tulu-3-sft-prompts-ultrafeedback
收藏资源简介:
该数据集是Tulu 3偏好混合的一部分,包含来自Tulu 3 SFT的提示和使用多种模型生成的内容。数据集的特征包括ID、提示、被选中的内容和角色以及被拒绝的内容和角色。数据集分为多个拆分,每个拆分包含不同数量的字节和示例。生成完成和偏好的方法结合了在线和离线数据,并使用Ultrafeedback模板和LLM判断进行偏好注释。数据集根据ODC-BY-1.0许可证授权,适用于研究和教育用途。
This dataset is a component of the Tulu 3 Preference Mixture, containing prompts sourced from Tulu 3 SFT and content generated by multiple models. The dataset's attributes include ID, prompt, chosen content and its associated role, as well as rejected content and its associated role. The dataset is divided into multiple splits, each with a distinct number of bytes and examples. The methods for generating completions and preference annotations combine online and offline data, and utilize Ultrafeedback templates and LLM judgments for preference labeling. This dataset is licensed under ODC-BY-1.0 and is intended for research and educational use.




