Intro to Notebooks
Substratus notebooks allow you to access high-performance GPUs while developing machine learning code. In this guide you will explore how to launch a containerized notebook from your local machine.
Prerequisites
This guide assumes you already have a Substratus cluster installed.
Also make sure to install up-to-date Substratus kubectl plugins:
wget -O - https://raw.githubusercontent.com/substratusai/substratus/main/install/kubectl-plugins.sh | bash
Clone the Example Notebook
Clone the example notebook repo.
git clone https://github.com/substratusai/example-minimal-notebook
cd example-minimal-notebook
By convention in Substratus source code is stored in src/. A Dockerfile is used to describe the environment the notebook will run in. When launching a notebook, Substratus will look for a notebook.yaml file. This file can be used to configure things like the compute resources allocated to the notebook.
ls
# Dockerfile README.md notebook.yaml src
Launch a Remote Notebook
With the following command Substratus will upload, build, and serve your notebook. The -d flag instructs the command to launch from the current directory. Once launched, try making a change to a file in the notebook and saving it.
kubectl notebook -d .
Substratus continuously syncs any changes you made in your remote notebook back to your local directory.
git status --short
# M src/hello.ipynb
Once you are done, press Ctrl-C to terminate the notebook session. This will suspend the remote notebook (you will see the browser session end).
Cleanup
Your remote notebook will continue to exist in the cluster in a suspended state (not consuming any compute resources). You can completely remove the Notebook object if you are done with it.
kubectl delete -f ./notebook.yaml
Notebooks will eventually support scratch space (see GitHub issue #66). Deleting the notebook will free the attached disks once this is supported.