Benchmark Datasets for Offline LLM Applications

该数据集包含五个用于评估离线大型语言模型应用中对抗性查询的集合，它们作为系统提示使用。这些数据集由指令和示例对组成，用于生成提示。规模上，共有五个数据集，其任务是进行提示泄露攻击评估。

This dataset collection contains five sets that serve as system prompts for evaluating adversarial queries in offline large language model applications. Each of these datasets is composed of instruction and example pairs for generating prompts. In total, there are five such datasets, whose core task is to evaluate prompt leakage attacks.