{
  "abstract": "The key-value (KV) cache is the dominant memory bottleneck of long-context large language model (LLM) inference, growing linearly with context length. We show that uniform KV quantization on a fractional-bit grid does not degrade gracefully: under a prespecified multi-seed statistical protocol, Llama-3.1-8B-Instruct with an affine quantizer is statistically indistinguishable from FP16 KV down to 2.322 code bits/value and collapses at 2.0 bits - a quality cliff in (2.0, 2.322] that reappears in generation-time quantization and multi-turn dialogue and transfers to Mistral-7B. The cliff reframes importance-aware mixed precision: above it, eight model-internal importance indicators are statistically interchangeable, so the benefit of mixing is grid interpolation, reaching average precisions uniform quantization cannot realize. SemKV preserves every token, ranks tokens by a model-internal score, and assigns two adjacent above-cliff precisions, achieving a measured 6.0x storage reduction with no statistically detectable quality difference from full KV (n=900, three seeds), and outperforming FP16 token pruning granted a 1.5x larger memory budget. Replacing the affine base with a distortion-optimized quantizer (TurboQuant-MSE) lowers the cliff in every protocol tested, raising the no-detectable-loss operating point to 7.9x. The recipe: measure the cliff for the target deployment setting, then interpolate above it.",
  "abstract_shown": true,
  "affiliations": [],
  "article_type": null,
  "arxiv": "2608.28911",
  "author_total": 3,
  "authors": [
    {
      "name": "Daeha Lee"
    },
    {
      "name": "Do-Hyung Kim"
    },
    {
      "name": "Jae-Hong Kim"
    }
  ],
  "bibtex": "https://research.draconitic.com/paper/az771359jx352.bib",
  "cited_by": 0,
  "doi": "10.48550/arxiv.2608.28911",
  "fulltext_url": null,
  "id": "az771359jx352",
  "issns": [],
  "landing_url": "https://arxiv.org/abs/2608.28911",
  "language": null,
  "licence": "LicenseRef-arXiv-nonexclusive-1.0",
  "licence_label": "arXiv licence",
  "licence_source": "article",
  "licence_url": "http://arxiv.org/licenses/nonexclusive-distrib/1.0/",
  "metadata_licence": "CC0-1.0",
  "native_subjects": [
    [
      "arxiv",
      "cs.CL"
    ],
    [
      "arxiv",
      "cs.IT"
    ],
    [
      "arxiv",
      "cs.LG"
    ],
    [
      "arxiv",
      "math.IT"
    ],
    [
      "openalex",
      "1707|Advanced Neural Network Applications"
    ]
  ],
  "openalex": "W7204930987",
  "pdf_url": "https://arxiv.org/pdf/2608.28911",
  "pmcid": null,
  "pmid": null,
  "published": "2026-08-28",
  "publisher": null,
  "ris": "https://research.draconitic.com/paper/az771359jx352.ris",
  "sources": [
    {
      "first_seen": "2026-10-30T12:17:48Z",
      "native_id": "2608.28911",
      "source": "arxiv",
      "url": "https://arxiv.org/abs/2608.28911"
    }
  ],
  "subjects": [
    {
      "field": "mathematics",
      "origin": "arxiv:math.IT",
      "subfield": "general"
    }
  ],
  "title": "SemKV: Semantic Mixed-Precision KV Cache Quantization Guided by the Quality Cliff for Long-Context LLM Inference",
  "title_text": "SemKV: Semantic Mixed-Precision KV Cache Quantization Guided by the Quality Cliff for Long-Context LLM Inference",
  "type": "preprint",
  "updated_at": "2026-10-30T12:17:48Z",
  "url": "https://research.draconitic.com/paper/az771359jx352",
  "venue": "arXiv",
  "versions": []
}