{ "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": "2610.03198", "title": "KV$^2$: A Self-Refining KV Cache", "summary": "KV^2 replaces both lightweight proxy-only eviction scoring and full-context reconstruction scoring for query-agnostic KV-cache compression: a cheap proxy scorer selects informative tokens, and only that subset is reprocessed to compute final eviction scores. The cost is an extra selective-reconstruction pass at compression time plus implementation complexity over single-pass proxy scoring, though it is cheaper at compression time than reprocessing the full prompt. It can fail if the lightweight proxy scorer discards tokens that later queries need, since eviction decisions are query-agnostic and errors made at compression time are unrecoverable across all downstream queries.", "rationale": "The 2%-budget setting directly addresses single-GPU long-context serving with a reusable prefill, but the quality gain is demonstrated only on RULER, NIAH, and LongBench, so the reader should validate proxy-scorer reliability on their own prompts before production adoption.", "technique": "Two-stage query-agnostic KV cache compression", "family": "cache_kv", "recommendation": "testar", "published": "2026-10-02", "as_of": "2026-10-18", "editorial_status": "indexed", "source_url": "https://arxiv.org/abs/2610.03198", "independent_implementations": 0, "total_implementations": 1, "stars": 121, "citations": null, "claims": [ { "claim_id": "claim-01", "statement": "KV^2 replaces both lightweight proxy-only eviction scoring and full-context reconstruction scoring for query-agnostic KV-cache compression: a cheap proxy scorer selects informative tokens, and only that subset is reprocessed to compute final eviction scores. The cost is an extra selective-reconstruction pass at compression time plus implementation complexity over single-pass proxy scoring, though it is cheaper at compression time than reprocessing the full prompt. It can fail if the lightweight proxy scorer discards tokens that later queries need, since eviction decisions are query-agnostic and errors made at compression time are unrecoverable across all downstream queries.", "basis": "inferred", "result": "", "baseline": "", "conditions": "", "source_url": "https://arxiv.org/abs/2610.03198", "source_page": null, "source_excerpt": "" }, { "claim_id": "claim-02", "statement": "Operates at 2%-10% KV-cache budgets (a 50x reduction at 2%); on RULER 16K at 2% budget it exceeds the next-best baseline by over 40 percentage points, with lower compression-stage runtime and peak memory than full-context reconstruction scoring.", "basis": "inferred", "result": "", "baseline": "", "conditions": "", "source_url": "https://arxiv.org/abs/2610.03198", "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": "" }