Red-team Query on HarmQA
0.11ASR (%)NSPO
Evaluation results
| Method | Links | |
|---|---|---|
| NSPOBackbone=Qwen2.5-7B-Instruct2025.12 | 0.11 | |
| NSPOBackbone=Llama3-8B-Instruct2025.12 | 0.12 | |
| W-DOORBackbone=Qwen2.5-7B-Instruct2025.12 | 0.31 | |
| BFPOBackbone=Llama3-8B-Instruct2025.12 | 0.34 | |
| W-DOORBackbone=Llama3-8B-Instruct2025.12 | 0.66 | |
| BFPOBackbone=Qwen2.5-7B-Instruct2025.12 | 0.74 | |
| PeCANBackbone=Llama3-8B-Instruct2025.12 | 1.13 | |
| DPO-SBackbone=Llama3-8B-Instruct2025.12 | 1.2 | |
| MoCANBackbone=Llama3-8B-Instruct2025.12 | 1.41 | |
| Llama3-8B-InstructBackbone=Llama3-8B-Instruct2025.12 | 1.89 | |
| SafeRLHFBackbone=Qwen2.5-7B-Instruct2025.12 | 2.76 | |
| DPO-SBackbone=Qwen2.5-7B-Instruct2025.12 | 2.83 | |
| Qwen2.5-7B-InstructBackbone=Qwen2.5-7B-Instruct2025.12 | 3.26 | |
| DPO-MixBackbone=Llama3-8B-Instruct2025.12 | 3.4 | |
| PeCANBackbone=Qwen2.5-7B-Instruct2025.12 | 3.42 | |
| DPO-MixBackbone=Qwen2.5-7B-Instruct2025.12 | 3.52 | |
| MoCANBackbone=Qwen2.5-7B-Instruct2025.12 | 3.55 | |
| DPO-HBackbone=Qwen2.5-7B-Instruct2025.12 | 3.65 | |
| SafeRLHFBackbone=Llama3-8B-Instruct2025.12 | 4.72 | |
| DPO-HBackbone=Llama3-8B-Instruct2025.12 | 24.11 |