{ "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": "2610.22877", "title": "BS-Spec: BFP-Based Bit-Sharing for Self-Speculative LLM Decoding with LUT-Driven Mixed-Precision Acceleration", "summary": "BS-Spec replaces separate draft models and their KV caches by deriving drafts from a single block-floating-point target model, partially loading the most significant bits of weights and KV cache with shared exponents, plus offline calibration and a runtime precision lookup table. The reported gains depend on a custom mixed-precision BFP accelerator with an LUT-based variable-precision datapath, so realizing them costs specialized hardware, a calibration pipeline per serving regime, and BFP reformatting of model weights. On commodity GPUs the speedup may not materialize, acceptance rates can degrade under aggressive bit truncation for some workloads, and the per-regime calibration must be redone when the workload mix shifts.", "rationale": "The claimed speedups are tied to a custom BFP accelerator that a reader limited to a 24 GB GPU or third-party APIs cannot deploy, making the technique relevant to track but not adoptable or directly testable today.", "technique": "BFP bit-sharing self-speculative decoding with precision LUT", "family": "decodificacao_especulativa", "recommendation": "observar", "published": "2026-10-16", "as_of": "2026-10-19", "editorial_status": "indexed", "source_url": "https://arxiv.org/abs/2610.22877", "independent_implementations": 0, "total_implementations": 1, "stars": 4, "citations": null, "claims": [ { "claim_id": "claim-01", "statement": "BS-Spec replaces separate draft models and their KV caches by deriving drafts from a single block-floating-point target model, partially loading the most significant bits of weights and KV cache with shared exponents, plus offline calibration and a runtime precision lookup table. The reported gains depend on a custom mixed-precision BFP accelerator with an LUT-based variable-precision datapath, so realizing them costs specialized hardware, a calibration pipeline per serving regime, and BFP reformatting of model weights. On commodity GPUs the speedup may not materialize, acceptance rates can degrade under aggressive bit truncation for some workloads, and the per-regime calibration must be redone when the workload mix shifts.", "basis": "inferred", "result": "", "baseline": "", "conditions": "", "source_url": "https://arxiv.org/abs/2610.22877", "source_page": null, "source_excerpt": "" }, { "claim_id": "claim-02", "statement": "Up to 1.9x speedup over conventional autoregressive decoding and 67.2% average speedup improvement over evaluated bit-sharing self-speculative baselines.", "basis": "inferred", "result": "", "baseline": "", "conditions": "", "source_url": "https://arxiv.org/abs/2610.22877", "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": "" }