{ "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": "2610.03007", "title": "AvoKV-E: Payload-Aware KV Cache Eviction for Long Reasoning", "summary": "AvoKV-E replaces routing-only KV eviction policies (recurrence-, redundancy-, or attention-read-based scoring) with a training-free policy that delays eviction eligibility for recent states and ranks entries by normalized read pressure, key redundancy, and value-payload magnitude. It costs additional per-step scoring computation and implementation complexity in the inference path, with no training or extra model weights; memory cost is the eviction metadata plus the unchanged chosen cache budget. It can fail if its scoring heuristics miscalibrate on workloads unlike the evaluated long-reasoning traces, if the delay window is mistuned for short outputs, or if gains do not replicate outside the tested models and tightest-budget regimes.", "rationale": "The method is training-free and directly compatible with single-GPU or API-served long-reasoning inference, but the reader should validate accuracy at their target cache budget on their own reasoning workloads before production adoption.", "technique": "Payload-aware KV cache eviction for long reasoning", "family": "cache_kv", "recommendation": "testar", "published": "2026-10-02", "as_of": "2026-10-18", "editorial_status": "indexed", "source_url": "https://arxiv.org/abs/2610.03007", "independent_implementations": 0, "total_implementations": 0, "stars": 0, "citations": null, "claims": [ { "claim_id": "claim-01", "statement": "AvoKV-E replaces routing-only KV eviction policies (recurrence-, redundancy-, or attention-read-based scoring) with a training-free policy that delays eviction eligibility for recent states and ranks entries by normalized read pressure, key redundancy, and value-payload magnitude. It costs additional per-step scoring computation and implementation complexity in the inference path, with no training or extra model weights; memory cost is the eviction metadata plus the unchanged chosen cache budget. It can fail if its scoring heuristics miscalibrate on workloads unlike the evaluated long-reasoning traces, if the delay window is mistuned for short outputs, or if gains do not replicate outside the tested models and tightest-budget regimes.", "basis": "inferred", "result": "", "baseline": "", "conditions": "", "source_url": "https://arxiv.org/abs/2610.03007", "source_page": null, "source_excerpt": "" }, { "claim_id": "claim-02", "statement": "Matches or exceeds redundancy-aware, recurrence-based, and thought-adaptive eviction baselines at matched active-KV budgets, with largest gains in the tightest-cache regime; no absolute factor is reported in the abstract.", "basis": "inferred", "result": "", "baseline": "", "conditions": "", "source_url": "https://arxiv.org/abs/2610.03007", "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": "" }