Model techniques map
Techniquesinference & servingserving parallelism

specific method · filed under inference & serving

Prefill-decode disaggregation

Runs prefill and decoding on separate workers or pools; the evidence includes NIXL KV transfer as one configuration.

Also called Prefill–Decode (PD) disaggregation.

sources
4
models
3
labs adopt it
2
strongest
used

How sources treat it

One count per evidence span, weakest treatment to strongest.

used 4

Documented in

Further reading

Picked by hand, not extracted: where to read more, not evidence for anything on this page.

Evidence

4 spans quoted from the sources, strongest treatment first.

Prefill-decode disaggregation... we adopt it for Nemotron 3 Ultra

usedinference servingin Nemotron 3 UltraNVIDIA

Split one 8-GPU node into a prefill pool (GPUs 0-3) and a decode pool (GPUs 4-7), bridged by NIXL KV transfer.

usedinference servingin GLM-5.3-FlashZ.ai

GLM-5, therefore, leverages slime’s Prefill–Decode (PD) disaggregation

usedinference servingin GLM-5Z.ai

Prefill-decode disaggregation, which runs the two phases on separate workers with their own parallelism and scheduling, is the established response, and we adopt it for Nemotron 3 Ultra.

usedinference servingin Nemotron 3 UltraNVIDIA

Filed alongside

Other methods under inference & serving :: serving parallelism.