The TensorFlow models are trained and validated in the development environment. Once released, they need to be hosted somewhere to be made available to application engineers and software engineers to integrate into various applications. TensorFlow provides a high-performance server for this purpose, known as TensorFlow Serving.
For serving TensorFlow models in production, one would need to save them after training offline and then restore the trained models in the production environment. A TensorFlow model consists of the following files when saved:
- meta-graph: The meta-graph represents the protocol buffer definition of the graph. The meta-graph is saved in files with the .meta extension.
- checkpoint: The checkpoint represents the values of various variables. The checkpoint is saved in two files: one with the .index extension and...
