Sinkhorn: Make LLMs even smaller through quantisation while maintaining accuracy | Not Hacker News!