ACIDE/user-vlm-lazy-dpo
收藏Hugging Face2025-02-15 更新2025-04-12 收录
下载链接：
https://hf-mirror.com/datasets/ACIDE/user-vlm-lazy-dpo
下载链接
链接失效反馈官方服务：
资源简介：
DPO数据集是为了增强User-VLM模型的公平性和道德一致性而设计的。它包括两个主要的子数据集：BiasVision-DPO和VLM-DPO，分别用于改进模型性能和减少偏见。BiasVision-DPO包含12K条 entries，旨在解决User-VLM中的性别、种族偏见等不适当的问题。VLM-DPO包含5.4K条通用DPO entries，用于正则化模型、防止过拟合和增强公平性。
The DPO dataset is designed to enhance the fairness and ethical alignment of User-VLM models. It consists of two primary sub-datasets: BiasVision-DPO and VLM-DPO, which are used to improve model performance and reduce biases respectively. BiasVision-DPO contains 12K entries aimed at addressing biases in User-VLM, particularly sexist, racist, controversial, and inappropriate questions. VLM-DPO comprises 5.4K general-purpose DPO entries that help regularize the model, mitigate overfitting, prevent catastrophic forgetting, and enhance fairness.
提供机构：
ACIDE
