Image
Summary
Developing and testing new AI methods often requires large-scale computing resources. Therefore, with support from ENCCS, Modulai uses free access to European supercomputers to investigate how smaller AI models can be trained more efficiently.
Challenge: making smaller AI models more practical
Modulai develops AI solutions for real business use. In particular, the company focuses on smaller language models that are easier to deploy, cheaper to operate, and more energy efficient than large-scale alternatives. To reach this goal, Modulai applies methods where smaller models learn from larger, more advanced ones. However, this approach requires substantial computational resources, especially during training and evaluation.
Why supercomputers are needed
Training and testing methods for transferring knowledge from large language models to smaller ones require many experiments on large datasets. As a result, standard computing environments struggle to handle these workloads. To work efficiently and reliably, Modulai therefore needs access to high-performance computing resources.
ENCCS support and access to resources
ENCCS supports the company in identifying suitable resources and in applying for access to European supercomputers through EuroHPC. This access enables Modulai to carry out planned experiments in a stable and reproducible way.
Expected impact on AI development
By using smaller models for targeted tasks, Modulai expects to significantly reduce operational costs. For example, estimates indicate that inference costs could drop to around ten percent compared to larger models. At the same time, the company expects lower energy consumption and reduced infrastructure requirements.
Expected cost and energy benefits
With access to supercomputers, Modulai plans to develop effective methods for tailoring smaller language models to specific domains. Consequently, these models are expected to perform at a level comparable to larger models on selected tasks. In addition, they should be easier to deploy in secure, on-premise environments, such as internal systems in the financial sector.
ENCCS supports companies at an early stage
ENCCS guides Modulai in identifying suitable computing resources and supports the company throughout the application process for European supercomputers via EuroHPC. As a result, Modulai can carry out planned experiments in a more stable and reproducible way.
Do you need more computational power for your projects?
Is your organisation exploring similar challenges? ENCCS supports companies and public organisations in accessing and using European supercomputing resources for free.