🌟 olmOCR: a tool for processing PDF documents.

olmOCR is a project designed to convert PDF files and document images into structured Markdown text. It can handle equations, tables, and handwritten text, preserving the correct reading order even in the most complex multi-column layouts.

olmOCR is trained with heuristics to handle common parsing and metadata errors and supports SGLang and vLLM, where it can scale from one to hundreds of GPUs, making it a unique solution for large-scale tasks.

The key advantage of olmOCR is its cost-effectiveness. Processing 1 million PDF pages will cost only $190 (with GPU rental), which is about 1/32 of the cost of using the GPT-4o API for the same volume.

The development team created a unique method called "document anchoring" to improve the quality of the extracted text. It uses text and metadata from PDF files to improve the accuracy of processing. Image regions and text blocks are extracted, concatenated and inserted into the model prompt. When VLM requests a plain text version of the document, the "anchored" text is used along with the rasterized page image.

In tests, olmOCR showed high results compared to Marker, MinerU and GOT-OCR 2.0. During testing, olmOCR was preferred in 61.3% of cases against Marker, in 58.6% against GOT-OCR and in 71.4% against MinerU.

▶️ olmOCR release:

🟢 Model olmOCR-7B-0225-preview - retrained Qwen2-VL-7B-Instruct on dataset olmOCR-mix-0225;

🟢 Dataset olmOCR-mix-0225 - over 250 thousand pages of digital books and documents from the public domain, recognized using gpt-4o-2024-08-06 and a special prompt strategy that preserves all digital content of each page.

🟢 A set of codes for inference and training.


▶️ Recommended environment for inference:

🟠 NVIDIA GPU (RTX 4090 and above)
🟠 30 GB free space on SSD \ HDD
🟠 installed package poppler-utils
🟠 sglang with flashinfer for GPU inference

▶️ Local installation and launch:

 # Install dependencies
sudo apt-get update
sudo apt-get install poppler-utils ttf-mscorefonts-installer msttcorefonts fonts-crosextra-caladea fonts-crosextra-carlito gsfonts lcdf-typetools

# Set up a conda env
conda create -n olmocr python=3.11
conda activate olmocr

git clone https://github.com/allenai/olmocr.git
cd olmocr
pip install -e .

# Convert a Single PDF
python -m olmocr.pipeline ./localworkspace --pdfs tests/gnarly_pdfs/test.pdf

# Convert Multiple PDFs
python -m olmocr.pipeline ./localworkspace --pdfs tests/gnarly_pdfs/*.pdf

📌 Licensing: Apache 2.0 License.


🟡 Article
🟡 Demo
🟡 Model
🟡 Arxiv
🟡 Discord Community
🖥 Github