Model techniques map
Techniquesinference & servinginference kernel

implementation detail · filed under inference & serving

Distributed shared memory for cross-SM attention data exchange

Using distributed shared memory within thread-block clusters to exchange attention data across SMs.

source
1
models
2
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.

the second kernel utilizes distributed shared memory within thread-block clusters, enabling high-speed data exchange across SMs

usedunclearin DeepSeek-V4DeepSeek

Filed alongside

Other methods under inference & serving :: inference kernel.