TensorFlow is an open-source software library that across a range of tasks is used for data flow programming. It is also a symbolic math library, that is used for machine learning applications such as neural networks as well.

The Open-Source Software Library has had many releases, that came forward with more improvements and fixes. A new in the list has recently Arrived. Marking the release of a New Version Of Tensorflow 1.5.0.

In this article, we'll take a look at the major changes, improvements and fixes and more that have taken place in this new version:

Breaking Changes in the new release:

• Prebuilt binaries that are now built against CUDA 9 and cuDNN 7.

• These Linux binaries are built making the use of ubuntu 16 containers, potentially introducing glibc incompatibility issues with ubuntu 14.

• Starting from 1.6 release, these prebuilt binaries will make the use of AVX instructions.This may result in the breaking of TF on older CPUs.

Major Features And Improvements in Tensorflow 1.5.0:

• Eager execution mode, the preview version of the same is now made to be available.

• TensorFlow Lite, which is a module for using TensorFlow models to do inference in mobile applications, has also seen an expansion: dev preview is now available as well.

• CUDA 9 and cuDNN 7 support.

Bug Fixes as well as some Other Changes:

• The addition of auto_correlation to tf.contrib.distributions.

• Addition of the DenseFlipout probabilistic layer which uses weight perturbations to achieve decorrelated minibatch gradients.

• Restandardization of DenseVariational as a simpler template for other probabilistic layers.

• Make both the tf.contrib.distributions QuadratureCompound classes to support batch.

• Stream:: BlockHostUntilDone now returns Status rather than bool.

• Customization of request timeouts for the GCS file system.

Also, going back to the release, one of the best things about the new TensorFlow release 1.5.0 is the Eager Execution mode, which allows TensorFlow to be used much more similar to that of NumPy. This will, in return, result in the alleviation of some of the cumbersome computational graph constructions and session running that has been a hallmark of TensorFlow since the start.

For More Information: GitHub