LLM Agents Automate Quantum CIM Modeling Using All-Domestic Hardware

In a new arXiv preprint, Wang Rui and Lu Diannan present a system that empowers a practical quantum Coherent Ising Machine (CIM) with an all-domestic-core agentic large model. The CIM uses a femtosecond laser and is typically difficult for non-specialists to model, requiring tedious constraint weight tuning and methodology iteration. By leveraging LangGraph and LangChain frameworks, the researchers built an LLM-driven agentic system that can autonomously perform QUBO/Ising model calibration, iterate constraint weight decisions, and rapidly validate previously published schemes—all without human intervention.

Notably, the entire pipeline runs on domestic large models and domestically developed CIM hardware, marking a fully indigenous integration. The paper also uncovers an unexpected beneficial feedback loop: knowledge accumulated from agent-assisted quantum computing iterations reciprocally enhances the agent’s own problem-solving capability. This discovery suggests a new paradigm where quantum optimization and AI agents co-evolve, laying groundwork for more accessible, automated quantum computing while highlighting current challenges at the intersection of LLMs and quantum hardware.