stack/deploy/docker-compose.yaml

version: '3.8'

services:
  stack-2.9:
    build:
      context: .
      dockerfile: Dockerfile
      args:
        - PYTHON_VERSION=3.10
        - VLLM_VERSION=0.6.3
        - CUDA_VERSION=12.1.0
    container_name: stack-2.9-server
    restart: unless-stopped
    ports:
      - "${STACK_PORT:-8000}:8000"
    environment:
      # Model configuration
      - MODEL_ID=${MODEL_ID:-TheBloke/Llama-2-7B-Chat-AWQ}
      - HUGGING_FACE_TOKEN=${HUGGING_FACE_TOKEN:-}
      - QUANTIZATION=${QUANTIZATION:-awq}

      # vLLM engine parameters
      - TENSOR_PARALLEL_SIZE=${TENSOR_PARALLEL_SIZE:-1}
      - GPU_MEMORY_UTILIZATION=${GPU_MEMORY_UTILIZATION:-0.9}
      - MAX_MODEL_LEN=${MAX_MODEL_LEN:-4096}
      - MAX_NUM_SEQS=${MAX_NUM_SEQS:-64}
      - MAX_NUM_BATCHED_TOKENS=${MAX_NUM_BATCHED_TOKENS:-4096}
      - ENFORCE_EAGER=${ENFORCE_EAGER:-false}
      - DISABLE_LOG_STATS=${DISABLE_LOG_STATS:-false}

      # Server configuration
      - HOST=${HOST:-0.0.0.0}
      - PORT=${PORT:-8000}
      - MODEL_CACHE_DIR=${MODEL_CACHE_DIR:-/home/vllm/.cache/huggingface}

      # Performance tuning
      - OMP_NUM_THREADS=${OMP_NUM_THREADS:-4}
      - CUDA_LAUNCH_BLOCKING=${CUDA_LAUNCH_BLOCKING:-0}
      - CUDNN_LOGINFO_DBG=1

    volumes:
      # Model cache persistence
      - model_cache:/home/vllm/.cache/huggingface:rw
      # Optional: mount custom models
      - ./models:/app/models:ro
    networks:
      - stack-network
    # GPU configuration - uncomment for GPU support
    deploy:
      resources:
        reservations:
          devices:
            - driver: nvidia
              count: all
              capabilities: [gpu]
    # Runtime configuration
    runtime: nvidia
    # Health check
    healthcheck:
      test: ["CMD", "curl", "-f", "http://localhost:8000/health"]
      interval: 30s
      timeout: 10s
      retries: 3
      start_period: 60s
    # Resource limits
    # mem_limit: ${MEM_LIMIT:-8g}
    # mem_reservation: ${MEM_RESERVATION:-4g}

volumes:
  model_cache:
    driver: local

networks:
  stack-network:
    driver: bridge