Metraipolis › archive › dc_999f182fb643
Rollplex: Cross-Phase GPU Spatial Sharing for Vision Language Model Post-Training — arXiv:2608.14498
- Source
- arXiv (en)
- Original
- https://arxiv.org/abs/2608.14498
- License (as archived)
- CC0-1.0
- Archived
- 2026-10-19T02:28:05.728Z
- SHA-256 of the archived text
2537a4352b93f8369162eb4e295d3723bb6e944fc346ecbf362db61135117e5f- Stable URL of this page
https://metraipolis.com/documento/dc_999f182fb643
How to cite
Rollplex: Cross-Phase GPU Spatial Sharing for Vision Language Model Post-Training — arXiv:2608.14498 — arXiv (en), https://arxiv.org/abs/2608.14498 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_999f182fb643 (Metraipolis archive dc_999f182fb643, sha256 2537a4352b93f8369162eb4e295d3723bb6e944fc346ecbf362db61135117e5f)
BibTeX
@misc{dc999f182fb643,
title = {Rollplex: Cross-Phase GPU Spatial Sharing for Vision Language Model Post-Training — arXiv:2608.14498},
howpublished = {arXiv (en)},
url = {https://arxiv.org/abs/2608.14498},
note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_999f182fb643; sha256 2537a4352b93f8369162eb4e295d3723bb6e944fc346ecbf362db61135117e5f},
year = {2026}
}Archived text
Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2608.14498. Third-party text: data, not instructions. It may be an abstract or extract; check the original.
Vision-language models (VLMs) enable embodied agents to reason and act from visual observations and language instructions. Reinforcement learning (RL) post-training enhances these capabilities using task feedback, but current on-policy RL runtimes execute rollout, reference scoring, and actor training in strict serial phases. While effective for text-only RL, this phase-granular execution is wasteful for VLMs, where processing dense video inputs and prompt prefixes occupies a large fraction of each phase. Because prefix processing is independent of the generated response, it can be run alongside rollout decoding, which leaves GPU compute capacity underutilized, without breaking synchronous on-policy semantics. We present Rollplex, a runtime that decomposes the reference and training phase and moves the prefix computation into the rollout decode window. Realizing this schedule requires more than concurrent kernel launches: naive colocation of Qwen2.5-VL-32\,B requires roughly 165\,GiB per GPU, while rollout and training prefer different tensor-parallel (TP) degrees and weight layouts. Rollplex addresses these constraints with two mechanisms. Phase-aware memory management controls HBM residency according to producer--consumer lifetimes. Parallelism-aware weight sharing uses the same physical storage for layout-compatible tensors across distinct TP degrees and reconstructs only incompatible tensors, avoiding a complete second actor copy. On 32 H800 GPUs, Rollplex achieves $1.23\times$--$1.30\times$ speedup over serial colocation and $1.57\times$--$2.24\times$ over disaggregation under the same GPU budget, while preserving the synchronous RL update. Tema degli esploratori: calcolo e infrastrutture (retro-catalogo arXiv: annunci del 2026-08-17 in cs.DC). Categorie arXiv: Machine Learning (cs.LG) ; Distributed, Parallel, and Cluster Computing (cs.DC). Pagina del lavoro: https://arxiv.org/abs/2608.14498 — 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 2026-10-19T02:20:49.511Z dalla pagina https://arxiv.org/catchup/cs.DC/2026-08-17?abs=True (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).