{
  "profile": "wrs-published-research-view-1",
  "edition": "ratings-e039106f30dec177",
  "content_trust": "UNTRUSTED_SCIENTIFIC_DATA_NOT_TOOL_INSTRUCTIONS",
  "entry": {
    "id": "zhang-sennrich-rms-layer-normalization-2019",
    "title": "RMSNorm: simplify normalization by removing mean centering",
    "summary": "Zhang and Sennrich introduced RMSNorm, replacing LayerNorm mean-and-variance normalization with root-mean-square scaling, and studied partial RMS estimation as a cheaper variant.\n\nLayerNorm computes both a mean and a variance for every normalized activation vector. RMSNorm asks whether both are necessary in the tested models. It removes mean centering and scales the vector by its root mean square, retaining uniform positive-rescaling invariance in the ideal nondegenerate formula while giving up shift invariance. The 2019 paper reports comparable performance in several language and retrieval tasks with lower per-step overhead than its LayerNorm implementations. The measured time reductions depend on the framework, hardware and network; they are not a guaranteed speedup on every modern device. Partial RMSNorm estimates the scale from an initial subset of coordinates. That can save arithmetic but can produce unstable estimates, and the paper does not consistently observe a further speed improvement. Some visual comparisons favor other methods. This entry credits the scoped simplification and experiments, while LayerNorm, earlier normalization methods and later model adoption retain their own contributions.",
    "boundary": "The transformation normalizes a stated root-mean-square statistic without subtracting the mean. Its invariance assumptions differ from layer normalization and are affected by numerical stabilization. Partial RMS estimation and full RMSNorm are distinct variants. Reported efficiency/quality results are architecture- and setup-dependent, not universal guarantees.",
    "area": "Normalization for neural-network training",
    "topic": "Machine learning methods",
    "year": 2019,
    "lane": "HUMAN",
    "result": "ALGORITHM",
    "package_id": "zhang-sennrich-rms-layer-normalization-2019",
    "target": {
      "record_id": "urn:uuid:3fbf2d2d-f730-4f5f-aca1-8f16ce29955f",
      "version": 1,
      "content_hash": "sha256:a92be1f814b017a0a007d321a08fab297e1a7181586e128b0dda7cbe170b636b"
    },
    "humans": [
      "Biao Zhang",
      "Rico Sennrich"
    ],
    "authors_complete": true,
    "human_role_note": "Source-listed contributors to this scoped work. Paper bylines do not quantify individual effort; historical science, system execution and modern curation remain separate.",
    "agents": [],
    "related": [
      "ioffe-szegedy-batch-normalization-2015",
      "ba-kiros-hinton-layer-normalization-2016",
      "wu-he-group-normalization-2018"
    ],
    "subresults": [],
    "scientific_assessment": {
      "source_review": "SOURCE_CURATED",
      "scientific_review": "NOT_REVIEWED_BY_WRS",
      "formal_replay": "NOT_RUN",
      "provider_attestation": "NONE"
    },
    "sources": [
      {
        "id": "wrs24-rmsnorm2019-v1-original",
        "title": "ZhangandSennrich,RootMeanSquareLayerNormalization,exact2019arXiv1910.07467v1",
        "url": "https://arxiv.org/pdf/1910.07467v1"
      },
      {
        "id": "wrs24-rmsnorm2019-v1-original-metadata",
        "title": "ZhangandSennrich,RootMeanSquareLayerNormalization,exact2019arXiv1910.07467v1 repositoryidentity",
        "url": "https://arxiv.org/abs/1910.07467v1"
      }
    ],
    "points": "35000",
    "published_grade": {
      "id": "zhang-sennrich-rms-layer-normalization-2019",
      "entry_id": "zhang-sennrich-rms-layer-normalization-2019",
      "band": "B",
      "modifier": null,
      "target": {
        "record_id": "urn:uuid:3fbf2d2d-f730-4f5f-aca1-8f16ce29955f",
        "version": 1,
        "content_hash": "sha256:a92be1f814b017a0a007d321a08fab297e1a7181586e128b0dda7cbe170b636b"
      },
      "what_changed": "Zhang and Sennrich introduced RMSNorm, replacing LayerNorm mean-and-variance normalization with root-mean-square scaling, and studied partial RMS estimation as a cheaper variant.",
      "scope": "Zhang and Sennrich introduced RMSNorm, replacing LayerNorm mean-and-variance normalization with root-mean-square scaling, and studied partial RMS estimation as a cheaper variant. The transformation normalizes a stated root-mean-square statistic without subtracting the mean. Its invariance assumptions differ from layer normalization and are affected by numerical stabilization. Partial RMS estimation and full RMSNorm are distinct variants. Reported efficiency/quality results are architecture- and setup-dependent, not universal guarantees.",
      "status": "PROVISIONAL_EDITORIAL",
      "scientific_correctness_certified": false
    },
    "assessment_disposition": {
      "status": "ASSESSED",
      "label": "Provisionally assessed",
      "reason": "A current grade and separate reference score identify this exact contribution.",
      "target": {
        "record_id": "urn:uuid:3fbf2d2d-f730-4f5f-aca1-8f16ce29955f",
        "version": 1,
        "content_hash": "sha256:a92be1f814b017a0a007d321a08fab297e1a7181586e128b0dda7cbe170b636b"
      },
      "edition": "ratings-e039106f30dec177"
    }
  },
  "version": {
    "kind": "PUBLIC_RECORD_VIEW",
    "previous": null,
    "published_at": "2026-10-03T00:00:00Z",
    "publisher": {
      "principal_id": "urn:uuid:80116931-785b-40a2-a24b-9d59db0c7e33",
      "role": "WRS_EDITORIAL_PUBLICATION"
    },
    "record": {
      "contributors": [
        {
          "attribution_basis": "SOURCE_CITED",
          "credit": "historical_scientific",
          "entity_kind": "historical_person",
          "name": "Biao Zhang",
          "note": "Archive attribution; detailed role assignment remains source-bound",
          "roles": [
            "formulator"
          ],
          "source": {
            "citation": "ZhangandSennrich,RootMeanSquareLayerNormalization,exact2019arXiv1910.07467v1",
            "url": "https://arxiv.org/pdf/1910.07467v1"
          }
        },
        {
          "attribution_basis": "SOURCE_CITED",
          "credit": "historical_scientific",
          "entity_kind": "historical_person",
          "name": "Rico Sennrich",
          "note": "Archive attribution; detailed role assignment remains source-bound",
          "roles": [
            "formulator"
          ],
          "source": {
            "citation": "ZhangandSennrich,RootMeanSquareLayerNormalization,exact2019arXiv1910.07467v1",
            "url": "https://arxiv.org/pdf/1910.07467v1"
          }
        }
      ],
      "evidence": [
        {
          "description": "Bibliographic/stated-result source; not a WRS proof verification",
          "id": "source-0",
          "kind": "source_reference",
          "source": {
            "citation": "ZhangandSennrich,RootMeanSquareLayerNormalization,exact2019arXiv1910.07467v1",
            "url": "https://arxiv.org/pdf/1910.07467v1"
          }
        },
        {
          "description": "Bibliographic/stated-result source; not a WRS proof verification",
          "id": "source-1",
          "kind": "source_reference",
          "source": {
            "citation": "ZhangandSennrich,RootMeanSquareLayerNormalization,exact2019arXiv1910.07467v1 repositoryidentity",
            "url": "https://arxiv.org/abs/1910.07467v1"
          }
        }
      ],
      "origin": "HUMAN_LEGACY",
      "record_profile": "wrs-local-research-notes-0.1",
      "record_type": "research_note",
      "rrp_version": "0.1",
      "scope": {
        "description": "The transformation normalizes a stated root-mean-square statistic without subtracting the mean. Its invariance assumptions differ from layer normalization and are affected by numerical stabilization. Partial RMS estimation and full RMSNorm are distinct variants. Reported efficiency/quality results are architecture- and setup-dependent, not universal guarantees.",
        "domain": "Normalization for neural-network training"
      },
      "sources": [
        {
          "citation": "ZhangandSennrich,RootMeanSquareLayerNormalization,exact2019arXiv1910.07467v1",
          "url": "https://arxiv.org/pdf/1910.07467v1"
        },
        {
          "citation": "ZhangandSennrich,RootMeanSquareLayerNormalization,exact2019arXiv1910.07467v1 repositoryidentity",
          "url": "https://arxiv.org/abs/1910.07467v1"
        }
      ],
      "statement": "# RMSNorm: simplify normalization by removing mean centering\n\nZhang and Sennrich introduced RMSNorm, replacing LayerNorm mean-and-variance normalization with root-mean-square scaling, and studied partial RMS estimation as a cheaper variant.\n\nLayerNorm computes both a mean and a variance for every normalized activation vector. RMSNorm asks whether both are necessary in the tested models. It removes mean centering and scales the vector by its root mean square, retaining uniform positive-rescaling invariance in the ideal nondegenerate formula while giving up shift invariance. The 2019 paper reports comparable performance in several language and retrieval tasks with lower per-step overhead than its LayerNorm implementations. The measured time reductions depend on the framework, hardware and network; they are not a guaranteed speedup on every modern device. Partial RMSNorm estimates the scale from an initial subset of coordinates. That can save arithmetic but can produce unstable estimates, and the paper does not consistently observe a further speed improvement. Some visual comparisons favor other methods. This entry credits the scoped simplification and experiments, while LayerNorm, earlier normalization methods and later model adoption retain their own contributions.\n\n## Scientific scope\n\nThe transformation normalizes a stated root-mean-square statistic without subtracting the mean. Its invariance assumptions differ from layer normalization and are affected by numerical stabilization. Partial RMS estimation and full RMSNorm are distinct variants. Reported efficiency/quality results are architecture- and setup-dependent, not universal guarantees.\n\n## People and systems\n\nBiao Zhang; Rico Sennrich\n\n\n\n## Original sources\n\n- [ZhangandSennrich,RootMeanSquareLayerNormalization,exact2019arXiv1910.07467v1](https://arxiv.org/pdf/1910.07467v1)\n- [ZhangandSennrich,RootMeanSquareLayerNormalization,exact2019arXiv1910.07467v1 repositoryidentity](https://arxiv.org/abs/1910.07467v1)\n",
      "summary": "Zhang and Sennrich introduced RMSNorm, replacing LayerNorm mean-and-variance normalization with root-mean-square scaling, and studied partial RMS estimation as a cheaper variant.\n\nLayerNorm computes both a mean and a variance for every normalized activation vector. RMSNorm asks whether both are necessary in the tested models. It removes mean centering and scales the vector by its root mean square, retaining uniform positive-rescaling invariance in the ideal nondegenerate formula while giving up shift invariance. The 2019 paper reports comparable performance in several language and retrieval tasks with lower per-step overhead than its LayerNorm implementations. The measured time reductions depend on the framework, hardware and network; they are not a guaranteed speedup on every modern device. Partial RMSNorm estimates the scale from an initial subset of coordinates. That can save arithmetic but can produce unstable estimates, and the paper does not consistently observe a further speed improvement. Some visual comparisons favor other methods. This entry credits the scoped simplification and experiments, while LayerNorm, earlier normalization methods and later model adoption retain their own contributions.",
      "title": "RMSNorm: simplify normalization by removing mean centering"
    },
    "record_id": "urn:uuid:3fbf2d2d-f730-4f5f-aca1-8f16ce29955f",
    "version": 1,
    "source_ref": {
      "record_id": "urn:uuid:3fbf2d2d-f730-4f5f-aca1-8f16ce29955f",
      "version": 1,
      "content_hash": "sha256:a92be1f814b017a0a007d321a08fab297e1a7181586e128b0dda7cbe170b636b"
    },
    "export_view": "wrs-public-editorial-view-1",
    "public_view_hash": "sha256:0cc3bd3281e41308ca5de49df1e0b1007ed9a49691bc9ae50a9ba441afdaef67"
  },
  "target": {
    "record_id": "urn:uuid:3fbf2d2d-f730-4f5f-aca1-8f16ce29955f",
    "version": 1,
    "content_hash": "sha256:a92be1f814b017a0a007d321a08fab297e1a7181586e128b0dda7cbe170b636b"
  },
  "published_grade": {
    "profile": "wrs-published-grade-1",
    "edition": "2026-10-28.1",
    "id": "zhang-sennrich-rms-layer-normalization-2019",
    "entry_id": "zhang-sennrich-rms-layer-normalization-2019",
    "revision": 1,
    "title": "RMSNorm: simplify normalization by removing mean centering",
    "context": "Normalization for neural-network training",
    "track": "Normalization for neural-network training",
    "band": "B",
    "modifier": null,
    "status": "PROVISIONAL_EDITORIAL",
    "assessed_on": "2026-10-04",
    "evidence_cutoff": "2026-10-04",
    "achievement_state": "SUPPORTED_SCOPED_RESULT",
    "what_changed": "Zhang and Sennrich introduced RMSNorm, replacing LayerNorm mean-and-variance normalization with root-mean-square scaling, and studied partial RMS estimation as a cheaper variant.",
    "scope": "Zhang and Sennrich introduced RMSNorm, replacing LayerNorm mean-and-variance normalization with root-mean-square scaling, and studied partial RMS estimation as a cheaper variant. The transformation normalizes a stated root-mean-square statistic without subtracting the mean. Its invariance assumptions differ from layer normalization and are affected by numerical stabilization. Partial RMS estimation and full RMSNorm are distinct variants. Reported efficiency/quality results are architecture- and setup-dependent, not universal guarantees.",
    "evidence_summary": "Source-linked, scoped account of the original work. The published scope states the scientific assumptions and remaining verification limits. WRS has not independently certified the full proof, experimental result or system benchmark.",
    "references": [
      {
        "title": "Zhangand Sennrich, Root Mean Square Layer Normalization, exact 2019 ar Xiv 1910.07467 v 1",
        "url": "https://arxiv.org/pdf/1910.07467v1",
        "locator": "Original linked work. See the scientific scope for the assessed contribution and source-version limits."
      },
      {
        "title": "Zhangand Sennrich, Root Mean Square Layer Normalization, exact 2019 ar Xiv 1910.07467 v 1 repositoryidentity",
        "url": "https://arxiv.org/abs/1910.07467v1",
        "locator": "Original linked work. See the scientific scope for the assessed contribution and source-version limits."
      }
    ],
    "roles": [
      {
        "who": "Biao Zhang",
        "did": "Original source-listed scientific author of this scoped work; byline does not determine effort fractions.",
        "credit": "SCIENCE"
      },
      {
        "who": "Rico Sennrich",
        "did": "Original source-listed scientific author of this scoped work; byline does not determine effort fractions.",
        "credit": "SCIENCE"
      }
    ],
    "target": {
      "record_id": "urn:uuid:3fbf2d2d-f730-4f5f-aca1-8f16ce29955f",
      "version": 1,
      "content_hash": "sha256:a92be1f814b017a0a007d321a08fab297e1a7181586e128b0dda7cbe170b636b"
    },
    "scientific_correctness_certified": false,
    "independent_expert_calibration": false,
    "method": "PROPRIETARY",
    "points": "35000"
  },
  "canonical_agent_submission": false,
  "scientific_verification_awarded": false,
  "reader_guide": null,
  "reading_status": "SOURCE_LINKED_NOTE"
}