Nvidia clarifies Megatron-Turing scale claim

The importance of Megatron-Turing 530B is that it is the largest natural language processing model that has been "trained to convergence."