "episode": 35840,
1395
+ "epoch": 0.15352981494174092,
1396
+ "eps": 7,
1397
+ "loss/policy_avg": 0.04901011660695076,
1398
+ "lr": 2.19140625e-06,
1399
+ "objective/entropy": -57.22145462036133,
1400
+ "objective/kl": 28.202184677124023,
1401
+ "objective/non_score_reward": -1.410109281539917,
1402
+ "objective/rlhf_reward": 11.595409393310547,
1403
+ "objective/scores": 13.005517959594727,
1404
+ "policy/approxkl_avg": 0.05339674651622772,
1405
+ "policy/clipfrac_avg": 0.197265625,
1406
+ "policy/entropy_avg": 0.2128574401140213,
1407
+ "step": 69,
1408
+ "val/clipfrac_avg": 0.0,
1409
+ "val/num_eos_tokens": 2659,
1410
+ "val/ratio": 0.9999070167541504,
1411
+ "val/ratio_var": 6.147514000076626e-07
1412
+ },
1413
+ {
1414
+ "episode": 36352,
1415
+ "epoch": 0.15572309801233722,
1416
+ "eps": 7,
1417
+ "loss/policy_avg": 0.04001376032829285,
1418
+ "lr": 2.1796875e-06,
1419
+ "objective/entropy": -55.01459503173828,
1420
+ "objective/kl": 29.15834617614746,
1421
+ "objective/non_score_reward": -1.4579174518585205,
1422
+ "objective/rlhf_reward": 11.658880233764648,
1423
+ "objective/scores": 13.11679744720459,
1424
+ "policy/approxkl_avg": 0.0480184480547905,
1425
+ "policy/clipfrac_avg": 0.208984375,
1426
+ "policy/entropy_avg": 0.20534545183181763,
1427
+ "step": 70,
1428
+ "val/clipfrac_avg": 0.0,
1429
+ "val/num_eos_tokens": 2526,
1430
+ "val/ratio": 1.0001248121261597,
1431
+ "val/ratio_var": 7.817943696863949e-07
1432
+ },
1433
+ {
1434
+ "episode": 36864,
1435
+ "epoch": 0.15791638108293352,
1436
+ "eps": 7,
1437
+ "loss/policy_avg": 0.0806807279586792,
1438
+ "lr": 2.16796875e-06,
1439
+ "objective/entropy": -54.07442092895508,
1440
+ "objective/kl": 28.838031768798828,
1441
+ "objective/non_score_reward": -1.4419015645980835,
1442
+ "objective/rlhf_reward": 11.803167343139648,
1443
+ "objective/scores": 13.24506950378418,
1444
+ "policy/approxkl_avg": 0.05658765137195587,
1445
+ "policy/clipfrac_avg": 0.16796875,
1446
+ "policy/entropy_avg": 0.22964796423912048,
1447
+ "step": 71,
1448
+ "val/clipfrac_avg": 0.0,
1449
+ "val/num_eos_tokens": 2643,
1450
+ "val/ratio": 1.000180721282959,
1451
+ "val/ratio_var": 8.247698701779882e-07
1452
+ },
1453
+ {
1454
+ "episode": 37376,
1455
+ "epoch": 0.1601096641535298,
1456
+ "eps": 7,
1457
+ "loss/policy_avg": 0.051692895591259,
1458
+ "lr": 2.15625e-06,
1459
+ "objective/entropy": -54.76805114746094,
1460
+ "objective/kl": 29.931236267089844,
1461
+ "objective/non_score_reward": -1.4965617656707764,
1462
+ "objective/rlhf_reward": 11.834163665771484,
1463
+ "objective/scores": 13.330726623535156,
1464
+ "policy/approxkl_avg": 0.0646805614233017,
1465
+ "policy/clipfrac_avg": 0.19140625,
1466
+ "policy/entropy_avg": 0.2366558015346527,
1467
+ "step": 72,
1468
+ "val/clipfrac_avg": 0.0,
1469
+ "val/num_eos_tokens": 2777,
1470
+ "val/ratio": 0.9999984502792358,
1471
+ "val/ratio_var": 1.2423626003510435e-06
1472
+ },
1473
+ {
1474
+ "episode": 37888,
1475
+ "epoch": 0.1623029472241261,
1476
+ "eps": 7,
1477
+ "loss/policy_avg": 0.05296178162097931,
1478
+ "lr": 2.14453125e-06,
1479
+ "objective/entropy": -52.60663604736328,
1480
+ "objective/kl": 29.616146087646484,
1481
+ "objective/non_score_reward": -1.4808073043823242,
1482
+ "objective/rlhf_reward": 11.590347290039062,
1483
+ "objective/scores": 13.071154594421387,
1484
+ "policy/approxkl_avg": 0.05852410942316055,
1485
+ "policy/clipfrac_avg": 0.203125,
1486
+ "policy/entropy_avg": 0.22921660542488098,
1487
+ "step": 73,
1488
+ "val/clipfrac_avg": 0.0,
1489
+ "val/num_eos_tokens": 2599,
1490
+ "val/ratio": 0.9999311566352844,
1491
+ "val/ratio_var": 9.6972451046895e-07
1492
+ },
1493
+ {
1494
+ "episode": 38400,
1495
+ "epoch": 0.1644962302947224,
1496
+ "eps": 7,
1497
+ "loss/policy_avg": 0.05771177262067795,
1498
+ "lr": 2.1328125e-06,
1499
+ "objective/entropy": -50.87690734863281,
1500
+ "objective/kl": 31.555557250976562,
1501
+ "objective/non_score_reward": -1.5777779817581177,
1502
+ "objective/rlhf_reward": 11.474925994873047,
1503
+ "objective/scores": 13.052703857421875,
1504
+ "policy/approxkl_avg": 0.06892445683479309,
1505
+ "policy/clipfrac_avg": 0.23828125,
1506
+ "policy/entropy_avg": 0.2290726602077484,
1507
+ "step": 74,
1508
+ "val/clipfrac_avg": 0.0,
1509
+ "val/num_eos_tokens": 2545,
1510
+ "val/ratio": 1.0000228881835938,
1511
+ "val/ratio_var": 8.888110301086272e-07
1512
+ },
1513
+ {
1514
+ "episode": 38912,
1515
+ "epoch": 0.1666895133653187,
1516
+ "eps": 7,
1517
+ "loss/policy_avg": 0.08321105688810349,
1518
+ "lr": 2.12109375e-06,
1519
+ "objective/entropy": -49.94180679321289,
1520
+ "objective/kl": 31.343868255615234,
1521
+ "objective/non_score_reward": -1.5671935081481934,
1522
+ "objective/rlhf_reward": 11.681865692138672,
1523
+ "objective/scores": 13.249059677124023,
1524
+ "policy/approxkl_avg": 0.06739464402198792,
1525
+ "policy/clipfrac_avg": 0.205078125,
1526
+ "policy/entropy_avg": 0.22561725974082947,
1527
+ "step": 75,
1528
+ "val/clipfrac_avg": 0.0,
1529
+ "val/num_eos_tokens": 2453,
1530
+ "val/ratio": 1.0000121593475342,
1531
+ "val/ratio_var": 1.293211198571953e-06
1532
+ },
1533
+ {
1534
+ "episode": 39424,
1535
+ "epoch": 0.168882796435915,
1536
+ "eps": 7,
1537
+ "loss/policy_avg": 0.04158315062522888,
1538
+ "lr": 2.109375e-06,
1539
+ "objective/entropy": -51.920650482177734,
1540
+ "objective/kl": 30.741943359375,
1541
+ "objective/non_score_reward": -1.5370972156524658,
1542
+ "objective/rlhf_reward": 11.58804988861084,
1543
+ "objective/scores": 13.125146865844727,
1544
+ "policy/approxkl_avg": 0.06920449435710907,
1545
+ "policy/clipfrac_avg": 0.2265625,
1546
+ "policy/entropy_avg": 0.23201118409633636,
1547
+ "step": 76,
1548
+ "val/clipfrac_avg": 0.0,
1549
+ "val/num_eos_tokens": 2530,
1550
+ "val/ratio": 1.000053882598877,
1551
+ "val/ratio_var": 1.389662315887108e-06
1552
+ },
1553
+ {
1554
+ "episode": 39936,
1555
+ "epoch": 0.1710760795065113,
1556
+ "eps": 7,
1557
+ "loss/policy_avg": 0.053401652723550797,
1558
+ "lr": 2.09765625e-06,
1559
+ "objective/entropy": -54.08863830566406,
1560
+ "objective/kl": 29.898433685302734,
1561
+ "objective/non_score_reward": -1.4949218034744263,
1562
+ "objective/rlhf_reward": 11.703346252441406,
1563
+ "objective/scores": 13.19826889038086,
1564
+ "policy/approxkl_avg": 0.05423176288604736,
1565
+ "policy/clipfrac_avg": 0.185546875,
1566
+ "policy/entropy_avg": 0.23785756528377533,
1567
+ "step": 77,
1568
+ "val/clipfrac_avg": 0.0,
1569
+ "val/num_eos_tokens": 2552,
1570
+ "val/ratio": 1.000126838684082,
1571
+ "val/ratio_var": 9.200340400639107e-07
1572
+ },
1573
+ {
1574
+ "episode": 40448,
1575
+ "epoch": 0.17326936257710762,
1576
+ "eps": 7,
1577
+ "loss/policy_avg": 0.016359925270080566,
1578
+ "lr": 2.0859375e-06,
1579
+ "objective/entropy": -52.485328674316406,
1580
+ "objective/kl": 31.477571487426758,
1581
+ "objective/non_score_reward": -1.5738786458969116,
1582
+ "objective/rlhf_reward": 11.632445335388184,
1583
+ "objective/scores": 13.206323623657227,
1584
+ "policy/approxkl_avg": 0.05758652836084366,
1585
+ "policy/clipfrac_avg": 0.244140625,
1586
+ "policy/entropy_avg": 0.22791028022766113,
1587
+ "step": 78,
1588
+ "val/clipfrac_avg": 0.0,
1589
+ "val/num_eos_tokens": 2451,
1590
+ "val/ratio": 0.9999771118164062,
1591
+ "val/ratio_var": 1.3394280813372461e-06
1592
+ },
1593
+ {
1594
+ "episode": 40960,
1595
+ "epoch": 0.17546264564770392,
1596
+ "eps": 7,
1597
+ "loss/policy_avg": 0.04443100094795227,
1598
+ "lr": 2.07421875e-06,
1599
+ "objective/entropy": -52.247257232666016,
1600
+ "objective/kl": 31.95646095275879,
1601
+ "objective/non_score_reward": -1.597823143005371,
1602
+ "objective/rlhf_reward": 11.39157485961914,
1603
+ "objective/scores": 12.989398002624512,
1604
+ "policy/approxkl_avg": 0.062236685305833817,
1605
+ "policy/clipfrac_avg": 0.2421875,
1606
+ "policy/entropy_avg": 0.2292611300945282,
1607
+ "step": 79,
1608
+ "val/clipfrac_avg": 0.0,
1609
+ "val/num_eos_tokens": 2326,
1610
+ "val/ratio": 0.9999898672103882,
1611
+ "val/ratio_var": 1.2004873042315012e-06
1612
+ },
1613
+ {
1614
+ "episode": 41472,
1615
+ "epoch": 0.17765592871830022,
1616
+ "eps": 7,
1617
+ "loss/policy_avg": 0.06968129426240921,
1618
+ "lr": 2.0625e-06,
1619
+ "objective/entropy": -53.13126754760742,
1620
+ "objective/kl": 31.035659790039062,
1621
+ "objective/non_score_reward": -1.5517830848693848,
1622
+ "objective/rlhf_reward": 11.682025909423828,
1623
+ "objective/scores": 13.233808517456055,
1624
+ "policy/approxkl_avg": 0.05690852180123329,
1625
+ "policy/clipfrac_avg": 0.20703125,
1626
+ "policy/entropy_avg": 0.22946974635124207,
1627
+ "step": 80,
1628
+ "val/clipfrac_avg": 0.0,
1629
+ "val/num_eos_tokens": 2409,
1630
+ "val/ratio": 0.9998325109481812,
1631
+ "val/ratio_var": 6.52003791401512e-07
1632
+ },
1633
+ {
1634
+ "episode": 41984,
1635
+ "epoch": 0.1798492117888965,
1636
+ "eps": 7,
1637
+ "loss/policy_avg": 0.05149594694375992,
1638
+ "lr": 2.05078125e-06,
1639
+ "objective/entropy": -53.41246032714844,
1640
+ "objective/kl": 30.750011444091797,
1641
+ "objective/non_score_reward": -1.5375006198883057,
1642
+ "objective/rlhf_reward": 11.637171745300293,
1643
+ "objective/scores": 13.17467212677002,
1644
+ "policy/approxkl_avg": 0.07379427552223206,
1645
+ "policy/clipfrac_avg": 0.224609375,
1646
+ "policy/entropy_avg": 0.23126959800720215,
1647
+ "step": 81,
1648
+ "val/clipfrac_avg": 0.0,
1649
+ "val/num_eos_tokens": 2298,
1650
+ "val/ratio": 1.0000648498535156,
1651
+ "val/ratio_var": 1.0951293916150462e-06
1652
+ },
1653
+ {
1654
+ "episode": 42496,
1655
+ "epoch": 0.1820424948594928,
1656
+ "eps": 7,
1657
+ "loss/policy_avg": 0.1227312758564949,
1658
+ "lr": 2.0390625e-06,
1659
+ "objective/entropy": -51.953941345214844,
1660
+ "objective/kl": 31.699325561523438,
1661
+ "objective/non_score_reward": -1.5849663019180298,
1662
+ "objective/rlhf_reward": 11.240765571594238,
1663
+ "objective/scores": 12.82573127746582,
1664
+ "policy/approxkl_avg": 0.08355057239532471,
1665
+ "policy/clipfrac_avg": 0.224609375,
1666
+ "policy/entropy_avg": 0.21501347422599792,
1667
+ "step": 82,
1668
+ "val/clipfrac_avg": 0.0,
1669
+ "val/num_eos_tokens": 2209,
1670
+ "val/ratio": 1.0003668069839478,
1671
+ "val/ratio_var": 1.6002859410946257e-05
1672
+ },
1673
+ {
1674
+ "episode": 43008,
1675
+ "epoch": 0.1842357779300891,
1676
+ "eps": 7,
1677
+ "loss/policy_avg": 0.035134680569171906,
1678
+ "lr": 2.02734375e-06,
1679
+ "objective/entropy": -51.80181121826172,
1680
+ "objective/kl": 31.36351776123047,
1681
+ "objective/non_score_reward": -1.568176031112671,
1682
+ "objective/rlhf_reward": 11.594715118408203,
1683
+ "objective/scores": 13.162890434265137,
1684
+ "policy/approxkl_avg": 0.05513919144868851,
1685
+ "policy/clipfrac_avg": 0.212890625,
1686
+ "policy/entropy_avg": 0.2203870713710785,
1687
+ "step": 83,
1688
+ "val/clipfrac_avg": 0.0,
1689
+ "val/num_eos_tokens": 2286,
1690
+ "val/ratio": 0.999825119972229,
1691
+ "val/ratio_var": 7.327400339818269e-07
1692
+ },
1693
+ {
1694
+ "episode": 43520,
1695
+ "epoch": 0.1864290610006854,
1696
+ "eps": 7,
1697
+ "loss/policy_avg": 0.047856032848358154,
1698
+ "lr": 2.015625e-06,
1699
+ "objective/entropy": -50.72874450683594,
1700
+ "objective/kl": 31.98059844970703,
1701
+ "objective/non_score_reward": -1.5990298986434937,
1702
+ "objective/rlhf_reward": 11.84218978881836,
1703
+ "objective/scores": 13.4412202835083,
1704
+ "policy/approxkl_avg": 0.06686994433403015,
1705
+ "policy/clipfrac_avg": 0.20703125,
1706
+ "policy/entropy_avg": 0.208661288022995,
1707
+ "step": 84,
1708
+ "val/clipfrac_avg": 0.0,
1709
+ "val/num_eos_tokens": 2292,
1710
+ "val/ratio": 1.0000088214874268,
1711
+ "val/ratio_var": 8.189380196199636e-07
1712
+ },
1713
+ {
1714
+ "episode": 44032,
1715
+ "epoch": 0.1886223440712817,
1716
+ "eps": 7,
1717
+ "loss/policy_avg": 0.06779906898736954,
1718
+ "lr": 2.00390625e-06,
1719
+ "objective/entropy": -53.01400375366211,
1720
+ "objective/kl": 31.55933380126953,
1721
+ "objective/non_score_reward": -1.5779666900634766,
1722
+ "objective/rlhf_reward": 11.665364265441895,
1723
+ "objective/scores": 13.243330955505371,
1724
+ "policy/approxkl_avg": 0.052923060953617096,
1725
+ "policy/clipfrac_avg": 0.177734375,
1726
+ "policy/entropy_avg": 0.2091614007949829,
1727
+ "step": 85,
1728
+ "val/clipfrac_avg": 0.0,
1729
+ "val/num_eos_tokens": 2185,
1730
+ "val/ratio": 0.9999080896377563,
1731
+ "val/ratio_var": 7.456224579982518e-07
1732
+ },
1733
+ {
1734
+ "episode": 44544,
1735
+ "epoch": 0.190815627141878,
1736
+ "eps": 7,
1737
+ "loss/policy_avg": 0.05880901217460632,
1738
+ "lr": 1.9921875e-06,
1739
+ "objective/entropy": -49.737144470214844,
1740
+ "objective/kl": 31.93135643005371,
1741
+ "objective/non_score_reward": -1.596567988395691,
1742
+ "objective/rlhf_reward": 11.870726585388184,
1743
+ "objective/scores": 13.467294692993164,
1744
+ "policy/approxkl_avg": 0.055838484317064285,
1745
+ "policy/clipfrac_avg": 0.20703125,
1746
+ "policy/entropy_avg": 0.2014000117778778,
1747
+ "step": 86,
1748
+ "val/clipfrac_avg": 0.0,
1749
+ "val/num_eos_tokens": 2277,
1750
+ "val/ratio": 1.0000534057617188,
1751
+ "val/ratio_var": 1.0206571232629358e-06
1752
+ },
1753
+ {
1754
+ "episode": 45056,
1755
+ "epoch": 0.1930089102124743,
1756
+ "eps": 7,
1757
+ "loss/policy_avg": 0.04301924630999565,
1758
+ "lr": 1.98046875e-06,
1759
+ "objective/entropy": -49.397239685058594,
1760
+ "objective/kl": 32.40669250488281,
1761
+ "objective/non_score_reward": -1.6203346252441406,
1762
+ "objective/rlhf_reward": 11.55942153930664,
1763
+ "objective/scores": 13.179756164550781,
1764
+ "policy/approxkl_avg": 0.056858062744140625,
1765
+ "policy/clipfrac_avg": 0.2109375,
1766
+ "policy/entropy_avg": 0.18657711148262024,
1767
+ "step": 87,
1768
+ "val/clipfrac_avg": 0.0,
1769
+ "val/num_eos_tokens": 2092,
1770
+ "val/ratio": 0.9999588131904602,
1771
+ "val/ratio_var": 9.056823273567716e-07
1772
+ },
1773
+ {
1774
+ "episode": 45568,
1775
+ "epoch": 0.1952021932830706,
1776
+ "eps": 7,
1777
+ "loss/policy_avg": 0.04341074824333191,
1778
+ "lr": 1.96875e-06,
1779
+ "objective/entropy": -50.30104064941406,
1780
+ "objective/kl": 31.91788673400879,
1781
+ "objective/non_score_reward": -1.5958943367004395,
1782
+ "objective/rlhf_reward": 11.957464218139648,
1783
+ "objective/scores": 13.553359985351562,
1784
+ "policy/approxkl_avg": 0.054110683500766754,
1785
+ "policy/clipfrac_avg": 0.193359375,
1786
+ "policy/entropy_avg": 0.177315354347229,
1787
+ "step": 88,
1788
+ "val/clipfrac_avg": 0.0,
1789
+ "val/num_eos_tokens": 2108,
1790
+ "val/ratio": 1.0000953674316406,
1791
+ "val/ratio_var": 8.026260616134095e-07
1792
+ },
1793
+ {
1794
+ "episode": 46080,
1795
+ "epoch": 0.1973954763536669,
1796
+ "eps": 7,
1797
+ "loss/policy_avg": 0.060787640511989594,
1798
+ "lr": 1.95703125e-06,
1799
+ "objective/entropy": -45.673580169677734,
1800
+ "objective/kl": 32.983306884765625,
1801
+ "objective/non_score_reward": -1.649165391921997,
1802
+ "objective/rlhf_reward": 11.565816879272461,
1803
+ "objective/scores": 13.214982986450195,
1804
+ "policy/approxkl_avg": 0.061830729246139526,
1805
+ "policy/clipfrac_avg": 0.21875,
1806
+ "policy/entropy_avg": 0.19462166726589203,
1807
+ "step": 89,
1808
+ "val/clipfrac_avg": 0.0,
1809
+ "val/num_eos_tokens": 2096,
1810
+ "val/ratio": 1.000004768371582,
1811
+ "val/ratio_var": 7.845932259442634e-07
1812
+ },
1813
+ {
1814
+ "episode": 46592,
1815
+ "epoch": 0.19958875942426318,
1816
+ "eps": 7,
1817
+ "loss/policy_avg": 0.06810497492551804,
1818
+ "lr": 1.9453125e-06,
1819
+ "objective/entropy": -49.181297302246094,
1820
+ "objective/kl": 32.277130126953125,
1821
+ "objective/non_score_reward": -1.6138564348220825,
1822
+ "objective/rlhf_reward": 11.766349792480469,
1823
+ "objective/scores": 13.380206108093262,
1824
+ "policy/approxkl_avg": 0.05695688724517822,
1825
+ "policy/clipfrac_avg": 0.203125,
1826
+ "policy/entropy_avg": 0.17930074036121368,
1827
+ "step": 90,
1828
+ "val/clipfrac_avg": 0.0,
1829
+ "val/num_eos_tokens": 2083,
1830
+ "val/ratio": 0.9998986721038818,
1831
+ "val/ratio_var": 8.221533107644063e-07
1832
+ },
1833
+ {
1834
+ "episode": 47104,
1835
+ "epoch": 0.20178204249485948,
1836
+ "eps": 7,
1837
+ "loss/policy_avg": 0.055701229721307755,
1838
+ "lr": 1.93359375e-06,
1839
+ "objective/entropy": -45.867530822753906,
1840
+ "objective/kl": 31.220354080200195,
1841
+ "objective/non_score_reward": -1.5610177516937256,
1842
+ "objective/rlhf_reward": 11.842193603515625,
1843
+ "objective/scores": 13.40321159362793,
1844
+ "policy/approxkl_avg": 0.05533070117235184,
1845
+ "policy/clipfrac_avg": 0.173828125,
1846
+ "policy/entropy_avg": 0.18107563257217407,
1847
+ "step": 91,
1848
+ "val/clipfrac_avg": 0.0,
1849
+ "val/num_eos_tokens": 2156,
1850
+ "val/ratio": 1.0001351833343506,
1851
+ "val/ratio_var": 8.131133313327155e-07
1852
+ },
1853
+ {
1854
+ "episode": 47616,
1855
+ "epoch": 0.20397532556545578,
1856
+ "eps": 7,
1857
+ "loss/policy_avg": 0.05227220430970192,
1858
+ "lr": 1.921875e-06,
1859
+ "objective/entropy": -48.75312805175781,
1860
+ "objective/kl": 32.26356506347656,
1861
+ "objective/non_score_reward": -1.6131783723831177,
1862
+ "objective/rlhf_reward": 11.66547966003418,
1863
+ "objective/scores": 13.278657913208008,
1864
+ "policy/approxkl_avg": 0.052241113036870956,
1865
+ "policy/clipfrac_avg": 0.185546875,
1866
+ "policy/entropy_avg": 0.193963885307312,
1867
+ "step": 92,
1868
+ "val/clipfrac_avg": 0.0,
1869
+ "val/num_eos_tokens": 2168,
1870
+ "val/ratio": 0.999929666519165,
1871
+ "val/ratio_var": 7.763456437714922e-07
1872
+ },
1873
+ {
1874
+ "episode": 48128,
1875
+ "epoch": 0.2061686086360521,
1876
+ "eps": 7,
1877
+ "loss/policy_avg": 0.1222270280122757,
1878
+ "lr": 1.91015625e-06,
1879
+ "objective/entropy": -51.514556884765625,
1880
+ "objective/kl": 30.764408111572266,
1881
+ "objective/non_score_reward": -1.5382204055786133,
1882
+ "objective/rlhf_reward": 11.856256484985352,
1883
+ "objective/scores": 13.394476890563965,
1884
+ "policy/approxkl_avg": 0.05982975289225578,
1885
+ "policy/clipfrac_avg": 0.1875,
1886
+ "policy/entropy_avg": 0.1786767691373825,
1887
+ "step": 93,
1888
+ "val/clipfrac_avg": 0.0,
1889
+ "val/num_eos_tokens": 2209,
1890
+ "val/ratio": 1.000001072883606,
1891
+ "val/ratio_var": 1.080703782463388e-06
1892
+ },
1893
+ {
1894
+ "episode": 48640,
1895
+ "epoch": 0.2083618917066484,
1896
+ "eps": 7,
1897
+ "loss/policy_avg": 0.05552385002374649,
1898
+ "lr": 1.8984375e-06,
1899
+ "objective/entropy": -50.57196044921875,
1900
+ "objective/kl": 31.101146697998047,
1901
+ "objective/non_score_reward": -1.555057406425476,
1902
+ "objective/rlhf_reward": 11.832839965820312,
1903
+ "objective/scores": 13.387897491455078,
1904
+ "policy/approxkl_avg": 0.056622885167598724,
1905
+ "policy/clipfrac_avg": 0.203125,
1906
+ "policy/entropy_avg": 0.1913204789161682,
1907
+ "step": 94,
1908
+ "val/clipfrac_avg": 0.0,
1909
+ "val/num_eos_tokens": 2301,
1910
+ "val/ratio": 1.000207781791687,
1911
+ "val/ratio_var": 1.146110207628226e-06
1912
+ },
1913
+ {
1914
+ "episode": 49152,
1915
+ "epoch": 0.2105551747772447,
1916
+ "eps": 7,
1917
+ "loss/policy_avg": 0.06297577172517776,
1918
+ "lr": 1.8867187500000001e-06,
1919
+ "objective/entropy": -46.39762878417969,
1920
+ "objective/kl": 31.752248764038086,
1921
+ "objective/non_score_reward": -1.587612509727478,
1922
+ "objective/rlhf_reward": 11.969014167785645,
1923
+ "objective/scores": 13.556626319885254,
1924
+ "policy/approxkl_avg": 0.05632961913943291,
1925
+ "policy/clipfrac_avg": 0.2109375,
1926
+ "policy/entropy_avg": 0.18543684482574463,
1927
+ "step": 95,
1928
+ "val/clipfrac_avg": 0.0,
1929
+ "val/num_eos_tokens": 2353,
1930
+ "val/ratio": 1.0002882480621338,
1931
+ "val/ratio_var": 1.0207173772869282e-06
1932
+ },
1933
+ {
1934
+ "episode": 49664,
1935
+ "epoch": 0.212748457847841,
1936
+ "eps": 7,
1937
+ "loss/policy_avg": 0.1230374351143837,
1938
+ "lr": 1.875e-06,
1939
+ "objective/entropy": -49.59297180175781,
1940
+ "objective/kl": 31.085838317871094,
1941
+ "objective/non_score_reward": -1.554291844367981,
1942
+ "objective/rlhf_reward": 11.586030960083008,
1943
+ "objective/scores": 13.140321731567383,
1944
+ "policy/approxkl_avg": 0.05181390047073364,
1945
+ "policy/clipfrac_avg": 0.15625,
1946
+ "policy/entropy_avg": 0.2025001049041748,
1947
+ "step": 96,
1948
+ "val/clipfrac_avg": 0.0,
1949
+ "val/num_eos_tokens": 2391,
1950
+ "val/ratio": 0.9999078512191772,
1951
+ "val/ratio_var": 8.279580470116343e-07
1952
+ },
1953
+ {
1954
+ "episode": 50176,
1955
+ "epoch": 0.2149417409184373,
1956
+ "eps": 7,
1957
+ "loss/policy_avg": 0.09635881334543228,
1958
+ "lr": 1.86328125e-06,
1959
+ "objective/entropy": -49.964263916015625,
1960
+ "objective/kl": 31.440513610839844,
1961
+ "objective/non_score_reward": -1.5720256567001343,
1962
+ "objective/rlhf_reward": 11.82887077331543,
1963
+ "objective/scores": 13.400896072387695,
1964
+ "policy/approxkl_avg": 0.08038289844989777,
1965
+ "policy/clipfrac_avg": 0.1953125,
1966
+ "policy/entropy_avg": 0.19088959693908691,
1967
+ "step": 97,
1968
+ "val/clipfrac_avg": 0.0,
1969
+ "val/num_eos_tokens": 2472,
1970
+ "val/ratio": 0.9999905228614807,
1971
+ "val/ratio_var": 5.620386787086318e-07
1972
+ },
1973
+ {
1974
+ "episode": 50688,
1975
+ "epoch": 0.2171350239890336,
1976
+ "eps": 7,
1977
+ "loss/policy_avg": 0.06588736176490784,
1978
+ "lr": 1.8515625000000001e-06,
1979
+ "objective/entropy": -47.49836730957031,
1980
+ "objective/kl": 32.423866271972656,
1981
+ "objective/non_score_reward": -1.6211934089660645,
1982
+ "objective/rlhf_reward": 11.744735717773438,
1983
+ "objective/scores": 13.365928649902344,
1984
+ "policy/approxkl_avg": 0.05758340656757355,
1985
+ "policy/clipfrac_avg": 0.21875,
1986
+ "policy/entropy_avg": 0.2113109529018402,
1987
+ "step": 98,
1988
+ "val/clipfrac_avg": 0.0,
1989
+ "val/num_eos_tokens": 2630,
1990
+ "val/ratio": 1.0000933408737183,
1991
+ "val/ratio_var": 6.861681072223291e-07
1992
+ },
1993
+ {
1994
+ "episode": 51200,
1995
+ "epoch": 0.21932830705962988,
1996
+ "eps": 7,
1997
+ "loss/policy_avg": 0.04643254354596138,
1998
+ "lr": 1.83984375e-06,
1999
+ "objective/entropy": -48.634422302246094,
2000
+ "objective/kl": 30.64898681640625,
2001
+ "objective/non_score_reward": -1.5324492454528809,
2002
+ "objective/rlhf_reward": 12.014122009277344,
2003
+ "objective/scores": 13.54656982421875,
2004
+ "policy/approxkl_avg": 0.049847401678562164,
2005
+ "policy/clipfrac_avg": 0.197265625,
2006
+ "policy/entropy_avg": 0.20576263964176178,
2007
+ "step": 99,
2008
+ "val/clipfrac_avg": 0.0,
2009
+ "val/num_eos_tokens": 2796,
2010
+ "val/ratio": 0.9998978972434998,
2011
+ "val/ratio_var": 7.073323331496795e-07
2012
+ },
2013
+ {
2014
+ "episode": 51712,
2015
+ "epoch": 0.22152159013022618,
2016
+ "eps": 7,
2017
+ "loss/policy_avg": 0.062006328254938126,
2018
+ "lr": 1.828125e-06,
2019
+ "objective/entropy": -44.873626708984375,
2020
+ "objective/kl": 32.74094772338867,
2021
+ "objective/non_score_reward": -1.637047529220581,
2022
+ "objective/rlhf_reward": 11.676469802856445,
2023
+ "objective/scores": 13.313518524169922,
2024
+ "policy/approxkl_avg": 0.05966964364051819,
2025
+ "policy/clipfrac_avg": 0.2109375,
2026
+ "policy/entropy_avg": 0.20745235681533813,
2027
+ "step": 100,
2028
+ "val/clipfrac_avg": 0.0,
2029
+ "val/num_eos_tokens": 2748,
2030
+ "val/ratio": 1.0000770092010498,
2031
+ "val/ratio_var": 8.392998438466748e-07
2032
+ },
2033
+ {
2034
+ "episode": 52224,
2035
+ "epoch": 0.22371487320082248,
2036
+ "eps": 7,
2037
+ "loss/policy_avg": 0.07344357669353485,
2038
+ "lr": 1.81640625e-06,
2039
+ "objective/entropy": -48.71068572998047,
2040
+ "objective/kl": 31.158803939819336,
2041
+ "objective/non_score_reward": -1.5579402446746826,
2042
+ "objective/rlhf_reward": 12.125553131103516,
2043
+ "objective/scores": 13.683492660522461,
2044
+ "policy/approxkl_avg": 0.07100490480661392,
2045
+ "policy/clipfrac_avg": 0.189453125,
2046
+ "policy/entropy_avg": 0.21341751515865326,
2047
+ "step": 101,
2048
+ "val/clipfrac_avg": 0.0,
2049
+ "val/num_eos_tokens": 2885,
2050
+ "val/ratio": 1.00006103515625,
2051
+ "val/ratio_var": 7.003912401160051e-07
2052
+ },
2053
+ {
2054
+ "episode": 52736,
2055
+ "epoch": 0.22590815627141878,
2056
+ "eps": 7,
2057
+ "loss/policy_avg": 0.06474415212869644,
2058
+ "lr": 1.8046875e-06,
2059
+ "objective/entropy": -48.79777908325195,
2060
+ "objective/kl": 31.459522247314453,
2061
+ "objective/non_score_reward": -1.5729761123657227,
2062
+ "objective/rlhf_reward": 11.911396026611328,
2063
+ "objective/scores": 13.48437213897705,
2064
+ "policy/approxkl_avg": 0.05279037356376648,
2065
+ "policy/clipfrac_avg": 0.1875,
2066
+ "policy/entropy_avg": 0.21712711453437805,
2067
+ "step": 102,
2068
+ "val/clipfrac_avg": 0.0,
2069
+ "val/num_eos_tokens": 2853,
2070
+ "val/ratio": 0.999960720539093,
2071
+ "val/ratio_var": 6.893490649417799e-07
2072
+ },
2073
+ {
2074
+ "episode": 53248,
2075
+ "epoch": 0.22810143934201507,
2076
+ "eps": 7,
2077
+ "loss/policy_avg": 0.02789628878235817,
2078
+ "lr": 1.79296875e-06,
2079
+ "objective/entropy": -48.6159553527832,
2080
+ "objective/kl": 32.19043731689453,
2081
+ "objective/non_score_reward": -1.6095221042633057,
2082
+ "objective/rlhf_reward": 11.970061302185059,
2083
+ "objective/scores": 13.579584121704102,
2084
+ "policy/approxkl_avg": 0.06396406888961792,
2085
+ "policy/clipfrac_avg": 0.20703125,
2086
+ "policy/entropy_avg": 0.21447110176086426,
2087
+ "step": 103,
2088
+ "val/clipfrac_avg": 0.0,
2089
+ "val/num_eos_tokens": 2813,
2090
+ "val/ratio": 0.9999403357505798,
2091
+ "val/ratio_var": 7.448740575455304e-07
2092
+ },
2093
+ {
2094
+ "episode": 53760,
2095
+ "epoch": 0.23029472241261137,
2096
+ "eps": 7,
2097
+ "loss/policy_avg": 0.04925355315208435,
2098
+ "lr": 1.78125e-06,
2099
+ "objective/entropy": -48.03702926635742,
2100
+ "objective/kl": 31.010663986206055,
2101
+ "objective/non_score_reward": -1.550533413887024,
2102
+ "objective/rlhf_reward": 11.89893627166748,
2103
+ "objective/scores": 13.449468612670898,
2104
+ "policy/approxkl_avg": 0.04896912723779678,
2105
+ "policy/clipfrac_avg": 0.2109375,
2106
+ "policy/entropy_avg": 0.21880042552947998,
2107
+ "step": 104,
2108
+ "val/clipfrac_avg": 0.0,
2109
+ "val/num_eos_tokens": 2957,
2110
+ "val/ratio": 1.000004529953003,
2111
+ "val/ratio_var": 8.116120966406015e-07
2112
+ },
2113
+ {
2114
+ "episode": 54272,
2115
+ "epoch": 0.23248800548320767,
2116
+ "eps": 7,
2117
+ "loss/policy_avg": 0.06632091104984283,
2118
+ "lr": 1.76953125e-06,
2119
+ "objective/entropy": -47.25006866455078,
2120
+ "objective/kl": 31.43054962158203,
2121
+ "objective/non_score_reward": -1.5715274810791016,
2122
+ "objective/rlhf_reward": 11.807076454162598,
2123
+ "objective/scores": 13.3786039352417,
2124
+ "policy/approxkl_avg": 0.066850945353508,
2125
+ "policy/clipfrac_avg": 0.208984375,
2126
+ "policy/entropy_avg": 0.21371319890022278,
2127
+ "step": 105,
2128
+ "val/clipfrac_avg": 0.0,
2129
+ "val/num_eos_tokens": 2918,
2130
+ "val/ratio": 0.999915361404419,
2131
+ "val/ratio_var": 1.0229657618765486e-06
2132
+ },
2133
+ {
2134
+ "episode": 54784,
2135
+ "epoch": 0.23468128855380396,
2136
+ "eps": 7,
2137
+ "loss/policy_avg": 0.06832720339298248,
2138
+ "lr": 1.7578125e-06,
2139
+ "objective/entropy": -49.64140319824219,
2140
+ "objective/kl": 31.527286529541016,
2141
+ "objective/non_score_reward": -1.5763643980026245,
2142
+ "objective/rlhf_reward": 11.714372634887695,
2143
+ "objective/scores": 13.290736198425293,
2144
+ "policy/approxkl_avg": 0.0627245381474495,
2145
+ "policy/clipfrac_avg": 0.189453125,
2146
+ "policy/entropy_avg": 0.21471458673477173,
2147
+ "step": 106,
2148
+ "val/clipfrac_avg": 0.0,
2149
+ "val/num_eos_tokens": 3010,
2150
+ "val/ratio": 0.9998877048492432,
2151
+ "val/ratio_var": 6.754234505024215e-07
2152
+ },
2153
+ {
2154
+ "episode": 55296,
2155
+ "epoch": 0.2368745716244003,
2156
+ "eps": 7,
2157
+ "loss/policy_avg": 0.05563385412096977,
2158
+ "lr": 1.74609375e-06,
2159
+ "objective/entropy": -48.93874740600586,
2160
+ "objective/kl": 31.15831756591797,
2161
+ "objective/non_score_reward": -1.5579159259796143,
2162
+ "objective/rlhf_reward": 12.166226387023926,
2163
+ "objective/scores": 13.724142074584961,
2164
+ "policy/approxkl_avg": 0.04459884762763977,
2165
+ "policy/clipfrac_avg": 0.181640625,
2166
+ "policy/entropy_avg": 0.21589021384716034,
2167
+ "step": 107,
2168
+ "val/clipfrac_avg": 0.0,
2169
+ "val/num_eos_tokens": 3050,
2170
+ "val/ratio": 1.0001426935195923,
2171
+ "val/ratio_var": 8.754975056035619e-07
2172
+ },
2173
+ {
2174
+ "episode": 55808,
2175
+ "epoch": 0.23906785469499658,
2176
+ "eps": 7,
2177
+ "loss/policy_avg": 0.07894216477870941,
2178
+ "lr": 1.734375e-06,
2179
+ "objective/entropy": -48.76978302001953,
2180
+ "objective/kl": 31.012630462646484,
2181
+ "objective/non_score_reward": -1.5506315231323242,
2182
+ "objective/rlhf_reward": 11.895812034606934,
2183
+ "objective/scores": 13.446443557739258,
2184
+ "policy/approxkl_avg": 0.0571858212351799,
2185
+ "policy/clipfrac_avg": 0.224609375,
2186
+ "policy/entropy_avg": 0.21034139394760132,
2187
+ "step": 108,
2188
+ "val/clipfrac_avg": 0.0,
2189
+ "val/num_eos_tokens": 2946,
2190
+ "val/ratio": 0.9999710321426392,
2191
+ "val/ratio_var": 1.362170110041916e-06
2192
+ },
2193
+ {
2194
+ "episode": 56320,
2195
+ "epoch": 0.24126113776559288,
2196
+ "eps": 7,
2197
+ "loss/policy_avg": 0.05730992928147316,
2198
+ "lr": 1.72265625e-06,
2199
+ "objective/entropy": -50.21818923950195,
2200
+ "objective/kl": 31.104856491088867,
2201
+ "objective/non_score_reward": -1.555242896080017,
2202
+ "objective/rlhf_reward": 11.923521041870117,
2203
+ "objective/scores": 13.478763580322266,
2204
+ "policy/approxkl_avg": 0.06301895529031754,
2205
+ "policy/clipfrac_avg": 0.22265625,
2206
+ "policy/entropy_avg": 0.20434075593948364,
2207
+ "step": 109,
2208
+ "val/clipfrac_avg": 0.0,
2209
+ "val/num_eos_tokens": 3144,
2210
+ "val/ratio": 0.9999529123306274,
2211
+ "val/ratio_var": 8.132472544275515e-07
2212
+ },
2213
+ {
2214
+ "episode": 56832,
2215
+ "epoch": 0.24345442083618918,
2216
+ "eps": 7,
2217
+ "loss/policy_avg": 0.05458497628569603,
2218
+ "lr": 1.7109375e-06,
2219
+ "objective/entropy": -49.27418899536133,
2220
+ "objective/kl": 30.324630737304688,
2221
+ "objective/non_score_reward": -1.5162315368652344,
2222
+ "objective/rlhf_reward": 12.01490592956543,
2223
+ "objective/scores": 13.531137466430664,
2224
+ "policy/approxkl_avg": 0.052419357001781464,
2225
+ "policy/clipfrac_avg": 0.201171875,
2226
+ "policy/entropy_avg": 0.21094170212745667,
2227
+ "step": 110,
2228
+ "val/clipfrac_avg": 0.0,
2229
+ "val/num_eos_tokens": 3055,
2230
+ "val/ratio": 1.0001752376556396,
2231
+ "val/ratio_var": 1.080083507076779e-06
2232
+ },
2233
+ {
2234
+ "episode": 57344,
2235
+ "epoch": 0.24564770390678548,
2236
+ "eps": 7,
2237
+ "loss/policy_avg": 0.038599204272031784,
2238
+ "lr": 1.69921875e-06,
2239
+ "objective/entropy": -46.80632781982422,
2240
+ "objective/kl": 31.30860137939453,
2241
+ "objective/non_score_reward": -1.5654300451278687,
2242
+ "objective/rlhf_reward": 11.990469932556152,
2243
+ "objective/scores": 13.555900573730469,
2244
+ "policy/approxkl_avg": 0.05481341481208801,
2245
+ "policy/clipfrac_avg": 0.1953125,
2246
+ "policy/entropy_avg": 0.21038272976875305,
2247
+ "step": 111,
2248
+ "val/clipfrac_avg": 0.0,
2249
+ "val/num_eos_tokens": 3174,
2250
+ "val/ratio": 1.00007164478302,
2251
+ "val/ratio_var": 8.40473433072475e-07
2252
+ },
2253
+ {
2254
+ "episode": 57856,
2255
+ "epoch": 0.24784098697738177,
2256
+ "eps": 7,
2257
+ "loss/policy_avg": 0.06401612609624863,
2258
+ "lr": 1.6875e-06,
2259
+ "objective/entropy": -49.942535400390625,
2260
+ "objective/kl": 31.591415405273438,
2261
+ "objective/non_score_reward": -1.5795707702636719,
2262
+ "objective/rlhf_reward": 11.994038581848145,
2263
+ "objective/scores": 13.573609352111816,
2264
+ "policy/approxkl_avg": 0.05404981970787048,
2265
+ "policy/clipfrac_avg": 0.189453125,
2266
+ "policy/entropy_avg": 0.2190382182598114,
2267
+ "step": 112,
2268
+ "val/clipfrac_avg": 0.0,
2269
+ "val/num_eos_tokens": 3124,
2270
+ "val/ratio": 1.0002111196517944,
2271
+ "val/ratio_var": 7.723161843387061e-07
2272
+ },
2273
+ {
2274
+ "episode": 58368,
2275
+ "epoch": 0.25003427004797807,
2276
+ "eps": 7,
2277
+ "loss/policy_avg": 0.03556077182292938,
2278
+ "lr": 1.67578125e-06,
2279
+ "objective/entropy": -50.223976135253906,
2280
+ "objective/kl": 31.2110538482666,
2281
+ "objective/non_score_reward": -1.560552716255188,
2282
+ "objective/rlhf_reward": 11.979619979858398,
2283
+ "objective/scores": 13.540172576904297,
2284
+ "policy/approxkl_avg": 0.05359281599521637,
2285
+ "policy/clipfrac_avg": 0.216796875,
2286
+ "policy/entropy_avg": 0.20700502395629883,
2287
+ "step": 113,
2288
+ "val/clipfrac_avg": 0.0,
2289
+ "val/num_eos_tokens": 3152,
2290
+ "val/ratio": 1.000028371810913,
2291
+ "val/ratio_var": 8.244367677434639e-07
2292
+ },
2293
+ {
2294
+ "episode": 58880,
2295
+ "epoch": 0.25222755311857437,
2296
+ "eps": 7,
2297
+ "loss/policy_avg": 0.010434262454509735,
2298
+ "lr": 1.6640625e-06,
2299
+ "objective/entropy": -46.467777252197266,
2300
+ "objective/kl": 31.24257469177246,
2301
+ "objective/non_score_reward": -1.5621287822723389,
2302
+ "objective/rlhf_reward": 11.877924919128418,
2303
+ "objective/scores": 13.440053939819336,
2304
+ "policy/approxkl_avg": 0.06819692254066467,
2305
+ "policy/clipfrac_avg": 0.234375,
2306
+ "policy/entropy_avg": 0.2004886418581009,
2307
+ "step": 114,
2308
+ "val/clipfrac_avg": 0.0,
2309
+ "val/num_eos_tokens": 3142,
2310
+ "val/ratio": 0.9997619390487671,
2311
+ "val/ratio_var": 7.186379207269056e-07
2312
+ },
2313
+ {
2314
+ "episode": 59392,
2315
+ "epoch": 0.25442083618917066,
2316
+ "eps": 7,
2317
+ "loss/policy_avg": 0.036439333111047745,
2318
+ "lr": 1.6523437500000001e-06,
2319
+ "objective/entropy": -49.938045501708984,
2320
+ "objective/kl": 30.839561462402344,
2321
+ "objective/non_score_reward": -1.5419780015945435,
2322
+ "objective/rlhf_reward": 11.90538501739502,
2323
+ "objective/scores": 13.447362899780273,
2324
+ "policy/approxkl_avg": 0.05128248780965805,
2325
+ "policy/clipfrac_avg": 0.201171875,
2326
+ "policy/entropy_avg": 0.20895111560821533,
2327
+ "step": 115,
2328
+ "val/clipfrac_avg": 0.0,
2329
+ "val/num_eos_tokens": 3327,
2330
+ "val/ratio": 1.0000596046447754,
2331
+ "val/ratio_var": 8.727883482606558e-07
2332
+ },
2333
+ {
2334
+ "episode": 59904,
2335
+ "epoch": 0.25661411925976696,
2336
+ "eps": 7,
2337
+ "loss/policy_avg": 0.03798438236117363,
2338
+ "lr": 1.640625e-06,
2339
+ "objective/entropy": -49.26465606689453,
2340
+ "objective/kl": 29.985055923461914,
2341
+ "objective/non_score_reward": -1.4992527961730957,
2342
+ "objective/rlhf_reward": 12.096814155578613,
2343
+ "objective/scores": 13.59606647491455,
2344
+ "policy/approxkl_avg": 0.058065369725227356,
2345
+ "policy/clipfrac_avg": 0.203125,
2346
+ "policy/entropy_avg": 0.20883342623710632,
2347
+ "step": 116,
2348
+ "val/clipfrac_avg": 0.0,
2349
+ "val/num_eos_tokens": 3162,
2350
+ "val/ratio": 0.9998736381530762,
2351
+ "val/ratio_var": 7.116667575246538e-07
2352
+ },
2353
+ {
2354
+ "episode": 60416,
2355
+ "epoch": 0.25880740233036326,
2356
+ "eps": 7,
2357
+ "loss/policy_avg": 0.07099050283432007,
2358
+ "lr": 1.62890625e-06,
2359
+ "objective/entropy": -47.13051223754883,
2360
+ "objective/kl": 31.923290252685547,
2361
+ "objective/non_score_reward": -1.5961644649505615,
2362
+ "objective/rlhf_reward": 11.913225173950195,
2363
+ "objective/scores": 13.509389877319336,
2364
+ "policy/approxkl_avg": 0.04841821268200874,
2365
+ "policy/clipfrac_avg": 0.181640625,
2366
+ "policy/entropy_avg": 0.202653706073761,
2367
+ "step": 117,
2368
+ "val/clipfrac_avg": 0.0,
2369
+ "val/num_eos_tokens": 3143,
2370
+ "val/ratio": 1.0000693798065186,
2371
+ "val/ratio_var": 5.713557698072691e-07
2372
+ },
2373
+ {
2374
+ "episode": 60928,
2375
+ "epoch": 0.26100068540095955,
2376
+ "eps": 7,
2377
+ "loss/policy_avg": 0.029071252793073654,
2378
+ "lr": 1.6171875000000001e-06,
2379
+ "objective/entropy": -52.13547134399414,
2380
+ "objective/kl": 29.580509185791016,
2381
+ "objective/non_score_reward": -1.4790253639221191,
2382
+ "objective/rlhf_reward": 12.157207489013672,
2383
+ "objective/scores": 13.636232376098633,
2384
+ "policy/approxkl_avg": 0.04795447736978531,
2385
+ "policy/clipfrac_avg": 0.16796875,
2386
+ "policy/entropy_avg": 0.1991284042596817,
2387
+ "step": 118,
2388
+ "val/clipfrac_avg": 0.0,
2389
+ "val/num_eos_tokens": 3306,
2390
+ "val/ratio": 0.9998592138290405,
2391
+ "val/ratio_var": 5.243380769570649e-07
2392
+ },
2393
+ {
2394
+ "episode": 61440,
2395
+ "epoch": 0.26319396847155585,
2396
+ "eps": 7,
2397
+ "loss/policy_avg": 0.029355045408010483,
2398
+ "lr": 1.60546875e-06,
2399
+ "objective/entropy": -51.54493713378906,
2400
+ "objective/kl": 31.228269577026367,
2401
+ "objective/non_score_reward": -1.5614135265350342,
2402
+ "objective/rlhf_reward": 11.880817413330078,
2403
+ "objective/scores": 13.442231178283691,
2404
+ "policy/approxkl_avg": 0.05728745460510254,
2405
+ "policy/clipfrac_avg": 0.216796875,
2406
+ "policy/entropy_avg": 0.21894299983978271,
2407
+ "step": 119,
2408
+ "val/clipfrac_avg": 0.0,
2409
+ "val/num_eos_tokens": 3491,
2410
+ "val/ratio": 1.0003288984298706,
2411
+ "val/ratio_var": 4.954263204126619e-06
2412
+ },
2413
+ {
2414
+ "episode": 61952,
2415
+ "epoch": 0.26538725154215215,
2416
+ "eps": 7,
2417
+ "loss/policy_avg": 0.017333194613456726,
2418
+ "lr": 1.59375e-06,
2419
+ "objective/entropy": -51.9672966003418,
2420
+ "objective/kl": 30.101791381835938,
2421
+ "objective/non_score_reward": -1.5050896406173706,
2422
+ "objective/rlhf_reward": 12.025577545166016,
2423
+ "objective/scores": 13.530667304992676,
2424
+ "policy/approxkl_avg": 0.0607779435813427,
2425
+ "policy/clipfrac_avg": 0.181640625,
2426
+ "policy/entropy_avg": 0.2030687928199768,
2427
+ "step": 120,
2428
+ "val/clipfrac_avg": 0.0,
2429
+ "val/num_eos_tokens": 3547,
2430
+ "val/ratio": 0.9999595880508423,
2431
+ "val/ratio_var": 6.275835175983957e-07
2432
+ },
2433
+ {
2434
+ "episode": 62464,
2435
+ "epoch": 0.26758053461274844,
2436
+ "eps": 7,
2437
+ "loss/policy_avg": 0.04087621718645096,
2438
+ "lr": 1.5820312500000001e-06,
2439
+ "objective/entropy": -51.47197723388672,
2440
+ "objective/kl": 30.397109985351562,
2441
+ "objective/non_score_reward": -1.5198554992675781,
2442
+ "objective/rlhf_reward": 12.032506942749023,
2443
+ "objective/scores": 13.552363395690918,
2444
+ "policy/approxkl_avg": 0.041269607841968536,
2445
+ "policy/clipfrac_avg": 0.1875,
2446
+ "policy/entropy_avg": 0.20843926072120667,
2447
+ "step": 121,
2448
+ "val/clipfrac_avg": 0.0,
2449
+ "val/num_eos_tokens": 3534,
2450
+ "val/ratio": 1.0000314712524414,
2451
+ "val/ratio_var": 6.925745310581988e-07
2452
+ },
2453
+ {
2454
+ "episode": 62976,
2455
+ "epoch": 0.26977381768334474,
2456
+ "eps": 7,
2457
+ "loss/policy_avg": 0.07551665604114532,
2458
+ "lr": 1.5703125e-06,
2459
+ "objective/entropy": -51.15126037597656,
2460
+ "objective/kl": 30.321029663085938,
2461
+ "objective/non_score_reward": -1.5160515308380127,
2462
+ "objective/rlhf_reward": 11.992143630981445,
2463
+ "objective/scores": 13.508195877075195,
2464
+ "policy/approxkl_avg": 0.052853114902973175,
2465
+ "policy/clipfrac_avg": 0.171875,
2466
+ "policy/entropy_avg": 0.2155429720878601,
2467
+ "step": 122,
2468
+ "val/clipfrac_avg": 0.0,
2469
+ "val/num_eos_tokens": 3662,
2470
+ "val/ratio": 1.0000662803649902,
2471
+ "val/ratio_var": 1.1502866072987672e-06
2472
+ },
2473
+ {
2474
+ "episode": 63488,
2475
+ "epoch": 0.27196710075394104,
2476
+ "eps": 7,
2477
+ "loss/policy_avg": 0.03800806403160095,
2478
+ "lr": 1.55859375e-06,
2479
+ "objective/entropy": -49.645782470703125,
2480
+ "objective/kl": 30.97983169555664,
2481
+ "objective/non_score_reward": -1.5489915609359741,
2482
+ "objective/rlhf_reward": 12.060916900634766,
2483
+ "objective/scores": 13.609909057617188,
2484
+ "policy/approxkl_avg": 0.04618716239929199,
2485
+ "policy/clipfrac_avg": 0.201171875,
2486
+ "policy/entropy_avg": 0.2078637331724167,
2487
+ "step": 123,
2488
+ "val/clipfrac_avg": 0.0,
2489
+ "val/num_eos_tokens": 3616,
2490
+ "val/ratio": 0.9998440742492676,
2491
+ "val/ratio_var": 7.410563398480008e-07
2492
+ },
2493
+ {
2494
+ "episode": 64000,
2495
+ "epoch": 0.27416038382453733,
2496
+ "eps": 7,
2497
+ "loss/policy_avg": 0.032623402774333954,
2498
+ "lr": 1.5468750000000001e-06,
2499
+ "objective/entropy": -51.170257568359375,
2500
+ "objective/kl": 31.303388595581055,
2501
+ "objective/non_score_reward": -1.5651694536209106,
2502
+ "objective/rlhf_reward": 11.771775245666504,
2503
+ "objective/scores": 13.336943626403809,
2504
+ "policy/approxkl_avg": 0.05362207815051079,
2505
+ "policy/clipfrac_avg": 0.208984375,
2506
+ "policy/entropy_avg": 0.21997956931591034,
2507
+ "step": 124,
2508
+ "val/clipfrac_avg": 0.0,
2509
+ "val/num_eos_tokens": 3862,
2510
+ "val/ratio": 1.000096321105957,
2511
+ "val/ratio_var": 7.600924618600402e-07
2512
+ },
2513
+ {
2514
+ "episode": 64512,
2515
+ "epoch": 0.27635366689513363,
2516
+ "eps": 7,
2517
+ "loss/policy_avg": 0.031085459515452385,
2518
+ "lr": 1.53515625e-06,
2519
+ "objective/entropy": -49.02000427246094,
2520
+ "objective/kl": 31.002422332763672,
2521
+ "objective/non_score_reward": -1.5501210689544678,
2522
+ "objective/rlhf_reward": 12.034141540527344,
2523
+ "objective/scores": 13.58426284790039,
2524
+ "policy/approxkl_avg": 0.05696885287761688,
2525
+ "policy/clipfrac_avg": 0.18359375,
2526
+ "policy/entropy_avg": 0.20467829704284668,
2527
+ "step": 125,
2528
+ "val/clipfrac_avg": 0.0,
2529
+ "val/num_eos_tokens": 3761,
2530
+ "val/ratio": 1.0001591444015503,
2531
+ "val/ratio_var": 3.589534117054427e-06
2532
+ },
2533
+ {
2534
+ "episode": 65024,
2535
+ "epoch": 0.27854694996572993,
2536
+ "eps": 7,
2537
+ "loss/policy_avg": 0.07421186566352844,
2538
+ "lr": 1.5234375e-06,
2539
+ "objective/entropy": -47.81622314453125,
2540
+ "objective/kl": 31.5286922454834,
2541
+ "objective/non_score_reward": -1.5764347314834595,
2542
+ "objective/rlhf_reward": 11.903038024902344,
2543
+ "objective/scores": 13.479473114013672,
2544
+ "policy/approxkl_avg": 0.05290243774652481,
2545
+ "policy/clipfrac_avg": 0.205078125,
2546
+ "policy/entropy_avg": 0.2120390087366104,
2547
+ "step": 126,
2548
+ "val/clipfrac_avg": 0.0,
2549
+ "val/num_eos_tokens": 3738,
2550
+ "val/ratio": 0.9998422265052795,
2551
+ "val/ratio_var": 8.801233661870356e-07
2552
+ },
2553
+ {
2554
+ "episode": 65536,
2555
+ "epoch": 0.2807402330363262,
2556
+ "eps": 7,
2557
+ "loss/policy_avg": 0.03595343977212906,
2558
+ "lr": 1.5117187500000001e-06,
2559
+ "objective/entropy": -48.716453552246094,
2560
+ "objective/kl": 30.322792053222656,
2561
+ "objective/non_score_reward": -1.5161396265029907,
2562
+ "objective/rlhf_reward": 12.159992218017578,
2563
+ "objective/scores": 13.676132202148438,
2564
+ "policy/approxkl_avg": 0.04783044382929802,
2565
+ "policy/clipfrac_avg": 0.16796875,
2566
+ "policy/entropy_avg": 0.19197297096252441,
2567
+ "step": 127,
2568
+ "val/clipfrac_avg": 0.0,
2569
+ "val/num_eos_tokens": 3735,
2570
+ "val/ratio": 0.9998753666877747,
2571
+ "val/ratio_var": 7.83547932314832e-07
2572
+ }
2573
+ ],
2574
+ "logging_steps": 100,
2575
+ "max_steps": 128,
2576
+ "num_input_tokens_seen": 0,
2577
+ "num_train_epochs": 1.122960932145305,
2578
+ "save_steps": 26,
2579
+ "stateful_callbacks": {
2580
+ "TrainerControl": {
2581
+ "args": {
2582
+ "should_epoch_stop": false,
2583
+ "should_evaluate": false,
2584
+ "should_log": false,
2585
+ "should_save": true,
2586
+ "should_training_stop": true
2587
+ },
2588
+ "attributes": {}
2589
+ }
2590
+ },
2591
+ "total_flos": 0,
2592
+ "train_batch_size": null,
2593
+ "trial_name": null,
2594
+ "trial_params": null
2595
+ }
checkpoint-128/training_args.bin ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:d4ae92ff556ef2b4cbc99a3ca0fe3f1b097e1f615db1191f4fcc5c2dcf24c00c
3
+ size 6520