{
  "$schema": "https://themachinepress.com/schemas/story-v1.schema.json",
  "schema_version": "1.0.0",
  "document_type": "machine_press_story",
  "story": {
    "story_id": "mp-2026-11-06-015",
    "source_story_id": "tmp-brief-autocompress",
    "edition_id": "mp-2026-11-06-morning-0119",
    "edition_url": "https://themachinepress.com/edition/2026-11-06",
    "position": 17,
    "story_type": "ticker",
    "section": "chips-infrastructure",
    "editorial_classification": "editorial",
    "headline": "Compression Targets the Hardware, Not the Parameter Count",
    "slug": "compression-targets-the-hardware-not-the-parameter-count",
    "dek": "AutoCompress searches architectures against latency, throughput and memory models validated on actual serving stacks.",
    "summary": "AutoCompress searches architectures against latency, throughput and memory models validated on actual serving stacks.",
    "body_text": "The authors report higher average accuracy than Minitron across RTX 5070, RTX 5080 and H100 settings, and compress Nemotron 3.5 Super from 120 billion to 74 billion parameters for one B200 at 128K context while retaining 95 percent of its mean benchmark score.",
    "why_it_matters": "AutoCompress searches architectures against latency, throughput and memory models validated on actual serving stacks.",
    "limitations": [],
    "importance": 8,
    "canonical_url": "https://themachinepress.com/story/mp-2026-11-06-015/compression-targets-the-hardware-not-the-parameter-count",
    "json_url": "https://themachinepress.com/story/mp-2026-11-06-015.json",
    "first_published_at": "2026-11-06T09:00:00.000-05:00",
    "modified_at": "2026-11-06T09:00:00.000-05:00",
    "content_status": "new",
    "is_carryover": false,
    "carryover_reason": null,
    "key_claims": [
      {
        "claim_id": "claim-mp-2026-11-06-015-001",
        "text": "AutoCompress searches architectures against latency, throughput and memory models validated on actual serving stacks.",
        "source_ids": [
          "source-2026-11-06-017"
        ],
        "qualification": null
      }
    ],
    "source_ids": [
      "source-2026-11-06-017"
    ],
    "tags": [
      "model compression",
      "hardware",
      "serving"
    ],
    "image_url": null,
    "corrections": []
  },
  "sources": [
    {
      "source_id": "source-2026-11-06-017",
      "title": "arXiv preprint 2611.08290, original v1",
      "publisher": "arXiv",
      "url": "https://arxiv.org/abs/2611.08290v1",
      "canonical_url": "https://arxiv.org/abs/2611.08290v1",
      "source_type": "primary_research",
      "is_primary_source": true,
      "published_at": "2026-11-05T11:53:48.000-05:00",
      "accessed_at": "2026-11-06T08:21:03.209-05:00",
      "supports_claim_ids": [
        "claim-mp-2026-11-06-015-001"
      ]
    }
  ],
  "corrections": [],
  "publisher": {
    "name": "The Machine Press",
    "url": "https://themachinepress.com",
    "description": "A daily newspaper for the age of artificial intelligence."
  },
  "cite_this_report": {
    "title": "Compression Targets the Hardware, Not the Parameter Count",
    "publisher": "The Machine Press",
    "published_at": "2026-11-06T09:00:00.000-05:00",
    "canonical_url": "https://themachinepress.com/story/mp-2026-11-06-015/compression-targets-the-hardware-not-the-parameter-count"
  }
}