Swiss Universities Publish First Fully Open AI Language Model

(CONNECT) Researchers from the Swiss Federal Institutes of Technology in Zurich(ETH) and Lausanne(EPFL), together with the Swiss Supercomputing Center(CSCS) in Lugano, have launched the open and transparent language model Apertus. According to a press release, this is the first Large Language Model (LLM) in Switzerland to be supported by artificial intelligence. The language model will serve as the basis for further applications such as chatbots, translation systems and digital learning tools.

As its Latin name for "open" indicates, Apertus is a freely accessible, transparent and comprehensively documented system. It is available in two model sizes with 8 billion or 70 billion parameters. The smaller version is suitable for individual use. However, both models are managed under a user-friendly permissive open source license.

"Apertus was developed as a contribution to the common good. It is one of the few fully open LLMs of this size and is the first of its kind to combine multilingualism, transparency and compliance as fundamental design principles", Imanol Schlag, Technical Director of the LLM project and Research Scientist at ETH, is quoted in the press release.

The universities are taking a new approach with Apertus. "Apertus is not a classic technology transfer from research to product. Rather, we see the model as a catalyst for innovation and as a means of expanding AI expertise in research, society and business," says Thomas Schulthess, Director of CSCS and Professor at ETH Zurich.

Apertus is accessible via the Swisscom platform or via Hugging Face, a French-American platform for AI models, and is particularly suitable for research and teaching. Daniel Dobos, Research Director at Swisscom, is delighted that Swisscom is involved in a "secure and responsible AI ecosystem that serves the common good and strengthens Switzerland's digital sovereignty". ce/ww