hotchpotch/japanese-splade-v1-hard-negatives
收藏Hugging Face2024-12-23 更新2024-12-21 收录
下载链接:
https://hf-mirror.com/datasets/hotchpotch/japanese-splade-v1-hard-negatives
下载链接
链接失效反馈官方服务:
资源简介:
该数据集用于训练日本語SPLADE v2模型,包括多个子数据集:mmarco-collection、mmarco-dataset、mqa-collection、mqa-dataset、msmarco-ja-collection和msmarco-ja-dataset。每个数据集都有详细的配置信息,如特征、分割、下载大小和数据集大小。数据集使用了SPLADE模型进行硬负样本挖掘,并使用BAAI/bge-reranker-v2-m3模型进行重排序评分。数据来源于hpprc/emb和hpprc/msmarco-ja,并继承了各自数据集的许可证。
This dataset is used for training the Japanese SPLADE v2 model and includes multiple sub-datasets (mmarco-collection, mmarco-dataset, mqa-collection, mqa-dataset, msmarco-ja-collection, msmarco-ja-dataset). These datasets are used for hard negative mining and reranking, employing the hotchpotch/japanese-splade-base-v1, v1_5 models and the BAAI/bge-reranker-v2-m3 model. The features of the datasets include text, original row ID, positive and negative sample IDs, and their similarity scores.
提供机构:
hotchpotch



