uukuguy/MindSpeed-Infinity-Instruct-7M

Infinity Instruct数据集是基于Infinity Instruct项目构建的，包含数百万条指令，旨在匹配MindSpeed-LLM的多轮对话微调格式。数据集分为基础数据集和聊天数据集，基础数据集通过从开源数据集中选择和迭代指令构建，聊天数据集则包含从少量高质量种子数据演化而来的约100万条指令。数据集采用cc-by-sa-4.0许可证。

The Infinity Instruct dataset is built upon the Infinity Instruct project, containing millions of instructions aimed at matching the multi-round dialogue finetuning format of MindSpeed-LLM. The dataset is divided into a foundational dataset and a chat dataset. The foundational dataset is constructed by selecting and iterating instructions from open-source datasets, while the chat dataset includes about 1 million instructions evolved from a small subset of high-quality seed data. The dataset is licensed under cc-by-sa-4.0.