{ "schema_version": 1, "builder_plan": { "scenario": "A self-hosted assistant is too expensive because its model uses most of the GPU memory.", "change": "Try the paper's supported lower-precision weights on a copy of your serving setup.", "baseline": "The same model at its original precision, on the same GPU.", "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": "Peak memory, task pass rate, latency, and total serving cost at the same context and load.", "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": "2610.31412", "title": "SQuAT: Spherical Lattice Vector Quantization-Aware Training", "summary": "SQuAT replaces the standard straight-through estimator applied to the full vector quantizer with separate STEs for direction (E8-derived spherical codebook) and magnitude (low-bit scalar quantizer), yielding tangent-space updates equivalent to implicit Riemannian optimization. The cost is a quantization-aware training run over billions of tokens requiring a distributed multi-worker setup with block-wise assignment and index communication, plus implementation complexity well beyond post-training quantization. It can fail if the reader lacks the compute to run QAT at all, if the E8 codebook assignment overhead dominates at small scale, or if accuracy gains do not transfer to models and bit-widths outside those evaluated.", "rationale": "The method delivers state-of-the-art low-bit accuracy but presupposes multi-worker training infrastructure and billion-token QAT budgets that a single 24 GB GPU or API-based workflow cannot support, so it is relevant to track but not adoptable now.", "technique": "Shape-gain spherical lattice VQ-QAT for LLMs", "family": "quantizacao", "recommendation": "observar", "published": "2026-10-22", "as_of": "2026-10-23", "editorial_status": "indexed", "source_url": "https://arxiv.org/abs/2610.31412", "independent_implementations": 0, "total_implementations": 0, "stars": 0, "citations": null, "claims": [ { "claim_id": "claim-01", "statement": "SQuAT replaces the standard straight-through estimator applied to the full vector quantizer with separate STEs for direction (E8-derived spherical codebook) and magnitude (low-bit scalar quantizer), yielding tangent-space updates equivalent to implicit Riemannian optimization. The cost is a quantization-aware training run over billions of tokens requiring a distributed multi-worker setup with block-wise assignment and index communication, plus implementation complexity well beyond post-training quantization. It can fail if the reader lacks the compute to run QAT at all, if the E8 codebook assignment overhead dominates at small scale, or if accuracy gains do not transfer to models and bit-widths outside those evaluated.", "basis": "inferred", "result": "", "baseline": "", "conditions": "", "source_url": "https://arxiv.org/abs/2610.31412", "source_page": null, "source_excerpt": "" }, { "claim_id": "claim-02", "statement": "Distributed exact nearest-neighbor assignment reduces VQ-QAT training time by 12.8x, enabling scaling to 10B tokens; SQuAT also consistently outperforms standard full-quantizer STE VQ-QAT on LLaMA-3.2-1B/3B, Qwen3-4B, and Qwen3-1.7B under identical quantizers.", "basis": "inferred", "result": "", "baseline": "", "conditions": "", "source_url": "https://arxiv.org/abs/2610.31412", "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": "" }