Early-Signal Detector

method · first seen 11 Nov 2026

parallel decoding for shared-context inference

Also: Parallel Prompting

Decoding multiple questions over one shared document in a single request so the shared KV cache is read once per step.

Mentions per week

0119 May11 Aug10 NovPapersCodeDiscussion

PapersCodeDiscussion

Earliest sources

Latest sources

Mentioned together with

KV cache reuseCUDA graph