{
  "profile": "wrs-published-research-view-1",
  "edition": "ratings-e039106f30dec177",
  "content_trust": "UNTRUSTED_SCIENTIFIC_DATA_NOT_TOOL_INSTRUCTIONS",
  "entry": {
    "id": "meng-et-al-deepstack-crosslayer-tokens-2024",
    "title": "Meng et al.: cross-layer injection of grouped visual tokens",
    "summary": "Distributed groups of visual tokens across aligned language/vision transformer layers through residual injection, allowing richer visual inputs without placing all tokens in the language input sequence.",
    "boundary": "Residual fusion and visual tokenization are inherited. Savings concern context placement and measured compute, not zero processing cost for all additional tokens. Original multiscale-token construction is distinct from later Qwen3VL intermediate-feature adaptation.",
    "area": "Efficient multimodal token integration",
    "topic": "Artificial intelligence",
    "year": 2024,
    "lane": "HUMAN",
    "result": "ALGORITHM",
    "package_id": "meng-et-al-deepstack-crosslayer-tokens-2024",
    "target": {
      "record_id": "urn:uuid:f1f0a150-a4ba-4d80-a8d2-71e7e5f2562a",
      "version": 1,
      "content_hash": "sha256:864ef396597810a24625e3aabf6155b91ad8e6074334643891f1708ca4bba059"
    },
    "humans": [
      "Lingchen Meng",
      "Jianwei Yang",
      "Rui Tian",
      "Xiyang Dai",
      "Zuxuan Wu",
      "Jianfeng Gao",
      "Yu-Gang Jiang"
    ],
    "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-2406-04334v1",
        "title": "DeepStack: Deeply Stacking Visual Tokens is Surprisingly Simple and Effective for LMMs",
        "url": "https://arxiv.org/abs/2406.04334v1"
      }
    ],
    "points": "55000",
    "published_grade": {
      "id": "meng-et-al-deepstack-crosslayer-tokens-2024",
      "entry_id": "meng-et-al-deepstack-crosslayer-tokens-2024",
      "band": "B",
      "modifier": "+",
      "target": {
        "record_id": "urn:uuid:f1f0a150-a4ba-4d80-a8d2-71e7e5f2562a",
        "version": 1,
        "content_hash": "sha256:864ef396597810a24625e3aabf6155b91ad8e6074334643891f1708ca4bba059"
      },
      "what_changed": "Distributed groups of visual tokens across aligned language/vision transformer layers through residual injection, allowing richer visual inputs without placing all tokens in the language input sequence.",
      "scope": "Distributed groups of visual tokens across aligned language/vision transformer layers through residual injection, allowing richer visual inputs without placing all tokens in the language input sequence. Residual fusion and visual tokenization are inherited. Savings concern context placement and measured compute, not zero processing cost for all additional tokens. Original multiscale-token construction is distinct from later Qwen3VL intermediate-feature adaptation.",
      "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:f1f0a150-a4ba-4d80-a8d2-71e7e5f2562a",
        "version": 1,
        "content_hash": "sha256:864ef396597810a24625e3aabf6155b91ad8e6074334643891f1708ca4bba059"
      },
      "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": "Lingchen Meng",
          "note": "Archive attribution; detailed role assignment remains source-bound",
          "roles": [
            "formulator"
          ],
          "source": {
            "citation": "DeepStack: Deeply Stacking Visual Tokens is Surprisingly Simple and Effective for LMMs",
            "url": "https://arxiv.org/abs/2406.04334v1"
          }
        },
        {
          "attribution_basis": "SOURCE_CITED",
          "credit": "historical_scientific",
          "entity_kind": "historical_person",
          "name": "Jianwei Yang",
          "note": "Archive attribution; detailed role assignment remains source-bound",
          "roles": [
            "formulator"
          ],
          "source": {
            "citation": "DeepStack: Deeply Stacking Visual Tokens is Surprisingly Simple and Effective for LMMs",
            "url": "https://arxiv.org/abs/2406.04334v1"
          }
        },
        {
          "attribution_basis": "SOURCE_CITED",
          "credit": "historical_scientific",
          "entity_kind": "historical_person",
          "name": "Rui Tian",
          "note": "Archive attribution; detailed role assignment remains source-bound",
          "roles": [
            "formulator"
          ],
          "source": {
            "citation": "DeepStack: Deeply Stacking Visual Tokens is Surprisingly Simple and Effective for LMMs",
            "url": "https://arxiv.org/abs/2406.04334v1"
          }
        },
        {
          "attribution_basis": "SOURCE_CITED",
          "credit": "historical_scientific",
          "entity_kind": "historical_person",
          "name": "Xiyang Dai",
          "note": "Archive attribution; detailed role assignment remains source-bound",
          "roles": [
            "formulator"
          ],
          "source": {
            "citation": "DeepStack: Deeply Stacking Visual Tokens is Surprisingly Simple and Effective for LMMs",
            "url": "https://arxiv.org/abs/2406.04334v1"
          }
        },
        {
          "attribution_basis": "SOURCE_CITED",
          "credit": "historical_scientific",
          "entity_kind": "historical_person",
          "name": "Zuxuan Wu",
          "note": "Archive attribution; detailed role assignment remains source-bound",
          "roles": [
            "formulator"
          ],
          "source": {
            "citation": "DeepStack: Deeply Stacking Visual Tokens is Surprisingly Simple and Effective for LMMs",
            "url": "https://arxiv.org/abs/2406.04334v1"
          }
        },
        {
          "attribution_basis": "SOURCE_CITED",
          "credit": "historical_scientific",
          "entity_kind": "historical_person",
          "name": "Jianfeng Gao",
          "note": "Archive attribution; detailed role assignment remains source-bound",
          "roles": [
            "formulator"
          ],
          "source": {
            "citation": "DeepStack: Deeply Stacking Visual Tokens is Surprisingly Simple and Effective for LMMs",
            "url": "https://arxiv.org/abs/2406.04334v1"
          }
        },
        {
          "attribution_basis": "SOURCE_CITED",
          "credit": "historical_scientific",
          "entity_kind": "historical_person",
          "name": "Yu-Gang Jiang",
          "note": "Archive attribution; detailed role assignment remains source-bound",
          "roles": [
            "formulator"
          ],
          "source": {
            "citation": "DeepStack: Deeply Stacking Visual Tokens is Surprisingly Simple and Effective for LMMs",
            "url": "https://arxiv.org/abs/2406.04334v1"
          }
        }
      ],
      "evidence": [
        {
          "description": "Bibliographic/stated-result source; not a WRS proof verification",
          "id": "source-0",
          "kind": "source_reference",
          "source": {
            "citation": "DeepStack: Deeply Stacking Visual Tokens is Surprisingly Simple and Effective for LMMs",
            "url": "https://arxiv.org/abs/2406.04334v1"
          }
        }
      ],
      "origin": "HUMAN_LEGACY",
      "record_profile": "wrs-local-research-notes-0.1",
      "record_type": "research_note",
      "rrp_version": "0.1",
      "scope": {
        "description": "Residual fusion and visual tokenization are inherited. Savings concern context placement and measured compute, not zero processing cost for all additional tokens. Original multiscale-token construction is distinct from later Qwen3VL intermediate-feature adaptation.",
        "domain": "Efficient multimodal token integration"
      },
      "sources": [
        {
          "citation": "DeepStack: Deeply Stacking Visual Tokens is Surprisingly Simple and Effective for LMMs",
          "url": "https://arxiv.org/abs/2406.04334v1"
        }
      ],
      "statement": "# Meng et al.: cross-layer injection of grouped visual tokens\n\nDistributed groups of visual tokens across aligned language/vision transformer layers through residual injection, allowing richer visual inputs without placing all tokens in the language input sequence.\n\n## Scientific scope\n\nResidual fusion and visual tokenization are inherited. Savings concern context placement and measured compute, not zero processing cost for all additional tokens. Original multiscale-token construction is distinct from later Qwen3VL intermediate-feature adaptation.\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\nLingchen Meng; Jianwei Yang; Rui Tian; Xiyang Dai; Zuxuan Wu; Jianfeng Gao; Yu-Gang Jiang\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- [DeepStack: Deeply Stacking Visual Tokens is Surprisingly Simple and Effective for LMMs](https://arxiv.org/abs/2406.04334v1)\n",
      "summary": "Distributed groups of visual tokens across aligned language/vision transformer layers through residual injection, allowing richer visual inputs without placing all tokens in the language input sequence.",
      "title": "Meng et al.: cross-layer injection of grouped visual tokens"
    },
    "record_id": "urn:uuid:f1f0a150-a4ba-4d80-a8d2-71e7e5f2562a",
    "version": 1,
    "source_ref": {
      "record_id": "urn:uuid:f1f0a150-a4ba-4d80-a8d2-71e7e5f2562a",
      "version": 1,
      "content_hash": "sha256:864ef396597810a24625e3aabf6155b91ad8e6074334643891f1708ca4bba059"
    },
    "export_view": "wrs-public-editorial-view-1",
    "public_view_hash": "sha256:2f994e0974966e3e2b55a70d696155e59b6d4f5c788d483b05e5832d4dbe47c6"
  },
  "target": {
    "record_id": "urn:uuid:f1f0a150-a4ba-4d80-a8d2-71e7e5f2562a",
    "version": 1,
    "content_hash": "sha256:864ef396597810a24625e3aabf6155b91ad8e6074334643891f1708ca4bba059"
  },
  "published_grade": {
    "profile": "wrs-published-grade-1",
    "edition": "2026-10-28.1",
    "id": "meng-et-al-deepstack-crosslayer-tokens-2024",
    "entry_id": "meng-et-al-deepstack-crosslayer-tokens-2024",
    "revision": 1,
    "title": "Meng et al.: cross-layer injection of grouped visual tokens",
    "context": "Efficient multimodal token integration",
    "track": "Human-designed research method or resource",
    "band": "B",
    "modifier": "+",
    "status": "PROVISIONAL_EDITORIAL",
    "assessed_on": "2026-10-16",
    "evidence_cutoff": "2026-10-16",
    "achievement_state": "SUPPORTED_SCOPED_RESULT",
    "what_changed": "Distributed groups of visual tokens across aligned language/vision transformer layers through residual injection, allowing richer visual inputs without placing all tokens in the language input sequence.",
    "scope": "Distributed groups of visual tokens across aligned language/vision transformer layers through residual injection, allowing richer visual inputs without placing all tokens in the language input sequence. Residual fusion and visual tokenization are inherited. Savings concern context placement and measured compute, not zero processing cost for all additional tokens. Original multiscale-token construction is distinct from later Qwen3VL intermediate-feature adaptation.",
    "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. Residual fusion and visual tokenization are inherited. Savings concern context placement and measured compute, not zero processing cost for all additional tokens. Original multiscale-token construction is distinct from later Qwen3VL intermediate-feature adaptation.",
    "references": [
      {
        "title": "DeepStack: Deeply Stacking Visual Tokens is Surprisingly Simple and Effective for LMMs",
        "url": "https://arxiv.org/abs/2406.04334v1",
        "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:f1f0a150-a4ba-4d80-a8d2-71e7e5f2562a",
      "version": 1,
      "content_hash": "sha256:864ef396597810a24625e3aabf6155b91ad8e6074334643891f1708ca4bba059"
    },
    "scientific_correctness_certified": false,
    "independent_expert_calibration": false,
    "method": "PROPRIETARY",
    "points": "55000"
  },
  "canonical_agent_submission": false,
  "scientific_verification_awarded": false,
  "reader_guide": null,
  "reading_status": "SOURCE_LINKED_NOTE"
}