logoalt Hacker News

codexontoday at 7:17 AM0 repliesview on HN

It most likely will be quantized. A cerebras wafer only has 44gb ram, and linking them together vastly reduces the speedup.