DBlake-BoxedLogic/

Image-2-Markdown

100,000 (page image, OCR markdown) pairs for training/evaluating document OCR and image-to-markdown models.

image: a rendered image of a single PDF page (grayscale, longest side 1536px). text: the OCR transcription in olmOCR's markdown dialect (LaTeX for equations, HTML for tables, tags for figures/diagrams). pdf_relpath, url, page_number, primary_language, is_table, is_diagram: carried-over metadata.