{ "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": "2609.38166", "title": "LeapQuant: Efficient Linear Attention with Accurate Recurrent State Quantization", "summary": "LeapQuant replaces full-precision recurrent-state reads and updates in hybrid linear-attention models (Gated DeltaNet, Kimi Delta Attention) with 8-bit quantized states, using per-window quantization with buffered high-precision updates plus a few high-precision Compensator Tokens for outlier rows and columns. It is training-free and costs implementation complexity: custom kernels, window buffering logic, and residual smoothing must be integrated into the inference path. It can fail if the deployed stack lacks the paper's kernels (the 1.47x figure depends on specific GPU kernels), if the reader's models use standard attention rather than linear-attention hybrids, or if error accumulation resurfaces on workloads whose state outlier structure differs from the evaluated Qwen, Kimi, and GLM families.", "rationale": "The method is training-free and targets exactly the single-GPU long-context inference regime the reader operates in, but the end-to-end gain is modest, hardware-kernel-dependent, and only relevant if the reader serves linear-attention hybrid models, so workload-specific validation is required before adoption.", "technique": "Recurrent-state quantization for linear attention", "family": "quantizacao", "recommendation": "testar", "published": "2026-09-29", "as_of": "2026-10-18", "editorial_status": "indexed", "source_url": "https://arxiv.org/abs/2609.38166", "independent_implementations": 0, "total_implementations": 1, "stars": 1, "citations": null, "claims": [ { "claim_id": "claim-01", "statement": "LeapQuant replaces full-precision recurrent-state reads and updates in hybrid linear-attention models (Gated DeltaNet, Kimi Delta Attention) with 8-bit quantized states, using per-window quantization with buffered high-precision updates plus a few high-precision Compensator Tokens for outlier rows and columns. It is training-free and costs implementation complexity: custom kernels, window buffering logic, and residual smoothing must be integrated into the inference path. It can fail if the deployed stack lacks the paper's kernels (the 1.47x figure depends on specific GPU kernels), if the reader's models use standard attention rather than linear-attention hybrids, or if error accumulation resurfaces on workloads whose state outlier structure differs from the evaluated Qwen, Kimi, and GLM families.", "basis": "inferred", "result": "", "baseline": "", "conditions": "", "source_url": "https://arxiv.org/abs/2609.38166", "source_page": null, "source_excerpt": "" }, { "claim_id": "claim-02", "statement": "Near-lossless 8-bit recurrent-state quantization with 2.05-3.70x kernel-level speedups and 1.47x end-to-end inference speedup on B200, RTX PRO 6000, and RTX 5090 GPUs, at accuracy comparable to FP32.", "basis": "inferred", "result": "", "baseline": "", "conditions": "", "source_url": "https://arxiv.org/abs/2609.38166", "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": "" }