Triangle104/Atsunori-HelpSteer2-DPO
收藏Hugging Face2025-02-11 更新2025-02-15 收录
下载链接：
https://hf-mirror.com/datasets/Triangle104/Atsunori-HelpSteer2-DPO
下载链接
链接失效反馈官方服务：
资源简介：
HelpSteer2-DPO数据集是基于原始nvidia/HelpSteer2数据集转换而来，它将回应按照帮助性评分转换为偏好对，用于训练DPO（Difference Policy Optimization）。数据集包括7221个训练样本和373个验证样本，每个样本包含一个提示和两个回应，其中一个被选中作为更有帮助的回应，另一个被拒绝。
The HelpSteer2-DPO dataset is a conversion from the original nvidia/HelpSteer2 dataset into preference pairs based on the helpfulness score for training DPO (Difference Policy Optimization). It includes 7,221 training samples and 373 validation samples, each containing a prompt and two responses, one chosen as the more helpful response and the other rejected.
提供机构：
Triangle104
