shareAI/ultra-chat_clean

该数据集是基于ultrachat原始数据集（英文，多轮对话）进行清洗过滤得到的。清洗规则包括过滤安全对齐文本、身份认知信息、代码、数学、符号和网址等内容。数据集提供了两种清洗后的样本文件：ultra-chat_clean.jsonl和ultra-chat_clean_common.jsonl，并且可以转换为sharegpt格式的多轮对话数据集。

This dataset is a cleaned and filtered version of the original ultrachat dataset (English, multi-turn dialogue). It mainly removes content such as refusal to answer, apologies, identity recognition information, code, mathematics, URLs, and special symbols. The dataset is divided into two versions: ultra-chat_clean.jsonl removes refusal to answer, apologies, and identity recognition information; ultra-chat_clean_common.jsonl further removes code, mathematics, URLs, and special symbols. The dataset format is firefly, which can be converted to the sharegpt format using the provided scripts.