Nalandadata/nalanda-image-qa

Nalanda Image QA数据集是一个包含1,000个多模态STEM科学问答对的样本，这些样本来自一个更大的22,679个科学问答对的集合。数据集覆盖物理、化学、生物和数学四个学科，包含三种类型的图像：问题图、解答图和选项图。每个样本包含唯一的标识符、学科领域、难度级别、图像路径以及对话记录。数据集用于训练Nalandadata/nalanda-image-vl模型，该模型在STEM科学问答任务上表现优异。Nalanda Image QA is a sample of 1,000 multimodal STEM science question-answer pairs with images, stratified from a larger collection of 22,679 science QA pairs. It covers Physics, Chemistry, Biology, and Mathematics with three types of images: question diagrams, solution illustrations, and option images. Each entry includes a unique question identifier, subject area, difficulty level, image paths, and conversation turns. The dataset was used to train Nalandadata/nalanda-image-vl, a model achieving significant improvements in STEM science QA tasks.