Case Study: University of Amsterdam develops multilingual AI image model with SURF’s LUMI supercomputer

University of Amsterdam trains NeoBabel on 1,000 GPUs with SURF

The University of Amsterdam, represented by a PhD student, faced the challenge of developing an inclusive, multilingual text-to-image AI model without relying on English translations. The project required a level of computing power that exceeded the capacity of the national Dutch supercomputer, Snellius. SURF provided access to the European supercomputer LUMI to help address this significant computational need.

By leveraging SURF's support and guidance to utilize LUMI, the researchers successfully trained the NeoBabel model. This solution outperformed larger models from major Big Tech companies in accurately generating images from non-English prompts. The project's success, achieved with a model four times smaller than its competitors, demonstrated that a powerful and inclusive AI model could be developed at a Dutch university with SURF's assistance.

University of Amsterdam