Sweaterdog/GRaPE-Base-Mix
收藏Hugging Face2026-02-13 更新2025-10-25 收录
下载链接：
https://hf-mirror.com/datasets/Sweaterdog/GRaPE-Base-Mix
下载链接
链接失效反馈官方服务：
资源简介：
GRaPE Base Mix数据集用于初始化训练GRaPE基础模型，使其在编程、数学、音乐理论、指令遵循、工具使用和长文本上下文等多个领域具备会话和智能能力。该数据集包含最多137,000个token，由3M个来自不同数据集的示例组成，且示例已随机打乱。
The GRaPE Base Mix dataset is used for the initial training of the GRaPE base model, enabling it to be conversational and intelligent in multiple fields such as programming, mathematics, music theory, instruction following, tool usage, and long context. The dataset contains a maximum of 137,000 tokens and consists of 3M examples from various datasets, which have been shuffled for smooth training.
提供机构：
Sweaterdog
