{
  "$schema": "https://themachinepress.com/schemas/story-v1.schema.json",
  "schema_version": "1.0.0",
  "document_type": "machine_press_story",
  "story": {
    "story_id": "mp-2026-10-08-017",
    "source_story_id": "tmp-sidebar-residualquant-kv-cache",
    "edition_id": "mp-2026-10-08-morning-0090",
    "edition_url": "https://themachinepress.com/edition/2026-10-08",
    "position": 19,
    "story_type": "ticker",
    "section": "infrastructure",
    "editorial_classification": "editorial",
    "headline": "Two-Bit Residuals Cut Looped-Transformer Cache Storage",
    "slug": "two-bit-residuals-cut-looped-transformer-cache-storage",
    "dek": "ResidualQuant used final-loop states as a reference and encoded similar earlier loops as low-precision differences.",
    "summary": "ResidualQuant used final-loop states as a reference and encoded similar earlier loops as low-precision differences.",
    "body_text": "Mixed-precision tests retained accuracy near BF16 while reducing theoretical KV storage 80.7%. On an RTX 5090, lower memory traffic improved fixed-batch decode throughput up to 2.73 times and the larger feasible batch raised peak throughput up to 4.15 times in the reported models.",
    "why_it_matters": "ResidualQuant used final-loop states as a reference and encoded similar earlier loops as low-precision differences.",
    "limitations": [],
    "importance": 6,
    "canonical_url": "https://themachinepress.com/story/mp-2026-10-08-017/two-bit-residuals-cut-looped-transformer-cache-storage",
    "json_url": "https://themachinepress.com/story/mp-2026-10-08-017.json",
    "first_published_at": "2026-10-08T09:00:00.000-04:00",
    "modified_at": "2026-10-08T09:00:00.000-04:00",
    "content_status": "new",
    "is_carryover": false,
    "carryover_reason": null,
    "key_claims": [
      {
        "claim_id": "claim-mp-2026-10-08-017-001",
        "text": "ResidualQuant used final-loop states as a reference and encoded similar earlier loops as low-precision differences.",
        "source_ids": [
          "source-2026-10-08-019"
        ],
        "qualification": null
      }
    ],
    "source_ids": [
      "source-2026-10-08-019"
    ],
    "tags": [
      "KV cache",
      "quantization",
      "looped transformers"
    ],
    "image_url": null,
    "corrections": []
  },
  "sources": [
    {
      "source_id": "source-2026-10-08-019",
      "title": "arXiv preprint 2610.10381",
      "publisher": "arXiv",
      "url": "https://arxiv.org/abs/2610.10381",
      "canonical_url": "https://arxiv.org/abs/2610.10381",
      "source_type": "primary_research",
      "is_primary_source": true,
      "published_at": "2026-10-07T12:40:21.000-04:00",
      "accessed_at": "2026-10-08T08:26:48.909-04:00",
      "supports_claim_ids": [
        "claim-mp-2026-10-08-017-001"
      ]
    }
  ],
  "corrections": [],
  "publisher": {
    "name": "The Machine Press",
    "url": "https://themachinepress.com",
    "description": "A daily newspaper for the age of artificial intelligence."
  },
  "cite_this_report": {
    "title": "Two-Bit Residuals Cut Looped-Transformer Cache Storage",
    "publisher": "The Machine Press",
    "published_at": "2026-10-08T09:00:00.000-04:00",
    "canonical_url": "https://themachinepress.com/story/mp-2026-10-08-017/two-bit-residuals-cut-looped-transformer-cache-storage"
  }
}