services:
  ldh-deepspeed-test:
    build: 
      context: .
      dockerfile: Dockerfile
      # args:
      #   HTTP_PROXY: "http://127.0.0.1:15777"
      #   HTTPS_PROXY: "http://127.0.0.1:15777"
      #   cache-from: "type=local"
    image: ldh/deepspeed:test
    container_name: ldh-deepspeed-test
    shm_size: '1024gb'
    deploy:
      resources:
        reservations:
          devices:
            - driver: nvidia
              count: all
              capabilities: [gpu]
    # runtime: nvidia
    # ipc: host
    pid: host
    environment:
      - NVIDIA_VISIBLE_DEVICES=all
      - NVIDIA_DRIVER_CAPABILITIES=compute,utility
    # stdin_open: true
    # tty: true
    # privileged: true
    cap_add:
      - ALL
    volumes:
      - /mnt/beegfs:/root/shared/beegfs
      - /mnt/yrfs:/root/shared/yrfs
    # ports:
    #   - "22242:22242"
    #   - "5000:5000"
    network_mode: host
    command: ["/usr/sbin/sshd", "-D"]
    # command: ["/bin/bash", "-c", "while true; do sleep 1000; done"]