lsz05/MrTyDiJaRetrievalLite

MrTyDiJaRetrievalLite是一个基于TyDi构建的多语言基准数据集，专门用于文档检索任务。这是日语的轻量级版本，减少了语料库规模（93,382个文档），并使用了来自5个高性能模型的硬负样本。数据集的任务类别包括文本检索、多项选择QA和问答，领域涵盖百科全书、非小说和书面材料。数据集来源于sbintuitions/JMTEB-lite，语言为日语（jpn），许可证为apache-2.0。数据集包含三个配置：corpus（语料库）、qrels（查询相关文档）和queries（查询），每个配置都有详细的特征和分割信息。

Mr.TyDi is a multilingual benchmark dataset built on TyDi for document retrieval tasks. This is the lightweight Japanese version with a reduced corpus (93,382 documents) constructed using hard negatives from 5 high-performance models. The task categories include text retrieval, multiple-choice QA, and question answering, covering domains such as encyclopaedic, non-fiction, and written materials. The dataset is sourced from sbintuitions/JMTEB-lite, with the language being Japanese (jpn) and licensed under apache-2.0. The dataset includes three configurations: corpus, qrels, and queries, each with detailed features and split information.