Izvajalec / Course provider: Institute of Information Sciences (IZUM)
Predavatelji / Instructors: Darin Lah (IZUM), Žiga Zebec (IZUM), Teo Prica (IZUM), Samo Miklavc (IZUM), Dejvid Črešnar (IZUM), Samo Lorenčič (IZUM)
Learning objectives: Give basic information on HPC Vega and the new AI. It covers the basics, from creating SSH keys, logging into Vega, EuroHPC registration in comparison to SLING registration, basic Linux commands, and job submission and management. Additionally, it covers an introduction to Vega's new cloud-like AI/HPC architecture, built on OpenStack, for running containerized AI workloads. The event will also allow participants to provide information about their AI projects, along with any issues, challenges, or questions, in advance, enabling our HPC and AI-infrastructure expert teams to take a detailed look at their projects.
Course content: The workshop content will be tailored to participant AI projects, building on past use cases and established best practices for running machine learning and AI workloads on HPC systems, including efficient GPU job submission, resource usage, and common pitfalls. Attention will be given to Vega's new cloud-like AI/HPC platform, which combines OpenStack for infrastructure provisioning with Kubernetes for container orchestration, enabling scalable, reproducible, and portable AI workflows alongside traditional Slurm-based batch jobs. Practical guidance will be provided on how to improve, streamline, and accelerate AI training, fine-tuning, and inference workflows on Vega.
Learning outcomes: As a result, participants will gain a solid foundation for working independently on HPC Vega for AI workloads, a clear understanding of core HPC and cloud-native concepts including Slurm, OpenStack, and Kubernetes, and practical insights into optimizing their AI jobs to improve performance and overall efficiency in a hybrid HPC/cloud environment.