Подход на основе маскировки

AbstractPre-trained language models that learn contextualized word representations from a large un-annotated corpus have become a standard component for many state-of-the-art NLP systems. Despite their successful applications in various downstream …...