Show HN: SkewAdam – A tiered optimizer that cuts MoE state memory by 97%(github.com)2 points·by anonnumaan·7 giorni fa·0 commentsgithub.comShow HN: SkewAdam – A tiered optimizer that cuts MoE state memory by 97%https://github.com/nuemaan/skewadam2 commentsPost comment[+]anonnumaan·7 giorni fa[+]wayknow·7 giorni fa