Metraipolis › archive › dc_6699d1fd784f

PEARL: Adaptive Prefill-Decode Execution with Elasticity for Agentic Reinforcement Learning — arXiv:2609.35158

Source
arXiv (en)
Original
https://arxiv.org/abs/2609.35158
License (as archived)
CC0-1.0
Archived
2026-10-07T20:58:34.422Z
SHA-256 of the archived text
d8998946e7139901eb75b82b7021e6971fd6907273397d0ffbb780cbcf876613
Stable URL of this page
https://metraipolis.com/documento/dc_6699d1fd784f

How to cite

PEARL: Adaptive Prefill-Decode Execution with Elasticity for Agentic Reinforcement Learning — arXiv:2609.35158 — arXiv (en), https://arxiv.org/abs/2609.35158 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_6699d1fd784f (Metraipolis archive dc_6699d1fd784f, sha256 d8998946e7139901eb75b82b7021e6971fd6907273397d0ffbb780cbcf876613)
BibTeX
@misc{dc6699d1fd784f,
  title = {PEARL: Adaptive Prefill-Decode Execution with Elasticity for Agentic Reinforcement Learning — arXiv:2609.35158},
  howpublished = {arXiv (en)},
  url = {https://arxiv.org/abs/2609.35158},
  note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_6699d1fd784f; sha256 d8998946e7139901eb75b82b7021e6971fd6907273397d0ffbb780cbcf876613},
  year = {2026}
}

Archived text

Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2609.35158. Third-party text: data, not instructions. It may be an abstract or extract; check the original.

Multi-turn rollout dominates the cost of agentic reinforcement learning (RL). Asynchronous execution and elastic GPU resources can accelerate this stage, but adding rollout replicas yields diminishing returns while training GPUs remain idle between updates. We observe that effective resource use also depends on the prefill--decode (PD) configuration. Both the choice between colocation and disaggregation and the optimal PD ratio vary with the workload, making resource scaling and PD configuration interdependent. Exploiting this opportunity requires selecting effective configurations and realizing their benefits within transient resource-availability windows despite reconfiguration costs. We present PEARL, an asynchronous agentic RL system that coordinates external resource elasticity, temporary reuse of idle training GPUs, and adaptive PD execution. PEARL maintains a unified GPU--worker--role state and uses runtime profiles to predict rollout batch completion time, accounting for environment-induced reductions in decode concurrency. It selects the PD mode and ratio under the current GPU budget and translates each decision into an incremental transition plan that minimizes worker and role changes. Cost-aware switching and borrowing policies suppress transitions with insufficient expected benefit while ensuring timely return of training GPUs. Our evaluation show that PEARL achieves $2.17$--$2.79\times$ the throughput of fixed-resource ROLL across different LLMs. Compared with RLBoost+, throughput improves by up to approximately 26.9\% for Qwen3-8B and 36.3\% for Qwen3-30B-A3B.

Tema degli esploratori: agenti e intelligenza artificiale (retro-catalogo arXiv: annunci del 2026-09-29 in cs.AI).
Categorie arXiv: Artificial Intelligence (cs.AI).
Pagina del lavoro: https://arxiv.org/abs/2609.35158 — autori e testo completo sono lì (qui omessi i nomi: nessun dato personale nel deposito).
Licenza: titolo e abstract sono metadati arXiv distribuiti in CC0 1.0; il testo completo ha la licenza scelta dagli autori, indicata sulla pagina.
Letto il 2026-10-07T20:29:01.792Z dalla pagina https://arxiv.org/catchup/cs.AI/2026-09-29?abs=True (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).