agentlans/BAAI-Infinity-Instruct

BAAI/Infinity-Instruct核心数据集是[BAAI/Infinity-Instruct](https://huggingface.co/datasets/BAAI/Infinity-Instruct)的7M_Core配置的非官方修订版。该子集代表了Infinity Instruct训练中使用的核心数据。据作者所述，7M核心数据集“仅使用1.4M指令就达到了完整数据集性能的95.7%”。更新内容包括：修复了对话顺序和标签问题，移除了空白、去重并打乱了样本，删除了URL、电子邮件和电话号码，并添加了已验证的语言检测列。需要注意的是，删除操作可能会影响某些代码、计算或Markdown渲染。

The BAAI/Infinity-Instruct Core Dataset is an unofficial revision of the [BAAI/Infinity-Instruct](https://huggingface.co/datasets/BAAI/Infinity-Instruct) `7M_Core` configuration. This subset represents core data used in Infinity Instruct training. As noted by the authors, the 7M Core achieves 95.7% of the full dataset’s performance with only 1.4M instructions. Updates include fixing conversation order and tag issues, removing blanks, deduplicating, and shuffling samples, redacting URLs, emails, and phone numbers, and adding a `language` column for verified language detection. Note that redaction may affect some code, calculations, or Markdown rendering.