Detoxification on SafeEdit
95.78Detoxification PerformanceSTA
Evaluation Results
| Method | Links | ||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|
| STAModel=Gemma-2-9b-it2025.05 | 95.78 | — | — | — | — | — | — | — | — | — | |
| CAAModel=Gemma-2-9b-it2025.05 | 91.48 | — | — | — | — | — | — | — | — | — | |
| SAE_AXBENCHModel=Gemma-2-9b-it2025.05 | 90.74 | — | — | — | — | — | — | — | — | — | |
| STAModel=Gemma-2-9b-pt2025.05 | 89.93 | — | — | — | — | — | — | — | — | — | |
| SAE_AXBENCHModel=Gemma-2-9b-pt2025.05 | 86.81 | — | — | — | — | — | — | — | — | — | |
| CAAModel=Gemma-2-9b-pt2025.05 | 85.78 | — | — | — | — | — | — | — | — | — | |
| Prompt_handModel=Gemma-2-9b-it2025.05 | 78.74 | — | — | — | — | — | — | — | — | — | |
| Prompt_autoModel=Gemma-2-9b-it2025.05 | 75.56 | — | — | — | — | — | — | — | — | — | |
| Prompt_handModel=Gemma-2-9b-pt2025.05 | 72.52 | — | — | — | — | — | — | — | — | — | |
| STAModel=Llama-3.1-8B2025.05 | 70.81 | — | — | — | — | — | — | — | — | — | |
| SAE_AXBENCHModel=Llama-3.1-8B2025.05 | 70.74 | — | — | — | — | — | — | — | — | — | |
| VanillaModel=Gemma-2-9b-it2025.05 | 70.37 | — | — | — | — | — | — | — | — | — | |
| CAAModel=Llama-3.1-8B2025.05 | 68.67 | — | — | — | — | — | — | — | — | — | |
| Prompt_autoModel=Gemma-2-9b-pt2025.05 | 64.15 | — | — | — | — | — | — | — | — | — | |
| Prompt_handModel=Llama-3.1-8B2025.05 | 63.7 | — | — | — | — | — | — | — | — | — | |
| VanillaModel=Gemma-2-9b-pt2025.05 | 62.3 | — | — | — | — | — | — | — | — | — | |
| Prompt_autoModel=Llama-3.1-8B2025.05 | 61.63 | — | — | — | — | — | — | — | — | — | |
| VanillaModel=Llama-3.1-8B2025.05 | 59.78 | — | — | — | — | — | — | — | — | — | |
| OrthoProjTarget Model=LLaMA-7B-Chat2024.06 | — | 50.7 | 57.9 | 50.2 | 47.5 | 67 | 50.1 | 74.9 | 65.7 | 66.4 | |
| OrthoProjTarget Model=LLaMA2-13B-Chat2024.06 | — | 51.1 | 82.6 | 50.6 | 72.4 | 52.3 | 58 | 51.4 | 65.1 | 75.5 | |
| PaCETarget Model=LLaMA-7B-Chat2024.06 | — | 69.6 | 46.2 | 58.2 | 75.3 | 94.2 | 62.3 | 80.8 | 72.8 | 88.3 | |
| PaCETarget Model=LLaMA2-13B-Chat2024.06 | — | 93.7 | 97.9 | 97.7 | 94.9 | 98.9 | 96.6 | 99.3 | 90.8 | 98.9 | |
| PromptingTarget Model=LLaMA-7B-Chat2024.06 | — | 82.5 | 47.3 | 57.8 | 65.2 | 75.1 | 54.8 | 72 | 72.4 | 56.1 | |
| PromptingTarget Model=LLaMA2-13B-Chat2024.06 | — | 35.8 | 68.3 | 59.3 | 52.5 | 73.5 | 23.4 | 78 | 71.1 | 66.5 | |
| VanillaTarget Model=LLaMA-7B-Chat2024.06 | — | 17.6 | 19.5 | 10.1 | 7.79 | 11.3 | 17.2 | 22.6 | 11.8 | 17.2 | |
| VanillaTarget Model=LLaMA2-13B-Chat2024.06 | — | 8.01 | 23.7 | 13.6 | 19.8 | 18.3 | 21.6 | 13.6 | 14 | 16.7 | |
| VecAddTarget Model=LLaMA-7B-Chat2024.06 | — | 50.9 | 58.9 | 59 | 53.9 | 66.1 | 55 | 60.7 | 61.7 | 66.4 | |
| VecAddTarget Model=LLaMA2-13B-Chat2024.06 | — | 76.6 | 71.4 | 70 | 64.3 | 87.2 | 66.9 | 47.4 | 74.5 | 71.1 |