Refusal Evaluation on XSTest Unsafe
100Refusal RateDefault
Evaluation Results
| Method | Links | |
|---|---|---|
| DefaultBase Model=Llama32026.04 | 100 | |
| SSDBase Model=Llama32026.04 | 100 | |
| AdaCDBase Model=Llama32026.04 | 100 | |
| SSDBase Model=Gemma22026.04 | 100 | |
| AdaCDBase Model=Gemma22026.04 | 100 | |
| AdaCDBase Model=Qwen32026.04 | 100 | |
| DefaultBase Model=Gemma22026.04 | 99.5 | |
| PromptBase Model=Gemma22026.04 | 99.5 | |
| SurgicalBase Model=Gemma22026.04 | 99.5 | |
| DefaultBase Model=Qwen32026.04 | 99.5 | |
| SSDBase Model=Qwen32026.04 | 99.5 | |
| SurgicalBase Model=Qwen32026.04 | 99.5 | |
| Categorical SteeringBackbone=REFUSE-LLAMA, Method=Categorical Steering2026.03 | 99 | |
| PromptBase Model=Llama32026.04 | 99 | |
| SelfCDBase Model=Gemma22026.04 | 99 | |
| PromptBase Model=Qwen32026.04 | 99 | |
| SelfCDBase Model=Qwen32026.04 | 98.5 | |
| SurgicalBase Model=Llama32026.04 | 98 | |
| SelfCDBase Model=Llama32026.04 | 98 | |
| Low-Rank CombinationBackbone=REFUSE-LLAMA, Method=Low-Rank Combination2026.03 | 97 | |
| REFUSE-LLAMABackbone=REFUSE-LLAMA2026.03 | 94.5 | |
| Low-Rank CombinationBackbone=LLAMA 3 8B INSTRUCT, Method=Low-Rank Combination2026.03 | 90 | |
| LLAMA 3 8B INSTRUCTBackbone=LLAMA 3 8B INSTRUCT2026.03 | 89.5 | |
| Low-Rank CombinationBackbone=DEEPSEEK R1 DISTILL LLAMA, Method=Low-Rank Combination2026.03 | 83 | |
| DEEPSEEK R1 DISTILL LLAMABackbone=DEEPSEEK R1 DISTILL LLAMA2026.03 | 80.5 |