Safety Evaluation on LatHarmful
7.07ASRLisa
Evaluation results
| Method | Links | ||
|---|---|---|---|
| LisaBackbone=Qwen-2.5-7B-Instruct2026.05 | 7.07 | 1.22 | |
| SafeGradBackbone=Qwen-2.5-7B-Instruct2026.05 | 7.68 | 1.24 | |
| Qwen-2.5-7B-InstructBackbone=Qwen-2.5-7B-Instruct2026.05 | 7.88 | 1.23 | |
| SPARDBackbone=Qwen-2.5-7B-Instruct2026.05 | 8.28 | 1.31 | |
| LisaBase Model=LLaMA-3.2-3B-Instruct, Defense Method=Lisa2026.05 | 9.9 | 1.36 | |
| PTSTBackbone=Qwen-2.5-7B-Instruct2026.05 | 10.71 | 1.35 | |
| SPARDBase Model=LLaMA-3.2-3B-Instruct, Defense Method=SPARD2026.05 | 11.31 | 1.43 | |
| LLaMA-3.2-3B-InstructBase Model=LLaMA-3.2-3B-Instruct2026.05 | 16.36 | 1.55 | |
| SFTBackbone=Qwen-2.5-7B-Instruct2026.05 | 25.66 | 1.96 | |
| SafeInstrBackbone=Qwen-2.5-7B-Instruct2026.05 | 26.06 | 1.95 | |
| SafeGradBase Model=LLaMA-3.2-3B-Instruct, Defense Method=SafeGrad2026.05 | 78.79 | 4.06 | |
| SafeInstrBase Model=LLaMA-3.2-3B-Instruct, Defense Method=SafeInstr2026.05 | 90.1 | 4.56 | |
| PTSTBase Model=LLaMA-3.2-3B-Instruct, Defense Method=PTST2026.05 | 97.78 | 4.82 | |
| SFTBase Model=LLaMA-3.2-3B-Instruct, Defense Method=SFT2026.05 | 98.99 | 4.91 |