ResearchBenchmarksEmbedding Attack on HarmBench Mistral-7BFollow9MSSStandard1.723.615.57.39Feb 13, 2025Evaluation ResultsMethodMethodLinksMSSASSASRLFSStandardInitialization=StandardInitialization=Standard2025.02914.281001.39BaselineInitialization=BaselineInitialization=Baseline2025.0248.281002.225-CRIInitialization=25-CRIInitialization=25-CRI2025.022.52.761000.461-CRIInitialization=1-CRIInitialization=1-CRI2025.0222.481000.665-CRIInitialization=5-CRIInitialization=5-CRI2025.0222.31000.41