Enhancing Reasoning Capabilities of LLMs via Principled Synthetic Logic Corpus

TL;DR

This paper introduces a principled synthetic logic training approach that significantly improves large language models' reasoning abilities across multiple benchmarks by using a carefully designed logical reasoning corpus.

Contribution

The paper presents a novel training method, ALT, and a synthetic corpus, FLD$_{ imes 2}$, to enhance LLM reasoning through high-quality, diverse logical samples based on symbolic logic principles.

Findings

Up to 30-point improvement on logical reasoning benchmarks

Up to 10-point gains on math and coding tasks

5-point increase on the BBH benchmark

Abstract

Large language models (LLMs) are capable of solving a wide range of tasks, yet they have struggled with reasoning. To address this, we propose $\textbf{Additional Logic Training (ALT)}$, which aims to enhance LLMs' reasoning capabilities by program-generated logical reasoning samples. We first establish principles for designing high-quality samples by integrating symbolic logic theory and previous empirical insights. Then, based on these principles, we construct a synthetic corpus named $\textbf{Formal Logic Deduction Diverse}$ ($\textbf{FLD}$$_{\times 2}$), comprising numerous samples of multi-step deduction with unknown facts, diverse reasoning rules, diverse linguistic expressions, and challenging distractors. Finally, we empirically show that ALT on FLD$_{\times2}$ substantially enhances the reasoning capabilities of state-of-the-art LLMs, including LLaMA-3.1-70B. Improvements include gains of up to 30 points on logical reasoning benchmarks, up to 10 points on math and coding benchmarks, and 5 points on the benchmark suite BBH.