wzx111 commited on
Commit
2607a31
·
verified ·
1 Parent(s): fc0d3bf

Model save

Browse files
Files changed (3) hide show
  1. all_results.json +8 -0
  2. train_results.json +8 -0
  3. trainer_state.json +1440 -0
all_results.json ADDED
@@ -0,0 +1,8 @@
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "total_flos": 0.0,
3
+ "train_loss": 2.0794413089752197,
4
+ "train_runtime": 3642.2415,
5
+ "train_samples": 1592,
6
+ "train_samples_per_second": 0.874,
7
+ "train_steps_per_second": 0.013
8
+ }
train_results.json ADDED
@@ -0,0 +1,8 @@
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "total_flos": 0.0,
3
+ "train_loss": 2.0794413089752197,
4
+ "train_runtime": 3642.2415,
5
+ "train_samples": 1592,
6
+ "train_samples_per_second": 0.874,
7
+ "train_steps_per_second": 0.013
8
+ }
trainer_state.json ADDED
@@ -0,0 +1,1440 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "best_global_step": null,
3
+ "best_metric": null,
4
+ "best_model_checkpoint": null,
5
+ "epoch": 2.0,
6
+ "eval_steps": 500,
7
+ "global_step": 48,
8
+ "is_hyper_param_search": false,
9
+ "is_local_process_zero": true,
10
+ "is_world_process_zero": true,
11
+ "log_history": [
12
+ {
13
+ "adpo/anchor_kl": 0.0,
14
+ "adpo/anchor_updates": 0.0,
15
+ "adpo/mean_tau": 0.5728723406791687,
16
+ "completions/clipped_ratio": 0.3515625,
17
+ "completions/max_length": 1024.0,
18
+ "completions/max_terminated_length": 1023.0,
19
+ "completions/mean_length": 820.990234375,
20
+ "completions/mean_terminated_length": 710.9246826171875,
21
+ "completions/min_length": 237.0,
22
+ "completions/min_terminated_length": 237.0,
23
+ "entropy": 0.10525308595970273,
24
+ "epoch": 0.041666666666666664,
25
+ "frac_reward_zero_std": 0.15625,
26
+ "grad_norm": 403.2952880859375,
27
+ "learning_rate": 0.0,
28
+ "loss": 2.0794,
29
+ "num_tokens": 476003.0,
30
+ "reward": 0.5155274868011475,
31
+ "reward_std": 0.28498148918151855,
32
+ "rewards/<lambda>/mean": 0.5155274868011475,
33
+ "rewards/<lambda>/std": 0.506263017654419,
34
+ "sampling/importance_sampling_ratio/max": 2.0,
35
+ "sampling/importance_sampling_ratio/mean": 1.0000114440917969,
36
+ "sampling/importance_sampling_ratio/min": 0.0859043225646019,
37
+ "sampling/sampling_logp_difference/max": 2.4545211791992188,
38
+ "sampling/sampling_logp_difference/mean": 0.00970321986824274,
39
+ "step": 1,
40
+ "step_time": 77.00020869076252
41
+ },
42
+ {
43
+ "adpo/anchor_kl": 0.0,
44
+ "adpo/anchor_updates": 0.0,
45
+ "completions/clipped_ratio": 0.345703125,
46
+ "completions/max_length": 1024.0,
47
+ "completions/max_terminated_length": 1023.0,
48
+ "completions/mean_length": 849.328125,
49
+ "completions/mean_terminated_length": 757.038818359375,
50
+ "completions/min_length": 295.0,
51
+ "completions/min_terminated_length": 295.0,
52
+ "entropy": 0.10956642962992191,
53
+ "epoch": 0.08333333333333333,
54
+ "frac_reward_zero_std": 0.140625,
55
+ "grad_norm": 413.4783935546875,
56
+ "learning_rate": 3e-06,
57
+ "loss": 2.0794,
58
+ "num_tokens": 966003.0,
59
+ "reward": 0.4997570812702179,
60
+ "reward_std": 0.3475712537765503,
61
+ "rewards/<lambda>/mean": 0.4997570812702179,
62
+ "rewards/<lambda>/std": 0.506650984287262,
63
+ "sampling/importance_sampling_ratio/max": 2.0,
64
+ "sampling/importance_sampling_ratio/mean": 0.9999332427978516,
65
+ "sampling/importance_sampling_ratio/min": 0.19995339214801788,
66
+ "sampling/sampling_logp_difference/max": 1.6096709966659546,
67
+ "sampling/sampling_logp_difference/mean": 0.010090015828609467,
68
+ "step": 2,
69
+ "step_time": 60.77968284860253
70
+ },
71
+ {
72
+ "adpo/anchor_kl": 0.0,
73
+ "adpo/anchor_updates": 0.0,
74
+ "completions/clipped_ratio": 0.478515625,
75
+ "completions/max_length": 1024.0,
76
+ "completions/max_terminated_length": 1024.0,
77
+ "completions/mean_length": 885.806640625,
78
+ "completions/mean_terminated_length": 759.0,
79
+ "completions/min_length": 288.0,
80
+ "completions/min_terminated_length": 288.0,
81
+ "entropy": 0.11405191803351045,
82
+ "epoch": 0.125,
83
+ "frac_reward_zero_std": 0.125,
84
+ "grad_norm": 371.73321533203125,
85
+ "learning_rate": 6e-06,
86
+ "loss": 2.0794,
87
+ "num_tokens": 1482840.0,
88
+ "reward": 0.3516584038734436,
89
+ "reward_std": 0.25603362917900085,
90
+ "rewards/<lambda>/mean": 0.351658433675766,
91
+ "rewards/<lambda>/std": 0.48609307408332825,
92
+ "sampling/importance_sampling_ratio/max": 2.0,
93
+ "sampling/importance_sampling_ratio/mean": 0.9999924898147583,
94
+ "sampling/importance_sampling_ratio/min": 0.10895482450723648,
95
+ "sampling/sampling_logp_difference/max": 2.2168219089508057,
96
+ "sampling/sampling_logp_difference/mean": 0.010214481502771378,
97
+ "step": 3,
98
+ "step_time": 64.36457661911845
99
+ },
100
+ {
101
+ "adpo/anchor_kl": 0.0,
102
+ "adpo/anchor_updates": 0.0,
103
+ "completions/clipped_ratio": 0.3515625,
104
+ "completions/max_length": 1024.0,
105
+ "completions/max_terminated_length": 1022.0,
106
+ "completions/mean_length": 835.21875,
107
+ "completions/mean_terminated_length": 732.867431640625,
108
+ "completions/min_length": 392.0,
109
+ "completions/min_terminated_length": 392.0,
110
+ "entropy": 0.11456855991855264,
111
+ "epoch": 0.16666666666666666,
112
+ "frac_reward_zero_std": 0.171875,
113
+ "grad_norm": 485.5037841796875,
114
+ "learning_rate": 9e-06,
115
+ "loss": 2.0794,
116
+ "num_tokens": 1965512.0,
117
+ "reward": 0.5115566253662109,
118
+ "reward_std": 0.281027227640152,
119
+ "rewards/<lambda>/mean": 0.5115566253662109,
120
+ "rewards/<lambda>/std": 0.506432831287384,
121
+ "sampling/importance_sampling_ratio/max": 2.0,
122
+ "sampling/importance_sampling_ratio/mean": 1.0000736713409424,
123
+ "sampling/importance_sampling_ratio/min": 0.12385714799165726,
124
+ "sampling/sampling_logp_difference/max": 2.0886263847351074,
125
+ "sampling/sampling_logp_difference/mean": 0.0101640485227108,
126
+ "step": 4,
127
+ "step_time": 59.65729798562825
128
+ },
129
+ {
130
+ "adpo/anchor_kl": 0.0,
131
+ "adpo/anchor_updates": 0.0,
132
+ "completions/clipped_ratio": 0.38671875,
133
+ "completions/max_length": 1024.0,
134
+ "completions/max_terminated_length": 1024.0,
135
+ "completions/mean_length": 842.9140625,
136
+ "completions/mean_terminated_length": 728.7261352539062,
137
+ "completions/min_length": 326.0,
138
+ "completions/min_terminated_length": 326.0,
139
+ "entropy": 0.11538639292120934,
140
+ "epoch": 0.20833333333333334,
141
+ "frac_reward_zero_std": 0.28125,
142
+ "grad_norm": 478.4039611816406,
143
+ "learning_rate": 1.2e-05,
144
+ "loss": 2.0794,
145
+ "num_tokens": 2454060.0,
146
+ "reward": 0.5511578321456909,
147
+ "reward_std": 0.22717225551605225,
148
+ "rewards/<lambda>/mean": 0.5511578321456909,
149
+ "rewards/<lambda>/std": 0.5034298896789551,
150
+ "sampling/importance_sampling_ratio/max": 2.0,
151
+ "sampling/importance_sampling_ratio/mean": 0.9998940229415894,
152
+ "sampling/importance_sampling_ratio/min": 0.2624821066856384,
153
+ "sampling/sampling_logp_difference/max": 1.3375723361968994,
154
+ "sampling/sampling_logp_difference/mean": 0.009679680690169334,
155
+ "step": 5,
156
+ "step_time": 61.57523806579411
157
+ },
158
+ {
159
+ "adpo/anchor_kl": 0.0,
160
+ "adpo/anchor_updates": 0.0,
161
+ "completions/clipped_ratio": 0.33984375,
162
+ "completions/max_length": 1024.0,
163
+ "completions/max_terminated_length": 1023.0,
164
+ "completions/mean_length": 829.62109375,
165
+ "completions/mean_terminated_length": 735.473388671875,
166
+ "completions/min_length": 317.0,
167
+ "completions/min_terminated_length": 317.0,
168
+ "entropy": 0.12515308475121856,
169
+ "epoch": 0.25,
170
+ "frac_reward_zero_std": 0.28125,
171
+ "grad_norm": 382.7787780761719,
172
+ "learning_rate": 1.5e-05,
173
+ "loss": 2.0794,
174
+ "num_tokens": 2938818.0,
175
+ "reward": 0.5940355062484741,
176
+ "reward_std": 0.2690280079841614,
177
+ "rewards/<lambda>/mean": 0.5940355062484741,
178
+ "rewards/<lambda>/std": 0.497295081615448,
179
+ "sampling/importance_sampling_ratio/max": 2.0,
180
+ "sampling/importance_sampling_ratio/mean": 0.9999448657035828,
181
+ "sampling/importance_sampling_ratio/min": 0.12196172028779984,
182
+ "sampling/sampling_logp_difference/max": 2.104048013687134,
183
+ "sampling/sampling_logp_difference/mean": 0.009653620421886444,
184
+ "step": 6,
185
+ "step_time": 70.90387387387455
186
+ },
187
+ {
188
+ "adpo/anchor_kl": 0.0,
189
+ "adpo/anchor_updates": 0.0,
190
+ "completions/clipped_ratio": 0.431640625,
191
+ "completions/max_length": 1024.0,
192
+ "completions/max_terminated_length": 1024.0,
193
+ "completions/mean_length": 818.556640625,
194
+ "completions/mean_terminated_length": 662.5326538085938,
195
+ "completions/min_length": 285.0,
196
+ "completions/min_terminated_length": 285.0,
197
+ "entropy": 0.1711420053616166,
198
+ "epoch": 0.2916666666666667,
199
+ "frac_reward_zero_std": 0.234375,
200
+ "grad_norm": 398.9694519042969,
201
+ "learning_rate": 1.4979992127476638e-05,
202
+ "loss": 2.0794,
203
+ "num_tokens": 3419503.0,
204
+ "reward": 0.4641575515270233,
205
+ "reward_std": 0.20721769332885742,
206
+ "rewards/<lambda>/mean": 0.4641575813293457,
207
+ "rewards/<lambda>/std": 0.5078222155570984,
208
+ "sampling/importance_sampling_ratio/max": 2.0,
209
+ "sampling/importance_sampling_ratio/mean": 0.9999407529830933,
210
+ "sampling/importance_sampling_ratio/min": 0.14981676638126373,
211
+ "sampling/sampling_logp_difference/max": 1.8983423709869385,
212
+ "sampling/sampling_logp_difference/mean": 0.012385254725813866,
213
+ "step": 7,
214
+ "step_time": 60.95480399392545
215
+ },
216
+ {
217
+ "adpo/anchor_kl": 0.0,
218
+ "adpo/anchor_updates": 0.0,
219
+ "completions/clipped_ratio": 0.19140625,
220
+ "completions/max_length": 1024.0,
221
+ "completions/max_terminated_length": 1010.0,
222
+ "completions/mean_length": 666.703125,
223
+ "completions/mean_terminated_length": 582.1256103515625,
224
+ "completions/min_length": 216.0,
225
+ "completions/min_terminated_length": 216.0,
226
+ "entropy": 0.1797881731763482,
227
+ "epoch": 0.3333333333333333,
228
+ "frac_reward_zero_std": 0.25,
229
+ "grad_norm": 347.8717346191406,
230
+ "learning_rate": 1.4920075260563328e-05,
231
+ "loss": 2.0794,
232
+ "num_tokens": 3824727.0,
233
+ "reward": 0.6354062557220459,
234
+ "reward_std": 0.261150598526001,
235
+ "rewards/<lambda>/mean": 0.6354062557220459,
236
+ "rewards/<lambda>/std": 0.4872872829437256,
237
+ "sampling/importance_sampling_ratio/max": 2.0,
238
+ "sampling/importance_sampling_ratio/mean": 0.999925971031189,
239
+ "sampling/importance_sampling_ratio/min": 0.09418575465679169,
240
+ "sampling/sampling_logp_difference/max": 2.3624863624572754,
241
+ "sampling/sampling_logp_difference/mean": 0.012666234746575356,
242
+ "step": 8,
243
+ "step_time": 53.79076077602804
244
+ },
245
+ {
246
+ "adpo/anchor_kl": 0.0,
247
+ "adpo/anchor_updates": 0.0,
248
+ "completions/clipped_ratio": 0.2890625,
249
+ "completions/max_length": 1024.0,
250
+ "completions/max_terminated_length": 1019.0,
251
+ "completions/mean_length": 731.34375,
252
+ "completions/mean_terminated_length": 613.4725341796875,
253
+ "completions/min_length": 212.0,
254
+ "completions/min_terminated_length": 212.0,
255
+ "entropy": 0.1969222454354167,
256
+ "epoch": 0.375,
257
+ "frac_reward_zero_std": 0.15625,
258
+ "grad_norm": 388.7760925292969,
259
+ "learning_rate": 1.4820569081669455e-05,
260
+ "loss": 2.0794,
261
+ "num_tokens": 4282671.0,
262
+ "reward": 0.5200888514518738,
263
+ "reward_std": 0.28363779187202454,
264
+ "rewards/<lambda>/mean": 0.5200888514518738,
265
+ "rewards/<lambda>/std": 0.5074349045753479,
266
+ "sampling/importance_sampling_ratio/max": 2.0,
267
+ "sampling/importance_sampling_ratio/mean": 0.9999901652336121,
268
+ "sampling/importance_sampling_ratio/min": 0.0017239645821973681,
269
+ "sampling/sampling_logp_difference/max": 6.363128662109375,
270
+ "sampling/sampling_logp_difference/mean": 0.013784973882138729,
271
+ "step": 9,
272
+ "step_time": 58.84364004805684
273
+ },
274
+ {
275
+ "adpo/anchor_kl": 0.0,
276
+ "adpo/anchor_updates": 0.0,
277
+ "completions/clipped_ratio": 0.3125,
278
+ "completions/max_length": 1024.0,
279
+ "completions/max_terminated_length": 1023.0,
280
+ "completions/mean_length": 669.052734375,
281
+ "completions/mean_terminated_length": 507.71307373046875,
282
+ "completions/min_length": 139.0,
283
+ "completions/min_terminated_length": 139.0,
284
+ "entropy": 0.18559871893376112,
285
+ "epoch": 0.4166666666666667,
286
+ "frac_reward_zero_std": 0.359375,
287
+ "grad_norm": 385.1412658691406,
288
+ "learning_rate": 1.4682004499313044e-05,
289
+ "loss": 2.0794,
290
+ "num_tokens": 4716146.0,
291
+ "reward": 0.5486379861831665,
292
+ "reward_std": 0.1545945405960083,
293
+ "rewards/<lambda>/mean": 0.5486379861831665,
294
+ "rewards/<lambda>/std": 0.5042667388916016,
295
+ "sampling/importance_sampling_ratio/max": 2.0,
296
+ "sampling/importance_sampling_ratio/mean": 1.000030517578125,
297
+ "sampling/importance_sampling_ratio/min": 0.29496005177497864,
298
+ "sampling/sampling_logp_difference/max": 1.2209153175354004,
299
+ "sampling/sampling_logp_difference/mean": 0.013065028935670853,
300
+ "step": 10,
301
+ "step_time": 60.04833007417619
302
+ },
303
+ {
304
+ "adpo/anchor_kl": 0.0,
305
+ "adpo/anchor_updates": 0.0,
306
+ "adpo/mean_tau": 0.5050484538078308,
307
+ "completions/clipped_ratio": 0.373046875,
308
+ "completions/max_length": 1024.0,
309
+ "completions/max_terminated_length": 1022.0,
310
+ "completions/mean_length": 733.859375,
311
+ "completions/mean_terminated_length": 561.22119140625,
312
+ "completions/min_length": 164.0,
313
+ "completions/min_terminated_length": 164.0,
314
+ "entropy": 0.17730128951370716,
315
+ "epoch": 0.4583333333333333,
316
+ "frac_reward_zero_std": 0.234375,
317
+ "grad_norm": 369.3194885253906,
318
+ "learning_rate": 1.450512081549411e-05,
319
+ "loss": 2.0794,
320
+ "num_tokens": 5176258.0,
321
+ "reward": 0.5165244936943054,
322
+ "reward_std": 0.22097265720367432,
323
+ "rewards/<lambda>/mean": 0.5165244936943054,
324
+ "rewards/<lambda>/std": 0.5072101950645447,
325
+ "sampling/importance_sampling_ratio/max": 2.0,
326
+ "sampling/importance_sampling_ratio/mean": 0.9999682903289795,
327
+ "sampling/importance_sampling_ratio/min": 0.31634521484375,
328
+ "sampling/sampling_logp_difference/max": 1.3067893981933594,
329
+ "sampling/sampling_logp_difference/mean": 0.012410955503582954,
330
+ "step": 11,
331
+ "step_time": 56.4488173853606
332
+ },
333
+ {
334
+ "adpo/anchor_kl": 0.0,
335
+ "adpo/anchor_updates": 0.0,
336
+ "completions/clipped_ratio": 0.3671875,
337
+ "completions/max_length": 1024.0,
338
+ "completions/max_terminated_length": 1019.0,
339
+ "completions/mean_length": 781.01953125,
340
+ "completions/mean_terminated_length": 640.0308837890625,
341
+ "completions/min_length": 263.0,
342
+ "completions/min_terminated_length": 263.0,
343
+ "entropy": 0.15673367399722338,
344
+ "epoch": 0.5,
345
+ "frac_reward_zero_std": 0.21875,
346
+ "grad_norm": 333.9354248046875,
347
+ "learning_rate": 1.4290861781198601e-05,
348
+ "loss": 2.0794,
349
+ "num_tokens": 5639724.0,
350
+ "reward": 0.5360932946205139,
351
+ "reward_std": 0.24962013959884644,
352
+ "rewards/<lambda>/mean": 0.5360932946205139,
353
+ "rewards/<lambda>/std": 0.5061600804328918,
354
+ "sampling/importance_sampling_ratio/max": 2.0,
355
+ "sampling/importance_sampling_ratio/mean": 0.9998836517333984,
356
+ "sampling/importance_sampling_ratio/min": 0.20600874722003937,
357
+ "sampling/sampling_logp_difference/max": 1.8993070125579834,
358
+ "sampling/sampling_logp_difference/mean": 0.010962596163153648,
359
+ "step": 12,
360
+ "step_time": 61.27043263055384
361
+ },
362
+ {
363
+ "adpo/anchor_kl": 0.0,
364
+ "adpo/anchor_updates": 0.0,
365
+ "completions/clipped_ratio": 0.4609375,
366
+ "completions/max_length": 1024.0,
367
+ "completions/max_terminated_length": 1014.0,
368
+ "completions/mean_length": 816.0703125,
369
+ "completions/mean_terminated_length": 651.5507202148438,
370
+ "completions/min_length": 273.0,
371
+ "completions/min_terminated_length": 273.0,
372
+ "entropy": 0.1673039924353361,
373
+ "epoch": 0.5416666666666666,
374
+ "frac_reward_zero_std": 0.1875,
375
+ "grad_norm": 335.7733459472656,
376
+ "learning_rate": 1.4040370561078558e-05,
377
+ "loss": 2.0794,
378
+ "num_tokens": 6133624.0,
379
+ "reward": 0.4917134642601013,
380
+ "reward_std": 0.2636360228061676,
381
+ "rewards/<lambda>/mean": 0.49171343445777893,
382
+ "rewards/<lambda>/std": 0.5088064074516296,
383
+ "sampling/importance_sampling_ratio/max": 2.0,
384
+ "sampling/importance_sampling_ratio/mean": 0.9998838901519775,
385
+ "sampling/importance_sampling_ratio/min": 0.04999053478240967,
386
+ "sampling/sampling_logp_difference/max": 2.9959216117858887,
387
+ "sampling/sampling_logp_difference/mean": 0.011442139744758606,
388
+ "step": 13,
389
+ "step_time": 72.87974170222878
390
+ },
391
+ {
392
+ "adpo/anchor_kl": 0.0,
393
+ "adpo/anchor_updates": 0.0,
394
+ "completions/clipped_ratio": 0.43359375,
395
+ "completions/max_length": 1024.0,
396
+ "completions/max_terminated_length": 1023.0,
397
+ "completions/mean_length": 822.033203125,
398
+ "completions/mean_terminated_length": 667.4241333007812,
399
+ "completions/min_length": 257.0,
400
+ "completions/min_terminated_length": 257.0,
401
+ "entropy": 0.16570829413831234,
402
+ "epoch": 0.5833333333333334,
403
+ "frac_reward_zero_std": 0.1875,
404
+ "grad_norm": 353.3135986328125,
405
+ "learning_rate": 1.3754983634174084e-05,
406
+ "loss": 2.0794,
407
+ "num_tokens": 6617785.0,
408
+ "reward": 0.4657760262489319,
409
+ "reward_std": 0.276570200920105,
410
+ "rewards/<lambda>/mean": 0.4657760262489319,
411
+ "rewards/<lambda>/std": 0.506278395652771,
412
+ "sampling/importance_sampling_ratio/max": 2.0,
413
+ "sampling/importance_sampling_ratio/mean": 1.000022053718567,
414
+ "sampling/importance_sampling_ratio/min": 0.13505807518959045,
415
+ "sampling/sampling_logp_difference/max": 2.0020503997802734,
416
+ "sampling/sampling_logp_difference/mean": 0.011564884334802628,
417
+ "step": 14,
418
+ "step_time": 63.08116350695491
419
+ },
420
+ {
421
+ "adpo/anchor_kl": 0.0,
422
+ "adpo/anchor_updates": 0.0,
423
+ "completions/clipped_ratio": 0.4296875,
424
+ "completions/max_length": 1024.0,
425
+ "completions/max_terminated_length": 1022.0,
426
+ "completions/mean_length": 818.72265625,
427
+ "completions/mean_terminated_length": 667.7876586914062,
428
+ "completions/min_length": 201.0,
429
+ "completions/min_terminated_length": 201.0,
430
+ "entropy": 0.1621287539601326,
431
+ "epoch": 0.625,
432
+ "frac_reward_zero_std": 0.171875,
433
+ "grad_norm": 311.2161560058594,
434
+ "learning_rate": 1.3436223663219406e-05,
435
+ "loss": 2.0794,
436
+ "num_tokens": 7104427.0,
437
+ "reward": 0.500846266746521,
438
+ "reward_std": 0.31220686435699463,
439
+ "rewards/<lambda>/mean": 0.500846266746521,
440
+ "rewards/<lambda>/std": 0.5075346231460571,
441
+ "sampling/importance_sampling_ratio/max": 2.0,
442
+ "sampling/importance_sampling_ratio/mean": 1.0000503063201904,
443
+ "sampling/importance_sampling_ratio/min": 0.02315940335392952,
444
+ "sampling/sampling_logp_difference/max": 3.7653543949127197,
445
+ "sampling/sampling_logp_difference/mean": 0.011532379314303398,
446
+ "step": 15,
447
+ "step_time": 67.25278560817242
448
+ },
449
+ {
450
+ "adpo/anchor_kl": 0.0,
451
+ "adpo/anchor_updates": 0.0,
452
+ "completions/clipped_ratio": 0.23046875,
453
+ "completions/max_length": 1024.0,
454
+ "completions/max_terminated_length": 1021.0,
455
+ "completions/mean_length": 706.595703125,
456
+ "completions/mean_terminated_length": 618.5532836914062,
457
+ "completions/min_length": 158.0,
458
+ "completions/min_terminated_length": 158.0,
459
+ "entropy": 0.15026380773633718,
460
+ "epoch": 0.6666666666666666,
461
+ "frac_reward_zero_std": 0.25,
462
+ "grad_norm": 335.7610168457031,
463
+ "learning_rate": 1.3085791370578364e-05,
464
+ "loss": 2.0794,
465
+ "num_tokens": 7529476.0,
466
+ "reward": 0.6287662982940674,
467
+ "reward_std": 0.29228705167770386,
468
+ "rewards/<lambda>/mean": 0.6287662982940674,
469
+ "rewards/<lambda>/std": 0.4878396987915039,
470
+ "sampling/importance_sampling_ratio/max": 2.0,
471
+ "sampling/importance_sampling_ratio/mean": 1.000042200088501,
472
+ "sampling/importance_sampling_ratio/min": 0.007269550114870071,
473
+ "sampling/sampling_logp_difference/max": 4.924060821533203,
474
+ "sampling/sampling_logp_difference/mean": 0.011203760281205177,
475
+ "step": 16,
476
+ "step_time": 71.91690382920206
477
+ },
478
+ {
479
+ "adpo/anchor_kl": 0.0,
480
+ "adpo/anchor_updates": 0.0,
481
+ "completions/clipped_ratio": 0.12109375,
482
+ "completions/max_length": 1024.0,
483
+ "completions/max_terminated_length": 1024.0,
484
+ "completions/mean_length": 580.720703125,
485
+ "completions/mean_terminated_length": 519.6466674804688,
486
+ "completions/min_length": 200.0,
487
+ "completions/min_terminated_length": 200.0,
488
+ "entropy": 0.14060428459197283,
489
+ "epoch": 0.7083333333333334,
490
+ "frac_reward_zero_std": 0.484375,
491
+ "grad_norm": 409.5798034667969,
492
+ "learning_rate": 1.2705556464154755e-05,
493
+ "loss": 2.0794,
494
+ "num_tokens": 7879509.0,
495
+ "reward": 0.802566409111023,
496
+ "reward_std": 0.2324337363243103,
497
+ "rewards/<lambda>/mean": 0.802566409111023,
498
+ "rewards/<lambda>/std": 0.40114688873291016,
499
+ "sampling/importance_sampling_ratio/max": 2.0,
500
+ "sampling/importance_sampling_ratio/mean": 0.9999762177467346,
501
+ "sampling/importance_sampling_ratio/min": 0.11966455727815628,
502
+ "sampling/sampling_logp_difference/max": 2.1230628490448,
503
+ "sampling/sampling_logp_difference/mean": 0.010822593234479427,
504
+ "step": 17,
505
+ "step_time": 49.497211031615734
506
+ },
507
+ {
508
+ "adpo/anchor_kl": 0.0,
509
+ "adpo/anchor_updates": 0.0,
510
+ "completions/clipped_ratio": 0.08984375,
511
+ "completions/max_length": 1024.0,
512
+ "completions/max_terminated_length": 1011.0,
513
+ "completions/mean_length": 556.521484375,
514
+ "completions/mean_terminated_length": 510.37554931640625,
515
+ "completions/min_length": 134.0,
516
+ "completions/min_terminated_length": 134.0,
517
+ "entropy": 0.13123633759096265,
518
+ "epoch": 0.75,
519
+ "frac_reward_zero_std": 0.421875,
520
+ "grad_norm": 315.4649963378906,
521
+ "learning_rate": 1.2297547661691685e-05,
522
+ "loss": 2.0794,
523
+ "num_tokens": 8214992.0,
524
+ "reward": 0.7499439120292664,
525
+ "reward_std": 0.232870951294899,
526
+ "rewards/<lambda>/mean": 0.7499439120292664,
527
+ "rewards/<lambda>/std": 0.43581071496009827,
528
+ "sampling/importance_sampling_ratio/max": 2.0,
529
+ "sampling/importance_sampling_ratio/mean": 0.9999969005584717,
530
+ "sampling/importance_sampling_ratio/min": 0.03257989510893822,
531
+ "sampling/sampling_logp_difference/max": 3.4240598678588867,
532
+ "sampling/sampling_logp_difference/mean": 0.010380487889051437,
533
+ "step": 18,
534
+ "step_time": 49.246268317103386
535
+ },
536
+ {
537
+ "adpo/anchor_kl": 0.0,
538
+ "adpo/anchor_updates": 0.0,
539
+ "completions/clipped_ratio": 0.0546875,
540
+ "completions/max_length": 1024.0,
541
+ "completions/max_terminated_length": 1013.0,
542
+ "completions/mean_length": 471.37890625,
543
+ "completions/mean_terminated_length": 439.7561950683594,
544
+ "completions/min_length": 28.0,
545
+ "completions/min_terminated_length": 28.0,
546
+ "entropy": 0.1348801627755165,
547
+ "epoch": 0.7916666666666666,
548
+ "frac_reward_zero_std": 0.40625,
549
+ "grad_norm": 296.21246337890625,
550
+ "learning_rate": 1.1863941866684647e-05,
551
+ "loss": 2.0794,
552
+ "num_tokens": 8523618.0,
553
+ "reward": 0.759402871131897,
554
+ "reward_std": 0.2495279312133789,
555
+ "rewards/<lambda>/mean": 0.759402871131897,
556
+ "rewards/<lambda>/std": 0.43060389161109924,
557
+ "sampling/importance_sampling_ratio/max": 2.0,
558
+ "sampling/importance_sampling_ratio/mean": 0.9998505115509033,
559
+ "sampling/importance_sampling_ratio/min": 0.02500835806131363,
560
+ "sampling/sampling_logp_difference/max": 3.6885452270507812,
561
+ "sampling/sampling_logp_difference/mean": 0.011401316151022911,
562
+ "step": 19,
563
+ "step_time": 54.79210297949612
564
+ },
565
+ {
566
+ "adpo/anchor_kl": 0.0,
567
+ "adpo/anchor_updates": 0.0,
568
+ "completions/clipped_ratio": 0.09375,
569
+ "completions/max_length": 1024.0,
570
+ "completions/max_terminated_length": 1005.0,
571
+ "completions/mean_length": 419.9609375,
572
+ "completions/mean_terminated_length": 359.6314697265625,
573
+ "completions/min_length": 35.0,
574
+ "completions/min_terminated_length": 35.0,
575
+ "entropy": 0.13178576761856675,
576
+ "epoch": 0.8333333333333334,
577
+ "frac_reward_zero_std": 0.53125,
578
+ "grad_norm": 377.0572509765625,
579
+ "learning_rate": 1.1407052553659478e-05,
580
+ "loss": 2.0794,
581
+ "num_tokens": 8805422.0,
582
+ "reward": 0.7694751620292664,
583
+ "reward_std": 0.15304654836654663,
584
+ "rewards/<lambda>/mean": 0.7694751620292664,
585
+ "rewards/<lambda>/std": 0.4239923357963562,
586
+ "sampling/importance_sampling_ratio/max": 2.0,
587
+ "sampling/importance_sampling_ratio/mean": 1.0001556873321533,
588
+ "sampling/importance_sampling_ratio/min": 0.014293007552623749,
589
+ "sampling/sampling_logp_difference/max": 4.247984886169434,
590
+ "sampling/sampling_logp_difference/mean": 0.011536775156855583,
591
+ "step": 20,
592
+ "step_time": 57.35540423169732
593
+ },
594
+ {
595
+ "adpo/anchor_kl": 0.0,
596
+ "adpo/anchor_updates": 0.0,
597
+ "adpo/mean_tau": 0.3482546806335449,
598
+ "completions/clipped_ratio": 0.078125,
599
+ "completions/max_length": 1024.0,
600
+ "completions/max_terminated_length": 1009.0,
601
+ "completions/mean_length": 497.177734375,
602
+ "completions/mean_terminated_length": 452.5317687988281,
603
+ "completions/min_length": 33.0,
604
+ "completions/min_terminated_length": 33.0,
605
+ "entropy": 0.12525559728965163,
606
+ "epoch": 0.875,
607
+ "frac_reward_zero_std": 0.4375,
608
+ "grad_norm": 393.7420959472656,
609
+ "learning_rate": 1.0929317424784789e-05,
610
+ "loss": 2.0794,
611
+ "num_tokens": 9134265.0,
612
+ "reward": 0.7311800122261047,
613
+ "reward_std": 0.21489663422107697,
614
+ "rewards/<lambda>/mean": 0.7311800122261047,
615
+ "rewards/<lambda>/std": 0.4474351406097412,
616
+ "sampling/importance_sampling_ratio/max": 2.0,
617
+ "sampling/importance_sampling_ratio/mean": 0.9998689889907837,
618
+ "sampling/importance_sampling_ratio/min": 0.09945565462112427,
619
+ "sampling/sampling_logp_difference/max": 2.3080434799194336,
620
+ "sampling/sampling_logp_difference/mean": 0.011143107898533344,
621
+ "step": 21,
622
+ "step_time": 54.157339030876756
623
+ },
624
+ {
625
+ "adpo/anchor_kl": 0.0,
626
+ "adpo/anchor_updates": 0.0,
627
+ "completions/clipped_ratio": 0.080078125,
628
+ "completions/max_length": 1024.0,
629
+ "completions/max_terminated_length": 951.0,
630
+ "completions/mean_length": 462.52734375,
631
+ "completions/mean_terminated_length": 413.6518249511719,
632
+ "completions/min_length": 62.0,
633
+ "completions/min_terminated_length": 62.0,
634
+ "entropy": 0.11509952088817954,
635
+ "epoch": 0.9166666666666666,
636
+ "frac_reward_zero_std": 0.40625,
637
+ "grad_norm": 397.80389404296875,
638
+ "learning_rate": 1.043328540367617e-05,
639
+ "loss": 2.0794,
640
+ "num_tokens": 9435239.0,
641
+ "reward": 0.7673884630203247,
642
+ "reward_std": 0.2196895182132721,
643
+ "rewards/<lambda>/mean": 0.7673884630203247,
644
+ "rewards/<lambda>/std": 0.42547982931137085,
645
+ "sampling/importance_sampling_ratio/max": 2.0,
646
+ "sampling/importance_sampling_ratio/mean": 1.0001059770584106,
647
+ "sampling/importance_sampling_ratio/min": 0.06498447805643082,
648
+ "sampling/sampling_logp_difference/max": 2.7336068153381348,
649
+ "sampling/sampling_logp_difference/mean": 0.010541493073105812,
650
+ "step": 22,
651
+ "step_time": 56.50692548789084
652
+ },
653
+ {
654
+ "adpo/anchor_kl": 0.0,
655
+ "adpo/anchor_updates": 0.0,
656
+ "completions/clipped_ratio": 0.33203125,
657
+ "completions/max_length": 1024.0,
658
+ "completions/max_terminated_length": 1022.0,
659
+ "completions/mean_length": 682.1328125,
660
+ "completions/mean_terminated_length": 512.1988525390625,
661
+ "completions/min_length": 40.0,
662
+ "completions/min_terminated_length": 40.0,
663
+ "entropy": 0.12456677202135324,
664
+ "epoch": 0.9583333333333334,
665
+ "frac_reward_zero_std": 0.1875,
666
+ "grad_norm": 377.5677795410156,
667
+ "learning_rate": 9.921603035785846e-06,
668
+ "loss": 2.0794,
669
+ "num_tokens": 9853347.0,
670
+ "reward": 0.47955191135406494,
671
+ "reward_std": 0.2642451822757721,
672
+ "rewards/<lambda>/mean": 0.47955194115638733,
673
+ "rewards/<lambda>/std": 0.5069260001182556,
674
+ "sampling/importance_sampling_ratio/max": 2.0,
675
+ "sampling/importance_sampling_ratio/mean": 1.0001275539398193,
676
+ "sampling/importance_sampling_ratio/min": 6.08826042025612e-07,
677
+ "sampling/sampling_logp_difference/max": 14.31173324584961,
678
+ "sampling/sampling_logp_difference/mean": 0.010669536888599396,
679
+ "step": 23,
680
+ "step_time": 55.66705738380551
681
+ },
682
+ {
683
+ "adpo/anchor_kl": 0.0,
684
+ "adpo/anchor_updates": 0.0,
685
+ "completions/clipped_ratio": 0.4765625,
686
+ "completions/max_length": 1024.0,
687
+ "completions/max_terminated_length": 1024.0,
688
+ "completions/mean_length": 805.478515625,
689
+ "completions/mean_terminated_length": 609.3992309570312,
690
+ "completions/min_length": 191.0,
691
+ "completions/min_terminated_length": 191.0,
692
+ "entropy": 0.13493244536221027,
693
+ "epoch": 1.0,
694
+ "frac_reward_zero_std": 0.09375,
695
+ "grad_norm": 376.5946960449219,
696
+ "learning_rate": 9.397000367937605e-06,
697
+ "loss": 2.0794,
698
+ "num_tokens": 10351232.0,
699
+ "reward": 0.32667607069015503,
700
+ "reward_std": 0.26685771346092224,
701
+ "rewards/<lambda>/mean": 0.3266761004924774,
702
+ "rewards/<lambda>/std": 0.4794226586818695,
703
+ "sampling/importance_sampling_ratio/max": 2.0,
704
+ "sampling/importance_sampling_ratio/mean": 0.9999583959579468,
705
+ "sampling/importance_sampling_ratio/min": 0.15206119418144226,
706
+ "sampling/sampling_logp_difference/max": 1.883472204208374,
707
+ "sampling/sampling_logp_difference/mean": 0.010822313837707043,
708
+ "step": 24,
709
+ "step_time": 64.30626124888659
710
+ },
711
+ {
712
+ "adpo/anchor_kl": 0.0,
713
+ "adpo/anchor_updates": 0.0,
714
+ "completions/clipped_ratio": 0.439453125,
715
+ "completions/max_length": 1024.0,
716
+ "completions/max_terminated_length": 1009.0,
717
+ "completions/mean_length": 705.517578125,
718
+ "completions/mean_terminated_length": 455.83624267578125,
719
+ "completions/min_length": 123.0,
720
+ "completions/min_terminated_length": 123.0,
721
+ "entropy": 0.06874244613572955,
722
+ "epoch": 1.0416666666666667,
723
+ "frac_reward_zero_std": 0.046875,
724
+ "grad_norm": 322.31121826171875,
725
+ "learning_rate": 8.862276382345772e-06,
726
+ "loss": 2.0794,
727
+ "num_tokens": 10768113.0,
728
+ "reward": 0.45595476031303406,
729
+ "reward_std": 0.3999968469142914,
730
+ "rewards/<lambda>/mean": 0.45595476031303406,
731
+ "rewards/<lambda>/std": 0.5178707242012024,
732
+ "sampling/importance_sampling_ratio/max": 2.0,
733
+ "sampling/importance_sampling_ratio/mean": 0.9998959302902222,
734
+ "sampling/importance_sampling_ratio/min": 0.028519853949546814,
735
+ "sampling/sampling_logp_difference/max": 3.557154893875122,
736
+ "sampling/sampling_logp_difference/mean": 0.00642294529825449,
737
+ "step": 25,
738
+ "step_time": 59.545914109796286
739
+ },
740
+ {
741
+ "adpo/anchor_kl": 0.0,
742
+ "adpo/anchor_updates": 0.0,
743
+ "completions/clipped_ratio": 0.17578125,
744
+ "completions/max_length": 1024.0,
745
+ "completions/max_terminated_length": 1016.0,
746
+ "completions/mean_length": 551.46875,
747
+ "completions/mean_terminated_length": 450.69195556640625,
748
+ "completions/min_length": 146.0,
749
+ "completions/min_terminated_length": 146.0,
750
+ "entropy": 0.09305409202352166,
751
+ "epoch": 1.0833333333333333,
752
+ "frac_reward_zero_std": 0.1875,
753
+ "grad_norm": 386.7756042480469,
754
+ "learning_rate": 8.32028406283406e-06,
755
+ "loss": 2.0794,
756
+ "num_tokens": 11105609.0,
757
+ "reward": 0.7143889665603638,
758
+ "reward_std": 0.3812097907066345,
759
+ "rewards/<lambda>/mean": 0.7143889665603638,
760
+ "rewards/<lambda>/std": 0.45940032601356506,
761
+ "sampling/importance_sampling_ratio/max": 2.0,
762
+ "sampling/importance_sampling_ratio/mean": 0.9999741315841675,
763
+ "sampling/importance_sampling_ratio/min": 0.005678437650203705,
764
+ "sampling/sampling_logp_difference/max": 5.171079158782959,
765
+ "sampling/sampling_logp_difference/mean": 0.008768249303102493,
766
+ "step": 26,
767
+ "step_time": 50.71110679022968
768
+ },
769
+ {
770
+ "adpo/anchor_kl": 0.0,
771
+ "adpo/anchor_updates": 0.0,
772
+ "completions/clipped_ratio": 0.0625,
773
+ "completions/max_length": 1024.0,
774
+ "completions/max_terminated_length": 1021.0,
775
+ "completions/mean_length": 517.19140625,
776
+ "completions/mean_terminated_length": 483.4042053222656,
777
+ "completions/min_length": 155.0,
778
+ "completions/min_terminated_length": 155.0,
779
+ "entropy": 0.10468809772282839,
780
+ "epoch": 1.125,
781
+ "frac_reward_zero_std": 0.1875,
782
+ "grad_norm": 404.017578125,
783
+ "learning_rate": 7.773915172932443e-06,
784
+ "loss": 2.0794,
785
+ "num_tokens": 11433715.0,
786
+ "reward": 0.7240350246429443,
787
+ "reward_std": 0.3405328392982483,
788
+ "rewards/<lambda>/mean": 0.7240350246429443,
789
+ "rewards/<lambda>/std": 0.4502919316291809,
790
+ "sampling/importance_sampling_ratio/max": 2.0,
791
+ "sampling/importance_sampling_ratio/mean": 0.9999520778656006,
792
+ "sampling/importance_sampling_ratio/min": 0.003656858578324318,
793
+ "sampling/sampling_logp_difference/max": 5.611150741577148,
794
+ "sampling/sampling_logp_difference/mean": 0.01007787138223648,
795
+ "step": 27,
796
+ "step_time": 54.12670726701617
797
+ },
798
+ {
799
+ "adpo/anchor_kl": 0.0,
800
+ "adpo/anchor_updates": 0.0,
801
+ "completions/clipped_ratio": 0.02734375,
802
+ "completions/max_length": 1024.0,
803
+ "completions/max_terminated_length": 1016.0,
804
+ "completions/mean_length": 423.765625,
805
+ "completions/mean_terminated_length": 406.89154052734375,
806
+ "completions/min_length": 134.0,
807
+ "completions/min_terminated_length": 134.0,
808
+ "entropy": 0.09546938771381974,
809
+ "epoch": 1.1666666666666667,
810
+ "frac_reward_zero_std": 0.28125,
811
+ "grad_norm": 394.0153503417969,
812
+ "learning_rate": 7.226084827067558e-06,
813
+ "loss": 2.0794,
814
+ "num_tokens": 11705723.0,
815
+ "reward": 0.7384281158447266,
816
+ "reward_std": 0.3273390829563141,
817
+ "rewards/<lambda>/mean": 0.7384281158447266,
818
+ "rewards/<lambda>/std": 0.441993772983551,
819
+ "sampling/importance_sampling_ratio/max": 2.0,
820
+ "sampling/importance_sampling_ratio/mean": 0.9999547004699707,
821
+ "sampling/importance_sampling_ratio/min": 0.10396554321050644,
822
+ "sampling/sampling_logp_difference/max": 2.26369571685791,
823
+ "sampling/sampling_logp_difference/mean": 0.009469419717788696,
824
+ "step": 28,
825
+ "step_time": 47.505339877679944
826
+ },
827
+ {
828
+ "adpo/anchor_kl": 0.0,
829
+ "adpo/anchor_updates": 0.0,
830
+ "completions/clipped_ratio": 0.052734375,
831
+ "completions/max_length": 1024.0,
832
+ "completions/max_terminated_length": 1007.0,
833
+ "completions/mean_length": 475.0234375,
834
+ "completions/mean_terminated_length": 444.4618835449219,
835
+ "completions/min_length": 178.0,
836
+ "completions/min_terminated_length": 178.0,
837
+ "entropy": 0.09483881155028939,
838
+ "epoch": 1.2083333333333333,
839
+ "frac_reward_zero_std": 0.203125,
840
+ "grad_norm": 357.1606750488281,
841
+ "learning_rate": 6.679715937165944e-06,
842
+ "loss": 2.0794,
843
+ "num_tokens": 12005911.0,
844
+ "reward": 0.6610323190689087,
845
+ "reward_std": 0.3293628990650177,
846
+ "rewards/<lambda>/mean": 0.6610323190689087,
847
+ "rewards/<lambda>/std": 0.47705599665641785,
848
+ "sampling/importance_sampling_ratio/max": 2.0,
849
+ "sampling/importance_sampling_ratio/mean": 1.0000580549240112,
850
+ "sampling/importance_sampling_ratio/min": 0.07158884406089783,
851
+ "sampling/sampling_logp_difference/max": 2.6368160247802734,
852
+ "sampling/sampling_logp_difference/mean": 0.008950229734182358,
853
+ "step": 29,
854
+ "step_time": 50.50793307647109
855
+ },
856
+ {
857
+ "adpo/anchor_kl": 0.0,
858
+ "adpo/anchor_updates": 0.0,
859
+ "completions/clipped_ratio": 0.0546875,
860
+ "completions/max_length": 1024.0,
861
+ "completions/max_terminated_length": 989.0,
862
+ "completions/mean_length": 497.115234375,
863
+ "completions/mean_terminated_length": 467.15081787109375,
864
+ "completions/min_length": 192.0,
865
+ "completions/min_terminated_length": 192.0,
866
+ "entropy": 0.09122372791171074,
867
+ "epoch": 1.25,
868
+ "frac_reward_zero_std": 0.296875,
869
+ "grad_norm": 382.6507568359375,
870
+ "learning_rate": 6.137723617654227e-06,
871
+ "loss": 2.0794,
872
+ "num_tokens": 12320426.0,
873
+ "reward": 0.715755820274353,
874
+ "reward_std": 0.29924681782722473,
875
+ "rewards/<lambda>/mean": 0.7157557606697083,
876
+ "rewards/<lambda>/std": 0.45485544204711914,
877
+ "sampling/importance_sampling_ratio/max": 2.0,
878
+ "sampling/importance_sampling_ratio/mean": 0.9999900460243225,
879
+ "sampling/importance_sampling_ratio/min": 0.07345468550920486,
880
+ "sampling/sampling_logp_difference/max": 2.61108660697937,
881
+ "sampling/sampling_logp_difference/mean": 0.008717824704945087,
882
+ "step": 30,
883
+ "step_time": 60.74604574404657
884
+ },
885
+ {
886
+ "adpo/anchor_kl": 0.0,
887
+ "adpo/anchor_updates": 0.0,
888
+ "adpo/mean_tau": 0.26636892557144165,
889
+ "completions/clipped_ratio": 0.099609375,
890
+ "completions/max_length": 1024.0,
891
+ "completions/max_terminated_length": 1017.0,
892
+ "completions/mean_length": 582.728515625,
893
+ "completions/mean_terminated_length": 533.9110717773438,
894
+ "completions/min_length": 202.0,
895
+ "completions/min_terminated_length": 202.0,
896
+ "entropy": 0.12519589345902205,
897
+ "epoch": 1.2916666666666667,
898
+ "frac_reward_zero_std": 0.109375,
899
+ "grad_norm": 320.775634765625,
900
+ "learning_rate": 5.602999632062394e-06,
901
+ "loss": 2.0794,
902
+ "num_tokens": 12680367.0,
903
+ "reward": 0.5093868970870972,
904
+ "reward_std": 0.3450024425983429,
905
+ "rewards/<lambda>/mean": 0.5093868970870972,
906
+ "rewards/<lambda>/std": 0.506705105304718,
907
+ "sampling/importance_sampling_ratio/max": 2.0,
908
+ "sampling/importance_sampling_ratio/mean": 0.999984860420227,
909
+ "sampling/importance_sampling_ratio/min": 0.11154952645301819,
910
+ "sampling/sampling_logp_difference/max": 2.193286657333374,
911
+ "sampling/sampling_logp_difference/mean": 0.010777526535093784,
912
+ "step": 31,
913
+ "step_time": 52.976465525105596
914
+ },
915
+ {
916
+ "adpo/anchor_kl": 0.0,
917
+ "adpo/anchor_updates": 0.0,
918
+ "completions/clipped_ratio": 0.046875,
919
+ "completions/max_length": 1024.0,
920
+ "completions/max_terminated_length": 1022.0,
921
+ "completions/mean_length": 577.169921875,
922
+ "completions/mean_terminated_length": 555.1946411132812,
923
+ "completions/min_length": 246.0,
924
+ "completions/min_terminated_length": 246.0,
925
+ "entropy": 0.11813855404034257,
926
+ "epoch": 1.3333333333333333,
927
+ "frac_reward_zero_std": 0.125,
928
+ "grad_norm": 352.9529113769531,
929
+ "learning_rate": 5.078396964214155e-06,
930
+ "loss": 2.0794,
931
+ "num_tokens": 13039750.0,
932
+ "reward": 0.5626362562179565,
933
+ "reward_std": 0.34678590297698975,
934
+ "rewards/<lambda>/mean": 0.5626362562179565,
935
+ "rewards/<lambda>/std": 0.5023864507675171,
936
+ "sampling/importance_sampling_ratio/max": 2.0,
937
+ "sampling/importance_sampling_ratio/mean": 0.9999432563781738,
938
+ "sampling/importance_sampling_ratio/min": 0.00921670626848936,
939
+ "sampling/sampling_logp_difference/max": 4.686737537384033,
940
+ "sampling/sampling_logp_difference/mean": 0.010188741609454155,
941
+ "step": 32,
942
+ "step_time": 52.343729170039296
943
+ },
944
+ {
945
+ "adpo/anchor_kl": 0.0,
946
+ "adpo/anchor_updates": 0.0,
947
+ "completions/clipped_ratio": 0.14453125,
948
+ "completions/max_length": 1024.0,
949
+ "completions/max_terminated_length": 1023.0,
950
+ "completions/mean_length": 636.32421875,
951
+ "completions/mean_terminated_length": 571.059326171875,
952
+ "completions/min_length": 215.0,
953
+ "completions/min_terminated_length": 215.0,
954
+ "entropy": 0.13468809612095356,
955
+ "epoch": 1.375,
956
+ "frac_reward_zero_std": 0.046875,
957
+ "grad_norm": 514.3561401367188,
958
+ "learning_rate": 4.566714596323831e-06,
959
+ "loss": 2.0794,
960
+ "num_tokens": 13449044.0,
961
+ "reward": 0.44942206144332886,
962
+ "reward_std": 0.33521586656570435,
963
+ "rewards/<lambda>/mean": 0.44942206144332886,
964
+ "rewards/<lambda>/std": 0.5056480765342712,
965
+ "sampling/importance_sampling_ratio/max": 2.0,
966
+ "sampling/importance_sampling_ratio/mean": 0.9999642372131348,
967
+ "sampling/importance_sampling_ratio/min": 0.05576265975832939,
968
+ "sampling/sampling_logp_difference/max": 2.886650800704956,
969
+ "sampling/sampling_logp_difference/mean": 0.011390320956707,
970
+ "step": 33,
971
+ "step_time": 56.10497439280152
972
+ },
973
+ {
974
+ "adpo/anchor_kl": 0.0,
975
+ "adpo/anchor_updates": 0.0,
976
+ "completions/clipped_ratio": 0.15234375,
977
+ "completions/max_length": 1024.0,
978
+ "completions/max_terminated_length": 1024.0,
979
+ "completions/mean_length": 612.12890625,
980
+ "completions/mean_terminated_length": 538.1060180664062,
981
+ "completions/min_length": 189.0,
982
+ "completions/min_terminated_length": 189.0,
983
+ "entropy": 0.12763040326535702,
984
+ "epoch": 1.4166666666666667,
985
+ "frac_reward_zero_std": 0.140625,
986
+ "grad_norm": 547.2996215820312,
987
+ "learning_rate": 4.0706825752152114e-06,
988
+ "loss": 2.0794,
989
+ "num_tokens": 13853374.0,
990
+ "reward": 0.5017865896224976,
991
+ "reward_std": 0.32174602150917053,
992
+ "rewards/<lambda>/mean": 0.5017865300178528,
993
+ "rewards/<lambda>/std": 0.506568968296051,
994
+ "sampling/importance_sampling_ratio/max": 2.0,
995
+ "sampling/importance_sampling_ratio/mean": 0.9999485611915588,
996
+ "sampling/importance_sampling_ratio/min": 0.08450578153133392,
997
+ "sampling/sampling_logp_difference/max": 2.470935344696045,
998
+ "sampling/sampling_logp_difference/mean": 0.01078584510833025,
999
+ "step": 34,
1000
+ "step_time": 57.154975183308125
1001
+ },
1002
+ {
1003
+ "adpo/anchor_kl": 0.0,
1004
+ "adpo/anchor_updates": 0.0,
1005
+ "completions/clipped_ratio": 0.244140625,
1006
+ "completions/max_length": 1024.0,
1007
+ "completions/max_terminated_length": 1024.0,
1008
+ "completions/mean_length": 700.619140625,
1009
+ "completions/mean_terminated_length": 596.16796875,
1010
+ "completions/min_length": 168.0,
1011
+ "completions/min_terminated_length": 168.0,
1012
+ "entropy": 0.1245508803986013,
1013
+ "epoch": 1.4583333333333333,
1014
+ "frac_reward_zero_std": 0.0625,
1015
+ "grad_norm": 382.90838623046875,
1016
+ "learning_rate": 3.592947446340524e-06,
1017
+ "loss": 2.0794,
1018
+ "num_tokens": 14296467.0,
1019
+ "reward": 0.4405647814273834,
1020
+ "reward_std": 0.3587495684623718,
1021
+ "rewards/<lambda>/mean": 0.4405647814273834,
1022
+ "rewards/<lambda>/std": 0.5057622790336609,
1023
+ "sampling/importance_sampling_ratio/max": 2.0,
1024
+ "sampling/importance_sampling_ratio/mean": 1.0000665187835693,
1025
+ "sampling/importance_sampling_ratio/min": 0.011985463090240955,
1026
+ "sampling/sampling_logp_difference/max": 4.424060821533203,
1027
+ "sampling/sampling_logp_difference/mean": 0.010092010721564293,
1028
+ "step": 35,
1029
+ "step_time": 61.529238210991025
1030
+ },
1031
+ {
1032
+ "adpo/anchor_kl": 0.0,
1033
+ "adpo/anchor_updates": 0.0,
1034
+ "completions/clipped_ratio": 0.361328125,
1035
+ "completions/max_length": 1024.0,
1036
+ "completions/max_terminated_length": 1013.0,
1037
+ "completions/mean_length": 794.794921875,
1038
+ "completions/mean_terminated_length": 665.1223754882812,
1039
+ "completions/min_length": 282.0,
1040
+ "completions/min_terminated_length": 282.0,
1041
+ "entropy": 0.11921501066535711,
1042
+ "epoch": 1.5,
1043
+ "frac_reward_zero_std": 0.09375,
1044
+ "grad_norm": 417.32421875,
1045
+ "learning_rate": 3.136058133315355e-06,
1046
+ "loss": 2.0794,
1047
+ "num_tokens": 14766986.0,
1048
+ "reward": 0.35132455825805664,
1049
+ "reward_std": 0.28514936566352844,
1050
+ "rewards/<lambda>/mean": 0.35132455825805664,
1051
+ "rewards/<lambda>/std": 0.48842233419418335,
1052
+ "sampling/importance_sampling_ratio/max": 2.0,
1053
+ "sampling/importance_sampling_ratio/mean": 1.00003981590271,
1054
+ "sampling/importance_sampling_ratio/min": 8.279879182282457e-08,
1055
+ "sampling/sampling_logp_difference/max": 16.306852340698242,
1056
+ "sampling/sampling_logp_difference/mean": 0.009398296475410461,
1057
+ "step": 36,
1058
+ "step_time": 62.635166535153985
1059
+ },
1060
+ {
1061
+ "adpo/anchor_kl": 0.0,
1062
+ "adpo/anchor_updates": 0.0,
1063
+ "completions/clipped_ratio": 0.37890625,
1064
+ "completions/max_length": 1024.0,
1065
+ "completions/max_terminated_length": 1011.0,
1066
+ "completions/mean_length": 783.37890625,
1067
+ "completions/mean_terminated_length": 648.1068725585938,
1068
+ "completions/min_length": 250.0,
1069
+ "completions/min_terminated_length": 250.0,
1070
+ "entropy": 0.11975136725232005,
1071
+ "epoch": 1.5416666666666665,
1072
+ "frac_reward_zero_std": 0.046875,
1073
+ "grad_norm": 386.8076171875,
1074
+ "learning_rate": 2.702452338308317e-06,
1075
+ "loss": 2.0794,
1076
+ "num_tokens": 15244148.0,
1077
+ "reward": 0.3197505474090576,
1078
+ "reward_std": 0.3237205743789673,
1079
+ "rewards/<lambda>/mean": 0.3197505474090576,
1080
+ "rewards/<lambda>/std": 0.4779842495918274,
1081
+ "sampling/importance_sampling_ratio/max": 2.0,
1082
+ "sampling/importance_sampling_ratio/mean": 0.9998108148574829,
1083
+ "sampling/importance_sampling_ratio/min": 0.060896072536706924,
1084
+ "sampling/sampling_logp_difference/max": 2.79858660697937,
1085
+ "sampling/sampling_logp_difference/mean": 0.00945335067808628,
1086
+ "step": 37,
1087
+ "step_time": 68.87949313037097
1088
+ },
1089
+ {
1090
+ "adpo/anchor_kl": 0.0,
1091
+ "adpo/anchor_updates": 0.0,
1092
+ "completions/clipped_ratio": 0.330078125,
1093
+ "completions/max_length": 1024.0,
1094
+ "completions/max_terminated_length": 1023.0,
1095
+ "completions/mean_length": 810.37890625,
1096
+ "completions/mean_terminated_length": 705.1253662109375,
1097
+ "completions/min_length": 320.0,
1098
+ "completions/min_terminated_length": 320.0,
1099
+ "entropy": 0.12328301183879375,
1100
+ "epoch": 1.5833333333333335,
1101
+ "frac_reward_zero_std": 0.046875,
1102
+ "grad_norm": 353.4709777832031,
1103
+ "learning_rate": 2.2944435358452453e-06,
1104
+ "loss": 2.0794,
1105
+ "num_tokens": 15722342.0,
1106
+ "reward": 0.3540261387825012,
1107
+ "reward_std": 0.32076478004455566,
1108
+ "rewards/<lambda>/mean": 0.3540261387825012,
1109
+ "rewards/<lambda>/std": 0.48843950033187866,
1110
+ "sampling/importance_sampling_ratio/max": 2.0,
1111
+ "sampling/importance_sampling_ratio/mean": 0.9999909996986389,
1112
+ "sampling/importance_sampling_ratio/min": 0.0012162798084318638,
1113
+ "sampling/sampling_logp_difference/max": 6.711958408355713,
1114
+ "sampling/sampling_logp_difference/mean": 0.00941707007586956,
1115
+ "step": 38,
1116
+ "step_time": 62.24681276269257
1117
+ },
1118
+ {
1119
+ "adpo/anchor_kl": 0.0,
1120
+ "adpo/anchor_updates": 0.0,
1121
+ "completions/clipped_ratio": 0.341796875,
1122
+ "completions/max_length": 1024.0,
1123
+ "completions/max_terminated_length": 1022.0,
1124
+ "completions/mean_length": 793.314453125,
1125
+ "completions/mean_terminated_length": 674.73291015625,
1126
+ "completions/min_length": 301.0,
1127
+ "completions/min_terminated_length": 301.0,
1128
+ "entropy": 0.1196233481168747,
1129
+ "epoch": 1.625,
1130
+ "frac_reward_zero_std": 0.0625,
1131
+ "grad_norm": 348.0384826660156,
1132
+ "learning_rate": 1.914208629421636e-06,
1133
+ "loss": 2.0794,
1134
+ "num_tokens": 16195975.0,
1135
+ "reward": 0.4032697081565857,
1136
+ "reward_std": 0.35333114862442017,
1137
+ "rewards/<lambda>/mean": 0.4032697081565857,
1138
+ "rewards/<lambda>/std": 0.5001312494277954,
1139
+ "sampling/importance_sampling_ratio/max": 2.0,
1140
+ "sampling/importance_sampling_ratio/mean": 0.9999759197235107,
1141
+ "sampling/importance_sampling_ratio/min": 0.04630627483129501,
1142
+ "sampling/sampling_logp_difference/max": 3.0724778175354004,
1143
+ "sampling/sampling_logp_difference/mean": 0.009164238348603249,
1144
+ "step": 39,
1145
+ "step_time": 67.44319609180093
1146
+ },
1147
+ {
1148
+ "adpo/anchor_kl": 0.0,
1149
+ "adpo/anchor_updates": 0.0,
1150
+ "completions/clipped_ratio": 0.2109375,
1151
+ "completions/max_length": 1024.0,
1152
+ "completions/max_terminated_length": 1023.0,
1153
+ "completions/mean_length": 678.208984375,
1154
+ "completions/mean_terminated_length": 593.591552734375,
1155
+ "completions/min_length": 206.0,
1156
+ "completions/min_terminated_length": 206.0,
1157
+ "entropy": 0.11372271459549665,
1158
+ "epoch": 1.6666666666666665,
1159
+ "frac_reward_zero_std": 0.03125,
1160
+ "grad_norm": 353.4413146972656,
1161
+ "learning_rate": 1.563776336780595e-06,
1162
+ "loss": 2.0794,
1163
+ "num_tokens": 16606490.0,
1164
+ "reward": 0.42268919944763184,
1165
+ "reward_std": 0.3676312565803528,
1166
+ "rewards/<lambda>/mean": 0.42268919944763184,
1167
+ "rewards/<lambda>/std": 0.5036536455154419,
1168
+ "sampling/importance_sampling_ratio/max": 2.0,
1169
+ "sampling/importance_sampling_ratio/mean": 0.9999003410339355,
1170
+ "sampling/importance_sampling_ratio/min": 0.028015917167067528,
1171
+ "sampling/sampling_logp_difference/max": 3.5749824047088623,
1172
+ "sampling/sampling_logp_difference/mean": 0.0093317162245512,
1173
+ "step": 40,
1174
+ "step_time": 71.98558804579079
1175
+ },
1176
+ {
1177
+ "adpo/anchor_kl": 0.0,
1178
+ "adpo/anchor_updates": 0.0,
1179
+ "adpo/mean_tau": 0.3997672200202942,
1180
+ "completions/clipped_ratio": 0.08984375,
1181
+ "completions/max_length": 1024.0,
1182
+ "completions/max_terminated_length": 1015.0,
1183
+ "completions/mean_length": 568.919921875,
1184
+ "completions/mean_terminated_length": 523.9978637695312,
1185
+ "completions/min_length": 197.0,
1186
+ "completions/min_terminated_length": 197.0,
1187
+ "entropy": 0.10392343206331134,
1188
+ "epoch": 1.7083333333333335,
1189
+ "frac_reward_zero_std": 0.078125,
1190
+ "grad_norm": 337.94818115234375,
1191
+ "learning_rate": 1.2450163658259165e-06,
1192
+ "loss": 2.0794,
1193
+ "num_tokens": 16950481.0,
1194
+ "reward": 0.4955139458179474,
1195
+ "reward_std": 0.3674989342689514,
1196
+ "rewards/<lambda>/mean": 0.4955139756202698,
1197
+ "rewards/<lambda>/std": 0.5090046525001526,
1198
+ "sampling/importance_sampling_ratio/max": 2.0,
1199
+ "sampling/importance_sampling_ratio/mean": 0.9999813437461853,
1200
+ "sampling/importance_sampling_ratio/min": 0.0006764940917491913,
1201
+ "sampling/sampling_logp_difference/max": 7.298586845397949,
1202
+ "sampling/sampling_logp_difference/mean": 0.008984290063381195,
1203
+ "step": 41,
1204
+ "step_time": 49.45478970184922
1205
+ },
1206
+ {
1207
+ "adpo/anchor_kl": 0.0,
1208
+ "adpo/anchor_updates": 0.0,
1209
+ "completions/clipped_ratio": 0.052734375,
1210
+ "completions/max_length": 1024.0,
1211
+ "completions/max_terminated_length": 1017.0,
1212
+ "completions/mean_length": 548.701171875,
1213
+ "completions/mean_terminated_length": 522.2412719726562,
1214
+ "completions/min_length": 158.0,
1215
+ "completions/min_terminated_length": 158.0,
1216
+ "entropy": 0.10423749126493931,
1217
+ "epoch": 1.75,
1218
+ "frac_reward_zero_std": 0.0625,
1219
+ "grad_norm": 302.0933532714844,
1220
+ "learning_rate": 9.596294389214422e-07,
1221
+ "loss": 2.0794,
1222
+ "num_tokens": 17281960.0,
1223
+ "reward": 0.5015671253204346,
1224
+ "reward_std": 0.3655853867530823,
1225
+ "rewards/<lambda>/mean": 0.5015671253204346,
1226
+ "rewards/<lambda>/std": 0.5068206191062927,
1227
+ "sampling/importance_sampling_ratio/max": 2.0,
1228
+ "sampling/importance_sampling_ratio/mean": 0.9999548196792603,
1229
+ "sampling/importance_sampling_ratio/min": 0.09954588860273361,
1230
+ "sampling/sampling_logp_difference/max": 2.3071365356445312,
1231
+ "sampling/sampling_logp_difference/mean": 0.00906403362751007,
1232
+ "step": 42,
1233
+ "step_time": 50.07575857639313
1234
+ },
1235
+ {
1236
+ "adpo/anchor_kl": 0.0,
1237
+ "adpo/anchor_updates": 0.0,
1238
+ "completions/clipped_ratio": 0.03515625,
1239
+ "completions/max_length": 1024.0,
1240
+ "completions/max_terminated_length": 1014.0,
1241
+ "completions/mean_length": 457.927734375,
1242
+ "completions/mean_terminated_length": 437.3502197265625,
1243
+ "completions/min_length": 162.0,
1244
+ "completions/min_terminated_length": 162.0,
1245
+ "entropy": 0.10703387530520558,
1246
+ "epoch": 1.7916666666666665,
1247
+ "frac_reward_zero_std": 0.171875,
1248
+ "grad_norm": 313.5877990722656,
1249
+ "learning_rate": 7.091382188014004e-07,
1250
+ "loss": 2.0794,
1251
+ "num_tokens": 17583699.0,
1252
+ "reward": 0.6239477396011353,
1253
+ "reward_std": 0.3349877893924713,
1254
+ "rewards/<lambda>/mean": 0.6239477396011353,
1255
+ "rewards/<lambda>/std": 0.49007174372673035,
1256
+ "sampling/importance_sampling_ratio/max": 2.0,
1257
+ "sampling/importance_sampling_ratio/mean": 1.00001859664917,
1258
+ "sampling/importance_sampling_ratio/min": 0.05251529812812805,
1259
+ "sampling/sampling_logp_difference/max": 2.946650743484497,
1260
+ "sampling/sampling_logp_difference/mean": 0.009622866287827492,
1261
+ "step": 43,
1262
+ "step_time": 55.025071216747165
1263
+ },
1264
+ {
1265
+ "adpo/anchor_kl": 0.0,
1266
+ "adpo/anchor_updates": 0.0,
1267
+ "completions/clipped_ratio": 0.033203125,
1268
+ "completions/max_length": 1024.0,
1269
+ "completions/max_terminated_length": 1008.0,
1270
+ "completions/mean_length": 411.361328125,
1271
+ "completions/mean_terminated_length": 390.32122802734375,
1272
+ "completions/min_length": 160.0,
1273
+ "completions/min_terminated_length": 160.0,
1274
+ "entropy": 0.1085237069055438,
1275
+ "epoch": 1.8333333333333335,
1276
+ "frac_reward_zero_std": 0.21875,
1277
+ "grad_norm": 287.4824523925781,
1278
+ "learning_rate": 4.948791845058906e-07,
1279
+ "loss": 2.0794,
1280
+ "num_tokens": 17861100.0,
1281
+ "reward": 0.6151071786880493,
1282
+ "reward_std": 0.2960226833820343,
1283
+ "rewards/<lambda>/mean": 0.6151071786880493,
1284
+ "rewards/<lambda>/std": 0.49125242233276367,
1285
+ "sampling/importance_sampling_ratio/max": 2.0,
1286
+ "sampling/importance_sampling_ratio/mean": 0.9999807476997375,
1287
+ "sampling/importance_sampling_ratio/min": 0.040546316653490067,
1288
+ "sampling/sampling_logp_difference/max": 3.205310344696045,
1289
+ "sampling/sampling_logp_difference/mean": 0.009774073027074337,
1290
+ "step": 44,
1291
+ "step_time": 56.13218645751476
1292
+ },
1293
+ {
1294
+ "adpo/anchor_kl": 0.0,
1295
+ "adpo/anchor_updates": 0.0,
1296
+ "completions/clipped_ratio": 0.04296875,
1297
+ "completions/max_length": 1024.0,
1298
+ "completions/max_terminated_length": 1005.0,
1299
+ "completions/mean_length": 465.1484375,
1300
+ "completions/mean_terminated_length": 440.05712890625,
1301
+ "completions/min_length": 183.0,
1302
+ "completions/min_terminated_length": 183.0,
1303
+ "entropy": 0.11649307142943144,
1304
+ "epoch": 1.875,
1305
+ "frac_reward_zero_std": 0.21875,
1306
+ "grad_norm": 351.9586486816406,
1307
+ "learning_rate": 3.1799550068695616e-07,
1308
+ "loss": 2.0794,
1309
+ "num_tokens": 18173544.0,
1310
+ "reward": 0.5606979131698608,
1311
+ "reward_std": 0.29378432035446167,
1312
+ "rewards/<lambda>/mean": 0.5606979131698608,
1313
+ "rewards/<lambda>/std": 0.5026035904884338,
1314
+ "sampling/importance_sampling_ratio/max": 2.0,
1315
+ "sampling/importance_sampling_ratio/mean": 0.9999479055404663,
1316
+ "sampling/importance_sampling_ratio/min": 0.04839687421917915,
1317
+ "sampling/sampling_logp_difference/max": 3.028320074081421,
1318
+ "sampling/sampling_logp_difference/mean": 0.010193913243710995,
1319
+ "step": 45,
1320
+ "step_time": 53.353074584156275
1321
+ },
1322
+ {
1323
+ "adpo/anchor_kl": 0.0,
1324
+ "adpo/anchor_updates": 0.0,
1325
+ "completions/clipped_ratio": 0.025390625,
1326
+ "completions/max_length": 1024.0,
1327
+ "completions/max_terminated_length": 996.0,
1328
+ "completions/mean_length": 436.63671875,
1329
+ "completions/mean_terminated_length": 421.3346862792969,
1330
+ "completions/min_length": 188.0,
1331
+ "completions/min_terminated_length": 188.0,
1332
+ "entropy": 0.10410326765850186,
1333
+ "epoch": 1.9166666666666665,
1334
+ "frac_reward_zero_std": 0.203125,
1335
+ "grad_norm": 352.49700927734375,
1336
+ "learning_rate": 1.7943091833054704e-07,
1337
+ "loss": 2.0794,
1338
+ "num_tokens": 18461262.0,
1339
+ "reward": 0.6479157209396362,
1340
+ "reward_std": 0.3350326418876648,
1341
+ "rewards/<lambda>/mean": 0.6479157209396362,
1342
+ "rewards/<lambda>/std": 0.48279955983161926,
1343
+ "sampling/importance_sampling_ratio/max": 2.0,
1344
+ "sampling/importance_sampling_ratio/mean": 0.9999201893806458,
1345
+ "sampling/importance_sampling_ratio/min": 0.00018599774921312928,
1346
+ "sampling/sampling_logp_difference/max": 8.589776039123535,
1347
+ "sampling/sampling_logp_difference/mean": 0.009486394003033638,
1348
+ "step": 46,
1349
+ "step_time": 55.959206158295274
1350
+ },
1351
+ {
1352
+ "adpo/anchor_kl": 0.0,
1353
+ "adpo/anchor_updates": 0.0,
1354
+ "completions/clipped_ratio": 0.306640625,
1355
+ "completions/max_length": 1024.0,
1356
+ "completions/max_terminated_length": 1016.0,
1357
+ "completions/mean_length": 693.673828125,
1358
+ "completions/mean_terminated_length": 547.5858764648438,
1359
+ "completions/min_length": 183.0,
1360
+ "completions/min_terminated_length": 183.0,
1361
+ "entropy": 0.1179198045283556,
1362
+ "epoch": 1.9583333333333335,
1363
+ "frac_reward_zero_std": 0.09375,
1364
+ "grad_norm": 361.5752258300781,
1365
+ "learning_rate": 7.992473943667311e-08,
1366
+ "loss": 2.0794,
1367
+ "num_tokens": 18885279.0,
1368
+ "reward": 0.40157318115234375,
1369
+ "reward_std": 0.30041205883026123,
1370
+ "rewards/<lambda>/mean": 0.40157318115234375,
1371
+ "rewards/<lambda>/std": 0.4995480477809906,
1372
+ "sampling/importance_sampling_ratio/max": 2.0,
1373
+ "sampling/importance_sampling_ratio/mean": 1.0000332593917847,
1374
+ "sampling/importance_sampling_ratio/min": 0.11990858614444733,
1375
+ "sampling/sampling_logp_difference/max": 2.121025562286377,
1376
+ "sampling/sampling_logp_difference/mean": 0.009333573281764984,
1377
+ "step": 47,
1378
+ "step_time": 50.76096178218722
1379
+ },
1380
+ {
1381
+ "adpo/anchor_kl": 0.0,
1382
+ "adpo/anchor_updates": 0.0,
1383
+ "completions/clipped_ratio": 0.4375,
1384
+ "completions/max_length": 1024.0,
1385
+ "completions/max_terminated_length": 1024.0,
1386
+ "completions/mean_length": 819.15234375,
1387
+ "completions/mean_terminated_length": 662.8819580078125,
1388
+ "completions/min_length": 208.0,
1389
+ "completions/min_terminated_length": 208.0,
1390
+ "entropy": 0.1297450098209083,
1391
+ "epoch": 2.0,
1392
+ "frac_reward_zero_std": 0.046875,
1393
+ "grad_norm": 411.93304443359375,
1394
+ "learning_rate": 2.0007872523362668e-08,
1395
+ "loss": 2.0794,
1396
+ "num_tokens": 19390165.0,
1397
+ "reward": 0.26859036087989807,
1398
+ "reward_std": 0.2463715523481369,
1399
+ "rewards/<lambda>/mean": 0.26859036087989807,
1400
+ "rewards/<lambda>/std": 0.4558042287826538,
1401
+ "sampling/importance_sampling_ratio/max": 2.0,
1402
+ "sampling/importance_sampling_ratio/mean": 0.9999411702156067,
1403
+ "sampling/importance_sampling_ratio/min": 0.17340362071990967,
1404
+ "sampling/sampling_logp_difference/max": 1.7521333694458008,
1405
+ "sampling/sampling_logp_difference/mean": 0.009593246504664421,
1406
+ "step": 48,
1407
+ "step_time": 65.81808259896934
1408
+ },
1409
+ {
1410
+ "epoch": 2.0,
1411
+ "step": 48,
1412
+ "total_flos": 0.0,
1413
+ "train_loss": 2.0794413089752197,
1414
+ "train_runtime": 3642.2415,
1415
+ "train_samples_per_second": 0.874,
1416
+ "train_steps_per_second": 0.013
1417
+ }
1418
+ ],
1419
+ "logging_steps": 1,
1420
+ "max_steps": 48,
1421
+ "num_input_tokens_seen": 19390165,
1422
+ "num_train_epochs": 2,
1423
+ "save_steps": 500,
1424
+ "stateful_callbacks": {
1425
+ "TrainerControl": {
1426
+ "args": {
1427
+ "should_epoch_stop": false,
1428
+ "should_evaluate": false,
1429
+ "should_log": false,
1430
+ "should_save": true,
1431
+ "should_training_stop": true
1432
+ },
1433
+ "attributes": {}
1434
+ }
1435
+ },
1436
+ "total_flos": 0.0,
1437
+ "train_batch_size": 8,
1438
+ "trial_name": null,
1439
+ "trial_params": null
1440
+ }