docker-compose.yml

services:
  vllm-cpu:
    build:
      context: .
      dockerfile: Dockerfile
    image: vllm-cpu-optimized:latest
    container_name: vllm-smollm2

    # Command to serve the model (arguments passed to the entrypoint)
    command:
      - --model
      - ${MODEL_NAME:-HuggingFaceTB/SmolLM2-360M-Instruct}
      - --host
      - "0.0.0.0"
      - --port
      - "8000"
      - --dtype
      - ${DTYPE:-auto}
      - --max-model-len
      - ${MAX_MODEL_LEN:-2048}
      - --max-num-seqs
      - ${MAX_NUM_SEQS:-8}
      - --tensor-parallel-size
      - "1"
      - --pipeline-parallel-size
      - "1"
      - --disable-log-requests
      - --trust-remote-code

    ports:
      - "${VLLM_PORT:-8009}:8000"

    environment:
      # vLLM CPU-specific settings (inherited from Dockerfile)
      - VLLM_TARGET_DEVICE=cpu
      - VLLM_CPU_KVCACHE_SPACE=${KVCACHE_SPACE:-1}

      # Thread control for optimal CPU performance
      - OMP_NUM_THREADS=${OMP_THREADS:-2}
      - OPENBLAS_NUM_THREADS=1
      - MKL_NUM_THREADS=1

      # HuggingFace settings
      - HF_HOME=/workspace/.cache/huggingface
      - TRANSFORMERS_CACHE=/workspace/.cache/huggingface

      # Optional: HuggingFace token for gated models
      # - HF_TOKEN=${HF_TOKEN}

    volumes:
      # Cache HuggingFace models to avoid re-downloading
      - hf-cache:/workspace/.cache/huggingface
      # Optional: Mount local models directory
      # - ./models:/workspace/models:ro

    # Resource limits optimized for macOS
    deploy:
      resources:
        limits:
          cpus: '${CPU_LIMIT:-4.0}'
          memory: ${MEMORY_LIMIT:-8G}
        reservations:
          cpus: '${CPU_RESERVATION:-2.0}'
          memory: ${MEMORY_RESERVATION:-4G}

    # Health check
    healthcheck:
      test: ["CMD", "curl", "-f", "http://localhost:8000/health"]
      interval: 30s
      timeout: 10s
      retries: 3
      start_period: 60s

    # Restart policy
    restart: unless-stopped

    # Network configuration
    networks:
      - vllm-network

  # Gradio Chatbot Interface
  chatbot:
    build:
      context: .
      dockerfile: Dockerfile.chatbot
    image: vllm-chatbot:latest
    container_name: vllm-chatbot

    ports:
      - "7860:7860"

    environment:
      - VLLM_BASE_URL=http://vllm-cpu:8000/v1
      - MODEL_NAME=${MODEL_NAME:-HuggingFaceTB/SmolLM2-360M-Instruct}

    depends_on:
      vllm-cpu:
        condition: service_healthy

    restart: unless-stopped

    networks:
      - vllm-network

  # Optional: Simple web UI for testing (uncomment to enable)
  # webui:
  #   image: ghcr.io/open-webui/open-webui:main
  #   container_name: vllm-webui
  #   ports:
  #     - "3000:8080"
  #   environment:
  #     - OPENAI_API_BASE=http://vllm-cpu:8000/v1
  #     - OPENAI_API_KEY=dummy
  #   depends_on:
  #     - vllm-cpu
  #   networks:
  #     - vllm-network

volumes:
  hf-cache:
    driver: local

networks:
  vllm-network:
    driver: bridge