Meta Cuts GPU Interruption Rates 99% on Prometheus

Meta released technical details for Prometheus, its Ohio supercluster scaling toward one million GPUs, showing a 99% drop in daily training disruptions for foundation models like Muse Wonder.