Harmful Query on HarmB
0.18ASR (%)NSPO
Evaluation Results
| Method | Links | |
|---|---|---|
| NSPOBackbone=Llama3-8B-Instruct2025.12 | 0.18 | |
| NSPOBackbone=Qwen2.5-7B-Instruct2025.12 | 0.5 | |
| W-DOORBackbone=Qwen2.5-7B-Instruct2025.12 | 0.51 | |
| W-DOORBackbone=Llama3-8B-Instruct2025.12 | 2.81 | |
| BFPOBackbone=Llama3-8B-Instruct2025.12 | 4.16 | |
| DPO-SBackbone=Llama3-8B-Instruct2025.12 | 5.67 | |
| MoCANBackbone=Llama3-8B-Instruct2025.12 | 6.83 | |
| Llama3-8B-InstructBackbone=Llama3-8B-Instruct2025.12 | 7.5 | |
| PeCANBackbone=Llama3-8B-Instruct2025.12 | 8.83 | |
| BFPOBackbone=Qwen2.5-7B-Instruct2025.12 | 11.33 | |
| Qwen2.5-7B-InstructBackbone=Qwen2.5-7B-Instruct2025.12 | 13.83 | |
| DPO-HBackbone=Qwen2.5-7B-Instruct2025.12 | 14.67 | |
| DPO-MixBackbone=Llama3-8B-Instruct2025.12 | 15.58 | |
| MoCANBackbone=Qwen2.5-7B-Instruct2025.12 | 17.83 | |
| DPO-SBackbone=Qwen2.5-7B-Instruct2025.12 | 18.33 | |
| PeCANBackbone=Qwen2.5-7B-Instruct2025.12 | 18.48 | |
| DPO-MixBackbone=Qwen2.5-7B-Instruct2025.12 | 18.67 | |
| SafeRLHFBackbone=Qwen2.5-7B-Instruct2025.12 | 32.25 | |
| SafeRLHFBackbone=Llama3-8B-Instruct2025.12 | 41.34 | |
| DPO-HBackbone=Llama3-8B-Instruct2025.12 | 57.32 |