Harmful Query on PKU-Safe
0.52ASRNSPO
Evaluation Results
| Method | Links | |
|---|---|---|
| NSPOBackbone=Qwen2.5-7B-Instruct2025.12 | 0.52 | |
| BFPOBackbone=Qwen2.5-7B-Instruct2025.12 | 1.35 | |
| BFPOBackbone=Llama3-8B-Instruct2025.12 | 1.86 | |
| NSPOBackbone=Llama3-8B-Instruct2025.12 | 2.23 | |
| DPO-SBackbone=Llama3-8B-Instruct2025.12 | 2.24 | |
| Qwen2.5-7B-InstructBackbone=Qwen2.5-7B-Instruct2025.12 | 2.95 | |
| DPO-SBackbone=Qwen2.5-7B-Instruct2025.12 | 3.11 | |
| PeCANBackbone=Llama3-8B-Instruct2025.12 | 3.22 | |
| DPO-HBackbone=Qwen2.5-7B-Instruct2025.12 | 3.26 | |
| Llama3-8B-InstructBackbone=Llama3-8B-Instruct2025.12 | 3.29 | |
| W-DOORBackbone=Llama3-8B-Instruct2025.12 | 3.31 | |
| MoCANBackbone=Llama3-8B-Instruct2025.12 | 3.38 | |
| MoCANBackbone=Qwen2.5-7B-Instruct2025.12 | 3.51 | |
| PeCANBackbone=Qwen2.5-7B-Instruct2025.12 | 3.55 | |
| DPO-MixBackbone=Qwen2.5-7B-Instruct2025.12 | 3.71 | |
| W-DOORBackbone=Qwen2.5-7B-Instruct2025.12 | 3.72 | |
| SafeRLHFBackbone=Qwen2.5-7B-Instruct2025.12 | 3.86 | |
| DPO-MixBackbone=Llama3-8B-Instruct2025.12 | 3.96 | |
| SafeRLHFBackbone=Llama3-8B-Instruct2025.12 | 6.5 | |
| DPO-HBackbone=Llama3-8B-Instruct2025.12 | 30.23 |