Keras Tuner crashes with Keras 3.0

[To3No7]

Describe the bug
When I was trying our Keras 3.0 it seems that Keras Tuner is crashing due to compatibility problems

First it seems to fail on check “isinstance(model, keras.models.Model)”

    165         # Stop if "build()" does not return a valid model.
    166         if not isinstance(model, keras.models.Model):
--> 167             raise errors.FatalTypeError(
    168                 "Expected the model-building function, or HyperModel.build() "
    169                 "to return a valid Keras Model instance. "

FatalTypeError: Expected the model-building function, or HyperModel.build() to return a valid Keras Model instance. Received: <Sequential name=sequential_1, built=True> of type <class 'keras.src.models.sequential.Sequential'>.

Then there is some problem what is returned from model.fit()

FatalTypeError: Expected the return value of HyperModel.fit() to be one of float, dict, keras.callbacks.History, or a list of one of these types. Recevied return value: <keras.src.callbacks.history.History object at 0x7f5974212050> of type <class 'keras.src.callbacks.history.History'>.

If one get past these two then there are problems with the new way to handle save and load models. C.f. https://groups.google.com/g/keras-users/c/jgGcX730WGE

RuntimeError: Number of consecutive failures exceeded the limit of 3.
Traceback (most recent call last):
  File "~/.local/lib/python3.10/site-packages/keras_tuner/engine/base_tuner.py", line 273, in _try_run_and_update_trial
    self._run_and_update_trial(trial, *fit_args, **fit_kwargs)
  File "~/.local/lib/python3.10/site-packages/keras_tuner/engine/base_tuner.py", line 238, in _run_and_update_trial
    results = self.run_trial(trial, *fit_args, **fit_kwargs)
  File "/tmp/ipykernel_22160/2305993176.py", line 9, in run_trial
    return super(MyTuner, self).run_trial(trial, *args, **kwargs)
  File "~/.local/lib/python3.10/site-packages/keras_tuner/engine/tuner.py", line 314, in run_trial
    obj_value = self._build_and_fit_model(trial, *args, **copied_kwargs)
  File "~/.local/lib/python3.10/site-packages/keras_tuner/engine/tuner.py", line 233, in _build_and_fit_model
    results = self.hypermodel.fit(hp, model, *args, **kwargs)
  File "~/.local/lib/python3.10/site-packages/keras_tuner/engine/hypermodel.py", line 149, in fit
    return model.fit(*args, **kwargs)
  File "~/.local/lib/python3.10/site-packages/keras/src/utils/traceback_utils.py", line 123, in error_handler
    raise e.with_traceback(filtered_tb) from None
  File "~/.local/lib/python3.10/site-packages/keras_tuner/engine/tuner_utils.py", line 76, in on_epoch_end
    self._save_model()
  File "~/.local/lib/python3.10/site-packages/keras_tuner/engine/tuner_utils.py", line 86, in _save_model
    self.model.save_weights(write_filepath)
ValueError: The filename must end in `.weights.h5`. Received: filepath=output/FashionMNIST/trial_02/checkpoint

To Reproduce
I believe this will be true for any example running with Keras 3.0, but here is a colab notebook that shows this behaviour for the first type of crash, if that is fixed then the others will appear!
https://colab.research.google.com/drive/1v-RblPL4XiYY_AmGW47xLL6v5sbq2ggY?usp=sharing

Expected behavior
No crash!

Additional context
TensorFlow version: 2.16.0-dev20231022
Keras version: 3.0.0.dev2023102203
Keras Tuner version: 1.4.5

Would you like to help us fix it?
Would be interested, but not sure I have the knowledge to do that.

[haifeng-jin]

Thank you so much for the issue, @To3No7 !
I am looking into this.
KerasTuner v1.4.5 is build with keras-core instead of keras-nightly.
We will support the keras-nightly soon.

[To3No7]

I was testing KerasTuner today see colab file and now it works again, good job!

Keras Tuner version: 1.4.7.dev0
TensorFlow version: 2.16.0-dev20231111
Keras version: 3.0.0.dev2023111003