List your company

Harmful input detection on JailbreakBench Conv

100MCA AccuracyReGA
50.0863.047688.96Jun 2, 2025

Evaluation results

MethodLinks
ReGAModel=qwen2025.0610092
ReGAModel=baichuan2025.069898
ReGAModel=llama2025.069793
ReGAModel=mistral2025.069585
ReGAModel=vicuna2025.069494
ReGAModel=Average2025.068984
ReGAModel=koala2025.065243