Jekyll2026-10-19T03:40:11+00:00https://shangzhen-zhu.com/feed.xmlblankShangzhen Zhu, PhD student at the University of Illinois Urbana-Champaign. Research on how much numerical precision softmax attention in LLMs really needs, and on faster attention kernels. How Much Softmax Do You Need in LLMs?2026-10-18T00:00:00+00:002026-10-18T00:00:00+00:00https://shangzhen-zhu.com/blog/2026/how-much-softmaxThis post lives at /softmax/.]]>