fluently-sets/ultrathink

Ultrathink数据集是一个为LLM模型进行通用SFT训练而设计的文本和Alpaca类型的数据集。包含391K行数据，使用Parquet文件格式存储。该数据集用英语编写，采用灵活的多许可证方式，主要许可证为MIT。数据集能够帮助模型学会像OpenAI/o1一样思考和推理，提高问题分析和解决能力，优化答案的结构和沉着。数据集是通过合并多个Alpaca数据集而成的，每个参与合并的数据集都有自己的许可证。

The Ultrathink dataset is a text and Alpaca type dataset designed for universal SFT training of LLM models. It contains 391K rows of data, stored in Parquet file format. The dataset is written in English and uses a flexible multi-license approach, with the main license being MIT. The dataset helps models learn to think and reason like OpenAI/o1, improve problem analysis and solving skills, and optimize the structure and composure of answers. The dataset was created by merging multiple Alpaca datasets, each with its own license.