llm-semantic-router/halueval-llm-spans

HaluEval LLM Spans数据集是从HaluEval摘要数据中提取的跨度级幻觉检测数据集。包含10,000个样本，带有LLM检测到的幻觉跨度和RAGTruth标准化提示。该数据集使用Qwen2.5-72B-Instruct将HaluEval的二元幻觉标签转换为细粒度的跨度级注释，提示已标准化为RAGTruth格式，以便与幻觉检测模型兼容。数据集主要用于训练和评估摘要任务中的幻觉检测器，以及研究LLM在摘要中的幻觉模式。

The HaluEval LLM Spans Dataset is a span-level hallucination detection dataset derived from HaluEval summarization data. It contains 10,000 samples with LLM-detected hallucination spans and RAGTruth-normalized prompts. This dataset converts HaluEvals binary hallucination labels into fine-grained span-level annotations using Qwen2.5-72B-Instruct. The prompts have been normalized to RAGTruth format for compatibility with hallucination detection models. The dataset is primarily intended for training and evaluating hallucination detectors in summarization tasks, as well as for studying LLM hallucination patterns in summarization.