Code Generation on HumanEval (HE, HE+, pass@8)
69.5HE ScoreDFT
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| DFTBackbone=Qwen-Math-7B2026.05 | 69.5 | 64 | 70.3 | |
| InfoSFTBackbone=Qwen-Math-7B2026.05 | 68.9 | 64 | 71.9 | |
| SFTBackbone=Qwen-Math-7B2026.05 | 66.5 | 59.4 | 71.6 | |
| baseBackbone=Qwen-Math-7B2026.05 | 65.2 | 59.8 | 71.5 | |
| DFTBackbone=Qwen-Math-1.5B2026.05 | 46.7 | 40.3 | 58.9 | |
| InfoSFTBackbone=Qwen-Math-1.5B2026.05 | 46.7 | 41.2 | 58.8 | |
| InfoSFTBackbone=Llama-3.1-8B2026.05 | 44.8 | 37.8 | 64.6 | |
| SFTBackbone=Qwen-Math-1.5B2026.05 | 43 | 36.6 | 58.8 | |
| DFTBackbone=Llama-3.1-8B2026.05 | 41.8 | 35 | 54.4 | |
| baseBackbone=Qwen-Math-1.5B2026.05 | 40.9 | 35.4 | 56.7 | |
| SFTBackbone=Llama-3.1-8B2026.05 | 40.9 | 34.5 | 66.3 | |
| baseBackbone=Llama-3.1-8B2026.05 | 38.4 | 31.1 | 66 |