Machine learning on HPC: Advanced Workflows and Tools

Curious how to run AI jobs efficiently on the LUMI supercomputer? Wondering how to manage hyperparameter search or train models on multiple GPUs? Join us online on Thursday, August 21 (13:00–15:30 Helsinki time) for a talk/demo session on advanced AI workflows in HPC environments. You'll learn how to launch AI tasks on LUMI with Slurm, perform hyperparameter search, use job arrays, and train models with HuggingFace Accelerate using multi-GPU setups and quantization techniques. This event is organised by Aalto Scientific Computing in collaboration with CSC IT Center for Science and LUMI AI Factory. No ECTS credits are available for this events; the event is talks/demo only. LUMI AI Factory is funded jointly by the EuroHPC Joint Undertaking, and the Participating States Finland, the Czech Republic, Denmark, Estonia, Norway, and Poland.

Schedule

• 13:00 - 13:30: Moving AI Jobs to LUMI: Slurm and Job Launchers
  • Speakers: Mats Sjöberg (CSC)
  • Overview of how to run AI tasks on LUMI, advanced Slurm usage, job launchers, and lessons learned beyond the basic Practical Deep Learning course. [Slides]
• 13:30 - 13:45: Q&A
• 13:45 - 14:15: Hyperparameter Search on HPC
  • Speakers: Oskar Taubert (CSC) & Simo Tuomisto (Aalto Scientific Computing)
  • Practical approaches to hyperparameter optimization like running a parameter sweep as array jobs or using search frameworks like Propulate. [Slides1] [Slides2] [Code]
• 14:15 - 14:30: Q&A
• 14:30 - 15:00: Training with HuggingFace Accelerate: Multi-GPU and Quantization
  • Speakers: Simo Tuomisto (ASC) & Yu Tian (ASC)
  • Live demo of using HuggingFace Accelerate for scaling training, quantization techniques, and model deployment tips. [Slides]
• 15:00 - 15:30: Open Q&A + Panel discussion

Video recordings

You can watch a recording of the video at this link.

About the speakers

Mats Sjöberg works as a Senior Machine Learning Specialist at CSC, where he develops and supports the machine learning environments in CSC's supercomputers. 

Oskar Taubert works as a Machine Learning Specialist at CSC, where he supports scientific projects focused on scalable machine learning.

Yu Tian works as a research software engineer at Aalto university, supporting researchers working with machine learning and medical research.

Simo Tuomisto works as a research software engineer at Aalto university, contributing to research projects on computational physics and deep learning.

For info and questions please contact Enrico Glerean or any of the speakers listed.