services:

  vllm:
    image: {{ vllm_image }}:{{ vllm_image_tag }}
    deploy:
      resources:
        reservations:
          devices:
          - driver: nvidia
            count: all
            capabilities: [gpu]
    network_mode: host
    ipc: host
    command:
    # TODO: Fix identation... again
    - {{ vllm_model }}
    {% if vllm_args %}
    {{ vllm_args|to_nice_yaml|indent(4) }}
    {% endif %}
    volumes:
    - {{ vllm_model_dir }}:/root/.cache/huggingface
    tty: true
    restart: unless-stopped
    {% if vllm_container_env %}
    environment:
    {{ vllm_container_env|to_nice_yaml|indent(6) }}
    {% endif %}
    {% if vllm_container_labels %}
    labels:
    {{ vllm_container_labels|to_nice_yaml|indent(4) }}
    {% endif %}

#vllm pull hf.co/unsloth/Apriel-1.5-15b-Thinker-GGUF:UD-Q4_K_XL