UltraFeedback Binarized (UFB) dataset
收藏arXiv2025-09-30 收录
下载链接:
https://huggingface.co/datasets/HuggingFaceH4/ultrafeedback_binarized
下载链接
链接失效反馈官方服务:
资源简介:
该数据集是一个用于语言模型对齐任务的基准偏好数据集,包含了20,000个训练示例和1,024个评估示例。每个示例都由一个提示和两个回应组成:一个是被选中的回应,另一个是被拒绝的回应。该数据集的任务是利用人类偏好来进行语言模型的对齐。
This dataset is a benchmark preference dataset for language model alignment tasks, containing 20,000 training instances and 1,024 evaluation instances. Each instance consists of a prompt and two responses: one is the chosen response, and the other is the rejected response. The task of this dataset is to leverage human preferences to align language models.
提供机构:
Hugging Face



