{
  "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": "2610.18642",
  "title": "MultiAlloc: Budget-Aware Precision Allocation for KV Cache Quantization",
  "summary": "It replaces fixed-granularity KV cache quantization policies (uniform layer-wise or token-wise precision assignment) with a profiling-guided allocator that splits a fixed KV memory budget between layer-wise precision enhancement and token-wise preservation. The cost is a one-time offline profiling pass to estimate layer sensitivity and token importance, plus implementation complexity for mixed-granularity quantization; there is no online search overhead at inference. It can fail if the profiling distribution diverges from production traffic (token-importance and layer-sensitivity estimates may not transfer), if the serving stack lacks kernels supporting mixed per-layer and per-token precision, or if the abstract's unquantified gains do not replicate on the reader's model and workload.",
  "rationale": "The approach targets exactly the reader's constraint (a fixed KV memory budget on a single GPU) and avoids online search, but the claim is unquantified and the benefit depends on workload-specific profiling and kernel support, so it merits a small-scale validation before production adoption.",
  "technique": "Budget-aware mixed-granularity KV cache quantization",
  "family": "cache_kv",
  "recommendation": "testar",
  "published": "2026-10-13",
  "as_of": "2026-10-18",
  "editorial_status": "indexed",
  "source_url": "https://arxiv.org/abs/2610.18642",
  "independent_implementations": 0,
  "total_implementations": 0,
  "stars": 0,
  "citations": null,
  "claims": [
    {
      "claim_id": "claim-01",
      "statement": "It replaces fixed-granularity KV cache quantization policies (uniform layer-wise or token-wise precision assignment) with a profiling-guided allocator that splits a fixed KV memory budget between layer-wise precision enhancement and token-wise preservation. The cost is a one-time offline profiling pass to estimate layer sensitivity and token importance, plus implementation complexity for mixed-granularity quantization; there is no online search overhead at inference. It can fail if the profiling distribution diverges from production traffic (token-importance and layer-sensitivity estimates may not transfer), if the serving stack lacks kernels supporting mixed per-layer and per-token precision, or if the abstract's unquantified gains do not replicate on the reader's model and workload.",
      "basis": "inferred",
      "result": "",
      "baseline": "",
      "conditions": "",
      "source_url": "https://arxiv.org/abs/2610.18642",
      "source_page": null,
      "source_excerpt": ""
    },
    {
      "claim_id": "claim-02",
      "statement": "MultiAlloc 'consistently improves over layer-wise, token-wise, and recent multi-granularity KV cache quantization methods under matched memory budgets' across reasoning and long-context benchmarks on Mistral, Llama, DeepSeek-Qwen, and QwQ; no numeric accuracy or compression factor is stated in the abstract.",
      "basis": "inferred",
      "result": "",
      "baseline": "",
      "conditions": "",
      "source_url": "https://arxiv.org/abs/2610.18642",
      "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": ""
}