August 2019
Intermediate to advanced
202 pages
5h 9m
English
As we've seen in preceding sections, the SavedModel format is used to produce a reproducible representation of the current compute graph (the dataflow graph). This representation is independent of the specific code that was used to create the compute graph. It is also independent of the specific process used to construct this graph. For example, the SavedModel format doesn't actively distinguish between compute graphs created using native TensorFlow operations, Keras, or even tf.function. Though we interchangeably refer to this compute graph as a model, technically, it can also be considered a combination of a trained mathematical model and some additional code written around it for performing supporting ...
Read now
Unlock full access