Decide Then Retrieve: A Training-Free Framework with Uncertainty-Guided Triggering and Dual-Path Retrieval

TL;DR

Decide Then Retrieve (DTR) is a training-free framework that adaptively triggers retrieval based on uncertainty and employs dual-path retrieval to improve knowledge integration in large language models, reducing noise and enhancing performance.

Contribution

DTR introduces a novel, training-free approach that dynamically determines retrieval necessity and selection, improving upon traditional RAG methods with adaptive mechanisms.

Findings

Consistently improves EM and F1 scores across benchmarks.

Reduces unnecessary retrievals, increasing efficiency.

Effective across multiple model scales and retrievers.

Abstract

Retrieval-augmented generation (RAG) enhances large language models (LLMs) by incorporating external knowledge, but existing approaches indiscriminately trigger retrieval and rely on single-path evidence construction, often introducing noise and limiting performance gains. In this work, we propose Decide Then Retrieve (DTR), a training-free framework that adaptively determines when retrieval is necessary and how external information should be selected. DTR leverages generation uncertainty to guide retrieval triggering and introduces a dual-path retrieval mechanism with adaptive information selection to better handle sparse and ambiguous queries. Extensive experiments across five open-domain QA benchmarks, multiple model scales, and different retrievers demonstrate that DTR consistently improves EM and F1 over standard RAG and strong retrieval-enhanced baselines, while reducing unnecessary retrievals. The code and data used in this paper are available at https://github.com/ChenWangHKU/DTR.