kilian-group/phantom-wiki-v1

PhantomWiki是一个合成数据集，生成一个虚构的宇宙，包含角色和事实，模仿粉丝维基的架构。这个数据集包括问题-答案对、文本语料库和事实数据库，用于评估大型语言模型的推理和检索能力。数据集设计为每次评估时生成一个新实例，以防止数据泄露和过度拟合。

PhantomWiki is a synthetic dataset that generates a fictional universe populated with characters and facts, mirroring the structure of a fan wiki. This dataset includes question-answer pairs, text corpora, and a fact database for evaluating the reasoning and retrieval capabilities of large language models. It is designed to generate a new instance for each evaluation to prevent data leakage and overfitting.