Safety Defense on Safety Evaluation Text-only
0.4Training Time (hr)EVA (text)
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| EVA (text)Base Model=Vicuna-7B-v1.52026.05 | 0.4 | 1 | 6.7 | |
| SafeDecodingBase Model=Vicuna-7B-v1.52026.05 | 0.6 | 1.07 | 10.7 | |
| Circuit BreakersBase Model=Vicuna-7B-v1.52026.05 | 1 | 1 | 7.3 | |
| LoRABase Model=Vicuna-7B-v1.52026.05 | 1.5 | 1 | 23.2 | |
| LEDBase Model=Vicuna-7B-v1.52026.05 | 6.2 | 1 | 8.8 |