“The latest research has broken through the technology of generating visual Q&A data for electronic documents, significantly improving the document reading performance of multimodal large-scale language models.”