Machine-Learning TensorFlow 2: Faster training with the DirectML plugin

As early as October 2021, the developers from Redmond announced the general availability of the DirectML backend for machine learning framework TensorFlow in version 1.15. But as the Microsoft employees now share in a developer blog, the team has since worked hard to speed up machine learning training in TensorFlow 2. Now they have presented a preview release of the Direct ML PluggableDevice for TensorFlow 2 on PyPi, which is supposed to bring about exactly this acceleration.

TensorFlow DirectML is a fork maintained by Microsoft of the machine learning framework originally developed by Google. This package uses the DirectX API DirectML for hardware acceleration.



The PluggableDevice aims to enable users of the latest version of TensorFlow to accelerate model training on a wide range of DX12-capable GPUs (Image: Microsoft).

The TensorFlow DirectML plugin installs DirectML as the PluggableDevice backend on TensorFlow 2. DirectML is a machine learning library that enables model acceleration on any DirectX 12 compatible GPU. According to the statements in the blog entry, it can cause model acceleration on all GPUs compatible with DirectX 12. This should also work with cards from AMD, Intel and NVIDIA.

Using the TensorFlow DirectML plugin for TensorFlow 2.9 should be easy: The plugin package can be installed via PyPI without requiring changes to existing scripts. The plugin works with the TensorFlow core and should easily integrate with versions 2.9 and newer of the packages tensorflow or tensorflow-cpu integrate. According to the announcement, developers can register their existing GPU. If you want to try the plugin for yourself, you can download it from GitHub.

Simultaneously with the release, the development team also announces that the TensorFlow DirectML plugin is available as an open-source package. According to Microsoft, this first preview of the plug-in package should support most basic machine learning models. The team announces extended model support and performance optimizations for later versions. This news and all other details can be found in the entry in Microsoft’s developer blog.


(fms)

To home page

#MachineLearning #TensorFlow #Faster #training #DirectML #plugin

Leave a Comment

Your email address will not be published.