{ "schema_version": 1, "builder_plan": { "scenario": "Your self-hosted model spends too long generating answers token by token.", "change": "Check whether a compatible draft model and serving integration exist before renting compute.", "baseline": "The same target with speculation off, then your existing drafter if you have one.", "steps": [ "Locate the paper's implementation and check its license, model support, data needs, and hardware requirements. Stop here if you cannot obtain the required artifacts.", "Save a small set of representative product tasks and expected outcomes. Keep evaluation tasks out of any training or tuning, and choose quality and cost limits before running.", "Run the baseline and the proposed change on identical inputs. Keep other settings fixed and repeat paired runs to expose variation.", "Save per-task outputs, failures, timings, and all usage costs. Review regressions before expanding the test." ], "measure": "Accepted draft length, tokens per second, tail latency, output quality, and total GPU cost.", "decision": "Keep the change only if it meets your preselected quality and latency limits and improves the metric you care about after setup and operating costs. Otherwise retain the baseline. This is an exploration plan, not a validated recommendation." }, "builder_review": null, "arxiv_id": "2610.20392", "title": "FOLD: Position-Free Far Memory with Local Order for Long-Context Speculative Decoding", "summary": "FOLD replaces retraining or window-restriction of speculative drafters on long inputs by keeping exact local positions, collapsing distant KV entries into a far-memory at a shared position, and sharpening attention within it. It is training-free, so it costs no fine-tuning budget; the overhead is modified drafter attention logic and hyperparameters controlling far-memory sharpening and its weight versus the local window. It can fail through attention dilution on very long histories, sensitivity of the sharpening and weighting hyperparameters to workload, and gains that depend on having a compatible target-drafter speculative decoding stack already in production.", "rationale": "The method is training-free and cheap to try, but its benefit is conditional on running speculative decoding with long inputs, so it warrants a workload-specific pilot rather than unconditional adoption.", "technique": "Training-free far-memory drafting for speculative decoding", "family": "decodificacao_especulativa", "recommendation": "testar", "published": "2026-10-14", "as_of": "2026-10-18", "editorial_status": "indexed", "source_url": "https://arxiv.org/abs/2610.20392", "independent_implementations": 0, "total_implementations": 0, "stars": 0, "citations": null, "claims": [ { "claim_id": "claim-01", "statement": "FOLD replaces retraining or window-restriction of speculative drafters on long inputs by keeping exact local positions, collapsing distant KV entries into a far-memory at a shared position, and sharpening attention within it. It is training-free, so it costs no fine-tuning budget; the overhead is modified drafter attention logic and hyperparameters controlling far-memory sharpening and its weight versus the local window. It can fail through attention dilution on very long histories, sensitivity of the sharpening and weighting hyperparameters to workload, and gains that depend on having a compatible target-drafter speculative decoding stack already in production.", "basis": "inferred", "result": "", "baseline": "", "conditions": "", "source_url": "https://arxiv.org/abs/2610.20392", "source_page": null, "source_excerpt": "" }, { "claim_id": "claim-02", "statement": "Best case raises speedup over autoregressive decoding from 2.08x to 4.22x (about 2.03x relative improvement), predominantly the highest speedup across six target-drafter pairs on benchmarks up to 256K tokens.", "basis": "inferred", "result": "", "baseline": "", "conditions": "", "source_url": "https://arxiv.org/abs/2610.20392", "source_page": null, "source_excerpt": "" } ], "exposure_map": [ { "dimension": "quality", "basis": "not_evaluated", "finding": "", "interpretation": "", "limitation": "", "next_check": "", "source_url": "", "source_page": null, "source_excerpt": "" }, { "dimension": "compute", "basis": "not_evaluated", "finding": "", "interpretation": "", "limitation": "", "next_check": "", "source_url": "", "source_page": null, "source_excerpt": "" }, { "dimension": "latency", "basis": "not_evaluated", "finding": "", "interpretation": "", "limitation": "", "next_check": "", "source_url": "", "source_page": null, "source_excerpt": "" }, { "dimension": "operations", "basis": "not_evaluated", "finding": "", "interpretation": "", "limitation": "", "next_check": "", "source_url": "", "source_page": null, "source_excerpt": "" }, { "dimension": "compatibility", "basis": "not_evaluated", "finding": "", "interpretation": "", "limitation": "", "next_check": "", "source_url": "", "source_page": null, "source_excerpt": "" }, { "dimension": "security", "basis": "not_evaluated", "finding": "", "interpretation": "", "limitation": "", "next_check": "", "source_url": "", "source_page": null, "source_excerpt": "" }, { "dimension": "data_and_training", "basis": "not_evaluated", "finding": "", "interpretation": "", "limitation": "", "next_check": "", "source_url": "", "source_page": null, "source_excerpt": "" }, { "dimension": "reproducibility", "basis": "not_evaluated", "finding": "", "interpretation": "", "limitation": "", "next_check": "", "source_url": "", "source_page": null, "source_excerpt": "" } ], "risks": [], "minimum_test": [], "open_questions": [], "independent_tests": [], "lab_reproductions": [], "report_available": false, "validation_tier": "unknown", "evidence_tier": "unknown", "infrastructure_basis": "unknown", "equations": [], "equations_status": "not_fetched", "core_kind": "", "equations_fetched_at": "", "equations_source_url": "", "equations_source_sha256": "", "exposures_reviewed_at": "", "exposures_source_sha256": "" }