pararater/paracore
收藏Hugging Face2025-10-13 更新2025-10-25 收录
下载链接：
https://hf-mirror.com/datasets/pararater/paracore
下载链接
链接失效反馈官方服务：
资源简介：
ParaCore是一个经过ParaRater过滤的平行语料库，它通过选择最有价值的平行语对来增强大型语言模型的跨语言迁移能力。该数据集包含英语和多种其他语言的平行数据，每种语言大约有10亿个token，英文数据来源于Common Crawl，翻译使用Qwen3-8B模型。
ParaCore is a parallel corpus filtered by ParaRater, which enhances the cross-lingual transfer ability of large language models by selecting the most valuable parallel pairs. The dataset includes parallel data in English and several other languages, with each language containing about 1 billion tokens. The English data is sourced from Common Crawl, and translation is performed using the Qwen3-8B model.
提供机构：
pararater
