TensorFlow Lite’s Developer Preview Finally Shared By Google

TensorFlow Lite’s Developer Preview Finally Shared By Google

Today, TensorFlow Lite, a developer preview of the software library, was released by Google.

Google Launches Mobile Web Specialist Certification For Developers
AI Breakthroughs That Took Place In 2017
Samsung Purchases Another AI Company In A Move To Improve Bixby
Back in May, when it was announced at Google I/O that there would be a brand new version of TensorFlow designed and created with mobile devices in mind, developers were raving about it.
Today, TensorFlow Lite, a developer preview of the software library, was released by Google.
The purpose of TensorFlow Lite was to create a software library more suitable for embedded devices and smartphones, thanks to the lightweight design and build. There are versions of TF Lite for both iOS and Android app developers.
Training models are not the focus. Instead, TensorFlow Lite enables less robust devices to benefit from machine learning models providing low-latency inference. This essentially means that the focus will not be the application of TensorFlow for using existing data to learn completely new capabilities, as most mobile devices lack sufficient power to handle this, but rather it will focus on applying models’ existing capabilities to new data.
It has been outlined by Google that when they created the new TF Lite, the emphasis was on a lightweight tool that would be able to start quickly and enhance various mobile devices’ load times. Android Neural Networks API is supported by TF Lite.
There is a lot more to expect from the software library though, as it is not a full release. So, expect more additions in the future. For the moment though, Google is confident that the new TF Lite has been sufficiently redesigned and ready to function with Smart Reply, Inception v3 and MobileNet along with other popular vision and natural language processing models.
As stated in a post from the TensorFlow Lite team, their aim was to ensure the library would function properly on the most common models, which is why they released the constrained preview edition. Their future plans are to expand its functions based on what users need, with the aim of enabling the deployment of TF on a variety of different embedded devices and mobiles and simplifying the experience for developers.
Developers eager to get started with the library should check out the documentation for TF Lite today.

COMMENTS