Linear-Next/Linear-Next-Datasets

Linear Next基准使用了一系列高质量的数据集，包括用于通用语言建模任务的DCLM-pro数据集、涵盖多个教育领域内容的Cosmopedia-v2和Fineweb-edu数据集、用于代码理解和生成任务的The Stack v2数据集、包含数学内容和问题的Finemath数据集，以及专注于逻辑推理和问题解决的Natural Reasoning数据集。

Linear Next benchmark utilizes a collection of high-quality datasets, including the DCLM-pro dataset for general language modeling tasks, Cosmopedia-v2 and Fineweb-edu datasets covering a variety of educational content, The Stack v2 dataset for code understanding and generation tasks, the Finemath dataset containing mathematical content and problems, and the Natural Reasoning dataset focused on logical reasoning and problem-solving.