提示码本：面向语言模型指令精炼的离散组合优化
Prompt Codebooks: Discrete Compositional Optimization for Language Model Instruction Refinement
AI总结：
提出Prompt Codebooks (PCO)框架，将自动提示优化重构为离散组合学习，通过可重用的自然语言本能单元实现实例级路由和结构化反馈，在多个基准上提升性能并压缩提示长度。
摘要：
自动提示优化（APO）显著提升了基于LLM的智能体工作流。然而，现有方法将每个任务的提示视为一个整体、实例无关的字符串，通过全局编辑进行优化，导致更新脆弱且无法复用学到的子行为。我们提出提示码本（PCO），一种新颖的组合式提示优化框架，将APO重构为在有限自然语言本能（原子、可重用的指令单元）词汇表上的离散学习。PCO将提示构建知识组织在离散码本中，通过基于LLM的编码器将每个输入路由到少量条目；生成器将它们组合成冻结目标模型的提示；评论器输出结构化判决，通过归因分解为每个变量的文本梯度，在语言值极小极大目标下联合训练编码器、生成器和码本。得到的路由是实例级的：同一任务的不同输入接收不同的本能组合，这种机制在实例无关方法下结构上无法表达。在Qwen3-8B和LLaMA-3.1-8B上的六个基准测试中，PCO相比零样本提升高达+30.36分，在HotpotQA上超越最强先前基线（GEPA）达+3.34分，总体平均提升+1.11分，并且仅使用K=16个本能即可将部署提示长度相比MIPROv2压缩最多14.1倍，相比GEPA压缩3.0倍。
英文摘要：
Automatic prompt optimization (APO) has driven significant gains in LLM-based agentic workflows. However, most existing methods treat each task's prompt as a monolithic, instance-blind string optimized through global edits, producing brittle updates and preventing the reuse of learned sub-behaviors. We propose Prompt Codebook Optimization (PCO), a novel compositional prompt optimization framework that recasts APO as discrete learning over a finite vocabulary of natural-language instincts--atomic, reusable instruction units. PCO organizes prompt-construction knowledge in a discrete codebook and routes each input to a small subset of entries via an LLM-based encoder; a generator composes them into a prompt for the executor; a critic emits a structured verdict that decomposes by attribution into per-variable textual gradients, jointly training the encoder, generator, critic, and codebook under a language-valued min-max objective. The resulting routing is per-instance: different inputs in the same task receive different instinct compositions. Across six benchmarks, PCO improves aggregate performance over zero-shot by +13.50 points on Qwen3-8B and +11.80 points on LLaMA-3.1-8B. Crucially, PCO surpasses GEPA on HotpotQA by +6.34 points (Qwen3-8B) and +4.27 points (LLaMA-3.1-8B), while simultaneously reducing deployed prompt length by up to 14.1$\times$ vs. MIPROv2 and 3.0$\times$ vs. GEPA.
机构、总结、中文摘要均由AI分析而来，可能存在错误，请以论文正文为准。
