- Link: thesequence.substack.com/p/tensorflow-210-is-here
- Publication date: Sunday September 11th, 2022
TensorFlow and PyTorch have become the two most popular deep learning frameworks within the data science community. Their dominance in the market is strong. And it increases as both frameworks have been very fast to incorporate cutting-edge deep learning methods and ML engineering techniques that can help accelerate the implementation of deep learning solutions. As a result, each release of these frameworks drives a lot of attention across the data science space. Last week was TensorFlow’s turn with the release of its 2.10 version.
TensorFlow 2.10 has plenty of interesting features, but none were more notable than the improvements in Keras’ capabilities. The new version of TensorFlow includes improves Keras’ attention layers with features such as causal attention and implicit masking. The release also includes a new Keras optimizer API and changes in the Keras initializers to make them both deterministic and stale. Outside Keras, TensorFlow 2.10 includes hardware optimizations such as the support for the Compute Library for the Arm® Architecture (ACL) as well as a wider GPU support for Windows. Other notable features include a stable release of the popular TensorFlow Decision Forests (TF-DF), including JavaScript and Go inference APIs.
TensorFlow 2.10 shouldn’t be considered a major release but certainly incorporates many features that have been highly demanded by the developer community. An improved experience for Keras developers, wider hardware topology coverage, and improved libraries were at the center of this release. As usual, the frantic pace of innovation in TensorFlow and PyTorch does not disappoint.