Self-Guided Function Calling in Large Language Models via Stepwise Experience Recall

TL;DR

This paper introduces SEER, a self-guided, stepwise experience recall method that incrementally improves large language models' ability to perform multi-step tool usage by continually expanding its experience pool.

Contribution

The paper presents SEER, a novel approach that dynamically updates an experience pool for LLMs, reducing manual effort and enhancing multi-step tool interaction performance.

Findings

SEER improves accuracy by 6.1% on easy questions and 4.7% on hard questions.

SEER achieves 7.44% and 23.38% accuracy gains on real-world domains with different models.

The method reduces reliance on manual demonstrations and curates experience dynamically.

Abstract

Function calling enables large language models (LLMs) to interact with external systems by leveraging tools and APIs. When faced with multi-step tool usage, LLMs still struggle with tool selection, parameter generation, and tool-chain planning. Existing methods typically rely on manually designing task-specific demonstrations, or retrieving from a curated library. These approaches demand substantial expert effort and prompt engineering becomes increasingly complex and inefficient as tool diversity and task difficulty scale. To address these challenges, we propose a self-guided method, Stepwise Experience Recall (SEER), which performs fine-grained, stepwise retrieval from a continually updated experience pool. Instead of relying on static or manually curated library, SEER incrementally augments the experience pool with past successful trajectories, enabling continuous expansion of the pool and improved model performance over time. Evaluated on the ToolQA benchmark, SEER achieves an average improvement of 6.1% on easy and 4.7% on hard questions. We further test SEER on $\tau$-bench, which includes two real-world domains. Powered by Qwen2.5-7B and Qwen2.5-72B models, SEER demonstrates substantial accuracy gains of 7.44% and 23.38%, respectively.