Running Machine Learning Workloads on HPC Using Containers

This hands-on-workshop introduces participants to the practical use of containerization technologies—Apptainer, with an overview of Charliecloud—for running machine learning (ML) workloads in High-Performance Computing (HPC) environments. Containers have become essential tools for addressing challenges in software stack management, portability, and reproducibility, in complex ML software stacks.

Participants will gain direct, hands-on experience by deploying pre-built containerized machine learning applications (e.g., TensorFlow/PyTorch) on HPC systems using Apptainer. The session will demonstrate how container technologies streamline ML workflows and integrate with HPC schedulers (e.g., using SLURM) without sacrificing performance. While Apptainer will be the focus for in-session exercises, Charliecloud will be discussed as an alternative solution, highlighting its design philosophy, features, and typical use cases.

Hands-on exercises will be conducted on the Midway system, with an emphasis on running pre-built containers to maximize practical experience within the 2-hour timeframe. Resources for building custom containers and extending access to the examples will be provided via the GitHub repository for continued learning after the workshop.

Objectives:

By the end of this workshop, participants will be able to:

• Understand the core concepts and benefits of containerization for ML workloads in HPC environments.
• Identify key features of Apptainer and recognize common use cases.
• Describe Charliecloud as an alternative container solution and understand its distinguishing characteristics.
• Download and run pre-built containers for popular ML frameworks (e.g., TensorFlow, PyTorch) using Apptainer.
• Submit and execute an Apptainer containerized ML workload using SLURM.
• Apply best practices for using containers in shared HPC environments.

Level: Intermediate

Duration: 2 hours

Github Repository: https://github.com/rcc-uchicago/hpc-ml-containers-workshop

Prerequisite: Basic understanding of programming or scripting languages. Some familiarity with Linux. Must have an active account of RCC or access to any other cluster to take full advantage.

Know what’s Happening Next — before everyone else does.

Get App

Host or Publisher Research Computing Center