ValueError when loading models that reuse weights

[K1521]

if I create a model wich reuses layers I get a Error when trying to load it again.

import tensorflow as tf
from keras.models import load_model,save_model
from keras import layers

inputs = layers.Input(shape=(10,))
x=inputs

t=layers.Dense(10)
x = t(x)
x = layers.Dense(10)(x)
x = t(x)
model=tf.keras.Model(inputs, x)

model.summary()
save_model(model,'testmodel.keras')

model2=load_model('testmodel.keras')
model2.summary()

I also found out how to fix it:
https://github.com/keras-team/keras/blob/d3671cf276d838599dd8acec9616845ac262d52a/keras/src/models/functional.py#L687C3-L690C56

This ValueError has to be a IndexError (like in the legacy case in the code above).

That way it can be caught here:
https://github.com/keras-team/keras/blob/d3671cf276d838599dd8acec9616845ac262d52a/keras/src/models/functional.py#L517C1-L525C36

:)

[mehtamansi29]

Hi @K1521 -

Thanks for reporting this issue. Here While creating the model you can reuse the layer weight store in variable(t) t=layers.Dense(10) instead of declaring new layer x = layers.Dense(10)(x).
Attached gist here for your reference.

[K1521]

I know that it works if you feed the layer in itself, but that is not what i want to do. I have reduced a larger example down to this example.
As i have said , i have solved the problem by replacing the ValueError with a IndexError. That way the code is more similar to the legacy case here:

keras/keras/src/models/functional.py

Line 662 in d3671cf

raise IndexError(

i think somebody just translated the legacy case wrong