PreprintConference on Empirical Methods in Natural Language Processing2025
MAmmoTH-VL 2
Yiming Jia, Jiachen Li, Xiang Yue, et al.
MAmmoTH-VL 2 introduces VisualWebInstruct, a method using web search and LLMs to create a large multimodal instruction dataset, boosting VLM reasoning performance.
36Mar 1, 2025Fine TuningMultimodal
arXiv