Works on My Machine.
Runs on the Cluster?

Apptainer for ML on HPC

Wednesday, November 25, 2026 · 9:00–13:00 (CET)
Saarland University, building E1 1, room 407 · also online via livestream · in English

by Alexander Blatt · NHR consultant for ML @ CSHPC

Your training script runs fine on your laptop or lab machine. But how do you get the same environment onto an HPC cluster, with GPUs, shared file systems and a job scheduler? In this hands-on workshop, you will learn how to package your machine learning setup into a container with Apptainer and run it reliably and efficiently on a real NHR cluster.

Who should attend?

Anyone who wants to move their ML project from a local machine to an HPC cluster. No prior experience with containers is needed.

What you will learn

Why containers on HPC. Why "it works on my machine" often breaks on a cluster, and why Apptainer is the container tool of choice on HPC systems.

Building your own images. How to turn your Python and PyTorch environment into a container image, and how to build it quickly.

Inside vs. outside the container. Which files, software and settings come from the host, and how to avoid the pitfalls this causes.

Running on the cluster. How to run containerized jobs on CPUs and GPUs with Slurm.

Best practices. How to keep your setup reproducible, efficient and friendly to a shared cluster.

Format

The workshop starts with an introductory talk, followed by a hands-on session on a real NHR Süd-West GPU cluster. After the workshop, there will also be time until 14:00 for questions about your own projects.

Limited hands-on places. The talk is open to all registered participants, on-site or online. The hands-on session is limited to 30 places due to the number of available GPUs, and on-site participants have priority. Remaining hands-on places will be offered to online participants after registration closes on November 18, in order of registration.

Prerequisites

Basic Linux command line skills
Basic knowledge of Slurm and PyTorch is recommended
On-site: a laptop with a working eduroam connection

Registered participants receive setup instructions by email before the workshop.

Registration

Eligibility	Members of academic institutions
Registration fee	None
Organization	CSHPC, Saarland University, as part of NHR Süd-West
Contact	alexander.blatt@uni-saarland.de

The organization and coordination of this event are carried out by NHR Süd-West, a member of the National High Performance Computing (NHR) initiative in Germany. For more information, visit https://nhrsw.de/.