manifoldlabs/Infinity-Instruct

Infinity Instruct数据集是一个大规模、高质量的指令数据集，旨在提升模型在文本生成任务中的表现。数据集通过指令选择和指令进化两种策略进行迭代，包含基础数据集和聊天数据集，分别用于提升模型在代码、数学等下游任务中的表现和在实际对话场景中的指令遵循能力。数据集提供了详细的标签信息，如语言、能力类型、来源等，用户可以根据这些标签构建自己的数据子集。数据集包含多个版本，如3M、7M、0625等，涵盖了不同的训练数据量。

Infinity Instruct is a large-scale, high-quality instruction dataset designed to support model performance on various downstream tasks. The dataset includes multiple versions such as 3M, 7M, 0625, and Gen, each used for different tasks and model training. The dataset supports both English and Chinese languages, and each instruction data is labeled with information such as language, ability type, and source. The construction process of the dataset includes instruction selection and instruction evolution strategies to ensure the high quality and diversity of the data.