Awesome
GPU Environment Management for JupyterLab
A JupyterLab extension for managing GPU environments using genv.
The genv extension lets you interactively control, configure and monitor the GPU resources that your Jupyter Notebooks are using.
🏃🏻 Be an early runner in the genv community!
Join our Slack channel with the creators of genv and start building your models faster!
- Installation and setup support as well as best practice tips and tricks directly for your use-case
- Discuss possible features
- Monthly coffee breaks to get to know the rest of the community
Looking forward to seeing you as a part of the community!
Table of Contents
Getting Started
Read the genv reference to get started.
Installation
Requirements
JupyterLab >= 3.0
Conda
If you are using Conda, it is best to install the jupyterlab_genv
package from the channel conda-forge:
conda install -c conda-forge jupyterlab_genv
Pip
Alternatively, you can install jupyterlab_genv
from PyPI using pip
:
pip install jupyterlab_genv
Install genv Kernels
After installing jupyterlab_genv
, you will need to install genv Jupyter kernels using:
python -m jupyterlab_genv install
Usage
Activate Your Environment
To activate your environment, you will have to select a genv kernel.
Then, click the GPUs
button on the Jupyter Notebook toolbar.
A dialog should pop up where you can choose either to create a new environment for your Jupyter Notebook, or to use an existing one.
Then, you can open a terminal activated in your environment. From there you will be able to configure the environment and attach devices.
Attach GPUs to Your Environment
Configuring the environment and attaching devices is done from the genv terminal.
Make sure to restart your kernel after running the command in the terminal for it to take effect.
See Devices and Environments
You can open the devices and environments widgets to see information.
Open the command palette (Command/Ctrl Shift C
) and type GPUs
.
Development
Setup
You will need to create a virtual environment once using the command:
conda create -n jupyterlab_genv --override-channels --strict-channel-priority -c conda-forge -c nodefaults jupyterlab=3 cookiecutter nodejs jupyter-packaging git
Then, activate the virtual environment when you want to work on the project:
conda activate jupyterlab_genv
Install
Use the following commands to install the Python package and enable it in JupyterLab:
# Install package in development mode
pip install -e .
# Link your development version of the extension with JupyterLab
jupyter labextension develop . --overwrite
# Server extension must be manually installed in develop mode
jupyter server extension enable jupyterlab_genv
If you make any changes you will need to rebuild the extension Typescript source using:
jlpm build
Alternatively, you can watch the source directory using:
jlpm watch
With the jlpm watch
command running, every saved change will immediately be built locally and available in your running JupyterLab. Refresh JupyterLab to load the change in your browser (you may need to wait several seconds for the extension to be rebuilt).
Run
Run JupyterLab using the command:
jupyter lab
Running
SHELL=bash jupyter lab --no-browser
is even better
Uninstall
# Server extension must be manually disabled in develop mode
jupyter server extension disable jupyterlab_genv
pip uninstall jupyterlab_genv
In development mode, you will also need to remove the symlink created by jupyter labextension develop
command.
To find its location, you can run jupyter labextension list
to figure out where the labextensions
folder is located.
Then you can remove the symlink named jupyterlab_genv
within that folder.
Reference
List all kernel provisioners
jupyter kernelspec provisioners
Install a kernel provisioner
To add a kernel provisioner to a kernel spec, edit its kernel.json
file.
For example, to install a kernel provisioner for the python3
kernel spec, run:
vim $CONDA_PREFIX/share/jupyter/kernels/python3/kernel.json
And add:
"metadata": {
"kernel_provisioner": {
"provisioner_name": "genv-provisioner"
}
}
List all available kernel specs
ls -la $CONDA_PREFIX/share/jupyter/kernels/
List all running kernels
ls -la $(jupyter --runtime-dir)/kernel-*.json
List Jupyter server extensions
jupyter server extension list
List JupyterLab extensions
jupyter labextension list
Publish
The jupyterlab_genv
package is manually published to both PyPI and conda-forge.
We do not publish the frontend part as an npm package because the Python package is a prebuilt server extension, and the frontend part alone is useless.
Also make sure to update the changelog (here's how) and lint the project by running npm run lint
.
Bump Version
The cookiecutter template uses tbump
for bumping the version.
However, for some reason this does not work at the moment, and we bump the version manually.
Search for the current version in the project files and replace the relevant instances. Here is a list of files that you should update:
- package.json
- package-lock.json
- pyproject.toml (also here for future
tbump
support) - jupyterlab_genv/_version.py
After pushing these changes, create a release on GitHub.
PyPI
Prerequisites
pip install build twine tbump
Create a Python Package
Create a Python source package (.tar.gz
) and the binary package (.whl
) in the dist/
directory using:
python -m build
python setup.py sdist bdist_wheel
is deprecated and will not work for this package.
Then, upload the package to PyPI using:
twine upload dist/*
We upload to PyPI with the organizational user runai
Conda
The Conda package is managed using its feedstock.
After publishing to PyPI, update the version and sha256 fields in the recipe meta.yaml
file.
A few minutes after pushing these changes, you should be able to see that the Conda package version was updated.
You can get the SHA256 hash from PyPI