The number of unique tokens (words or word pieces) a model can recognize and process; larger vocabularies provide better coverage of a language.
Quality of non-English language understanding and generation