PyTorch 1.8 Released With AMD ROCm Binaries
PyTorch 1.8 brings improvements to distributed training with pipeline parallelism support that will bolster models not fitting on a single GPU device. ability to extend the PyTorch dispatcher for a new back-end in C++, and perhaps most exciting is AMD GPU binaries now being official/
Starting with PyTorch 1.8, AMD ROCm wheels are provided for an easy onboarding process of AMD GPU support for this machine learning library. It's great seeing the recent trend continue of more AI projects and other open-source initiatives finally making official their ROCm support.
PyTorch 1.8 details can be found on PyTorch.org.