Jailbreak Defense on Template
1.02Harmful ScoreSelf-Reminder
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| Self-ReminderModel=Llama22024.02 | 1.02 | 0 | |
| SafeDecodingModel=Llama22024.02 | 1.02 | 0 | |
| Self-ExaminationModel=Llama22024.02 | 1.03 | 0 | |
| RetokenizationModel=Llama22024.02 | 1.03 | 0.03 | |
| ICDModel=Llama22024.02 | 1.05 | 0 | |
| No DefenseModel=Llama22024.02 | 1.06 | 0 | |
| PPLModel=Llama22024.02 | 1.06 | 0 | |
| ParaphraseModel=Llama22024.02 | 1.1 | 0.11 | |
| Self-ExaminationModel=Vicuna2024.02 | 1.44 | 0.12 | |
| SafeDecodingModel=Vicuna2024.02 | 1.44 | 0.05 | |
| ParaphraseModel=Vicuna2024.02 | 2.31 | 0.32 | |
| RetokenizationModel=Vicuna2024.02 | 2.58 | 0.53 | |
| Self-ReminderModel=Vicuna2024.02 | 3.55 | 0.35 | |
| ICDModel=Vicuna2024.02 | 3.56 | 0.38 | |
| No DefenseModel=Vicuna2024.02 | 3.63 | 0.4 | |
| PPLModel=Vicuna2024.02 | 3.63 | 0.4 |