Omne-R1: Learning to Reason with Memory for Multi-hop Question Answering

TL;DR

Omne-R1 is a novel multi-hop question answering model that leverages advanced reasoning, multi-stage training, and domain-independent knowledge graphs to improve performance on complex questions across diverse domains.

Contribution

The paper introduces Omne-R1, a new reasoning approach with a multi-stage training process and domain-independent knowledge graphs for enhanced multi-hop QA.

Findings

Significant performance improvements on 3+ hop questions

Effective generalization across diverse knowledge domains

Successful auto-generation of QA pairs for training

Abstract

This paper introduces Omne-R1, a novel approach designed to enhance multi-hop question answering capabilities on schema-free knowledge graphs by integrating advanced reasoning models. Our method employs a multi-stage training workflow, including two reinforcement learning phases and one supervised fine-tuning phase. We address the challenge of limited suitable knowledge graphs and QA data by constructing domain-independent knowledge graphs and auto-generating QA pairs. Experimental results show significant improvements in answering multi-hop questions, with notable performance gains on more complex 3+ hop questions. Our proposed training framework demonstrates strong generalization abilities across diverse knowledge domains.