Malicious Query Safety Evaluation on AdvBench
100Refusal RateSSD
Evaluation Results
| Method | Links | |
|---|---|---|
| SSDBase Model=Qwen32026.04 | 100 | |
| SSDBase Model=Llama32026.04 | 99.81 | |
| SSDBase Model=Gemma22026.04 | 99.81 | |
| DefaultBase Model=Qwen32026.04 | 99.81 | |
| DefaultBase Model=Gemma22026.04 | 99.62 | |
| AdaCDBase Model=Gemma22026.04 | 99.62 | |
| PromptBase Model=Qwen32026.04 | 99.42 | |
| PromptBase Model=Gemma22026.04 | 99.23 | |
| AdaCDBase Model=Qwen32026.04 | 99.04 | |
| DefaultBase Model=Llama32026.04 | 98.85 | |
| PromptBase Model=Llama32026.04 | 98.08 | |
| SurgicalBase Model=Qwen32026.04 | 97.5 | |
| AdaCDBase Model=Llama32026.04 | 97.31 | |
| SurgicalBase Model=Gemma22026.04 | 97.12 | |
| SurgicalBase Model=Llama32026.04 | 95.96 | |
| SelfCDBase Model=Gemma22026.04 | 94.62 | |
| SelfCDBase Model=Qwen32026.04 | 93.65 | |
| SelfCDBase Model=Llama32026.04 | 91.54 |