This week in Geneva, the International Open-Source LLM Builders Summit convened around 50 organizations focused on open-source large language models (LLMs) and trustworthy AI. Hosted by the AI centers at EPFL and ETH Zurich, the summit highlighted the emergence of open LLMs as credible alternatives to proprietary systems. Attendees previewed a forthcoming multilingual LLM, co-developed by Swiss researchers, emphasizing transparency and accessibility. The model will support over 1,000 languages and be released under an open license, promoting responsible data practices compliant with Swiss and EU regulations. Built on the advanced CSCS “Alps” supercomputer, the LLM benefits from over 15 trillion training tokens, ensuring high performance. The model, available in 8 billion and 70 billion parameters, will facilitate diverse applications in science, education, and industry. Set for release this summer, accompanying documentation will aid users in leveraging the model for innovation and collaboration across sectors, thus fostering a vibrant international AI ecosystem.
Source link
