Pretrained model not accepting optimizer

[asegurat]

Hello everybody.

I am trying to optimize the hyperparameters of a pre-trained model (based on BERT) with TensorFlow 2.15.
A few months ago, this code was working for me, but now it is not working.

def build_model(hp):
  hf_model_name = "dccuchile/bert-base-spanish-wwm-cased"
  model = TFBertForSequenceClassification.from_pretrained(hf_model_name)

  # Adjustable hyperparameters
  hp_learning_rate = hp.Float("learning_rate", min_value=1e-6, max_value=1e-4, sampling="log")
  hp_epsilon = hp.Float("epsilon", min_value=1e-9, max_value=1e-7, sampling="log")
  hp_epochs = hp.Int("epochs", min_value=2, max_value=9, step=1)

  # Fixed hyperparameters
  loss = keras.losses.SparseCategoricalCrossentropy(from_logits=True)
  metric = keras.metrics.SparseCategoricalAccuracy('accuracy')

  optimizer = keras.optimizers.Adam(learning_rate=hp_learning_rate, epsilon=hp_epsilon, clipnorm=1.0)
  model.compile(optimizer=optimizer, loss=loss, metrics=[metric])

  return model

from kerastuner.tuners import RandomSearch

tuner = RandomSearch(
    build_model,
    objective='val_accuracy',
    max_trials=20,
    directory=drive_path + "results",
    project_name="bert_hyperparameter_tuning"
)

tuner.search_space_summary()

I get the following error messages:

---------------------------------------------------------------------------
ValueError                                Traceback (most recent call last)
[<ipython-input-30-600d78f7bf50>](https://localhost:8080/#) in <cell line: 3>()
      1 from kerastuner.tuners import RandomSearch
      2 
----> 3 tuner = RandomSearch(
      4     build_model,
      5     objective='val_accuracy',

8 frames
[/usr/local/lib/python3.10/dist-packages/tf_keras/src/optimizers/__init__.py](https://localhost:8080/#) in get(identifier, **kwargs)
    332         )
    333     else:
--> 334         raise ValueError(
    335             f"Could not interpret optimizer identifier: {identifier}"
    336         )

ValueError: Could not interpret optimizer identifier: <keras.src.optimizers.adam.Adam object at 0x7ed054b91030>

Is there any solution for this problem?
In the transformer library, something similar happened to me, and the (temporary) solution is to set an environment variable.

TF_USE_LEGACY_KERAS=1

Thanks in advance.