specific method · filed under inference & serving
Multi-stage candidate filtering
Filters sampled candidate solutions through multiple stages before choosing competition submissions.
Also called multi-stage filtering pipeline.
- sources
- 2
- model
- 1
- lab adopt it
- 1
- strongest
- used
How sources treat it
One count per evidence span, weakest treatment to strongest.
used 2
Documented in
Evidence
2 spans quoted from the sources, strongest treatment first.
Specifically, we first sampled 500 candidate solutions for each problem, then applied a multi-stage filtering pipeline.
usedevaluation onlyin DeepSeek-V3.2DeepSeek
we first sampled 500 candidate solutions for each problem, then applied a multi-stage filtering pipeline
usedunclearin DeepSeek-V3.2-SpecialeDeepSeek
Filed alongside
Other methods under inference & serving :: decoding strategy.
Speculative decodingMulti-Token PredictionDSparkEAGLEDFlashPresence PenaltyBest-of-N scaffoldingNEXTN speculative decodingMulti-layer EAGLESpeculative samplingStandardized sampling configurationRecursive shared MTP-head draftingTask-specific sampling parametersChat Prefix CompletionConcurrency-aware draft-length tuningDistribution-matched draft-model fine-tuningEAGLE-3-style draft-model fine-tuningFused recurrent replay kernelGrammar-constrained decodingKV-cache sharing between drafter and targetLongest-trace selectionMTP-1 speculative decodingSame-checkpoint target and draft weightsThroughput-aware dynamic verification-length scheduling