{
  "$schema": "https://themachinepress.com/schemas/story-v1.schema.json",
  "schema_version": "1.0.0",
  "document_type": "machine_press_story",
  "story": {
    "story_id": "mp-2026-07-11-007",
    "source_story_id": "tmp-story-vllm-transformers-backend",
    "edition_id": "mp-2026-07-11-morning-0002",
    "edition_url": "https://themachinepress.com/edition/2026-07-11",
    "position": 7,
    "story_type": "dispatch",
    "section": "open-source",
    "editorial_classification": "editorial",
    "headline": "One Model Definition Reaches Native Serving Speed",
    "slug": "one-model-definition-reaches-native-serving-speed",
    "dek": "Hugging Face says the Transformers backend for vLLM can now meet or beat hand-written model ports across several tested Qwen3 configurations.",
    "summary": "Hugging Face says the Transformers backend for vLLM can now meet or beat hand-written model ports across several tested Qwen3 configurations.",
    "body_text": "Hugging Face's updated vLLM integration uses graph analysis and source transformation to recognize model patterns, fuse operations, infer parallel plans, and route compatible work through optimized vLLM kernels. In the authors' tests, the Transformers backend matched or exceeded vLLM's hand-written implementations for three Qwen3 deployments ranging from a single-GPU dense model to a 235-billion-parameter mixture-of-experts system. The measurements cover selected compatible architectures, and linear-attention models remain unsupported. The engineering gain is a shorter path from one readable model implementation to high-performance serving.",
    "why_it_matters": "Hugging Face says the Transformers backend for vLLM can now meet or beat hand-written model ports across several tested Qwen3 configurations.",
    "limitations": [
      "The measurements cover selected compatible architectures, and linear-attention models remain unsupported."
    ],
    "importance": 7,
    "canonical_url": "https://themachinepress.com/story/mp-2026-07-11-007/one-model-definition-reaches-native-serving-speed",
    "json_url": "https://themachinepress.com/story/mp-2026-07-11-007.json",
    "first_published_at": "2026-07-11T09:00:00.000-04:00",
    "modified_at": "2026-07-11T23:17:20.883-04:00",
    "content_status": "updated",
    "is_carryover": false,
    "carryover_reason": null,
    "key_claims": [
      {
        "claim_id": "claim-mp-2026-07-11-007-001",
        "text": "Hugging Face says the Transformers backend for vLLM can now meet or beat hand-written model ports across several tested Qwen3 configurations.",
        "source_ids": [
          "source-2026-07-11-007"
        ],
        "qualification": "The measurements cover selected compatible architectures, and linear-attention models remain unsupported."
      }
    ],
    "source_ids": [
      "source-2026-07-11-007"
    ],
    "tags": [
      "vLLM",
      "Transformers",
      "inference",
      "open source",
      "model serving"
    ],
    "image_url": "https://themachinepress.com/issues/2026-07-11/bank-data-center-server-aisle-001.webp",
    "corrections": [
      {
        "correction_id": "correction-2026-07-11-001",
        "revision": 2,
        "summary": "Correction, July 11, 2026: An earlier automated version repeated five ticker briefs and nine dispatches from the July 10 edition. Those items were removed and replaced with previously unpublished editorial stories; the two front-page stories and three genuinely new dispatches were preserved.",
        "published_at": "2026-07-11T23:17:20.883-04:00",
        "affected_story_ids": [],
        "affected_claim_ids": []
      }
    ]
  },
  "sources": [
    {
      "source_id": "source-2026-07-11-007",
      "title": "Hugging Face: Native-speed vLLM Transformers modeling backend",
      "publisher": "Hugging Face",
      "url": "https://huggingface.co/blog/native-speed-vllm-transformers-backend",
      "canonical_url": "https://huggingface.co/blog/native-speed-vllm-transformers-backend",
      "source_type": "repository",
      "is_primary_source": true,
      "published_at": "2026-07-07T20:00:00.000-04:00",
      "accessed_at": "2026-07-11T12:30:00.000-04:00",
      "supports_claim_ids": [
        "claim-mp-2026-07-11-007-001"
      ]
    }
  ],
  "corrections": [
    {
      "correction_id": "correction-2026-07-11-001",
      "revision": 2,
      "summary": "Correction, July 11, 2026: An earlier automated version repeated five ticker briefs and nine dispatches from the July 10 edition. Those items were removed and replaced with previously unpublished editorial stories; the two front-page stories and three genuinely new dispatches were preserved.",
      "published_at": "2026-07-11T23:17:20.883-04:00",
      "affected_story_ids": [],
      "affected_claim_ids": []
    }
  ],
  "publisher": {
    "name": "The Machine Press",
    "url": "https://themachinepress.com",
    "description": "A daily newspaper for the age of artificial intelligence."
  },
  "cite_this_report": {
    "title": "One Model Definition Reaches Native Serving Speed",
    "publisher": "The Machine Press",
    "published_at": "2026-07-11T09:00:00.000-04:00",
    "canonical_url": "https://themachinepress.com/story/mp-2026-07-11-007/one-model-definition-reaches-native-serving-speed"
  }
}