Hacker News
new
past
comments
ask
show
jobs
points
by
_aavaa_
13 hours ago
|
comments
by
natrys
13 hours ago
|
[-]
I think we don't have a good draft model for better speculative decoding yet (e.g. DFlash 2). Once we do, it will be faster.
reply
by
_aavaa_
12 hours ago
|
parent
|
[-]
It very well could be faster, but right now it isn’t.
reply