Mathematical Reasoning on Minerva Math Dataset (avg.@8)
43.66Average Accuracy @8SFT
Evaluation Results
| Method | Links | |
|---|---|---|
| SFTBackbone=Qwen2.5-7B-Instruct2025.08 | 43.66 | |
| PSFTwarm-upBackbone=Qwen2.5-7B-Instruct2025.08 | 43.38 | |
| PSFTBackbone=Qwen2.5-7B-Instruct2025.08 | 43.33 | |
| SFT-KLBackbone=Qwen2.5-7B-Instruct2025.08 | 42.19 | |
| BaseBackbone=Qwen2.5-7B-Instruct2025.08 | 40.53 | |
| PSFTwarm-upBackbone=Llama3.1-8B-Instruct2025.08 | 33.64 | |
| PSFTBackbone=Llama3.1-8B-Instruct2025.08 | 32.4 | |
| SFTBackbone=Llama3.1-8B-Instruct2025.08 | 32.17 | |
| SFT-KLBackbone=Llama3.1-8B-Instruct2025.08 | 26.75 | |
| BaseBackbone=Llama3.1-8B-Instruct2025.08 | 25.87 |