answerdotai/MMARCO-japanese-32-scored-triplets

该数据集包含多个数据分割，每个分割包含query_id、pos_id、neg_ids、pos_score和neg_scores等特征。数据分割包括M3_full_logits、M3_full_normalized、M3_320k_normalized、monot5_320k_normalized、jpbase_320k_normalized、jplarge_320k_normalized、jpsmall_320k_normalized、colbertv2_320k_normalized和M3jp_320k_normalized。每个分割的字节数和样本数各不相同。数据集的下载大小为3889910104字节，数据集大小为4561920000字节。该数据集与JaColBERTv2.5相关，后者是一种在资源受限的情况下优化多向量检索器以创建最先进的日语检索器的方法。

This dataset contains multiple splits, each with features such as query_id, pos_id, neg_ids, pos_score, and neg_scores. The splits include M3_full_logits, M3_full_normalized, M3_320k_normalized, monot5_320k_normalized, jpbase_320k_normalized, jplarge_320k_normalized, jpsmall_320k_normalized, colbertv2_320k_normalized, and M3jp_320k_normalized. Each split has varying numbers of bytes and examples. The download size of the dataset is 3889910104 bytes, and the dataset size is 4561920000 bytes. This dataset is related to JaColBERTv2.5, which is a method for optimizing multi-vector retrievers to create state-of-the-art Japanese retrievers under constrained resources.