ToolGrad: Efficient Tool-use Dataset Generation with Textual "Gradients"
This addresses the problem of low-efficiency dataset generation for tool-use in LLMs, offering a more reliable and cost-effective solution, though it appears incremental as it builds on prior synthesis methods.
The paper tackles the inefficiency and annotation failures in generating tool-use datasets for LLMs by introducing ToolGrad, an agentic framework that inverts the paradigm to first construct valid tool-use chains using textual gradients and then synthesize queries, resulting in ToolGrad-5k with 100% pass rate and models outperforming baselines on OOD benchmarks.
Prior work synthesizes tool-use LLM datasets by first generating a user query, followed by complex tool-use annotations like 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-5k, a dataset generated with more complex tool use, lower cost, and 100% pass rate. Experiments show that models trained on ToolGrad-5k outperform those on expensive baseline datasets and proprietary LLMs, even on OOD benchmarks.