Fail to load models with custom layers about keras HOT 4 CLOSED

@jianyizh your custom layer would need a build method. Keras 3 is more strict about this than tf_keras.

from keras.

@divyashreepathihalli I added the build method in the issue description, still the same error.

from keras.

The V3 Keras weights saving format does indeed rely on object class names in order to load weights correctly. So here's what you can do (each one of these options will work):

• Rename your custom layer class Dense, the same as the original layer class.
• Transfer weights without relying on the .weights.h5 format, for instance by doing model.set_weights(other_model.get_weights()). You can use npz to store the output of get_weights() to disk (it's a list of numpy arrays).
• Edit the .weights.h5 file to rename the target "Dense" layer to "OptimizedDense" before loading.

from keras.

Are you satisfied with the resolution of your issue?
Yes
No

from keras.