This allows RoBERTa to improve on the masked language modeling objective compared with BERT and leads to better downstream task performance.
確定! 回上一頁