Docker #
It is recommended to use Docker. You can create your virtual environment with venv since Python is pre-installed on the OMUI Server. However, this is not recommended because the Python version may be out of date.
This page explains the basic usage of Docker. To learn more, check the official documentation.
Rootless Docker Configuration #
Do this only once. Not required if already installed.
You can use Rootless Docker on the OMUI Server. Rootless Docker enables you to manage Docker containers or images for individual users.
Please run the following command to configure Rootless Docker:
$ rootless-docker-install
Docker Container #
The following command creates a new Docker container and run [COMMAND] in the container:
$ docker container run [OPTIONS] IMAGE [COMMAND] [ARG...]
Typical options include:
# Remove the container automatically when you stop the container
--rm
# Bind-mount the SRC to the DST
--volume SRC:DST
# Set the working directory in the container
--workdir PATH
# GPU allocation
--gpus DEVICE
This example creates a Docker container using nvidia/cuda:12.5.1-cudnn-devel-ubuntu22.04 image downloaded from
Docker Hub, and runs nvidia-smi command in the container.
$ docker container run \
--rm \
--gpus '"device=0"' \
nvidia/cuda:12.5.1-cudnn-devel-ubuntu22.04 \
nvidia-smi
Docker Image #
To prepare a Docker image, create your own Docker image from scratch using a Dockerfile or download it from the Docker registry. Docker Hub and NVIDIA NGC Catalog are the popular Docker registries.
When selecting a Docker image that includes the CUDA Toolkit, please check the CUDA Toolkit version included in the image. You can check the supported CUDA Toolkit version by running the
nvidia-smicommand on the server. The version displayed in the upper right corner is the latest version of the supported CUDA Toolkit.( Version displayed in
nvidia-smi≦ CUDA Toolkit version you use)
You can check the loaded images by running:
$ docker images
REPOSITORY TAG IMAGE ID CREATED SIZE
ubuntu latest 35a88802559d 5 weeks ago 78.1MB
You can remove Docker images by running:
$ docker image rm IMAGE
Please remove images that are no longer in use.