candidate solutions, best-of-n selection traces, and verified CUDA kernels
general family · filed under inference & serving
Best-of-N selection
Generates multiple candidates and selects the preferred candidate, with the evidence also describing its use as evaluation scaffolding and as selection traces.
The sources on this page only mention it in passing. The line above is the extraction's one-line gloss. See decoding strategy for the methods filed beside it.
Also called Best of K scaffolding, best-of-N scaffolding, best-of-n selection traces.
Documented in
Evidence
4 quotes from 3 sources, one card per model, each opening with its strongest.
best-of-n selection traces
gpt-oss-120b
OpenAIRecommendation: Apply best-of-N scaffolding broadly to all evaluations.
1 more from this model
Tested internal 'Best of K' scaffolding in cyber evaluations.
Further reading
Picked by hand, not extracted: where to read more, not evidence for anything on this page.
- Training Verifiers to Solve Math Word Problems (Cobbe et al., 2021) paper arxiv.orgcanonical best-of-n selection with verifiers