Marker

State-of-the-art deep learning model that converts PDF and ebooks into clean Markdown and JSON.

40,807 Stars2,952 ForksPython / PyTorchRelease: v2.1.0

Download Assets Official Website Documentation

Tool Overview

Marker is a deep-learning-powered conversion pipeline that turns PDF, EPUB, and book documents into clean, structured Markdown, LaTeX formulas, and JSON. It eliminates boilerplate, fixes multi-column reading orders, transcribes complex mathematical equations, and extracts high-resolution figures.

System Requirements: Python 3.9+ with PyTorch (GPU recommended for batch inference)

Key Capabilities & Features

Converts PDF, EPUB, and mobi to clean Markdown 10x faster than Nougat

Accurate LaTeX equation parsing from mathematical textbooks and papers

Deep learning table recognition and formatting into clean Markdown tables

Identifies and strips running headers, page numbers, and footnotes

Extracts embedded images and diagrams with exact references

Supports multi-column academic papers and complex typography

Package Manager & CLI Installation

Install and update Marker using your preferred operating system package manager or container runtime:

$ pip install marker-pdf

Verified Official Release Downloads

v2.1.0

All binaries are sourced directly from the official developer release assets without re-packaging or modifications.

Source Code ZIPVerified

marker-v2.1.0-source.zip

Size: GitHub Archive

Download Source

Client-Side SHA-256 Binary Integrity Verifier

100% Client-Side. Verified locally in your browser using the Web Crypto API. No files are uploaded.

Drag & drop downloaded binary here, or browse file

Computes SHA-256 hash in real-time to guarantee file authenticity against developer releases.

Security Verification Status: Verified AI Model & Source

Downloads link directly to the official GitHub releases hosted by VikParuchuri/marker. Files are checked for authentic digital developer signatures and verified against open-source repository tags.

Latest Release Changelog (v2.1.0)

**Surya 2.5.** Marker 2.1.0 requires `surya-ocr` 0.23.0, which uses Surya 2.5 (`datalab-to/surya-ocr-2.5`) by default: 1.7x as many pages per second as Surya 2 on one H100, and 85.9 on OmniParseBench, the best of any open model under 1B parameters. See the [announcement](https://www.datalab.to/blog/surya-2-5). - To stay on Surya 2: `SURYA_MODEL_CHECKPOINT=datalab-to/surya-ocr-2`. ## What's Changed * CI fixes by @VikParuchuri in https://github.com/datalab-to/marker/pull/1064 * readme updates by @VikParuchuri in https://github.com/datalab-to/marker/pull/1132 * readme updates by @VikParuchuri in https://github.com/datalab-to/marker/pull/1133 * Surya 2.5: surya-ocr 0.23.0, release 2.1.0 by @hunter-heidenreich in https://github.com/datalab-to/marker/pull/1135 * Surya 2.5, release 2.1.0 by @hunter-heidenreich in https://github.com/datalab-to/marker/pull/1136 ## New Contributors * @hunter-heidenreich made their first contribution in https://github.com/datalab-to/marker/pull/1135 **Full Changelog**: https://github.com/datalab-to/marker/compare/v2.0.0...v2.1.0