text-machine-lab/diverse-not-short

Diverse, not Short数据集是一个用于提高语言模型响应词汇多样性的偏好学习数据集。数据集包括两种配置：Diverse-NS和Diverse-NS-Lite，前者使用熵作为多样性度量标准，ArmoRM分数作为质量度量标准；后者使用TTR测量多样性，MAAS指数测量质量。每个数据点都包括模型名称、系统提示、用户提示、被拒绝的响应和选中的响应。

The Diverse, not Short dataset is a preference learning dataset designed to enhance the lexical diversity of language model responses. It includes two configurations: Diverse-NS and Diverse-NS-Lite, with the former using entropy for diversity measurement and ArmoRM scores for quality, and the latter using TTR for diversity and MAAS index for quality. Each data point consists of the model name, system prompt, user prompt, rejected response, and chosen response.