Scientific Reasoning on GPQA (avg.@8)
33.27Avg.@8PSFTwarm-up
Evaluation Results
| Method | Links | |
|---|---|---|
| PSFTwarm-upBackbone=Qwen2.5-7B-Instruct2025.08 | 33.27 | |
| PSFTBackbone=Qwen2.5-7B-Instruct2025.08 | 33.21 | |
| SFT-KLBackbone=Qwen2.5-7B-Instruct2025.08 | 32.95 | |
| SFTBackbone=Qwen2.5-7B-Instruct2025.08 | 32.89 | |
| BaseBackbone=Qwen2.5-7B-Instruct2025.08 | 31.38 | |
| PSFTBackbone=Llama3.1-8B-Instruct2025.08 | 26.89 | |
| BaseBackbone=Llama3.1-8B-Instruct2025.08 | 24.62 | |
| PSFTwarm-upBackbone=Llama3.1-8B-Instruct2025.08 | 23.99 | |
| SFTBackbone=Llama3.1-8B-Instruct2025.08 | 19.38 | |
| SFT-KLBackbone=Llama3.1-8B-Instruct2025.08 | 18.18 |