• Jupyter is a large umbrella project that covers many different software offerings and tools, including the popular Jupyter Notebook and JupyterLab web-based notebook authoring and editing applications.

• Kernels are programming language specific processes that run independently and interact with the Jupyter Applications and their user interfaces.

  • They can be considered as the computational engine for notebooks.

  • The kernel process is a standalone process that runs in the background and executes the code that you write in your notebooks. The kernel process is responsible for running the code and returning the results to the frontend.

• ARCC provides the Jupyter with OnDemand workshop.

The general process involves updating the conda environment to include kernel related packages, and then configuring the kernel spec to allow it to be picked up by the Jupyter service.

The process creates a kernel for an individual user from the previously created shared Conda environment.

Each individual from the shared space will need to follow this process to create their own kernel.

If anyone updated the packages within this shared Conda environment, then the update will effect everyone. Be warned.

[]$ module purge
[]$ module load miniconda3/24.3.0
[]$ conda activate /cluster/medbow/project/<project-name>/software/tensorflow/2.16
(/cluster/medbow/project/<project-name>/software/tensorflow/2.16) []$ export PYTHONUSERBASE=$CONDA_PREFIX
(/cluster/medbow/project/<project-name>/software/tensorflow/2.16) []$ conda install ipykernel
...
(/cluster/medbow/project/<project-name>/software/tensorflow/2.16) []$ python -m ipykernel install --user --name=TF2.16-local

(/cluster/medbow/project/<project-name>/software/tensorflow/2.16) []$ conda deactivate
[]$ 

Using the python -m ipykernel install --user --name=TF2.16-local above will result in the following output of the form:

() (/project/<project-name>/software/tensorflow/2.16) []$ python -m ipykernel install --user --name=TF2.16-local
Installed kernelspec TF2.16-local in /cluster/medbow/home/<username>/.local/share/jupyter/kernels/tf2.16-local

Notice: This file is created in the user's home folder, and can be accessed and viewed by the user:

[]$ ls /cluster/medbow/home/<username>/.local/share/jupyter/kernels/tf2.16-local/
kernel.json  logo-32x32.png  logo-64x64.png  logo-svg.svg

[]$ cat /cluster/medbow/home/<username>/.local/share/jupyter/kernels/tf2.16-local/kernel.json
{
 "argv": [
  "/project/<project-name>/software/tensorflow/2.16/bin/python",
  "-Xfrozen_modules=off",
  "-m",
  "ipykernel_launcher",
  "-f",
  "{connection_file}"
 ],
 "display_name": "TF2.16-local",
 "language": "python",
 "metadata": {
  "debugger": true
 }
}

Suggestion: Using “-local” will help to distinguish that this is a local kernel, based on a Conda environment, that the user has created, and not something provided by ARCC.

Side Note: Installing the ipykernel package creates a kernel spec related folder within the Conda environment, under: share/jupyter/kernels/ named python3.

Running python -m ipykernel will copy and update this folder.

Kernel names can only contain ASCII letters and numbers and these separators: - . _ (hyphen, period, and underscore)

Names can not use whitespace such as --name="TF2.16 (local)"

From OnDemand start a Jupyter session.

Notice how the newly configured kernel is now available.

import tensorflow as tf; 
print("TensorFlow Version: " + str( tf.__version__))

TensorFlow Version: 2.16.1