keras 中的层规范化重命名

Question

I am trying to update a code that generates a model checkpoint from TF1 to TF2. However, the TF2 code renames the layer normalization as follows:

• ../../layer_normalization_4/../..
• ../../layer_normalization_1/../..
• ../../layer_normalization_5/../..

It adds a suffix _n to the variable name. I would like it to be only../../layer_normalization/../.., without _n. Doing some research, i've noticed that this may be a Keras (tf2) behaviour.

Thanks in advance

Answer 1

You can't unfortunately, all Keras layers must have unique names eg:

import tensorflow as tf
from tensorflow.keras.layers import Dense, Input
from tensorflow.keras.models import Model

input = Input(shape=(50,))
model = Dense(200,activation='relu', name='dense1')(input)
model = Dense(200,activation='relu',name='dense1')(model)
output = Dense(2,activation='sigmoid')(model)
model = Model(input,output)

Will give the error:

ValueError: The name "dense1" is used 2 times in the model. All layer names should be unique.