MM-Instruct
收藏arXiv2025-09-30 收录
下载链接：
https://github.com/jihaonew/MM-Instruct
下载链接
链接失效反馈官方服务：
资源简介：
该数据集是一个大规模、多样化的高质量视觉指导数据集，旨在提升大型多模态模型遵循指导的能力。该数据集融合了生成数据与LLaVA-1.5原始数据，用于指导调整，并展示了在遵循指导方面的能力提升。其规模涵盖了从两个数据集中抽取的40万张图片生成的23.4万条指导调整数据，任务重点在于多模态模型的指导遵循及评估。
This dataset is a large-scale, diverse high-quality visual instruction dataset designed to enhance the instruction-following capabilities of large multimodal models. It combines synthetic data and the original LLaVA-1.5 data for instruction tuning, and demonstrates improved performance in instruction following. It contains 234,000 instruction tuning samples generated from 400,000 images sampled from the two datasets, with the task focusing on the instruction following and evaluation of multimodal models.
