通过异步调度加速大模型智能体自演化，提升效率三倍以上。
FlashEvolve: Accelerating Agent Self-Evolution with Asynchronous Stage Orchestration
- 采用异步工作流替代同步执行，实现各阶段并行重叠。
- 在GEPA任务上，本地部署提速3.5倍，API服务提速4.9倍。
- 支持可修复的过时产物，适合高并发智能体演化场景。
基于大语言模型的智能体自演化能有效改进非参数化成果，但其实际耗时仍是主要瓶颈。我们发现该瓶颈源于各阶段的同步执行以及每个大模型密集阶段内的负载不均。为此提出FlashEvolve框架，用异步工作者与队列取代同步执行，使不同阶段和步骤得以重叠运行。为应对异步带来的数据过时问题，FlashEvolve追踪产物版本，并对过时产物采用更新、丢弃或修补等策略。与强化学习中的权重空间过时不同，语言空间过时可被检视和修复：过时产物不仅是延迟的工作，更是大模型反思、修正并转化为有效演化信号的可读证据。FlashEvolve还通过推测性阶段完成和自适应工作流控制进一步提升吞吐量与词元效率。在GEPA任务上，相比同步GEPA，FlashEvolve在本地vLLM环境下将提案吞吐量提升3.5倍，在API服务中提升4.9倍。该设计同样适用于ACE和Meta-Harness。
原文摘要 · Abstract (English)
LLM-based evolution has emerged as a promising way to improve agents by refining non-parametric artifacts, but its wall-clock cost remains a major bottleneck. We identify that this cost comes from synchronized stage execution and imbalance inside each LLM-heavy stage. We present FlashEvolve, an efficient framework that replaces synchronized execution with asynchronous workers and queues, allowing different stages and steps to overlap. To handle data staleness introduced by asynchrony, FlashEvolve tracks artifact versions and applies different policies to update, discard, or patch stale artifacts. Unlike weight-space staleness in asynchronous RL, language-space staleness is inspectable and repairable: a stale artifact is not just delayed work, but readable evidence that the LLM can reflect on, revise, and turn into useful evolution signal. FlashEvolve further improves throughput and token efficiency with speculative stage completion and adaptive workflow control. On GEPA workloads, FlashEvolve improves proposal throughput by $3.5\times$ on local vLLM and $4.9\times$ on API serving over synchronous GEPA. The same design also applies to ACE and Meta-Harness.
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。