Switzerland Releases National-Level Open-Source Large Language Model Apertus

Switzerland has released the national-level open-source large language model Apertus. Jointly developed by institutions including the Swiss Federal Institute of Technology in Lausanne (EPFL), the model is fully open-source. Its training data covers over 1,000 languages, with a total of 15 trillion tokens, of which 40% are non-English data. Two versions have been launched: one with 8 billion parameters and another with 70 billion parameters.