Verified project record

allenai/olmocr

olmocr converts PDF, PNG, and JPEG documents into clean Markdown while preserving equations, tables, and reading order. It is designed to process documents at scale for LLM datasets and training, at a cost of less than $200 USD per million pages.

Project overview

The project combines a 7B parameter VLM with a batch pipeline that supports multi-node coordination, making it viable for converting millions of PDFs into structured Markdown for LLM training data.

Project type: Document Processing · Data Processing
Use cases: Documents & Office
Deployment: Refer to project documentation
License: Apache-2.0

Best for

Developers, data teams, and researchers who need to convert PDFs and image-based documents into clean Markdown or Dolma format for LLM datasets.

Key capabilities

A benchmark suite covering over 7,000 test cases across 1,400 documents to help measure performance of OCR systems.
Manager for running millions of PDFs through a batch inference pipeline with support for multi-node coordination.
Includes reusable scripts for prompt strategies, language filtering, SEO spam removal, SFT finetuning, GRPO RL training, synthetic data generation, and viewing Dolma docs.
Allows offloading inference to an external OpenAI-compatible API server or verified external model provider.

Limitations and risks

Local inference requires a GPU, based on a 7B parameter VLM. A recent NVIDIA GPU with at least 12 GB of GPU RAM and 30 GB of free disk space is needed.
Requirements are difficult to install in an existing Python environment.
The Docker image with the model is very large, approximately 30 GB.

Getting started

Install dependencies in a clean Python environment, then install the Python package. Local GPU execution requires a recent NVIDIA GPU with at least 12 GB of GPU RAM and 30 GB of free disk space.
Pull the Docker image from Docker Hub (alleninstituteforai/olmocr). The image with the model is approximately 30 GB.
Prepare input PDFs, then run the olmocr command to generate Markdown output. The CLI provides direct execution but involves optional arguments for scaling and external servers.

Evidence and sources

GitHub project description: Toolkit for linearizing PDFs for LLM datasets/training
README: - Efficient, less than $200 USD per million pages converted - (Based on a 7B parameter VLM, so it requires a GPU)
README: - Feb 25, 2025 - v0.1.58 - Initial public launch and demo.
README: We also ship a comprehensive benchmark suite covering over 7,000 test cases across 1,400 documents to help measure performance of OCR systems.
README: Try the online demo: [https://olmocr.allenai.org/](https://olmocr.allenai.org/)

Reviewed at: 2026-07-26T20:37:12Z

Open original GitHub project