TensorFlow 2.1.0 adds experimental features and breaking changes
TensorFlow 2.1.0 has been released, following two release candidates. The final version of the machine learning platform includes new features and breaking changes. Meanwhile, Python 2.7 has reached its end of life and is no longer supported by TensorFlow. Let’s take a look at what else has changed.
TensorFlow has reached version 2.1.0. The open source library for machine learning includes breaking changes and bug fixes as well as additional features for the tensorflow pip package, tf.keras, tf.data and more.
Support for Python 2.7 has been dropped as of January 1, 2020, the official end of life date for this language version. The current version of Python is 3.8.1, released in December 2019.
What’s new in TensorFlow 2.1.0
In the latest TensorFlow release, the tensorflow pip package has received an update: GPU support is now included by default for Linux and Windows on machines with and without NVIDIA GPUs.
For Windows users, there is another change: Visual Studio 2019 version 16.4 is now used to create official tensorflow pip packages, as it offers the new
/d2ReducedOptimizeHugeFunctions compiler flag. These pip packages require “Microsoft Visual C++ Redistributable for Visual Studio 2015, 2017 and 2019,” available at Microsoft.
tf.keras has received a number of updates, including experimental support for mixed precision on GPUs and Cloud TPUs. Mixed precision refers to the use of both 16-bit and 32-bit floating point types in models during training, which should make them run faster and use less memory.
Further updates have been made to tf.data, tf.debugging, tf.distribute and TensorRT. Among the bug fixes in TensorFlow 2.1.0 is that Keras
model.load_weights now accepts
skip_mismatch as an argument. This option was previously only available in external Keras.
See the full changelog for more details.