datalab-to/marker: Convert PDF to markdown + JSON quickly with high accuracy

Marker converts PDFs, images, PPTX, DOCX, XLSX, HTML, and EPUB into markdown, JSON, chunks, or HTML with high accuracy.
It extracts tables, forms, equations, inline math, links, references, code blocks, and images while removing headers, footers, and other artifacts.
Marker supports structured extraction from a JSON schema and can optionally use an LLM to improve accuracy.
The LLM-assisted hybrid mode can merge tables across pages, better handle inline math, format tables properly, and extract form values.
By default, the LLM mode uses Gemini 2.0 Flash, but any Gemini or Ollama model can be used.
Marker runs on GPU, CPU, or MPS, and its performance improves substantially in batch mode.
The project claims projected throughput of about 25 pages per second on an H100 in batch processing.
Installation requires Python 3.10+ and PyTorch, with an extra package for non-PDF document types.
Command-line options include page range selection, output format choice, OCR controls, image extraction toggles, debug mode, and custom processors/configuration.
The code is GPL and the model weights use a modified AI Pubs Open Rail-M license, with broader commercial licensing available for paid use.

Your notes

Save this item to your library to add private notes.