Beijing-AISI/panda-bench

PandaBench是一个评估大型语言模型安全的基准数据集，专注于越狱攻击、防御机制和评估方法。该数据集提供了各种LLM在不同越狱攻击和防御机制下的评估结果，帮助研究人员比较防御机制的效果，分析系统的安全性和能力权衡，评估LLM的鲁棒性，以及开发新的防御算法。

PandaBench is a comprehensive benchmark for evaluating the safety of Large Language Models (LLMs), focusing on jailbreak attacks, defense mechanisms, and evaluation methodologies. The dataset provides evaluation results of various LLMs against different jailbreak attacks and defense mechanisms, assisting researchers in comparing the effectiveness of defense mechanisms, analyzing the trade-offs between safety and capability of defensive systems, evaluating the robustness of LLMs, and developing new defense algorithms.