Jailbreak Defense on AdvBench (test)
1Harmful ScoreICD
Evaluation Results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| ICDModel=Vicuna2024.02 | 1 | 0 | — | — | — | |
| SafeDecodingModel=Vicuna2024.02 | 1 | 0 | — | — | — | |
| No DefenseModel=Llama22024.02 | 1 | 0 | — | — | — | |
| PPLModel=Llama22024.02 | 1 | 0 | — | — | — | |
| ParaphraseModel=Llama22024.02 | 1 | 2 | — | — | — | |
| RetokenizationModel=Llama22024.02 | 1 | 0 | — | — | — | |
| Self-ReminderModel=Llama22024.02 | 1 | 0 | — | — | — | |
| ICDModel=Llama22024.02 | 1 | 0 | — | — | — | |
| SafeDecodingModel=Llama22024.02 | 1 | 0 | — | — | — | |
| Self-ExaminationModel=Llama22024.02 | 1.04 | 0 | — | — | — | |
| Self-ReminderModel=Vicuna2024.02 | 1.06 | 0 | — | — | — | |
| Self-ExaminationModel=Vicuna2024.02 | 1.14 | 0 | — | — | — | |
| No DefenseModel=Vicuna2024.02 | 1.34 | 8 | — | — | — | |
| PPLModel=Vicuna2024.02 | 1.34 | 8 | — | — | — | |
| ParaphraseModel=Vicuna2024.02 | 1.58 | 14 | — | — | — | |
| RetokenizationModel=Vicuna2024.02 | 1.58 | 30 | — | — | — | |
| CaC-diverseModel=Vicuna2024.05 | — | — | 2 | 0 | 80 | |
| CaC-diverseModel=Llama22024.05 | — | — | 2 | 0 | 0 | |
| CaC-selfModel=Vicuna2024.05 | — | — | 2 | 0 | 88 | |
| CaC-selfModel=Llama22024.05 | — | — | 0 | 0 | 0 | |
| CaC-self-instructModel=Vicuna2024.05 | — | — | 1 | 0 | 29 | |
| CaC-self-instructModel=Llama22024.05 | — | — | 0 | 0 | 0 | |
| ICDModel=Vicuna2024.05 | — | — | 4 | 17 | 86 | |
| ICDModel=Llama22024.05 | — | — | 0 | 0 | 0 | |
| No defenseModel=Vicuna2024.05 | — | — | 95 | 90 | 91 | |
| No defenseModel=Llama22024.05 | — | — | 38 | 41 | 12 | |
| RAINModel=Vicuna2024.05 | — | — | 72 | 55 | — | |
| Self-reminderModel=Vicuna2024.05 | — | — | 94 | 59 | 88 | |
| Self-reminderModel=Llama22024.05 | — | — | 0 | 0 | 0 |