FP8 versus INT8 for efficient deep learning inference

van Baalen, Mart, Kuzmin, Andrey, Nair, Suparna S · arXiv · 2023