QuasarResearch/Quasar-RP-DPO-v0.1

Quasar-RP-DPO-v0.1数据集是一个角色扮演偏好数据集，主要用于训练和校准语言模型。它专注于提高创造性输出，如散文、诗歌和整体叙事创造力，并且也适用于开发奖励模型和偏好优化技术（例如直接偏好优化DPO）。数据集包含2642个条目，每个条目由首选和被拒绝的响应组成，这些响应是通过抓取Chaiverse平台上六个表现最佳的语言模型的输出来收集的。请注意，数据集中包含成人内容和敏感材料，可能不适合所有受众。

The Quasar-RP-DPO-v0.1 dataset is a collection of roleplaying preference data primarily designed for training and aligning language models. It focuses on enhancing creative outputs such as prose, poetry, and overall narrative creativity, and is also suitable for developing reward models and preference optimization techniques like Direct Preference Optimization (DPO). The dataset contains 2642 entries, each comprising a pair of preferred (chosen) and less preferred (rejected) responses, collected by scraping the outputs of six top-performing language models on the Chaiverse platform. Please note that the dataset includes adult content and sensitive material, which may not be appropriate for all audiences.