ServiceNow-AI/Curriculum_DPO_preferences

Curri-DPO数据集包含两种任务类型：文本到文本生成和文本生成，使用英语。数据集规模在1K到10K之间。数据集由Ultrafeedback和Open-Assistant两个部分组成，每个部分都包含按难度分级的偏好对，分为简单、中等和困难三个等级。数据格式为用户和助手之间的对话，包括助手的偏好回应和非偏好回应。

The Curri-DPO dataset includes two task types: text2text-generation and text-generation, using English. The dataset size is between 1K and 10K. The dataset consists of two parts, Ultrafeedback and Open-Assistant, each containing preference pairs categorized by difficulty levels: easy, medium, and hard. The data format is structured as conversations between users and assistants, including the assistants preferred and non-preferred responses.