allenai/pixmo-ask-model-anything

PixMo-AskModelAnything是一个用于视觉语言模型指令调优的数据集，包含人类编写的关于多样化图像的长篇问答对。它是PixMo数据集集合的一部分，并用于训练Molmo系列模型。数据集中的每一行包含一个图像URL和一个问答对。图像URL可能会重复，因为许多图像有多个问答对。数据集还包括图像哈希值，以支持验证下载的图像是否与注释的图像匹配。

PixMo-AskModelAnything is an instruction-tuning dataset for vision-language models, containing human-authored question-answer pairs about diverse images with long-form answers. Each row in the dataset contains an image URL and a Q/A pair, with image URLs potentially repeating. The dataset also includes SHA-256 hashes of images to support image matching checks.