Dynamic adaptive ensemble decoding under uncertainty for large language models

Weitao Ma

doi:10.1016/j.knosys.2026.116931

Dynamic adaptive ensemble decoding under uncertainty for large language models

2026-08-29

0

PRE

AI

DOI：10.1016/j.knosys.2026.116931

Original

Original request for help

Save

Abstract

En 中文

• We propose GUIDE, a two-stage framework for efficient and robust LLM ensembles. • We devise uncertainty-gated triggers and dual-granularity filtering to reduce computation and reject discordant signals. • Extensive experiments on diverse generative and multiple-choice benchmarks validate GUIDE’s effectiveness and robustness.

Keywords:

Large language models

Model ensemble

Uncertainty estimation

Dynamic adaptive ensemble decoding under uncertainty for large language models

Abstract

Journal

Organization

Cited Papers