implementation detail · filed under inference & serving
Fused recurrent replay kernel
Combines replayed tokens, the bonus token, and the next draft window in one recurrent fused kernel.
Also called single fused kernel.
- source
- 1
- model
- 1
- lab adopt it
- 1
- strongest
- core
How sources treat it
One count per evidence span, weakest treatment to strongest.
core 1
Documented in
Evidence
1 span quoted from the sources, strongest treatment first.
The replayed tokens, the bonus token, and the next draft window share one recurrent loop inside a single fused kernel
coreinference servingin Kimi K3
Filed alongside
Other methods under inference & serving :: decoding strategy.
Speculative decodingMulti-Token PredictionDSparkEAGLEDFlashPresence PenaltyBest-of-N scaffoldingNEXTN speculative decodingMulti-layer EAGLESpeculative samplingStandardized sampling configurationMulti-stage candidate filteringRecursive shared MTP-head draftingTask-specific sampling parametersChat Prefix CompletionConcurrency-aware draft-length tuningDistribution-matched draft-model fine-tuningEAGLE-3-style draft-model fine-tuningGrammar-constrained decodingKV-cache sharing between drafter and targetLongest-trace selectionMTP-1 speculative decodingSame-checkpoint target and draft weightsThroughput-aware dynamic verification-length scheduling