What's New in TensorFlow 2.0

Understanding a model from an inference perspective

One thing that developers implementing machine learning-based applications can rely on to make life easy is that the process to serve models to users is more or less the same, regardless of the actual computations in the models being served. This implies that, if implemented correctly, engineers potentially wouldn't have to rebuild the deployment pipelines every time data scientists update the models. This can be achieved by leveraging the power of abstractions. A key abstraction here is the format in which models are stored and loaded. By introducing a standardized format, TF 2.0 makes it easy to train a model in one environment and then use it across platforms. In TF 2.0, the standard ...