[bert4keras].weights权重文件转换成.ckpt格式，便于build_transformer_model方法加载

方法步骤：

1. 建立同样的模型

   bert = build_transformer_model(
       config_path,
       checkpoint_path,
       model='roformer',
       application='unilm',
       with_pool='linear',
       with_mlm='linear',
       dropout_rate=0.2,
       ignore_invalid_weights=True,
       return_keras_model=False # 必须；参考源码：为True，返回transformer.model；为False，返回transformer
   )

2. load_weights加载你之前保存的权重

   bert.model.load_weights('best_model.weights')

3. save_weights_as_checkpoint保存为新的权重

   bert.save_weights_as_checkpoint(filename='./bert_model.ckpt')