TensorFlow provides a PluggableDevice mechanism that enables modular, plug-and-play integration of device-specific code.
AMD adopted PluggableDevice when developing the zentf plugin for inference on AMD EPYC™ CPUs. zentf adds custom kernel implementations and operations specific to AMD EPYC™ CPUs to TensorFlow via its kernel and op registration C APIs.
zentf is a supplemental package to be installed alongside standard TensorFlow packages with TensorFlow version 2.19.0. From a TensorFlow developer’s perspective, the zentf approach simplifies the process of leveraging ZenDNN optimizations.
This section provides instructions to setup zentf v5.1.