services:
  qwen36-262k:
    build:
      context: .
      dockerfile: docker/vllm/Dockerfile
      args:
        VLLM_IMAGE: "${VLLM_IMAGE:-vllm/vllm-openai:cu130-nightly-aarch64}"
    image: qwen36/qwen36-35b-a3b-fp8-262k:latest
    container_name: qwen36-262k
    restart: unless-stopped
    network_mode: host
    ipc: host
    shm_size: "32gb"
    env_file: .env
    environment:
      NVIDIA_VISIBLE_DEVICES: all
      NVIDIA_DRIVER_CAPABILITIES: compute,utility
    volumes:
      - ./models/hf:/root/.cache/huggingface
      - ./logs:/logs
    deploy:
      resources:
        reservations:
          devices:
            - driver: nvidia
              count: all
              capabilities: [gpu]