FlashEvolve: Accelerating Agent Self-Evolution with Asynchronous Stage Orchestration
Fuente:
arXiv
Gespeichert in:
| Hauptverfasser: | , , , , , , , , , , |
|---|---|
| Format: | Preprint |
| Veröffentlicht: |
2026
|
| Schlagworte: | |
| Online-Zugang: | |
| Tags: |
Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
|
| _version_ | 1866915995915911168 |
|---|---|
| author | Hu, Zhengding Lu, Mingge Wang, Zhen Ruan, Jixuan Chen, Chang Pan, Zaifeng Guan, Yue Wang, Ruiyi Yu, Zhongkai Zhang, Chao Ding, Yufei |
| author_facet | Hu, Zhengding Lu, Mingge Wang, Zhen Ruan, Jixuan Chen, Chang Pan, Zaifeng Guan, Yue Wang, Ruiyi Yu, Zhongkai Zhang, Chao Ding, Yufei |
| contents | 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. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2605_08520 |
| institution | arXiv |
| publishDate | 2026 |
| record_format | arxiv |
| spellingShingle | FlashEvolve: Accelerating Agent Self-Evolution with Asynchronous Stage Orchestration Hu, Zhengding Lu, Mingge Wang, Zhen Ruan, Jixuan Chen, Chang Pan, Zaifeng Guan, Yue Wang, Ruiyi Yu, Zhongkai Zhang, Chao Ding, Yufei Machine Learning Distributed, Parallel, and Cluster Computing 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. |
| title | FlashEvolve: Accelerating Agent Self-Evolution with Asynchronous Stage Orchestration |
| topic | Machine Learning Distributed, Parallel, and Cluster Computing |
| url | https://arxiv.org/abs/2605.08520 |