DocuLingo combines "Document" processing with "Linguistic" intelligence, offering a powerful end-to-end parsing solution based on multimodal large language models. It enhances RAG (Retrieval Augmented Generation) workflows by intelligently parsing and structuring content from various document formats, including accurate extraction of formulas, tables, and images. While primarily optimized for Qwen2.5-VL, DocuLingo supports integration with other Vision Language Models (VLMs) through flexible configuration options, bridging the gap between document understanding and language processing capabilities.
- PDF to HTML Conversion: Convert PDF documents to HTML format with preserved images
- PDF to Markdown Conversion: Transform PDF content into clean, structured markdown
- Office Document Support: Process DOCX, PPTX, and other office formats with proper image extraction
- Intelligent Layout Analysis: Maintains document structure including tables, lists, and formatting
- Multi-language Support: Works effectively with documents in multiple languages
- Customizable Processing Parameters: Adjust DPI, token limits, and other settings based on your needs
- Python 3.11 or higher
- LibreOffice (for processing Office documents)
- Clone the repository:
git clone https://github.com/Niraya666/DocuLingo.git
cd DocuLingo- Create a conda environment:
conda create -n doculingo python=3.11
conda activate doculingo- Install required packages:
pip install -r requirements.txtbrew install --cask libreoffice
brew install poppler
# Create symbolic link
sudo ln -s /Applications/LibreOffice.app/Contents/MacOS/soffice /usr/local/bin/libreofficeapt-get update
apt-get install -y libreoffice
apt-get install -y unoconv
# Install Chinese fonts
apt-get update
apt-get install -y fonts-wqy-zenhei fonts-wqy-microhei
apt-get install -y fonts-noto-cjk- Copy the example environment file:
cp .env.example .env- Configure your API keys and model preferences:
OPENAI_API_KEY=sk-xxx
API_BASE=your-api-base
VISION_MODEL=Qwen/Qwen2-VL-72B-Instruct
TEXT_MODEL=Qwen/Qwen2.5-72B-Instruct
python main.py \
--pdf_path your-file-path \
--output_dir path-to-save \
--doc_type qwen_vl_htmlpython main.py \
--pdf_path your-file-path \
--output_dir path-to-save \
--dpi 150 \
--max_tokens 4096 \
--doc_type qwen_vl_html \
--convert_officeNote: It's recommended to use absolute paths for file locations.
--pdf_path Path to the input PDF or Office file
--output_dir Directory to save intermediate images
--dpi DPI for PDF to image conversion (default: 150)
--max_tokens Maximum tokens for LLM processing (default: 4096)
--doc_type Document type for processing (default: qwen_vl_html)
--convert_office Enable Office format conversion using LibreOffice
You can adjust the concurrency and retry parameters in .env:
MAX_RETRIES: 3 # Maximum number of retry attempts for failed requests
MAX_WORKERS: 2 # Maximum number of concurrent workers for parallel processing
- Qwen2.5-VL for the powerful multimodal language model