MinerU

MinerU is an open-source tool for high-precision document content extraction: it converts complex PDFs into structured, LLM-ready formats like Markdown and JSON.

This is MinerU, your reliable, open-source solution for document content extraction, built by opendatalab. It tackles the tough challenge of transforming complex documents, like scientific PDFs, into clean, machine-readable data (Markdown or JSON) for Agentic workflows and LLM pre-training. The system employs a multi-module parsing strategy, combining advanced techniques for layout detection, formula recognition, and table recognition to ensure high accuracy and consistency. Specifically, MinerU addresses symbol conversion issues in scientific literature, providing cleaner training data that has shown to deliver a +1.08 percentage point accuracy gain in LLM pretraining compared to traditional methods. It supports multi-language recognition and offers flexible deployment via pip, Docker, or a zero-install web version.

https://github.com/opendatalab/MinerU