JupyterLab GPU Dashboards

A JupyterLab extension for displaying dashboards of GPU usage.

Built with JupyterLab and Bokeh Server

What's here

This repository contains two sets of code:

• Python code defining a Bokeh Server application that generates the dashboards in the jupyterlab_nvdashboard/ directory
• TypeScript code integrating these dashboards into JupyterLab in the src/ directory

You should be able to modify only the Python code to edit the dashboards without modifying the TypeScript code.

Prerequisites

• JupyterLab 1.0
• bokeh
• pynvml

Installation

This extension has a server-side (Python) and a client-side (Typescript) component, and we must install both in order for it to work.

Note: Currently nvdashboard does not support Windows

To install the server-side component, run the following in your terminal

pip install jupyterlab-nvdashboard

To install the client-side component, run

jupyter labextension install jupyterlab-nvdashboard

Development

To install the server-side part, run the following in your terminal from the repository directory:

pip install -e .

In order to install the client-side component (requires node version 8 or later), run the following in the repository directory:

jlpm install
jlpm run build
jupyter labextension install .

To rebuild the package and the JupyterLab app:

jlpm run build
jupyter lab build

Publishing

This application is distributed as two subpackages.

The JupyterLab frontend part is published to npm, and the server-side part to both PyPI and Anaconda (nightlies).

Releases for both packages are handled by gpuCI. Nightly builds are triggered when a push to a versioned branch occurs (i.e. branch-0.5). Stable builds are triggered when a push to the main branch occurs.