Code Generation on HumanEval (Accuracy %)
86.6Accuracy (%)MIXSD
Evaluation results
| Method | Links | |
|---|---|---|
| MIXSDModel backbone=Qwen3-4B-It, Fine-tuning method=MIXSD, Mixing rate λ=02026.05 | 86.6 | |
| MIXSDModel backbone=Qwen3-4B-It, Fine-tuning method=MIXSD, Mixing rate λ=0.72026.05 | 86.6 | |
| BaseModel backbone=Qwen3-8B, Fine-tuning method=Base2026.05 | 86.6 | |
| BaseModel backbone=Qwen3-4B-It, Fine-tuning method=Base2026.05 | 86 | |
| MIXSDModel backbone=Qwen3-8B, Fine-tuning method=MIXSD, Mixing rate λ=0.32026.05 | 86 | |
| SFTModel backbone=Qwen3-4B-It, Fine-tuning method=SFT2026.05 | 84.8 | |
| MIXSDModel backbone=Qwen3-4B-It, Fine-tuning method=MIXSD, Mixing rate λ=0.32026.05 | 84.8 | |
| MIXSDModel backbone=Qwen3-4B-It, Fine-tuning method=MIXSD, Mixing rate λ=0.52026.05 | 84.8 | |
| OPSDModel backbone=Qwen3-8B, Fine-tuning method=OPSD2026.05 | 83.5 | |
| MIXSDModel backbone=Qwen3-8B, Fine-tuning method=MIXSD, Mixing rate λ=0.52026.05 | 83.5 | |
| MIXSDModel backbone=Qwen3-8B, Fine-tuning method=MIXSD, Mixing rate λ=02026.05 | 82.9 | |
| SFTModel backbone=Qwen3-8B, Fine-tuning method=SFT2026.05 | 81.7 | |
| MIXSDModel backbone=Qwen3-8B, Fine-tuning method=MIXSD, Mixing rate λ=0.72026.05 | 78.7 | |
| OPSDModel backbone=Qwen3-4B-It, Fine-tuning method=OPSD2026.05 | 76.2 | |
| BaseModel backbone=Qwen3-1.7B, Fine-tuning method=Base2026.05 | 60.4 | |
| MIXSDModel backbone=Qwen3-1.7B, Fine-tuning method=MIXSD, Mixing rate λ=0.52026.05 | 48.8 | |
| MIXSDModel backbone=Qwen3-1.7B, Fine-tuning method=MIXSD, Mixing rate λ=0.32026.05 | 45.1 | |
| MIXSDModel backbone=Qwen3-1.7B, Fine-tuning method=MIXSD, Mixing rate λ=0.72026.05 | 43.9 | |
| SFTModel backbone=Qwen3-1.7B, Fine-tuning method=SFT2026.05 | 11.6 | |
| OPSDModel backbone=Qwen3-1.7B, Fine-tuning method=OPSD2026.05 | 1.2 | |
| MIXSDModel backbone=Qwen3-1.7B, Fine-tuning method=MIXSD, Mixing rate λ=02026.05 | 1.2 |