Y
HN Search
Hacker News Search
new
|
comments
|
top
|
jobs
khoss
searching Neon…
1.
▲
2.
▲
3.
▲
4.
▲
5.
▲
6.
▲
5 ms
·
1.
▲
by
khoss
1mo ago
I think the key application here is spec decoding. This solves the deprecation problem pretty nicely, and will likely have huge performance benefits
2.
▲
by
khoss
2y ago
AMD actually also does inference on PL, with reasonable commercial success, actually. Have a look at FINN.