Autoformalization on Gaokao Formal
74.2Mean ScoreSFT+GRPO-0%
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| SFT+GRPO-0%sampling_strategy=Best-of-8, data_overlap=0%2026.04 | 74.2 | 80.9 | |
| Mathesis-7Bsampling_strategy=Best-of-82026.04 | 73.2 | 78.5 | |
| Kimina-7Bsampling_strategy=Best-of-82026.04 | 73.1 | 76.8 | |
| SFTsampling_strategy=Best-of-82026.04 | 71.9 | 80.5 | |
| SFT+GRPO-30%sampling_strategy=Best-of-8, data_overlap=30%2026.04 | 71.6 | 79.1 | |
| SFT+GRPO-100%sampling_strategy=Best-of-8, data_overlap=100%2026.04 | 71.6 | 80.4 | |
| GRPO-onlysampling_strategy=Best-of-82026.04 | 42.3 | 71.2 | |
| Basesampling_strategy=Best-of-82026.04 | 21.5 | 67.8 |