A Collection of datasets for OCR and document understanding tasks. Specifically curated for training vision-language models.