LLucass commited on
Commit
a8f925a
·
verified ·
1 Parent(s): 91aaa73

Model save

Browse files
README.md ADDED
@@ -0,0 +1,68 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ base_model: deepseek-ai/DeepSeek-R1-Distill-Qwen-1.5B
3
+ library_name: transformers
4
+ model_name: Gaussian
5
+ tags:
6
+ - generated_from_trainer
7
+ - trl
8
+ - grpo
9
+ licence: license
10
+ ---
11
+
12
+ # Model Card for Gaussian
13
+
14
+ This model is a fine-tuned version of [deepseek-ai/DeepSeek-R1-Distill-Qwen-1.5B](https://huggingface.co/deepseek-ai/DeepSeek-R1-Distill-Qwen-1.5B).
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="LLucass/Gaussian", 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/lavatorywang-nus/openr1/runs/wuysuu4f)
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
38
+ - Transformers: 4.50.0
39
+ - Pytorch: 2.5.1
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 Gallouédec},
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": -0.004574703625403345,
4
+ "train_runtime": 8361.7209,
5
+ "train_samples": 7000,
6
+ "train_samples_per_second": 1.148,
7
+ "train_steps_per_second": 0.012
8
+ }
generation_config.json ADDED
@@ -0,0 +1,9 @@
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "_from_model_config": true,
3
+ "bos_token_id": 151646,
4
+ "do_sample": true,
5
+ "eos_token_id": 151643,
6
+ "temperature": 0.6,
7
+ "top_p": 0.95,
8
+ "transformers_version": "4.50.0"
9
+ }
train_results.json ADDED
@@ -0,0 +1,8 @@
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "total_flos": 0.0,
3
+ "train_loss": -0.004574703625403345,
4
+ "train_runtime": 8361.7209,
5
+ "train_samples": 7000,
6
+ "train_samples_per_second": 1.148,
7
+ "train_steps_per_second": 0.012
8
+ }
trainer_state.json ADDED
@@ -0,0 +1,2243 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "best_global_step": null,
3
+ "best_metric": null,
4
+ "best_model_checkpoint": null,
5
+ "epoch": 0.11428571428571428,
6
+ "eval_steps": 500,
7
+ "global_step": 100,
8
+ "is_hyper_param_search": false,
9
+ "is_local_process_zero": true,
10
+ "is_world_process_zero": true,
11
+ "log_history": [
12
+ {
13
+ "completion_length": 2700.4271850585938,
14
+ "cov_mean": -6.0587970438064076e-05,
15
+ "cov_std": 0.35307812318205833,
16
+ "entropy": 0.3681640625,
17
+ "epoch": 0.001142857142857143,
18
+ "grad_norm": 0.45492634177207947,
19
+ "kl": 0.0,
20
+ "learning_rate": 1e-07,
21
+ "loss": -0.0382,
22
+ "reward": 0.7604166893288493,
23
+ "reward_std": 0.4268697127699852,
24
+ "rewards/accuracy_reward": 0.25000001303851604,
25
+ "rewards/format_reward": 0.5104166669771075,
26
+ "step": 1,
27
+ "w_high_ratio": 0.2200421690940857,
28
+ "w_low_ratio": 0.03663695091381669,
29
+ "w_max": 2.1593789756298065,
30
+ "w_mean": 1.4711343348026276,
31
+ "w_min": 6.525355682266089e-35,
32
+ "w_std": 0.2659660503268242
33
+ },
34
+ {
35
+ "completion_length": 3127.3958435058594,
36
+ "cov_mean": -2.155053698515985e-05,
37
+ "cov_std": 0.310540571808815,
38
+ "entropy": 0.353515625,
39
+ "epoch": 0.002285714285714286,
40
+ "grad_norm": 0.5142282843589783,
41
+ "kl": 0.0,
42
+ "learning_rate": 2e-07,
43
+ "loss": 0.0049,
44
+ "reward": 0.6458333637565374,
45
+ "reward_std": 0.4249730706214905,
46
+ "rewards/accuracy_reward": 0.2812500102445483,
47
+ "rewards/format_reward": 0.3645833386108279,
48
+ "step": 2,
49
+ "w_high_ratio": 0.05183619633316994,
50
+ "w_low_ratio": 0.036958135198801756,
51
+ "w_max": 1.8325217366218567,
52
+ "w_mean": 1.2113382518291473,
53
+ "w_min": 0.0,
54
+ "w_std": 0.20957503467798233
55
+ },
56
+ {
57
+ "completion_length": 3733.9583740234375,
58
+ "cov_mean": -2.010253456319333e-05,
59
+ "cov_std": 0.2787674851715565,
60
+ "entropy": 0.451171875,
61
+ "epoch": 0.0034285714285714284,
62
+ "grad_norm": 0.28526851534843445,
63
+ "kl": 4.315376281738281e-05,
64
+ "learning_rate": 3e-07,
65
+ "loss": 0.0391,
66
+ "reward": 0.16666667070239782,
67
+ "reward_std": 0.31075509265065193,
68
+ "rewards/accuracy_reward": 0.031250000931322575,
69
+ "rewards/format_reward": 0.13541666697710752,
70
+ "step": 3,
71
+ "w_high_ratio": 0.0,
72
+ "w_low_ratio": 0.03727753087878227,
73
+ "w_max": 1.426508367061615,
74
+ "w_mean": 1.069275826215744,
75
+ "w_min": 0.0,
76
+ "w_std": 0.17276715114712715
77
+ },
78
+ {
79
+ "completion_length": 2456.5104370117188,
80
+ "cov_mean": 6.050579213479068e-05,
81
+ "cov_std": 0.4746721386909485,
82
+ "entropy": 0.404296875,
83
+ "epoch": 0.004571428571428572,
84
+ "grad_norm": 0.4348958432674408,
85
+ "kl": 3.764033317565918e-05,
86
+ "learning_rate": 4e-07,
87
+ "loss": -0.0128,
88
+ "reward": 0.8125000223517418,
89
+ "reward_std": 0.4816601872444153,
90
+ "rewards/accuracy_reward": 0.1354166679084301,
91
+ "rewards/format_reward": 0.677083358168602,
92
+ "step": 4,
93
+ "w_high_ratio": 0.09904679656028748,
94
+ "w_low_ratio": 0.03708651335909963,
95
+ "w_max": 2.0615443289279938,
96
+ "w_mean": 1.430301010608673,
97
+ "w_min": 2.2357144385761694e-24,
98
+ "w_std": 0.25941959768533707
99
+ },
100
+ {
101
+ "completion_length": 3495.9584350585938,
102
+ "cov_mean": -0.00011635808687060489,
103
+ "cov_std": 0.43240153789520264,
104
+ "entropy": 0.4658203125,
105
+ "epoch": 0.005714285714285714,
106
+ "grad_norm": 0.44890686869621277,
107
+ "kl": 4.1797757148742676e-05,
108
+ "learning_rate": 5e-07,
109
+ "loss": -0.077,
110
+ "reward": 0.38541667722165585,
111
+ "reward_std": 0.41880372166633606,
112
+ "rewards/accuracy_reward": 0.0416666679084301,
113
+ "rewards/format_reward": 0.34375001303851604,
114
+ "step": 5,
115
+ "w_high_ratio": 0.057075776159763336,
116
+ "w_low_ratio": 0.04613090865314007,
117
+ "w_max": 1.8747033178806305,
118
+ "w_mean": 1.2025023698806763,
119
+ "w_min": 2.0185204017920427e-32,
120
+ "w_std": 0.24863697960972786
121
+ },
122
+ {
123
+ "completion_length": 3429.7604370117188,
124
+ "cov_mean": 7.305060944418074e-05,
125
+ "cov_std": 0.3747733160853386,
126
+ "entropy": 0.4658203125,
127
+ "epoch": 0.006857142857142857,
128
+ "grad_norm": 0.323794424533844,
129
+ "kl": 4.607439041137695e-05,
130
+ "learning_rate": 6e-07,
131
+ "loss": 0.0289,
132
+ "reward": 0.4791666828095913,
133
+ "reward_std": 0.4919559434056282,
134
+ "rewards/accuracy_reward": 0.1666666716337204,
135
+ "rewards/format_reward": 0.3125000111758709,
136
+ "step": 6,
137
+ "w_high_ratio": 0.0,
138
+ "w_low_ratio": 0.045196511782705784,
139
+ "w_max": 1.4718832075595856,
140
+ "w_mean": 1.11675626039505,
141
+ "w_min": 0.25,
142
+ "w_std": 0.20970172435045242
143
+ },
144
+ {
145
+ "completion_length": 3298.1875610351562,
146
+ "cov_mean": 1.544893893878907e-05,
147
+ "cov_std": 0.5214613378047943,
148
+ "entropy": 0.3837890625,
149
+ "epoch": 0.008,
150
+ "grad_norm": 0.3830799460411072,
151
+ "kl": 2.524256706237793e-05,
152
+ "learning_rate": 7e-07,
153
+ "loss": -0.0429,
154
+ "reward": 0.802083358168602,
155
+ "reward_std": 0.5995434895157814,
156
+ "rewards/accuracy_reward": 0.2083333432674408,
157
+ "rewards/format_reward": 0.5937500223517418,
158
+ "step": 7,
159
+ "w_high_ratio": 0.10538750886917114,
160
+ "w_low_ratio": 0.0529888030141592,
161
+ "w_max": 1.5837838649749756,
162
+ "w_mean": 1.266415923833847,
163
+ "w_min": 1.0439673559219887e-43,
164
+ "w_std": 0.28801921010017395
165
+ },
166
+ {
167
+ "completion_length": 2903.697998046875,
168
+ "cov_mean": -8.183626050595194e-06,
169
+ "cov_std": 0.2509700208902359,
170
+ "entropy": 0.355712890625,
171
+ "epoch": 0.009142857142857144,
172
+ "grad_norm": 0.2550622820854187,
173
+ "kl": 1.93864107131958e-05,
174
+ "learning_rate": 8e-07,
175
+ "loss": -0.0211,
176
+ "reward": 0.7812500298023224,
177
+ "reward_std": 0.3037422299385071,
178
+ "rewards/accuracy_reward": 0.3333333432674408,
179
+ "rewards/format_reward": 0.4479166865348816,
180
+ "step": 8,
181
+ "w_high_ratio": 0.152647502720356,
182
+ "w_low_ratio": 0.020135570783168077,
183
+ "w_max": 1.6893351376056671,
184
+ "w_mean": 1.325122356414795,
185
+ "w_min": 0.5,
186
+ "w_std": 0.13287343829870224
187
+ },
188
+ {
189
+ "completion_length": 3427.0001220703125,
190
+ "cov_mean": 2.1584107344096992e-05,
191
+ "cov_std": 0.41377831995487213,
192
+ "entropy": 0.4404296875,
193
+ "epoch": 0.010285714285714285,
194
+ "grad_norm": 0.36070960760116577,
195
+ "kl": 4.297494888305664e-05,
196
+ "learning_rate": 9e-07,
197
+ "loss": -0.0144,
198
+ "reward": 0.4375000149011612,
199
+ "reward_std": 0.5150812789797783,
200
+ "rewards/accuracy_reward": 0.11458333395421505,
201
+ "rewards/format_reward": 0.322916679084301,
202
+ "step": 9,
203
+ "w_high_ratio": 0.0,
204
+ "w_low_ratio": 0.053231239318847656,
205
+ "w_max": 1.665610283613205,
206
+ "w_mean": 1.168578028678894,
207
+ "w_min": 0.0,
208
+ "w_std": 0.2691279575228691
209
+ },
210
+ {
211
+ "completion_length": 2921.260498046875,
212
+ "cov_mean": -0.00011126142680950579,
213
+ "cov_std": 0.37009984999895096,
214
+ "entropy": 0.34130859375,
215
+ "epoch": 0.011428571428571429,
216
+ "grad_norm": 0.463008314371109,
217
+ "kl": 2.7805566787719727e-05,
218
+ "learning_rate": 1e-06,
219
+ "loss": -0.0304,
220
+ "reward": 0.604166679084301,
221
+ "reward_std": 0.49994633346796036,
222
+ "rewards/accuracy_reward": 0.19791667349636555,
223
+ "rewards/format_reward": 0.4062500111758709,
224
+ "step": 10,
225
+ "w_high_ratio": 0.10105682164430618,
226
+ "w_low_ratio": 0.039817056618630886,
227
+ "w_max": 2.203812062740326,
228
+ "w_mean": 1.3322690725326538,
229
+ "w_min": 0.0,
230
+ "w_std": 0.24996469356119633
231
+ },
232
+ {
233
+ "completion_length": 3739.4583740234375,
234
+ "cov_mean": 1.6300052948281518e-05,
235
+ "cov_std": 0.22599393874406815,
236
+ "entropy": 0.3876953125,
237
+ "epoch": 0.012571428571428572,
238
+ "grad_norm": 0.2399490475654602,
239
+ "kl": 2.893805503845215e-05,
240
+ "learning_rate": 9.997258721585931e-07,
241
+ "loss": 0.0375,
242
+ "reward": 0.1979166716337204,
243
+ "reward_std": 0.3496348336338997,
244
+ "rewards/accuracy_reward": 0.07291666697710752,
245
+ "rewards/format_reward": 0.12500000558793545,
246
+ "step": 11,
247
+ "w_high_ratio": 0.0,
248
+ "w_low_ratio": 0.03310772683471441,
249
+ "w_max": 1.3216401040554047,
250
+ "w_mean": 1.05658358335495,
251
+ "w_min": 0.25,
252
+ "w_std": 0.13504411466419697
253
+ },
254
+ {
255
+ "completion_length": 2672.3959045410156,
256
+ "cov_mean": -1.72590916918125e-05,
257
+ "cov_std": 0.4432706832885742,
258
+ "entropy": 0.38525390625,
259
+ "epoch": 0.013714285714285714,
260
+ "grad_norm": 0.5923348665237427,
261
+ "kl": 3.87728214263916e-05,
262
+ "learning_rate": 9.989038226169207e-07,
263
+ "loss": -0.0398,
264
+ "reward": 0.7916666865348816,
265
+ "reward_std": 0.44830071553587914,
266
+ "rewards/accuracy_reward": 0.15625,
267
+ "rewards/format_reward": 0.6354166865348816,
268
+ "step": 12,
269
+ "w_high_ratio": 0.18827838450670242,
270
+ "w_low_ratio": 0.039877891540527344,
271
+ "w_max": 2.278617113828659,
272
+ "w_mean": 1.533039778470993,
273
+ "w_min": 0.0,
274
+ "w_std": 0.30229373276233673
275
+ },
276
+ {
277
+ "completion_length": 3129.9168090820312,
278
+ "cov_mean": 5.8586666455084924e-05,
279
+ "cov_std": 0.3728199526667595,
280
+ "entropy": 0.3896484375,
281
+ "epoch": 0.014857142857142857,
282
+ "grad_norm": 0.7794428467750549,
283
+ "kl": 2.4527311325073242e-05,
284
+ "learning_rate": 9.975348529157229e-07,
285
+ "loss": -0.0268,
286
+ "reward": 0.6145833507180214,
287
+ "reward_std": 0.3987155184149742,
288
+ "rewards/accuracy_reward": 0.1979166716337204,
289
+ "rewards/format_reward": 0.4166666716337204,
290
+ "step": 13,
291
+ "w_high_ratio": 0.1489243432879448,
292
+ "w_low_ratio": 0.03552113357000053,
293
+ "w_max": 1.913963943719864,
294
+ "w_mean": 1.3093808591365814,
295
+ "w_min": 0.0,
296
+ "w_std": 0.22065079398453236
297
+ },
298
+ {
299
+ "completion_length": 2925.8750610351562,
300
+ "cov_mean": 7.129182449716609e-06,
301
+ "cov_std": 0.3760656714439392,
302
+ "entropy": 0.353515625,
303
+ "epoch": 0.016,
304
+ "grad_norm": 0.6446069478988647,
305
+ "kl": 2.3916363716125488e-05,
306
+ "learning_rate": 9.956206309337066e-07,
307
+ "loss": -0.0123,
308
+ "reward": 0.6562500298023224,
309
+ "reward_std": 0.41464143246412277,
310
+ "rewards/accuracy_reward": 0.17708334233611822,
311
+ "rewards/format_reward": 0.4791666716337204,
312
+ "step": 14,
313
+ "w_high_ratio": 0.04337232932448387,
314
+ "w_low_ratio": 0.04529819544404745,
315
+ "w_max": 1.892830729484558,
316
+ "w_mean": 1.2675927579402924,
317
+ "w_min": 7.264961823368798e-41,
318
+ "w_std": 0.26389942318201065
319
+ },
320
+ {
321
+ "completion_length": 2977.6458740234375,
322
+ "cov_mean": 2.068816957034869e-05,
323
+ "cov_std": 0.2604588344693184,
324
+ "entropy": 0.3759765625,
325
+ "epoch": 0.017142857142857144,
326
+ "grad_norm": 0.4150949716567993,
327
+ "kl": 3.0860304832458496e-05,
328
+ "learning_rate": 9.931634888554935e-07,
329
+ "loss": -0.0032,
330
+ "reward": 0.6041666902601719,
331
+ "reward_std": 0.25453949719667435,
332
+ "rewards/accuracy_reward": 0.20833333861082792,
333
+ "rewards/format_reward": 0.3958333358168602,
334
+ "step": 15,
335
+ "w_high_ratio": 0.0,
336
+ "w_low_ratio": 0.027457581600174308,
337
+ "w_max": 1.3896198272705078,
338
+ "w_mean": 1.1120553016662598,
339
+ "w_min": 0.0,
340
+ "w_std": 0.1376371644437313
341
+ },
342
+ {
343
+ "completion_length": 3860.1771850585938,
344
+ "cov_mean": 3.589540312987083e-05,
345
+ "cov_std": 0.1900031939148903,
346
+ "entropy": 0.45751953125,
347
+ "epoch": 0.018285714285714287,
348
+ "grad_norm": 0.25400859117507935,
349
+ "kl": 3.0994415283203125e-05,
350
+ "learning_rate": 9.901664203302124e-07,
351
+ "loss": 0.0468,
352
+ "reward": 0.1458333395421505,
353
+ "reward_std": 0.24955713748931885,
354
+ "rewards/accuracy_reward": 0.06250000279396772,
355
+ "rewards/format_reward": 0.08333333674818277,
356
+ "step": 16,
357
+ "w_high_ratio": 0.0,
358
+ "w_low_ratio": 0.028868926223367453,
359
+ "w_max": 1.2315952479839325,
360
+ "w_mean": 1.0204393565654755,
361
+ "w_min": 0.25,
362
+ "w_std": 0.10705284960567951
363
+ },
364
+ {
365
+ "completion_length": 2540.4895935058594,
366
+ "cov_mean": -0.0001341435610129338,
367
+ "cov_std": 0.47027114033699036,
368
+ "entropy": 0.45166015625,
369
+ "epoch": 0.019428571428571427,
370
+ "grad_norm": 0.46505773067474365,
371
+ "kl": 5.486607551574707e-05,
372
+ "learning_rate": 9.866330768241983e-07,
373
+ "loss": -0.0001,
374
+ "reward": 0.812500037252903,
375
+ "reward_std": 0.49661170691251755,
376
+ "rewards/accuracy_reward": 0.2291666716337204,
377
+ "rewards/format_reward": 0.5833333507180214,
378
+ "step": 17,
379
+ "w_high_ratio": 0.15490177273750305,
380
+ "w_low_ratio": 0.04533949866890907,
381
+ "w_max": 2.4269833266735077,
382
+ "w_mean": 1.407887727022171,
383
+ "w_min": 6.344169701078957e-36,
384
+ "w_std": 0.2766948938369751
385
+ },
386
+ {
387
+ "completion_length": 3100.604248046875,
388
+ "cov_mean": 9.316997420683037e-06,
389
+ "cov_std": 0.345074113458395,
390
+ "entropy": 0.36865234375,
391
+ "epoch": 0.02057142857142857,
392
+ "grad_norm": 0.26780965924263,
393
+ "kl": 3.306567668914795e-05,
394
+ "learning_rate": 9.825677631722435e-07,
395
+ "loss": 0.0093,
396
+ "reward": 0.5208333507180214,
397
+ "reward_std": 0.4629998616874218,
398
+ "rewards/accuracy_reward": 0.11458333861082792,
399
+ "rewards/format_reward": 0.4062500074505806,
400
+ "step": 18,
401
+ "w_high_ratio": 0.12114312499761581,
402
+ "w_low_ratio": 0.046439004596322775,
403
+ "w_max": 2.0130829215049744,
404
+ "w_mean": 1.2991048991680145,
405
+ "w_min": 0.0,
406
+ "w_std": 0.22966066375374794
407
+ },
408
+ {
409
+ "completion_length": 3111.229248046875,
410
+ "cov_mean": -1.3287149158713873e-10,
411
+ "cov_std": 0.305859737098217,
412
+ "entropy": 0.37939453125,
413
+ "epoch": 0.021714285714285714,
414
+ "grad_norm": 0.2747480869293213,
415
+ "kl": 4.106760025024414e-05,
416
+ "learning_rate": 9.779754323328192e-07,
417
+ "loss": -0.0265,
418
+ "reward": 0.8437500251457095,
419
+ "reward_std": 0.44304458051919937,
420
+ "rewards/accuracy_reward": 0.3854166818782687,
421
+ "rewards/format_reward": 0.4583333535119891,
422
+ "step": 19,
423
+ "w_high_ratio": 0.007512019481509924,
424
+ "w_low_ratio": 0.03380460711196065,
425
+ "w_max": 1.7184137105941772,
426
+ "w_mean": 1.2587754130363464,
427
+ "w_min": 0.0,
428
+ "w_std": 0.17892248183488846
429
+ },
430
+ {
431
+ "completion_length": 2602.0000610351562,
432
+ "cov_mean": -4.431823981576599e-06,
433
+ "cov_std": 0.3394951745867729,
434
+ "entropy": 0.334716796875,
435
+ "epoch": 0.022857142857142857,
436
+ "grad_norm": 0.2722320854663849,
437
+ "kl": 7.30752944946289e-05,
438
+ "learning_rate": 9.728616793536587e-07,
439
+ "loss": -0.016,
440
+ "reward": 0.989583358168602,
441
+ "reward_std": 0.41970477253198624,
442
+ "rewards/accuracy_reward": 0.2812500074505806,
443
+ "rewards/format_reward": 0.708333358168602,
444
+ "step": 20,
445
+ "w_high_ratio": 0.053037188947200775,
446
+ "w_low_ratio": 0.03638318553566933,
447
+ "w_max": 1.8807734847068787,
448
+ "w_mean": 1.3341735899448395,
449
+ "w_min": 0.28736552596092224,
450
+ "w_std": 0.21230074390769005
451
+ },
452
+ {
453
+ "completion_length": 2938.9896240234375,
454
+ "cov_mean": 1.115940267482074e-05,
455
+ "cov_std": 0.3257220685482025,
456
+ "entropy": 0.42919921875,
457
+ "epoch": 0.024,
458
+ "grad_norm": 0.9565598964691162,
459
+ "kl": 0.00023573637008666992,
460
+ "learning_rate": 9.672327345550543e-07,
461
+ "loss": 0.0067,
462
+ "reward": 0.604166692122817,
463
+ "reward_std": 0.4190382584929466,
464
+ "rewards/accuracy_reward": 0.17708333674818277,
465
+ "rewards/format_reward": 0.4270833460614085,
466
+ "step": 21,
467
+ "w_high_ratio": 0.2210865318775177,
468
+ "w_low_ratio": 0.03242270671762526,
469
+ "w_max": 1.8820680975914001,
470
+ "w_mean": 1.4674345552921295,
471
+ "w_min": 2.9579121420725567e-37,
472
+ "w_std": 0.23061984032392502
473
+ },
474
+ {
475
+ "completion_length": 1817.8125305175781,
476
+ "cov_mean": 3.471795207588002e-05,
477
+ "cov_std": 0.4554404318332672,
478
+ "entropy": 0.37548828125,
479
+ "epoch": 0.025142857142857144,
480
+ "grad_norm": 0.6403908729553223,
481
+ "kl": 0.0001913309097290039,
482
+ "learning_rate": 9.610954559391704e-07,
483
+ "loss": -0.0476,
484
+ "reward": 1.1041667014360428,
485
+ "reward_std": 0.36877378076314926,
486
+ "rewards/accuracy_reward": 0.28125000558793545,
487
+ "rewards/format_reward": 0.8229166716337204,
488
+ "step": 22,
489
+ "w_high_ratio": 0.3237294889986515,
490
+ "w_low_ratio": 0.03283138480037451,
491
+ "w_max": 3.265183061361313,
492
+ "w_mean": 1.977644294500351,
493
+ "w_min": 5.22007475166266e-30,
494
+ "w_std": 0.34112248569726944
495
+ },
496
+ {
497
+ "completion_length": 2787.4896850585938,
498
+ "cov_mean": 3.710087821673369e-05,
499
+ "cov_std": 0.3974616304039955,
500
+ "entropy": 0.38525390625,
501
+ "epoch": 0.026285714285714287,
502
+ "grad_norm": 0.6307149529457092,
503
+ "kl": 0.00013697147369384766,
504
+ "learning_rate": 9.54457320834625e-07,
505
+ "loss": -0.0959,
506
+ "reward": 0.6041666716337204,
507
+ "reward_std": 0.4507869631052017,
508
+ "rewards/accuracy_reward": 0.14583333395421505,
509
+ "rewards/format_reward": 0.4583333358168602,
510
+ "step": 23,
511
+ "w_high_ratio": 0.18567950278520584,
512
+ "w_low_ratio": 0.042684963904321194,
513
+ "w_max": 2.490270048379898,
514
+ "w_mean": 1.4617709517478943,
515
+ "w_min": 1.9980297628321344e-34,
516
+ "w_std": 0.30704520270228386
517
+ },
518
+ {
519
+ "completion_length": 3070.3125610351562,
520
+ "cov_mean": 5.4191557865124196e-05,
521
+ "cov_std": 0.5094943791627884,
522
+ "entropy": 0.367431640625,
523
+ "epoch": 0.027428571428571427,
524
+ "grad_norm": 0.5826663374900818,
525
+ "kl": 0.0001633763313293457,
526
+ "learning_rate": 9.473264167865171e-07,
527
+ "loss": -0.02,
528
+ "reward": 0.8229166939854622,
529
+ "reward_std": 0.5960981249809265,
530
+ "rewards/accuracy_reward": 0.2708333358168602,
531
+ "rewards/format_reward": 0.5520833507180214,
532
+ "step": 24,
533
+ "w_high_ratio": 0.16660767421126366,
534
+ "w_low_ratio": 0.05264063738286495,
535
+ "w_max": 2.2747649252414703,
536
+ "w_mean": 1.4129657745361328,
537
+ "w_min": 0.0,
538
+ "w_std": 0.3425128310918808
539
+ },
540
+ {
541
+ "completion_length": 3037.8125,
542
+ "cov_mean": -3.0258611332101282e-05,
543
+ "cov_std": 0.2974156364798546,
544
+ "entropy": 0.4462890625,
545
+ "epoch": 0.02857142857142857,
546
+ "grad_norm": 0.33458662033081055,
547
+ "kl": 0.00022605061531066895,
548
+ "learning_rate": 9.397114317029974e-07,
549
+ "loss": 0.0202,
550
+ "reward": 0.6875000149011612,
551
+ "reward_std": 0.4009811282157898,
552
+ "rewards/accuracy_reward": 0.2395833432674408,
553
+ "rewards/format_reward": 0.4479166865348816,
554
+ "step": 25,
555
+ "w_high_ratio": 0.1122930608689785,
556
+ "w_low_ratio": 0.034513187827542424,
557
+ "w_max": 1.991903692483902,
558
+ "w_mean": 1.2860748171806335,
559
+ "w_min": 0.25,
560
+ "w_std": 0.2155727706849575
561
+ },
562
+ {
563
+ "completion_length": 3117.4584350585938,
564
+ "cov_mean": -8.455652073280362e-05,
565
+ "cov_std": 0.23132415860891342,
566
+ "entropy": 0.42626953125,
567
+ "epoch": 0.029714285714285714,
568
+ "grad_norm": 0.41856279969215393,
569
+ "kl": 6.198883056640625e-05,
570
+ "learning_rate": 9.316216432703916e-07,
571
+ "loss": -0.053,
572
+ "reward": 0.6562500074505806,
573
+ "reward_std": 0.20796867460012436,
574
+ "rewards/accuracy_reward": 0.1979166716337204,
575
+ "rewards/format_reward": 0.4583333432674408,
576
+ "step": 26,
577
+ "w_high_ratio": 0.04478337988257408,
578
+ "w_low_ratio": 0.011211750796064734,
579
+ "w_max": 1.7470227479934692,
580
+ "w_mean": 1.224214792251587,
581
+ "w_min": 0.25,
582
+ "w_std": 0.10643218830227852
583
+ },
584
+ {
585
+ "completion_length": 3314.1041870117188,
586
+ "cov_mean": 2.3687035081820795e-05,
587
+ "cov_std": 0.44449782371520996,
588
+ "entropy": 0.44921875,
589
+ "epoch": 0.030857142857142857,
590
+ "grad_norm": 0.3447716236114502,
591
+ "kl": 0.0005403757095336914,
592
+ "learning_rate": 9.230669076497687e-07,
593
+ "loss": -0.0255,
594
+ "reward": 0.6354166865348816,
595
+ "reward_std": 0.5543549209833145,
596
+ "rewards/accuracy_reward": 0.1875000037252903,
597
+ "rewards/format_reward": 0.447916679084301,
598
+ "step": 27,
599
+ "w_high_ratio": 0.04345417395234108,
600
+ "w_low_ratio": 0.047908838372677565,
601
+ "w_max": 1.680276095867157,
602
+ "w_mean": 1.1971288323402405,
603
+ "w_min": 0.0,
604
+ "w_std": 0.23072999715805054
605
+ },
606
+ {
607
+ "completion_length": 3026.5626220703125,
608
+ "cov_mean": 4.19466396124335e-05,
609
+ "cov_std": 0.4268927723169327,
610
+ "entropy": 0.4091796875,
611
+ "epoch": 0.032,
612
+ "grad_norm": 0.4165506958961487,
613
+ "kl": 0.00021666288375854492,
614
+ "learning_rate": 9.140576474687263e-07,
615
+ "loss": -0.0277,
616
+ "reward": 0.8437500298023224,
617
+ "reward_std": 0.4840777963399887,
618
+ "rewards/accuracy_reward": 0.3750000149011612,
619
+ "rewards/format_reward": 0.4687500149011612,
620
+ "step": 28,
621
+ "w_high_ratio": 0.09859117493033409,
622
+ "w_low_ratio": 0.04558967752382159,
623
+ "w_max": 2.2461937069892883,
624
+ "w_mean": 1.3320654332637787,
625
+ "w_min": 0.0,
626
+ "w_std": 0.2779119908809662
627
+ },
628
+ {
629
+ "completion_length": 3620.1563110351562,
630
+ "cov_mean": -3.5699086538443225e-05,
631
+ "cov_std": 0.34214310348033905,
632
+ "entropy": 0.43017578125,
633
+ "epoch": 0.03314285714285714,
634
+ "grad_norm": 0.3975992202758789,
635
+ "kl": 0.0004680156707763672,
636
+ "learning_rate": 9.046048391230247e-07,
637
+ "loss": 0.0814,
638
+ "reward": 0.2291666716337204,
639
+ "reward_std": 0.39027639478445053,
640
+ "rewards/accuracy_reward": 0.0625,
641
+ "rewards/format_reward": 0.1666666716337204,
642
+ "step": 29,
643
+ "w_high_ratio": 0.0,
644
+ "w_low_ratio": 0.04788992367684841,
645
+ "w_max": 1.3650976717472076,
646
+ "w_mean": 1.0994165241718292,
647
+ "w_min": 0.25,
648
+ "w_std": 0.22099098190665245
649
+ },
650
+ {
651
+ "completion_length": 3238.17724609375,
652
+ "cov_mean": 5.147313459019642e-06,
653
+ "cov_std": 0.466504842042923,
654
+ "entropy": 0.38134765625,
655
+ "epoch": 0.03428571428571429,
656
+ "grad_norm": 0.3571697175502777,
657
+ "kl": 0.0008358955383300781,
658
+ "learning_rate": 8.9471999940354e-07,
659
+ "loss": 0.0066,
660
+ "reward": 0.7395833693444729,
661
+ "reward_std": 0.6347565650939941,
662
+ "rewards/accuracy_reward": 0.2708333423361182,
663
+ "rewards/format_reward": 0.4687500074505806,
664
+ "step": 30,
665
+ "w_high_ratio": 0.09309378266334534,
666
+ "w_low_ratio": 0.04863026458770037,
667
+ "w_max": 1.7810659110546112,
668
+ "w_mean": 1.2216298282146454,
669
+ "w_min": 0.0,
670
+ "w_std": 0.26985964179039
671
+ },
672
+ {
673
+ "completion_length": 3252.6771240234375,
674
+ "cov_mean": -2.1283684873196762e-05,
675
+ "cov_std": 0.31266890093684196,
676
+ "entropy": 0.376953125,
677
+ "epoch": 0.03542857142857143,
678
+ "grad_norm": 0.28267669677734375,
679
+ "kl": 0.00033402442932128906,
680
+ "learning_rate": 8.844151714648274e-07,
681
+ "loss": 0.0136,
682
+ "reward": 0.53125,
683
+ "reward_std": 0.3965660147368908,
684
+ "rewards/accuracy_reward": 0.1979166679084301,
685
+ "rewards/format_reward": 0.3333333358168602,
686
+ "step": 31,
687
+ "w_high_ratio": 0.13101918064057827,
688
+ "w_low_ratio": 0.03200887469574809,
689
+ "w_max": 1.9720379710197449,
690
+ "w_mean": 1.2918123006820679,
691
+ "w_min": 0.25,
692
+ "w_std": 0.22479910403490067
693
+ },
694
+ {
695
+ "completion_length": 3411.0208740234375,
696
+ "cov_mean": -2.1311116142896935e-06,
697
+ "cov_std": 0.37892643362283707,
698
+ "entropy": 0.439453125,
699
+ "epoch": 0.036571428571428574,
700
+ "grad_norm": 0.245382159948349,
701
+ "kl": 0.00054168701171875,
702
+ "learning_rate": 8.737029101523929e-07,
703
+ "loss": 0.0361,
704
+ "reward": 0.6666666716337204,
705
+ "reward_std": 0.5180360972881317,
706
+ "rewards/accuracy_reward": 0.2916666669771075,
707
+ "rewards/format_reward": 0.3750000037252903,
708
+ "step": 32,
709
+ "w_high_ratio": 0.0408792570233345,
710
+ "w_low_ratio": 0.0450265696272254,
711
+ "w_max": 1.5775762796401978,
712
+ "w_mean": 1.1778713762760162,
713
+ "w_min": 0.0,
714
+ "w_std": 0.22701822221279144
715
+ },
716
+ {
717
+ "completion_length": 3525.760498046875,
718
+ "cov_mean": -4.6630036194983404e-05,
719
+ "cov_std": 0.3494578003883362,
720
+ "entropy": 0.37939453125,
721
+ "epoch": 0.037714285714285714,
722
+ "grad_norm": 0.2607223689556122,
723
+ "kl": 0.0014134645462036133,
724
+ "learning_rate": 8.625962667065487e-07,
725
+ "loss": 0.0251,
726
+ "reward": 0.47916669212281704,
727
+ "reward_std": 0.5007270351052284,
728
+ "rewards/accuracy_reward": 0.18750001024454832,
729
+ "rewards/format_reward": 0.2916666818782687,
730
+ "step": 33,
731
+ "w_high_ratio": 0.038857411593198776,
732
+ "w_low_ratio": 0.04762307275086641,
733
+ "w_max": 1.5454095602035522,
734
+ "w_mean": 1.1393149495124817,
735
+ "w_min": 0.0,
736
+ "w_std": 0.22297161258757114
737
+ },
738
+ {
739
+ "completion_length": 2453.9791870117188,
740
+ "cov_mean": -9.226416841556784e-05,
741
+ "cov_std": 0.4420240595936775,
742
+ "entropy": 0.43701171875,
743
+ "epoch": 0.038857142857142854,
744
+ "grad_norm": 0.5145274996757507,
745
+ "kl": 0.0017969608306884766,
746
+ "learning_rate": 8.511087728614862e-07,
747
+ "loss": -0.018,
748
+ "reward": 0.9583333684131503,
749
+ "reward_std": 0.4463650956749916,
750
+ "rewards/accuracy_reward": 0.375000006519258,
751
+ "rewards/format_reward": 0.5833333432674408,
752
+ "step": 34,
753
+ "w_high_ratio": 0.13110801950097084,
754
+ "w_low_ratio": 0.052741190418601036,
755
+ "w_max": 2.1099403500556946,
756
+ "w_mean": 1.343121200799942,
757
+ "w_min": 0.0,
758
+ "w_std": 0.31580859050154686
759
+ },
760
+ {
761
+ "completion_length": 3289.4376220703125,
762
+ "cov_mean": 5.6672024584258907e-05,
763
+ "cov_std": 0.3820243999361992,
764
+ "entropy": 0.42529296875,
765
+ "epoch": 0.04,
766
+ "grad_norm": 0.35655274987220764,
767
+ "kl": 0.0010710954666137695,
768
+ "learning_rate": 8.392544243589427e-07,
769
+ "loss": 0.0185,
770
+ "reward": 0.6458333432674408,
771
+ "reward_std": 0.5987429097294807,
772
+ "rewards/accuracy_reward": 0.2708333395421505,
773
+ "rewards/format_reward": 0.3750000037252903,
774
+ "step": 35,
775
+ "w_high_ratio": 0.0,
776
+ "w_low_ratio": 0.05319366045296192,
777
+ "w_max": 1.4592802226543427,
778
+ "w_mean": 1.0992730259895325,
779
+ "w_min": 3.853570776893247e-45,
780
+ "w_std": 0.22482672333717346
781
+ },
782
+ {
783
+ "completion_length": 3621.78125,
784
+ "cov_mean": 7.363205986621324e-05,
785
+ "cov_std": 0.19353903084993362,
786
+ "entropy": 0.4765625,
787
+ "epoch": 0.04114285714285714,
788
+ "grad_norm": 0.30453404784202576,
789
+ "kl": 0.001312553882598877,
790
+ "learning_rate": 8.270476638965461e-07,
791
+ "loss": -0.0147,
792
+ "reward": 0.1666666716337204,
793
+ "reward_std": 0.17689276114106178,
794
+ "rewards/accuracy_reward": 0.010416666977107525,
795
+ "rewards/format_reward": 0.15625,
796
+ "step": 36,
797
+ "w_high_ratio": 0.006425275932997465,
798
+ "w_low_ratio": 0.022115232422947884,
799
+ "w_max": 1.4147455394268036,
800
+ "w_mean": 1.117747575044632,
801
+ "w_min": 0.5,
802
+ "w_std": 0.1310933344066143
803
+ },
804
+ {
805
+ "completion_length": 3511.2813110351562,
806
+ "cov_mean": -8.677186542627169e-06,
807
+ "cov_std": 0.2782720513641834,
808
+ "entropy": 0.43408203125,
809
+ "epoch": 0.04228571428571429,
810
+ "grad_norm": 0.357716828584671,
811
+ "kl": 0.0007710456848144531,
812
+ "learning_rate": 8.145033635316128e-07,
813
+ "loss": -0.0037,
814
+ "reward": 0.21875000279396772,
815
+ "reward_std": 0.20087094232439995,
816
+ "rewards/accuracy_reward": 0.010416666977107525,
817
+ "rewards/format_reward": 0.20833333861082792,
818
+ "step": 37,
819
+ "w_high_ratio": 0.07502041757106781,
820
+ "w_low_ratio": 0.02649637684226036,
821
+ "w_max": 1.848203331232071,
822
+ "w_mean": 1.2203471958637238,
823
+ "w_min": 0.25,
824
+ "w_std": 0.18464426696300507
825
+ },
826
+ {
827
+ "completion_length": 3627.8438110351562,
828
+ "cov_mean": -6.837155524408445e-05,
829
+ "cov_std": 0.14785446226596832,
830
+ "entropy": 0.44921875,
831
+ "epoch": 0.04342857142857143,
832
+ "grad_norm": 0.28619271516799927,
833
+ "kl": 0.0005115270614624023,
834
+ "learning_rate": 8.01636806561836e-07,
835
+ "loss": 0.0051,
836
+ "reward": 0.3645833358168602,
837
+ "reward_std": 0.25648824870586395,
838
+ "rewards/accuracy_reward": 0.1770833395421505,
839
+ "rewards/format_reward": 0.1875000111758709,
840
+ "step": 38,
841
+ "w_high_ratio": 0.051015257835388184,
842
+ "w_low_ratio": 0.01901705004274845,
843
+ "w_max": 1.441689670085907,
844
+ "w_mean": 1.1396986544132233,
845
+ "w_min": 0.5,
846
+ "w_std": 0.09881949052214622
847
+ },
848
+ {
849
+ "completion_length": 3050.979248046875,
850
+ "cov_mean": 4.009085705547477e-05,
851
+ "cov_std": 0.20474716648459435,
852
+ "entropy": 0.366943359375,
853
+ "epoch": 0.044571428571428574,
854
+ "grad_norm": 0.2579488158226013,
855
+ "kl": 0.0016760826110839844,
856
+ "learning_rate": 7.884636689049422e-07,
857
+ "loss": -0.0486,
858
+ "reward": 0.7083333358168602,
859
+ "reward_std": 0.2621734067797661,
860
+ "rewards/accuracy_reward": 0.21875,
861
+ "rewards/format_reward": 0.4895833358168602,
862
+ "step": 39,
863
+ "w_high_ratio": 0.04251348227262497,
864
+ "w_low_ratio": 0.022377901477739215,
865
+ "w_max": 1.5512709319591522,
866
+ "w_mean": 1.1656133532524109,
867
+ "w_min": 0.25,
868
+ "w_std": 0.12140309438109398
869
+ },
870
+ {
871
+ "completion_length": 2983.6876220703125,
872
+ "cov_mean": -2.4692089027666952e-05,
873
+ "cov_std": 0.3223262317478657,
874
+ "entropy": 0.39697265625,
875
+ "epoch": 0.045714285714285714,
876
+ "grad_norm": 0.3634110987186432,
877
+ "kl": 0.003978729248046875,
878
+ "learning_rate": 7.75e-07,
879
+ "loss": 0.0235,
880
+ "reward": 0.7083333507180214,
881
+ "reward_std": 0.39669927954673767,
882
+ "rewards/accuracy_reward": 0.20833334140479565,
883
+ "rewards/format_reward": 0.5000000149011612,
884
+ "step": 40,
885
+ "w_high_ratio": 0.023237863555550575,
886
+ "w_low_ratio": 0.03736424632370472,
887
+ "w_max": 1.7330412566661835,
888
+ "w_mean": 1.253759890794754,
889
+ "w_min": 0.0,
890
+ "w_std": 0.2265719212591648
891
+ },
892
+ {
893
+ "completion_length": 3342.916748046875,
894
+ "cov_mean": 7.446667950716801e-06,
895
+ "cov_std": 0.36464616656303406,
896
+ "entropy": 0.38232421875,
897
+ "epoch": 0.046857142857142854,
898
+ "grad_norm": 0.3088231682777405,
899
+ "kl": 0.0006549358367919922,
900
+ "learning_rate": 7.612622032536507e-07,
901
+ "loss": 0.01,
902
+ "reward": 0.416666679084301,
903
+ "reward_std": 0.49482717365026474,
904
+ "rewards/accuracy_reward": 0.10416667070239782,
905
+ "rewards/format_reward": 0.3125000037252903,
906
+ "step": 41,
907
+ "w_high_ratio": 0.02740619331598282,
908
+ "w_low_ratio": 0.045960213989019394,
909
+ "w_max": 1.735625982284546,
910
+ "w_mean": 1.2091110348701477,
911
+ "w_min": 0.0,
912
+ "w_std": 0.24962139129638672
913
+ },
914
+ {
915
+ "completion_length": 3026.229248046875,
916
+ "cov_mean": 2.7950623916694894e-05,
917
+ "cov_std": 0.22710799425840378,
918
+ "entropy": 0.46923828125,
919
+ "epoch": 0.048,
920
+ "grad_norm": 0.188913494348526,
921
+ "kl": 0.0005321502685546875,
922
+ "learning_rate": 7.472670160550848e-07,
923
+ "loss": 0.0293,
924
+ "reward": 0.36458334140479565,
925
+ "reward_std": 0.2087036520242691,
926
+ "rewards/accuracy_reward": 0.010416666977107525,
927
+ "rewards/format_reward": 0.3541666744276881,
928
+ "step": 42,
929
+ "w_high_ratio": 0.0,
930
+ "w_low_ratio": 0.03042191034182906,
931
+ "w_max": 1.318351686000824,
932
+ "w_mean": 1.0878586322069168,
933
+ "w_min": 0.25,
934
+ "w_std": 0.1288738138973713
935
+ },
936
+ {
937
+ "completion_length": 3391.8333740234375,
938
+ "cov_mean": 1.0224231118627358e-05,
939
+ "cov_std": 0.3281352184712887,
940
+ "entropy": 0.421875,
941
+ "epoch": 0.04914285714285714,
942
+ "grad_norm": 0.3027113378047943,
943
+ "kl": 0.0014162063598632812,
944
+ "learning_rate": 7.330314893841101e-07,
945
+ "loss": 0.0293,
946
+ "reward": 0.4479166716337204,
947
+ "reward_std": 0.4176873341202736,
948
+ "rewards/accuracy_reward": 0.18750000558793545,
949
+ "rewards/format_reward": 0.2604166716337204,
950
+ "step": 43,
951
+ "w_high_ratio": 0.0,
952
+ "w_low_ratio": 0.0375382611528039,
953
+ "w_max": 1.406785786151886,
954
+ "w_mean": 1.0772785246372223,
955
+ "w_min": 0.0,
956
+ "w_std": 0.17778386548161507
957
+ },
958
+ {
959
+ "completion_length": 2956.250030517578,
960
+ "cov_mean": -4.8457440243510064e-05,
961
+ "cov_std": 0.4899379573762417,
962
+ "entropy": 0.3984375,
963
+ "epoch": 0.05028571428571429,
964
+ "grad_norm": 0.4807187020778656,
965
+ "kl": 0.001889944076538086,
966
+ "learning_rate": 7.185729670371604e-07,
967
+ "loss": 0.0382,
968
+ "reward": 0.7395833684131503,
969
+ "reward_std": 0.4934317544102669,
970
+ "rewards/accuracy_reward": 0.2395833432674408,
971
+ "rewards/format_reward": 0.5000000176951289,
972
+ "step": 44,
973
+ "w_high_ratio": 0.09932601824402809,
974
+ "w_low_ratio": 0.052049466874450445,
975
+ "w_max": 2.26826411485672,
976
+ "w_mean": 1.3322851359844208,
977
+ "w_min": 0.0,
978
+ "w_std": 0.30705559998750687
979
+ },
980
+ {
981
+ "completion_length": 3707.854248046875,
982
+ "cov_mean": -4.304830235923873e-05,
983
+ "cov_std": 0.290540874004364,
984
+ "entropy": 0.4033203125,
985
+ "epoch": 0.05142857142857143,
986
+ "grad_norm": 0.16171735525131226,
987
+ "kl": 0.002239227294921875,
988
+ "learning_rate": 7.039090644965509e-07,
989
+ "loss": 0.0065,
990
+ "reward": 0.5000000279396772,
991
+ "reward_std": 0.4716969132423401,
992
+ "rewards/accuracy_reward": 0.21875000279396772,
993
+ "rewards/format_reward": 0.2812500102445483,
994
+ "step": 45,
995
+ "w_high_ratio": 0.0,
996
+ "w_low_ratio": 0.03661782294511795,
997
+ "w_max": 1.3389239609241486,
998
+ "w_mean": 1.0591736733913422,
999
+ "w_min": 0.0,
1000
+ "w_std": 0.1518448255956173
1001
+ },
1002
+ {
1003
+ "completion_length": 3539.2188110351562,
1004
+ "cov_mean": 0.00013026446868025232,
1005
+ "cov_std": 0.34402996301651,
1006
+ "entropy": 0.5166015625,
1007
+ "epoch": 0.052571428571428575,
1008
+ "grad_norm": 0.28467056155204773,
1009
+ "kl": 0.0012564659118652344,
1010
+ "learning_rate": 6.890576474687263e-07,
1011
+ "loss": 0.0542,
1012
+ "reward": 0.2395833395421505,
1013
+ "reward_std": 0.31017982587218285,
1014
+ "rewards/accuracy_reward": 0.041666666977107525,
1015
+ "rewards/format_reward": 0.1979166716337204,
1016
+ "step": 46,
1017
+ "w_high_ratio": 0.038347791880369186,
1018
+ "w_low_ratio": 0.04659654130227864,
1019
+ "w_max": 1.562936246395111,
1020
+ "w_mean": 1.0923520624637604,
1021
+ "w_min": 6.825020352992891e-31,
1022
+ "w_std": 0.19784706830978394
1023
+ },
1024
+ {
1025
+ "completion_length": 3168.5313110351562,
1026
+ "cov_mean": -1.4116008060227614e-05,
1027
+ "cov_std": 0.3409600518643856,
1028
+ "entropy": 0.4072265625,
1029
+ "epoch": 0.053714285714285714,
1030
+ "grad_norm": 0.21553435921669006,
1031
+ "kl": 0.0014462471008300781,
1032
+ "learning_rate": 6.740368101176495e-07,
1033
+ "loss": 0.0063,
1034
+ "reward": 0.7083333432674408,
1035
+ "reward_std": 0.505032941699028,
1036
+ "rewards/accuracy_reward": 0.28125,
1037
+ "rewards/format_reward": 0.4270833358168602,
1038
+ "step": 47,
1039
+ "w_high_ratio": 0.1824447140097618,
1040
+ "w_low_ratio": 0.0349124651402235,
1041
+ "w_max": 1.8283918797969818,
1042
+ "w_mean": 1.3520298600196838,
1043
+ "w_min": 0.25,
1044
+ "w_std": 0.17590370774269104
1045
+ },
1046
+ {
1047
+ "completion_length": 3045.5209350585938,
1048
+ "cov_mean": 1.4324002222565468e-05,
1049
+ "cov_std": 0.41333913803100586,
1050
+ "entropy": 0.40966796875,
1051
+ "epoch": 0.054857142857142854,
1052
+ "grad_norm": 0.514603316783905,
1053
+ "kl": 0.0032389163970947266,
1054
+ "learning_rate": 6.588648530198504e-07,
1055
+ "loss": -0.0291,
1056
+ "reward": 0.5729166902601719,
1057
+ "reward_std": 0.4265441298484802,
1058
+ "rewards/accuracy_reward": 0.20833334419876337,
1059
+ "rewards/format_reward": 0.3645833460614085,
1060
+ "step": 48,
1061
+ "w_high_ratio": 0.14455852285027504,
1062
+ "w_low_ratio": 0.04336892580613494,
1063
+ "w_max": 2.106689363718033,
1064
+ "w_mean": 1.3398694097995758,
1065
+ "w_min": 0.0,
1066
+ "w_std": 0.29295507818460464
1067
+ },
1068
+ {
1069
+ "completion_length": 2561.3646850585938,
1070
+ "cov_mean": -5.681005814039963e-05,
1071
+ "cov_std": 0.3279518857598305,
1072
+ "entropy": 0.36865234375,
1073
+ "epoch": 0.056,
1074
+ "grad_norm": 0.25349730253219604,
1075
+ "kl": 0.0052852630615234375,
1076
+ "learning_rate": 6.435602608679916e-07,
1077
+ "loss": 0.0033,
1078
+ "reward": 0.833333358168602,
1079
+ "reward_std": 0.46440815925598145,
1080
+ "rewards/accuracy_reward": 0.21875001303851604,
1081
+ "rewards/format_reward": 0.6145833432674408,
1082
+ "step": 49,
1083
+ "w_high_ratio": 0.02398517169058323,
1084
+ "w_low_ratio": 0.04019799269735813,
1085
+ "w_max": 1.7071016132831573,
1086
+ "w_mean": 1.2629328966140747,
1087
+ "w_min": 0.25,
1088
+ "w_std": 0.19682494550943375
1089
+ },
1090
+ {
1091
+ "completion_length": 3245.2083740234375,
1092
+ "cov_mean": 4.621147024863603e-05,
1093
+ "cov_std": 0.35216138139367104,
1094
+ "entropy": 0.35205078125,
1095
+ "epoch": 0.05714285714285714,
1096
+ "grad_norm": 0.20161262154579163,
1097
+ "kl": 0.0010981559753417969,
1098
+ "learning_rate": 6.281416799501187e-07,
1099
+ "loss": 0.014,
1100
+ "reward": 0.6354166967794299,
1101
+ "reward_std": 0.420846089720726,
1102
+ "rewards/accuracy_reward": 0.2812500149011612,
1103
+ "rewards/format_reward": 0.3541666818782687,
1104
+ "step": 50,
1105
+ "w_high_ratio": 0.0,
1106
+ "w_low_ratio": 0.042036826722323895,
1107
+ "w_max": 1.5023921430110931,
1108
+ "w_mean": 1.146658033132553,
1109
+ "w_min": 0.0,
1110
+ "w_std": 0.206118730828166
1111
+ },
1112
+ {
1113
+ "completion_length": 2383.6980590820312,
1114
+ "cov_mean": -2.697939362406032e-05,
1115
+ "cov_std": 0.38119756057858467,
1116
+ "entropy": 0.42822265625,
1117
+ "epoch": 0.05828571428571429,
1118
+ "grad_norm": 0.4689660966396332,
1119
+ "kl": 0.0037221908569335938,
1120
+ "learning_rate": 6.126278954320294e-07,
1121
+ "loss": 0.011,
1122
+ "reward": 0.7604166865348816,
1123
+ "reward_std": 0.39414726570248604,
1124
+ "rewards/accuracy_reward": 0.20833333861082792,
1125
+ "rewards/format_reward": 0.5520833432674408,
1126
+ "step": 51,
1127
+ "w_high_ratio": 0.0406239777803421,
1128
+ "w_low_ratio": 0.04253132501617074,
1129
+ "w_max": 1.5954174101352692,
1130
+ "w_mean": 1.217557042837143,
1131
+ "w_min": 0.0,
1132
+ "w_std": 0.24302981421351433
1133
+ },
1134
+ {
1135
+ "completion_length": 3188.3021240234375,
1136
+ "cov_mean": -1.9340616745466832e-05,
1137
+ "cov_std": 0.42421338707208633,
1138
+ "entropy": 0.41259765625,
1139
+ "epoch": 0.05942857142857143,
1140
+ "grad_norm": 0.43333640694618225,
1141
+ "kl": 0.0021028518676757812,
1142
+ "learning_rate": 5.97037808470444e-07,
1143
+ "loss": -0.0265,
1144
+ "reward": 0.7500000298023224,
1145
+ "reward_std": 0.56574647128582,
1146
+ "rewards/accuracy_reward": 0.3229166716337204,
1147
+ "rewards/format_reward": 0.4270833507180214,
1148
+ "step": 52,
1149
+ "w_high_ratio": 0.0,
1150
+ "w_low_ratio": 0.04618662968277931,
1151
+ "w_max": 1.4080191552639008,
1152
+ "w_mean": 1.1331122517585754,
1153
+ "w_min": 0.25,
1154
+ "w_std": 0.21432508528232574
1155
+ },
1156
+ {
1157
+ "completion_length": 2943.291748046875,
1158
+ "cov_mean": 1.8985145061378716e-05,
1159
+ "cov_std": 0.42384568601846695,
1160
+ "entropy": 0.41259765625,
1161
+ "epoch": 0.060571428571428575,
1162
+ "grad_norm": 0.28105950355529785,
1163
+ "kl": 0.001844644546508789,
1164
+ "learning_rate": 5.813904131848564e-07,
1165
+ "loss": 0.0064,
1166
+ "reward": 0.895833358168602,
1167
+ "reward_std": 0.5318989902734756,
1168
+ "rewards/accuracy_reward": 0.30208334885537624,
1169
+ "rewards/format_reward": 0.5937500149011612,
1170
+ "step": 53,
1171
+ "w_high_ratio": 0.0625,
1172
+ "w_low_ratio": 0.04191893618553877,
1173
+ "w_max": 1.6293854713439941,
1174
+ "w_mean": 1.2208082973957062,
1175
+ "w_min": 0.25,
1176
+ "w_std": 0.20713179931044579
1177
+ },
1178
+ {
1179
+ "completion_length": 3035.8438110351562,
1180
+ "cov_mean": 6.566583215317223e-05,
1181
+ "cov_std": 0.39858507737517357,
1182
+ "entropy": 0.3798828125,
1183
+ "epoch": 0.061714285714285715,
1184
+ "grad_norm": 0.38175392150878906,
1185
+ "kl": 0.0011749267578125,
1186
+ "learning_rate": 5.657047735161255e-07,
1187
+ "loss": -0.0521,
1188
+ "reward": 0.9791667088866234,
1189
+ "reward_std": 0.5917666852474213,
1190
+ "rewards/accuracy_reward": 0.42708334885537624,
1191
+ "rewards/format_reward": 0.552083358168602,
1192
+ "step": 54,
1193
+ "w_high_ratio": 0.044216301292181015,
1194
+ "w_low_ratio": 0.03513730876147747,
1195
+ "w_max": 1.7533048391342163,
1196
+ "w_mean": 1.3249602317810059,
1197
+ "w_min": 0.0,
1198
+ "w_std": 0.22004129365086555
1199
+ },
1200
+ {
1201
+ "completion_length": 3354.666748046875,
1202
+ "cov_mean": -3.56279870175058e-05,
1203
+ "cov_std": 0.30405301228165627,
1204
+ "entropy": 0.419921875,
1205
+ "epoch": 0.06285714285714286,
1206
+ "grad_norm": 0.3018040060997009,
1207
+ "kl": 0.0012395381927490234,
1208
+ "learning_rate": 5.5e-07,
1209
+ "loss": -0.0185,
1210
+ "reward": 0.5208333469927311,
1211
+ "reward_std": 0.3702617585659027,
1212
+ "rewards/accuracy_reward": 0.1979166716337204,
1213
+ "rewards/format_reward": 0.3229166679084301,
1214
+ "step": 55,
1215
+ "w_high_ratio": 0.0,
1216
+ "w_low_ratio": 0.03356400621123612,
1217
+ "w_max": 1.4872715771198273,
1218
+ "w_mean": 1.1252524256706238,
1219
+ "w_min": 0.0,
1220
+ "w_std": 0.17813249677419662
1221
+ },
1222
+ {
1223
+ "completion_length": 3199.7188720703125,
1224
+ "cov_mean": -1.4941413155611372e-05,
1225
+ "cov_std": 0.25671473518013954,
1226
+ "entropy": 0.40185546875,
1227
+ "epoch": 0.064,
1228
+ "grad_norm": 0.21876569092273712,
1229
+ "kl": 0.0006089210510253906,
1230
+ "learning_rate": 5.342952264838747e-07,
1231
+ "loss": -0.0325,
1232
+ "reward": 0.6458333395421505,
1233
+ "reward_std": 0.27983931452035904,
1234
+ "rewards/accuracy_reward": 0.1979166716337204,
1235
+ "rewards/format_reward": 0.4479166828095913,
1236
+ "step": 56,
1237
+ "w_high_ratio": 0.0,
1238
+ "w_low_ratio": 0.028304541716352105,
1239
+ "w_max": 1.453190177679062,
1240
+ "w_mean": 1.1167612671852112,
1241
+ "w_min": 1.400437147815929e-36,
1242
+ "w_std": 0.14085076563060284
1243
+ },
1244
+ {
1245
+ "completion_length": 3545.4271850585938,
1246
+ "cov_mean": 1.6042108654801268e-05,
1247
+ "cov_std": 0.19959648326039314,
1248
+ "entropy": 0.31787109375,
1249
+ "epoch": 0.06514285714285714,
1250
+ "grad_norm": 0.15603993833065033,
1251
+ "kl": 0.0003676414489746094,
1252
+ "learning_rate": 5.186095868151436e-07,
1253
+ "loss": 0.0067,
1254
+ "reward": 0.4270833395421505,
1255
+ "reward_std": 0.26762784644961357,
1256
+ "rewards/accuracy_reward": 0.13541667256504297,
1257
+ "rewards/format_reward": 0.29166667722165585,
1258
+ "step": 57,
1259
+ "w_high_ratio": 0.0,
1260
+ "w_low_ratio": 0.028780631721019745,
1261
+ "w_max": 1.3290077447891235,
1262
+ "w_mean": 1.1172043681144714,
1263
+ "w_min": 0.25,
1264
+ "w_std": 0.12668126448988914
1265
+ },
1266
+ {
1267
+ "completion_length": 2554.1458740234375,
1268
+ "cov_mean": -8.6602381998091e-05,
1269
+ "cov_std": 0.4178408235311508,
1270
+ "entropy": 0.333740234375,
1271
+ "epoch": 0.06628571428571428,
1272
+ "grad_norm": 0.4292741119861603,
1273
+ "kl": 0.0062389373779296875,
1274
+ "learning_rate": 5.02962191529556e-07,
1275
+ "loss": -0.045,
1276
+ "reward": 0.9375000447034836,
1277
+ "reward_std": 0.5022420063614845,
1278
+ "rewards/accuracy_reward": 0.28125000558793545,
1279
+ "rewards/format_reward": 0.6562500223517418,
1280
+ "step": 58,
1281
+ "w_high_ratio": 0.08225274085998535,
1282
+ "w_low_ratio": 0.038829353637993336,
1283
+ "w_max": 1.8886817693710327,
1284
+ "w_mean": 1.3429239392280579,
1285
+ "w_min": 0.0,
1286
+ "w_std": 0.24752679839730263
1287
+ },
1288
+ {
1289
+ "completion_length": 3169.0625610351562,
1290
+ "cov_mean": 4.032192100567045e-05,
1291
+ "cov_std": 0.30262381583452225,
1292
+ "entropy": 0.34619140625,
1293
+ "epoch": 0.06742857142857143,
1294
+ "grad_norm": 0.38411930203437805,
1295
+ "kl": 0.0004928112030029297,
1296
+ "learning_rate": 4.873721045679706e-07,
1297
+ "loss": 0.0209,
1298
+ "reward": 0.5000000102445483,
1299
+ "reward_std": 0.27465011551976204,
1300
+ "rewards/accuracy_reward": 0.1770833358168602,
1301
+ "rewards/format_reward": 0.3229166669771075,
1302
+ "step": 59,
1303
+ "w_high_ratio": 0.15833117440342903,
1304
+ "w_low_ratio": 0.027412693481892347,
1305
+ "w_max": 2.2743676006793976,
1306
+ "w_mean": 1.3636261224746704,
1307
+ "w_min": 0.25,
1308
+ "w_std": 0.22530828416347504
1309
+ },
1310
+ {
1311
+ "completion_length": 3205.885498046875,
1312
+ "cov_mean": 5.204351282372954e-06,
1313
+ "cov_std": 0.21868624165654182,
1314
+ "entropy": 0.35986328125,
1315
+ "epoch": 0.06857142857142857,
1316
+ "grad_norm": 0.2159898579120636,
1317
+ "kl": 0.0013399124145507812,
1318
+ "learning_rate": 4.7185832004988133e-07,
1319
+ "loss": 0.0012,
1320
+ "reward": 0.552083358168602,
1321
+ "reward_std": 0.3438131511211395,
1322
+ "rewards/accuracy_reward": 0.13541667070239782,
1323
+ "rewards/format_reward": 0.416666679084301,
1324
+ "step": 60,
1325
+ "w_high_ratio": 0.0,
1326
+ "w_low_ratio": 0.0294471038505435,
1327
+ "w_max": 1.4707800149917603,
1328
+ "w_mean": 1.1660968363285065,
1329
+ "w_min": 0.25,
1330
+ "w_std": 0.13705550134181976
1331
+ },
1332
+ {
1333
+ "completion_length": 3322.7084350585938,
1334
+ "cov_mean": 3.926075260096695e-05,
1335
+ "cov_std": 0.3601963371038437,
1336
+ "entropy": 0.3564453125,
1337
+ "epoch": 0.06971428571428571,
1338
+ "grad_norm": 0.39628517627716064,
1339
+ "kl": 0.0009188652038574219,
1340
+ "learning_rate": 4.5643973913200837e-07,
1341
+ "loss": -0.0714,
1342
+ "reward": 0.5833333414047956,
1343
+ "reward_std": 0.515319935977459,
1344
+ "rewards/accuracy_reward": 0.1666666716337204,
1345
+ "rewards/format_reward": 0.41666666977107525,
1346
+ "step": 61,
1347
+ "w_high_ratio": 0.03236403688788414,
1348
+ "w_low_ratio": 0.03914234112016857,
1349
+ "w_max": 1.7150523364543915,
1350
+ "w_mean": 1.2098670601844788,
1351
+ "w_min": 0.0,
1352
+ "w_std": 0.2282575462013483
1353
+ },
1354
+ {
1355
+ "completion_length": 3006.885498046875,
1356
+ "cov_mean": 2.1906609163124813e-05,
1357
+ "cov_std": 0.46718820184469223,
1358
+ "entropy": 0.33935546875,
1359
+ "epoch": 0.07085714285714285,
1360
+ "grad_norm": 0.30309396982192993,
1361
+ "kl": 0.0017642974853515625,
1362
+ "learning_rate": 4.4113514698014953e-07,
1363
+ "loss": -0.035,
1364
+ "reward": 0.8229167014360428,
1365
+ "reward_std": 0.6976551413536072,
1366
+ "rewards/accuracy_reward": 0.27083334140479565,
1367
+ "rewards/format_reward": 0.5520833432674408,
1368
+ "step": 62,
1369
+ "w_high_ratio": 0.0,
1370
+ "w_low_ratio": 0.05965683329850435,
1371
+ "w_max": 1.593172937631607,
1372
+ "w_mean": 1.1934809386730194,
1373
+ "w_min": 1.0754965713692971e-43,
1374
+ "w_std": 0.28461313247680664
1375
+ },
1376
+ {
1377
+ "completion_length": 2580.229248046875,
1378
+ "cov_mean": 5.573407906922512e-05,
1379
+ "cov_std": 0.3945640027523041,
1380
+ "entropy": 0.39111328125,
1381
+ "epoch": 0.072,
1382
+ "grad_norm": 0.45718619227409363,
1383
+ "kl": 0.001735687255859375,
1384
+ "learning_rate": 4.2596318988235037e-07,
1385
+ "loss": -0.0807,
1386
+ "reward": 0.958333358168602,
1387
+ "reward_std": 0.5150914192199707,
1388
+ "rewards/accuracy_reward": 0.31250001303851604,
1389
+ "rewards/format_reward": 0.6458333432674408,
1390
+ "step": 63,
1391
+ "w_high_ratio": 0.08918174356222153,
1392
+ "w_low_ratio": 0.03547394182533026,
1393
+ "w_max": 2.0218554735183716,
1394
+ "w_mean": 1.3539214432239532,
1395
+ "w_min": 0.0,
1396
+ "w_std": 0.23917281255126
1397
+ },
1398
+ {
1399
+ "completion_length": 3340.0313110351562,
1400
+ "cov_mean": 7.548666690126993e-05,
1401
+ "cov_std": 0.31813300400972366,
1402
+ "entropy": 0.4248046875,
1403
+ "epoch": 0.07314285714285715,
1404
+ "grad_norm": 0.5389249920845032,
1405
+ "kl": 0.0018744468688964844,
1406
+ "learning_rate": 4.1094235253127374e-07,
1407
+ "loss": -0.0757,
1408
+ "reward": 0.5104166865348816,
1409
+ "reward_std": 0.4255366772413254,
1410
+ "rewards/accuracy_reward": 0.1979166753590107,
1411
+ "rewards/format_reward": 0.3125000074505806,
1412
+ "step": 64,
1413
+ "w_high_ratio": 0.15633241832256317,
1414
+ "w_low_ratio": 0.0320257218554616,
1415
+ "w_max": 1.8200311660766602,
1416
+ "w_mean": 1.3009037971496582,
1417
+ "w_min": 0.25,
1418
+ "w_std": 0.21954189240932465
1419
+ },
1420
+ {
1421
+ "completion_length": 2969.541748046875,
1422
+ "cov_mean": -2.4962878342194017e-05,
1423
+ "cov_std": 0.2727944627404213,
1424
+ "entropy": 0.353759765625,
1425
+ "epoch": 0.07428571428571429,
1426
+ "grad_norm": 0.2332458198070526,
1427
+ "kl": 0.0015840530395507812,
1428
+ "learning_rate": 3.9609093550344907e-07,
1429
+ "loss": 0.012,
1430
+ "reward": 0.666666679084301,
1431
+ "reward_std": 0.357291292399168,
1432
+ "rewards/accuracy_reward": 0.2083333358168602,
1433
+ "rewards/format_reward": 0.4583333432674408,
1434
+ "step": 65,
1435
+ "w_high_ratio": 0.0,
1436
+ "w_low_ratio": 0.03699250426143408,
1437
+ "w_max": 1.5260689854621887,
1438
+ "w_mean": 1.164745956659317,
1439
+ "w_min": 1.0748096548620851e-35,
1440
+ "w_std": 0.14831626787781715
1441
+ },
1442
+ {
1443
+ "completion_length": 2311.2709197998047,
1444
+ "cov_mean": 2.157548601644521e-05,
1445
+ "cov_std": 0.23550356552004814,
1446
+ "entropy": 0.31787109375,
1447
+ "epoch": 0.07542857142857143,
1448
+ "grad_norm": 0.5557933449745178,
1449
+ "kl": 0.0011758804321289062,
1450
+ "learning_rate": 3.8142703296283953e-07,
1451
+ "loss": -0.0116,
1452
+ "reward": 0.9062500260770321,
1453
+ "reward_std": 0.29818330332636833,
1454
+ "rewards/accuracy_reward": 0.38541667722165585,
1455
+ "rewards/format_reward": 0.520833333954215,
1456
+ "step": 66,
1457
+ "w_high_ratio": 0.10277381911873817,
1458
+ "w_low_ratio": 0.03059569653123617,
1459
+ "w_max": 1.97393798828125,
1460
+ "w_mean": 1.278551459312439,
1461
+ "w_min": 0.0,
1462
+ "w_std": 0.20061488449573517
1463
+ },
1464
+ {
1465
+ "completion_length": 3726.135498046875,
1466
+ "cov_mean": 8.957304089562967e-07,
1467
+ "cov_std": 0.12987565249204636,
1468
+ "entropy": 0.35205078125,
1469
+ "epoch": 0.07657142857142857,
1470
+ "grad_norm": 0.11042757332324982,
1471
+ "kl": 0.0013402700424194336,
1472
+ "learning_rate": 3.6696851061588994e-07,
1473
+ "loss": -0.0059,
1474
+ "reward": 0.1770833358168602,
1475
+ "reward_std": 0.15668470412492752,
1476
+ "rewards/accuracy_reward": 0.03125,
1477
+ "rewards/format_reward": 0.1458333358168602,
1478
+ "step": 67,
1479
+ "w_high_ratio": 0.0,
1480
+ "w_low_ratio": 0.01588357798755169,
1481
+ "w_max": 1.2971015870571136,
1482
+ "w_mean": 1.0769366472959518,
1483
+ "w_min": 0.5,
1484
+ "w_std": 0.08244858682155609
1485
+ },
1486
+ {
1487
+ "completion_length": 2420.979232788086,
1488
+ "cov_mean": 1.4521759112540167e-05,
1489
+ "cov_std": 0.2692869156599045,
1490
+ "entropy": 0.37353515625,
1491
+ "epoch": 0.07771428571428571,
1492
+ "grad_norm": 0.3267475962638855,
1493
+ "kl": 0.0064449310302734375,
1494
+ "learning_rate": 3.5273298394491515e-07,
1495
+ "loss": -0.0214,
1496
+ "reward": 0.822916679084301,
1497
+ "reward_std": 0.39656912535429,
1498
+ "rewards/accuracy_reward": 0.2187500074505806,
1499
+ "rewards/format_reward": 0.6041666716337204,
1500
+ "step": 68,
1501
+ "w_high_ratio": 0.028063589707016945,
1502
+ "w_low_ratio": 0.029913824051618576,
1503
+ "w_max": 1.8040810525417328,
1504
+ "w_mean": 1.316351056098938,
1505
+ "w_min": 0.25,
1506
+ "w_std": 0.17742961645126343
1507
+ },
1508
+ {
1509
+ "completion_length": 2790.5730590820312,
1510
+ "cov_mean": 3.09187362290686e-05,
1511
+ "cov_std": 0.42568420618772507,
1512
+ "entropy": 0.4453125,
1513
+ "epoch": 0.07885714285714286,
1514
+ "grad_norm": 0.7317398190498352,
1515
+ "kl": 0.006710052490234375,
1516
+ "learning_rate": 3.387377967463493e-07,
1517
+ "loss": -0.0192,
1518
+ "reward": 0.4895833358168602,
1519
+ "reward_std": 0.41857269406318665,
1520
+ "rewards/accuracy_reward": 0.08333333674818277,
1521
+ "rewards/format_reward": 0.4062500074505806,
1522
+ "step": 69,
1523
+ "w_high_ratio": 0.19593259878456593,
1524
+ "w_low_ratio": 0.036022237967699766,
1525
+ "w_max": 3.16190242767334,
1526
+ "w_mean": 1.5170519053936005,
1527
+ "w_min": 0.0,
1528
+ "w_std": 0.3000602051615715
1529
+ },
1530
+ {
1531
+ "completion_length": 3255.7709350585938,
1532
+ "cov_mean": -6.321061573544284e-05,
1533
+ "cov_std": 0.31040842086076736,
1534
+ "entropy": 0.34765625,
1535
+ "epoch": 0.08,
1536
+ "grad_norm": 0.2253309190273285,
1537
+ "kl": 0.0019707679748535156,
1538
+ "learning_rate": 3.250000000000001e-07,
1539
+ "loss": 0.038,
1540
+ "reward": 0.5104166679084301,
1541
+ "reward_std": 0.3943600133061409,
1542
+ "rewards/accuracy_reward": 0.09375000093132257,
1543
+ "rewards/format_reward": 0.4166666716337204,
1544
+ "step": 70,
1545
+ "w_high_ratio": 0.0,
1546
+ "w_low_ratio": 0.04103634599596262,
1547
+ "w_max": 1.2795763909816742,
1548
+ "w_mean": 1.0765459537506104,
1549
+ "w_min": 0.0,
1550
+ "w_std": 0.15424982458353043
1551
+ },
1552
+ {
1553
+ "completion_length": 2893.093795776367,
1554
+ "cov_mean": -5.907902482249483e-05,
1555
+ "cov_std": 0.28848100267350674,
1556
+ "entropy": 0.39111328125,
1557
+ "epoch": 0.08114285714285714,
1558
+ "grad_norm": 0.4800524115562439,
1559
+ "kl": 0.0017595291137695312,
1560
+ "learning_rate": 3.115363310950578e-07,
1561
+ "loss": 0.0284,
1562
+ "reward": 0.562500030733645,
1563
+ "reward_std": 0.3276159390807152,
1564
+ "rewards/accuracy_reward": 0.18750000651925802,
1565
+ "rewards/format_reward": 0.37500000558793545,
1566
+ "step": 71,
1567
+ "w_high_ratio": 0.07317740470170975,
1568
+ "w_low_ratio": 0.03526141308248043,
1569
+ "w_max": 2.095757156610489,
1570
+ "w_mean": 1.278771072626114,
1571
+ "w_min": 4.3955273748636295e-36,
1572
+ "w_std": 0.23910818807780743
1573
+ },
1574
+ {
1575
+ "completion_length": 3460.354248046875,
1576
+ "cov_mean": 6.12462517892709e-05,
1577
+ "cov_std": 0.36787012219429016,
1578
+ "entropy": 0.4931640625,
1579
+ "epoch": 0.08228571428571428,
1580
+ "grad_norm": 0.39541420340538025,
1581
+ "kl": 0.0038938522338867188,
1582
+ "learning_rate": 2.9836319343816397e-07,
1583
+ "loss": -0.0245,
1584
+ "reward": 0.3333333469927311,
1585
+ "reward_std": 0.4593120887875557,
1586
+ "rewards/accuracy_reward": 0.06250000093132257,
1587
+ "rewards/format_reward": 0.2708333395421505,
1588
+ "step": 72,
1589
+ "w_high_ratio": 0.040276188403367996,
1590
+ "w_low_ratio": 0.046780284494161606,
1591
+ "w_max": 1.6123959422111511,
1592
+ "w_mean": 1.1790215075016022,
1593
+ "w_min": 0.0,
1594
+ "w_std": 0.24864742159843445
1595
+ },
1596
+ {
1597
+ "completion_length": 3808.166748046875,
1598
+ "cov_mean": 2.382581669735373e-05,
1599
+ "cov_std": 0.26704882457852364,
1600
+ "entropy": 0.4677734375,
1601
+ "epoch": 0.08342857142857144,
1602
+ "grad_norm": 0.39250648021698,
1603
+ "kl": 0.0005078315734863281,
1604
+ "learning_rate": 2.854966364683872e-07,
1605
+ "loss": 0.0333,
1606
+ "reward": 0.21875001024454832,
1607
+ "reward_std": 0.34505901485681534,
1608
+ "rewards/accuracy_reward": 0.0729166679084301,
1609
+ "rewards/format_reward": 0.14583333488553762,
1610
+ "step": 73,
1611
+ "w_high_ratio": 0.0,
1612
+ "w_low_ratio": 0.031687673181295395,
1613
+ "w_max": 1.3024969696998596,
1614
+ "w_mean": 1.03611221909523,
1615
+ "w_min": 0.0,
1616
+ "w_std": 0.13818097487092018
1617
+ },
1618
+ {
1619
+ "completion_length": 3584.8438110351562,
1620
+ "cov_mean": -4.775809429702349e-05,
1621
+ "cov_std": 0.3642261065542698,
1622
+ "entropy": 0.42138671875,
1623
+ "epoch": 0.08457142857142858,
1624
+ "grad_norm": 0.22599299252033234,
1625
+ "kl": 0.002000570297241211,
1626
+ "learning_rate": 2.729523361034538e-07,
1627
+ "loss": 0.0465,
1628
+ "reward": 0.552083358168602,
1629
+ "reward_std": 0.47496967762708664,
1630
+ "rewards/accuracy_reward": 0.2604166679084301,
1631
+ "rewards/format_reward": 0.2916666753590107,
1632
+ "step": 74,
1633
+ "w_high_ratio": 0.0,
1634
+ "w_low_ratio": 0.04971347749233246,
1635
+ "w_max": 1.3916961252689362,
1636
+ "w_mean": 1.079592913389206,
1637
+ "w_min": 0.0,
1638
+ "w_std": 0.20160867273807526
1639
+ },
1640
+ {
1641
+ "completion_length": 3403.7188720703125,
1642
+ "cov_mean": 4.775901288667228e-06,
1643
+ "cov_std": 0.31741658970713615,
1644
+ "entropy": 0.39208984375,
1645
+ "epoch": 0.08571428571428572,
1646
+ "grad_norm": 0.28842759132385254,
1647
+ "kl": 0.002251148223876953,
1648
+ "learning_rate": 2.6074557564105724e-07,
1649
+ "loss": -0.0149,
1650
+ "reward": 0.5312500074505806,
1651
+ "reward_std": 0.3348959684371948,
1652
+ "rewards/accuracy_reward": 0.1770833358168602,
1653
+ "rewards/format_reward": 0.3541666744276881,
1654
+ "step": 75,
1655
+ "w_high_ratio": 0.05775141716003418,
1656
+ "w_low_ratio": 0.04404309671372175,
1657
+ "w_max": 1.7450213432312012,
1658
+ "w_mean": 1.203758329153061,
1659
+ "w_min": 0.25,
1660
+ "w_std": 0.19945883378386497
1661
+ },
1662
+ {
1663
+ "completion_length": 3129.885498046875,
1664
+ "cov_mean": 1.2703612810582854e-05,
1665
+ "cov_std": 0.2526189871132374,
1666
+ "entropy": 0.396484375,
1667
+ "epoch": 0.08685714285714285,
1668
+ "grad_norm": 0.15439313650131226,
1669
+ "kl": 0.0004563331604003906,
1670
+ "learning_rate": 2.488912271385139e-07,
1671
+ "loss": 0.0167,
1672
+ "reward": 0.572916679084301,
1673
+ "reward_std": 0.2542962171137333,
1674
+ "rewards/accuracy_reward": 0.1041666716337204,
1675
+ "rewards/format_reward": 0.4687500074505806,
1676
+ "step": 76,
1677
+ "w_high_ratio": 0.0,
1678
+ "w_low_ratio": 0.03176691988483071,
1679
+ "w_max": 1.4051032662391663,
1680
+ "w_mean": 1.1220950484275818,
1681
+ "w_min": 0.25,
1682
+ "w_std": 0.1384560540318489
1683
+ },
1684
+ {
1685
+ "completion_length": 3342.510498046875,
1686
+ "cov_mean": 4.0359717786486726e-05,
1687
+ "cov_std": 0.2773597203195095,
1688
+ "entropy": 0.3876953125,
1689
+ "epoch": 0.088,
1690
+ "grad_norm": 0.39251646399497986,
1691
+ "kl": 0.0005764961242675781,
1692
+ "learning_rate": 2.374037332934512e-07,
1693
+ "loss": 0.0041,
1694
+ "reward": 0.552083333954215,
1695
+ "reward_std": 0.26390674337744713,
1696
+ "rewards/accuracy_reward": 0.1458333432674408,
1697
+ "rewards/format_reward": 0.40625000558793545,
1698
+ "step": 77,
1699
+ "w_high_ratio": 0.0,
1700
+ "w_low_ratio": 0.035081148613244295,
1701
+ "w_max": 1.4559779465198517,
1702
+ "w_mean": 1.158086121082306,
1703
+ "w_min": 0.0,
1704
+ "w_std": 0.16358545422554016
1705
+ },
1706
+ {
1707
+ "completion_length": 3630.3438720703125,
1708
+ "cov_mean": 2.0063042029505596e-05,
1709
+ "cov_std": 0.3748911917209625,
1710
+ "entropy": 0.39599609375,
1711
+ "epoch": 0.08914285714285715,
1712
+ "grad_norm": 0.50286865234375,
1713
+ "kl": 0.013364076614379883,
1714
+ "learning_rate": 2.2629708984760706e-07,
1715
+ "loss": 0.0002,
1716
+ "reward": 0.4479166939854622,
1717
+ "reward_std": 0.5592645406723022,
1718
+ "rewards/accuracy_reward": 0.1770833432674408,
1719
+ "rewards/format_reward": 0.2708333469927311,
1720
+ "step": 78,
1721
+ "w_high_ratio": 0.04053416848182678,
1722
+ "w_low_ratio": 0.047788419760763645,
1723
+ "w_max": 1.5594379305839539,
1724
+ "w_mean": 1.1108087003231049,
1725
+ "w_min": 0.0,
1726
+ "w_std": 0.2227778285741806
1727
+ },
1728
+ {
1729
+ "completion_length": 2405.6354598999023,
1730
+ "cov_mean": 9.158174270851305e-05,
1731
+ "cov_std": 0.34057875350117683,
1732
+ "entropy": 0.326171875,
1733
+ "epoch": 0.09028571428571429,
1734
+ "grad_norm": 0.4699649214744568,
1735
+ "kl": 0.0014543533325195312,
1736
+ "learning_rate": 2.1558482853517253e-07,
1737
+ "loss": 0.0159,
1738
+ "reward": 0.8750000223517418,
1739
+ "reward_std": 0.34952742606401443,
1740
+ "rewards/accuracy_reward": 0.250000006519258,
1741
+ "rewards/format_reward": 0.6250000298023224,
1742
+ "step": 79,
1743
+ "w_high_ratio": 0.11352726817131042,
1744
+ "w_low_ratio": 0.03820635436568409,
1745
+ "w_max": 1.81002277135849,
1746
+ "w_mean": 1.309929370880127,
1747
+ "w_min": 0.0,
1748
+ "w_std": 0.2257107077166438
1749
+ },
1750
+ {
1751
+ "completion_length": 3596.3750610351562,
1752
+ "cov_mean": 7.016459858277813e-05,
1753
+ "cov_std": 0.21266964823007584,
1754
+ "entropy": 0.46337890625,
1755
+ "epoch": 0.09142857142857143,
1756
+ "grad_norm": 0.224221870303154,
1757
+ "kl": 0.0014083385467529297,
1758
+ "learning_rate": 2.0528000059645995e-07,
1759
+ "loss": 0.0048,
1760
+ "reward": 0.5520833348855376,
1761
+ "reward_std": 0.33429770916700363,
1762
+ "rewards/accuracy_reward": 0.20833333395421505,
1763
+ "rewards/format_reward": 0.3437500009313226,
1764
+ "step": 80,
1765
+ "w_high_ratio": 0.0,
1766
+ "w_low_ratio": 0.029426061548292637,
1767
+ "w_max": 1.2386472523212433,
1768
+ "w_mean": 1.0539521872997284,
1769
+ "w_min": 0.25,
1770
+ "w_std": 0.11237771809101105
1771
+ },
1772
+ {
1773
+ "completion_length": 3377.5521240234375,
1774
+ "cov_mean": 1.5553408047708217e-05,
1775
+ "cov_std": 0.22893034294247627,
1776
+ "entropy": 0.52197265625,
1777
+ "epoch": 0.09257142857142857,
1778
+ "grad_norm": 0.27431702613830566,
1779
+ "kl": 0.0027017593383789062,
1780
+ "learning_rate": 1.9539516087697517e-07,
1781
+ "loss": 0.0442,
1782
+ "reward": 0.3125000027939677,
1783
+ "reward_std": 0.3139677047729492,
1784
+ "rewards/accuracy_reward": 0.0833333358168602,
1785
+ "rewards/format_reward": 0.22916666697710752,
1786
+ "step": 81,
1787
+ "w_high_ratio": 0.0,
1788
+ "w_low_ratio": 0.03366563096642494,
1789
+ "w_max": 1.3624085187911987,
1790
+ "w_mean": 1.1045191884040833,
1791
+ "w_min": 0.25,
1792
+ "w_std": 0.13904165104031563
1793
+ },
1794
+ {
1795
+ "completion_length": 3133.4583740234375,
1796
+ "cov_mean": -1.667316132625274e-05,
1797
+ "cov_std": 0.27454316802322865,
1798
+ "entropy": 0.408203125,
1799
+ "epoch": 0.09371428571428571,
1800
+ "grad_norm": 0.38514095544815063,
1801
+ "kl": 0.0017423629760742188,
1802
+ "learning_rate": 1.8594235253127372e-07,
1803
+ "loss": -0.0213,
1804
+ "reward": 0.5937500149011612,
1805
+ "reward_std": 0.35102832317352295,
1806
+ "rewards/accuracy_reward": 0.18750000465661287,
1807
+ "rewards/format_reward": 0.4062500074505806,
1808
+ "step": 82,
1809
+ "w_high_ratio": 0.0,
1810
+ "w_low_ratio": 0.034213229548186064,
1811
+ "w_max": 1.6117319464683533,
1812
+ "w_mean": 1.1832448840141296,
1813
+ "w_min": 9.42250974980716e-31,
1814
+ "w_std": 0.1813708432018757
1815
+ },
1816
+ {
1817
+ "completion_length": 3142.1771240234375,
1818
+ "cov_mean": 4.071597049914999e-05,
1819
+ "cov_std": 0.17724395170807838,
1820
+ "entropy": 0.45751953125,
1821
+ "epoch": 0.09485714285714286,
1822
+ "grad_norm": 0.3409688472747803,
1823
+ "kl": 0.001071929931640625,
1824
+ "learning_rate": 1.7693309235023127e-07,
1825
+ "loss": 0.0102,
1826
+ "reward": 0.38541667722165585,
1827
+ "reward_std": 0.20087094232439995,
1828
+ "rewards/accuracy_reward": 0.11458333395421505,
1829
+ "rewards/format_reward": 0.2708333386108279,
1830
+ "step": 83,
1831
+ "w_high_ratio": 0.0,
1832
+ "w_low_ratio": 0.023759393487125635,
1833
+ "w_max": 1.2598195374011993,
1834
+ "w_mean": 1.0758444666862488,
1835
+ "w_min": 0.25,
1836
+ "w_std": 0.10355101153254509
1837
+ },
1838
+ {
1839
+ "completion_length": 3299.4271850585938,
1840
+ "cov_mean": 1.610395520401653e-05,
1841
+ "cov_std": 0.3381447046995163,
1842
+ "entropy": 0.43505859375,
1843
+ "epoch": 0.096,
1844
+ "grad_norm": 0.3248639404773712,
1845
+ "kl": 0.0011587142944335938,
1846
+ "learning_rate": 1.6837835672960831e-07,
1847
+ "loss": -0.0241,
1848
+ "reward": 0.7500000298023224,
1849
+ "reward_std": 0.4175455719232559,
1850
+ "rewards/accuracy_reward": 0.3020833358168602,
1851
+ "rewards/format_reward": 0.4479166865348816,
1852
+ "step": 84,
1853
+ "w_high_ratio": 0.0,
1854
+ "w_low_ratio": 0.034003185108304024,
1855
+ "w_max": 1.4201359450817108,
1856
+ "w_mean": 1.1546694040298462,
1857
+ "w_min": 0.25,
1858
+ "w_std": 0.19938229769468307
1859
+ },
1860
+ {
1861
+ "completion_length": 3368.1146240234375,
1862
+ "cov_mean": 1.1272636584180873e-05,
1863
+ "cov_std": 0.3433619923889637,
1864
+ "entropy": 0.353515625,
1865
+ "epoch": 0.09714285714285714,
1866
+ "grad_norm": 0.2291695475578308,
1867
+ "kl": 0.0006651878356933594,
1868
+ "learning_rate": 1.6028856829700258e-07,
1869
+ "loss": 0.0156,
1870
+ "reward": 0.552083333954215,
1871
+ "reward_std": 0.5098650455474854,
1872
+ "rewards/accuracy_reward": 0.17708334233611822,
1873
+ "rewards/format_reward": 0.3750000102445483,
1874
+ "step": 85,
1875
+ "w_high_ratio": 0.05620652809739113,
1876
+ "w_low_ratio": 0.04038320202380419,
1877
+ "w_max": 1.5647819340229034,
1878
+ "w_mean": 1.2357110977172852,
1879
+ "w_min": 1.9246834407501362e-42,
1880
+ "w_std": 0.2121213562786579
1881
+ },
1882
+ {
1883
+ "completion_length": 3338.4479370117188,
1884
+ "cov_mean": 6.590973612219386e-06,
1885
+ "cov_std": 0.2853589281439781,
1886
+ "entropy": 0.45654296875,
1887
+ "epoch": 0.09828571428571428,
1888
+ "grad_norm": 0.3797212839126587,
1889
+ "kl": 0.0021305084228515625,
1890
+ "learning_rate": 1.5267358321348285e-07,
1891
+ "loss": -0.0236,
1892
+ "reward": 0.5208333432674408,
1893
+ "reward_std": 0.299550324678421,
1894
+ "rewards/accuracy_reward": 0.1770833395421505,
1895
+ "rewards/format_reward": 0.34375,
1896
+ "step": 86,
1897
+ "w_high_ratio": 0.05717267841100693,
1898
+ "w_low_ratio": 0.02882188046351075,
1899
+ "w_max": 1.963057816028595,
1900
+ "w_mean": 1.2316765785217285,
1901
+ "w_min": 0.25,
1902
+ "w_std": 0.1924368217587471
1903
+ },
1904
+ {
1905
+ "completion_length": 3156.2188110351562,
1906
+ "cov_mean": 8.500453259330243e-05,
1907
+ "cov_std": 0.5130489468574524,
1908
+ "entropy": 0.5185546875,
1909
+ "epoch": 0.09942857142857142,
1910
+ "grad_norm": 0.46258026361465454,
1911
+ "kl": 0.00295257568359375,
1912
+ "learning_rate": 1.4554267916537495e-07,
1913
+ "loss": -0.0212,
1914
+ "reward": 0.5833333432674408,
1915
+ "reward_std": 0.5054471418261528,
1916
+ "rewards/accuracy_reward": 0.14583333674818277,
1917
+ "rewards/format_reward": 0.4375000074505806,
1918
+ "step": 87,
1919
+ "w_high_ratio": 0.13361193798482418,
1920
+ "w_low_ratio": 0.05879375524818897,
1921
+ "w_max": 1.8866381645202637,
1922
+ "w_mean": 1.3355560600757599,
1923
+ "w_min": 0.0,
1924
+ "w_std": 0.33218564838171005
1925
+ },
1926
+ {
1927
+ "completion_length": 2856.8959350585938,
1928
+ "cov_mean": -5.945291923126206e-05,
1929
+ "cov_std": 0.5394940003752708,
1930
+ "entropy": 0.44140625,
1931
+ "epoch": 0.10057142857142858,
1932
+ "grad_norm": 0.890120804309845,
1933
+ "kl": 0.02105236053466797,
1934
+ "learning_rate": 1.3890454406082956e-07,
1935
+ "loss": -0.0362,
1936
+ "reward": 0.8958333730697632,
1937
+ "reward_std": 0.7090653479099274,
1938
+ "rewards/accuracy_reward": 0.3645833432674408,
1939
+ "rewards/format_reward": 0.5312500149011612,
1940
+ "step": 88,
1941
+ "w_high_ratio": 0.21772565320134163,
1942
+ "w_low_ratio": 0.049323974177241325,
1943
+ "w_max": 2.266294479370117,
1944
+ "w_mean": 1.4745216369628906,
1945
+ "w_min": 0.0,
1946
+ "w_std": 0.3573007620871067
1947
+ },
1948
+ {
1949
+ "completion_length": 3663.3021850585938,
1950
+ "cov_mean": -5.7599154160925536e-05,
1951
+ "cov_std": 0.2993681877851486,
1952
+ "entropy": 0.43505859375,
1953
+ "epoch": 0.10171428571428572,
1954
+ "grad_norm": 0.3626137971878052,
1955
+ "kl": 0.00238800048828125,
1956
+ "learning_rate": 1.3276726544494571e-07,
1957
+ "loss": 0.0099,
1958
+ "reward": 0.3645833469927311,
1959
+ "reward_std": 0.47493190318346024,
1960
+ "rewards/accuracy_reward": 0.12500000651925802,
1961
+ "rewards/format_reward": 0.23958334513008595,
1962
+ "step": 89,
1963
+ "w_high_ratio": 0.0,
1964
+ "w_low_ratio": 0.0442591137252748,
1965
+ "w_max": 1.4542400240898132,
1966
+ "w_mean": 1.079471230506897,
1967
+ "w_min": 0.0,
1968
+ "w_std": 0.18644139170646667
1969
+ },
1970
+ {
1971
+ "completion_length": 2692.8125610351562,
1972
+ "cov_mean": 4.647310163363727e-06,
1973
+ "cov_std": 0.3419715017080307,
1974
+ "entropy": 0.5390625,
1975
+ "epoch": 0.10285714285714286,
1976
+ "grad_norm": 0.36119693517684937,
1977
+ "kl": 0.0059909820556640625,
1978
+ "learning_rate": 1.2713832064634125e-07,
1979
+ "loss": 0.0442,
1980
+ "reward": 0.5729166697710752,
1981
+ "reward_std": 0.3239624537527561,
1982
+ "rewards/accuracy_reward": 0.06250000279396772,
1983
+ "rewards/format_reward": 0.5104166744276881,
1984
+ "step": 90,
1985
+ "w_high_ratio": 0.14126155525445938,
1986
+ "w_low_ratio": 0.02876184810884297,
1987
+ "w_max": 2.2300167977809906,
1988
+ "w_mean": 1.42653489112854,
1989
+ "w_min": 1.471363387541058e-44,
1990
+ "w_std": 0.21830591186881065
1991
+ },
1992
+ {
1993
+ "completion_length": 3451.5313110351562,
1994
+ "cov_mean": 1.8957917745865416e-05,
1995
+ "cov_std": 0.4000442400574684,
1996
+ "entropy": 0.45654296875,
1997
+ "epoch": 0.104,
1998
+ "grad_norm": 0.4046555459499359,
1999
+ "kl": 0.0024518966674804688,
2000
+ "learning_rate": 1.220245676671809e-07,
2001
+ "loss": -0.0665,
2002
+ "reward": 0.6770833730697632,
2003
+ "reward_std": 0.40345774590969086,
2004
+ "rewards/accuracy_reward": 0.260416679084301,
2005
+ "rewards/format_reward": 0.4166666865348816,
2006
+ "step": 91,
2007
+ "w_high_ratio": 0.06697376817464828,
2008
+ "w_low_ratio": 0.0342584103345871,
2009
+ "w_max": 1.4452155828475952,
2010
+ "w_mean": 1.1833300292491913,
2011
+ "w_min": 0.25,
2012
+ "w_std": 0.2148173823952675
2013
+ },
2014
+ {
2015
+ "completion_length": 3021.2188110351562,
2016
+ "cov_mean": -6.440309425670421e-05,
2017
+ "cov_std": 0.45011676102876663,
2018
+ "entropy": 0.38720703125,
2019
+ "epoch": 0.10514285714285715,
2020
+ "grad_norm": 0.44917768239974976,
2021
+ "kl": 0.007047891616821289,
2022
+ "learning_rate": 1.1743223682775649e-07,
2023
+ "loss": 0.0015,
2024
+ "reward": 0.7291666865348816,
2025
+ "reward_std": 0.5440528020262718,
2026
+ "rewards/accuracy_reward": 0.19791667256504297,
2027
+ "rewards/format_reward": 0.5312500074505806,
2028
+ "step": 92,
2029
+ "w_high_ratio": 0.060439030872657895,
2030
+ "w_low_ratio": 0.046312456484884024,
2031
+ "w_max": 2.021807760000229,
2032
+ "w_mean": 1.2982419431209564,
2033
+ "w_min": 0.0,
2034
+ "w_std": 0.2590548172593117
2035
+ },
2036
+ {
2037
+ "completion_length": 3865.6250610351562,
2038
+ "cov_mean": 1.7783165276341606e-05,
2039
+ "cov_std": 0.23176611587405205,
2040
+ "entropy": 0.52880859375,
2041
+ "epoch": 0.10628571428571429,
2042
+ "grad_norm": 0.2683483958244324,
2043
+ "kl": 0.002590179443359375,
2044
+ "learning_rate": 1.1336692317580158e-07,
2045
+ "loss": 0.0487,
2046
+ "reward": 0.1145833358168602,
2047
+ "reward_std": 0.27467070147395134,
2048
+ "rewards/accuracy_reward": 0.031250000931322575,
2049
+ "rewards/format_reward": 0.08333333488553762,
2050
+ "step": 93,
2051
+ "w_high_ratio": 0.0,
2052
+ "w_low_ratio": 0.03534638648852706,
2053
+ "w_max": 1.232116162776947,
2054
+ "w_mean": 1.0145132839679718,
2055
+ "w_min": 4.369921683443164e-37,
2056
+ "w_std": 0.12999964877963066
2057
+ },
2058
+ {
2059
+ "completion_length": 3234.9271850585938,
2060
+ "cov_mean": 1.053587311616866e-05,
2061
+ "cov_std": 0.20603487268090248,
2062
+ "entropy": 0.52490234375,
2063
+ "epoch": 0.10742857142857143,
2064
+ "grad_norm": 0.25439760088920593,
2065
+ "kl": 0.0029878616333007812,
2066
+ "learning_rate": 1.0983357966978745e-07,
2067
+ "loss": 0.0214,
2068
+ "reward": 0.4270833348855376,
2069
+ "reward_std": 0.17300797998905182,
2070
+ "rewards/accuracy_reward": 0.14583333395421505,
2071
+ "rewards/format_reward": 0.281250006519258,
2072
+ "step": 94,
2073
+ "w_high_ratio": 0.0574149414896965,
2074
+ "w_low_ratio": 0.021458564791828394,
2075
+ "w_max": 1.955292284488678,
2076
+ "w_mean": 1.259522944688797,
2077
+ "w_min": 0.25,
2078
+ "w_std": 0.14511701837182045
2079
+ },
2080
+ {
2081
+ "completion_length": 3710.2396240234375,
2082
+ "cov_mean": 1.0393607226433232e-05,
2083
+ "cov_std": 0.3523143231868744,
2084
+ "entropy": 0.42578125,
2085
+ "epoch": 0.10857142857142857,
2086
+ "grad_norm": 0.20859070122241974,
2087
+ "kl": 0.0004982948303222656,
2088
+ "learning_rate": 1.068365111445064e-07,
2089
+ "loss": 0.0209,
2090
+ "reward": 0.34375,
2091
+ "reward_std": 0.45749304443597794,
2092
+ "rewards/accuracy_reward": 0.11458333488553762,
2093
+ "rewards/format_reward": 0.2291666716337204,
2094
+ "step": 95,
2095
+ "w_high_ratio": 0.0,
2096
+ "w_low_ratio": 0.04463845491409302,
2097
+ "w_max": 1.3612166047096252,
2098
+ "w_mean": 1.0816101431846619,
2099
+ "w_min": 0.25,
2100
+ "w_std": 0.20336205884814262
2101
+ },
2102
+ {
2103
+ "completion_length": 3114.041748046875,
2104
+ "cov_mean": -6.73702224958106e-05,
2105
+ "cov_std": 0.36486806720495224,
2106
+ "entropy": 0.4013671875,
2107
+ "epoch": 0.10971428571428571,
2108
+ "grad_norm": 0.3484714925289154,
2109
+ "kl": 0.002678394317626953,
2110
+ "learning_rate": 1.0437936906629334e-07,
2111
+ "loss": -0.0171,
2112
+ "reward": 0.6979166865348816,
2113
+ "reward_std": 0.46633855253458023,
2114
+ "rewards/accuracy_reward": 0.26041666977107525,
2115
+ "rewards/format_reward": 0.4375000074505806,
2116
+ "step": 96,
2117
+ "w_high_ratio": 0.1971779614686966,
2118
+ "w_low_ratio": 0.04042420396581292,
2119
+ "w_max": 2.1471228301525116,
2120
+ "w_mean": 1.42051500082016,
2121
+ "w_min": 1.9142914218395184e-30,
2122
+ "w_std": 0.24729400128126144
2123
+ },
2124
+ {
2125
+ "completion_length": 3457.760498046875,
2126
+ "cov_mean": -1.2597635304700816e-05,
2127
+ "cov_std": 0.2245815545320511,
2128
+ "entropy": 0.41552734375,
2129
+ "epoch": 0.11085714285714286,
2130
+ "grad_norm": 0.24951718747615814,
2131
+ "kl": 0.0011386871337890625,
2132
+ "learning_rate": 1.0246514708427701e-07,
2133
+ "loss": -0.0394,
2134
+ "reward": 0.5625000055879354,
2135
+ "reward_std": 0.30785489082336426,
2136
+ "rewards/accuracy_reward": 0.23958333395421505,
2137
+ "rewards/format_reward": 0.32291666977107525,
2138
+ "step": 97,
2139
+ "w_high_ratio": 0.0,
2140
+ "w_low_ratio": 0.02458228124305606,
2141
+ "w_max": 1.5016373097896576,
2142
+ "w_mean": 1.1687426269054413,
2143
+ "w_min": 0.25,
2144
+ "w_std": 0.14521248266100883
2145
+ },
2146
+ {
2147
+ "completion_length": 3134.1251220703125,
2148
+ "cov_mean": 8.062847064138623e-05,
2149
+ "cov_std": 0.3362714499235153,
2150
+ "entropy": 0.384765625,
2151
+ "epoch": 0.112,
2152
+ "grad_norm": 0.3256186544895172,
2153
+ "kl": 0.0006456375122070312,
2154
+ "learning_rate": 1.0109617738307911e-07,
2155
+ "loss": -0.0102,
2156
+ "reward": 0.572916679084301,
2157
+ "reward_std": 0.3148062378168106,
2158
+ "rewards/accuracy_reward": 0.14583333395421505,
2159
+ "rewards/format_reward": 0.4270833432674408,
2160
+ "step": 98,
2161
+ "w_high_ratio": 0.00467883562669158,
2162
+ "w_low_ratio": 0.039873864501714706,
2163
+ "w_max": 1.795419156551361,
2164
+ "w_mean": 1.266855150461197,
2165
+ "w_min": 0.25,
2166
+ "w_std": 0.22055331617593765
2167
+ },
2168
+ {
2169
+ "completion_length": 3086.7395935058594,
2170
+ "cov_mean": 1.7768036741472315e-05,
2171
+ "cov_std": 0.15420139580965042,
2172
+ "entropy": 0.35205078125,
2173
+ "epoch": 0.11314285714285714,
2174
+ "grad_norm": 0.3351471722126007,
2175
+ "kl": 0.0013794898986816406,
2176
+ "learning_rate": 1.002741278414069e-07,
2177
+ "loss": 0.0477,
2178
+ "reward": 0.5000000223517418,
2179
+ "reward_std": 0.22891659289598465,
2180
+ "rewards/accuracy_reward": 0.1875000074505806,
2181
+ "rewards/format_reward": 0.3125,
2182
+ "step": 99,
2183
+ "w_high_ratio": 0.055527687072753906,
2184
+ "w_low_ratio": 0.02032439224421978,
2185
+ "w_max": 1.390455573797226,
2186
+ "w_mean": 1.1158559024333954,
2187
+ "w_min": 0.5,
2188
+ "w_std": 0.11466998234391212
2189
+ },
2190
+ {
2191
+ "completion_length": 3004.6354370117188,
2192
+ "cov_mean": 2.6663584918651395e-05,
2193
+ "cov_std": 0.3363041579723358,
2194
+ "entropy": 0.36181640625,
2195
+ "epoch": 0.11428571428571428,
2196
+ "grad_norm": 0.40786388516426086,
2197
+ "kl": 0.002788543701171875,
2198
+ "learning_rate": 1e-07,
2199
+ "loss": -0.0227,
2200
+ "reward": 0.7187500149011612,
2201
+ "reward_std": 0.4770490266382694,
2202
+ "rewards/accuracy_reward": 0.2500000074505806,
2203
+ "rewards/format_reward": 0.4687500149011612,
2204
+ "step": 100,
2205
+ "w_high_ratio": 0.01440071314573288,
2206
+ "w_low_ratio": 0.04455719655379653,
2207
+ "w_max": 1.8046407103538513,
2208
+ "w_mean": 1.2536596357822418,
2209
+ "w_min": 2.0318827732709848e-44,
2210
+ "w_std": 0.2585573047399521
2211
+ },
2212
+ {
2213
+ "epoch": 0.11428571428571428,
2214
+ "step": 100,
2215
+ "total_flos": 0.0,
2216
+ "train_loss": -0.004574703625403345,
2217
+ "train_runtime": 8361.7209,
2218
+ "train_samples_per_second": 1.148,
2219
+ "train_steps_per_second": 0.012
2220
+ }
2221
+ ],
2222
+ "logging_steps": 1,
2223
+ "max_steps": 100,
2224
+ "num_input_tokens_seen": 0,
2225
+ "num_train_epochs": 1,
2226
+ "save_steps": 50,
2227
+ "stateful_callbacks": {
2228
+ "TrainerControl": {
2229
+ "args": {
2230
+ "should_epoch_stop": false,
2231
+ "should_evaluate": false,
2232
+ "should_log": false,
2233
+ "should_save": true,
2234
+ "should_training_stop": true
2235
+ },
2236
+ "attributes": {}
2237
+ }
2238
+ },
2239
+ "total_flos": 0.0,
2240
+ "train_batch_size": 6,
2241
+ "trial_name": null,
2242
+ "trial_params": null
2243
+ }