{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:MO3T3DH2LSIM5XFZRFKEW25JE2","short_pith_number":"pith:MO3T3DH2","schema_version":"1.0","canonical_sha256":"63b73d8cfa5c90cedcb989544b6ba9269e1a9c372236f8f168dd137b7247c33e","source":{"kind":"arxiv","id":"2502.16927","version":2},"attestation_state":"computed","paper":{"title":"BigMac: A Communication-Efficient Mixture-of-Experts Model Structure for Fast Training and Inference","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.LG","authors_text":"Cheng Li, Hongrui Zhan, Jiaan Zhu, Lin Zhang, Shengnan Wang, Youhui Bai, Zewen Jin, Zhenyu Ming","submitted_at":"2025-02-24T07:37:29Z","abstract_excerpt":"The Mixture-of-Experts (MoE) structure scales the Transformer-based large language models (LLMs) and improves their performance with only the sub-linear increase in computation resources. Recently, a fine-grained DeepSeekMoE structure is proposed, which can further improve the computing efficiency of MoE without performance degradation. However, the All-to-All communication introduced by MoE has become a bottleneck, especially for the fine-grained structure, which typically involves and activates more experts, hence contributing to heavier communication overhead.\n  In this paper, we propose a "},"verification_status":{"content_addressed":true,"pith_receipt":true,"author_attested":false,"weak_author_claims":0,"strong_author_claims":0,"externally_anchored":false,"storage_verified":false,"citation_signatures":0,"replication_records":0,"graph_snapshot":true,"references_resolved":false,"formal_links_present":false},"canonical_record":{"source":{"id":"2502.16927","kind":"arxiv","version":2},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2025-02-24T07:37:29Z","cross_cats_sorted":["cs.AI"],"title_canon_sha256":"a1c2159ab191e515f656641ccfa8aed2d4158b624ea3e5c81a39bebb1248344b","abstract_canon_sha256":"4a7dede24ca46267e1480c7f0b6d57659ce66068c86487d3dd016d01526e528d"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T10:26:14.533165Z","signature_b64":"HqJF/GBzdvJ//lXMqYpZ0SuOUTXcqM8kpHzlTSbeqUXuaNjtain7IO4SyZSmtUCYZ+hCDeaoCkduRpNd/xW/Dg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"63b73d8cfa5c90cedcb989544b6ba9269e1a9c372236f8f168dd137b7247c33e","last_reissued_at":"2026-07-05T10:26:14.532620Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T10:26:14.532620Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"BigMac: A Communication-Efficient Mixture-of-Experts Model Structure for Fast Training and Inference","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.LG","authors_text":"Cheng Li, Hongrui Zhan, Jiaan Zhu, Lin Zhang, Shengnan Wang, Youhui Bai, Zewen Jin, Zhenyu Ming","submitted_at":"2025-02-24T07:37:29Z","abstract_excerpt":"The Mixture-of-Experts (MoE) structure scales the Transformer-based large language models (LLMs) and improves their performance with only the sub-linear increase in computation resources. Recently, a fine-grained DeepSeekMoE structure is proposed, which can further improve the computing efficiency of MoE without performance degradation. However, the All-to-All communication introduced by MoE has become a bottleneck, especially for the fine-grained structure, which typically involves and activates more experts, hence contributing to heavier communication overhead.\n  In this paper, we propose a "},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2502.16927","kind":"arxiv","version":2},"verdict":{"id":null,"model_set":{},"created_at":null,"strongest_claim":"","one_line_summary":"","pipeline_version":null,"weakest_assumption":"","pith_extraction_headline":""},"integrity":{"clean":true,"summary":{"advisory":0,"critical":0,"by_detector":{},"informational":0},"endpoint":"/pith/2502.16927/integrity.json","findings":[],"available":true,"detectors_run":[],"snapshot_sha256":"c28c3603d3b5d939e8dc4c7e95fa8dfce3d595e45f758748cecf8e644a296938"},"references":{"count":0,"sample":[],"resolved_work":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57","internal_anchors":0},"formal_canon":{"evidence_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"},"aliases":[{"alias_kind":"arxiv","alias_value":"2502.16927","created_at":"2026-07-05T10:26:14.532691+00:00"},{"alias_kind":"arxiv_version","alias_value":"2502.16927v2","created_at":"2026-07-05T10:26:14.532691+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2502.16927","created_at":"2026-07-05T10:26:14.532691+00:00"},{"alias_kind":"pith_short_12","alias_value":"MO3T3DH2LSIM","created_at":"2026-07-05T10:26:14.532691+00:00"},{"alias_kind":"pith_short_16","alias_value":"MO3T3DH2LSIM5XFZ","created_at":"2026-07-05T10:26:14.532691+00:00"},{"alias_kind":"pith_short_8","alias_value":"MO3T3DH2","created_at":"2026-07-05T10:26:14.532691+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":1,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2605.06206","citing_title":"Federation of Experts: Communication Efficient Distributed Inference for Large Language Models","ref_index":14,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/MO3T3DH2LSIM5XFZRFKEW25JE2","json":"https://pith.science/pith/MO3T3DH2LSIM5XFZRFKEW25JE2.json","graph_json":"https://pith.science/api/pith-number/MO3T3DH2LSIM5XFZRFKEW25JE2/graph.json","events_json":"https://pith.science/api/pith-number/MO3T3DH2LSIM5XFZRFKEW25JE2/events.json","paper":"https://pith.science/paper/MO3T3DH2"},"agent_actions":{"view_html":"https://pith.science/pith/MO3T3DH2LSIM5XFZRFKEW25JE2","download_json":"https://pith.science/pith/MO3T3DH2LSIM5XFZRFKEW25JE2.json","view_paper":"https://pith.science/paper/MO3T3DH2","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2502.16927&json=true","fetch_graph":"https://pith.science/api/pith-number/MO3T3DH2LSIM5XFZRFKEW25JE2/graph.json","fetch_events":"https://pith.science/api/pith-number/MO3T3DH2LSIM5XFZRFKEW25JE2/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/MO3T3DH2LSIM5XFZRFKEW25JE2/action/timestamp_anchor","attest_storage":"https://pith.science/pith/MO3T3DH2LSIM5XFZRFKEW25JE2/action/storage_attestation","attest_author":"https://pith.science/pith/MO3T3DH2LSIM5XFZRFKEW25JE2/action/author_attestation","sign_citation":"https://pith.science/pith/MO3T3DH2LSIM5XFZRFKEW25JE2/action/citation_signature","submit_replication":"https://pith.science/pith/MO3T3DH2LSIM5XFZRFKEW25JE2/action/replication_record"}},"created_at":"2026-07-05T10:26:14.532691+00:00","updated_at":"2026-07-05T10:26:14.532691+00:00"}