Abstract: To address the challenges of low-cost deployment and reliability in real-world knowledge graph generation, this study proposes a few-shot knowledge graph question generation method through cooperation between large and small models. By constructing a chain of thought based on the OODA loop and designing an efficient deployment framework for smaller models, the method enables precise, controllable generation in resource-constrained operational environments at reduced costs. Experimental validation across multiple public datasets and military scenarios confirms the effectiveness of this approach, providing a feasible solution to the challenges of deployment difficulty and uncontrollable generation in resource-scarce scenarios.