NVIDIA's AI Workbench is a complete development environment that simplifies the development experience and makes it easy to manage and scale workloads across local and remote servers.
AI workbench runs workloads in containers. These containers can be ones that you create, NVIDIA NIMs, or from other sources. The AI Workbench installation will mange the container environment for you. We recommend podman for various reasons but docker works great too.
You will need an SSH key for this process to work. If you don't already have one, you can create it with the following command:
ssh-keygen -t rsa -b 4096 -C "your_email@example.com"
There are two distinct installation processes that need to be considered: local and remote. You can think of your local installation as the client and your remote installation(s) as your server(s). You can connect to multiple remote instances from one local client. Unless you have a powerful local computer, we recommend keeping your local instance lightweight and using it for basic things, with all of your heavy lifting happening on your remote instances.
The local installation tends to be pretty straightforward, follow the local instlalation instructions here.
There are many GPU providers available. The first goal of this project is to make it easier for you to use those resource by simplifying the remote installation process. Because remote environments vary widely, the installation also varies widely.
Once you have your remote server setup, execute the following:
git clone https://github.com/brainwavecollective/nvidia-ai-workbench.git
#add your SSH Key to: ~/nvidia-ai-workbench/my_public_key.pub
./nvidia-ai-workbench/install.sh
| Provider | Status | Notes |
|---|---|---|
| Runpod | Does not work | We like the provider generally but can't access host b/c we're isolated in a container |
| Lambda Labs | Confirmed | Launch Jupyter server and then launch terminal through UI |
| Massed Compute | Confirmed | RTX A6000 [Spot], Base Ubuntu Desktop 22.04 Navigate to desktop URL and open terminal |