用扩散模型迭代优化提示词，不改模型也能提升大模型表现
Prompt Optimization Via Diffusion Language Models
- 通过掩码去噪逐步优化提示词，基于用户交互痕迹实现细粒度调整
- 在τ-bench、SST-2等数据集上，优化后提示词使GPT-4o-mini性能显著提升
- 适度扩散步数平衡效果与稳定性，适用于各类冻结模型的提示优化
我们提出一种基于扩散的提示词优化框架，利用扩散语言模型（DLMs）通过掩码去噪机制，迭代优化系统提示词。该方法基于用户查询、模型响应及可选反馈等交互痕迹进行条件建模，无需梯度访问或修改下游语言模型，即可实现细粒度的提示词更新。在多个基准测试（如τ-bench、SST-2、SST-5）中，经DLM优化的提示词持续提升了冻结目标语言模型（如GPT-4o-mini）的表现。进一步实验表明，适中的扩散步数能实现优化质量与稳定性的最佳平衡。结果表明，基于扩散的提示优化是一种通用、模型无关且可扩展的方法，可通过迭代优化有效提升大模型性能。
原文摘要 · Abstract (English)
We propose a diffusion-based framework for prompt optimization that leverages Diffusion Language Models (DLMs) to iteratively refine system prompts through masked denoising. By conditioning on interaction traces, including user queries, model responses, and optional feedback, our method enables flexible, span-level prompt updates without requiring gradient access or modifying the downstream language model. Across diverse benchmarks (e.g., $τ$-bench, SST-2, SST-5), DLM-optimized prompts consistently improve the performance of a frozen target LLM (e.g., GPT-4o-mini). We further show that moderate diffusion step counts provide the best balance between refinement quality and stability. These results highlight diffusion-based prompt optimization as a general, model-agnostic, and scalable approach for enhancing LLM performance through iterative prompt refinement.
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。