Unrecognized keyword arguments passed to GRU: {'time_major': False} using TensorFlow 2.18 and Python 3.12

[HkThinker]

Hi everyone,

I encountered an issue when trying to load a Keras model saved as an .h5 file. Here's the setup:

• Python version: 3.12

• TensorFlow version: 2.18

• Code:

  from tensorflow.keras.models import load_model
  
  # Trying to load a pre-saved model
  model = load_model('./resources/pd_voice_gru.h5')

• Error message:

  ValueError: Unrecognized keyword arguments passed to GRU: {'time_major': False}
  

What I've tried:

1. Checked the TensorFlow and Keras versions compatibility.
2. Attempted to manually modify the .h5 file (unsuccessfully).
3. Tried to re-save the model in a different format but don't have the original training script.

Questions:

1. Why is time_major causing this error in TensorFlow 2.18?
2. Is there a way to ignore or bypass this parameter during model loading?
3. If the issue is due to version incompatibility, which TensorFlow version should I use?

Any help or suggestions would be greatly appreciated!

Thank you!

[fchollet]

That option has been removed in Keras 3 (it was extremely niche). It looks like your saved model file makes a reference to the arg but doesn't use the option (time_major=False).

The solution is to remove the arg. If you have still access to the code, you can:

1. Reload the model in Keras 2
2. Save the weights only (model.save_weights(fname))
3. Reload the model in Keras 3 (typically no code changes are necessary, but occasionally might be needed)
4. Reload the weights file