Y
Hacker News
new
|
ask
|
show
|
jobs
by
storus
246 days ago
M5 is supposed to support FP4 natively which would explain the speed up on Q4 quantized models (down from BF16).