tf_bert convert to pytorch_bert

Update for latest transformers, add modeling_bert.py:116:

if name in ['output_weights', 'output_bias']:
    name = 'classifier/' + name

and convert_bert_original_tf_checkpoint_to_pytorch.py

config.num_labels = 1089 # 微调的标签个数，由自身微调的bert决定
print(f"Building PyTorch model from configuration: {config}")
model = BertForSequenceClassification(config)

参考github transformers issues