Accelerating Distributed Training in Azure Machine Learning service using SR-IOV

Hi
we know

Single Root I/O Virtualization (https://www.lynx.com/embedded-systems-learning-center/what-is-sr-iov-and-why-is-it-important-for-embedded-devices) is the complex name for a technology beginning to find its way into embedded devices,so.
Today’s state-of-the-art deep learning models like BERT require distributed multi-machine training to reduce training time from weeks to days.(machine learning) Interconnect is one of the key components to reduce communication overhead and achieve good scaling efficiency in distributed multi-machine training.