EPFL, ETH Zurich, and the Swiss National Supercomputing Centre (CSCS) have unveiled Apertus, Switzerland’s first large-scale open multilingual language model, marking a pivotal moment in generative AI. Released today, Apertus provides a foundational tool for developers and organizations to create applications like chatbots and translation systems, emphasizing transparency and diversity. The model is openly accessible, allowing AI researchers and enthusiasts to adapt its architecture, model weights, and training data. Available in two versions—8 billion and 70 billion parameters—Apertus supports educational and commercial uses under a permissive open-source license. With training on 15 trillion tokens, including many underrepresented languages, Apertus prioritizes ethical standards and compliance with Swiss regulations. It aims to drive innovation, fostering AI expertise across sectors. The model’s launch coincides with the Swiss {ai} Weeks initiative, promoting responsible AI applications. Apertus exemplifies a commitment to open, trustworthy AI, inspiring future enhancements and collaborations.
Source link
