LUMI SmartSim Workshop

Using machine learning (ML) within HPC applications offers the advantage of improving the overall efficiency of HPC workflows by allowing ML predictions to guide or replace parts of simulations.  However, there are challenges in coupling the two approaches, as HPC applications are typically written in C, C++, or Fortran and run on CPUs, whereas ML frameworks are typically Python-based and benefit from running on GPUs. 

This workshop will introduce SmartSim, an open-source framework that has been developed to help address these challenges and allow researchers to easily adapt their traditional HPC workloads to include ML. This is done with a distributed orchestrator, which allows data to be exchanged between the HPC application and the ML model.

Learning outcomes

In the morning session, attendees will discover how ML and HPC applications may be coupled together. The session will include an overview of SmartSim and how it can interact with simulation data in HPC applications in order to integrate ML models. Attendees will also hear about real examples of how SmartSim has been used in different fields of research, such as Computational Fluid Dynamics and Climate Modelling, and find out how SmartSim can be integrated within an application and how to deploy the resulting workflow.

The afternoon session will take a more hands-on approach where attendees can run examples of hybrid scientific simulation/AI workflows using SmartSim.  Attendees will learn how to launch SmartSim leveraging the system’s native batch scheduler and job launcher and gain a practical understanding of the modifications required to adapt HPC applications to use ML. Additionally, examples will be provided where attendees can run real applications that utilise SmartSim.

Target group

LUMI users interested in integrating machine learning into existing HPC workflows.

Organisers

HPE, LUMI User Support Team (LUST), SURF Utrecht