Google Releases TensorFlow 1.7.0! All You Need to Know

Overview

• TensorFlow 1.7.0 has been released to the general public!
• It features TensorRT integration with TensorFlow
• The TensorFlow Debugger Plugin, a GUI for the TensorFlow Debugger
• Other features include eager mode coming out of contrib, easy customization of gradient computation, better text processing, and many other things

Introduction

Google has released the latest update to it’s ultra-popular library TensorFlow – version 1.7.0. TensorFlow is an open-source library written in python, C++ and CUDA. The last big update was TensorFlow 1.5 which we covered at the time of the release here.

There are a few fascinating additions to the latest update, including the integration of TensorRT with TensorFlow. Let’s check out a few features below.

TensorRT integration with TensorFlow

TensorRT is a library that optimizes deep learning models for inference and creates a runtime for deployment on GPUs in production environments. In tests performed by Google, they found that “ResNet-50 performed 8x faster under 7 ms latency with the TensorFlow-TensorRT integration using NVIDIA Volta Tensor Cores as compared with running TensorFlow only”.

TensorFlow Debugger Plugin

This is still in alpha mode. It’s a graphical user interface of the TensorFlow Debugger. This feature will enable users to:

• pause and resume execution at specified nodes or numbers of steps
• visualize values of tensors over time
• associate tensors with specific lines in python code

Eager Mode Moves out of Contrib

One of the biggest additions in the last two version was the introduction of Eager Execution. It enables users to execute TensorFlow operations as soon as they are called from python. This is not only welcome by researchers experienced with the language, but makes things easier for newcomers as well. With this update, Eager mode has come out of contrib. Try using the below command:

tf.enable_eager_execution()

You can read more about the release on the TensorFlow GitHub page here.

Other major features

• Graph rewrites emulating fixed-point quantization compatible with TensorFlow Lite, supported by new tf.contrib.quantize package.
• Easily customize gradient computation with tf.custom_gradient.
• Experimental support for reading a sqlite database as a Dataset with new tf.contrib.data.SqlDataset.
• Distributed Mutex / CriticalSection added to tf.contrib.framework.CriticalSection.
• Better text processing with tf.regex_replace.
• Easy, efficient sequence input with tf.contrib.data.bucket_by_sequence_length

How do I install TensorFlow?

It just takes a line of code:

$  pip install --ignore-installed --upgrade tensorflow

Our take on this

I can’t wait to get our hands on this! I am a huge fan of visualizations so I’m especially intrigued by the TensorFlow Debugger dashboard that gives an overview of the Tensor values.

We will update this article with how the latest version performs for us soon.

Subscribe to AVBytes here to get regular data science, machine learning and AI updates in your inbox!

Pranav Dar 29 Mar 2018

Senior Editor at Analytics Vidhya. Data visualization practitioner who loves reading and delving deeper into the data science and machine learning arts. Always looking for new ways to improve processes using ML and AI.