Example devcontainer for Deep Learning projects. This repo contains a Dockerfile/devcontainer.json/requirements.txt combination to build a container for deep learning projects containing a collection of useful packages. You can read more about why I think devcontainers are a great idea and the design decisions I made in the linked blog post.

Yes, you will need to have Docker with the Nvidia container runtime installed. Installation instructions can be found here. Depending on your setup, you might also need to have rootless docker configured. I am also assuming that you have VScode with the remote development extension pack installed to take advantage of the Devcontainers feature.

Simply run Remote containers: Clone Repository in Container Volume from the VScode command palette to clone the repository in an isolated volume, build the image, and spin up a container. Alternatively, you can clone the repository normally and run Remote containers: Open Folder in Container - I sometimes find the second method to be a bit more reliable.

The above steps will give you a VSCode window attached to the container to use as a development environment. If you want to use docker in a more traditional way, i.e build an image and run it as a traditional executable, you can first build with:

docker build -t deep-learning-devcontainers/production:latest -f .devcontainer/Dockerfile .

Then run with:

docker run --privileged --gpus=all deep-learning-devcontainers/production:latest

This will run the main.py file with all dependencies installed and return the result.

This container gives you a Ubuntu 20.04 environment with CUDA 11.3, Python 3.8 and the packages from the requirements.txt file. The main.py file simply includes some code for checking that the deep learning libraries installed can successfully see your GPU.

You will need a GPU + drivers capable of CUDA 11.3 - you can check this by running nvidia-smi and ensuring CUDA Version = 11.3 or greater.

If your project uses python virtual environments with a requirements.txt file, all you will need to do is copy the .devcontainer directory from this project into your own project. You can then make any adjustments you want to the Dockerfile and devcontainer.json. I wrote an accompanying blog post explaining some relevant concepts for this.