Unlocking Implicit Experience: Synthesizing Tool-Use Trajectories from Text

TL;DR

This paper introduces GEM, a novel text-based data synthesis pipeline that extracts multi-turn tool-use trajectories from textual corpora to improve LLM tool utilization, achieving significant performance gains and efficiency.

Contribution

We propose GEM, a scalable method for generating multi-turn tool-use data from text, including a specialized Trajectory Synthesizer for efficient trajectory generation.

Findings

GEM-32B improves multi-turn benchmark performance by 16.5%.

Text corpora contain rich problem-solving trajectories useful for training.

The Trajectory Synthesizer matches pipeline quality with lower latency.

Abstract

Enabling Large Language Models (LLMs) to effectively utilize tools in multi-turn interactions is essential for building capable autonomous agents. However, acquiring diverse and realistic multi-turn tool-use data remains a significant challenge. In this work, we propose a novel text-based paradigm. We observe that textual corpora naturally contain rich, multi-step problem-solving experiences, which can serve as an untapped, scalable, and authentic data source for multi-turn tool-use tasks. Based on this insight, we introduce GEM, a data synthesis pipeline that enables the generation and extraction of multi-turn tool-use trajectories from text corpora through a four-stage process: relevance filtering, workflow & tool extraction, trajectory grounding, and complexity refinement. To reduce the computational cost, we further train a specialized Trajectory Synthesizer via supervised fine-tuning. This model distills the complex generation pipeline into an efficient, end-to-end trajectory generator. Experiments demonstrate that our GEM-32B achieve a 16.5% improvement on the BFCL V3 Multi-turn benchmark. Our models partially surpass the performance of models trained on {\tau} - bench (Airline and Retail) in-domain data, highlighting the superior generalization capability derived from our text-based synthesis paradigm. Notably, our Trajectory Synthesizer matches the quality of the full pipeline while significantly reducing inference latency and costs.