Muqeeth commited on
Commit
29330f7
·
verified ·
1 Parent(s): cfaccce

Training in progress, step 52, checkpoint

Browse files
checkpoint-52/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-52/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-52/model.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:fa97109fb7464ff6b2c22a126bd0d3b1ec3512b7e3fb0fe6d99be668e6ae1252
3
+ size 1621370224
checkpoint-52/optimizer.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:338afd25e7181ecc3e40034762305ee1f9cc9a007372cb7cd91e0a4110118913
3
+ size 3242916026
checkpoint-52/rng_state.pth ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:bdbac6ebcb8a13c75ff51d45035dae363302c36eb1d2c5eb32f8dfbe90c4915e
3
+ size 14244
checkpoint-52/scheduler.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:15adc088a35512de96eaf76adc88eb47505d933e9ec793220480f6c500342492
3
+ size 1064
checkpoint-52/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-52/tokenizer.json ADDED
The diff for this file is too large to render. See raw diff
 
checkpoint-52/tokenizer_config.json ADDED
@@ -0,0 +1,220 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "add_bos_token": false,
3
+ "add_eos_token": false,
4
+ "add_prefix_space": false,
5
+ "added_tokens_decoder": {
6
+ "0": {
7
+ "content": "<|endoftext|>",
8
+ "lstrip": false,
9
+ "normalized": false,
10
+ "rstrip": false,
11
+ "single_word": false,
12
+ "special": true
13
+ },
14
+ "1": {
15
+ "content": "<|padding|>",
16
+ "lstrip": false,
17
+ "normalized": false,
18
+ "rstrip": false,
19
+ "single_word": false,
20
+ "special": true
21
+ },
22
+ "50254": {
23
+ "content": " ",
24
+ "lstrip": false,
25
+ "normalized": true,
26
+ "rstrip": false,
27
+ "single_word": false,
28
+ "special": false
29
+ },
30
+ "50255": {
31
+ "content": " ",
32
+ "lstrip": false,
33
+ "normalized": true,
34
+ "rstrip": false,
35
+ "single_word": false,
36
+ "special": false
37
+ },
38
+ "50256": {
39
+ "content": " ",
40
+ "lstrip": false,
41
+ "normalized": true,
42
+ "rstrip": false,
43
+ "single_word": false,
44
+ "special": false
45
+ },
46
+ "50257": {
47
+ "content": " ",
48
+ "lstrip": false,
49
+ "normalized": true,
50
+ "rstrip": false,
51
+ "single_word": false,
52
+ "special": false
53
+ },
54
+ "50258": {
55
+ "content": " ",
56
+ "lstrip": false,
57
+ "normalized": true,
58
+ "rstrip": false,
59
+ "single_word": false,
60
+ "special": false
61
+ },
62
+ "50259": {
63
+ "content": " ",
64
+ "lstrip": false,
65
+ "normalized": true,
66
+ "rstrip": false,
67
+ "single_word": false,
68
+ "special": false
69
+ },
70
+ "50260": {
71
+ "content": " ",
72
+ "lstrip": false,
73
+ "normalized": true,
74
+ "rstrip": false,
75
+ "single_word": false,
76
+ "special": false
77
+ },
78
+ "50261": {
79
+ "content": " ",
80
+ "lstrip": false,
81
+ "normalized": true,
82
+ "rstrip": false,
83
+ "single_word": false,
84
+ "special": false
85
+ },
86
+ "50262": {
87
+ "content": " ",
88
+ "lstrip": false,
89
+ "normalized": true,
90
+ "rstrip": false,
91
+ "single_word": false,
92
+ "special": false
93
+ },
94
+ "50263": {
95
+ "content": " ",
96
+ "lstrip": false,
97
+ "normalized": true,
98
+ "rstrip": false,
99
+ "single_word": false,
100
+ "special": false
101
+ },
102
+ "50264": {
103
+ "content": " ",
104
+ "lstrip": false,
105
+ "normalized": true,
106
+ "rstrip": false,
107
+ "single_word": false,
108
+ "special": false
109
+ },
110
+ "50265": {
111
+ "content": " ",
112
+ "lstrip": false,
113
+ "normalized": true,
114
+ "rstrip": false,
115
+ "single_word": false,
116
+ "special": false
117
+ },
118
+ "50266": {
119
+ "content": " ",
120
+ "lstrip": false,
121
+ "normalized": true,
122
+ "rstrip": false,
123
+ "single_word": false,
124
+ "special": false
125
+ },
126
+ "50267": {
127
+ "content": " ",
128
+ "lstrip": false,
129
+ "normalized": true,
130
+ "rstrip": false,
131
+ "single_word": false,
132
+ "special": false
133
+ },
134
+ "50268": {
135
+ "content": " ",
136
+ "lstrip": false,
137
+ "normalized": true,
138
+ "rstrip": false,
139
+ "single_word": false,
140
+ "special": false
141
+ },
142
+ "50269": {
143
+ "content": " ",
144
+ "lstrip": false,
145
+ "normalized": true,
146
+ "rstrip": false,
147
+ "single_word": false,
148
+ "special": false
149
+ },
150
+ "50270": {
151
+ "content": " ",
152
+ "lstrip": false,
153
+ "normalized": true,
154
+ "rstrip": false,
155
+ "single_word": false,
156
+ "special": false
157
+ },
158
+ "50271": {
159
+ "content": " ",
160
+ "lstrip": false,
161
+ "normalized": true,
162
+ "rstrip": false,
163
+ "single_word": false,
164
+ "special": false
165
+ },
166
+ "50272": {
167
+ "content": " ",
168
+ "lstrip": false,
169
+ "normalized": true,
170
+ "rstrip": false,
171
+ "single_word": false,
172
+ "special": false
173
+ },
174
+ "50273": {
175
+ "content": " ",
176
+ "lstrip": false,
177
+ "normalized": true,
178
+ "rstrip": false,
179
+ "single_word": false,
180
+ "special": false
181
+ },
182
+ "50274": {
183
+ "content": " ",
184
+ "lstrip": false,
185
+ "normalized": true,
186
+ "rstrip": false,
187
+ "single_word": false,
188
+ "special": false
189
+ },
190
+ "50275": {
191
+ "content": " ",
192
+ "lstrip": false,
193
+ "normalized": true,
194
+ "rstrip": false,
195
+ "single_word": false,
196
+ "special": false
197
+ },
198
+ "50276": {
199
+ "content": " ",
200
+ "lstrip": false,
201
+ "normalized": true,
202
+ "rstrip": false,
203
+ "single_word": false,
204
+ "special": false
205
+ }
206
+ },
207
+ "bos_token": "<|endoftext|>",
208
+ "clean_up_tokenization_spaces": true,
209
+ "eos_token": "<|endoftext|>",
210
+ "extra_special_tokens": {},
211
+ "max_length": 580,
212
+ "model_max_length": 1000000000000000019884624838656,
213
+ "pad_token": "<|padding|>",
214
+ "padding_side": "left",
215
+ "stride": 0,
216
+ "tokenizer_class": "GPTNeoXTokenizer",
217
+ "truncation_side": "right",
218
+ "truncation_strategy": "longest_first",
219
+ "unk_token": "<|endoftext|>"
220
+ }
checkpoint-52/trainer_state.json ADDED
@@ -0,0 +1,1075 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "best_global_step": null,
3
+ "best_metric": null,
4
+ "best_model_checkpoint": null,
5
+ "episode": 26624,
6
+ "epoch": 0.11405071967100754,
7
+ "eval_steps": 500,
8
+ "global_step": 52,
9
+ "is_hyper_param_search": false,
10
+ "is_local_process_zero": true,
11
+ "is_world_process_zero": true,
12
+ "log_history": [
13
+ {
14
+ "episode": 512,
15
+ "epoch": 0.0021932830705962986,
16
+ "eps": 6,
17
+ "loss/policy_avg": 0.15234875679016113,
18
+ "lr": 3e-06,
19
+ "objective/entropy": -45.407432556152344,
20
+ "objective/kl": 0.3935524821281433,
21
+ "objective/non_score_reward": -0.001967762364074588,
22
+ "objective/rlhf_reward": 4.677086353302002,
23
+ "objective/scores": 4.679054260253906,
24
+ "policy/approxkl_avg": 0.11690396815538406,
25
+ "policy/clipfrac_avg": 0.34375,
26
+ "policy/entropy_avg": 0.8878304958343506,
27
+ "step": 0,
28
+ "val/clipfrac_avg": 0.0,
29
+ "val/num_eos_tokens": 2729,
30
+ "val/ratio": 1.000101089477539,
31
+ "val/ratio_var": 2.341874278499745e-06
32
+ },
33
+ {
34
+ "episode": 1024,
35
+ "epoch": 0.004386566141192597,
36
+ "eps": 6,
37
+ "loss/policy_avg": 0.10656304657459259,
38
+ "lr": 2.9882812500000002e-06,
39
+ "objective/entropy": -49.59076690673828,
40
+ "objective/kl": 0.6760815978050232,
41
+ "objective/non_score_reward": -0.0033804080449044704,
42
+ "objective/rlhf_reward": 4.551654815673828,
43
+ "objective/scores": 4.55503511428833,
44
+ "policy/approxkl_avg": 0.12034796178340912,
45
+ "policy/clipfrac_avg": 0.32421875,
46
+ "policy/entropy_avg": 0.8024331331253052,
47
+ "step": 1,
48
+ "val/clipfrac_avg": 0.0,
49
+ "val/num_eos_tokens": 3517,
50
+ "val/ratio": 0.9998391270637512,
51
+ "val/ratio_var": 2.0723566649394343e-06
52
+ },
53
+ {
54
+ "episode": 1536,
55
+ "epoch": 0.006579849211788897,
56
+ "eps": 6,
57
+ "loss/policy_avg": 0.17722828686237335,
58
+ "lr": 2.9765625e-06,
59
+ "objective/entropy": -51.09684371948242,
60
+ "objective/kl": 1.214200735092163,
61
+ "objective/non_score_reward": -0.006071004085242748,
62
+ "objective/rlhf_reward": 4.974597930908203,
63
+ "objective/scores": 4.980669021606445,
64
+ "policy/approxkl_avg": 0.12037836015224457,
65
+ "policy/clipfrac_avg": 0.318359375,
66
+ "policy/entropy_avg": 0.8043380379676819,
67
+ "step": 2,
68
+ "val/clipfrac_avg": 0.0,
69
+ "val/num_eos_tokens": 3252,
70
+ "val/ratio": 1.0001596212387085,
71
+ "val/ratio_var": 2.1233340703474823e-06
72
+ },
73
+ {
74
+ "episode": 2048,
75
+ "epoch": 0.008773132282385195,
76
+ "eps": 6,
77
+ "loss/policy_avg": 0.11922580748796463,
78
+ "lr": 2.96484375e-06,
79
+ "objective/entropy": -44.18528366088867,
80
+ "objective/kl": 1.4338339567184448,
81
+ "objective/non_score_reward": -0.007169169839471579,
82
+ "objective/rlhf_reward": 4.915553569793701,
83
+ "objective/scores": 4.922722339630127,
84
+ "policy/approxkl_avg": 0.16550302505493164,
85
+ "policy/clipfrac_avg": 0.375,
86
+ "policy/entropy_avg": 0.7921671271324158,
87
+ "step": 3,
88
+ "val/clipfrac_avg": 0.0,
89
+ "val/num_eos_tokens": 2435,
90
+ "val/ratio": 0.9996773600578308,
91
+ "val/ratio_var": 2.4303119516844163e-06
92
+ },
93
+ {
94
+ "episode": 2560,
95
+ "epoch": 0.010966415352981495,
96
+ "eps": 6,
97
+ "loss/policy_avg": 0.15014484524726868,
98
+ "lr": 2.953125e-06,
99
+ "objective/entropy": -42.067405700683594,
100
+ "objective/kl": 2.0778450965881348,
101
+ "objective/non_score_reward": -0.01038922369480133,
102
+ "objective/rlhf_reward": 5.2342209815979,
103
+ "objective/scores": 5.244609832763672,
104
+ "policy/approxkl_avg": 0.1276472508907318,
105
+ "policy/clipfrac_avg": 0.31640625,
106
+ "policy/entropy_avg": 0.7496817111968994,
107
+ "step": 4,
108
+ "val/clipfrac_avg": 0.0,
109
+ "val/num_eos_tokens": 1877,
110
+ "val/ratio": 1.0003182888031006,
111
+ "val/ratio_var": 2.2540482405020157e-06
112
+ },
113
+ {
114
+ "episode": 3072,
115
+ "epoch": 0.013159698423577794,
116
+ "eps": 6,
117
+ "loss/policy_avg": 0.1916506588459015,
118
+ "lr": 2.94140625e-06,
119
+ "objective/entropy": -39.82306671142578,
120
+ "objective/kl": 3.079103708267212,
121
+ "objective/non_score_reward": -0.015395518392324448,
122
+ "objective/rlhf_reward": 5.611381530761719,
123
+ "objective/scores": 5.626776695251465,
124
+ "policy/approxkl_avg": 0.13991527259349823,
125
+ "policy/clipfrac_avg": 0.33203125,
126
+ "policy/entropy_avg": 0.7127301692962646,
127
+ "step": 5,
128
+ "val/clipfrac_avg": 0.0,
129
+ "val/num_eos_tokens": 1565,
130
+ "val/ratio": 0.9999192357063293,
131
+ "val/ratio_var": 1.5017312762211077e-06
132
+ },
133
+ {
134
+ "episode": 3584,
135
+ "epoch": 0.015352981494174092,
136
+ "eps": 6,
137
+ "loss/policy_avg": 0.1968970149755478,
138
+ "lr": 2.9296875e-06,
139
+ "objective/entropy": -32.000404357910156,
140
+ "objective/kl": 4.284768581390381,
141
+ "objective/non_score_reward": -0.021423842757940292,
142
+ "objective/rlhf_reward": 5.7136030197143555,
143
+ "objective/scores": 5.7350263595581055,
144
+ "policy/approxkl_avg": 0.1647772341966629,
145
+ "policy/clipfrac_avg": 0.3515625,
146
+ "policy/entropy_avg": 0.6934345364570618,
147
+ "step": 6,
148
+ "val/clipfrac_avg": 0.0,
149
+ "val/num_eos_tokens": 1364,
150
+ "val/ratio": 0.9998640418052673,
151
+ "val/ratio_var": 2.735638872763957e-06
152
+ },
153
+ {
154
+ "episode": 4096,
155
+ "epoch": 0.01754626456477039,
156
+ "eps": 6,
157
+ "loss/policy_avg": 0.20194123685359955,
158
+ "lr": 2.91796875e-06,
159
+ "objective/entropy": -35.15373992919922,
160
+ "objective/kl": 4.795171737670898,
161
+ "objective/non_score_reward": -0.023975860327482224,
162
+ "objective/rlhf_reward": 6.039769649505615,
163
+ "objective/scores": 6.063745498657227,
164
+ "policy/approxkl_avg": 0.19040407240390778,
165
+ "policy/clipfrac_avg": 0.365234375,
166
+ "policy/entropy_avg": 0.6659716367721558,
167
+ "step": 7,
168
+ "val/clipfrac_avg": 0.0,
169
+ "val/num_eos_tokens": 1595,
170
+ "val/ratio": 0.9995955228805542,
171
+ "val/ratio_var": 2.328864411538234e-06
172
+ },
173
+ {
174
+ "episode": 4608,
175
+ "epoch": 0.01973954763536669,
176
+ "eps": 6,
177
+ "loss/policy_avg": 0.2287895679473877,
178
+ "lr": 2.90625e-06,
179
+ "objective/entropy": -33.49115753173828,
180
+ "objective/kl": 6.028232097625732,
181
+ "objective/non_score_reward": -0.030141159892082214,
182
+ "objective/rlhf_reward": 6.286189556121826,
183
+ "objective/scores": 6.316330909729004,
184
+ "policy/approxkl_avg": 0.16508515179157257,
185
+ "policy/clipfrac_avg": 0.35546875,
186
+ "policy/entropy_avg": 0.6881623268127441,
187
+ "step": 8,
188
+ "val/clipfrac_avg": 0.0,
189
+ "val/num_eos_tokens": 1875,
190
+ "val/ratio": 0.9999189376831055,
191
+ "val/ratio_var": 2.560210077717784e-06
192
+ },
193
+ {
194
+ "episode": 5120,
195
+ "epoch": 0.02193283070596299,
196
+ "eps": 6,
197
+ "loss/policy_avg": 0.23305176198482513,
198
+ "lr": 2.89453125e-06,
199
+ "objective/entropy": -31.895954132080078,
200
+ "objective/kl": 7.27902889251709,
201
+ "objective/non_score_reward": -0.03639514371752739,
202
+ "objective/rlhf_reward": 6.584256172180176,
203
+ "objective/scores": 6.6206512451171875,
204
+ "policy/approxkl_avg": 0.15825235843658447,
205
+ "policy/clipfrac_avg": 0.333984375,
206
+ "policy/entropy_avg": 0.6986380219459534,
207
+ "step": 9,
208
+ "val/clipfrac_avg": 0.0,
209
+ "val/num_eos_tokens": 2233,
210
+ "val/ratio": 1.000351905822754,
211
+ "val/ratio_var": 2.54610608863004e-06
212
+ },
213
+ {
214
+ "episode": 5632,
215
+ "epoch": 0.024126113776559287,
216
+ "eps": 6,
217
+ "loss/policy_avg": 0.2743307948112488,
218
+ "lr": 2.8828125e-06,
219
+ "objective/entropy": -30.13780975341797,
220
+ "objective/kl": 8.618773460388184,
221
+ "objective/non_score_reward": -0.04309386759996414,
222
+ "objective/rlhf_reward": 6.671762466430664,
223
+ "objective/scores": 6.714856147766113,
224
+ "policy/approxkl_avg": 0.2310631275177002,
225
+ "policy/clipfrac_avg": 0.376953125,
226
+ "policy/entropy_avg": 0.6997334957122803,
227
+ "step": 10,
228
+ "val/clipfrac_avg": 0.0,
229
+ "val/num_eos_tokens": 2288,
230
+ "val/ratio": 0.9999397397041321,
231
+ "val/ratio_var": 2.7119385777041316e-06
232
+ },
233
+ {
234
+ "episode": 6144,
235
+ "epoch": 0.026319396847155587,
236
+ "eps": 6,
237
+ "loss/policy_avg": 0.47530221939086914,
238
+ "lr": 2.87109375e-06,
239
+ "objective/entropy": -25.868932723999023,
240
+ "objective/kl": 10.715995788574219,
241
+ "objective/non_score_reward": -0.05357997864484787,
242
+ "objective/rlhf_reward": 6.660526275634766,
243
+ "objective/scores": 6.714106559753418,
244
+ "policy/approxkl_avg": 0.23832854628562927,
245
+ "policy/clipfrac_avg": 0.33984375,
246
+ "policy/entropy_avg": 0.6960214376449585,
247
+ "step": 11,
248
+ "val/clipfrac_avg": 0.0,
249
+ "val/num_eos_tokens": 2465,
250
+ "val/ratio": 1.0001269578933716,
251
+ "val/ratio_var": 4.06193794333376e-06
252
+ },
253
+ {
254
+ "episode": 6656,
255
+ "epoch": 0.028512679917751884,
256
+ "eps": 6,
257
+ "loss/policy_avg": 0.2478591650724411,
258
+ "lr": 2.859375e-06,
259
+ "objective/entropy": -28.481605529785156,
260
+ "objective/kl": 11.356575012207031,
261
+ "objective/non_score_reward": -0.05678287148475647,
262
+ "objective/rlhf_reward": 7.32498025894165,
263
+ "objective/scores": 7.381762981414795,
264
+ "policy/approxkl_avg": 0.23808607459068298,
265
+ "policy/clipfrac_avg": 0.345703125,
266
+ "policy/entropy_avg": 0.6703023910522461,
267
+ "step": 12,
268
+ "val/clipfrac_avg": 0.0,
269
+ "val/num_eos_tokens": 2564,
270
+ "val/ratio": 0.999792754650116,
271
+ "val/ratio_var": 2.613600599943311e-06
272
+ },
273
+ {
274
+ "episode": 7168,
275
+ "epoch": 0.030705962988348184,
276
+ "eps": 6,
277
+ "loss/policy_avg": 0.18863295018672943,
278
+ "lr": 2.84765625e-06,
279
+ "objective/entropy": -31.87090301513672,
280
+ "objective/kl": 12.704984664916992,
281
+ "objective/non_score_reward": -0.06352491676807404,
282
+ "objective/rlhf_reward": 7.6809868812561035,
283
+ "objective/scores": 7.744511604309082,
284
+ "policy/approxkl_avg": 0.20181894302368164,
285
+ "policy/clipfrac_avg": 0.3828125,
286
+ "policy/entropy_avg": 0.6681352853775024,
287
+ "step": 13,
288
+ "val/clipfrac_avg": 0.0,
289
+ "val/num_eos_tokens": 2607,
290
+ "val/ratio": 1.0001722574234009,
291
+ "val/ratio_var": 3.1368097097583814e-06
292
+ },
293
+ {
294
+ "episode": 7680,
295
+ "epoch": 0.03289924605894448,
296
+ "eps": 6,
297
+ "loss/policy_avg": 0.23149724304676056,
298
+ "lr": 2.8359375e-06,
299
+ "objective/entropy": -31.151302337646484,
300
+ "objective/kl": 14.230264663696289,
301
+ "objective/non_score_reward": -0.07115132361650467,
302
+ "objective/rlhf_reward": 7.844120025634766,
303
+ "objective/scores": 7.915271282196045,
304
+ "policy/approxkl_avg": 0.23936033248901367,
305
+ "policy/clipfrac_avg": 0.40625,
306
+ "policy/entropy_avg": 0.6505059003829956,
307
+ "step": 14,
308
+ "val/clipfrac_avg": 0.0,
309
+ "val/num_eos_tokens": 2523,
310
+ "val/ratio": 0.999779224395752,
311
+ "val/ratio_var": 3.855011527775787e-06
312
+ },
313
+ {
314
+ "episode": 8192,
315
+ "epoch": 0.03509252912954078,
316
+ "eps": 6,
317
+ "loss/policy_avg": 0.25306057929992676,
318
+ "lr": 2.82421875e-06,
319
+ "objective/entropy": -33.95362091064453,
320
+ "objective/kl": 15.211322784423828,
321
+ "objective/non_score_reward": -0.07605661451816559,
322
+ "objective/rlhf_reward": 8.07334041595459,
323
+ "objective/scores": 8.149396896362305,
324
+ "policy/approxkl_avg": 0.2219444364309311,
325
+ "policy/clipfrac_avg": 0.376953125,
326
+ "policy/entropy_avg": 0.6534073352813721,
327
+ "step": 15,
328
+ "val/clipfrac_avg": 0.0,
329
+ "val/num_eos_tokens": 2392,
330
+ "val/ratio": 1.000091791152954,
331
+ "val/ratio_var": 3.7099030123499688e-06
332
+ },
333
+ {
334
+ "episode": 8704,
335
+ "epoch": 0.03728581220013708,
336
+ "eps": 6,
337
+ "loss/policy_avg": 0.3671952486038208,
338
+ "lr": 2.8125e-06,
339
+ "objective/entropy": -36.56987762451172,
340
+ "objective/kl": 15.896276473999023,
341
+ "objective/non_score_reward": -0.07948137074708939,
342
+ "objective/rlhf_reward": 8.115487098693848,
343
+ "objective/scores": 8.194968223571777,
344
+ "policy/approxkl_avg": 0.2001374065876007,
345
+ "policy/clipfrac_avg": 0.376953125,
346
+ "policy/entropy_avg": 0.5611248016357422,
347
+ "step": 16,
348
+ "val/clipfrac_avg": 0.0,
349
+ "val/num_eos_tokens": 2099,
350
+ "val/ratio": 1.0001428127288818,
351
+ "val/ratio_var": 2.9763129987259163e-06
352
+ },
353
+ {
354
+ "episode": 9216,
355
+ "epoch": 0.03947909527073338,
356
+ "eps": 6,
357
+ "loss/policy_avg": 0.2377510964870453,
358
+ "lr": 2.80078125e-06,
359
+ "objective/entropy": -38.7183837890625,
360
+ "objective/kl": 16.948450088500977,
361
+ "objective/non_score_reward": -0.08474224805831909,
362
+ "objective/rlhf_reward": 8.287287712097168,
363
+ "objective/scores": 8.372029304504395,
364
+ "policy/approxkl_avg": 0.19774162769317627,
365
+ "policy/clipfrac_avg": 0.33984375,
366
+ "policy/entropy_avg": 0.5588769912719727,
367
+ "step": 17,
368
+ "val/clipfrac_avg": 0.0,
369
+ "val/num_eos_tokens": 2173,
370
+ "val/ratio": 0.9999226331710815,
371
+ "val/ratio_var": 2.539955858082976e-06
372
+ },
373
+ {
374
+ "episode": 9728,
375
+ "epoch": 0.041672378341329676,
376
+ "eps": 6,
377
+ "loss/policy_avg": 0.30208033323287964,
378
+ "lr": 2.7890625e-06,
379
+ "objective/entropy": -37.54444122314453,
380
+ "objective/kl": 17.52410316467285,
381
+ "objective/non_score_reward": -0.08762051165103912,
382
+ "objective/rlhf_reward": 8.367655754089355,
383
+ "objective/scores": 8.455276489257812,
384
+ "policy/approxkl_avg": 0.20374038815498352,
385
+ "policy/clipfrac_avg": 0.33984375,
386
+ "policy/entropy_avg": 0.5387827157974243,
387
+ "step": 18,
388
+ "val/clipfrac_avg": 0.0,
389
+ "val/num_eos_tokens": 2168,
390
+ "val/ratio": 0.99991774559021,
391
+ "val/ratio_var": 3.0660403353977017e-06
392
+ },
393
+ {
394
+ "episode": 10240,
395
+ "epoch": 0.04386566141192598,
396
+ "eps": 6,
397
+ "loss/policy_avg": 0.22399874031543732,
398
+ "lr": 2.77734375e-06,
399
+ "objective/entropy": -39.96161651611328,
400
+ "objective/kl": 17.749210357666016,
401
+ "objective/non_score_reward": -0.08874605596065521,
402
+ "objective/rlhf_reward": 8.257219314575195,
403
+ "objective/scores": 8.345966339111328,
404
+ "policy/approxkl_avg": 0.2131994068622589,
405
+ "policy/clipfrac_avg": 0.359375,
406
+ "policy/entropy_avg": 0.5233398079872131,
407
+ "step": 19,
408
+ "val/clipfrac_avg": 0.0,
409
+ "val/num_eos_tokens": 2201,
410
+ "val/ratio": 1.0001959800720215,
411
+ "val/ratio_var": 3.099843979725847e-06
412
+ },
413
+ {
414
+ "episode": 10752,
415
+ "epoch": 0.04605894448252228,
416
+ "eps": 6,
417
+ "loss/policy_avg": 0.16000297665596008,
418
+ "lr": 2.765625e-06,
419
+ "objective/entropy": -40.21980285644531,
420
+ "objective/kl": 18.693737030029297,
421
+ "objective/non_score_reward": -0.09346868097782135,
422
+ "objective/rlhf_reward": 8.49026107788086,
423
+ "objective/scores": 8.58372974395752,
424
+ "policy/approxkl_avg": 0.2675846219062805,
425
+ "policy/clipfrac_avg": 0.38671875,
426
+ "policy/entropy_avg": 0.524081826210022,
427
+ "step": 20,
428
+ "val/clipfrac_avg": 0.0,
429
+ "val/num_eos_tokens": 2342,
430
+ "val/ratio": 0.9995419979095459,
431
+ "val/ratio_var": 3.2623988772684243e-06
432
+ },
433
+ {
434
+ "episode": 11264,
435
+ "epoch": 0.048252227553118573,
436
+ "eps": 6,
437
+ "loss/policy_avg": 0.23491168022155762,
438
+ "lr": 2.75390625e-06,
439
+ "objective/entropy": -38.962947845458984,
440
+ "objective/kl": 19.98727035522461,
441
+ "objective/non_score_reward": -0.099936343729496,
442
+ "objective/rlhf_reward": 8.68085765838623,
443
+ "objective/scores": 8.780793190002441,
444
+ "policy/approxkl_avg": 0.19925493001937866,
445
+ "policy/clipfrac_avg": 0.3359375,
446
+ "policy/entropy_avg": 0.5088595151901245,
447
+ "step": 21,
448
+ "val/clipfrac_avg": 0.0,
449
+ "val/num_eos_tokens": 2385,
450
+ "val/ratio": 0.9999128580093384,
451
+ "val/ratio_var": 3.195402769051725e-06
452
+ },
453
+ {
454
+ "episode": 11776,
455
+ "epoch": 0.05044551062371487,
456
+ "eps": 6,
457
+ "loss/policy_avg": 0.20272764563560486,
458
+ "lr": 2.7421875e-06,
459
+ "objective/entropy": -39.06739807128906,
460
+ "objective/kl": 20.73778533935547,
461
+ "objective/non_score_reward": -0.10368892550468445,
462
+ "objective/rlhf_reward": 8.66939640045166,
463
+ "objective/scores": 8.77308464050293,
464
+ "policy/approxkl_avg": 0.24808868765830994,
465
+ "policy/clipfrac_avg": 0.3671875,
466
+ "policy/entropy_avg": 0.5206277370452881,
467
+ "step": 22,
468
+ "val/clipfrac_avg": 0.0,
469
+ "val/num_eos_tokens": 2328,
470
+ "val/ratio": 0.9997999668121338,
471
+ "val/ratio_var": 2.6692378014558926e-06
472
+ },
473
+ {
474
+ "episode": 12288,
475
+ "epoch": 0.052638793694311174,
476
+ "eps": 6,
477
+ "loss/policy_avg": 0.1657373309135437,
478
+ "lr": 2.73046875e-06,
479
+ "objective/entropy": -41.76697540283203,
480
+ "objective/kl": 21.265926361083984,
481
+ "objective/non_score_reward": -0.10632962733507156,
482
+ "objective/rlhf_reward": 8.899429321289062,
483
+ "objective/scores": 9.005760192871094,
484
+ "policy/approxkl_avg": 0.21000100672245026,
485
+ "policy/clipfrac_avg": 0.330078125,
486
+ "policy/entropy_avg": 0.4891219437122345,
487
+ "step": 23,
488
+ "val/clipfrac_avg": 0.0,
489
+ "val/num_eos_tokens": 2460,
490
+ "val/ratio": 1.000035285949707,
491
+ "val/ratio_var": 3.269958597229561e-06
492
+ },
493
+ {
494
+ "episode": 12800,
495
+ "epoch": 0.05483207676490747,
496
+ "eps": 6,
497
+ "loss/policy_avg": 0.17451098561286926,
498
+ "lr": 2.71875e-06,
499
+ "objective/entropy": -40.41400146484375,
500
+ "objective/kl": 22.69107437133789,
501
+ "objective/non_score_reward": -0.11345535516738892,
502
+ "objective/rlhf_reward": 8.97903060913086,
503
+ "objective/scores": 9.092485427856445,
504
+ "policy/approxkl_avg": 0.2612704336643219,
505
+ "policy/clipfrac_avg": 0.37109375,
506
+ "policy/entropy_avg": 0.4792313873767853,
507
+ "step": 24,
508
+ "val/clipfrac_avg": 0.0,
509
+ "val/num_eos_tokens": 2361,
510
+ "val/ratio": 0.9997962713241577,
511
+ "val/ratio_var": 4.033691766380798e-06
512
+ },
513
+ {
514
+ "episode": 13312,
515
+ "epoch": 0.05702535983550377,
516
+ "eps": 6,
517
+ "loss/policy_avg": 0.3060297966003418,
518
+ "lr": 2.70703125e-06,
519
+ "objective/entropy": -39.72850036621094,
520
+ "objective/kl": 24.020402908325195,
521
+ "objective/non_score_reward": -0.12010201811790466,
522
+ "objective/rlhf_reward": 8.757214546203613,
523
+ "objective/scores": 8.877317428588867,
524
+ "policy/approxkl_avg": 0.20275026559829712,
525
+ "policy/clipfrac_avg": 0.333984375,
526
+ "policy/entropy_avg": 0.47031739354133606,
527
+ "step": 25,
528
+ "val/clipfrac_avg": 0.0,
529
+ "val/num_eos_tokens": 2365,
530
+ "val/ratio": 1.0004143714904785,
531
+ "val/ratio_var": 4.571053977997508e-06
532
+ },
533
+ {
534
+ "episode": 13824,
535
+ "epoch": 0.05921864290610007,
536
+ "eps": 6,
537
+ "loss/policy_avg": 0.2459714114665985,
538
+ "lr": 2.6953125e-06,
539
+ "objective/entropy": -42.67210388183594,
540
+ "objective/kl": 24.816062927246094,
541
+ "objective/non_score_reward": -0.12408032268285751,
542
+ "objective/rlhf_reward": 8.767488479614258,
543
+ "objective/scores": 8.891569137573242,
544
+ "policy/approxkl_avg": 0.16891685128211975,
545
+ "policy/clipfrac_avg": 0.314453125,
546
+ "policy/entropy_avg": 0.4268378019332886,
547
+ "step": 26,
548
+ "val/clipfrac_avg": 0.0,
549
+ "val/num_eos_tokens": 2338,
550
+ "val/ratio": 1.0001232624053955,
551
+ "val/ratio_var": 2.055706318060402e-06
552
+ },
553
+ {
554
+ "episode": 14336,
555
+ "epoch": 0.06141192597669637,
556
+ "eps": 6,
557
+ "loss/policy_avg": 0.2477143555879593,
558
+ "lr": 2.68359375e-06,
559
+ "objective/entropy": -40.98384094238281,
560
+ "objective/kl": 26.158355712890625,
561
+ "objective/non_score_reward": -0.1307917833328247,
562
+ "objective/rlhf_reward": 8.869270324707031,
563
+ "objective/scores": 9.000061988830566,
564
+ "policy/approxkl_avg": 0.21648374199867249,
565
+ "policy/clipfrac_avg": 0.35546875,
566
+ "policy/entropy_avg": 0.43942803144454956,
567
+ "step": 27,
568
+ "val/clipfrac_avg": 0.0,
569
+ "val/num_eos_tokens": 2325,
570
+ "val/ratio": 1.00002920627594,
571
+ "val/ratio_var": 3.6563492358254734e-06
572
+ },
573
+ {
574
+ "episode": 14848,
575
+ "epoch": 0.06360520904729267,
576
+ "eps": 6,
577
+ "loss/policy_avg": 0.18666958808898926,
578
+ "lr": 2.671875e-06,
579
+ "objective/entropy": -39.86817169189453,
580
+ "objective/kl": 26.283226013183594,
581
+ "objective/non_score_reward": -0.13141612708568573,
582
+ "objective/rlhf_reward": 9.001875877380371,
583
+ "objective/scores": 9.133292198181152,
584
+ "policy/approxkl_avg": 0.22604086995124817,
585
+ "policy/clipfrac_avg": 0.37109375,
586
+ "policy/entropy_avg": 0.4275825023651123,
587
+ "step": 28,
588
+ "val/clipfrac_avg": 0.0,
589
+ "val/num_eos_tokens": 2388,
590
+ "val/ratio": 0.999720573425293,
591
+ "val/ratio_var": 2.5127403660007985e-06
592
+ },
593
+ {
594
+ "episode": 15360,
595
+ "epoch": 0.06579849211788896,
596
+ "eps": 6,
597
+ "loss/policy_avg": 0.21501849591732025,
598
+ "lr": 2.66015625e-06,
599
+ "objective/entropy": -34.56945037841797,
600
+ "objective/kl": 28.54375457763672,
601
+ "objective/non_score_reward": -0.14271876215934753,
602
+ "objective/rlhf_reward": 8.915486335754395,
603
+ "objective/scores": 9.058204650878906,
604
+ "policy/approxkl_avg": 0.23621705174446106,
605
+ "policy/clipfrac_avg": 0.361328125,
606
+ "policy/entropy_avg": 0.42039746046066284,
607
+ "step": 29,
608
+ "val/clipfrac_avg": 0.0,
609
+ "val/num_eos_tokens": 2389,
610
+ "val/ratio": 0.9995044469833374,
611
+ "val/ratio_var": 4.271460511517944e-06
612
+ },
613
+ {
614
+ "episode": 15872,
615
+ "epoch": 0.06799177518848526,
616
+ "eps": 6,
617
+ "loss/policy_avg": 0.343304842710495,
618
+ "lr": 2.6484375e-06,
619
+ "objective/entropy": -35.42715072631836,
620
+ "objective/kl": 28.722532272338867,
621
+ "objective/non_score_reward": -0.14361265301704407,
622
+ "objective/rlhf_reward": 9.159289360046387,
623
+ "objective/scores": 9.302902221679688,
624
+ "policy/approxkl_avg": 0.221934974193573,
625
+ "policy/clipfrac_avg": 0.3046875,
626
+ "policy/entropy_avg": 0.3945683240890503,
627
+ "step": 30,
628
+ "val/clipfrac_avg": 0.0,
629
+ "val/num_eos_tokens": 2236,
630
+ "val/ratio": 1.0000040531158447,
631
+ "val/ratio_var": 3.5196928820369067e-06
632
+ },
633
+ {
634
+ "episode": 16384,
635
+ "epoch": 0.07018505825908156,
636
+ "eps": 6,
637
+ "loss/policy_avg": 0.4800252318382263,
638
+ "lr": 2.63671875e-06,
639
+ "objective/entropy": -33.33995056152344,
640
+ "objective/kl": 29.83435821533203,
641
+ "objective/non_score_reward": -0.14917179942131042,
642
+ "objective/rlhf_reward": 9.096776962280273,
643
+ "objective/scores": 9.245948791503906,
644
+ "policy/approxkl_avg": 0.267780601978302,
645
+ "policy/clipfrac_avg": 0.337890625,
646
+ "policy/entropy_avg": 0.401678204536438,
647
+ "step": 31,
648
+ "val/clipfrac_avg": 0.0,
649
+ "val/num_eos_tokens": 2179,
650
+ "val/ratio": 1.0002527236938477,
651
+ "val/ratio_var": 3.2502473459317116e-06
652
+ },
653
+ {
654
+ "episode": 16896,
655
+ "epoch": 0.07237834132967787,
656
+ "eps": 6,
657
+ "loss/policy_avg": 0.2809556722640991,
658
+ "lr": 2.6250000000000003e-06,
659
+ "objective/entropy": -34.06169509887695,
660
+ "objective/kl": 31.27130889892578,
661
+ "objective/non_score_reward": -0.156356543302536,
662
+ "objective/rlhf_reward": 9.190299034118652,
663
+ "objective/scores": 9.346654891967773,
664
+ "policy/approxkl_avg": 0.2590065896511078,
665
+ "policy/clipfrac_avg": 0.333984375,
666
+ "policy/entropy_avg": 0.39545533061027527,
667
+ "step": 32,
668
+ "val/clipfrac_avg": 0.0,
669
+ "val/num_eos_tokens": 2267,
670
+ "val/ratio": 1.0000057220458984,
671
+ "val/ratio_var": 3.7364695799624315e-06
672
+ },
673
+ {
674
+ "episode": 17408,
675
+ "epoch": 0.07457162440027416,
676
+ "eps": 6,
677
+ "loss/policy_avg": 0.22726668417453766,
678
+ "lr": 2.61328125e-06,
679
+ "objective/entropy": -29.238605499267578,
680
+ "objective/kl": 32.63374710083008,
681
+ "objective/non_score_reward": -0.16316872835159302,
682
+ "objective/rlhf_reward": 9.312760353088379,
683
+ "objective/scores": 9.475929260253906,
684
+ "policy/approxkl_avg": 0.25571638345718384,
685
+ "policy/clipfrac_avg": 0.328125,
686
+ "policy/entropy_avg": 0.38992634415626526,
687
+ "step": 33,
688
+ "val/clipfrac_avg": 0.0,
689
+ "val/num_eos_tokens": 2153,
690
+ "val/ratio": 0.9997001886367798,
691
+ "val/ratio_var": 2.696072442631703e-06
692
+ },
693
+ {
694
+ "episode": 17920,
695
+ "epoch": 0.07676490747087046,
696
+ "eps": 6,
697
+ "loss/policy_avg": 0.31760329008102417,
698
+ "lr": 2.6015625e-06,
699
+ "objective/entropy": -30.36035919189453,
700
+ "objective/kl": 33.105316162109375,
701
+ "objective/non_score_reward": -0.16552656888961792,
702
+ "objective/rlhf_reward": 9.344242095947266,
703
+ "objective/scores": 9.509769439697266,
704
+ "policy/approxkl_avg": 0.26468268036842346,
705
+ "policy/clipfrac_avg": 0.31640625,
706
+ "policy/entropy_avg": 0.3962373435497284,
707
+ "step": 34,
708
+ "val/clipfrac_avg": 0.0,
709
+ "val/num_eos_tokens": 2063,
710
+ "val/ratio": 0.9997377395629883,
711
+ "val/ratio_var": 3.3684978006931487e-06
712
+ },
713
+ {
714
+ "episode": 18432,
715
+ "epoch": 0.07895819054146676,
716
+ "eps": 6,
717
+ "loss/policy_avg": 0.29205620288848877,
718
+ "lr": 2.5898437500000003e-06,
719
+ "objective/entropy": -31.297311782836914,
720
+ "objective/kl": 34.06657409667969,
721
+ "objective/non_score_reward": -0.1703328639268875,
722
+ "objective/rlhf_reward": 9.36227035522461,
723
+ "objective/scores": 9.532604217529297,
724
+ "policy/approxkl_avg": 0.23790021240711212,
725
+ "policy/clipfrac_avg": 0.359375,
726
+ "policy/entropy_avg": 0.39074409008026123,
727
+ "step": 35,
728
+ "val/clipfrac_avg": 0.0,
729
+ "val/num_eos_tokens": 2030,
730
+ "val/ratio": 1.0001909732818604,
731
+ "val/ratio_var": 4.307842573325615e-06
732
+ },
733
+ {
734
+ "episode": 18944,
735
+ "epoch": 0.08115147361206305,
736
+ "eps": 6,
737
+ "loss/policy_avg": 0.27079224586486816,
738
+ "lr": 2.578125e-06,
739
+ "objective/entropy": -31.059741973876953,
740
+ "objective/kl": 34.23755645751953,
741
+ "objective/non_score_reward": -0.17118775844573975,
742
+ "objective/rlhf_reward": 9.575616836547852,
743
+ "objective/scores": 9.746803283691406,
744
+ "policy/approxkl_avg": 0.2780650556087494,
745
+ "policy/clipfrac_avg": 0.359375,
746
+ "policy/entropy_avg": 0.36144694685935974,
747
+ "step": 36,
748
+ "val/clipfrac_avg": 0.0,
749
+ "val/num_eos_tokens": 1946,
750
+ "val/ratio": 1.0000677108764648,
751
+ "val/ratio_var": 5.890160991839366e-06
752
+ },
753
+ {
754
+ "episode": 19456,
755
+ "epoch": 0.08334475668265935,
756
+ "eps": 6,
757
+ "loss/policy_avg": 0.35481345653533936,
758
+ "lr": 2.56640625e-06,
759
+ "objective/entropy": -29.504077911376953,
760
+ "objective/kl": 35.572723388671875,
761
+ "objective/non_score_reward": -0.17786362767219543,
762
+ "objective/rlhf_reward": 9.650212287902832,
763
+ "objective/scores": 9.828075408935547,
764
+ "policy/approxkl_avg": 0.27989301085472107,
765
+ "policy/clipfrac_avg": 0.341796875,
766
+ "policy/entropy_avg": 0.36728498339653015,
767
+ "step": 37,
768
+ "val/clipfrac_avg": 0.0,
769
+ "val/num_eos_tokens": 1913,
770
+ "val/ratio": 1.000013828277588,
771
+ "val/ratio_var": 3.834660674328916e-06
772
+ },
773
+ {
774
+ "episode": 19968,
775
+ "epoch": 0.08553803975325565,
776
+ "eps": 6,
777
+ "loss/policy_avg": 0.32023417949676514,
778
+ "lr": 2.5546875000000003e-06,
779
+ "objective/entropy": -32.94467544555664,
780
+ "objective/kl": 35.05192565917969,
781
+ "objective/non_score_reward": -0.17525961995124817,
782
+ "objective/rlhf_reward": 9.715032577514648,
783
+ "objective/scores": 9.89029312133789,
784
+ "policy/approxkl_avg": 0.28202196955680847,
785
+ "policy/clipfrac_avg": 0.32421875,
786
+ "policy/entropy_avg": 0.3595014214515686,
787
+ "step": 38,
788
+ "val/clipfrac_avg": 0.0,
789
+ "val/num_eos_tokens": 1895,
790
+ "val/ratio": 1.0004128217697144,
791
+ "val/ratio_var": 3.976154857809888e-06
792
+ },
793
+ {
794
+ "episode": 20480,
795
+ "epoch": 0.08773132282385196,
796
+ "eps": 6,
797
+ "loss/policy_avg": 0.3284844160079956,
798
+ "lr": 2.54296875e-06,
799
+ "objective/entropy": -33.950965881347656,
800
+ "objective/kl": 35.06300354003906,
801
+ "objective/non_score_reward": -0.1753150224685669,
802
+ "objective/rlhf_reward": 9.614059448242188,
803
+ "objective/scores": 9.789375305175781,
804
+ "policy/approxkl_avg": 0.2808968126773834,
805
+ "policy/clipfrac_avg": 0.3359375,
806
+ "policy/entropy_avg": 0.3299378454685211,
807
+ "step": 39,
808
+ "val/clipfrac_avg": 0.0,
809
+ "val/num_eos_tokens": 1822,
810
+ "val/ratio": 1.0001529455184937,
811
+ "val/ratio_var": 3.939061116398079e-06
812
+ },
813
+ {
814
+ "episode": 20992,
815
+ "epoch": 0.08992460589444826,
816
+ "eps": 6,
817
+ "loss/policy_avg": 0.25570446252822876,
818
+ "lr": 2.53125e-06,
819
+ "objective/entropy": -33.10383605957031,
820
+ "objective/kl": 37.40079116821289,
821
+ "objective/non_score_reward": -0.1870039403438568,
822
+ "objective/rlhf_reward": 9.649069786071777,
823
+ "objective/scores": 9.836073875427246,
824
+ "policy/approxkl_avg": 0.28119906783103943,
825
+ "policy/clipfrac_avg": 0.349609375,
826
+ "policy/entropy_avg": 0.33145490288734436,
827
+ "step": 40,
828
+ "val/clipfrac_avg": 0.0,
829
+ "val/num_eos_tokens": 1757,
830
+ "val/ratio": 0.999976396560669,
831
+ "val/ratio_var": 3.935067525162594e-06
832
+ },
833
+ {
834
+ "episode": 21504,
835
+ "epoch": 0.09211788896504455,
836
+ "eps": 6,
837
+ "loss/policy_avg": 0.3803982436656952,
838
+ "lr": 2.5195312500000003e-06,
839
+ "objective/entropy": -35.517391204833984,
840
+ "objective/kl": 35.55995178222656,
841
+ "objective/non_score_reward": -0.1777997612953186,
842
+ "objective/rlhf_reward": 9.623468399047852,
843
+ "objective/scores": 9.801267623901367,
844
+ "policy/approxkl_avg": 0.28167036175727844,
845
+ "policy/clipfrac_avg": 0.3828125,
846
+ "policy/entropy_avg": 0.33138006925582886,
847
+ "step": 41,
848
+ "val/clipfrac_avg": 0.0,
849
+ "val/num_eos_tokens": 1765,
850
+ "val/ratio": 1.000349760055542,
851
+ "val/ratio_var": 4.55302324553486e-06
852
+ },
853
+ {
854
+ "episode": 22016,
855
+ "epoch": 0.09431117203564085,
856
+ "eps": 6,
857
+ "loss/policy_avg": 0.2066161185503006,
858
+ "lr": 2.5078125e-06,
859
+ "objective/entropy": -34.937923431396484,
860
+ "objective/kl": 34.828224182128906,
861
+ "objective/non_score_reward": -0.17414110898971558,
862
+ "objective/rlhf_reward": 10.007875442504883,
863
+ "objective/scores": 10.18201732635498,
864
+ "policy/approxkl_avg": 0.25226256251335144,
865
+ "policy/clipfrac_avg": 0.349609375,
866
+ "policy/entropy_avg": 0.31548449397087097,
867
+ "step": 42,
868
+ "val/clipfrac_avg": 0.0,
869
+ "val/num_eos_tokens": 1692,
870
+ "val/ratio": 1.0003182888031006,
871
+ "val/ratio_var": 3.0859027901897207e-06
872
+ },
873
+ {
874
+ "episode": 22528,
875
+ "epoch": 0.09650445510623715,
876
+ "eps": 6,
877
+ "loss/policy_avg": 0.21718214452266693,
878
+ "lr": 2.49609375e-06,
879
+ "objective/entropy": -36.30521774291992,
880
+ "objective/kl": 35.64817810058594,
881
+ "objective/non_score_reward": -0.17824086546897888,
882
+ "objective/rlhf_reward": 9.524388313293457,
883
+ "objective/scores": 9.702629089355469,
884
+ "policy/approxkl_avg": 0.3034236431121826,
885
+ "policy/clipfrac_avg": 0.359375,
886
+ "policy/entropy_avg": 0.333159863948822,
887
+ "step": 43,
888
+ "val/clipfrac_avg": 0.0,
889
+ "val/num_eos_tokens": 1736,
890
+ "val/ratio": 0.9996454119682312,
891
+ "val/ratio_var": 3.5439479688648134e-06
892
+ },
893
+ {
894
+ "episode": 23040,
895
+ "epoch": 0.09869773817683344,
896
+ "eps": 6,
897
+ "loss/policy_avg": 0.20844724774360657,
898
+ "lr": 2.4843750000000002e-06,
899
+ "objective/entropy": -37.294647216796875,
900
+ "objective/kl": 34.449188232421875,
901
+ "objective/non_score_reward": -0.17224593460559845,
902
+ "objective/rlhf_reward": 9.562028884887695,
903
+ "objective/scores": 9.734274864196777,
904
+ "policy/approxkl_avg": 0.2691035270690918,
905
+ "policy/clipfrac_avg": 0.369140625,
906
+ "policy/entropy_avg": 0.3254753351211548,
907
+ "step": 44,
908
+ "val/clipfrac_avg": 0.0,
909
+ "val/num_eos_tokens": 1712,
910
+ "val/ratio": 0.9999078512191772,
911
+ "val/ratio_var": 3.873953573929612e-06
912
+ },
913
+ {
914
+ "episode": 23552,
915
+ "epoch": 0.10089102124742974,
916
+ "eps": 6,
917
+ "loss/policy_avg": 0.26617997884750366,
918
+ "lr": 2.47265625e-06,
919
+ "objective/entropy": -38.45973205566406,
920
+ "objective/kl": 33.36463165283203,
921
+ "objective/non_score_reward": -0.166823148727417,
922
+ "objective/rlhf_reward": 9.529053688049316,
923
+ "objective/scores": 9.695877075195312,
924
+ "policy/approxkl_avg": 0.2848384976387024,
925
+ "policy/clipfrac_avg": 0.32421875,
926
+ "policy/entropy_avg": 0.33362650871276855,
927
+ "step": 45,
928
+ "val/clipfrac_avg": 0.0,
929
+ "val/num_eos_tokens": 1776,
930
+ "val/ratio": 0.9998953938484192,
931
+ "val/ratio_var": 4.734669346362352e-06
932
+ },
933
+ {
934
+ "episode": 24064,
935
+ "epoch": 0.10308430431802605,
936
+ "eps": 6,
937
+ "loss/policy_avg": 0.14936941862106323,
938
+ "lr": 2.4609375e-06,
939
+ "objective/entropy": -41.61138153076172,
940
+ "objective/kl": 32.80962371826172,
941
+ "objective/non_score_reward": -0.16404810547828674,
942
+ "objective/rlhf_reward": 9.842905044555664,
943
+ "objective/scores": 10.006953239440918,
944
+ "policy/approxkl_avg": 0.237547367811203,
945
+ "policy/clipfrac_avg": 0.376953125,
946
+ "policy/entropy_avg": 0.311695396900177,
947
+ "step": 46,
948
+ "val/clipfrac_avg": 0.0,
949
+ "val/num_eos_tokens": 1758,
950
+ "val/ratio": 1.0000953674316406,
951
+ "val/ratio_var": 2.808139925036812e-06
952
+ },
953
+ {
954
+ "episode": 24576,
955
+ "epoch": 0.10527758738862235,
956
+ "eps": 6,
957
+ "loss/policy_avg": 0.30603936314582825,
958
+ "lr": 2.4492187500000002e-06,
959
+ "objective/entropy": -41.17787551879883,
960
+ "objective/kl": 32.624759674072266,
961
+ "objective/non_score_reward": -0.16312378644943237,
962
+ "objective/rlhf_reward": 9.724346160888672,
963
+ "objective/scores": 9.887470245361328,
964
+ "policy/approxkl_avg": 0.29364651441574097,
965
+ "policy/clipfrac_avg": 0.353515625,
966
+ "policy/entropy_avg": 0.3342776894569397,
967
+ "step": 47,
968
+ "val/clipfrac_avg": 0.0,
969
+ "val/num_eos_tokens": 1791,
970
+ "val/ratio": 1.000326156616211,
971
+ "val/ratio_var": 4.5322440200834535e-06
972
+ },
973
+ {
974
+ "episode": 25088,
975
+ "epoch": 0.10747087045921865,
976
+ "eps": 6,
977
+ "loss/policy_avg": 0.2642097771167755,
978
+ "lr": 2.4375e-06,
979
+ "objective/entropy": -39.200443267822266,
980
+ "objective/kl": 32.45161437988281,
981
+ "objective/non_score_reward": -0.1622580885887146,
982
+ "objective/rlhf_reward": 9.750046730041504,
983
+ "objective/scores": 9.912304878234863,
984
+ "policy/approxkl_avg": 0.30972546339035034,
985
+ "policy/clipfrac_avg": 0.345703125,
986
+ "policy/entropy_avg": 0.33516496419906616,
987
+ "step": 48,
988
+ "val/clipfrac_avg": 0.0,
989
+ "val/num_eos_tokens": 1872,
990
+ "val/ratio": 1.0000823736190796,
991
+ "val/ratio_var": 3.669098759928602e-06
992
+ },
993
+ {
994
+ "episode": 25600,
995
+ "epoch": 0.10966415352981494,
996
+ "eps": 6,
997
+ "loss/policy_avg": 0.2650497555732727,
998
+ "lr": 2.42578125e-06,
999
+ "objective/entropy": -42.08147430419922,
1000
+ "objective/kl": 32.27870178222656,
1001
+ "objective/non_score_reward": -0.16139349341392517,
1002
+ "objective/rlhf_reward": 9.627498626708984,
1003
+ "objective/scores": 9.788891792297363,
1004
+ "policy/approxkl_avg": 0.24272559583187103,
1005
+ "policy/clipfrac_avg": 0.34375,
1006
+ "policy/entropy_avg": 0.3419151306152344,
1007
+ "step": 49,
1008
+ "val/clipfrac_avg": 0.0,
1009
+ "val/num_eos_tokens": 1885,
1010
+ "val/ratio": 1.0000895261764526,
1011
+ "val/ratio_var": 3.1425495308212703e-06
1012
+ },
1013
+ {
1014
+ "episode": 26112,
1015
+ "epoch": 0.11185743660041124,
1016
+ "eps": 6,
1017
+ "loss/policy_avg": 0.19408683478832245,
1018
+ "lr": 2.4140625000000002e-06,
1019
+ "objective/entropy": -40.94256591796875,
1020
+ "objective/kl": 31.59925651550293,
1021
+ "objective/non_score_reward": -0.1579962968826294,
1022
+ "objective/rlhf_reward": 9.357413291931152,
1023
+ "objective/scores": 9.515409469604492,
1024
+ "policy/approxkl_avg": 0.2539810538291931,
1025
+ "policy/clipfrac_avg": 0.345703125,
1026
+ "policy/entropy_avg": 0.33615610003471375,
1027
+ "step": 50,
1028
+ "val/clipfrac_avg": 0.0,
1029
+ "val/num_eos_tokens": 1861,
1030
+ "val/ratio": 1.0000097751617432,
1031
+ "val/ratio_var": 4.071615421707975e-06
1032
+ },
1033
+ {
1034
+ "episode": 26624,
1035
+ "epoch": 0.11405071967100754,
1036
+ "eps": 6,
1037
+ "loss/policy_avg": 0.12506979703903198,
1038
+ "lr": 2.40234375e-06,
1039
+ "objective/entropy": -43.869224548339844,
1040
+ "objective/kl": 30.23246955871582,
1041
+ "objective/non_score_reward": -0.15116234123706818,
1042
+ "objective/rlhf_reward": 9.635049819946289,
1043
+ "objective/scores": 9.786212921142578,
1044
+ "policy/approxkl_avg": 0.22058835625648499,
1045
+ "policy/clipfrac_avg": 0.3828125,
1046
+ "policy/entropy_avg": 0.35689815878868103,
1047
+ "step": 51,
1048
+ "val/clipfrac_avg": 0.0,
1049
+ "val/num_eos_tokens": 1916,
1050
+ "val/ratio": 1.0003091096878052,
1051
+ "val/ratio_var": 3.508681174935191e-06
1052
+ }
1053
+ ],
1054
+ "logging_steps": 1,
1055
+ "max_steps": 128,
1056
+ "num_input_tokens_seen": 0,
1057
+ "num_train_epochs": 1.122960932145305,
1058
+ "save_steps": 26,
1059
+ "stateful_callbacks": {
1060
+ "TrainerControl": {
1061
+ "args": {
1062
+ "should_epoch_stop": false,
1063
+ "should_evaluate": false,
1064
+ "should_log": true,
1065
+ "should_save": true,
1066
+ "should_training_stop": false
1067
+ },
1068
+ "attributes": {}
1069
+ }
1070
+ },
1071
+ "total_flos": 0,
1072
+ "train_batch_size": null,
1073
+ "trial_name": null,
1074
+ "trial_params": null
1075
+ }
checkpoint-52/training_args.bin ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:30edc5fdb14278c841efedd8fdc4865c8766890f13483d7f2957be7405e19d4d
3
+ size 6584