I believe that's due to BERT's second training object – Next sentence classification. That objective seemingly trains the model to ...
確定! 回上一頁