I was pleasantly surprised to find out that Pytorch has a Vulkan backend. Unfortunately, it only does inferencing at fp32. Is this backend still being developed, and will it support training and half/mixed precision in the near(or distant) future?
1 Like