Building Better AI Apps with TF Hub

What is TF Hub

• Modeling is an important part

Modeling is an important part of any Machine Learning application, you would want to invest quite some time and effort to get this right. Also note I am not saying modeling is the only important thing, in reality there is a lot like building pipelines, serving models but we will not explore them in this blog. TF Hub helps you to do the model part better, faster, and easy.

• A way to easily discover models

You can see a GUI and a friendly version of TensorFlow Hub at tfhub.dev , you can see Hub here. You can filter models based on the type of problem you want to work for, the model format you need, and also based on the publisher. This definitely makes the model discovery process a lot easier for you.

You can find a lot of well tested and state of the art models right there on TF Hub. Each of the models there on TF Hub is well documented and thoroughly tested, so you as a developer can make the best use of it. A lot of models on TF Hub even have sample Colab Notebooks to show how they work.

• Ease in using and integrating with model APIs

What is even better is that you can easily integrate it with your model APIs, so you want to have flexibility while building models, and TF Hub helps you to do so. You will then see for yourself too how well TF Hub is integrated with Keras or Core TensorFlow APIs which really makes it super easy.

• A wide array of publishers

This may not be the first thing why you would like to consider TF Hub but its good to know the wide array of publishers TF Hub, some of the major publishers are listed in the image below.

• Without code dependencies

So a lot of times what happens is your codebases become very dependent or coupled, this can make the experimentation and iteration process a lot slower, so Hub defines artifacts for you which are not dependent on code, so you have a system which allows for faster iteration and experimentation too.

• A wide array of platforms

Another great thing about TF Hub is it does not matter if you use high-level Keras APIs or low-level APIs. You can also use it in your production-grade pipelines or systems, it integrates quite well with TensorFlow Extended too. You can also use TF.js models for web-based environments or node environments. Edge is taking off and TF Hub has covered you in that aspect too, you can use TF Lite pre-trained models to run your models directly in your mobile device and low power embedded systems. You can also discover models for Coral Edge TPUs. It is essentially just TF Lite and combining it with a powerful edge TPU.