Hacker News
new
past
comments
ask
show
jobs
points
by
regularfry
7 hours ago
|
comments
by
cubefox
6 hours ago
|
[-]
I assume that theoretically, 1-bit models could be most efficient because modern models switched from 32 bit to 16 bit to 8 bit per parameter (without quantization).
reply