@FedorShabashev @lemire @richardstartin Did I only dream this or was there a recent paper around language models that bragged about number of petaflops sustained during training?