Closed-book recall on KGFACT SMALL (train)
100Recall (%)MIXSD
Evaluation Results
| Method | Links | |
|---|---|---|
| MIXSDModel backbone=Qwen3-1.7B, Fine-tuning method=MIXSD, Mixing rate λ=0.32026.05 | 100 | |
| SFTModel backbone=Qwen3-4B-It, Fine-tuning method=SFT2026.05 | 100 | |
| OPSDModel backbone=Qwen3-4B-It, Fine-tuning method=OPSD2026.05 | 100 | |
| SFTModel backbone=Qwen3-8B, Fine-tuning method=SFT2026.05 | 100 | |
| MIXSDModel backbone=Qwen3-8B, Fine-tuning method=MIXSD, Mixing rate λ=02026.05 | 100 | |
| SFTModel backbone=Qwen3-1.7B, Fine-tuning method=SFT2026.05 | 99 | |
| OPSDModel backbone=Qwen3-1.7B, Fine-tuning method=OPSD2026.05 | 99 | |
| MIXSDModel backbone=Qwen3-4B-It, Fine-tuning method=MIXSD, Mixing rate λ=02026.05 | 99 | |
| OPSDModel backbone=Qwen3-8B, Fine-tuning method=OPSD2026.05 | 99 | |
| MIXSDModel backbone=Qwen3-8B, Fine-tuning method=MIXSD, Mixing rate λ=0.32026.05 | 99 | |
| MIXSDModel backbone=Qwen3-1.7B, Fine-tuning method=MIXSD, Mixing rate λ=0.52026.05 | 97 | |
| MIXSDModel backbone=Qwen3-8B, Fine-tuning method=MIXSD, Mixing rate λ=0.52026.05 | 97 | |
| MIXSDModel backbone=Qwen3-1.7B, Fine-tuning method=MIXSD, Mixing rate λ=02026.05 | 96 | |
| MIXSDModel backbone=Qwen3-4B-It, Fine-tuning method=MIXSD, Mixing rate λ=0.32026.05 | 94 | |
| MIXSDModel backbone=Qwen3-4B-It, Fine-tuning method=MIXSD, Mixing rate λ=0.52026.05 | 89 | |
| MIXSDModel backbone=Qwen3-4B-It, Fine-tuning method=MIXSD, Mixing rate λ=0.72026.05 | 85 | |
| MIXSDModel backbone=Qwen3-8B, Fine-tuning method=MIXSD, Mixing rate λ=0.72026.05 | 73 | |
| MIXSDModel backbone=Qwen3-1.7B, Fine-tuning method=MIXSD, Mixing rate λ=0.72026.05 | 71 | |
| BaseModel backbone=Qwen3-1.7B, Fine-tuning method=Base2026.05 | 0 | |
| BaseModel backbone=Qwen3-4B-It, Fine-tuning method=Base2026.05 | 0 | |
| BaseModel backbone=Qwen3-8B, Fine-tuning method=Base2026.05 | 0 |