Setup

Hardware (optional)¶

• An NVIDIA GPU

This workshop focuses on hardware acceleration for CUDA environments which require an NVIDIA GPU to use. While an NVIDIA GPU is not required to be able to work through the workshop, access to one will be required to run the examples.

Software¶

This workshop requires minimal software to be installed in advance:

• A computer running an 64 bit version of Linux, macOS, or Windows.
  • At the moment a laptop is required and an ARM based tablet will not be sufficient.
• Pixi
• Git, if not already familar with git and GitHub, please install gh as well to simplify the workflow.
• Brev CLI

Web Platforms (optional, but encouraged)¶

• GitHub

For this workshop we would like you to create your own Git repository where you add the results of your work as you move through the workshop so that you have a sharable form of what you have learned by the end. It doesn’t need to be GitHub (GitLab.com or some other alternative exist) but for the sake of consistency, the instructions will assume you are using GitHub.

Installation¶

Pixi¶

To install Pixi follow the installation instructions for your particular machine and then restart your shell.

curl -fsSL https://pixi.sh/install.sh | sh

powershell -ExecutionPolicy ByPass -c "irm -useb https://pixi.sh/install.ps1 | iex"

Pixi Shell completions¶

Additionally, install the Pixi shell completions for your particular shell choice.

Git¶

You probably already have Git installed on your machine. You can check with

command -v git

If the command doesn’t return a filepath to the git executable, first make sure you have Pixi installed, as described above, and then run

pixi global install git

You can now use the Git anywhere on your machine.

Brev¶

Brev CLI¶

For the portion of the workshop where GPUs will be used we’ll be working on an NVIDIA Brev instance. To install the CLI API for Brev, we’ll use pixi global so make sure you first have Pixi installed, as described above, and then run

pixi global install brev

You can now use the Brev CLI anywhere on your machine. Check out the CLI options with

brev --help

Setup a Personal GitHub Repository¶

To share your work from this workshop, we will create a GitHub repository to store your code and results. This will allow you to easily share your work with others and keep track of your progress, or share it with the Brev instance.

1. Create a personal GitHub account if you don’t have one yet.
2. Add a new repository to your account through this link: Create a new repository.
3. Name the new repository reproducible-ml-lesson, make it public, and give it a README and an open source license (e.g. MIT License).
4. As GitHub requires two-factor authentication, it is highly recommended that you generate an SSH key pair specifically for GitHub, add the generated SSH key to your GitHub account, and then use your SSH keys to connect with GitHub.

Now you can navigate to the newly created repository https://github.com/<username>/reproducible-ml-lesson and then clone to your local machine

# This assumes you have SSH keys setup with your GitHub account as recommended
git clone git@github.com:<username>/reproducible-ml-lesson.git
cd reproducible-ml-lesson

Now you have a GitHub repository set up to store your work from this workshop.

Prepare Brev Instance¶

Create an NVIDIA Brev account¶

To access the NVIDIA Brev instance you’ll also need to create an NVIDIA Brev account.

• Visit https://login.brev.nvidia.com/signin and fill in your email address and agree to the terms of use.
• Check your email address that you used to create the account for verification email with further instructions.

Login to Brev account with Brev CLI¶

To validate your Brev account and your Brev CLI install, login to your Brev account from the command line with brev login.

brev login

Prepare an NVIDIA Brev instance¶

Later on in the URSSI 2025 workshop, we’ll use a coupon code provided by NVIDIA to provision a new Brev GPU instance environment.

The particular configuration we’ll be using is:

• 1x NVIDIA L4 GPU
• 24GiB VRAM
• 16GiB Ram x 4 CPUS
• GCP

You can select it from the Brev new environment page

OR run the following command to create a new instance with the same configuration:

brev create reproducible-ml-workshop --gpu g2-standard-4:nvidia-l4:1

Access the NVIDIA Brev instance on your machine¶

Once the instance is created, get access to it with the following command:

# Start an ssh session connected to the instance
brev shell reproducible-ml-workshop

or

# Open the instance in VS Code
brev open reproducible-ml-workshop

Prepare your Brev instance¶

Once you have access to the Brev instance, you can use it like any other Linux machine and install any additional software you need. Please install the following software on your Brev instance:

# Pixi
curl -fsSL https://pixi.sh/install.sh | sh
echo -e '\neval "$(pixi completion --shell bash)"' >> ~/.bashrc
source ~/.bashrc
# Additional tools
pixi global install bat