ResearchBenchmarksLLM Safety Defense on BeaverTailsFollow10.39ASRMTSA-T39.825613.635317.44521.2547May 22, 2025Evaluation ResultsMethodMethodLinksASRMTSA-T3Target Model=Llama2-7B...Target Model=Llama2-7B-Chat, Iteration=32025.0510.39MTSA-T3Target Model=Zephyr-7B...Target Model=Zephyr-7B-Beta, Iteration=32025.0511.5BaselineTarget Model=Llama2-7B...Target Model=Llama2-7B-Chat2025.0521.5BaselineTarget Model=Zephyr-7B...Target Model=Zephyr-7B-Beta2025.0524.5