gpu-python/docker-compose.yml at main · nv-kriehl/gpu-python · GitHub

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
version: '3.8'

services:
  jupyterlab:
    container_name: jupyter-cuda
    user: root
    image: nvidia/cuda:12.6.2-devel-ubuntu22.04
    restart: always
    environment:
      - NVIDIA_VISIBLE_DEVICES=all
      - JUPYTER_TOKEN=token
      - NB_USER=jupyternb
      - NB_UID=1000
      - NB_GID=100
      - CHOWN_HOME=yes
    volumes:
      - ../gpu-python:/home/jupyternb
    working_dir: /home/jupyternb
    ports:
      - '8888:8888'
    command: >
      bash -c "
        apt-get update &&
        apt-get install -y python3 python3-pip &&
        pip3 install --upgrade pip &&
        pip3 install jupyterlab numpy numba matplotlib cupy-cuda12x nvmath-python numba_cuda pandas pynvml &&
        pip3 install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu124 &&
        pip3 install cudf-cu12 cuml-cu12 dask-cuda dask-cudf-cu12 dask aiohttp requests nvidia-libmathdx-cu12 --extra-index-url=https://pypi.nvidia.com &&
        ln -s /usr/local/cuda*/nvvm/lib64/libnvvm.so /usr/lib/libnvvm.so || true &&
        jupyter lab --ip=0.0.0.0 --port=8888 --no-browser --allow-root --NotebookApp.token=''"
    deploy:
      resources:
        reservations:
          devices:
            - driver: nvidia
              count: all
              capabilities: [gpu]