List your company

Jailbreak Defense on Prefix-Probe-600 Llama2-7B

1GCG ScorePPL
0.93681.36341.792.2166Oct 23, 2025

Evaluation results

MethodLinks
PPLBase Model=Llama2-7B2025.1011.081.241.1412.161.27
RetokenizationBase Model=Llama2-7B2025.1011.121.41.21.0311.13
SafeDecodingBase Model=Llama2-7B2025.10111.24111.41.11
SAIDBase Model=Llama2-7B2025.10111.1611.0111.03
ICDBase Model=Llama2-7B2025.101.1411113.041.36
ParaphraseBase Model=Llama2-7B2025.101.161.061.021.081.021.121.08
Self-ExaminationBase Model=Llama2-7B2025.101.61.0811.1211.81.27
No DefenseBase Model=Llama2-7B2025.102.581.081.241.212.21.55