Learning from Synthetic Data Improves Multi-hop Reasoning

TL;DR

This paper demonstrates that reinforcement learning fine-tuning on rule-generated synthetic data enhances large language models' multi-hop reasoning abilities, outperforming traditional data sources despite synthetic data containing fictional knowledge.

Contribution

It introduces a cost-effective method of using synthetic data for RL fine-tuning to improve reasoning in large language models, showing significant performance gains on real-world benchmarks.

Findings

Synthetic data improves real-world reasoning performance

LLMs learn to compose knowledge from synthetic data

Synthetic data is a scalable resource for reasoning skills

Abstract

Reinforcement Learning (RL) has been shown to significantly boost reasoning capabilities of large language models (LLMs) in math, coding, and multi-hop reasoning tasks. However, RL fine-tuning requires abundant high-quality verifiable data, often sourced from human annotations, generated from frontier LLMs, or scored by LLM-based verifiers. All three have considerable limitations: human-annotated datasets are small and expensive to curate, LLM-generated data is hallucination-prone and costly, and LLM-based verifiers are inaccurate and slow. In this work, we investigate a cheaper alternative: RL fine-tuning on rule-generated synthetic data for multi-hop reasoning tasks. We discover that LLMs fine-tuned on synthetic data perform significantly better on popular real-world question-answering benchmarks, despite the synthetic data containing only fictional knowledge. On stratifying performance by question difficulty, we find that synthetic data teaches LLMs to compose knowledge -- a fundamental and generalizable reasoning skill. Our work highlights rule-generated synthetic reasoning data as a free and scalable resource to improve LLM reasoning capabilities.