Harmful Query on SORRY
12.79ASR (%)PeCAN
Evaluation Results
| Method | Links | |
|---|---|---|
| PeCANBackbone=Llama3-8B-Instruct2025.12 | 12.79 | |
| DPO-SBackbone=Llama3-8B-Instruct2025.12 | 14.83 | |
| W-DOORBackbone=Qwen2.5-7B-Instruct2025.12 | 15 | |
| NSPOBackbone=Llama3-8B-Instruct2025.12 | 16.81 | |
| NSPOBackbone=Qwen2.5-7B-Instruct2025.12 | 19.54 | |
| DPO-MixBackbone=Llama3-8B-Instruct2025.12 | 23.26 | |
| BFPOBackbone=Llama3-8B-Instruct2025.12 | 29.73 | |
| W-DOORBackbone=Llama3-8B-Instruct2025.12 | 30.46 | |
| Llama3-8B-InstructBackbone=Llama3-8B-Instruct2025.12 | 34.15 | |
| BFPOBackbone=Qwen2.5-7B-Instruct2025.12 | 34.31 | |
| MoCANBackbone=Llama3-8B-Instruct2025.12 | 36.45 | |
| SafeRLHFBackbone=Qwen2.5-7B-Instruct2025.12 | 37.95 | |
| DPO-SBackbone=Qwen2.5-7B-Instruct2025.12 | 40.91 | |
| DPO-HBackbone=Qwen2.5-7B-Instruct2025.12 | 41.21 | |
| MoCANBackbone=Qwen2.5-7B-Instruct2025.12 | 41.33 | |
| DPO-MixBackbone=Qwen2.5-7B-Instruct2025.12 | 41.48 | |
| Qwen2.5-7B-InstructBackbone=Qwen2.5-7B-Instruct2025.12 | 41.88 | |
| SafeRLHFBackbone=Llama3-8B-Instruct2025.12 | 44.77 | |
| PeCANBackbone=Qwen2.5-7B-Instruct2025.12 | 45.42 | |
| DPO-HBackbone=Llama3-8B-Instruct2025.12 | 89.08 |