ME AI MESSAGE: PyTorch's official announcement states that at PyTorchCon North America 2026, AMD's Liz (@lizli202503) and Shekhar (@indianspeedster) will share their work on implementing scalable MXFP8 pretraining on the MI355X GPU. They will introduce optimizations of key MXFP8 operators in TorchAO, and how these kernels are integrated with the specified link (https://t.co/rwbt3VY4XO). The original article did not disclose specific optimization metrics, performance data, release date, or more technical details. (Source: ME)