docker-compose.yaml

version: "3.8"

services:
  nvidia-endpoint:
    build: .
    working_dir: /app
    # Run FastAPI via uvicorn, listening on all interfaces
    command: >
      python3 -m uvicorn nvidia_endpoint:app
      --host 0.0.0.0 --port 5050 --workers 1
    ports:
      - "5051:5050"               # host:container
    volumes:
      # Map your script into the container under a valid module name
      - ./nvidia-endpoint-server.py:/app/nvidia_endpoint.py:ro
      # (Optional) map requirements if you iterate frequently:
      # - ./requirements.txt:/app/requirements.txt:ro
    # Minimal, simple runtime—add your key only if you want auth on /nvidia/
    environment:
      # NVIDIA_API_KEY: "change-me"   # uncomment to require X-API-Key
    restart: unless-stopped
    # Modern GPU exposure in Compose v2 (no deprecated 'runtime: nvidia')
    # If you target multiple GPUs, change gpus to - "all" or list devices, e.g. "device=0,1".
    gpus:
      - "device=0"