Y
HN Search
Hacker News Search
new
|
comments
|
top
|
jobs
arpperzhao
searching PlanetScale…
1.
▲
2.
▲
3.
▲
4.
▲
5.
▲
6.
▲
2 ms
·
1.
▲
by
arpperzhao
19d ago
Conjecture: Does GLM only handle the prefill stage, compress its output hidden vectors (trained?), and then send them to Qwen for decoding?