Lris47/MLLMstage3-textcaps-scienceqa-textvqa

该数据集包含图像和相关的多轮对话记录。每条数据包括一个唯一标识符、一张图像、一段多轮对话（每轮包含角色和内容）以及数据来源。数据集划分为训练集（106679条）、验证集（16595条）和测试集（17618条）。This dataset contains images and associated multi-turn conversation records. Each data entry includes a unique identifier, an image, a multi-turn conversation (each turn consists of a role and content), and a data source. The dataset is split into training (106,679 examples), validation (16,595 examples), and test (17,618 examples) sets.