Saved in:
Bibliographic Details
Main Authors: Cheng, Aijia, Wang, Kailong, Shi, Ling, Zhao, Yongxin
Format: Preprint
Published: 2026
Subjects:
Online Access:https://arxiv.org/abs/2604.20316
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866913053956636672
author Cheng, Aijia
Wang, Kailong
Shi, Ling
Zhao, Yongxin
author_facet Cheng, Aijia
Wang, Kailong
Shi, Ling
Zhao, Yongxin
contents Function calling empowers large language models (LLMs) to interface with external tools, yet existing RL-based approaches suffer from misalignment between reasoning processes and tool-call decisions. We propose R2IF, a reasoning-aware RL framework for interpretable function calling, adopting a composite reward integrating format/correctness constraints, Chain-of-Thought Effectiveness Reward (CER), and Specification-Modification-Value (SMV) reward, optimized via GRPO. Experiments on BFCL/ACEBench show R2IF outperforms baselines by up to 34.62% (Llama3.2-3B on BFCL) with positive Average CoT Effectiveness (0.05 for Llama3.2-3B), enhancing both function-calling accuracy and interpretability for reliable tool-augmented LLM deployment.
format Preprint
id arxiv_https___arxiv_org_abs_2604_20316
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle R2IF: Aligning Reasoning with Decisions via Composite Rewards for Interpretable LLM Function Calling
Cheng, Aijia
Wang, Kailong
Shi, Ling
Zhao, Yongxin
Machine Learning
Function calling empowers large language models (LLMs) to interface with external tools, yet existing RL-based approaches suffer from misalignment between reasoning processes and tool-call decisions. We propose R2IF, a reasoning-aware RL framework for interpretable function calling, adopting a composite reward integrating format/correctness constraints, Chain-of-Thought Effectiveness Reward (CER), and Specification-Modification-Value (SMV) reward, optimized via GRPO. Experiments on BFCL/ACEBench show R2IF outperforms baselines by up to 34.62% (Llama3.2-3B on BFCL) with positive Average CoT Effectiveness (0.05 for Llama3.2-3B), enhancing both function-calling accuracy and interpretability for reliable tool-augmented LLM deployment.
title R2IF: Aligning Reasoning with Decisions via Composite Rewards for Interpretable LLM Function Calling
topic Machine Learning
url https://arxiv.org/abs/2604.20316