Rombo-Org/Optimized_Reasoning

Optimized_Reasoning数据集是为了解决现代大型语言模型在处理推理任务时效率不高、消耗大量token的问题而创建的。该数据集通过在每条数据前添加一个推理标签，来指示模型是否需要进行推理，以此减少token的使用并增强模型在推理方面的能力。数据集分为两种类型，分别针对简单查询和复杂查询，并经过过滤以确保模型能够区分易难查询，同时降低训练成本。

The Optimized_Reasoning dataset was created to address the issue of modern large language models not being efficient at handling reasoning tasks, consuming a large number of tokens in the process. The dataset includes a reasoning tag at the beginning of each data line to indicate whether the model should engage in reasoning, thereby reducing token usage and enhancing the models ability to reason. The dataset is divided into two types, one for simple queries and another for complex queries, and has been filtered to ensure the model can distinguish between easy and difficult queries while also reducing training costs.