cuda-ml

Ubuntu Machine Learning w/ CUDA Docker Development Environment

WARNING: Be sure to change the password for the user immediately after creating the container!

This Docker image contains the necessary PyTorch and TensorFlow tools pre-installed in Miniforge3 conda environments. See the Dockerfile for more info.

CUDA Version Tags

11.7.1 = s7117/ubuntu-ml:11.7
11.8.0 = s7117/ubuntu-ml:11.8
12.2.2 = s7117/ubuntu-ml:12.2

GPU Access Prerequisite: nvidia-container-toolkit

You must use the --gpus all argument to the docker run command to pass gpus access to the containter.

This requires installing the nvidia-container-toolkit prior to using the GPUs.

https://docs.nvidia.com/datacenter/cloud-native/container-toolkit/install-guide.html

Bind/Mount a Directory

If you wish to have a directory shared between the host machine and the Docker container do the following when you run the image for the first time:

docker run --gpus all --name <container_name> --hostname <hostname> --mount type=bind,source=<localdir>,target=<containerdir> -ti s7117/docker-ml:<cudaversion>

First Time Configuration/Installation

1. Run docker pull s7117/docker-ml:<cudaversion>
2. Run docker run --gpus all --name <container_name> --hostname <hostname> -ti s7117/docker-ml:<cudaversion>

Re-Enter the shell via docker exec or docker start

• docker start -ai <container_name>
• docker exec -u user -w /home/user -ti <container_name> /bin/bash
• docker exec <container_name> <executable>

Conda Environments:

• conda activate tfgpu - TensorFlow GPU
• conda activate torchgpu - Pytorch GPU

For PyTorch do the following:

1. conda create --name pytorch pytorch cudatoolkit=11.6
2. conda activate pytorch

Installing Other Packages:

The user password is temp2023. It is advised to reset this password as soon as possible upon creating the container.

Use passwd and temp2023 to change the default password to a new password.

Optional Setup:

Optionally, you can add the --restart unless-stopped to the docker run command to restart the docker container on docker startup.