{
  "profile": "wrs-published-research-view-1",
  "edition": "ratings-e039106f30dec177",
  "content_trust": "UNTRUSTED_SCIENTIFIC_DATA_NOT_TOOL_INSTRUCTIONS",
  "entry": {
    "id": "fu-et-al-neuron-distributed-training-2024",
    "title": "AWS Neuron distributed LLM training on Trainium",
    "summary": "Integrated inherited parallel-training methods with Trainium-aware mixed tensor-parallel degrees, automatic pipeline graph partitioning, delayed device materialization and asynchronous graph execution in Neuron/NxDT.",
    "boundary": "Not the invention of systolic arrays, tensor/pipeline/data parallelism or ZeRO. Some features require model-code changes; symbolic tracing excludes unresolved dynamic control flow and can require opaque module wrappers. Graph lookahead consumes host memory. Historical instance-cost/performance comparisons are selected-workload results, not current prices or universal superiority.",
    "area": "AI training systems",
    "topic": "Artificial intelligence",
    "year": 2024,
    "lane": "HUMAN",
    "result": "EMPIRICAL",
    "package_id": "fu-et-al-neuron-distributed-training-2024",
    "target": {
      "record_id": "urn:uuid:55e55737-f34f-4944-a02c-cadbc2706a59",
      "version": 1,
      "content_hash": "sha256:6e465b9f5427b80fa666dbc78b8ca5bfd600988e9a49e12de7b6ba6e49639a76"
    },
    "humans": [
      "Xinwei Fu",
      "Zhen Zhang",
      "Haozheng Fan",
      "Guangtai Huang",
      "Mohammad El-Shabani",
      "Randy Huang",
      "Rahul Solanki",
      "Fei Wu",
      "Ron Diamant",
      "Yida Wang"
    ],
    "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-aws-trainium-socc-2024",
        "title": "Distributed Training of Large Language Models on AWS Trainium",
        "url": "https://cdn.amazon.science/fa/fc/6c9a63824f1fa3655fc757825256/distributed-training-of-large-language-models-on-aws-trainium.pdf"
      }
    ],
    "points": "65000",
    "published_grade": {
      "id": "fu-et-al-neuron-distributed-training-2024",
      "entry_id": "fu-et-al-neuron-distributed-training-2024",
      "band": "B",
      "modifier": "+",
      "target": {
        "record_id": "urn:uuid:55e55737-f34f-4944-a02c-cadbc2706a59",
        "version": 1,
        "content_hash": "sha256:6e465b9f5427b80fa666dbc78b8ca5bfd600988e9a49e12de7b6ba6e49639a76"
      },
      "what_changed": "Integrated inherited parallel-training methods with Trainium-aware mixed tensor-parallel degrees, automatic pipeline graph partitioning, delayed device materialization and asynchronous graph execution in Neuron/NxDT.",
      "scope": "Integrated inherited parallel-training methods with Trainium-aware mixed tensor-parallel degrees, automatic pipeline graph partitioning, delayed device materialization and asynchronous graph execution in Neuron/NxDT. Not the invention of systolic arrays, tensor/pipeline/data parallelism or ZeRO. Some features require model-code changes; symbolic tracing excludes unresolved dynamic control flow and can require opaque module wrappers. Graph lookahead consumes host memory. Historical instance-cost/performance comparisons are selected-workload results, not current prices or universal superiority.",
      "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:55e55737-f34f-4944-a02c-cadbc2706a59",
        "version": 1,
        "content_hash": "sha256:6e465b9f5427b80fa666dbc78b8ca5bfd600988e9a49e12de7b6ba6e49639a76"
      },
      "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": "Xinwei Fu",
          "note": "Archive attribution; detailed role assignment remains source-bound",
          "roles": [
            "discoverer"
          ],
          "source": {
            "citation": "Distributed Training of Large Language Models on AWS Trainium",
            "url": "https://cdn.amazon.science/fa/fc/6c9a63824f1fa3655fc757825256/distributed-training-of-large-language-models-on-aws-trainium.pdf"
          }
        },
        {
          "attribution_basis": "SOURCE_CITED",
          "credit": "historical_scientific",
          "entity_kind": "historical_person",
          "name": "Zhen Zhang",
          "note": "Archive attribution; detailed role assignment remains source-bound",
          "roles": [
            "discoverer"
          ],
          "source": {
            "citation": "Distributed Training of Large Language Models on AWS Trainium",
            "url": "https://cdn.amazon.science/fa/fc/6c9a63824f1fa3655fc757825256/distributed-training-of-large-language-models-on-aws-trainium.pdf"
          }
        },
        {
          "attribution_basis": "SOURCE_CITED",
          "credit": "historical_scientific",
          "entity_kind": "historical_person",
          "name": "Haozheng Fan",
          "note": "Archive attribution; detailed role assignment remains source-bound",
          "roles": [
            "discoverer"
          ],
          "source": {
            "citation": "Distributed Training of Large Language Models on AWS Trainium",
            "url": "https://cdn.amazon.science/fa/fc/6c9a63824f1fa3655fc757825256/distributed-training-of-large-language-models-on-aws-trainium.pdf"
          }
        },
        {
          "attribution_basis": "SOURCE_CITED",
          "credit": "historical_scientific",
          "entity_kind": "historical_person",
          "name": "Guangtai Huang",
          "note": "Archive attribution; detailed role assignment remains source-bound",
          "roles": [
            "discoverer"
          ],
          "source": {
            "citation": "Distributed Training of Large Language Models on AWS Trainium",
            "url": "https://cdn.amazon.science/fa/fc/6c9a63824f1fa3655fc757825256/distributed-training-of-large-language-models-on-aws-trainium.pdf"
          }
        },
        {
          "attribution_basis": "SOURCE_CITED",
          "credit": "historical_scientific",
          "entity_kind": "historical_person",
          "name": "Mohammad El-Shabani",
          "note": "Archive attribution; detailed role assignment remains source-bound",
          "roles": [
            "discoverer"
          ],
          "source": {
            "citation": "Distributed Training of Large Language Models on AWS Trainium",
            "url": "https://cdn.amazon.science/fa/fc/6c9a63824f1fa3655fc757825256/distributed-training-of-large-language-models-on-aws-trainium.pdf"
          }
        },
        {
          "attribution_basis": "SOURCE_CITED",
          "credit": "historical_scientific",
          "entity_kind": "historical_person",
          "name": "Randy Huang",
          "note": "Archive attribution; detailed role assignment remains source-bound",
          "roles": [
            "discoverer"
          ],
          "source": {
            "citation": "Distributed Training of Large Language Models on AWS Trainium",
            "url": "https://cdn.amazon.science/fa/fc/6c9a63824f1fa3655fc757825256/distributed-training-of-large-language-models-on-aws-trainium.pdf"
          }
        },
        {
          "attribution_basis": "SOURCE_CITED",
          "credit": "historical_scientific",
          "entity_kind": "historical_person",
          "name": "Rahul Solanki",
          "note": "Archive attribution; detailed role assignment remains source-bound",
          "roles": [
            "discoverer"
          ],
          "source": {
            "citation": "Distributed Training of Large Language Models on AWS Trainium",
            "url": "https://cdn.amazon.science/fa/fc/6c9a63824f1fa3655fc757825256/distributed-training-of-large-language-models-on-aws-trainium.pdf"
          }
        },
        {
          "attribution_basis": "SOURCE_CITED",
          "credit": "historical_scientific",
          "entity_kind": "historical_person",
          "name": "Fei Wu",
          "note": "Archive attribution; detailed role assignment remains source-bound",
          "roles": [
            "discoverer"
          ],
          "source": {
            "citation": "Distributed Training of Large Language Models on AWS Trainium",
            "url": "https://cdn.amazon.science/fa/fc/6c9a63824f1fa3655fc757825256/distributed-training-of-large-language-models-on-aws-trainium.pdf"
          }
        },
        {
          "attribution_basis": "SOURCE_CITED",
          "credit": "historical_scientific",
          "entity_kind": "historical_person",
          "name": "Ron Diamant",
          "note": "Archive attribution; detailed role assignment remains source-bound",
          "roles": [
            "discoverer"
          ],
          "source": {
            "citation": "Distributed Training of Large Language Models on AWS Trainium",
            "url": "https://cdn.amazon.science/fa/fc/6c9a63824f1fa3655fc757825256/distributed-training-of-large-language-models-on-aws-trainium.pdf"
          }
        },
        {
          "attribution_basis": "SOURCE_CITED",
          "credit": "historical_scientific",
          "entity_kind": "historical_person",
          "name": "Yida Wang",
          "note": "Archive attribution; detailed role assignment remains source-bound",
          "roles": [
            "discoverer"
          ],
          "source": {
            "citation": "Distributed Training of Large Language Models on AWS Trainium",
            "url": "https://cdn.amazon.science/fa/fc/6c9a63824f1fa3655fc757825256/distributed-training-of-large-language-models-on-aws-trainium.pdf"
          }
        }
      ],
      "evidence": [
        {
          "description": "Bibliographic/stated-result source; not a WRS proof verification",
          "id": "source-0",
          "kind": "source_reference",
          "source": {
            "citation": "Distributed Training of Large Language Models on AWS Trainium",
            "url": "https://cdn.amazon.science/fa/fc/6c9a63824f1fa3655fc757825256/distributed-training-of-large-language-models-on-aws-trainium.pdf"
          }
        }
      ],
      "origin": "HUMAN_LEGACY",
      "record_profile": "wrs-local-research-notes-0.1",
      "record_type": "research_note",
      "rrp_version": "0.1",
      "scope": {
        "description": "Not the invention of systolic arrays, tensor/pipeline/data parallelism or ZeRO. Some features require model-code changes; symbolic tracing excludes unresolved dynamic control flow and can require opaque module wrappers. Graph lookahead consumes host memory. Historical instance-cost/performance comparisons are selected-workload results, not current prices or universal superiority.",
        "domain": "AI training systems"
      },
      "sources": [
        {
          "citation": "Distributed Training of Large Language Models on AWS Trainium",
          "url": "https://cdn.amazon.science/fa/fc/6c9a63824f1fa3655fc757825256/distributed-training-of-large-language-models-on-aws-trainium.pdf"
        }
      ],
      "statement": "# AWS Neuron distributed LLM training on Trainium\n\nIntegrated inherited parallel-training methods with Trainium-aware mixed tensor-parallel degrees, automatic pipeline graph partitioning, delayed device materialization and asynchronous graph execution in Neuron/NxDT.\n\n## Scientific scope\n\nNot the invention of systolic arrays, tensor/pipeline/data parallelism or ZeRO. Some features require model-code changes; symbolic tracing excludes unresolved dynamic control flow and can require opaque module wrappers. Graph lookahead consumes host memory. Historical instance-cost/performance comparisons are selected-workload results, not current prices or universal superiority.\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\nXinwei Fu; Zhen Zhang; Haozheng Fan; Guangtai Huang; Mohammad El-Shabani; Randy Huang; Rahul Solanki; Fei Wu; Ron Diamant; Yida Wang\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- [Distributed Training of Large Language Models on AWS Trainium](https://cdn.amazon.science/fa/fc/6c9a63824f1fa3655fc757825256/distributed-training-of-large-language-models-on-aws-trainium.pdf)\n",
      "summary": "Integrated inherited parallel-training methods with Trainium-aware mixed tensor-parallel degrees, automatic pipeline graph partitioning, delayed device materialization and asynchronous graph execution in Neuron/NxDT.",
      "title": "AWS Neuron distributed LLM training on Trainium"
    },
    "record_id": "urn:uuid:55e55737-f34f-4944-a02c-cadbc2706a59",
    "version": 1,
    "source_ref": {
      "record_id": "urn:uuid:55e55737-f34f-4944-a02c-cadbc2706a59",
      "version": 1,
      "content_hash": "sha256:6e465b9f5427b80fa666dbc78b8ca5bfd600988e9a49e12de7b6ba6e49639a76"
    },
    "export_view": "wrs-public-editorial-view-1",
    "public_view_hash": "sha256:64526babf839f59e73ee1bc8115b2bd4412ef87823d076afca362500388b7c89"
  },
  "target": {
    "record_id": "urn:uuid:55e55737-f34f-4944-a02c-cadbc2706a59",
    "version": 1,
    "content_hash": "sha256:6e465b9f5427b80fa666dbc78b8ca5bfd600988e9a49e12de7b6ba6e49639a76"
  },
  "published_grade": {
    "profile": "wrs-published-grade-1",
    "edition": "2026-10-28.1",
    "id": "fu-et-al-neuron-distributed-training-2024",
    "entry_id": "fu-et-al-neuron-distributed-training-2024",
    "revision": 1,
    "title": "AWS Neuron distributed LLM training on Trainium",
    "context": "AI training systems",
    "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": "Integrated inherited parallel-training methods with Trainium-aware mixed tensor-parallel degrees, automatic pipeline graph partitioning, delayed device materialization and asynchronous graph execution in Neuron/NxDT.",
    "scope": "Integrated inherited parallel-training methods with Trainium-aware mixed tensor-parallel degrees, automatic pipeline graph partitioning, delayed device materialization and asynchronous graph execution in Neuron/NxDT. Not the invention of systolic arrays, tensor/pipeline/data parallelism or ZeRO. Some features require model-code changes; symbolic tracing excludes unresolved dynamic control flow and can require opaque module wrappers. Graph lookahead consumes host memory. Historical instance-cost/performance comparisons are selected-workload results, not current prices or universal superiority.",
    "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. Not the invention of systolic arrays, tensor/pipeline/data parallelism or ZeRO. Some features require model-code changes; symbolic tracing excludes unresolved dynamic control flow and can require opaque module wrappers. Graph lookahead consumes host memory. Historical instance-cost/performance comparisons are selected-workload results, not current prices or universal superiority.",
    "references": [
      {
        "title": "Distributed Training of Large Language Models on AWS Trainium",
        "url": "https://cdn.amazon.science/fa/fc/6c9a63824f1fa3655fc757825256/distributed-training-of-large-language-models-on-aws-trainium.pdf",
        "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:55e55737-f34f-4944-a02c-cadbc2706a59",
      "version": 1,
      "content_hash": "sha256:6e465b9f5427b80fa666dbc78b8ca5bfd600988e9a49e12de7b6ba6e49639a76"
    },
    "scientific_correctness_certified": false,
    "independent_expert_calibration": false,
    "method": "PROPRIETARY",
    "points": "65000"
  },
  "canonical_agent_submission": false,
  "scientific_verification_awarded": false,
  "reader_guide": null,
  "reading_status": "SOURCE_LINKED_NOTE"
}