MidTool: Mid-training Data Synthesis for Agentic Tool Use
Fengqing Jiang · Yite Wang · Boyi Liu · Zhaoyang Wang · Canwen Xu · Zhewei Yao · Radha Poovendran · Yuxiong He
We present MidTool, an open pipeline for constructing large-scale mid-training corpora that endows base LLMs with general tool-use priors, yielding consistent downstream gains under both SFT and RL on agentic tool-use benchmarks.
摘要
Mid-training is increasingly recognized as a critical stage for shaping the capabilities of large language models. Recent work has shown that targeted mid-training can strengthen reasoning-intensive abilities such as math and science, and can also improve agentic capabilities in software-engineering settings. In this work, we study the parallel but less explored agentic capability: general tool use. We present MidTool, an open corpus construction pipeline for agentic tool-use mid-training that combines large-scale web, PDF, and code data with synthesized supervision from real-world tool APIs, MCP skills, and document-grounded workflows. MidTool is designed to teach models how to recognize tool affordances, ground arguments from context, compose tool call workflow, and recover from incomplete information. We mid-train Qwen3-4B-Base and Qwen3-8B-Base on MidTool-Mix, and then apply follow-up post-training with both supervised fine-tuning and reinforcement learning. Compared with baselines, MidTool-Mix consistently improves downstream performance under both SFT and RL on BFCL, $\tau^2$-Bench, and MCP Universe. These results suggest that general tool use, like other important LLM capabilities, benefits from dedicated mid-training rather than being left entirely to post-training.