May. 26, 2025
OpenEuroLLM secures compute on the LUMI Supercomputer
OpenEuroLLM has been awarded 1.5 million GPU hours of computing resources on the LUMI supercomputer through the Finnish LUMI Extreme Scale Access 2025 call. The computing resources will be available for one year, providing time for essential experimentation.
“As one of Europe’s premier supercomputing facilities, the compute allocation from LUMI represents an important advancement in the work towards the project’s goal of open LLMs for Europe. LUMI is one of Europe’s supercomputers that runs on AMD technology, adding not only one of the most performant compute facilities but also diversity to Europe’s technological infrastructure and enabling stronger resilience.” notes Peter Sarlin, co-founder and CEO of AMD Silo AI, and co-lead of the OpenEuroLLM project.
Work conducted with the help of this allocation will focus on key architectural and data related decisions for the next generation of European LLMs, including:
“With the help of this first compute grant for model architecture and dataset design we will be able to make the right choices as we work towards our first large model releases. Our team has extensive experience of LLM work on LUMI and I'm confident that we will be able to make rapid progress on our goals with the support of these resources.” says Sampo Pyysalo, Research Fellow at the University of Turku, leading the TurkuNLP research group.
This approach centers on experimenting with smaller-scale models to guide the development of larger models. By testing various approaches and architectures, the aim is to identify strategies that lead to effective LLMs. Such targeted research will help ensure the quality of the large-scale models that are the primary objective of the OpenEuroLLM project.
The results will benefit the community working on open LLMs and organizations developing applications based on these models, particularly in Europe. The OpenEuroLLM consortium looks forward to contributing to the advancement of open, multilingual LLMs and supporting innovation across the European tech ecosystem.
Image: Fade Creative
“As one of Europe’s premier supercomputing facilities, the compute allocation from LUMI represents an important advancement in the work towards the project’s goal of open LLMs for Europe. LUMI is one of Europe’s supercomputers that runs on AMD technology, adding not only one of the most performant compute facilities but also diversity to Europe’s technological infrastructure and enabling stronger resilience.” notes Peter Sarlin, co-founder and CEO of AMD Silo AI, and co-lead of the OpenEuroLLM project.
Work conducted with the help of this allocation will focus on key architectural and data related decisions for the next generation of European LLMs, including:
- Data pre-processing, filtering, and cleaning methods
- Effective data mixtures for LLM training
“With the help of this first compute grant for model architecture and dataset design we will be able to make the right choices as we work towards our first large model releases. Our team has extensive experience of LLM work on LUMI and I'm confident that we will be able to make rapid progress on our goals with the support of these resources.” says Sampo Pyysalo, Research Fellow at the University of Turku, leading the TurkuNLP research group.
This approach centers on experimenting with smaller-scale models to guide the development of larger models. By testing various approaches and architectures, the aim is to identify strategies that lead to effective LLMs. Such targeted research will help ensure the quality of the large-scale models that are the primary objective of the OpenEuroLLM project.
The results will benefit the community working on open LLMs and organizations developing applications based on these models, particularly in Europe. The OpenEuroLLM consortium looks forward to contributing to the advancement of open, multilingual LLMs and supporting innovation across the European tech ecosystem.
Image: Fade Creative