Under the backdrop of the booming development of artificial intelligence technology worldwide, Switzerland has recently officially launched its first fully open-source large language model - Apertus. This model was jointly developed by the Swiss Federal Institute of Technology Lausanne (EPFL), ETH Zurich, and the National Supercomputing Center (CSCS), marking an important step forward for Switzerland in terms of independent innovation in the field of artificial intelligence.
As a national-level large language model in Switzerland, Apertus has multilingual processing capabilities, covering more than 1000 languages, with 40% of the data coming from non-English languages. This feature not only enhances the inclusiveness of diversity but also enables Apertus to better serve the special language needs within Switzerland, such as Swiss German and Romansh. This open design aims to provide users with an alternative AI solution to international companies like OpenAI.
Notably, the development process of Apertus fully embodies the principles of transparency and openness. In addition to model weights, the research team has made the model architecture, training data, and training process public, using a permissive open-source license that supports educational, research, and commercial use. This measure greatly reduces the usage barrier for users, enabling more researchers and developers to participate in the innovation of artificial intelligence.
Apertus is trained on Switzerland's "Alps" supercomputer, which is equipped with over 10,000 NVIDIA Grace Hopper GPUs, and uses renewable energy, reflecting a commitment to sustainable development. According to the research team, Apertus offers two versions: one with 8 billion parameters, suitable for individual users and lightweight applications; and another with 70 billion parameters, designed for complex scenarios.
Currently, users can access Apertus through channels such as the Swisscom platform, Hugging Face, and Public AI Network. The Swiss government hopes to build artificial intelligence into a public infrastructure through this project, emphasizing compliance, transparency, and inclusive value, to enhance the innovative capabilities of research, education, and industry.
Address: https://huggingface.co/swiss-ai/Apertus-8B-Instruct-2509
