Hello! An article on TesorwFlow Mixed precision | TensorFlow Core says that mixed precision increases the speed of training while maintaining the quality of the model. Is there a reason not to use mixed precision during OpenNMT-tf training?
Thank you.