Model artifact – the SavedModel format

The SavedModel format is the default model serialization and deserialization format used by TensorFlow. In layman's terms, this can be understood as a container that holds everything there is to reproduce a model from scratch elsewhere without access to the original code that created it. We can use SavedModel to transfer trained models from the training to the inference phase or even to transfer state between different parts of the training process. In a nutshell, it can be said that SavedModel contains a complete TensorFlow program along with model weights and descriptions of the various compute operations described. While working with the Python API of TF 2.0, it is now possible to export certain native Python operations along with the model. This is facilitated largely by the tf.function and tf.autograph APIs. In the following...