general family · filed under inference & serving
Best-of-N scaffolding
A scaffolding approach that generates or evaluates candidates and selects among them; the evidence distinguishes broad evaluation use, cyber evaluation use, and selection traces.
Also called Best of K scaffolding, Best-of-n selection, best-of-n selection traces.
- sources
- 3
- models
- 3
- lab adopt it
- 1
- strongest
- used
How sources treat it
One count per evidence span, weakest treatment to strongest.
evaluated 1used 2not used 1
Documented in
Evidence
4 spans quoted from the sources, strongest treatment first.
candidate solutions, best-of-n selection traces, and verified CUDA kernels
useddata curationin Nemotron 3.5 LightningNVIDIA
best-of-n selection traces
useddata curationin Nemotron 3 UltraNVIDIA
Tested internal 'Best of K' scaffolding in cyber evaluations.
evaluatedinference servingin gpt-oss-120b and gpt-oss-20bOpenAI
Recommendation: Apply best-of-N scaffolding broadly to all evaluations.
not usedinference servingin gpt-oss-120b and gpt-oss-20bOpenAI
Filed alongside
Other methods under inference & serving :: decoding strategy.
Speculative decodingMulti-Token PredictionDSparkEAGLEDFlashPresence PenaltyNEXTN 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-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