mlxha commited on
Commit
2c4170f
·
verified ·
1 Parent(s): ce559e8

Model save

Browse files
Files changed (5) hide show
  1. README.md +68 -0
  2. all_results.json +8 -0
  3. generation_config.json +13 -0
  4. train_results.json +8 -0
  5. trainer_state.json +1087 -0
README.md ADDED
@@ -0,0 +1,68 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ base_model: Qwen/Qwen3-8B
3
+ library_name: transformers
4
+ model_name: Qwen3-8B-grpo-medmcqa-medi70
5
+ tags:
6
+ - generated_from_trainer
7
+ - trl
8
+ - grpo
9
+ licence: license
10
+ ---
11
+
12
+ # Model Card for Qwen3-8B-grpo-medmcqa-medi70
13
+
14
+ This model is a fine-tuned version of [Qwen/Qwen3-8B](https://huggingface.co/Qwen/Qwen3-8B).
15
+ It has been trained using [TRL](https://github.com/huggingface/trl).
16
+
17
+ ## Quick start
18
+
19
+ ```python
20
+ from transformers import pipeline
21
+
22
+ question = "If you had a time machine, but could only go to the past or the future once and never return, which would you choose and why?"
23
+ generator = pipeline("text-generation", model="mlxha/Qwen3-8B-grpo-medmcqa-medi70", device="cuda")
24
+ output = generator([{"role": "user", "content": question}], max_new_tokens=128, return_full_text=False)[0]
25
+ print(output["generated_text"])
26
+ ```
27
+
28
+ ## Training procedure
29
+
30
+ [<img src="https://raw.githubusercontent.com/wandb/assets/main/wandb-github-badge-28.svg" alt="Visualize in Weights & Biases" width="150" height="24"/>](https://wandb.ai/alexs-team/reasoning/runs/a353e548-7458-41bd-a49d-4ba5a41cdc37)
31
+
32
+
33
+ This model was trained with GRPO, a method introduced in [DeepSeekMath: Pushing the Limits of Mathematical Reasoning in Open Language Models](https://huggingface.co/papers/2402.03300).
34
+
35
+ ### Framework versions
36
+
37
+ - TRL: 0.18.0.dev0
38
+ - Transformers: 4.52.0.dev0
39
+ - Pytorch: 2.6.0
40
+ - Datasets: 3.6.0
41
+ - Tokenizers: 0.21.1
42
+
43
+ ## Citations
44
+
45
+ Cite GRPO as:
46
+
47
+ ```bibtex
48
+ @article{zhihong2024deepseekmath,
49
+ title = {{DeepSeekMath: Pushing the Limits of Mathematical Reasoning in Open Language Models}},
50
+ author = {Zhihong Shao and Peiyi Wang and Qihao Zhu and Runxin Xu and Junxiao Song and Mingchuan Zhang and Y. K. Li and Y. Wu and Daya Guo},
51
+ year = 2024,
52
+ eprint = {arXiv:2402.03300},
53
+ }
54
+
55
+ ```
56
+
57
+ Cite TRL as:
58
+
59
+ ```bibtex
60
+ @misc{vonwerra2022trl,
61
+ title = {{TRL: Transformer Reinforcement Learning}},
62
+ author = {Leandro von Werra and Younes Belkada and Lewis Tunstall and Edward Beeching and Tristan Thrush and Nathan Lambert and Shengyi Huang and Kashif Rasul and Quentin Gallou{\'e}dec},
63
+ year = 2020,
64
+ journal = {GitHub repository},
65
+ publisher = {GitHub},
66
+ howpublished = {\url{https://github.com/huggingface/trl}}
67
+ }
68
+ ```
all_results.json ADDED
@@ -0,0 +1,8 @@
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "total_flos": 0.0,
3
+ "train_loss": 22251.08962759373,
4
+ "train_runtime": 39419.1987,
5
+ "train_samples": 16885,
6
+ "train_samples_per_second": 0.428,
7
+ "train_steps_per_second": 0.004
8
+ }
generation_config.json ADDED
@@ -0,0 +1,13 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "bos_token_id": 151643,
3
+ "do_sample": true,
4
+ "eos_token_id": [
5
+ 151645,
6
+ 151643
7
+ ],
8
+ "pad_token_id": 151643,
9
+ "temperature": 0.6,
10
+ "top_k": 20,
11
+ "top_p": 0.95,
12
+ "transformers_version": "4.52.0.dev0"
13
+ }
train_results.json ADDED
@@ -0,0 +1,8 @@
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "total_flos": 0.0,
3
+ "train_loss": 22251.08962759373,
4
+ "train_runtime": 39419.1987,
5
+ "train_samples": 16885,
6
+ "train_samples_per_second": 0.428,
7
+ "train_steps_per_second": 0.004
8
+ }
trainer_state.json ADDED
@@ -0,0 +1,1087 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "best_global_step": null,
3
+ "best_metric": null,
4
+ "best_model_checkpoint": null,
5
+ "epoch": 0.9946714031971581,
6
+ "eval_steps": 500,
7
+ "global_step": 175,
8
+ "is_hyper_param_search": false,
9
+ "is_local_process_zero": true,
10
+ "is_world_process_zero": true,
11
+ "log_history": [
12
+ {
13
+ "clip_ratio/high_max": 0.0,
14
+ "clip_ratio/high_mean": 0.0,
15
+ "clip_ratio/low_mean": 0.0,
16
+ "clip_ratio/low_min": 0.0,
17
+ "clip_ratio/region_mean": 0.0,
18
+ "completions/clipped_ratio": 0.0,
19
+ "completions/max_length": 3742.0,
20
+ "completions/max_terminated_length": 3742.0,
21
+ "completions/mean_length": 1092.838623046875,
22
+ "completions/mean_terminated_length": 1092.838623046875,
23
+ "completions/min_length": 250.0,
24
+ "completions/min_terminated_length": 250.0,
25
+ "epoch": 0.005683836589698046,
26
+ "grad_norm": 0.13580130434100188,
27
+ "kl": 0.0012049674987792969,
28
+ "learning_rate": 0.0,
29
+ "loss": 0.0074,
30
+ "num_tokens": 954428.0,
31
+ "reward": 0.6531250476837158,
32
+ "reward_std": 0.16057492792606354,
33
+ "rewards/format_reward/mean": 1.0,
34
+ "rewards/format_reward/std": 0.0,
35
+ "rewards/mcq_accuracy_reward/mean": 0.453125,
36
+ "rewards/mcq_accuracy_reward/std": 0.49812230467796326,
37
+ "rewards/tag_count_reward/mean": 1.0,
38
+ "rewards/tag_count_reward/std": 0.0,
39
+ "step": 1
40
+ },
41
+ {
42
+ "clip_ratio/high_max": 0.0,
43
+ "clip_ratio/high_mean": 0.0,
44
+ "clip_ratio/low_mean": 0.0,
45
+ "clip_ratio/low_min": 0.0,
46
+ "clip_ratio/region_mean": 0.0,
47
+ "completions/clipped_ratio": 0.0,
48
+ "completions/max_length": 5508.0,
49
+ "completions/max_terminated_length": 5508.0,
50
+ "completions/mean_length": 1069.8079528808594,
51
+ "completions/mean_terminated_length": 1069.8079528808594,
52
+ "completions/min_length": 178.75,
53
+ "completions/min_terminated_length": 178.75,
54
+ "epoch": 0.028419182948490232,
55
+ "grad_norm": 0.14511271553818444,
56
+ "kl": 0.0012257099151611328,
57
+ "learning_rate": 1.111111111111111e-06,
58
+ "loss": 0.002,
59
+ "num_tokens": 4707030.0,
60
+ "reward": 0.6582194864749908,
61
+ "reward_std": 0.16218429803848267,
62
+ "rewards/format_reward/mean": 0.9993489682674408,
63
+ "rewards/format_reward/std": 0.018042195588350296,
64
+ "rewards/mcq_accuracy_reward/mean": 0.4583333358168602,
65
+ "rewards/mcq_accuracy_reward/std": 0.49666526168584824,
66
+ "rewards/tag_count_reward/mean": 0.99951171875,
67
+ "rewards/tag_count_reward/std": 0.013531646691262722,
68
+ "step": 5
69
+ },
70
+ {
71
+ "clip_ratio/high_max": 0.0,
72
+ "clip_ratio/high_mean": 0.0,
73
+ "clip_ratio/low_mean": 0.0,
74
+ "clip_ratio/low_min": 0.0,
75
+ "clip_ratio/region_mean": 0.0,
76
+ "completions/clipped_ratio": 0.0,
77
+ "completions/max_length": 4794.4,
78
+ "completions/max_terminated_length": 4794.4,
79
+ "completions/mean_length": 1049.7560302734375,
80
+ "completions/mean_terminated_length": 1049.7560302734375,
81
+ "completions/min_length": 121.8,
82
+ "completions/min_terminated_length": 121.8,
83
+ "epoch": 0.056838365896980464,
84
+ "grad_norm": 0.1398931239536535,
85
+ "kl": 0.0014247894287109375,
86
+ "learning_rate": 2.5e-06,
87
+ "loss": 0.004,
88
+ "num_tokens": 9314069.0,
89
+ "reward": 0.6792318463325501,
90
+ "reward_std": 0.166354176402092,
91
+ "rewards/format_reward/mean": 0.99921875,
92
+ "rewards/format_reward/std": 0.01741643026471138,
93
+ "rewards/mcq_accuracy_reward/mean": 0.47942708134651185,
94
+ "rewards/mcq_accuracy_reward/std": 0.4984712541103363,
95
+ "rewards/tag_count_reward/mean": 0.998828113079071,
96
+ "rewards/tag_count_reward/std": 0.023616421967744827,
97
+ "step": 10
98
+ },
99
+ {
100
+ "clip_ratio/high_max": 0.0,
101
+ "clip_ratio/high_mean": 0.0,
102
+ "clip_ratio/low_mean": 0.0,
103
+ "clip_ratio/low_min": 0.0,
104
+ "clip_ratio/region_mean": 0.0,
105
+ "completions/clipped_ratio": 0.0,
106
+ "completions/max_length": 4637.0,
107
+ "completions/max_terminated_length": 4637.0,
108
+ "completions/mean_length": 1036.5583740234374,
109
+ "completions/mean_terminated_length": 1036.5583740234374,
110
+ "completions/min_length": 227.6,
111
+ "completions/min_terminated_length": 227.6,
112
+ "epoch": 0.0852575488454707,
113
+ "grad_norm": 0.15112174245508642,
114
+ "kl": 0.0023210525512695314,
115
+ "learning_rate": 3.88888888888889e-06,
116
+ "loss": 0.0125,
117
+ "num_tokens": 13874013.0,
118
+ "reward": 0.6991276741027832,
119
+ "reward_std": 0.16582457721233368,
120
+ "rewards/format_reward/mean": 0.9994791746139526,
121
+ "rewards/format_reward/std": 0.014433756470680237,
122
+ "rewards/mcq_accuracy_reward/mean": 0.49921875,
123
+ "rewards/mcq_accuracy_reward/std": 0.49765779376029967,
124
+ "rewards/tag_count_reward/mean": 0.999609375,
125
+ "rewards/tag_count_reward/std": 0.009444911777973176,
126
+ "step": 15
127
+ },
128
+ {
129
+ "clip_ratio/high_max": 0.0,
130
+ "clip_ratio/high_mean": 0.0,
131
+ "clip_ratio/low_mean": 0.0,
132
+ "clip_ratio/low_min": 0.0,
133
+ "clip_ratio/region_mean": 0.0,
134
+ "completions/clipped_ratio": 0.0,
135
+ "completions/max_length": 5373.4,
136
+ "completions/max_terminated_length": 5373.4,
137
+ "completions/mean_length": 1068.2205932617187,
138
+ "completions/mean_terminated_length": 1068.2205932617187,
139
+ "completions/min_length": 213.8,
140
+ "completions/min_terminated_length": 213.8,
141
+ "epoch": 0.11367673179396093,
142
+ "grad_norm": 0.1748099542859131,
143
+ "kl": 0.005224609375,
144
+ "learning_rate": 4.999499509357132e-06,
145
+ "loss": -0.0021,
146
+ "num_tokens": 18546748.0,
147
+ "reward": 0.6684310674667359,
148
+ "reward_std": 0.1836001008749008,
149
+ "rewards/format_reward/mean": 0.9997395873069763,
150
+ "rewards/format_reward/std": 0.007216878235340118,
151
+ "rewards/mcq_accuracy_reward/mean": 0.46848957538604735,
152
+ "rewards/mcq_accuracy_reward/std": 0.49776612520217894,
153
+ "rewards/tag_count_reward/mean": 0.9996744751930237,
154
+ "rewards/tag_count_reward/std": 0.007962546683847905,
155
+ "step": 20
156
+ },
157
+ {
158
+ "clip_ratio/high_max": 0.0,
159
+ "clip_ratio/high_mean": 0.0,
160
+ "clip_ratio/low_mean": 0.0,
161
+ "clip_ratio/low_min": 0.0,
162
+ "clip_ratio/region_mean": 0.0,
163
+ "completions/clipped_ratio": 0.0,
164
+ "completions/max_length": 5827.2,
165
+ "completions/max_terminated_length": 5827.2,
166
+ "completions/mean_length": 1014.5508178710937,
167
+ "completions/mean_terminated_length": 1014.5508178710937,
168
+ "completions/min_length": 239.6,
169
+ "completions/min_terminated_length": 239.6,
170
+ "epoch": 0.14209591474245115,
171
+ "grad_norm": 0.17738027139049306,
172
+ "kl": 0.00942840576171875,
173
+ "learning_rate": 4.982003369106287e-06,
174
+ "loss": 0.0057,
175
+ "num_tokens": 23021767.0,
176
+ "reward": 0.6767448544502258,
177
+ "reward_std": 0.17676816284656524,
178
+ "rewards/format_reward/mean": 0.9994791626930237,
179
+ "rewards/format_reward/std": 0.010199552029371261,
180
+ "rewards/mcq_accuracy_reward/mean": 0.4768229126930237,
181
+ "rewards/mcq_accuracy_reward/std": 0.49895642399787904,
182
+ "rewards/tag_count_reward/mean": 0.9997395873069763,
183
+ "rewards/tag_count_reward/std": 0.003601375222206116,
184
+ "step": 25
185
+ },
186
+ {
187
+ "clip_ratio/high_max": 0.0,
188
+ "clip_ratio/high_mean": 0.0,
189
+ "clip_ratio/low_mean": 0.0,
190
+ "clip_ratio/low_min": 0.0,
191
+ "clip_ratio/region_mean": 0.0,
192
+ "completions/clipped_ratio": 0.0,
193
+ "completions/max_length": 5485.4,
194
+ "completions/max_terminated_length": 5485.4,
195
+ "completions/mean_length": 1067.72373046875,
196
+ "completions/mean_terminated_length": 1067.72373046875,
197
+ "completions/min_length": 248.0,
198
+ "completions/min_terminated_length": 248.0,
199
+ "epoch": 0.1705150976909414,
200
+ "grad_norm": 0.1559592819101095,
201
+ "kl": 0.01161041259765625,
202
+ "learning_rate": 4.939682729058839e-06,
203
+ "loss": 0.0027,
204
+ "num_tokens": 27697786.0,
205
+ "reward": 0.6473893880844116,
206
+ "reward_std": 0.16071836650371552,
207
+ "rewards/format_reward/mean": 1.0,
208
+ "rewards/format_reward/std": 0.0,
209
+ "rewards/mcq_accuracy_reward/mean": 0.44739583134651184,
210
+ "rewards/mcq_accuracy_reward/std": 0.49608185291290285,
211
+ "rewards/tag_count_reward/mean": 0.9999348998069764,
212
+ "rewards/tag_count_reward/std": 0.0018042195588350296,
213
+ "step": 30
214
+ },
215
+ {
216
+ "clip_ratio/high_max": 0.0,
217
+ "clip_ratio/high_mean": 0.0,
218
+ "clip_ratio/low_mean": 0.0,
219
+ "clip_ratio/low_min": 0.0,
220
+ "clip_ratio/region_mean": 0.0,
221
+ "completions/clipped_ratio": 0.0,
222
+ "completions/max_length": 5728.2,
223
+ "completions/max_terminated_length": 5728.2,
224
+ "completions/mean_length": 1163.4677490234376,
225
+ "completions/mean_terminated_length": 1163.4677490234376,
226
+ "completions/min_length": 255.2,
227
+ "completions/min_terminated_length": 255.2,
228
+ "epoch": 0.1989342806394316,
229
+ "grad_norm": 0.15172870795054275,
230
+ "kl": 0.01438446044921875,
231
+ "learning_rate": 4.872960871766826e-06,
232
+ "loss": -0.0006,
233
+ "num_tokens": 32744310.0,
234
+ "reward": 0.6907943248748779,
235
+ "reward_std": 0.1814502149820328,
236
+ "rewards/format_reward/mean": 0.9997395873069763,
237
+ "rewards/format_reward/std": 0.007216878235340118,
238
+ "rewards/mcq_accuracy_reward/mean": 0.49088541269302366,
239
+ "rewards/mcq_accuracy_reward/std": 0.4999137282371521,
240
+ "rewards/tag_count_reward/mean": 0.9993489503860473,
241
+ "rewards/tag_count_reward/std": 0.010171890631318093,
242
+ "step": 35
243
+ },
244
+ {
245
+ "clip_ratio/high_max": 0.0,
246
+ "clip_ratio/high_mean": 0.0,
247
+ "clip_ratio/low_mean": 0.0,
248
+ "clip_ratio/low_min": 0.0,
249
+ "clip_ratio/region_mean": 0.0,
250
+ "completions/clipped_ratio": 0.0002604166666666741,
251
+ "completions/max_length": 5812.4,
252
+ "completions/max_terminated_length": 5298.8,
253
+ "completions/mean_length": 1077.8024047851563,
254
+ "completions/mean_terminated_length": 1075.9230712890626,
255
+ "completions/min_length": 250.6,
256
+ "completions/min_terminated_length": 250.6,
257
+ "epoch": 0.22735346358792186,
258
+ "grad_norm": 0.1564558072390042,
259
+ "kl": 0.015766143798828125,
260
+ "learning_rate": 4.782505135862176e-06,
261
+ "loss": 0.0012,
262
+ "num_tokens": 37463479.0,
263
+ "reward": 0.6952149271965027,
264
+ "reward_std": 0.1775230199098587,
265
+ "rewards/format_reward/mean": 0.9997395873069763,
266
+ "rewards/format_reward/std": 0.007216878235340118,
267
+ "rewards/mcq_accuracy_reward/mean": 0.4953125059604645,
268
+ "rewards/mcq_accuracy_reward/std": 0.4984812021255493,
269
+ "rewards/tag_count_reward/mean": 0.9992838621139526,
270
+ "rewards/tag_count_reward/std": 0.01261383220553398,
271
+ "step": 40
272
+ },
273
+ {
274
+ "clip_ratio/high_max": 0.0,
275
+ "clip_ratio/high_mean": 0.0,
276
+ "clip_ratio/low_mean": 0.0,
277
+ "clip_ratio/low_min": 0.0,
278
+ "clip_ratio/region_mean": 0.0,
279
+ "completions/clipped_ratio": 0.0002604166666666741,
280
+ "completions/max_length": 4943.8,
281
+ "completions/max_terminated_length": 4187.4,
282
+ "completions/mean_length": 992.1836181640625,
283
+ "completions/mean_terminated_length": 990.3313232421875,
284
+ "completions/min_length": 230.0,
285
+ "completions/min_terminated_length": 230.0,
286
+ "epoch": 0.2557726465364121,
287
+ "grad_norm": 0.1740869664959226,
288
+ "kl": 0.01738433837890625,
289
+ "learning_rate": 4.669220241469573e-06,
290
+ "loss": 0.0129,
291
+ "num_tokens": 41846552.0,
292
+ "reward": 0.6533008456230164,
293
+ "reward_std": 0.17375805079936982,
294
+ "rewards/format_reward/mean": 0.9994791746139526,
295
+ "rewards/format_reward/std": 0.014433756470680237,
296
+ "rewards/mcq_accuracy_reward/mean": 0.4533854126930237,
297
+ "rewards/mcq_accuracy_reward/std": 0.49718082547187803,
298
+ "rewards/tag_count_reward/mean": 0.9996744871139527,
299
+ "rewards/tag_count_reward/std": 0.009021097794175148,
300
+ "step": 45
301
+ },
302
+ {
303
+ "clip_ratio/high_max": 0.0,
304
+ "clip_ratio/high_mean": 0.0,
305
+ "clip_ratio/low_mean": 0.0,
306
+ "clip_ratio/low_min": 0.0,
307
+ "clip_ratio/region_mean": 0.0,
308
+ "completions/clipped_ratio": 0.0,
309
+ "completions/max_length": 6464.0,
310
+ "completions/max_terminated_length": 6464.0,
311
+ "completions/mean_length": 1109.7568115234376,
312
+ "completions/mean_terminated_length": 1109.7568115234376,
313
+ "completions/min_length": 233.2,
314
+ "completions/min_terminated_length": 233.2,
315
+ "epoch": 0.2841918294849023,
316
+ "grad_norm": 0.15589282758117567,
317
+ "kl": 0.0189788818359375,
318
+ "learning_rate": 4.534239241377266e-06,
319
+ "loss": 0.001,
320
+ "num_tokens": 46687962.0,
321
+ "reward": 0.676927137374878,
322
+ "reward_std": 0.17989677786827088,
323
+ "rewards/format_reward/mean": 0.99921875,
324
+ "rewards/format_reward/std": 0.01741643026471138,
325
+ "rewards/mcq_accuracy_reward/mean": 0.47708333730697633,
326
+ "rewards/mcq_accuracy_reward/std": 0.4982982099056244,
327
+ "rewards/tag_count_reward/mean": 0.99921875,
328
+ "rewards/tag_count_reward/std": 0.016611021012067795,
329
+ "step": 50
330
+ },
331
+ {
332
+ "clip_ratio/high_max": 0.0,
333
+ "clip_ratio/high_mean": 0.0,
334
+ "clip_ratio/low_mean": 0.0,
335
+ "clip_ratio/low_min": 0.0,
336
+ "clip_ratio/region_mean": 0.0,
337
+ "completions/clipped_ratio": 0.0,
338
+ "completions/max_length": 5222.0,
339
+ "completions/max_terminated_length": 5222.0,
340
+ "completions/mean_length": 1135.95576171875,
341
+ "completions/mean_terminated_length": 1135.95576171875,
342
+ "completions/min_length": 249.8,
343
+ "completions/min_terminated_length": 249.8,
344
+ "epoch": 0.31261101243339257,
345
+ "grad_norm": 0.13369568955391997,
346
+ "kl": 0.0205780029296875,
347
+ "learning_rate": 4.378912188470374e-06,
348
+ "loss": 0.0046,
349
+ "num_tokens": 51627616.0,
350
+ "reward": 0.6671224713325501,
351
+ "reward_std": 0.160793936252594,
352
+ "rewards/format_reward/mean": 0.9997395873069763,
353
+ "rewards/format_reward/std": 0.007216878235340118,
354
+ "rewards/mcq_accuracy_reward/mean": 0.4671874940395355,
355
+ "rewards/mcq_accuracy_reward/std": 0.4979245364665985,
356
+ "rewards/tag_count_reward/mean": 0.999609375,
357
+ "rewards/tag_count_reward/std": 0.008064506202936172,
358
+ "step": 55
359
+ },
360
+ {
361
+ "clip_ratio/high_max": 0.0,
362
+ "clip_ratio/high_mean": 0.0,
363
+ "clip_ratio/low_mean": 0.0,
364
+ "clip_ratio/low_min": 0.0,
365
+ "clip_ratio/region_mean": 0.0,
366
+ "completions/clipped_ratio": 0.0002604166666666741,
367
+ "completions/max_length": 6132.4,
368
+ "completions/max_terminated_length": 5533.2,
369
+ "completions/mean_length": 1087.3320434570312,
370
+ "completions/mean_terminated_length": 1085.4773315429688,
371
+ "completions/min_length": 263.6,
372
+ "completions/min_terminated_length": 263.6,
373
+ "epoch": 0.3410301953818828,
374
+ "grad_norm": 0.1477789943624076,
375
+ "kl": 0.020989990234375,
376
+ "learning_rate": 4.204792632772754e-06,
377
+ "loss": -0.0003,
378
+ "num_tokens": 56378771.0,
379
+ "reward": 0.6983789801597595,
380
+ "reward_std": 0.17249701023101807,
381
+ "rewards/format_reward/mean": 0.9997395873069763,
382
+ "rewards/format_reward/std": 0.007216878235340118,
383
+ "rewards/mcq_accuracy_reward/mean": 0.4984375059604645,
384
+ "rewards/mcq_accuracy_reward/std": 0.4989359140396118,
385
+ "rewards/tag_count_reward/mean": 0.9996744871139527,
386
+ "rewards/tag_count_reward/std": 0.007507431134581566,
387
+ "step": 60
388
+ },
389
+ {
390
+ "clip_ratio/high_max": 0.0,
391
+ "clip_ratio/high_mean": 0.0,
392
+ "clip_ratio/low_mean": 0.0,
393
+ "clip_ratio/low_min": 0.0,
394
+ "clip_ratio/region_mean": 0.0,
395
+ "completions/clipped_ratio": 0.0,
396
+ "completions/max_length": 4786.4,
397
+ "completions/max_terminated_length": 4786.4,
398
+ "completions/mean_length": 1063.5450805664063,
399
+ "completions/mean_terminated_length": 1063.5450805664063,
400
+ "completions/min_length": 225.6,
401
+ "completions/min_terminated_length": 225.6,
402
+ "epoch": 0.369449378330373,
403
+ "grad_norm": 0.14015972377677116,
404
+ "kl": 0.0214569091796875,
405
+ "learning_rate": 4.0136220831513205e-06,
406
+ "loss": -0.0,
407
+ "num_tokens": 61039224.0,
408
+ "reward": 0.6768229961395263,
409
+ "reward_std": 0.1578069359064102,
410
+ "rewards/format_reward/mean": 1.0,
411
+ "rewards/format_reward/std": 0.0,
412
+ "rewards/mcq_accuracy_reward/mean": 0.47682292461395265,
413
+ "rewards/mcq_accuracy_reward/std": 0.4989953935146332,
414
+ "rewards/tag_count_reward/mean": 1.0,
415
+ "rewards/tag_count_reward/std": 0.0,
416
+ "step": 65
417
+ },
418
+ {
419
+ "clip_ratio/high_max": 0.0,
420
+ "clip_ratio/high_mean": 0.0,
421
+ "clip_ratio/low_mean": 0.0,
422
+ "clip_ratio/low_min": 0.0,
423
+ "clip_ratio/region_mean": 0.0,
424
+ "completions/clipped_ratio": 0.0,
425
+ "completions/max_length": 5206.6,
426
+ "completions/max_terminated_length": 5206.6,
427
+ "completions/mean_length": 1109.176611328125,
428
+ "completions/mean_terminated_length": 1109.176611328125,
429
+ "completions/min_length": 246.0,
430
+ "completions/min_terminated_length": 246.0,
431
+ "epoch": 0.3978685612788632,
432
+ "grad_norm": 0.16531093494639468,
433
+ "kl": 95.62167663574219,
434
+ "learning_rate": 3.807312589093701e-06,
435
+ "loss": 3.8239,
436
+ "num_tokens": 65872382.0,
437
+ "reward": 0.6960807919502259,
438
+ "reward_std": 0.16992745101451873,
439
+ "rewards/format_reward/mean": 1.0,
440
+ "rewards/format_reward/std": 0.0,
441
+ "rewards/mcq_accuracy_reward/mean": 0.49609375596046446,
442
+ "rewards/mcq_accuracy_reward/std": 0.4989615023136139,
443
+ "rewards/tag_count_reward/mean": 0.9998697876930237,
444
+ "rewards/tag_count_reward/std": 0.0025498881936073304,
445
+ "step": 70
446
+ },
447
+ {
448
+ "clip_ratio/high_max": 0.0,
449
+ "clip_ratio/high_mean": 0.0,
450
+ "clip_ratio/low_mean": 0.0,
451
+ "clip_ratio/low_min": 0.0,
452
+ "clip_ratio/region_mean": 0.0,
453
+ "completions/clipped_ratio": 0.0,
454
+ "completions/max_length": 5165.0,
455
+ "completions/max_terminated_length": 5165.0,
456
+ "completions/mean_length": 1079.5161743164062,
457
+ "completions/mean_terminated_length": 1079.5161743164062,
458
+ "completions/min_length": 254.6,
459
+ "completions/min_terminated_length": 254.6,
460
+ "epoch": 0.42628774422735344,
461
+ "grad_norm": 0.15112401029602476,
462
+ "kl": 0.0230438232421875,
463
+ "learning_rate": 3.5879276167728343e-06,
464
+ "loss": 0.0125,
465
+ "num_tokens": 70595996.0,
466
+ "reward": 0.6843750596046447,
467
+ "reward_std": 0.18003216683864592,
468
+ "rewards/format_reward/mean": 1.0,
469
+ "rewards/format_reward/std": 0.0,
470
+ "rewards/mcq_accuracy_reward/mean": 0.484375,
471
+ "rewards/mcq_accuracy_reward/std": 0.4970580399036407,
472
+ "rewards/tag_count_reward/mean": 1.0,
473
+ "rewards/tag_count_reward/std": 0.0,
474
+ "step": 75
475
+ },
476
+ {
477
+ "clip_ratio/high_max": 0.0,
478
+ "clip_ratio/high_mean": 0.0,
479
+ "clip_ratio/low_mean": 0.0,
480
+ "clip_ratio/low_min": 0.0,
481
+ "clip_ratio/region_mean": 0.0,
482
+ "completions/clipped_ratio": 0.0002604166666666741,
483
+ "completions/max_length": 5178.6,
484
+ "completions/max_terminated_length": 4963.4,
485
+ "completions/mean_length": 1061.828955078125,
486
+ "completions/mean_terminated_length": 1059.984521484375,
487
+ "completions/min_length": 231.0,
488
+ "completions/min_terminated_length": 231.0,
489
+ "epoch": 0.4547069271758437,
490
+ "grad_norm": 0.15923626474140243,
491
+ "kl": 0.02301025390625,
492
+ "learning_rate": 3.3576614106722473e-06,
493
+ "loss": 0.0028,
494
+ "num_tokens": 75247475.0,
495
+ "reward": 0.7145378351211548,
496
+ "reward_std": 0.1841348797082901,
497
+ "rewards/format_reward/mean": 0.9997395873069763,
498
+ "rewards/format_reward/std": 0.007216878235340118,
499
+ "rewards/mcq_accuracy_reward/mean": 0.5145833313465118,
500
+ "rewards/mcq_accuracy_reward/std": 0.49900742769241335,
501
+ "rewards/tag_count_reward/mean": 0.9998046875,
502
+ "rewards/tag_count_reward/std": 0.005412658676505089,
503
+ "step": 80
504
+ },
505
+ {
506
+ "clip_ratio/high_max": 0.0,
507
+ "clip_ratio/high_mean": 0.0,
508
+ "clip_ratio/low_mean": 0.0,
509
+ "clip_ratio/low_min": 0.0,
510
+ "clip_ratio/region_mean": 0.0,
511
+ "completions/clipped_ratio": 0.0,
512
+ "completions/max_length": 5323.8,
513
+ "completions/max_terminated_length": 5323.8,
514
+ "completions/mean_length": 1048.9911743164062,
515
+ "completions/mean_terminated_length": 1048.9911743164062,
516
+ "completions/min_length": 247.6,
517
+ "completions/min_terminated_length": 247.6,
518
+ "epoch": 0.48312611012433393,
519
+ "grad_norm": 0.16115740042560586,
520
+ "kl": 0.02386932373046875,
521
+ "learning_rate": 3.118817047192907e-06,
522
+ "loss": 0.0024,
523
+ "num_tokens": 79853433.0,
524
+ "reward": 0.6648047566413879,
525
+ "reward_std": 0.189658322930336,
526
+ "rewards/format_reward/mean": 0.9997395873069763,
527
+ "rewards/format_reward/std": 0.007216878235340118,
528
+ "rewards/mcq_accuracy_reward/mean": 0.46484375596046446,
529
+ "rewards/mcq_accuracy_reward/std": 0.4943417370319366,
530
+ "rewards/tag_count_reward/mean": 0.9998697996139526,
531
+ "rewards/tag_count_reward/std": 0.003608439117670059,
532
+ "step": 85
533
+ },
534
+ {
535
+ "clip_ratio/high_max": 0.0,
536
+ "clip_ratio/high_mean": 0.0,
537
+ "clip_ratio/low_mean": 0.0,
538
+ "clip_ratio/low_min": 0.0,
539
+ "clip_ratio/region_mean": 0.0,
540
+ "completions/clipped_ratio": 0.0,
541
+ "completions/max_length": 6016.2,
542
+ "completions/max_terminated_length": 6016.2,
543
+ "completions/mean_length": 1056.59873046875,
544
+ "completions/mean_terminated_length": 1056.59873046875,
545
+ "completions/min_length": 238.0,
546
+ "completions/min_terminated_length": 238.0,
547
+ "epoch": 0.5115452930728241,
548
+ "grad_norm": 0.21552374412647782,
549
+ "kl": 0.02350616455078125,
550
+ "learning_rate": 2.8737833997450658e-06,
551
+ "loss": 0.0042,
552
+ "num_tokens": 84486748.0,
553
+ "reward": 0.6861914873123169,
554
+ "reward_std": 0.16900237798690795,
555
+ "rewards/format_reward/mean": 1.0,
556
+ "rewards/format_reward/std": 0.0,
557
+ "rewards/mcq_accuracy_reward/mean": 0.4861979126930237,
558
+ "rewards/mcq_accuracy_reward/std": 0.4990255832672119,
559
+ "rewards/tag_count_reward/mean": 0.9999348998069764,
560
+ "rewards/tag_count_reward/std": 0.0018042195588350296,
561
+ "step": 90
562
+ },
563
+ {
564
+ "clip_ratio/high_max": 0.0,
565
+ "clip_ratio/high_mean": 0.0,
566
+ "clip_ratio/low_mean": 0.0,
567
+ "clip_ratio/low_min": 0.0,
568
+ "clip_ratio/region_mean": 0.0,
569
+ "completions/clipped_ratio": 0.0013020833333333482,
570
+ "completions/max_length": 6964.4,
571
+ "completions/max_terminated_length": 6690.8,
572
+ "completions/mean_length": 1151.6203369140626,
573
+ "completions/mean_terminated_length": 1142.4264404296875,
574
+ "completions/min_length": 236.8,
575
+ "completions/min_terminated_length": 236.8,
576
+ "epoch": 0.5399644760213144,
577
+ "grad_norm": 0.14156099172275863,
578
+ "kl": 0.022979736328125,
579
+ "learning_rate": 2.6250112457156296e-06,
580
+ "loss": 0.0076,
581
+ "num_tokens": 89489882.0,
582
+ "reward": 0.6658268809318543,
583
+ "reward_std": 0.16832855641841887,
584
+ "rewards/format_reward/mean": 0.9981770873069763,
585
+ "rewards/format_reward/std": 0.03182193115353584,
586
+ "rewards/mcq_accuracy_reward/mean": 0.46614583134651183,
587
+ "rewards/mcq_accuracy_reward/std": 0.4980164170265198,
588
+ "rewards/tag_count_reward/mean": 0.9986328125,
589
+ "rewards/tag_count_reward/std": 0.023866448923945426,
590
+ "step": 95
591
+ },
592
+ {
593
+ "clip_ratio/high_max": 0.0,
594
+ "clip_ratio/high_mean": 0.0,
595
+ "clip_ratio/low_mean": 0.0,
596
+ "clip_ratio/low_min": 0.0,
597
+ "clip_ratio/region_mean": 0.0,
598
+ "completions/clipped_ratio": 0.0,
599
+ "completions/max_length": 5372.6,
600
+ "completions/max_terminated_length": 5372.6,
601
+ "completions/mean_length": 1119.63466796875,
602
+ "completions/mean_terminated_length": 1119.63466796875,
603
+ "completions/min_length": 220.8,
604
+ "completions/min_terminated_length": 220.8,
605
+ "epoch": 0.5683836589698046,
606
+ "grad_norm": 0.13456208913450085,
607
+ "kl": 0.02239227294921875,
608
+ "learning_rate": 2.374988754284371e-06,
609
+ "loss": 0.0074,
610
+ "num_tokens": 94365687.0,
611
+ "reward": 0.6567123174667359,
612
+ "reward_std": 0.1713259369134903,
613
+ "rewards/format_reward/mean": 1.0,
614
+ "rewards/format_reward/std": 0.0,
615
+ "rewards/mcq_accuracy_reward/mean": 0.4567708373069763,
616
+ "rewards/mcq_accuracy_reward/std": 0.4970240592956543,
617
+ "rewards/tag_count_reward/mean": 0.9994140625,
618
+ "rewards/tag_count_reward/std": 0.011108423396945,
619
+ "step": 100
620
+ },
621
+ {
622
+ "clip_ratio/high_max": 0.0,
623
+ "clip_ratio/high_mean": 0.0,
624
+ "clip_ratio/low_mean": 0.0,
625
+ "clip_ratio/low_min": 0.0,
626
+ "clip_ratio/region_mean": 0.0,
627
+ "completions/clipped_ratio": 0.0002604166666666741,
628
+ "completions/max_length": 5585.0,
629
+ "completions/max_terminated_length": 5486.6,
630
+ "completions/mean_length": 1100.5560302734375,
631
+ "completions/mean_terminated_length": 1098.734423828125,
632
+ "completions/min_length": 252.2,
633
+ "completions/min_terminated_length": 252.2,
634
+ "epoch": 0.5968028419182948,
635
+ "grad_norm": 0.11958484360378753,
636
+ "kl": 0.02285919189453125,
637
+ "learning_rate": 2.1262166002549346e-06,
638
+ "loss": 0.004,
639
+ "num_tokens": 99170286.0,
640
+ "reward": 0.664778733253479,
641
+ "reward_std": 0.14817258715629578,
642
+ "rewards/format_reward/mean": 0.9997395873069763,
643
+ "rewards/format_reward/std": 0.007216878235340118,
644
+ "rewards/mcq_accuracy_reward/mean": 0.46484375596046446,
645
+ "rewards/mcq_accuracy_reward/std": 0.49491878151893615,
646
+ "rewards/tag_count_reward/mean": 0.9996093869209289,
647
+ "rewards/tag_count_reward/std": 0.010825317353010178,
648
+ "step": 105
649
+ },
650
+ {
651
+ "clip_ratio/high_max": 0.0,
652
+ "clip_ratio/high_mean": 0.0,
653
+ "clip_ratio/low_mean": 0.0,
654
+ "clip_ratio/low_min": 0.0,
655
+ "clip_ratio/region_mean": 0.0,
656
+ "completions/clipped_ratio": 0.0,
657
+ "completions/max_length": 4669.0,
658
+ "completions/max_terminated_length": 4669.0,
659
+ "completions/mean_length": 1052.46826171875,
660
+ "completions/mean_terminated_length": 1052.46826171875,
661
+ "completions/min_length": 245.8,
662
+ "completions/min_terminated_length": 245.8,
663
+ "epoch": 0.6252220248667851,
664
+ "grad_norm": 0.14676711793529598,
665
+ "kl": 19608371.221609496,
666
+ "learning_rate": 1.8811829528070935e-06,
667
+ "loss": 778784.15,
668
+ "num_tokens": 103796876.0,
669
+ "reward": 0.6645313143730164,
670
+ "reward_std": 0.18562877476215361,
671
+ "rewards/format_reward/mean": 0.9997395873069763,
672
+ "rewards/format_reward/std": 0.007216878235340118,
673
+ "rewards/mcq_accuracy_reward/mean": 0.46458333134651186,
674
+ "rewards/mcq_accuracy_reward/std": 0.49767143130302427,
675
+ "rewards/tag_count_reward/mean": 0.9997395873069763,
676
+ "rewards/tag_count_reward/std": 0.007216878235340118,
677
+ "step": 110
678
+ },
679
+ {
680
+ "clip_ratio/high_max": 0.0,
681
+ "clip_ratio/high_mean": 0.0,
682
+ "clip_ratio/low_mean": 0.0,
683
+ "clip_ratio/low_min": 0.0,
684
+ "clip_ratio/region_mean": 0.0,
685
+ "completions/clipped_ratio": 0.0002604166666666741,
686
+ "completions/max_length": 5137.8,
687
+ "completions/max_terminated_length": 4382.8,
688
+ "completions/mean_length": 1032.596630859375,
689
+ "completions/mean_terminated_length": 1030.736181640625,
690
+ "completions/min_length": 224.0,
691
+ "completions/min_terminated_length": 224.0,
692
+ "epoch": 0.6536412078152753,
693
+ "grad_norm": 0.1802276243424991,
694
+ "kl": 0.02183837890625,
695
+ "learning_rate": 1.6423385893277537e-06,
696
+ "loss": 0.0108,
697
+ "num_tokens": 108338703.0,
698
+ "reward": 0.6494206309318542,
699
+ "reward_std": 0.17672575116157532,
700
+ "rewards/format_reward/mean": 0.9997395873069763,
701
+ "rewards/format_reward/std": 0.007216878235340118,
702
+ "rewards/mcq_accuracy_reward/mean": 0.4494791626930237,
703
+ "rewards/mcq_accuracy_reward/std": 0.49625040888786315,
704
+ "rewards/tag_count_reward/mean": 0.9996744751930237,
705
+ "rewards/tag_count_reward/std": 0.006501290947198868,
706
+ "step": 115
707
+ },
708
+ {
709
+ "clip_ratio/high_max": 0.0,
710
+ "clip_ratio/high_mean": 0.0,
711
+ "clip_ratio/low_mean": 0.0,
712
+ "clip_ratio/low_min": 0.0,
713
+ "clip_ratio/region_mean": 0.0,
714
+ "completions/clipped_ratio": 0.0002604166666666741,
715
+ "completions/max_length": 5606.2,
716
+ "completions/max_terminated_length": 5407.0,
717
+ "completions/mean_length": 1034.7218994140626,
718
+ "completions/mean_terminated_length": 1032.8922119140625,
719
+ "completions/min_length": 242.0,
720
+ "completions/min_terminated_length": 242.0,
721
+ "epoch": 0.6820603907637656,
722
+ "grad_norm": 0.23206072060115124,
723
+ "kl": 0.02138214111328125,
724
+ "learning_rate": 1.4120723832271665e-06,
725
+ "loss": 0.0068,
726
+ "num_tokens": 112897891.0,
727
+ "reward": 0.7249414920806885,
728
+ "reward_std": 0.1745230883359909,
729
+ "rewards/format_reward/mean": 0.9997395873069763,
730
+ "rewards/format_reward/std": 0.007216878235340118,
731
+ "rewards/mcq_accuracy_reward/mean": 0.524999988079071,
732
+ "rewards/mcq_accuracy_reward/std": 0.49913292527198794,
733
+ "rewards/tag_count_reward/mean": 0.9996744751930237,
734
+ "rewards/tag_count_reward/std": 0.009021097794175148,
735
+ "step": 120
736
+ },
737
+ {
738
+ "clip_ratio/high_max": 0.0,
739
+ "clip_ratio/high_mean": 0.0,
740
+ "clip_ratio/low_mean": 0.0,
741
+ "clip_ratio/low_min": 0.0,
742
+ "clip_ratio/region_mean": 0.0,
743
+ "completions/clipped_ratio": 0.0002604166666666741,
744
+ "completions/max_length": 5944.4,
745
+ "completions/max_terminated_length": 5786.4,
746
+ "completions/mean_length": 1096.26279296875,
747
+ "completions/mean_terminated_length": 1094.3953369140625,
748
+ "completions/min_length": 231.0,
749
+ "completions/min_terminated_length": 231.0,
750
+ "epoch": 0.7104795737122558,
751
+ "grad_norm": 0.12807422733812146,
752
+ "kl": 0.020963287353515624,
753
+ "learning_rate": 1.1926874109063e-06,
754
+ "loss": 0.0023,
755
+ "num_tokens": 117693212.0,
756
+ "reward": 0.6751953840255738,
757
+ "reward_std": 0.17198645174503327,
758
+ "rewards/format_reward/mean": 0.9997395873069763,
759
+ "rewards/format_reward/std": 0.007216878235340118,
760
+ "rewards/mcq_accuracy_reward/mean": 0.4752604246139526,
761
+ "rewards/mcq_accuracy_reward/std": 0.49931321144104,
762
+ "rewards/tag_count_reward/mean": 0.999609375,
763
+ "rewards/tag_count_reward/std": 0.009444911777973176,
764
+ "step": 125
765
+ },
766
+ {
767
+ "clip_ratio/high_max": 0.0,
768
+ "clip_ratio/high_mean": 0.0,
769
+ "clip_ratio/low_mean": 0.0,
770
+ "clip_ratio/low_min": 0.0,
771
+ "clip_ratio/region_mean": 0.0,
772
+ "completions/clipped_ratio": 0.0,
773
+ "completions/max_length": 4637.4,
774
+ "completions/max_terminated_length": 4637.4,
775
+ "completions/mean_length": 1052.527099609375,
776
+ "completions/mean_terminated_length": 1052.527099609375,
777
+ "completions/min_length": 261.8,
778
+ "completions/min_terminated_length": 261.8,
779
+ "epoch": 0.738898756660746,
780
+ "grad_norm": 0.1465611288853758,
781
+ "kl": 0.021143341064453126,
782
+ "learning_rate": 9.863779168486797e-07,
783
+ "loss": 0.0095,
784
+ "num_tokens": 122313876.0,
785
+ "reward": 0.6882617950439454,
786
+ "reward_std": 0.16661006212234497,
787
+ "rewards/format_reward/mean": 1.0,
788
+ "rewards/format_reward/std": 0.0,
789
+ "rewards/mcq_accuracy_reward/mean": 0.48828125596046446,
790
+ "rewards/mcq_accuracy_reward/std": 0.4994440317153931,
791
+ "rewards/tag_count_reward/mean": 0.9998046875,
792
+ "rewards/tag_count_reward/std": 0.004354107566177845,
793
+ "step": 130
794
+ },
795
+ {
796
+ "clip_ratio/high_max": 0.0,
797
+ "clip_ratio/high_mean": 0.0,
798
+ "clip_ratio/low_mean": 0.0,
799
+ "clip_ratio/low_min": 0.0,
800
+ "clip_ratio/region_mean": 0.0,
801
+ "completions/clipped_ratio": 0.0,
802
+ "completions/max_length": 5632.6,
803
+ "completions/max_terminated_length": 5632.6,
804
+ "completions/mean_length": 1087.99873046875,
805
+ "completions/mean_terminated_length": 1087.99873046875,
806
+ "completions/min_length": 252.0,
807
+ "completions/min_terminated_length": 252.0,
808
+ "epoch": 0.7673179396092362,
809
+ "grad_norm": 0.1600247679302984,
810
+ "kl": 0.019998931884765626,
811
+ "learning_rate": 7.952073672272464e-07,
812
+ "loss": 0.0036,
813
+ "num_tokens": 127064943.0,
814
+ "reward": 0.6473828911781311,
815
+ "reward_std": 0.18547057807445527,
816
+ "rewards/format_reward/mean": 1.0,
817
+ "rewards/format_reward/std": 0.0,
818
+ "rewards/mcq_accuracy_reward/mean": 0.4473958373069763,
819
+ "rewards/mcq_accuracy_reward/std": 0.49676424264907837,
820
+ "rewards/tag_count_reward/mean": 0.9998697996139526,
821
+ "rewards/tag_count_reward/std": 0.003608439117670059,
822
+ "step": 135
823
+ },
824
+ {
825
+ "clip_ratio/high_max": 0.0,
826
+ "clip_ratio/high_mean": 0.0,
827
+ "clip_ratio/low_mean": 0.0,
828
+ "clip_ratio/low_min": 0.0,
829
+ "clip_ratio/region_mean": 0.0,
830
+ "completions/clipped_ratio": 0.0013020833333333259,
831
+ "completions/max_length": 6002.8,
832
+ "completions/max_terminated_length": 5989.2,
833
+ "completions/mean_length": 1068.9239868164063,
834
+ "completions/mean_terminated_length": 1059.5496948242187,
835
+ "completions/min_length": 238.2,
836
+ "completions/min_terminated_length": 238.2,
837
+ "epoch": 0.7957371225577264,
838
+ "grad_norm": 0.14840743240817333,
839
+ "kl": 0.0195098876953125,
840
+ "learning_rate": 6.210878115296267e-07,
841
+ "loss": -0.0015,
842
+ "num_tokens": 131744587.0,
843
+ "reward": 0.6677019000053406,
844
+ "reward_std": 0.1643421858549118,
845
+ "rewards/format_reward/mean": 0.9984375,
846
+ "rewards/format_reward/std": 0.023312175273895265,
847
+ "rewards/mcq_accuracy_reward/mean": 0.46796875,
848
+ "rewards/mcq_accuracy_reward/std": 0.49840737581253053,
849
+ "rewards/tag_count_reward/mean": 0.9988932251930237,
850
+ "rewards/tag_count_reward/std": 0.01567991152405739,
851
+ "step": 140
852
+ },
853
+ {
854
+ "clip_ratio/high_max": 0.0,
855
+ "clip_ratio/high_mean": 0.0,
856
+ "clip_ratio/low_mean": 0.0,
857
+ "clip_ratio/low_min": 0.0,
858
+ "clip_ratio/region_mean": 0.0,
859
+ "completions/clipped_ratio": 0.0,
860
+ "completions/max_length": 5690.0,
861
+ "completions/max_terminated_length": 5690.0,
862
+ "completions/mean_length": 1053.6370361328125,
863
+ "completions/mean_terminated_length": 1053.6370361328125,
864
+ "completions/min_length": 236.4,
865
+ "completions/min_terminated_length": 236.4,
866
+ "epoch": 0.8241563055062167,
867
+ "grad_norm": 0.16563780810558218,
868
+ "kl": 0.01880035400390625,
869
+ "learning_rate": 4.6576075862273445e-07,
870
+ "loss": 0.0026,
871
+ "num_tokens": 136367097.0,
872
+ "reward": 0.6843424916267395,
873
+ "reward_std": 0.17991138994693756,
874
+ "rewards/format_reward/mean": 1.0,
875
+ "rewards/format_reward/std": 0.0,
876
+ "rewards/mcq_accuracy_reward/mean": 0.48437499403953554,
877
+ "rewards/mcq_accuracy_reward/std": 0.4988525092601776,
878
+ "rewards/tag_count_reward/mean": 0.9996744871139527,
879
+ "rewards/tag_count_reward/std": 0.009021097794175148,
880
+ "step": 145
881
+ },
882
+ {
883
+ "clip_ratio/high_max": 0.0,
884
+ "clip_ratio/high_mean": 0.0,
885
+ "clip_ratio/low_mean": 0.0,
886
+ "clip_ratio/low_min": 0.0,
887
+ "clip_ratio/region_mean": 0.0,
888
+ "completions/clipped_ratio": 0.0,
889
+ "completions/max_length": 4469.6,
890
+ "completions/max_terminated_length": 4469.6,
891
+ "completions/mean_length": 1041.7265869140624,
892
+ "completions/mean_terminated_length": 1041.7265869140624,
893
+ "completions/min_length": 230.8,
894
+ "completions/min_terminated_length": 230.8,
895
+ "epoch": 0.8525754884547069,
896
+ "grad_norm": 0.14692099930627384,
897
+ "kl": 0.017971038818359375,
898
+ "learning_rate": 3.3077975853042704e-07,
899
+ "loss": 0.0001,
900
+ "num_tokens": 140943367.0,
901
+ "reward": 0.6906185746192932,
902
+ "reward_std": 0.1725286066532135,
903
+ "rewards/format_reward/mean": 1.0,
904
+ "rewards/format_reward/std": 0.0,
905
+ "rewards/mcq_accuracy_reward/mean": 0.49062501192092894,
906
+ "rewards/mcq_accuracy_reward/std": 0.4994231522083282,
907
+ "rewards/tag_count_reward/mean": 0.9999348998069764,
908
+ "rewards/tag_count_reward/std": 0.0018042195588350296,
909
+ "step": 150
910
+ },
911
+ {
912
+ "clip_ratio/high_max": 0.0,
913
+ "clip_ratio/high_mean": 0.0,
914
+ "clip_ratio/low_mean": 0.0,
915
+ "clip_ratio/low_min": 0.0,
916
+ "clip_ratio/region_mean": 0.0,
917
+ "completions/clipped_ratio": 0.0,
918
+ "completions/max_length": 4531.2,
919
+ "completions/max_terminated_length": 4531.2,
920
+ "completions/mean_length": 1035.3060302734375,
921
+ "completions/mean_terminated_length": 1035.3060302734375,
922
+ "completions/min_length": 257.0,
923
+ "completions/min_terminated_length": 257.0,
924
+ "epoch": 0.8809946714031972,
925
+ "grad_norm": 0.15437157339648844,
926
+ "kl": 0.018267059326171876,
927
+ "learning_rate": 2.174948641378244e-07,
928
+ "loss": 0.0147,
929
+ "num_tokens": 145497102.0,
930
+ "reward": 0.7106380820274353,
931
+ "reward_std": 0.17128031849861144,
932
+ "rewards/format_reward/mean": 0.9997395873069763,
933
+ "rewards/format_reward/std": 0.007216878235340118,
934
+ "rewards/mcq_accuracy_reward/mean": 0.5106770873069764,
935
+ "rewards/mcq_accuracy_reward/std": 0.4973214030265808,
936
+ "rewards/tag_count_reward/mean": 0.9998697996139526,
937
+ "rewards/tag_count_reward/std": 0.003608439117670059,
938
+ "step": 155
939
+ },
940
+ {
941
+ "clip_ratio/high_max": 0.0,
942
+ "clip_ratio/high_mean": 0.0,
943
+ "clip_ratio/low_mean": 0.0,
944
+ "clip_ratio/low_min": 0.0,
945
+ "clip_ratio/region_mean": 0.0,
946
+ "completions/clipped_ratio": 0.0,
947
+ "completions/max_length": 4793.6,
948
+ "completions/max_terminated_length": 4793.6,
949
+ "completions/mean_length": 1055.2679809570313,
950
+ "completions/mean_terminated_length": 1055.2679809570313,
951
+ "completions/min_length": 241.0,
952
+ "completions/min_terminated_length": 241.0,
953
+ "epoch": 0.9094138543516874,
954
+ "grad_norm": 0.15275141096890196,
955
+ "kl": 0.0176239013671875,
956
+ "learning_rate": 1.27039128233174e-07,
957
+ "loss": 0.0121,
958
+ "num_tokens": 150127331.0,
959
+ "reward": 0.7130144119262696,
960
+ "reward_std": 0.18018046617507935,
961
+ "rewards/format_reward/mean": 1.0,
962
+ "rewards/format_reward/std": 0.0,
963
+ "rewards/mcq_accuracy_reward/mean": 0.5130208253860473,
964
+ "rewards/mcq_accuracy_reward/std": 0.49882700443267824,
965
+ "rewards/tag_count_reward/mean": 0.9999348998069764,
966
+ "rewards/tag_count_reward/std": 0.0018042195588350296,
967
+ "step": 160
968
+ },
969
+ {
970
+ "clip_ratio/high_max": 0.0,
971
+ "clip_ratio/high_mean": 0.0,
972
+ "clip_ratio/low_mean": 0.0,
973
+ "clip_ratio/low_min": 0.0,
974
+ "clip_ratio/region_mean": 0.0,
975
+ "completions/clipped_ratio": 0.0002604166666666741,
976
+ "completions/max_length": 5728.6,
977
+ "completions/max_terminated_length": 5411.0,
978
+ "completions/mean_length": 1094.807080078125,
979
+ "completions/mean_terminated_length": 1092.956884765625,
980
+ "completions/min_length": 244.0,
981
+ "completions/min_terminated_length": 244.0,
982
+ "epoch": 0.9378330373001776,
983
+ "grad_norm": 0.1354642339017401,
984
+ "kl": 0.0183624267578125,
985
+ "learning_rate": 6.031727094116174e-08,
986
+ "loss": 0.0067,
987
+ "num_tokens": 154910862.0,
988
+ "reward": 0.6801628351211548,
989
+ "reward_std": 0.1546088457107544,
990
+ "rewards/format_reward/mean": 0.9997395873069763,
991
+ "rewards/format_reward/std": 0.007216878235340118,
992
+ "rewards/mcq_accuracy_reward/mean": 0.48020832538604735,
993
+ "rewards/mcq_accuracy_reward/std": 0.498971688747406,
994
+ "rewards/tag_count_reward/mean": 0.9998046875,
995
+ "rewards/tag_count_reward/std": 0.005412658676505089,
996
+ "step": 165
997
+ },
998
+ {
999
+ "clip_ratio/high_max": 0.0,
1000
+ "clip_ratio/high_mean": 0.0,
1001
+ "clip_ratio/low_mean": 0.0,
1002
+ "clip_ratio/low_min": 0.0,
1003
+ "clip_ratio/region_mean": 0.0,
1004
+ "completions/clipped_ratio": 0.0,
1005
+ "completions/max_length": 4582.0,
1006
+ "completions/max_terminated_length": 4582.0,
1007
+ "completions/mean_length": 1007.6719116210937,
1008
+ "completions/mean_terminated_length": 1007.6719116210937,
1009
+ "completions/min_length": 239.0,
1010
+ "completions/min_terminated_length": 239.0,
1011
+ "epoch": 0.9662522202486679,
1012
+ "grad_norm": 0.13334679057257176,
1013
+ "kl": 0.018311309814453124,
1014
+ "learning_rate": 1.7996630893712675e-08,
1015
+ "loss": 0.0029,
1016
+ "num_tokens": 159352714.0,
1017
+ "reward": 0.7252474665641785,
1018
+ "reward_std": 0.1664626866579056,
1019
+ "rewards/format_reward/mean": 1.0,
1020
+ "rewards/format_reward/std": 0.0,
1021
+ "rewards/mcq_accuracy_reward/mean": 0.5252604126930237,
1022
+ "rewards/mcq_accuracy_reward/std": 0.4992336809635162,
1023
+ "rewards/tag_count_reward/mean": 0.9998697876930237,
1024
+ "rewards/tag_count_reward/std": 0.003608439117670059,
1025
+ "step": 170
1026
+ },
1027
+ {
1028
+ "clip_ratio/high_max": 0.0,
1029
+ "clip_ratio/high_mean": 0.0,
1030
+ "clip_ratio/low_mean": 0.0,
1031
+ "clip_ratio/low_min": 0.0,
1032
+ "clip_ratio/region_mean": 0.0,
1033
+ "completions/clipped_ratio": 0.0002604166666666741,
1034
+ "completions/max_length": 5149.6,
1035
+ "completions/max_terminated_length": 4739.2,
1036
+ "completions/mean_length": 1035.9330444335938,
1037
+ "completions/mean_terminated_length": 1034.0750610351563,
1038
+ "completions/min_length": 250.0,
1039
+ "completions/min_terminated_length": 250.0,
1040
+ "epoch": 0.9946714031971581,
1041
+ "grad_norm": 0.14564780944336908,
1042
+ "kl": 0.018206024169921876,
1043
+ "learning_rate": 5.004906428685008e-10,
1044
+ "loss": 0.0067,
1045
+ "num_tokens": 163907551.0,
1046
+ "reward": 0.6884701490402222,
1047
+ "reward_std": 0.1888585329055786,
1048
+ "rewards/format_reward/mean": 0.9997395873069763,
1049
+ "rewards/format_reward/std": 0.007216878235340118,
1050
+ "rewards/mcq_accuracy_reward/mean": 0.48854167461395265,
1051
+ "rewards/mcq_accuracy_reward/std": 0.49964500069618223,
1052
+ "rewards/tag_count_reward/mean": 0.9995442748069763,
1053
+ "rewards/tag_count_reward/std": 0.012629536911845207,
1054
+ "step": 175
1055
+ },
1056
+ {
1057
+ "epoch": 0.9946714031971581,
1058
+ "step": 175,
1059
+ "total_flos": 0.0,
1060
+ "train_loss": 22251.08962759373,
1061
+ "train_runtime": 39419.1987,
1062
+ "train_samples_per_second": 0.428,
1063
+ "train_steps_per_second": 0.004
1064
+ }
1065
+ ],
1066
+ "logging_steps": 5,
1067
+ "max_steps": 175,
1068
+ "num_input_tokens_seen": 163907551,
1069
+ "num_train_epochs": 1,
1070
+ "save_steps": 50,
1071
+ "stateful_callbacks": {
1072
+ "TrainerControl": {
1073
+ "args": {
1074
+ "should_epoch_stop": false,
1075
+ "should_evaluate": false,
1076
+ "should_log": false,
1077
+ "should_save": true,
1078
+ "should_training_stop": true
1079
+ },
1080
+ "attributes": {}
1081
+ }
1082
+ },
1083
+ "total_flos": 0.0,
1084
+ "train_batch_size": 8,
1085
+ "trial_name": null,
1086
+ "trial_params": null
1087
+ }