edgaras0x4e/deepseek-ocr-2-api repository overview

⁠DeepSeek-OCR-2 API

Self-hosted OCR API powered by DeepSeek-OCR-2⁠. Process PDFs and images, get clean markdown output via REST API.

⁠Features

• Multi-format support: PDF, PNG, JPG, JPEG, BMP, TIFF, WEBP
• GPU accelerated: NVIDIA CUDA support
• Async processing: Job queue with progress tracking
• High accuracy: State-of-the-art DeepSeek-OCR-2 model
• Optional API key: Built-in authentication
• Persistent storage: SQLite database with Docker volumes

⁠Requirements

• NVIDIA GPU with ~8GB VRAM
• Docker with NVIDIA Container Toolkit⁠

⁠Quick Start

Create docker-compose.yml:

services:
  deepseek-ocr2:
    image: edgaras0x4e/deepseek-ocr-2-api:latest
    container_name: deepseek-ocr2
    ports:
      - "9713:7860"
    volumes:
      - ocr-data:/data
    deploy:
      resources:
        reservations:
          devices:
            - driver: nvidia
              count: 1
              capabilities: [gpu]
    restart: unless-stopped

volumes:
  ocr-data:

Run:

docker compose up -d

The API will be available at http://localhost:9713. Model (~6.4GB) downloads on first startup.

⁠Usage

Submit a document:

curl -X POST http://localhost:9713/ocr -F "[email protected]"

Response:

{
  "job_id": "abc123...",
  "filename": "document.pdf",
  "status": "queued"
}

Check status:

curl http://localhost:9713/ocr/{job_id}

Get results:

curl http://localhost:9713/ocr/{job_id}/result

Returns markdown text for each page.

⁠API Endpoints

⁠Configuration

Environment variables:

environment:
  - API_KEY=your-secret-key   
  - OCR_DPI=300               
  - DB_PATH=/data/ocr.db
  - UPLOAD_DIR=/data/uploads

⁠Model Info

• Model: DeepSeek-OCR-2
• Size: 6.4GB download
• VRAM: ~8GB
• Architecture: DeepSeek-VL-v2 based

⁠Links

• Source Code: GitHub⁠

⁠License

MIT