google的bert预训练模型: BERT-Large, Uncased (Whole Word Masking): 24-layer, 1024-hidden, 16-heads, 340M parameters BERT-Large, Cased (Whole Word Maskin_牛客网_牛客在手,offer不愁