---
title: "Batching the Reflective Optimization Loop: Parallel Proposals Make GEPA Faster and Better"
source: "arXiv AI/CL/LG（论文）"
source_tier: "t2"
url: "https://arxiv.org/abs/2610.36774v1"
permalink: "https://feed.astralor.com/items/itm_54f5fb3ecd47c2cd610efc6e"
published: "2026-10-27T07:41:40.000Z"
score: 54
channel: "ai"
tags: ["AI Agent","Batching","Reflective Optimization Loop","Parallel Proposals Make","GEPA Faster","Better Reflective","Reflective","GEPA"]
---

# Batching the Reflective Optimization Loop: Parallel Proposals Make GEPA Faster and Better

## 速览

Reflective text optimization makes agentic systems better, but the optimization loop is itself a resource-hungry workload whose sequential structure leaves parallel evaluation capacity idle. We extend the open-source GE…

## 关键点

- Reflective text optimization makes agentic systems better, but the optimization loop is itself a resource-hungry workload whose sequential structure leaves parallel evaluation capacity idle.
- We extend the open-source GEPA optimizer to propose and evaluate a batch of candidates on each optimization step instead of one candidate at a time.
- In our sweep on two tasks, most batched runs finished in half the wall-clock time or less, and the fastest in about a quarter to a third.

## 全文

Reflective text optimization makes agentic systems better, but the optimization loop is itself a resource-hungry workload whose sequential structure leaves parallel evaluation capacity idle. We extend the open-source GEPA optimizer to propose and evaluate a batch of candidates on each optimization step instead of one candidate at a time. In our sweep on two tasks, most batched runs finished in half the wall-clock time or less, and the fastest in about a quarter to a third. Batched settings also achieved higher held-out test scores, from 68.9% to 72.1% on LiveBench-Math (with $2\times2$) and from 49.0% to 60.0% on HoVer (with $8\times1$). A simple runtime model explains the measured speedups under a finite worker pool, and batched settings also spend their metric-call and dollar budgets more efficiently. We frame the axes along which a reflective optimization step can scale in parallel, a first look at allocating a fixed optimization budget across them.

---

引用前请核对原文;本文由星际流动整理:https://feed.astralor.com/items/itm_54f5fb3ecd47c2cd610efc6e