TF saved model, can we predict without converting to ELWC first?

[RodrigoVillatoro]

I trained and saved a TF model following the approach found here: https://github.com/tensorflow/ranking/blob/master/tensorflow_ranking/examples/keras/antique_kpl_din.py

The model is saved this way:

model.save(
      filepath=saved_model_path,
      signatures=tfr.keras.saved_model.Signatures(
          model,
          context_feature_spec=context_feature_spec,
          example_feature_spec=example_feature_spec,
          mask_feature_name=_MASK)())

Later to predict I do the following:

loaded_model = tf.saved_model.load(saved_model_path)
predict_function = loaded_model.signatures['tensorflow/serving/predict']
ELWC = convert_to_elwc(df_to_rank)
predictions = predict_function(inputs=tf.convert_to_tensor([ELWC.SerializeToString()]))

The convert_to_elwc function takes a pandas dataframe and converts it to a ELWC proto by iterating on each of the examples. My question: is there a way to skip this conversion? (i.e. take the original dataframe.values and pass them to the model directly)?