Text Detoxification on Black-box mode
0.67Similarity ScoreCP
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| CPBackbone=Llama-2-7b2024.01 | 0.67 | 23.2 | |
| CPBackbone=Mistral-7b-Ins.2024.01 | 0.62 | 5.9 | |
| CPBackbone=Falcon-7b2024.01 | 0.54 | 16.6 | |
| CPBackbone=Mistral-7b2024.01 | 0.44 | 9.9 | |
| BaselineBackbone=Mistral-7b2024.01 | 0.4 | 24.1 | |
| CPBackbone=OPT-2.7b2024.01 | 0.34 | 6.2 | |
| CPBackbone=OPT-6.7b2024.01 | 0.29 | 4.3 | |
| CPBackbone=Falcon-7b-Ins.2024.01 | 0.26 | 3.1 |