services:
  qwen-08b:
    build: .
    container_name: qwen_08b_server
    ports:
      - "8000:8000"
    devices:
      - /dev/kfd:/dev/kfd
      - /dev/dri:/dev/dri
    volumes:
      - /opt/models:/models
    # -ngl 99 offloada tutti i layer sulla GPU AMD
    # --host 0.0.0.0 lo rende accessibile fuori dal container
    command: >
      -m /models/Qwen3.5-0.8B-UD-Q8_K_XL.gguf
      --host 0.0.0.0
      --port 8000
      -ngl 99
      -c 4096
    restart: unless-stopped