用图结构反向传播目标，让多个大模型协作更准更稳。
Graph-based Target Back-Propagation for Context Adaptation in Multi-LLM Agentic Systems
- 构建任务图为有向无环图，反向传播目标输出指导提示优化
- 在三个基准上超越强基线，计算开销相近
- 适合多模型协作系统，尤其需要精准反馈分配的场景
上下文自适应通过迭代修改可调提示来自动化大模型系统的提示工程，无需调整模型权重。将这一范式扩展到多大模型智能体系统至关重要：现有方法存在信用分配不准确且缺乏收敛保证的问题。本文提出图结构目标反向传播（GTBP），一种针对以有向无环图建模的智能体工作流的上下文自适应框架。GTBP通过工作流图将局部目标输出反向传播，并利用目标与实际输出的差异来引导分阶段提示更新机制。理论上，我们证明了GTBP的分阶段提示更新在迭代中趋于稳定，且足够强大的大模型优化器能降低整体目标函数。实验上，GTBP在三个基准测试中持续优于强基线，同时保持相当的计算成本。
原文摘要 · Abstract (English)
Context adaptation automates prompt engineering in LLM-based systems by iteratively revising tunable prompts from task feedback, without modifying model weights. Extending this paradigm to multi-LLM agentic systems is crucial: existing methods suffer from inaccurate credit assignment and lack convergence guarantees. We propose \textbf{G}raph-based \textbf{T}arget \textbf{B}ack-\textbf{P}ropagation (GTBP), a context adaptation framework for agentic workflows modeled as directed acyclic graphs. GTBP propagates local target outputs backward through the workflow graph and uses target--output discrepancies to guide a stage-wise prompt update mechanism. Theoretically, we show that GTBP's stage-wise prompt updates become stable over iterations, and that a sufficiently capable LLM optimizer can decrease the overall objective. Empirically, GTBP consistently outperforms strong baselines across three benchmarks while maintaining comparable computational cost.
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。