Switzerland launches its own open source AI model
By ForGeeks
The race for leadership in artificial intelligence has a new player – an entire country. Switzerland has unveiled Apertus, an open source national LLM model that is intended to be an alternative to solutions from companies like OpenAI. The name Apertus means “open” in Latin. The Federal Institute of Technology Lausanne (EPFL), ETH Zurich and the Swiss National Supercomputer Center (CSCS) – all of which are government organizations – were involved in the development.
“Today, Apertus is the leading public AI model: built by public institutions in the public interest. It is the best proof that artificial intelligence can be a form of public infrastructure, like roads, water or electricity,” said Joshua Tan, a proponent of the concept of AI as a public resource.
“Apertus is now the leading public AI model: built by public institutions in the public interest.
Apertus is designed to be completely open source: users can explore all stages of its training. In addition to the model itself, the source code, documentation and datasets used in the training are published. The entire process complies with Swiss data protection and copyright laws, making it particularly attractive to European companies. The Swiss Bankers Association has previously stated that a national LLM has “great long-term potential” as it will be able to fully comply with strict local data protection and bank secrecy requirements. For now, banks are using other models, but it is possible that they will switch to Apertus in the future.
Apertus is the only LLM in the world that has been developed by the Swiss bankers’ association.
The new model is open to everyone – researchers, enthusiasts and companies – who can adapt it to their needs. Apertus can be used to create chatbots, translators, education and training services. The model has been trained on 15 trillion tokens in over 1,000 languages, with 40% of the data being non-English texts, including Swiss German and Romansh. Importantly, only open sources were used in training, and its crawlers took into account data collection bans posted on websites. This distinguishes Apertus from some AI companies that have been accused of ignoring protocols and illegally using content.
Apertus has been accused of ignoring protocols and illegally using content.
The model is available in two versions, with 8 and 70 billion parameters. You can get access through Swisscom or the Hugging Face platform.
