services:
  sglang:
    image: pipelined-sglang
    build:
      context: .
      dockerfile: Dockerfile
      target: server
    volumes:
      - ./models:/app/models:ro
      - model-cache:/root/.cache
    shm_size: 16gb
    healthcheck:
      test:
        - CMD
        - curl
        - --fail
        - --silent
        - http://127.0.0.1:30000/model_info
      interval: 2s
      timeout: 2s
      retries: 300
      start_period: 10s

  lm-eval:
    image: pipelined-lm-eval
    build:
      context: .
      dockerfile: Dockerfile
      target: eval
    depends_on:
      sglang:
        condition: service_healthy
    volumes:
      - ./models:/app/models:ro
      - ./results:/app/results
      - model-cache:/root/.cache

volumes:
  model-cache:
    name: pipelined-model-cache
    external: true