Poster (After Call)

FZJ-2026-04710

http://join2-wiki.gsi.de/foswiki/pub/Main/Artwork/join2_logo100x88.png Best Practices for Training LLMs on HPC Systems

Penke, C. (Corresponding author)FZJ* ; Herten, A.FZJ* ; John, C. M.FZJ* ; Ebert, J. ; Kesselheim, S.FZJ*

2026

This record in other databases:  

Please use a persistent id in citations: doi:10.34734/FZJ-2026-04710

Abstract: The training of large language models (LLMs) requires substantial computational resources, complex software stacks, and carefully designed workflows to achieve scalability and efficiency. This poster presents best practices and insights gained from the training of open, multilingual LLMs optimized for European languages. We detail the use of high-performance computing (HPC) systems, primarily JUWELS Booster at the Jülich Supercomputing Centre, for training a 7-billion-parameter transformer model. We include details about system architecture, training infrastructure, software choices, profiling and benchmarking tools, as well as engineering and operational challenges. Measured throughput data is provided that sheds light on 3D parallelism and the impact of features such as flash attention during training.

---

Contributing Institute(s):

1. Jülich Supercomputing Center (JSC)

Research Program(s):

1. 5112 - Cross-Domain Algorithms, Tools, Methods Labs (ATMLs) and Research Groups (POF4-511) (POF4-511)
2. ATML-X-DEV - ATML Accelerating Devices (ATML-X-DEV) (ATML-X-DEV)
3. NOVAS - Novel System Architectures Design (NOVAS) (NOVAS)

Appears in the scientific report 2026

---

Database coverage:

---