implementation detail · not yet filed
Overlap pinned-host embedding gather with decoder computation
- source
- 1
- model
- 1
- lab adopt it
- 1
- strongest
- used
How sources treat it
One count per evidence span, weakest treatment to strongest.
used 1
Documented in
Evidence
1 span quoted from the sources, strongest treatment first.
A dedicated CUDA stream overlaps the gather with the first decoder block.
usedunclearin Qwen3.8-Flash-NextSGLang