Apertus: a fully open, transparent, multilingual language model
SMRTR summary
EPFL, ETH Zurich, and CSCS have released Apertus, a fully open-source language model where all components including architecture, training data, and model weights are completely accessible and documented. Trained on 15 trillion tokens across over 1,000 languages with 40% non-English content, Apertus comes in 8 billion and 70 billion parameter versions and prioritizes transparency, multilingualism, and compliance with data protection laws.
SMRTR provides this summary for quick context. The original article belongs to lobste.rs.
Read the original article