1、使用很小的学习率进行学习，且：
for l in bert_model.layers:
l.trainable = True
2、由于bert模型巨大，我们每次训练只能取batch=4进行训练，而训练4个epoch之后，可以freeze bert模型，单独训练softmax
for l in bert_model.layers:
l.trainable = False
__________________________________________________________________________________________________
Layer (type) Output Shape Param # Connected to
==================================================================================================
input_1 (InputLayer) (None, None) 0
__________________________________________________________________________________________________
input_2 (InputLayer) (None, None) 0
__________________________________________________________________________________________________
model_2 (Model) multiple 101677056 input_1[0][0]
input_2[0][0]
__________________________________________________________________________________________________
lambda_1 (Lambda) (None, 768) 0 model_2[1][0]
__________________________________________________________________________________________________
dense_1 (Dense) (None, 50) 38450 lambda_1[0][0]
==================================================================================================
Total params: 101,715,506
Trainable params: 38,450
Non-trainable params: 101,677,056
调整学习率，单独训练几个epoch
尤其是我们有一个类是其他，这种类别不确定的分类问题的时候，单独训练softmax有很大的帮助