MMFineReason: Closing the Multimodal Reasoning Gap via Open Data-Centric Methods
IntermediateHonglin Lin, Zheng Liu et al.Jan 29arXiv
MMFineReason is a huge, open dataset (1.8 million examples, 5.1 billion solution tokens) that teaches AIs to think step by step about pictures and text together.
#multimodal reasoning#vision-language models#chain-of-thought