tensorflow: tf.keras.models.load_model does not load optimizer's weigths
System information Tensorflow: 2.5.0
Describe current behaviour
When using tf.keras.models.load_model() on a savedmodel containing a keras model saved using tf.keras.Model.save(include_optimizer=True), the optimizer’s weights are NOT loaded
Describe expected behaviour The reconstructed model should contain the optimizer’s weights to be able to resume training.
Standalone code to reproduce issue https://colab.research.google.com/drive/1Y0y-pvE2QDw1Wj68XnbpLaUggXAXyzcx?usp=sharing
About this issue
- Original URL
- State: closed
- Created 3 years ago
- Comments: 17 (4 by maintainers)
@alvaro-garcia-carrasco
model.save_weightsmethod will help you to extract weights from saved model format .Please find the gist.Thanks@alvaro-garcia-carrasco
I saved model in hdf5 format and could be able to see the optimizer’s weights loaded in the reconstructed model.Please find the gist.Thanks
@UsharaniPagadala
The internals of how the graph and the variables are stored is not relevant. I am trying to load a SavedModel following the instructions in the documentation https://www.tensorflow.org/guide/keras/save_and_serialize#savedmodel_format and the optimizer’s weights are not being loaded, as shown in the standalone code I provided to reproduce the issue.