The ZenDNN plug-in for PyTorch (zentorch) enables inference optimizations for
deep learning workloads on AMD EPYC™ CPUs. It uses the ZenDNN library, which contains
deep learning operators tailored for high performance on AMD EPYC™ CPUs. The zentorch
extension to PyTorch has been developed to leverage the torch.compile graph compilation flow, and all optimizations can be enabled
by a call to torch.compile with zentorch as the backend.
Multiple passes of graph level optimizations run on the torch.fx graph and provide further performance acceleration.