openbmb/Ultra-FineWeb
收藏Hugging Face2025-12-10 更新2025-05-31 收录
下载链接：
https://hf-mirror.com/datasets/openbmb/Ultra-FineWeb
下载链接
链接失效反馈官方服务：
资源简介：
Ultra-FineWeb是一个大规模、高质量、且经过高效过滤的数据集。它通过提出的高效验证基于模型驱动的数据过滤管道，对FineWeb和Chinese FineWeb数据集进行处理，创建了约1万亿个英文token的Ultra-FineWeb-en数据集和约1200亿个中文token的Ultra-FineWeb-zh数据集。该数据集可以促进高质量的大型语言模型训练。
Ultra-FineWeb is a large-scale, high-quality, and efficiently-filtered dataset. It is created by applying an efficient verification-based high-quality filtering pipeline to the FineWeb and Chinese FineWeb datasets, resulting in the Ultra-FineWeb-en dataset with approximately 1 trillion English tokens and the Ultra-FineWeb-zh dataset with approximately 120 billion Chinese tokens. This dataset is designed to facilitate high-quality training for large language models.
提供机构：
openbmb
