Mathematical Reasoning on MATH500 (acc@1, pass@8)
69.7Accuracy @ 1InfoSFT
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| InfoSFTBackbone=Qwen-Math-7B2026.05 | 69.7 | 87.2 | |
| InfoSFTBackbone=Qwen-Math-1.5B2026.05 | 66.2 | 84.3 | |
| DFTBackbone=Qwen-Math-7B2026.05 | 65.4 | 83.2 | |
| SFTBackbone=Qwen-Math-1.5B2026.05 | 61.6 | 81.1 | |
| DFTBackbone=Qwen-Math-1.5B2026.05 | 59.2 | 77 | |
| SFTBackbone=Qwen-Math-7B2026.05 | 53.4 | 84.4 | |
| baseBackbone=Qwen-Math-7B2026.05 | 52.8 | 82.9 | |
| baseBackbone=Qwen-Math-1.5B2026.05 | 33.2 | 77.5 | |
| InfoSFTBackbone=Llama-3.1-8B2026.05 | 27.8 | 58.8 | |
| SFTBackbone=Llama-3.1-8B2026.05 | 24 | 57.4 | |
| DFTBackbone=Llama-3.1-8B2026.05 | 15.5 | 29.6 | |
| baseBackbone=Llama-3.1-8B2026.05 | 3.2 | 20.5 |