Refusal Evaluation on HarmfulQA
85.31Refusal RateCategorical Steering
Evaluation Results
| Method | Links | |
|---|---|---|
| Categorical SteeringBackbone=REFUSE-LLAMA, Method=Categorical Steering2026.03 | 85.31 | |
| Low-Rank CombinationBackbone=REFUSE-LLAMA, Method=Low-Rank Combination2026.03 | 75.87 | |
| Low-Rank CombinationBackbone=DEEPSEEK R1 DISTILL LLAMA, Method=Low-Rank Combination2026.03 | 75.61 | |
| DEEPSEEK R1 DISTILL LLAMABackbone=DEEPSEEK R1 DISTILL LLAMA2026.03 | 71.99 | |
| REFUSE-LLAMABackbone=REFUSE-LLAMA2026.03 | 66.07 | |
| Low-Rank CombinationBackbone=LLAMA 3 8B INSTRUCT, Method=Low-Rank Combination2026.03 | 61.58 | |
| LLAMA 3 8B INSTRUCTBackbone=LLAMA 3 8B INSTRUCT2026.03 | 60.15 |