Mathematical Reasoning on GSM, MATH, Minerva, Gaokao, Olympiad, College, Aime24, Amc23 (test)
54.6Average ScorePPO
Evaluation Results
| Method | Links | |
|---|---|---|
| PPOType=Online RL, Training Sample Size=400K, Base Model=Qwen2.5-Math-7B2025.05 | 54.6 | |
| SAI-DPOType=Offline RL, Training Sample Size=48K, Base Model=Qwen2.5-Math-7B2025.05 | 53.7 | |
| IDPOType=Offline RL, Training Sample Size=67K, Base Model=Qwen2.5-Math-7B2025.05 | 52 | |
| IRFTType=SFT, Training Sample Size=67K, Base Model=Qwen2.5-Math-7B2025.05 | 47.1 | |
| LIMOType=SFT, Training Sample Size=0.8K, Base Model=Qwen2.5-Math-7B2025.05 | 46.8 | |
| S1Type=SFT, Training Sample Size=1K, Base Model=Qwen2.5-Math-7B2025.05 | 45.9 |