deu05232/repro_msmarco-w-instructions_seed42

该数据集包含978,432个训练样本，每个样本由查询（query）、查询ID（query_id）、正例段落（positive_passages）和负例段落（negative_passages）组成。正例段落包括文档ID、解释、followir评分、联合ID、文本和标题；负例段落包括文档ID、文本和标题。此外，还包含仅指令字段（only_instruction）、仅查询字段（only_query）以及是否包含指令的布尔标志（has_instruction）。另外，还有一个new_negatives字段，结构与正例段落相同，可能用于增强负例。该数据集适用于信息检索、问答系统或相关领域的模型训练。The dataset contains 978,432 training samples, each consisting of a query, query_id, positive passages, and negative passages. Positive passages include docid, explanation, followir_score, joint_id, text, and title; negative passages include docid, text, and title. Additionally, it includes fields for only_instruction, only_query, and a boolean flag has_instruction. There is also a new_negatives field with the same structure as positive passages, potentially for enhanced negative sampling. This dataset is suitable for training models in information retrieval, question answering, or related tasks.