tamuzc/DisasterBench

DisasterBench是一个基准数据集，用于评估大型语言模型在类型化多智能体灾难响应工作流程中的结构化规划能力。它包含233个规划任务、26个任务接口智能体、一个类型化兼容图以及结构化可执行工作流程计划。每个数据示例包括任务ID、任务描述和结构化计划。DisasterBench is a benchmark for evaluating LLM structured planning over typed multi-agent disaster-response workflows. It consists of 233 planning tasks, 26 task-interface agents, a typed compatibility graph, and structured executable workflow plans. Each example includes task_id, task_desc, and structured_plan.