Infinity-Parser2 achieves end-to-end document parsing via multi-task RL

Infinity-Parser2 uses a controllable data-synthesis pipeline and multi-task reinforcement learning to perform end-to-end document parsing. The project includes the open-source Infinity-Doc2-5M corpus, containing 5 million bilingual samples with annotations for Markdown, HTML, LaTeX, and SMILES.