In this article, we will be discussing saving loading models using TensorFlow 2.0+. This is a beginner-intermediate level article meant for people who have just started out using TensorFlow for their deep learning projects.

Why do you need to save a model?

One of the very common mistakes people make as a beginner in deep learning is not saving their models.

Saving a deep learning model both during training and after training is a good practice. It saves your time and enhances the reproducibility of the model. Here are a few more reasons that you might consider for saving a model:

• Training modern deep learning models with millions of parameters and huge datasets can be expensive in terms of computation and time. Moreover, you can get different results/accuracy during different training. So it is always a good idea to use a saved model for displaying your results rather than training on the spot.
• Saving the different version of the same models allows you to inspect and understand the working of the model.
• You can use the same compiled model in different languages and platforms that support TensorFlow eg.: TensorFlow Lite and TensorFlow JS without converting any of your code.

TensorFlow happens to offer a number of ways to save a model. We will be discussing all of them in detail in the next few sections.

#python modules