Safety Defense on Unsafe Prompts 7 categories
100Hate RateDTVI
Evaluation Results
| Method | Links | ||||||||
|---|---|---|---|---|---|---|---|---|---|
| DTVI2026.03 | 100 | 77.78 | 89.33 | 82.46 | 94.74 | 82.76 | 92.86 | 88.56 | |
| SLD-MAXHyperparameter setting=MAX2026.03 | 79.25 | 68.57 | 75 | 82.14 | 84.51 | 64.29 | 81.48 | 76.46 | |
| UCEconfiguration=Erasing seven unsafe concepts2026.03 | 41.18 | 8.57 | 24 | 31.58 | 8 | 20.69 | 14.81 | 21.26 | |
| ESDconfiguration=Erasing seven unsafe concepts2026.03 | 37.74 | -2.86 | 3.95 | 1.75 | -12.33 | -3.45 | 15.38 | 5.74 |