AimonLabs/HDM-Bench

HDM-Bench（虚构检测模型基准）是一个设计用于评估和比较语言模型在识别由语言模型生成的文本中的事实错误和虚构内容能力的基准。该基准包含1000个文本样本，其中包含人工标注的标签分割。每个示例包括LLM的响应、源上下文（如适用）以及详细注释，说明为什么某个陈述被认为是虚构的。HDM-Bench (Hallucination Detection Model Benchmark) is a benchmark designed to evaluate and compare the performance of models in identifying factual inaccuracies and hallucinations in text generated by language models. The benchmark consists of 1000 text samples across with a split containing human-annotated labels. Each example includes the LLM response, source context (when applicable), and detailed annotation explaining why a statement is considered a hallucination.