MCML at LRZ — Pitchtalks, Meetup, and Guided Tour

On 12 May, MCML members visited the Leibniz Supercomputing Centre (LRZ) in Garching for an afternoon dedicated to the intersection of machine learning research and high-performance computing. The event combined three short research talks, an open Q&A, and a guided tour of the LRZ’s facilities, including SuperMUC-NG.

Pitchtalks:
The programme opened with three pitchtalks bringing together perspectives from both institutions:

• Michael Hoffmann (LRZ) presented Llama-GENBA-10B: A Trilingual Large Language Model, introducing recent work on multilingual model development carried out at the LRZ and the technical considerations involved in training a model spanning German, English, and Bavarian.
• Kanishk Bhatia (MCML) followed with AI Methods for Partial Differential Equations, outlining how modern machine learning approaches are being applied to one of the long-standing challenges in scientific computing, and where the boundaries between classical numerical methods and learned solvers currently lie.
• Ajay Navilarekal (LRZ) closed the talk session with Large-Scale LLM Training on SuperMUC-NG Phase 2, providing concrete insights into how Europe’s leading academic supercomputer is being prepared and used for training large language models at scale.

Q&A and Discussion
Following a short break, the speakers joined MCML attendees for an open Q&A. The discussion ranged from practical questions around scaling, data pipelines, and software stacks on SuperMUC-NG, to broader questions about how academic HPC centres and ML research groups can collaborate more closely as model and dataset sizes continue to grow.

Guided Tour
The afternoon concluded with guided tours of the LRZ, including a visit to the SuperMUC-NG machine room and an overview of the infrastructure currently being deployed for the next generation of AI workloads in Bavaria. For many MCML members, this was a first opportunity to see the systems behind the compute allocations that increasingly shape their research.

Outlook
The visit underlined how closely the future of machine learning research is tied to the availability of, and access to, world-class compute infrastructure. We thank the LRZ team for hosting us and for the generous insight into their work. MCML and LRZ will continue to deepen this collaboration through joint events, shared projects, and exchange between researchers — building stronger bridges between machine learning and high-performance computing in Munich and beyond.