services:
  ldh-megatron-deepspeed-test:
    image: hotwa/magadeep:latest
    shm_size: '128gb'
    deploy:
      resources:
        reservations:
          devices:
            - driver: nvidia
              count: all
              capabilities: [gpu]
    #runtime: nvidia
    environment:
      - NVIDIA_VISIBLE_DEVICES=all
      - NVIDIA_DRIVER_CAPABILITIES=compute,utility
    # stdin_open: true
    # tty: true
    privileged: true
    cap_add:
      - IPC_LOCK
    volumes:
      - /root/workspace:/root/data
      - /dev/infiniband:/dev/infiniband
    # ports:
    #   - "22242:22242"
    #   - "5000:5000"
    # networks:
    #   - ldh_overlay_network
    network_mode: host
    command: ["/usr/sbin/sshd", "-D"]

# networks:
#   ldh_overlay_network:
#     external: true