version: '3.8'

services:
  ubuntu-finetune:
    build: 
      context: .
      dockerfile: Dockerfile
      args: 
        PYTHON_VERSION: "3.10"
        CUDA_VERSION: "12.1.0"
        PYTORCH_VERSION: "2.3.0"
        TORCHVISION_VERSION: "0.18.0"
        TORCHAUDIO_VERSION: "2.3.0"
        DS_BUILD_OPS: 1
        USE_CUDA: 1
        USE_ROCM: 0
        USE_XPU: 0
        CUDA: cu121
        CUDA_ARCH_LIST: "80;86;89;90"
        SETUPTOOLS_VERSION: "69.5.1"
        DCUTLASS_NVCC_ARCHS: "80;86;89;90;90a" 
        DEEPSPEED_VERSION: "master"
        DEEPSPEED_INSTALL_FLAGS: "--allow_sudo"
    volumes:
      - ./binbbt:/bbtft
    container_name: ubuntu-finetune
    pull_policy: if_not_present
    restart: unless-stopped
    image: hotwa/deepspeed:pt23
    shm_size: '40gb'
    ports:
      - 3228:22
    environment:
      - NVIDIA_VISIBLE_DEVICES=all
      - NVIDIA_DRIVER_CAPABILITIES=compute,utility
      - TMPDIR=/var/tmp
    networks:
      - test-net
    deploy:
      resources:
        reservations:
          devices:
            - driver: nvidia
              count: all
              capabilities: [gpu]

# 修改为docker-swarm的网络
networks:
  test-net:
    external: true