New GPU format runs sparse neural networks 20% faster

Original: Sparser, Faster, Lighter Transformer Language Models

Source: /twell/ ↗