Mathematical Reasoning on MATH 500 (avg.@8)
84.68Avg Success Rate (@8)PSFTwarm-up
Evaluation Results
| Method | Links | |
|---|---|---|
| PSFTwarm-upBackbone=Qwen2.5-7B-Instruct2025.08 | 84.68 | |
| SFTBackbone=Qwen2.5-7B-Instruct2025.08 | 84.1 | |
| SFT-KLBackbone=Qwen2.5-7B-Instruct2025.08 | 83.55 | |
| PSFTBackbone=Qwen2.5-7B-Instruct2025.08 | 83.35 | |
| BaseBackbone=Qwen2.5-7B-Instruct2025.08 | 75.05 | |
| PSFTwarm-upBackbone=Llama3.1-8B-Instruct2025.08 | 74.15 | |
| SFTBackbone=Llama3.1-8B-Instruct2025.08 | 72.6 | |
| PSFTBackbone=Llama3.1-8B-Instruct2025.08 | 71.98 | |
| SFT-KLBackbone=Llama3.1-8B-Instruct2025.08 | 69.83 | |
| BaseBackbone=Llama3.1-8B-Instruct2025.08 | 48.55 |