KnowPrefix-Tuning: A Two-Stage Prefix-Tuning Framework for Knowledge-Grounded Dialogue Generation

TL;DR

KnowPrefix-Tuning introduces a two-stage prefix-tuning framework that leverages pre-trained language models' inherent knowledge for knowledge-grounded dialogue generation, eliminating the need for resource-intensive retrieval systems.

Contribution

It proposes a novel two-stage prefix-tuning method with an interactive re-parameterization mechanism, enabling efficient knowledge-grounded dialogue generation without retrieval.

Findings

Outperforms fine-tuning and lightweight tuning baselines

Achieves comparable performance to retrieval-based methods

Runs 3 times faster during inference

Abstract

Existing knowledge-grounded conversation systems generate responses typically in a retrieve-then-generate manner. They require a large knowledge base and a strong knowledge retrieval component, which is time- and resource-consuming. In this paper, we address the challenge by leveraging the inherent knowledge encoded in the pre-trained language models (PLMs). We propose Knowledgeable Prefix Tuning (KnowPrefix-Tuning), a two-stage tuning framework, bypassing the retrieval process in a knowledge-grounded conversation system by injecting prior knowledge into the lightweight knowledge prefix. The knowledge prefix is a sequence of continuous knowledge-specific vectors that can be learned during training. In addition, we propose a novel interactive re-parameterization mechanism that allows the prefix to interact fully with the PLM during the optimization of response generation. Experimental results demonstrate that KnowPrefix-Tuning outperforms fine-tuning and other lightweight tuning approaches, and performs comparably with strong retrieval-based baselines while being $3\times$ faster during inference.