{ "schema_version": 1, "builder_plan": { "scenario": "Long customer conversations exhaust memory or limit concurrent requests.", "change": "Change only the attention cache policy, after checking model and runtime support.", "baseline": "The same model with its ordinary attention cache.", "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 cache memory, long-context answer quality, first-token latency, and requests completed.", "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": "2610.20686", "title": "RoLaQ: Joint Rotation and Range Calibration for 2-Bit KV Caches in Multi-Head Latent Attention", "summary": "RoLaQ replaces BF16 or coarser 2-bit KV cache storage in multi-head latent attention with an offline joint rotation and quantization-range calibration, keeping model weights frozen. The cost is a one-time calibration procedure with prediction-matching and attention-output supervision, plus reliance on a custom SGLang implementation; runtime cost is a rotation applied to the latent cache. It can fail on models without MLA architecture, when the calibration data does not match the deployment distribution, or if the SGLang kernels are unavailable for the reader's serving stack.", "rationale": "The method targets a real bottleneck (KV memory at long context) on constrained hardware, but its benefit is limited to MLA-based models such as DeepSeek-V2-Lite and requires validating calibration quality and kernel support on the reader's specific workload before production adoption.", "technique": "2-bit KV cache quantization for MLA attention", "family": "cache_kv", "recommendation": "testar", "published": "2026-10-14", "as_of": "2026-10-18", "editorial_status": "indexed", "source_url": "https://arxiv.org/abs/2610.20686", "independent_implementations": 0, "total_implementations": 1, "stars": 4, "citations": 0, "claims": [ { "claim_id": "claim-01", "statement": "RoLaQ replaces BF16 or coarser 2-bit KV cache storage in multi-head latent attention with an offline joint rotation and quantization-range calibration, keeping model weights frozen. The cost is a one-time calibration procedure with prediction-matching and attention-output supervision, plus reliance on a custom SGLang implementation; runtime cost is a rotation applied to the latent cache. It can fail on models without MLA architecture, when the calibration data does not match the deployment distribution, or if the SGLang kernels are unavailable for the reader's serving stack.", "basis": "inferred", "result": "", "baseline": "", "conditions": "", "source_url": "https://arxiv.org/abs/2610.20686", "source_page": null, "source_excerpt": "" }, { "claim_id": "claim-02", "statement": "On DeepSeek-V2-Lite, an SGLang implementation provides 3.96x the BF16 KV capacity under the same memory budget and 2.51x decode throughput with 256K input tokens, with the highest average accuracy among compared 2-bit methods across three MLA models and six benchmarks.", "basis": "inferred", "result": "", "baseline": "", "conditions": "", "source_url": "https://arxiv.org/abs/2610.20686", "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": "" }