Toxicity Mitigation on DailyDialog
1.16RTR (Backdoor Non-toxic)Attack w/o defense
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| Attack w/o defenseDefense setting=Attack w/o defense2025.07 | 1.16 | 62.18 | 13.76 | 54.78 | |
| Multi-level filterDefense setting=Multi-level filter2025.07 | 0.16 | 15.56 | 1.7 | 18.08 | |
| Optimus (CH)Defense setting=Optimus (CH)2025.07 | 0.02 | 0.04 | 0.02 | 0.64 |