VIS-MPU-Agent/MMAT-1M

MMAT-1M是一个百万规模的多模态智能体调优数据集，通过整合五个公开的多模态问答数据集的子集构建而成。它支持链式思维推理、自我反思和动态工具使用等关键能力，旨在推动多模态模型在推理和基于工具的能力方面达到先进视觉语言系统的水平。MMAT-1M is a million-scale multimodal agent tuning dataset, constructed by consolidating subsets of five publicly available multimodal question-answer datasets. It supports key capabilities such as Chain-of-Thought reasoning, self-reflection, and dynamic tool usage, aiming to advance multimodal models toward enhanced reasoning and tool-based capabilities, aligning with the performance of state-of-the-art vision-language systems.