TInR: Exploring Tool-Internalized Reasoning in Large Language Models

TL;DR

This paper introduces TInR-U, a framework that internalizes external tool knowledge into large language models to improve reasoning efficiency and effectiveness, reducing reliance on external tool documentation.

Contribution

The paper proposes a novel three-phase training pipeline for internalizing tool knowledge into LLMs, enhancing reasoning capabilities without external tool reliance.

Findings

TInR-U outperforms existing methods in in-domain and out-of-domain tasks.

Internalized tool knowledge improves reasoning efficiency.

The framework demonstrates superior performance across diverse settings.

Abstract

Tool-Integrated Reasoning (TIR) has emerged as a promising direction by extending Large Language Models' (LLMs) capabilities with external tools during reasoning. Existing TIR methods typically rely on external tool documentation during reasoning. However, this leads to tool mastery difficulty, tool size constraints, and inference inefficiency. To mitigate these issues, we explore Tool-Internalized Reasoning (TInR), aiming at facilitating reasoning with tool knowledge internalized into LLMs. Achieving this goal presents notable requirements, including tool internalization and tool-reasoning coordination. To address them, we propose TInR-U, a tool-internalized reasoning framework for unified reasoning and tool usage. TInR-U is trained through a three-phase pipeline: 1) tool internalization with a bidirectional knowledge alignment strategy; 2) supervised fine-tuning warm-up using high-quality reasoning annotations, and 3) reinforcement learning with TInR-specific rewards. We comprehensively evaluate our method across in-domain and out-of-domain settings. Experiment results show that TInR-U achieves superior performance in both settings, highlighting its effectiveness and efficiency.