[论文] GEPA-Curator: Curated Learning for Self Improving Agents

论文概要

研究领域: NLP / Agent
作者: Zack Yu, Nicholas Lee, Minjae Lee, Lakshya Agarwal, Luca Manolache, Wonjun Kang, Matei Zaharia, Michael W. Mahoney, Kurt Keutzer, Sehoon Kim, Amir Gholami
发布时间: 2026-10-27
arXiv: 2610.37651

中文摘要

LLM越来越多地用于需要与复杂环境持续交互的多轮智能体工作负载。现有反思式提示优化器（如GEPA）从失败的rollout中学习，但智能体的失败也可能来自提示弱点、工具链限制、随机执行或嘈杂评判器。本文认为这类失败可能不提供有用的学习信号，甚至有害，因此策划哪些任务提供给提示优化器至关重要。GEPA-Curator包含两个组件：任务策划（过滤可能提供嘈杂或不可靠学习信号的任务）和先验成功检索（选择性检索过去的成功轨迹作为额外上下文提供给优化器）。在GoBrowse、ALFWorld、AppWorld和τ-bench四个基准上，使用Gemini 3.1 Pro和Claude Opus 5作为优化器，GEPA-Curator匹配或超过两个基线的平均性能，最高比GEPA提升20.7%，比ACE提升21.2%。

原文摘要

Large Language Models (LLMs) are increasingly used in multi-turn agentic workloads that require sustained interaction with complex environments. Existing reflective prompt optimizers such as GEPA learn from failed rollouts, but agentic failures can also come from prompt weaknesses, harness limitations, stochastic execution, or noisy judges. We argue that such failures may not provide useful learning signal, and may even be harmful; and thus it is essential to curate which tasks are provided to the prompt optimizer for better performance. We introduce GEPA-Curator, a framework that augments the prompt optimizer to select which agent experiences should shape prompt optimization. In particular, GEPA-Curator consists of two components: task curation, which filters out tasks that are likely to provide noisy or unreliable learning signals, and prior success retrieval, which selectively retrieves past successful trajectories for a task and provides them as additional context to the prompt optimizer. We evaluate GEPA-Curator on GoBrowse, ALFWorld, AppWorld, and \(\tau\)-bench, using both Gemini 3.1 Pro and Claude Opus 5 as optimizer models. GEPA-Curator matches or exceeds the mean performance of both baselines, with gains of up to 20.7% over GEPA and 21.2% over ACE.

自动采集于 2026-10-29

#论文 #arXiv #AI #小凯