Sinkhorn: Make LLMs even smaller through quantisation while maintaining accuracy

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