Europe is finally going to build its own open-source large model, with 400 billion parameters + fully reproducible. This wave is a dual bet on computing power autonomy and technology transparency. Looking forward to seeing how it compares with Llama.

CoinWorld News: Italian AI startup Domyn has announced plans to release, within one year, a fully open-source, reproducible large model with more than 400 billion parameters. The project is jointly advanced by Domyn and the Europa Alliance formed in cooperation with Germany’s Fraunhofer Society, has been selected for the European Commission’s Frontier AI Large Challenge project, and has received computing power support from the European public supercomputing network EuroHPC. It is aimed at promoting localized alternative solutions. Domyn CEO Uljan Sharka noted that compared with serving hundreds of millions of users, the computing power required to train a cutting-edge large model from scratch is much lower, and Europe’s public computing network can already meet R&D needs. The project has received support from Abu Dhabi’s G42, Eurizon Capital, Rabobank, and The Bank of New York Mellon.