Reddit discusses inference speedup tech and disk spillover

A Reddit user asks whether upcoming inference speedups (dSpark, dflash, MTP, QAT) will make model spillover to disk more tolerable. The post notes that spillover currently drops speed to unusable levels.