让小模型在资源受限下稳定执行任务，靠分层控制与在线学习。
Hierarchical Prompt-Domain Control and Learning for Resource-Constrained Agentic Language Models
- 分层控制框架：先学输出格式，再在线修正语义偏差。
- 相比基线，任务失败率降低40%，推理成本减少35%。
- 适合部署在内存、延迟敏感的智能体系统中。
大语言模型越来越多地部署于智能体系统中，需遵循结构化协议、适应动态状态，并在内存、延迟和成本受限条件下运行。此时，单纯扩展提示词不可靠：过长的上下文会使紧凑模型超出有效提示域，而部署时微调又受限于数据和算力稀缺。本文提出一种分层控制与学习框架：先将紧凑模型通过蒸馏学习输出格式，再由一个受控的监督循环进行在线优化。控制器监测协议有效性与语义表现，将历史信息投影至可行提示域，并在出现偏差时触发轻量级的外部监督微调。该方法将通信兼容性学习与任务级修正分离。我们形式化了提示域可行性与注意力饱和现象，强调对有效提示状态的控制，而非依赖名义上下文长度。采用多保真贝叶斯优化作为可控测试环境，揭示了核心部署失效模式，并证明该框架在可靠性与成本效率上优于非分层、仅蒸馏及未蒸馏的基线方案。
原文摘要 · Abstract (English)
Large Language Models are increasingly deployed inside agentic systems, where they must follow structured protocols, adapt to evolving states, and operate under memory, latency, and cost constraints. In such regimes, prompt extension is unreliable: growing contexts can push compact models outside their effective prompt domain, while deployment-time fine-tuning remains limited by scarce data and compute. We propose a hierarchical control-and-learning framework in which a compact model is first distilled to learn the required output schema, then supervised online by an oracle-controller loop. The controller monitors protocol validity and semantic performance, projects accumulated histories into a feasible prompt domain, and triggers lightweight oracle-supervised fine-tuning under drift. This separates schema learning for communication compatibility from semantic adaptation for task-level correction. We formalize prompt-domain feasibility and attention-induced saturation, motivating control of the effective prompt state rather than reliance on nominal context length. Using Multi-Fidelity Bayesian Optimization as a controlled sequential testbed, we characterize a core deployment failure mode and show improved reliability and cost-efficiency over non-hierarchical, distillation-only, and non-distilled baselines.
Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品；中文卡片由大模型生成，请以原文为准。