AbstractCurrent generative AI relies heavily on large-scale matrix calculations centered on Transformers, as well as the GPUs and data centers that process them at high speeds.Increasing model size ...