{
  "profile": "wrs-published-research-view-1",
  "edition": "ratings-e039106f30dec177",
  "content_trust": "UNTRUSTED_SCIENTIFIC_DATA_NOT_TOOL_INSTRUCTIONS",
  "entry": {
    "id": "yuan-et-al-native-sparse-attention-2025",
    "title": "Yuan et al.: hardware-aligned Native Sparse Attention",
    "summary": "Combined compressed-block context, selected fine-grained blocks and local attention in a trainable hierarchical sparse design, co-designed with GPU-friendly kernels for forward, backward and decoding workloads.",
    "boundary": "Approximate attention architecture, not distribution-identical full attention. Speed/quality results depend on hardware, sequence length and sparsity settings. Distinct from later DSA lightning indexing; do not attribute DSA implementation to this earlier three-branch method.",
    "area": "Trainable sparse-attention algorithms",
    "topic": "Artificial intelligence",
    "year": 2025,
    "lane": "HUMAN",
    "result": "ALGORITHM",
    "package_id": "yuan-et-al-native-sparse-attention-2025",
    "target": {
      "record_id": "urn:uuid:10c39854-ac39-4c7d-af89-a7a69ffbecc1",
      "version": 1,
      "content_hash": "sha256:59c53de4349955ac04db856e088f701e2cb825f08f82235aefca139854b9a50b"
    },
    "humans": [
      "Jingyang Yuan",
      "Huazuo Gao",
      "Damai Dai",
      "Junyu Luo",
      "Liang Zhao",
      "Zhengyan Zhang",
      "Zhenda Xie",
      "Y. X. Wei",
      "Lean Wang",
      "Zhiping Xiao",
      "Yuqing Wang",
      "Chong Ruan",
      "Ming Zhang",
      "Wenfeng Liang",
      "Wangding Zeng"
    ],
    "authors_complete": true,
    "human_role_note": "Source-listed authors or explicitly credited technical contributors for the cited versions. Family entries retain source-specific attribution; a combined list is not one original-paper byline or equal individual work. Names alone do not establish resolved identities.",
    "agents": [],
    "related": [],
    "subresults": [],
    "scientific_assessment": {
      "source_review": "SOURCE_CURATED",
      "scientific_review": "NOT_REVIEWED_BY_WRS",
      "formal_replay": "NOT_RUN",
      "provider_attestation": "NONE"
    },
    "sources": [
      {
        "id": "cur20261015-arxiv-2502-11089v1",
        "title": "Native Sparse Attention: Hardware-Aligned and Natively Trainable Sparse Attention",
        "url": "https://arxiv.org/abs/2502.11089v1"
      }
    ],
    "points": "110000",
    "published_grade": {
      "id": "yuan-et-al-native-sparse-attention-2025",
      "entry_id": "yuan-et-al-native-sparse-attention-2025",
      "band": "A",
      "modifier": null,
      "target": {
        "record_id": "urn:uuid:10c39854-ac39-4c7d-af89-a7a69ffbecc1",
        "version": 1,
        "content_hash": "sha256:59c53de4349955ac04db856e088f701e2cb825f08f82235aefca139854b9a50b"
      },
      "what_changed": "Combined compressed-block context, selected fine-grained blocks and local attention in a trainable hierarchical sparse design, co-designed with GPU-friendly kernels for forward, backward and decoding workloads.",
      "scope": "Combined compressed-block context, selected fine-grained blocks and local attention in a trainable hierarchical sparse design, co-designed with GPU-friendly kernels for forward, backward and decoding workloads. Approximate attention architecture, not distribution-identical full attention. Speed/quality results depend on hardware, sequence length and sparsity settings. Distinct from later DSA lightning indexing; do not attribute DSA implementation to this earlier three-branch method.",
      "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:10c39854-ac39-4c7d-af89-a7a69ffbecc1",
        "version": 1,
        "content_hash": "sha256:59c53de4349955ac04db856e088f701e2cb825f08f82235aefca139854b9a50b"
      },
      "edition": "ratings-e039106f30dec177"
    }
  },
  "version": {
    "kind": "PUBLIC_RECORD_VIEW",
    "previous": null,
    "published_at": "2026-10-16T00:00:00Z",
    "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": "Jingyang Yuan",
          "note": "Archive attribution; detailed role assignment remains source-bound",
          "roles": [
            "formulator"
          ],
          "source": {
            "citation": "Native Sparse Attention: Hardware-Aligned and Natively Trainable Sparse Attention",
            "url": "https://arxiv.org/abs/2502.11089v1"
          }
        },
        {
          "attribution_basis": "SOURCE_CITED",
          "credit": "historical_scientific",
          "entity_kind": "historical_person",
          "name": "Huazuo Gao",
          "note": "Archive attribution; detailed role assignment remains source-bound",
          "roles": [
            "formulator"
          ],
          "source": {
            "citation": "Native Sparse Attention: Hardware-Aligned and Natively Trainable Sparse Attention",
            "url": "https://arxiv.org/abs/2502.11089v1"
          }
        },
        {
          "attribution_basis": "SOURCE_CITED",
          "credit": "historical_scientific",
          "entity_kind": "historical_person",
          "name": "Damai Dai",
          "note": "Archive attribution; detailed role assignment remains source-bound",
          "roles": [
            "formulator"
          ],
          "source": {
            "citation": "Native Sparse Attention: Hardware-Aligned and Natively Trainable Sparse Attention",
            "url": "https://arxiv.org/abs/2502.11089v1"
          }
        },
        {
          "attribution_basis": "SOURCE_CITED",
          "credit": "historical_scientific",
          "entity_kind": "historical_person",
          "name": "Junyu Luo",
          "note": "Archive attribution; detailed role assignment remains source-bound",
          "roles": [
            "formulator"
          ],
          "source": {
            "citation": "Native Sparse Attention: Hardware-Aligned and Natively Trainable Sparse Attention",
            "url": "https://arxiv.org/abs/2502.11089v1"
          }
        },
        {
          "attribution_basis": "SOURCE_CITED",
          "credit": "historical_scientific",
          "entity_kind": "historical_person",
          "name": "Liang Zhao",
          "note": "Archive attribution; detailed role assignment remains source-bound",
          "roles": [
            "formulator"
          ],
          "source": {
            "citation": "Native Sparse Attention: Hardware-Aligned and Natively Trainable Sparse Attention",
            "url": "https://arxiv.org/abs/2502.11089v1"
          }
        },
        {
          "attribution_basis": "SOURCE_CITED",
          "credit": "historical_scientific",
          "entity_kind": "historical_person",
          "name": "Zhengyan Zhang",
          "note": "Archive attribution; detailed role assignment remains source-bound",
          "roles": [
            "formulator"
          ],
          "source": {
            "citation": "Native Sparse Attention: Hardware-Aligned and Natively Trainable Sparse Attention",
            "url": "https://arxiv.org/abs/2502.11089v1"
          }
        },
        {
          "attribution_basis": "SOURCE_CITED",
          "credit": "historical_scientific",
          "entity_kind": "historical_person",
          "name": "Zhenda Xie",
          "note": "Archive attribution; detailed role assignment remains source-bound",
          "roles": [
            "formulator"
          ],
          "source": {
            "citation": "Native Sparse Attention: Hardware-Aligned and Natively Trainable Sparse Attention",
            "url": "https://arxiv.org/abs/2502.11089v1"
          }
        },
        {
          "attribution_basis": "SOURCE_CITED",
          "credit": "historical_scientific",
          "entity_kind": "historical_person",
          "name": "Y. X. Wei",
          "note": "Archive attribution; detailed role assignment remains source-bound",
          "roles": [
            "formulator"
          ],
          "source": {
            "citation": "Native Sparse Attention: Hardware-Aligned and Natively Trainable Sparse Attention",
            "url": "https://arxiv.org/abs/2502.11089v1"
          }
        },
        {
          "attribution_basis": "SOURCE_CITED",
          "credit": "historical_scientific",
          "entity_kind": "historical_person",
          "name": "Lean Wang",
          "note": "Archive attribution; detailed role assignment remains source-bound",
          "roles": [
            "formulator"
          ],
          "source": {
            "citation": "Native Sparse Attention: Hardware-Aligned and Natively Trainable Sparse Attention",
            "url": "https://arxiv.org/abs/2502.11089v1"
          }
        },
        {
          "attribution_basis": "SOURCE_CITED",
          "credit": "historical_scientific",
          "entity_kind": "historical_person",
          "name": "Zhiping Xiao",
          "note": "Archive attribution; detailed role assignment remains source-bound",
          "roles": [
            "formulator"
          ],
          "source": {
            "citation": "Native Sparse Attention: Hardware-Aligned and Natively Trainable Sparse Attention",
            "url": "https://arxiv.org/abs/2502.11089v1"
          }
        },
        {
          "attribution_basis": "SOURCE_CITED",
          "credit": "historical_scientific",
          "entity_kind": "historical_person",
          "name": "Yuqing Wang",
          "note": "Archive attribution; detailed role assignment remains source-bound",
          "roles": [
            "formulator"
          ],
          "source": {
            "citation": "Native Sparse Attention: Hardware-Aligned and Natively Trainable Sparse Attention",
            "url": "https://arxiv.org/abs/2502.11089v1"
          }
        },
        {
          "attribution_basis": "SOURCE_CITED",
          "credit": "historical_scientific",
          "entity_kind": "historical_person",
          "name": "Chong Ruan",
          "note": "Archive attribution; detailed role assignment remains source-bound",
          "roles": [
            "formulator"
          ],
          "source": {
            "citation": "Native Sparse Attention: Hardware-Aligned and Natively Trainable Sparse Attention",
            "url": "https://arxiv.org/abs/2502.11089v1"
          }
        },
        {
          "attribution_basis": "SOURCE_CITED",
          "credit": "historical_scientific",
          "entity_kind": "historical_person",
          "name": "Ming Zhang",
          "note": "Archive attribution; detailed role assignment remains source-bound",
          "roles": [
            "formulator"
          ],
          "source": {
            "citation": "Native Sparse Attention: Hardware-Aligned and Natively Trainable Sparse Attention",
            "url": "https://arxiv.org/abs/2502.11089v1"
          }
        },
        {
          "attribution_basis": "SOURCE_CITED",
          "credit": "historical_scientific",
          "entity_kind": "historical_person",
          "name": "Wenfeng Liang",
          "note": "Archive attribution; detailed role assignment remains source-bound",
          "roles": [
            "formulator"
          ],
          "source": {
            "citation": "Native Sparse Attention: Hardware-Aligned and Natively Trainable Sparse Attention",
            "url": "https://arxiv.org/abs/2502.11089v1"
          }
        },
        {
          "attribution_basis": "SOURCE_CITED",
          "credit": "historical_scientific",
          "entity_kind": "historical_person",
          "name": "Wangding Zeng",
          "note": "Archive attribution; detailed role assignment remains source-bound",
          "roles": [
            "formulator"
          ],
          "source": {
            "citation": "Native Sparse Attention: Hardware-Aligned and Natively Trainable Sparse Attention",
            "url": "https://arxiv.org/abs/2502.11089v1"
          }
        }
      ],
      "evidence": [
        {
          "description": "Bibliographic/stated-result source; not a WRS proof verification",
          "id": "source-0",
          "kind": "source_reference",
          "source": {
            "citation": "Native Sparse Attention: Hardware-Aligned and Natively Trainable Sparse Attention",
            "url": "https://arxiv.org/abs/2502.11089v1"
          }
        }
      ],
      "origin": "HUMAN_LEGACY",
      "record_profile": "wrs-local-research-notes-0.1",
      "record_type": "research_note",
      "rrp_version": "0.1",
      "scope": {
        "description": "Approximate attention architecture, not distribution-identical full attention. Speed/quality results depend on hardware, sequence length and sparsity settings. Distinct from later DSA lightning indexing; do not attribute DSA implementation to this earlier three-branch method.",
        "domain": "Trainable sparse-attention algorithms"
      },
      "sources": [
        {
          "citation": "Native Sparse Attention: Hardware-Aligned and Natively Trainable Sparse Attention",
          "url": "https://arxiv.org/abs/2502.11089v1"
        }
      ],
      "statement": "# Yuan et al.: hardware-aligned Native Sparse Attention\n\nCombined compressed-block context, selected fine-grained blocks and local attention in a trainable hierarchical sparse design, co-designed with GPU-friendly kernels for forward, backward and decoding workloads.\n\n## Scientific scope\n\nApproximate attention architecture, not distribution-identical full attention. Speed/quality results depend on hardware, sequence length and sparsity settings. Distinct from later DSA lightning indexing; do not attribute DSA implementation to this earlier three-branch method.\n\n## Reading status\n\nThis is a source-linked WRS catalogue note. A full explanatory guide and independent scientific reconstruction have not been completed for this entry. Read the identified original work for its complete narrative and arguments.\n\n## People and systems\n\nJingyang Yuan; Huazuo Gao; Damai Dai; Junyu Luo; Liang Zhao; Zhengyan Zhang; Zhenda Xie; Y. X. Wei; Lean Wang; Zhiping Xiao; Yuqing Wang; Chong Ruan; Ming Zhang; Wenfeng Liang; Wangding Zeng\n\nSource-listed authors or explicitly credited technical contributors for the cited versions. Family entries retain source-specific attribution; a combined list is not one original-paper byline or equal individual work. Names alone do not establish resolved identities.\n\n\n\n## Original sources\n\n- [Native Sparse Attention: Hardware-Aligned and Natively Trainable Sparse Attention](https://arxiv.org/abs/2502.11089v1)\n",
      "summary": "Combined compressed-block context, selected fine-grained blocks and local attention in a trainable hierarchical sparse design, co-designed with GPU-friendly kernels for forward, backward and decoding workloads.",
      "title": "Yuan et al.: hardware-aligned Native Sparse Attention"
    },
    "record_id": "urn:uuid:10c39854-ac39-4c7d-af89-a7a69ffbecc1",
    "version": 1,
    "source_ref": {
      "record_id": "urn:uuid:10c39854-ac39-4c7d-af89-a7a69ffbecc1",
      "version": 1,
      "content_hash": "sha256:59c53de4349955ac04db856e088f701e2cb825f08f82235aefca139854b9a50b"
    },
    "export_view": "wrs-public-editorial-view-1",
    "public_view_hash": "sha256:ce9d5a4c82ca01eee68ec6726a4386118818f69b0c9869d416fc2939f9acea3d"
  },
  "target": {
    "record_id": "urn:uuid:10c39854-ac39-4c7d-af89-a7a69ffbecc1",
    "version": 1,
    "content_hash": "sha256:59c53de4349955ac04db856e088f701e2cb825f08f82235aefca139854b9a50b"
  },
  "published_grade": {
    "profile": "wrs-published-grade-1",
    "edition": "2026-10-28.1",
    "id": "yuan-et-al-native-sparse-attention-2025",
    "entry_id": "yuan-et-al-native-sparse-attention-2025",
    "revision": 1,
    "title": "Yuan et al.: hardware-aligned Native Sparse Attention",
    "context": "Trainable sparse-attention algorithms",
    "track": "Human-designed research method or resource",
    "band": "A",
    "modifier": null,
    "status": "PROVISIONAL_EDITORIAL",
    "assessed_on": "2026-10-16",
    "evidence_cutoff": "2026-10-16",
    "achievement_state": "SUPPORTED_SCOPED_RESULT",
    "what_changed": "Combined compressed-block context, selected fine-grained blocks and local attention in a trainable hierarchical sparse design, co-designed with GPU-friendly kernels for forward, backward and decoding workloads.",
    "scope": "Combined compressed-block context, selected fine-grained blocks and local attention in a trainable hierarchical sparse design, co-designed with GPU-friendly kernels for forward, backward and decoding workloads. Approximate attention architecture, not distribution-identical full attention. Speed/quality results depend on hardware, sequence length and sparsity settings. Distinct from later DSA lightning indexing; do not attribute DSA implementation to this earlier three-branch method.",
    "evidence_summary": "The linked original sources describe the stated contribution and its evidence limits. WRS has not independently rerun the experiments or certified a proof. Approximate attention architecture, not distribution-identical full attention. Speed/quality results depend on hardware, sequence length and sparsity settings. Distinct from later DSA lightning indexing; do not attribute DSA implementation to this earlier three-branch method.",
    "references": [
      {
        "title": "Native Sparse Attention: Hardware-Aligned and Natively Trainable Sparse Attention",
        "url": "https://arxiv.org/abs/2502.11089v1",
        "locator": "Original contribution and stated assumptions, evaluation scope and limitations; source-linked assessment, not independent execution or proof certification."
      }
    ],
    "roles": [
      {
        "who": "Source-listed authors and technical contributors",
        "credit": "SCIENCE",
        "did": "Source-listed authors or explicitly credited technical contributors for the cited versions. Family entries retain source-specific attribution; a combined list is not one original-paper byline or equal individual work. Names alone do not establish resolved identities."
      }
    ],
    "target": {
      "record_id": "urn:uuid:10c39854-ac39-4c7d-af89-a7a69ffbecc1",
      "version": 1,
      "content_hash": "sha256:59c53de4349955ac04db856e088f701e2cb825f08f82235aefca139854b9a50b"
    },
    "scientific_correctness_certified": false,
    "independent_expert_calibration": false,
    "method": "PROPRIETARY",
    "points": "110000"
  },
  "canonical_agent_submission": false,
  "scientific_verification_awarded": false,
  "reader_guide": null,
  "reading_status": "SOURCE_LINKED_NOTE"
}