Reproduce “Lingshu: A Generalist Foundation Model for Unified Multimodal Medical Understanding and Reasoning” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Weiwen Xu, Hou Pong Chan, Long Li, Mahani Aljunied, Ruifeng Yuan, Jianyu Wang, Chenghao Xiao, Guizhen Chen and 10 more
Lingshu-Med adapts Qwen2.5-VL through staged training on curated medical images, medical text, and general-domain data. The authors evaluate it on medical multimodal question answering and report a seven-benchmark mean accuracy of 66.6 for the 32B model, compared with 59.4 for InternVL3-38B. It does not lead on every task, and these benchmark results do not establish clinical outcomes.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.