This tutorial will cover running a Deep Learning (DL) model, based on the Pytorch framework, on the Voyager supercomputer. Voyager comprises an innovative architecture to optimize DL applications thanks to its Intel Habana Gaudi accelerators. We will discuss Voyager architecture and show how to launch jobs using containers in Kubernetes. Then, we will port a Pytorch application using Intel Gaudi libraries. We will discuss how to run Jupyter Notebooks on Voyager. Finally, we will show how to pull a Hugginface model and run inference on Voyager using Transformers/Diffusers and Optimum-Habana libraries.

--- 

Dr. Javier Hernandez-Nicolau, Data Scientist, SDSC

Dr. Javier Hernandez-Nicolau is a Data Scientist in the Scientific applications group at SDSC. He obtained his PhD in Plasma Physics and Nuclear Fusion at the University of Carlos III of Madrid (Spain) in 2019. Then, he worked as a Project Scientist at the University of California-Irvine. He developed and optimized MHD and gyrokinetic fusion plasma codes for several HPC systems using Fortan, OpenMP, and OpenACC. He joined SDSC in 2023, where he keeps collaborating with the Nuclear Fusion community, developing ML/AI applications. He also offers user support to deploy a large variety of AI models on the Voyager machine, which uses Intel Gaudi accelerators.