Extending Keras models for TFLite authoring inspection

I am referring to the following example.

The tutorial shows a nice way to get a handle on whether the model I am authoring will be compatible with TFLite (and its GPU delegate) or not. However, it shows it only with tf.functions.

How can we use Keras models here?

@khanhlvg

Do you mean to add an example with from_keras_model: