Adversarial Attack on Harmful Prompts
70.3ASRAutoDAN
Evaluation Results
| Method | Links | |
|---|---|---|
| AutoDANTarget Model=Zephyr-7b-r2d2, N=82025.10 | 70.3 | |
| REFDIVTarget Model=Zephyr-7b-r2d2, N=82025.10 | 81.9 | |
| AutoDANTarget Model=Llama3.1-70B, N=82025.10 | 85.8 | |
| AutoDANTarget Model=Phi-4-mini, N=82025.10 | 92.8 | |
| REFDIVTarget Model=Llama3.1-70B, N=82025.10 | 94.3 | |
| REFDIVTarget Model=Phi-4-mini, N=82025.10 | 95.7 | |
| AutoDANTarget Model=Vicuna-1.5-7b, N=82025.10 | 98.2 | |
| REFDIVTarget Model=Vicuna-1.5-7b, N=82025.10 | 98.6 |