services:
  deepseek_ocr_vllm:
    build:
      context: .
      dockerfile: Dockerfile
    image: deepseek-ocr-api:torch2.6.0-cuda12.6-cudnn9-vllm0.8.5
    container_name: deepseek_ocr_vllm
    working_dir: /workspace
    volumes:
      - ./:/workspace
    ports:
      - "11635:11635"
    gpus: all
    shm_size: "16g"
    ipc: "host"
    environment:
      - HF_HOME=/workspace/.cache/huggingface
      - CUDA_HOME=/usr/local/cuda
      - LD_LIBRARY_PATH=/usr/local/cuda/lib64:${LD_LIBRARY_PATH}
      - PIP_DISABLE_PIP_VERSION_CHECK=1
      - PYTHONUNBUFFERED=1
    tty: true
    restart: always
    networks:
      - llm_gateway_local_net

networks:
  llm_gateway_local_net:
    external: true