Hacker News new | ask | show | jobs
by codexon 7 days ago
It most likely will be quantized. A cerebras wafer only has 44gb ram, and linking them together vastly reduces the speedup.