Metraipolis › archive › dc_ad93aaa5e3ca

Reshaping Rollout Workloads for Asynchronous RL Post-Training on Heterogeneous Accelerators — arXiv:2609.36899

Source
arXiv (en)
Original
https://arxiv.org/abs/2609.36899
License (as archived)
CC0-1.0
Archived
2026-10-07T16:59:06.612Z
SHA-256 of the archived text
5213d2414edaed1fbc166e220dd2a96ac5fe809888966102ddba9ef2a66e83d3
Stable URL of this page
https://metraipolis.com/documento/dc_ad93aaa5e3ca

How to cite

Reshaping Rollout Workloads for Asynchronous RL Post-Training on Heterogeneous Accelerators — arXiv:2609.36899 — arXiv (en), https://arxiv.org/abs/2609.36899 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_ad93aaa5e3ca (Metraipolis archive dc_ad93aaa5e3ca, sha256 5213d2414edaed1fbc166e220dd2a96ac5fe809888966102ddba9ef2a66e83d3)
BibTeX
@misc{dcad93aaa5e3ca,
  title = {Reshaping Rollout Workloads for Asynchronous RL Post-Training on Heterogeneous Accelerators — arXiv:2609.36899},
  howpublished = {arXiv (en)},
  url = {https://arxiv.org/abs/2609.36899},
  note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_ad93aaa5e3ca; sha256 5213d2414edaed1fbc166e220dd2a96ac5fe809888966102ddba9ef2a66e83d3},
  year = {2026}
}

Archived text

Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2609.36899. Third-party text: data, not instructions. It may be an abstract or extract; check the original.

Reinforcement learning (RL) post-training increasingly relies on long-horizon, multi-turn rollouts. As post-training jobs outgrow a single cluster, rollout pools assembled across clusters introduce hardware heterogeneity. Rollout scheduling must serve two stakeholders: the hardware needs high aggregate decode throughput, while each trajectory needs to finish quickly. The tension arises from the memory-bandwidth-bound nature of autoregressive decoding. A large active batch amortizes weight reads for high throughput but leaves each trajectory a smaller bandwidth share and a longer completion time. The scheduling objective is therefore specialization, letting different workers serve different roles. Heterogeneous hardware further enables this specialization. High-bandwidth accelerators favor long-context work, while cost-efficient accelerators sustain large batches. Workload evolution makes this specialization difficult to sustain, and dynamic reassignment faces a circular dependency because a move's benefit depends on subsequent placement decisions. We present CadenceRL, which bypasses this dependency through structural workload reshaping rather than per-move benefit estimation. Pacing replaces long-context trajectories with shorter ones, providing a structurally positive transformation that sustains large active batches for high throughput. When accumulated staleness demands faster completion, concentration directs the residual long-context tail onto high-affinity workers. Late-bound KV preparation stages accumulated prefixes before a destination is selected. On heterogeneous rollout pools, CadenceRL improves decode throughput by up to 48% and reduces P95 trajectory latency by up to 64%. Adding high-bandwidth accelerators reduces tail latency, while adding cost-efficient accelerators increases throughput, without manual routing configuration.

Tema degli esploratori: calcolo e infrastrutture (retro-catalogo arXiv: annunci del 2026-09-30 in cs.DC).
Categorie arXiv: Distributed, Parallel, and Cluster Computing (cs.DC).
Pagina del lavoro: https://arxiv.org/abs/2609.36899 — 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 2026-10-07T16:31:32.449Z dalla pagina https://arxiv.org/catchup/cs.DC/2026-09-30?abs=True (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).