MMRC
收藏资源简介:
MMRC是一个多模态现实世界对话基准数据集,由上海人工智能实验室等机构创建。该数据集从现实世界场景中收集数据,包含5120个经过精心挑选的对话,每个对话都有28720个对应的人工标注问题,用于评估多模态大型语言模型在开放端对话中的六种核心能力,包括信息提取、跨轮推理、信息更新、图像管理、长期记忆回忆和拒绝回答。数据集涵盖了多种主题,确保了数据的多样性和代表性,适用于评估模型在现实对话中的表现。
MMRC is a multimodal real-world dialogue benchmark dataset created by institutions including Shanghai AI Laboratory. This dataset is collected from real-world scenarios, containing 5,120 carefully selected dialogues, each corresponding to 28,720 manually annotated questions. It is designed to evaluate six core capabilities of multimodal large language models in open-ended dialogue, including information extraction, cross-turn reasoning, information updating, image management, long-term memory recall, and refusal to answer. The dataset covers a wide range of topics to ensure data diversity and representativeness, making it suitable for assessing model performance in real-world dialogues.




