ResearchBenchmarksHarmful input detection on RepE-HarmfulHarmful input detection on RepE-HarmfulFollow100MCA AccuracyReGA68.876.98593.1Jun 2, 2025Evaluation resultsMethodMethodLinksMCA AccuracyMFP AccuracyReGAModel=qwenModel=qwen2025.0610098ReGAModel=vicunaModel=vicuna2025.069898ReGAModel=mistralModel=mistral2025.069895ReGAModel=llamaModel=llama2025.069696ReGAModel=baichuanModel=baichuan2025.069696ReGAModel=AverageModel=Average2025.069391ReGAModel=koalaModel=koala2025.067062