caskcsg/NExtLong-512K-dataset-subset
收藏Hugging Face2025-08-02 更新2025-04-12 收录
下载链接：
https://hf-mirror.com/datasets/caskcsg/NExtLong-512K-dataset-subset
下载链接
链接失效反馈官方服务：
资源简介：
NExtLong数据集是一系列通过负文档扩展方法合成的长上下文数据集，用于训练大型语言模型。这些数据集完全由合成数据组成，来源于fineweb-edu和Cosmopedia v2两个短文本数据集。数据集有不同的规模，包括64K、128K和512K合成数据，以及用于特定模型微调的子集和变体。
The NExtLong dataset is a series of long-context datasets synthesized using the Negative document Extension method, designed for training large language models. These datasets are entirely composed of synthetic data sourced from the fineweb-edu and Cosmopedia v2 short-text datasets. The datasets come in various sizes, including 64K, 128K, and 512K synthetic data, as well as subsets and variants for fine-tuning specific models.
提供机构：
caskcsg
