List your company

Jailbreak Defense on AdvBench (test)

1Harmful ScoreICD
0.97681.13341.291.4466Feb 14, 2024

Evaluation results

MethodLinks
ICDModel=Vicuna2024.0210———
SafeDecodingModel=Vicuna2024.0210———
No DefenseModel=Llama22024.0210———
PPLModel=Llama22024.0210———
ParaphraseModel=Llama22024.0212———
RetokenizationModel=Llama22024.0210———
Self-ReminderModel=Llama22024.0210———
ICDModel=Llama22024.0210———
SafeDecodingModel=Llama22024.0210———
Self-ExaminationModel=Llama22024.021.040———
Self-ReminderModel=Vicuna2024.021.060———
Self-ExaminationModel=Vicuna2024.021.140———
No DefenseModel=Vicuna2024.021.348———
PPLModel=Vicuna2024.021.348———
ParaphraseModel=Vicuna2024.021.5814———
RetokenizationModel=Vicuna2024.021.5830———
CaC-diverseModel=Vicuna2024.05——2080
CaC-diverseModel=Llama22024.05——200
CaC-selfModel=Vicuna2024.05——2088
CaC-selfModel=Llama22024.05——000
CaC-self-instructModel=Vicuna2024.05——1029
CaC-self-instructModel=Llama22024.05——000
ICDModel=Vicuna2024.05——41786
ICDModel=Llama22024.05——000
No defenseModel=Vicuna2024.05——959091
No defenseModel=Llama22024.05——384112
RAINModel=Vicuna2024.05——7255—
Self-reminderModel=Vicuna2024.05——945988
Self-reminderModel=Llama22024.05——000