Sinkhorn: Make LLMs even smaller through quantisation while maintaining accuracy (github.com)

Media commentary: https://venturebeat.com/ai/huaweis-new-open-source-technique...