olmOCR
PDF and image-to-Markdown converter built on a 7B vision-language model
What it is
olmOCR turns PDFs, PNGs, and JPEGs into clean Markdown text, keeping a natural reading order across multi-column layouts and figures while handling equations, tables, and handwriting and stripping headers and footers. It runs a 7B vision-language model locally on an NVIDIA GPU or against a remote vLLM server, and includes trainer code and a benchmark suite for OCR systems. Researchers and teams building LLM training datasets use it to extract text from large document collections at low cost per page.