ResearchBenchmarksSafety Alignment Evaluation on AdvBench (test)Follow1.6Harm Score (HS)AsFT1.0244.9128.812.688Jun 10, 2025Evaluation ResultsMethodMethodLinksHarm Score (HS)Failure Rate (FA)AsFTUtility Task=AGNEWSUtility Task=AGNEWS2025.061.683.7Safe LoRAUtility Task=AGNEWSUtility Task=AGNEWS2025.06482.3SafeInstrUtility Task=AGNEWSUtility Task=AGNEWS2025.064.484.4Lisa-alignedUtility Task=AGNEWSUtility Task=AGNEWS2025.064.882.6Lisa-baseUtility Task=AGNEWSUtility Task=AGNEWS2025.067.683.9SFTUtility Task=AGNEWSUtility Task=AGNEWS2025.0611.283.9BEAUtility Task=AGNEWSUtility Task=AGNEWS2025.061683.5