Save and restore transformer model

I trained a model on Jupyter Notebook using the transformer tutorial. However, I want to save the model and work on it more on my own project.

However, I encountered an issue in saving.

It fails to save the model using transformer.save, giving me this error:

WARNING:tensorflow:SavedModel saved prior to TF 2.5 detected when loading Keras model. Please ensure that you are saving the model with model.save() or tf.keras.models.save_model(), *NOT* tf.saved_model.save(). To confirm, there should be a file named "keras_metadata.pb" in the SavedModel directory.
---------------------------------------------------------------------------
ValueError                                Traceback (most recent call last)
[<ipython-input-96-b268ce7d2a7f>](https://localhost:8080/#) in <module>
      7 # transformrr.summary()
      8 # type(transformrr)
----> 9 new_model = tf.keras.models.load_model('transformer')
     10 
     11 type(transformer)

1 frames
[/usr/local/lib/python3.8/dist-packages/keras/saving/legacy/saved_model/load.py](https://localhost:8080/#) in _read_legacy_metadata(object_graph_def, metadata, path)
    221         ):
    222             if not proto.user_object.metadata:
--> 223                 raise ValueError(
    224                     "Unable to create a Keras model from SavedModel at "
    225                     f"{path}. This SavedModel was exported with "

ValueError: Unable to create a Keras model from SavedModel at transformer. This SavedModel was exported with `tf.saved_model.save`, and lacks the Keras metadata file. Please save your Keras model by calling `model.save` or `tf.keras.models.save_model`. Note that you can still load this SavedModel with `tf.saved_model.load`.

Ok sure, read that it was hard to serialize.

Then, I tried tf.saved_models.save

That saves and loads my models fine, but now my model is in SavedModel form.

I tried using it again for the translator but it gives this error:

ValueError Traceback (most recent call last)
in
2 ground_truth = 'this is a problem we have to solve .'
3
----> 4 translated_text, translated_tokens, attention_weights = translator(tf.constant(sentence))
5 print_translation(sentence, translated_text, ground_truth)

3 frames
/usr/local/lib/python3.8/dist-packages/tensorflow/python/saved_model/function_deserialization.py in restored_function_body(*args, **kwargs)
293 "Option {}:\n {}\n Keyword arguments: {}".format(
294 index + 1, _pretty_format_positional(positional), keyword))
--> 295 raise ValueError(
296 "Could not find matching concrete function to call loaded from the "
297 f"SavedModel. Got:\n {_pretty_format_positional(args)}\n Keyword "

ValueError: Could not find matching concrete function to call loaded from the SavedModel. Got:
Positional arguments (1 total):
* [<tf.Tensor 'inputs:0' shape=(1, 11) dtype=int64>,
<tf.Tensor 'inputs_1:0' shape=(1, 1) dtype=int64>]
Keyword arguments: {'training': False}

Expected these arguments to match one of the following 4 option(s):

Option 1:
Positional arguments (1 total):
* (TensorSpec(shape=(None, None), dtype=tf.int64, name='input_1'),
TensorSpec(shape=(None, None), dtype=tf.int64, name='input_2'))
Keyword arguments: {'training': False}

Option 2:
Positional arguments (1 total):
* (TensorSpec(shape=(None, None), dtype=tf.int64, name='inputs_0'),
TensorSpec(shape=(None, None), dtype=tf.int64, name='inputs_1'))
Keyword arguments: {'training': False}

Option 3:
Positional arguments (1 total):
* (TensorSpec(shape=(None, None), dtype=tf.int64, name='inputs_0'),
TensorSpec(shape=(None, None), dtype=tf.int64, name='inputs_1'))
Keyword arguments: {'training': True}

Option 4:
Positional arguments (1 total):
* (TensorSpec(shape=(None, None), dtype=tf.int64, name='input_1'),
TensorSpec(shape=(None, None), dtype=tf.int64, name='input_2'))
Keyword arguments: {'training': True}

I don't know where this is coming from, and I don't really know how to fix it.

I would appreciate it if someone can help me figure out how to properly save the model that keeps it in keras form, or guide me in figuring out how I can use the SavedModel form with the translator to make the translation work.

Oh, I also read about save_weights(). I tried it, but it keeps on giving me errors. If that's an easier solution, please guide me on how to do it correctly :')

Start by reproducing the save/load steps in the transformer tutorial notebook and compare the Keras and SavedModel entry points shown in the report. Inspect the model's saved input signatures and the translator call that fails. Done means the model can be restored in a documented form and the translator runs on the reported inputs.