deu05232/repro_msmarco-w-instructions_seed42-multipos-cross_batch

该数据集包含查询及其对应的正例和负例文档片段，用于训练信息检索或问答模型。每个样本包括查询ID、查询文本、正例（包含文档ID、解释、FollowIR分数、联合ID、文本和标题）、负例（文档ID、文本和标题）、仅指令、仅查询、是否有指令标志以及新的负例（包含解释和分数）。训练集有978,432个样本，总大小约9.6GB。This dataset contains queries with corresponding positive and negative passage samples, designed for training information retrieval or question answering models. Each sample includes query ID, query text, positive passages (with docid, explanation, followir_score, joint_id, text, title), negative passages (docid, text, title), only_instruction, only_query, has_instruction flag, and new_negatives (with explanation and scores). The training split has 978,432 examples, totaling approximately 9.6GB.