Detoxification on AdvBench harmful behavior set
99.17Safety ScorePaCE
Evaluation Results
| Method | Links | |
|---|---|---|
| PaCEBackbone=LlaMA2-13B-Chat2024.06 | 99.17 | |
| OrthoProjBackbone=LlaMA2-13B-Chat2024.06 | 96.72 | |
| PaCEBackbone=LlaMA2-7B-Chat2024.06 | 96.65 | |
| VecAddBackbone=LlaMA2-13B-Chat2024.06 | 95.33 | |
| VecAddBackbone=LlaMA2-7B-Chat2024.06 | 94.51 | |
| PEBackbone=LlaMA2-13B-Chat2024.06 | 93.86 | |
| OrthoProjBackbone=LlaMA2-7B-Chat2024.06 | 92.81 | |
| PEBackbone=LlaMA2-7B-Chat2024.06 | 91.9 | |
| VanillaBackbone=LlaMA2-13B-Chat2024.06 | 18.04 | |
| VanillaBackbone=LlaMA2-7B-Chat2024.06 | 11.72 |