Alvis introduction

This repository contains examples and excercises for how to run ML applications on the SNIC Alvis AI/ML HPC system. The aim is to let users get their feet wet quickly with something that they can adapt into their own jobs.

The examples are written in Python with the frameworks TensorFlow and PyTorch. More frameworks can be added if interest exists. You can express interest by creating an issue for this repository or by contacting support.

Accompanying introduction slides are available at:

• https://www.c3se.chalmers.se/documentation/intro-alvis/presentation.html
• https://www.c3se.chalmers.se/documentation/intro-alvis/slides

Tutorial overview

Doing the excercises are voluntary, but make sure to read the associated READMEs for each part to make sure that you're not missing something. The tutorial is located in the tutorial/ folder.

1. Connecting and submitting jobs
2. A simple ML example on the GPU
3. Loading data: provided, your own and from the web
4. Checkpointing
5. Profiling w/ TensorBoard
6. Single node, multiple GPUs
7. Multiple nodes, multiple GPUs

A note on notebooks

Large parts of this tutorial are written in Jupyter Notebooks, this fileformat is good for teaching and early development. However, when submitting jobs or handling a larger code-base it might be more convenient to use python files. To convert a notebook to python file use

jupyter nbconvert --to script my_notebook.ipynb

or with older versions of jupyter

jupyter nbconvert --to python my_notebook.ipynb

The automatic conversion can be used as a start to refactor the code from the notebook into more suitable format.