OpenEuroLLM is a groundbreaking European initiative launched to develop next-generation open-source language models, aiming to counter the dominance of AI technologies from Silicon Valley and China.
The project is coordinated by Jan Hajič from Charles University in Czechia and co-led by Peter Sarlin from AMD Silo AI in Finland, with a consortium of 20 partners including top European research institutions and companies.
With a total budget of €52 million, the initiative has secured €20.6 million in funding from the EU's Digital Europe Program, supplemented by additional resources through the STEP program.
OpenEuroLLM aims to enhance Europe's digital sovereignty amid rapid advancements in AI by the US and China, addressing concerns over the risk to European culture and companies.
The project has been awarded the STEP seal and will leverage prior European project experiences and data repositories to build a robust AI framework.
The initiative intends to create high-performing, multilingual large language models that can be utilized across commercial, industrial, and public services throughout Europe.
The consortium focuses on developing multilingual open-source language models that cater to the diverse needs of businesses and public services in all EU member states.
A significant challenge lies in ensuring that these models serve all EU member states equally, particularly for low-resource languages.
Despite the limited computing capacity in Europe compared to the US and China, the project will utilize facilities within the EuroHPC network for model training.
OpenEuroLLM collaborates with open-source communities such as LAION, open-sci, and OpenML to ensure full transparency and openness in its models, software, data, and evaluations.
Development will occur within Europe's regulatory framework, ensuring compliance with European values while striving for technological excellence.
Ultimately, OpenEuroLLM represents Europe's strategic response to the global AI landscape, fostering innovation while maintaining control over AI technologies.
