Fetching the paper…

nmT5 -- Is parallel data still relevant for pre-training massively multilingual language models? · Around