Over-refusal Evaluation on CoCoNot Contrast
1.58Over-refusal RateCategorical Steering
Evaluation Results
| Method | Links | |
|---|---|---|
| Categorical SteeringBackbone=REFUSE-LLAMA, Method=Categorical Steering2026.03 | 1.58 | |
| Low-Rank CombinationBackbone=LLAMA 3 8B INSTRUCT, Method=Low-Rank Combination2026.03 | 2.37 | |
| LLAMA 3 8B INSTRUCTBackbone=LLAMA 3 8B INSTRUCT2026.03 | 3.69 | |
| Low-Rank CombinationBackbone=REFUSE-LLAMA, Method=Low-Rank Combination2026.03 | 7.12 | |
| Low-Rank CombinationBackbone=DEEPSEEK R1 DISTILL LLAMA, Method=Low-Rank Combination2026.03 | 9.5 | |
| REFUSE-LLAMABackbone=REFUSE-LLAMA2026.03 | 11.87 | |
| DEEPSEEK R1 DISTILL LLAMABackbone=DEEPSEEK R1 DISTILL LLAMA2026.03 | 12.66 |