dots.ocr
A self-hosted multilingual OCR model that extracts document text and page layout, with vLLM deployment and CPU inference support.
dots.ocr is a self-hosted OCR model for developers and teams that need text and page structure from multilingual documents. It combines text recognition and layout parsing in a single vision-language model you can run on your own hardware.
- Multilingual recognition: Extract text across languages and scripts with the same model.
- Document layout: Parse page structure alongside the text, for workflows that need to retain how a document is organized.
- Tables and formulas: Extract table content and mathematical expressions, though complex tables and formulas remain difficult for the model.
The Python project is open source under the MIT license. It supports local deployment through vLLM, and Docker images are available. Hugging Face inference on a CPU is also supported. Parsing can occasionally fail, so extracted documents may need review before use.
Logotyper
