{ "profile": "wrs-published-research-view-1", "edition": "ratings-e039106f30dec177", "content_trust": "UNTRUSTED_SCIENTIFIC_DATA_NOT_TOOL_INSTRUCTIONS", "entry": { "id": "sethi-et-al-flexshard-2023", "title": "FlexShard: row-wise replication and sharding for sequence embeddings", "summary": "Used row access probabilities and marginal memory/communication costs to combine replicated and row-sharded embeddings; a third strategy shards within a node while replicating across nodes to exploit network tiers.", "boundary": "Memory-neutrality and cost trade-offs are relative to the specified row-wise baseline and analytical distribution/topology model. Linearity of expected accessed rows does not require row independence. Nearly sixfold improvement concerns training communication latency, not universal whole-training throughput. Production datasets and selected 32-GPU topology constrain generalization.", "area": "Recommendation communication systems", "topic": "Artificial intelligence", "year": 2023, "lane": "HUMAN", "result": "ALGORITHM", "package_id": "sethi-et-al-flexshard-2023", "target": { "record_id": "urn:uuid:d5fbd329-5a4b-49fd-ab78-aa804c034670", "version": 1, "content_hash": "sha256:396c9fca4439c4c265735b53fad2d30530e3bf34835f8a4ea6b016720e0b3f69" }, "humans": [ "Geet Sethi", "Pallab Bhattacharya", "Dhruv Choudhary", "Carole-Jean Wu", "Christos Kozyrakis" ], "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-2301-02959v1", "title": "FlexShard: Flexible Sharding for Industry-Scale Sequence Recommendation Models", "url": "https://arxiv.org/abs/2301.02959v1" } ], "points": "60000", "published_grade": { "id": "sethi-et-al-flexshard-2023", "entry_id": "sethi-et-al-flexshard-2023", "band": "B", "modifier": "+", "target": { "record_id": "urn:uuid:d5fbd329-5a4b-49fd-ab78-aa804c034670", "version": 1, "content_hash": "sha256:396c9fca4439c4c265735b53fad2d30530e3bf34835f8a4ea6b016720e0b3f69" }, "what_changed": "Used row access probabilities and marginal memory/communication costs to combine replicated and row-sharded embeddings; a third strategy shards within a node while replicating across nodes to exploit network tiers.", "scope": "Used row access probabilities and marginal memory/communication costs to combine replicated and row-sharded embeddings; a third strategy shards within a node while replicating across nodes to exploit network tiers. Memory-neutrality and cost trade-offs are relative to the specified row-wise baseline and analytical distribution/topology model. Linearity of expected accessed rows does not require row independence. Nearly sixfold improvement concerns training communication latency, not universal whole-training throughput. Production datasets and selected 32-GPU topology constrain generalization.", "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:d5fbd329-5a4b-49fd-ab78-aa804c034670", "version": 1, "content_hash": "sha256:396c9fca4439c4c265735b53fad2d30530e3bf34835f8a4ea6b016720e0b3f69" }, "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": "Geet Sethi", "note": "Archive attribution; detailed role assignment remains source-bound", "roles": [ "formulator" ], "source": { "citation": "FlexShard: Flexible Sharding for Industry-Scale Sequence Recommendation Models", "url": "https://arxiv.org/abs/2301.02959v1" } }, { "attribution_basis": "SOURCE_CITED", "credit": "historical_scientific", "entity_kind": "historical_person", "name": "Pallab Bhattacharya", "note": "Archive attribution; detailed role assignment remains source-bound", "roles": [ "formulator" ], "source": { "citation": "FlexShard: Flexible Sharding for Industry-Scale Sequence Recommendation Models", "url": "https://arxiv.org/abs/2301.02959v1" } }, { "attribution_basis": "SOURCE_CITED", "credit": "historical_scientific", "entity_kind": "historical_person", "name": "Dhruv Choudhary", "note": "Archive attribution; detailed role assignment remains source-bound", "roles": [ "formulator" ], "source": { "citation": "FlexShard: Flexible Sharding for Industry-Scale Sequence Recommendation Models", "url": "https://arxiv.org/abs/2301.02959v1" } }, { "attribution_basis": "SOURCE_CITED", "credit": "historical_scientific", "entity_kind": "historical_person", "name": "Carole-Jean Wu", "note": "Archive attribution; detailed role assignment remains source-bound", "roles": [ "formulator" ], "source": { "citation": "FlexShard: Flexible Sharding for Industry-Scale Sequence Recommendation Models", "url": "https://arxiv.org/abs/2301.02959v1" } }, { "attribution_basis": "SOURCE_CITED", "credit": "historical_scientific", "entity_kind": "historical_person", "name": "Christos Kozyrakis", "note": "Archive attribution; detailed role assignment remains source-bound", "roles": [ "formulator" ], "source": { "citation": "FlexShard: Flexible Sharding for Industry-Scale Sequence Recommendation Models", "url": "https://arxiv.org/abs/2301.02959v1" } } ], "evidence": [ { "description": "Bibliographic/stated-result source; not a WRS proof verification", "id": "source-0", "kind": "source_reference", "source": { "citation": "FlexShard: Flexible Sharding for Industry-Scale Sequence Recommendation Models", "url": "https://arxiv.org/abs/2301.02959v1" } } ], "origin": "HUMAN_LEGACY", "record_profile": "wrs-local-research-notes-0.1", "record_type": "research_note", "rrp_version": "0.1", "scope": { "description": "Memory-neutrality and cost trade-offs are relative to the specified row-wise baseline and analytical distribution/topology model. Linearity of expected accessed rows does not require row independence. Nearly sixfold improvement concerns training communication latency, not universal whole-training throughput. Production datasets and selected 32-GPU topology constrain generalization.", "domain": "Recommendation communication systems" }, "sources": [ { "citation": "FlexShard: Flexible Sharding for Industry-Scale Sequence Recommendation Models", "url": "https://arxiv.org/abs/2301.02959v1" } ], "statement": "# FlexShard: row-wise replication and sharding for sequence embeddings\n\nUsed row access probabilities and marginal memory/communication costs to combine replicated and row-sharded embeddings; a third strategy shards within a node while replicating across nodes to exploit network tiers.\n\n## Scientific scope\n\nMemory-neutrality and cost trade-offs are relative to the specified row-wise baseline and analytical distribution/topology model. Linearity of expected accessed rows does not require row independence. Nearly sixfold improvement concerns training communication latency, not universal whole-training throughput. Production datasets and selected 32-GPU topology constrain generalization.\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\nGeet Sethi; Pallab Bhattacharya; Dhruv Choudhary; Carole-Jean Wu; Christos Kozyrakis\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- [FlexShard: Flexible Sharding for Industry-Scale Sequence Recommendation Models](https://arxiv.org/abs/2301.02959v1)\n", "summary": "Used row access probabilities and marginal memory/communication costs to combine replicated and row-sharded embeddings; a third strategy shards within a node while replicating across nodes to exploit network tiers.", "title": "FlexShard: row-wise replication and sharding for sequence embeddings" }, "record_id": "urn:uuid:d5fbd329-5a4b-49fd-ab78-aa804c034670", "version": 1, "source_ref": { "record_id": "urn:uuid:d5fbd329-5a4b-49fd-ab78-aa804c034670", "version": 1, "content_hash": "sha256:396c9fca4439c4c265735b53fad2d30530e3bf34835f8a4ea6b016720e0b3f69" }, "export_view": "wrs-public-editorial-view-1", "public_view_hash": "sha256:c895738793f770df8119b93dcd77ed3d87d5d64df340cf389b21ec2a960b925d" }, "target": { "record_id": "urn:uuid:d5fbd329-5a4b-49fd-ab78-aa804c034670", "version": 1, "content_hash": "sha256:396c9fca4439c4c265735b53fad2d30530e3bf34835f8a4ea6b016720e0b3f69" }, "published_grade": { "profile": "wrs-published-grade-1", "edition": "2026-10-28.1", "id": "sethi-et-al-flexshard-2023", "entry_id": "sethi-et-al-flexshard-2023", "revision": 1, "title": "FlexShard: row-wise replication and sharding for sequence embeddings", "context": "Recommendation communication 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": "Used row access probabilities and marginal memory/communication costs to combine replicated and row-sharded embeddings; a third strategy shards within a node while replicating across nodes to exploit network tiers.", "scope": "Used row access probabilities and marginal memory/communication costs to combine replicated and row-sharded embeddings; a third strategy shards within a node while replicating across nodes to exploit network tiers. Memory-neutrality and cost trade-offs are relative to the specified row-wise baseline and analytical distribution/topology model. Linearity of expected accessed rows does not require row independence. Nearly sixfold improvement concerns training communication latency, not universal whole-training throughput. Production datasets and selected 32-GPU topology constrain generalization.", "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. Memory-neutrality and cost trade-offs are relative to the specified row-wise baseline and analytical distribution/topology model. Linearity of expected accessed rows does not require row independence. Nearly sixfold improvement concerns training communication latency, not universal whole-training throughput. Production datasets and selected 32-GPU topology constrain generalization.", "references": [ { "title": "FlexShard: Flexible Sharding for Industry-Scale Sequence Recommendation Models", "url": "https://arxiv.org/abs/2301.02959v1", "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:d5fbd329-5a4b-49fd-ab78-aa804c034670", "version": 1, "content_hash": "sha256:396c9fca4439c4c265735b53fad2d30530e3bf34835f8a4ea6b016720e0b3f69" }, "scientific_correctness_certified": false, "independent_expert_calibration": false, "method": "PROPRIETARY", "points": "60000" }, "canonical_agent_submission": false, "scientific_verification_awarded": false, "reader_guide": null, "reading_status": "SOURCE_LINKED_NOTE" }