Sept. 3 (Bloomberg) -- The global AI race has added a new player, this time an entire country.SwitzerlandThe official release of the nationalOpen SourceLarge Language Model(LLM)ApertusApertus is a model of the Swiss Federal Institute of Technology in Lausanne (EPFL) and the Federal Institute of Technology in Zurich (ETH Zurich), and it is hoped that it will serve as an alternative to the models offered by companies like OpenAI. "Apertus, which comes from the Latin word for "open," was developed jointly by the Swiss Federal Institute of Technology in Lausanne (EPFL), the Swiss Federal Institute of Technology in Zurich (ETH Zurich), and the Swiss National Supercomputing Center (CSCS), all public institutions. The model was jointly developed by EPFL, ETH Zurich and CSCS, all three of which are public institutions.
"Currently, Apertus is the leading public AI model: built by public organizations to serve the public interest. This is our strongest demonstration to date - that AI is fully capable of becoming public infrastructure like roads, water, and electricity." said Joshua Tan, a core advocate in the push to make AI public infrastructure.
The Swiss R&D team designed Apertus to be fully open source, allowing users to view all aspects of the training process. In addition to the model itself, the team has also made public the full documentation of the training process, the source code, and the datasets used, and Apertus has been developed in strict compliance with Swiss data protection and copyright laws, making it perhaps a better choice for companies wishing to comply with European regulations. The Swiss Bankers Association has previously noted that locally developed big-language models have "great long-term potential" because they are better suited to Switzerland's strict local data protection regulations and banking secrecy. Swiss banks are already using other AI models to meet their business needs, and it remains to be seen whether they will turn to Apertus in the future.
According to 1AI, Apertus is open to all users: researchers, AI enthusiasts, and even enterprises can use the model for secondary development and customize functions according to their needs. For example, users can use it to develop chatbots, translation tools, or even educational or training applications.The training data of Apertus covers more than 1,000 languages, with a total token volume of 15 trillion, of which non-English data accounts for 40%, including Swiss German, Romansh and other local languages. According to an official statement from Switzerland, the model is trained using only publicly available data, and its data crawlers honor machine-readable "refuse to crawl" requests on the site. It should be noted that AI firms such as Perplexity have been accused of crawling websites and bypassing protocols used to block crawlers, and some have faced legal action for using content from news organizations and creators to train models without permission.
Currently, Apertus is available in two parameter sizes, 8 billion parameters and 70 billion parameters. The model is available through Swisscom, the Swiss ICT company, or Hugging Face, an AI platform.