The European Commission published, on Tuesday 21 July, the European Union’s institutional large language model (LLM) as an open model.
This model was developed from the Directorate-General for Translation’s “high-quality multilingual data” in order to guarantee “the fair and balanced representation of all 24 official EU languages, while also understanding the terminology, legislation and policy context that make the European Union unique”, announced the DGT’s Director-General, Christos Ellinides.
The development of this artificial intelligence model was made possible thanks to the power of the European supercomputers MeluXina (Luxembourg), Leonardo (Bologna) and MareNostrum (Barcelona), combined with the open model developed by the French company Mistral.
Two versions of the EU’s institutional LLM are now available via the European Language Data Space, a platform designed to create a reliable and efficient data market for the exchange of linguistic resources between the public and private sectors. (Original version in French by Ana Pisonero Hernández)