SmolDocling: An ultra-compact vision-language model for end-to-end multi-modal document conversion - Explained Simply | ArXiv Explained