Adding to the Q: Any good small open-source model with a high correctness of reading/extracting Tables and/of PDFs with more uncommon layouts.