{
  "schema_version": "1.1",
  "id": "archive:https://arxiv.org/abs/2608.14498v1",
  "slug": "2608-14498v1-1a3lotd",
  "url": "https://feed7.dev/p/2608-14498v1-1a3lotd",
  "title": "Rollplex: Cross-Phase GPU Spatial Sharing for Vision Language Model Post-Training",
  "why_included": "Rollplex overlaps VLM prefix work with rollout decoding and shares weights across training phases. On 32 H800 GPUs, it reports higher throughput without changing synchronous on-policy updates.",
  "summary": "Rollplex moves response-independent prefix computation into the rollout decode window. Phase-aware HBM management and parallelism-aware weight sharing address a naive Qwen2.5-VL-32B footprint of roughly **165 GiB per GPU**.",
  "practical_implication": "Teams post-training vision-language agents can inspect phase boundaries for independent prefix work and share compatible tensor storage across differing parallel layouts. The reported schedule preserves synchronous on-policy updates under the same GPU budget.",
  "agent_context": "Rollplex moves response-independent prefix computation into the rollout decode window. Phase-aware HBM management and parallelism-aware weight sharing address a naive Qwen2.5-VL-32B footprint of roughly **165 GiB per GPU**.\n\nTeams post-training vision-language agents can inspect phase boundaries for independent prefix work and share compatible tensor storage across differing parallel layouts. The reported schedule preserves synchronous on-policy updates under the same GPU budget.\n\nOn **32 H800 GPUs**, Rollplex reports **1.23×–1.30×** speedup over serial colocation and **1.57×–2.24×** over disaggregation. The material provides results for this hardware and workload setup, not broader cost or scaling evidence.",
  "source": {
    "name": "arXiv",
    "url": "https://arxiv.org/abs/2608.14498v1",
    "published_at": "2026-08-14T17:13:34.000Z"
  },
  "source_class": "blog_post",
  "content_type": "Paper",
  "layer": "infra",
  "domains": [
    "image",
    "video"
  ],
  "topics": [],
  "verification": {
    "status": "needs_review",
    "label": "Needs Review",
    "method": "unverified",
    "verified_at": null
  },
  "uncertainty": [
    "On **32 H800 GPUs**, Rollplex reports **1.23×–1.30×** speedup over serial colocation and **1.57×–2.24×** over disaggregation. The material provides results for this hardware and workload setup, not broader cost or scaling evidence."
  ],
  "connected_context": {
    "meaning": "Rollplex adds a concrete systems pattern for visual on-policy post-training: overlap response-independent prefix work with rollout decoding and share compatible tensor storage across phase-specific parallel layouts. It strengthens the case that phase scheduling can outperform both serial colocation and disaggregation under a fixed cluster budget, while narrowing the evidence to synchronous Qwen2.5-VL training on 32 H800s rather than general scaling or cost efficiency.",
    "corpus_size": 468,
    "generated_at": "2026-08-17T10:05:08.340Z",
    "connections": [
      {
        "title": "Taking Reinforcement Learning Cross Datacenter — Nan Jiang, Modal",
        "source_name": "AI Engineer",
        "source_url": "https://www.youtube.com/watch?v=maRzp4kImJ4",
        "feed7_url": "https://feed7.dev/p/taking-reinforcement-learning-cross-datacenter-nan-jiang-modal-05t7dwu",
        "reason": "Both reorganize rollout infrastructure, but at different boundaries: Rollplex spatially shares one GPU cluster while preserving synchronous updates, whereas the cross-datacenter proposal separates rollout workers and synchronizes sparse weight deltas."
      },
      {
        "title": "Accelerating Transformers Fine-Tuning with NVIDIA NeMo AutoModel",
        "source_name": "huggingface.co",
        "source_url": "https://huggingface.co/blog/nvidia/accelerating-fine-tuning-nvidia-nemo-automodel",
        "feed7_url": "https://feed7.dev/p/accelerating-fine-tuning-nvidia-nemo-automodel-0dj2ywc",
        "reason": "Both show that post-training throughput can improve through infrastructure changes without changing the model objective, but NeMo AutoModel addresses MoE fine-tuning APIs while Rollplex targets phase overlap and memory sharing in VLM rollouts, so their speedups are not directly comparable."
      },
      {
        "title": "OPD-V: Visual On-Policy Self-Distillation with Modality Balance",
        "source_name": "arXiv",
        "source_url": "https://arxiv.org/abs/2608.05131v1",
        "feed7_url": "https://feed7.dev/p/2608-05131v1-0m0x349",
        "reason": "OPD-V supplies an example of visual on-policy post-training whose rollout workload could benefit from Rollplex’s scheduling pattern; Rollplex addresses execution efficiency rather than OPD-V’s modality-balance objective."
      }
    ]
  },
  "lifecycle": "Current",
  "published_at": "2026-08-14T17:13:34.000Z",
  "modified_at": "2026-08-14T17:13:34.000Z",
  "supersedes": [],
  "expires_at": null,
  "formats": {
    "html": "https://feed7.dev/p/2608-14498v1-1a3lotd",
    "json": "https://feed7.dev/p/2608-14498v1-1a3lotd.json",
    "markdown": "https://feed7.dev/p/2608-14498v1-1a3lotd.md"
  }
}