canho/MSMarco_Negative_1k

该数据集名为MS MARCO Negative 1k，是从microsoft/ms_marco数据集的v1.1/train子集中随机抽取的1000个样本，并添加了negative_query和生成的negative_ans列。用于生成负面查询的文档是第一个可用的段落，如果没有可用的段落，则使用第一个非空段落。负面查询的类型包括500个explicit_negation和500个antonym。负面答案生成模型为gpt-4o，共写入了1000行数据。negative_ans是对negative_query的简洁生成预期回答。

This dataset is named MS MARCO Negative 1k and contains 1,000 random examples sampled from the microsoft/ms_marco datasets v1.1/train subset, with added negative_query and generated negative_ans columns. The document used for negative query generation is the first selected passage when available, otherwise the first non-empty passage. Negative query types include 500 explicit_negation and 500 antonym. The negative answer generation model is gpt-4o, and 1,000 rows were written. negative_ans is a concise generated expected answer to negative_query.