摘要：小型语言模型（SLM）部署成本低，并可在私有基础设施上运行，但基础模型在多轮工具调用方面通常能力不足；而对其进行微调则需要按 API 划分的数据，这类数据很少存在。现有合成方法对于大规模微调而言成本过高，因为它们往往需要为不同领域构建模拟运行环境，并且在每个生成的对话轮次中进行多次 LLM 调用。我们提出了一种全自动、轻量级的合成框架：该框架将每个 API 建模为有限状态机，将系统表示为抽象状态，这些状态决定何时可以调用各个工具，从而生成状态合法的工具序列；随后只需一次 LLM 调用，即可将这些序列转换为完整样本。我们并非优化多样性，而是针对轮次数、工具序列和任务复杂度设定目标分布。我们通过在生成的轨迹上微调 SLM 来衡量数据质量，结果表明，与未经变异的基线相比，基于 FSM 的生成方法显著提升了下游准确率；相较于现有工作，该方法以少 3.6-6.6$\times$ 的 token，实现了 70.7% 的完整准确率，而对比方法分别为 63.4% 和 53.7%。

Abstract:Small language models are inexpensive to serve and can run on private infrastructure, but base models are often not good enough at multi-turn tool calling, and fine-tuning them needs per-API data that rarely exists. Existing synthesis methods are too expensive for high-scale fine-tuning, as they often require mock operational environments for different domains and multiple LLM calls per generated conversation turn. We introduce a fully automated, lightweight synthesis framework that models each API as a finite-state machine, representing the system as abstract states that determine when each tool may be called, producing state-valid sequences of tools; sequences are translated into complete examples with a single LLM call. Rather than optimize diversity, we set a target distribution over the number of turns, the tool sequence and task complexity. We measure data quality by fine-tuning SLMs on generated trajectories, showing that our FSM-based generation significantly improves downstream accuracy over an unmutated baseline and, against existing works, reaches 70.7% full accuracy over 63.4% and 53.7% with 3.6-6.6$\times$ fewer tokens.