README.md

metadata

title: Typhoon OCR
emoji: 🌍
colorFrom: gray
colorTo: red
sdk: gradio
sdk_version: 5.29.1
app_file: app.py
pinned: false
license: apache-2.0
short_description: Convert Image & PDF to Markdown

Typhoon OCR

Typhoon OCR is a model for extracting structured markdown from images or PDFs. It supports document layout analysis and table extraction, returning results in markdown or HTML. This package is a simple Gradio website to demonstrate the performance of Typhoon OCR.

Features

• Upload a PDF or image (single page)
• Extracts and reconstructs document content as markdown
• Supports different prompt modes for layout or structure
• Language: English, Thai
• Uses a local or remote OpenAI-compatible API (e.g., vllm)

Install

pip install -r requirements.txt
# edit .env
# pip install vllm # optional for hosting a local server

Mac specific

brew install poppler
# The following binaries are required and provided by poppler:
# - pdfinfo
# - pdftoppm

Linux specific

sudo apt-get update
sudo apt-get install poppler-utils
# The following binaries are required and provided by poppler-utils:
# - pdfinfo
# - pdftoppm

Start vllm

vllm serve scb10x/typhoon-ocr-7b --served-model-name typhoon-ocr --dtype bfloat16 --port 8101

Run Gradio demo

python app.py

Dependencies

• openai
• python-dotenv
• ftfy
• pypdf
• gradio
• vllm (for hosting an inference server)
• pillow

License

This project is licensed under the Apache 2.0 License. See individual datasets and checkpoints for their respective licenses.