TensorFlow 1.6.0-rc0 Released!

Feb. 7, 2018, 1:44 a.m. By: Kirti Bakshi

TensorFlow 1.6.0

TensorFlow that as we know is a scalable fast and flexible open-source machine learning library that is made use for research and production has now released TensorFlow 1.6.0-rc0 which brings itself to us with many changes, fixes and more.

This new version of TensorFlow: 1.6.0 rc0 seems to go up very quickly. In its previous version 1.5.0, Eager Execution and TensorFlow Lite had been added. And talking of the 1.4.0 version, binary was provided with CUDA 9 and cuDNN 7 that started from 1.5.0. From this very version 1.6.0-rc0, binary will be provided to support AVX instruction of CPU as well.

A list of the few of the changes that came up with this release and more related to the same have been mentioned below:

Breaking Changes in 1.6.0-rc0:

  • Prebuilt binaries are now built against CUDA 9.0 and cuDNN 7.

  • Prebuilt binaries will make the use of AVX instructions as mentioned before. This may break TF on older CPUs.

Major Features And Improvements:

  • tf.estimator.{FinalExporter,LatestExporter} now export the stripped SavedModels as well. This will result in the improvement of the forward compatibility of the SavedModel.

  • FFT support added to XLA CPU/GPU.

Bug Fixes and Other Changes:

1. Documentation updates:

  • Addition of the second version of Getting Started, which is aimed at ML newcomers.

  • Clarification of the documentation on resize_images.align_corners parameter.

  • Additional documentation for TPUs.

2. Google Cloud Storage (GCS):

  • Addition of client-side throttle.

  • Addition of a FlushCaches() method to the FileSystem interface, with an implementation for GcsFileSystem.

3. Other Changes:

  • New Optimizer internal API for non-slot variables. Descendants of AdamOptimizer that access _beta[12]_power will need to be updated.

  • Addition of SeparableConv1D layer.

  • Addition of convolutional Flipout layers.

  • Addition of auto_correlation to distributions.

  • Addtion of SeparableConv1D layer.

  • Output variance over trees predictions for classifications tasks.

  • For pt and eval commands, allow writing tensor values to the filesystem as numpy files.

  • gRPC: Propagate truncated errors (instead of returning gRPC internal error).

  • Augment parallel_interleave to support 2 kinds of prefetching.

  • Improved XLA support for C64-related ops log, pow, atan2, tanh.

  • Addition of probabilistic convolutional layers.

Changes in its API:

  • Introduction of prepare_variance boolean with the default setting to False for backward compatibility.

  • Move layers_dense_variational_impl.py to layers_dense_variational.py.

For More Information: GitHub