Refusal Evaluation on WildJailbreak Adversarial Harmful
89.45Refusal RateLow-Rank Combination
Evaluation Results
| Method | Links | |
|---|---|---|
| Low-Rank CombinationBackbone=DEEPSEEK R1 DISTILL LLAMA, Method=Low-Rank Combination2026.03 | 89.45 | |
| DEEPSEEK R1 DISTILL LLAMABackbone=DEEPSEEK R1 DISTILL LLAMA2026.03 | 87.65 | |
| Low-Rank CombinationBackbone=LLAMA 3 8B INSTRUCT, Method=Low-Rank Combination2026.03 | 78.8 | |
| LLAMA 3 8B INSTRUCTBackbone=LLAMA 3 8B INSTRUCT2026.03 | 77.4 | |
| Low-Rank CombinationBackbone=REFUSE-LLAMA, Method=Low-Rank Combination2026.03 | 49.1 | |
| Categorical SteeringBackbone=REFUSE-LLAMA, Method=Categorical Steering2026.03 | 44.6 | |
| REFUSE-LLAMABackbone=REFUSE-LLAMA2026.03 | 20.5 |