deu05232/repro_msmarco-w-instructions_seed42-multipos-subset_add_version

该数据集是一个用于信息检索任务的数据集，包含查询及其相关文档的配对信息。主要特征包括：查询ID、查询文本、正面段落列表（每个段落包含文档ID、解释、FollowIR评分、联合ID、文本和标题）、负面段落列表（每个段落包含文档ID、文本和标题）、仅指令文本、仅查询文本、是否包含指令的布尔标志，以及新负面段落列表（结构与正面段落类似）。数据集分为训练集，包含约102.8万个示例，总大小约为10.2GB，下载大小约为5.9GB。适用于训练和评估检索模型，特别是针对查询与文档相关性排序的任务。

This dataset is designed for information retrieval tasks, containing paired information of queries and their relevant documents. Key features include: query ID, query text, positive passages list (each with docid, explanation, followir_score, joint_id, text, and title), negative passages list (each with docid, text, and title), only instruction text, only query text, a boolean flag indicating if instruction is present, and a new negatives list (with similar structure to positive passages). The dataset is split into a training set with approximately 1.028 million examples, total size of about 10.2GB, and download size of about 5.9GB. It is suitable for training and evaluating retrieval models, particularly for tasks involving query-document relevance ranking.