List your company

Jailbreak Defense on HEX-PHI

1.74Harmful ScoreRetokenization
0.97041.17021.371.5698Feb 14, 2024

Evaluation results

MethodLinks
RetokenizationModel=Vicuna2024.021.7433
ParaphraseModel=Vicuna2024.021.7123
Self-ExaminationModel=Vicuna2024.021.618
No DefenseModel=Vicuna2024.021.5817
PPLModel=Vicuna2024.021.5215
Self-ReminderModel=Vicuna2024.021.238
ICDModel=Vicuna2024.021.26
SafeDecodingModel=Vicuna2024.021.081
RetokenizationModel=Llama22024.021.0415
ICDModel=Llama22024.021.030
ParaphraseModel=Llama22024.021.023
No DefenseModel=Llama22024.021.012
PPLModel=Llama22024.021.012
Self-ExaminationModel=Llama22024.021.010
SafeDecodingModel=Llama22024.021.011
Self-ReminderModel=Llama22024.0210