Saved in:
Bibliographic Details
Main Authors: Zhou, Zhongyi, Uehara, Kohei, Zhang, Haoyu, Zhou, Jingtao, Gu, Lin, Du, Ruofei, Xu, Zheng, Harada, Tatsuya
Format: Preprint
Published: 2025
Subjects:
Online Access:https://arxiv.org/abs/2508.04086
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866913078540500992
author Zhou, Zhongyi
Uehara, Kohei
Zhang, Haoyu
Zhou, Jingtao
Gu, Lin
Du, Ruofei
Xu, Zheng
Harada, Tatsuya
author_facet Zhou, Zhongyi
Uehara, Kohei
Zhang, Haoyu
Zhou, Jingtao
Gu, Lin
Du, Ruofei
Xu, Zheng
Harada, Tatsuya
contents Prior work synthesizes tool-use LLM datasets by first generating a user query, followed by complex tool-use annotations like depth-first search (DFS). This leads to inevitable annotation failures and low efficiency in data generation. We introduce ToolGrad, an agentic framework that inverts this paradigm. ToolGrad first constructs valid tool-use chains through an iterative process guided by textual "gradients", and then synthesizes corresponding user queries. This "answer-first" approach led to ToolGrad-500, a dataset generated with more complex tool use, lower cost, and almost 100% pass rate. Experiments show that ToolGrad models outperform those trained on expensive baseline datasets and proprietary LLMs. The ToolGrad source code, dataset, and models are available at https://github.com/zhongyi-zhou/toolgrad.
format Preprint
id arxiv_https___arxiv_org_abs_2508_04086
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle ToolGrad: Efficient Tool-use Dataset Generation with Textual "Gradients"
Zhou, Zhongyi
Uehara, Kohei
Zhang, Haoyu
Zhou, Jingtao
Gu, Lin
Du, Ruofei
Xu, Zheng
Harada, Tatsuya
Computation and Language
Prior work synthesizes tool-use LLM datasets by first generating a user query, followed by complex tool-use annotations like depth-first search (DFS). This leads to inevitable annotation failures and low efficiency in data generation. We introduce ToolGrad, an agentic framework that inverts this paradigm. ToolGrad first constructs valid tool-use chains through an iterative process guided by textual "gradients", and then synthesizes corresponding user queries. This "answer-first" approach led to ToolGrad-500, a dataset generated with more complex tool use, lower cost, and almost 100% pass rate. Experiments show that ToolGrad models outperform those trained on expensive baseline datasets and proprietary LLMs. The ToolGrad source code, dataset, and models are available at https://github.com/zhongyi-zhou/toolgrad.
title ToolGrad: Efficient Tool-use Dataset Generation with Textual "Gradients"
topic Computation and Language
url https://arxiv.org/abs/2508.04086