YANAMI dataset targets stateful multi-turn multimodal interaction
The paper introduces YANAMI-400K, a dataset of 400,000 multi-turn multimodal dialogues containing more than 2 million images and 2.6 million turns across image fusion, editing, world modeling, and visual reasoning. It also presents YANAMI-Bench, with 400 multi-turn dialogues for evaluating dialogue-state understanding and visual realization. The authors report that fine-tuning BAGEL on YANAMI-400K improved editing, generation, and stateful interaction by 10.3%, 2.0%, and 13.23%, respectively.
YANAMI provides training and evaluation data focused on maintaining and updating multimodal dialogue state across turns, a capability the paper says existing unified multimodal models still struggle to handle.
Computer Science > Computer Vision and Pattern Recognition
arXiv:2611.05280 (cs)
[Submitted on 3 Nov 2026]
Title:YANAMI: A Large-Scale Multimodal Dataset and Benchmark for Multiturn Stateful Interaction
Authors:Shengyu Wang, Tengyue Nong, Junming Liu, Haosong Peng, Siyu Li, Ruiqi Lei, Fushuo Huo, Yuhao Pan, Yunfeng Fan, Ding Wang, Nan Cheng, Haozhao Wang, Wenchao Xu
View a PDF of the paper titled YANAMI: A Large-Scale Multimodal Dataset and Benchmark for Multiturn Stateful Interaction, by Shengyu Wang and 12 other authors