ACIDE/user-vlm-instruct

Instruct数据集是一个用于指令微调的数据集，包含四个子数据集，旨在增强模型对用户特定查询的解释和响应能力，同时保持泛化能力并防止过拟合。这些子数据集分别关注用户面部特征、通用视觉问答、针对Alexa的问答以及作为正则化机制的NLE-VQA数据集。

The Instruct Dataset of User-VLM is a dataset for instruction finetuning, consisting of four sub-datasets designed to enhance the models ability to interpret and respond to user-specific queries while maintaining generalization and preventing overfitting. These sub-datasets focus on user facial features, general visual question answering, Alexa-specific QA, and the NLE-VQA dataset serving as a regularization mechanism.