arXiv cs.CL· Zack Yu, Nicholas Lee, Minjae Lee, Lakshya Agarwal, Luca Manolache, Wonjun Kang, Matei Zaharia, Michael W. Mahoney, Kurt Keutzer, Sehoon Kim, Amir Gholami·· 13 小时前AI 评分34
GEPA-Curator：面向自我改进智能体的精选学习框架
GEPA-Curator: Curated Learning for Self Improving Agents
AI 导读
GEPA-Curator 通过任务筛选和成功轨迹检索，为提示词优化器挑选更可靠的学习信号，在 GoBrowse、ALFWorld、AppWorld 和 τ-bench 上使用 Gemini 3.1 Pro 与 Claude Opus 5 作为优化器模型，平均性能较 GEPA 提升最高 20.7%、较 ACE 提升 21.2%。
正文 · 原文
Authors:Zack Yu, Nicholas Lee, Minjae Lee, Lakshya Agarwal, Luca Manolache, Wonjun Kang, Matei Zaharia, Michael W. Mahoney, Kurt Keutzer, Sehoon Kim, Amir Gholami
Abstract: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.
| Subjects: | Computation and Language (cs.CL) |
| Cite as: | arXiv:2610.37651 [cs.CL] |
| (or arXiv:2610.37651v1 [cs.CL] for this version) | |
| https://doi.org/10.48550/arXiv.2610.37651 arXiv-issued DOI via DataCite (pending registration) |
Submission history
From: Nicholas Lee [view email]
[v1]
Tue, 27 Oct 2026 17:58:11 UTC (307 KB)
来源：arXiv cs.CL · arxiv.org