karashiiro/the-uwu-denpaset

该数据集是从多个其他工具使用、代码生成和思维链（CoT）数据集中各抽取1k行样本组成的合集，总计5k个样本。数据集经过Haiku 4.5的整理和转换，便于使用。为了质量控制，整个数据集都进行了推理回填。数据集包含消息（messages）和工具（tools）等特性，消息字段包括内容（content）、角色（role）、思考（thinking）和工具调用（tool_calls）等。数据集的语言为英语，大小在1K到10K之间，包含训练集，大小为25239764字节，包含5000个样本。

This dataset is a sample of 1k rows from each of several other tool use, code generation, and CoT datasets, for a total of 5k samples. It has been curated and transformed with Haiku 4.5 for your convenience. Reasoning has been backfilled through the entire dataset for quality control purposes. The dataset includes features such as messages and tools, with messages containing fields like content, role, thinking, and tool_calls. The language is English, and the size is between 1K and 10K, containing a training set of 25239764 bytes with 5000 examples.