SMARTS: A Small-to-large Model Coordinated Retrieval and Response Framework for Task-Oriented Dialogue

ICLR 2026 Conference Submission15118 Authors

Keywords: task-oriented, knowledge retrieval, model coordinated

Abstract: Task-Oriented Dialogue (TOD) systems are commonly used to assist users in achieving specific goals through human-computer interactions. Existing methods typically employ a single model during response generation to simultaneously learn response policy and perform fine-grained knowledge retrieval. However, these task-coupling methods often lead to suboptimal response generation, incorporating irrelevant knowledge and policy-violating styles, which makes optimizing TOD systems more difficult. To address this challenge, we propose a novel small-to-large model collaboration framework for task-oriented dialogue, named SMART. This framework utilizes a small language model to generate style-only responses without knowledge, while a large language model retrieves top-1 relevant knowledge from the knowledge base independently. The style-only responses are finally filled with top-1 relevant knowledge and form the completed responses to users. Finally, a re-thinker is designed to check the contextual relevance and knowledge accuracy of responses. Experiments on two public datasets demonstrate that SMART outperforms existing methods by an average of 5.79\% in Entity F1.

Primary Area: applications to computer vision, audio, language, and other modalities

Submission Number: 15118