{
  "schema_version": "1.1",
  "id": "archive:https://arxiv.org/abs/2610.42148v1",
  "slug": "2610-42148v1-1royypb",
  "url": "https://feed7.dev/p/2610-42148v1-1royypb",
  "title": "Flide: Accelerating Decision Inference with Exact State Forking",
  "why_included": "Flide targets decision-model inference, where ordinary generation engines can recompute shared prefixes or alter output probabilities. Its reported speedups suggest a specialized serving path for agent workloads with branching decisions.",
  "summary": "Flide flattens decision-request trees into a token stream and uses GPU-resident indices to navigate branches across standard attention and recurrent Gated DeltaNet layers. It reports up to **9.5×** faster answers than vLLM and SGLang, with outputs closest to a float32 reference in total variation distance.",
  "practical_implication": "If your agent workload makes repeated categorical decisions over shared history, consider whether a decision model and specialized inference engine could reduce latency. Flide also supports resuming expanding histories from earlier checkpoints.",
  "agent_context": "Flide flattens decision-request trees into a token stream and uses GPU-resident indices to navigate branches across standard attention and recurrent Gated DeltaNet layers. It reports up to **9.5×** faster answers than vLLM and SGLang, with outputs closest to a float32 reference in total variation distance.\n\nIf your agent workload makes repeated categorical decisions over shared history, consider whether a decision model and specialized inference engine could reduce latency. Flide also supports resuming expanding histories from earlier checkpoints.\n\nThe results are from evaluations on public hybrid and decision models; the material does not establish production performance across serving setups. Exactness is stated for exact arithmetic, while the reported comparison uses a float32 reference.",
  "source": {
    "name": "arXiv",
    "url": "https://arxiv.org/abs/2610.42148v1",
    "published_at": "2026-10-30T17:59:38.000Z"
  },
  "source_class": "blog_post",
  "content_type": "Paper",
  "layer": "infra",
  "domains": [
    "research"
  ],
  "topics": [
    "reasoning"
  ],
  "verification": {
    "status": "needs_review",
    "label": "Needs Review",
    "method": "unverified",
    "verified_at": null
  },
  "uncertainty": [
    "The results are from evaluations on public hybrid and decision models; the material does not establish production performance across serving setups. Exactness is stated for exact arithmetic, while the reported comparison uses a float32 reference."
  ],
  "connected_context": {
    "meaning": "LSA adds a learned latent-space index for reducing attention work at very long context lengths, with results at 10M tokens for a single layer. This is a distinct route from tree-indexed sparse attention, and the layer-level speedups and bounded accuracy evidence make end-to-end serving validation necessary before inferring agent-level gains.",
    "corpus_size": 974,
    "generated_at": "2026-11-02T11:02:31.143Z",
    "connections": []
  },
  "lifecycle": "Current",
  "published_at": "2026-10-30T17:59:38.000Z",
  "modified_at": "2026-10-30T17:59:38.000Z",
  "supersedes": [],
  "expires_at": null,
  "formats": {
    "html": "https://feed7.dev/p/2610-42148v1-1royypb",
    "json": "https://feed7.dev/p/2610-42148v1-1royypb.json",
    "markdown": "https://feed7.dev/p/2610-42148v1-1royypb.md"
  }
}