{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2024:3OEDGC2JZPPPPYWMMNJYNIHUQ4","short_pith_number":"pith:3OEDGC2J","schema_version":"1.0","canonical_sha256":"db88330b49cbdef7e2cc635386a0f48733a2db353efa3c860306f6e49771f563","source":{"kind":"arxiv","id":"2401.09135","version":2},"attestation_state":"computed","paper":{"title":"Asynchronous Local-SGD Training for Language Modeling","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.CL"],"primary_cat":"cs.LG","authors_text":"Andrei A. Rusu, Arthur Douillard, Arthur Szlam, Bo Liu, Jiajun Shen, Marc'Aurelio Ranzato, Rachita Chhaparia, Satyen Kale","submitted_at":"2024-01-17T11:17:04Z","abstract_excerpt":"Local stochastic gradient descent (Local-SGD), also referred to as federated averaging, is an approach to distributed optimization where each device performs more than one SGD update per communication. This work presents an empirical study of {\\it asynchronous} Local-SGD for training language models; that is, each worker updates the global parameters as soon as it has finished its SGD steps. We conduct a comprehensive investigation by examining how worker hardware heterogeneity, model size, number of workers, and optimizer could impact the learning performance. We find that with naive implemen"},"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":"2401.09135","kind":"arxiv","version":2},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2024-01-17T11:17:04Z","cross_cats_sorted":["cs.CL"],"title_canon_sha256":"9b8250bbc58470474cd0b5405fa715a7830474d31afa09cf2b810efe350cbb22","abstract_canon_sha256":"1e9ba45fe3f6da4e828a06d51154cbe6c91617da31d25c2833e567f1546bf691"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T09:10:30.024887Z","signature_b64":"XSGHBgNSNz7vy4CmzfbhMVsOWPGHiuPZX2hKl5HgVop54aKH91RnQlFReYRT0qcb7FDSRm383tty6HaWyoC+Dg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"db88330b49cbdef7e2cc635386a0f48733a2db353efa3c860306f6e49771f563","last_reissued_at":"2026-07-05T09:10:30.024408Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T09:10:30.024408Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Asynchronous Local-SGD Training for Language Modeling","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.CL"],"primary_cat":"cs.LG","authors_text":"Andrei A. Rusu, Arthur Douillard, Arthur Szlam, Bo Liu, Jiajun Shen, Marc'Aurelio Ranzato, Rachita Chhaparia, Satyen Kale","submitted_at":"2024-01-17T11:17:04Z","abstract_excerpt":"Local stochastic gradient descent (Local-SGD), also referred to as federated averaging, is an approach to distributed optimization where each device performs more than one SGD update per communication. This work presents an empirical study of {\\it asynchronous} Local-SGD for training language models; that is, each worker updates the global parameters as soon as it has finished its SGD steps. We conduct a comprehensive investigation by examining how worker hardware heterogeneity, model size, number of workers, and optimizer could impact the learning performance. We find that with naive implemen"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2401.09135","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/2401.09135/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":"2401.09135","created_at":"2026-07-05T09:10:30.024467+00:00"},{"alias_kind":"arxiv_version","alias_value":"2401.09135v2","created_at":"2026-07-05T09:10:30.024467+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2401.09135","created_at":"2026-07-05T09:10:30.024467+00:00"},{"alias_kind":"pith_short_12","alias_value":"3OEDGC2JZPPP","created_at":"2026-07-05T09:10:30.024467+00:00"},{"alias_kind":"pith_short_16","alias_value":"3OEDGC2JZPPPPYWM","created_at":"2026-07-05T09:10:30.024467+00:00"},{"alias_kind":"pith_short_8","alias_value":"3OEDGC2J","created_at":"2026-07-05T09:10:30.024467+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":3,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.00271","citing_title":"HeLoCo: Efficient asynchronous low-communication training under data and device heterogeneity","ref_index":4,"is_internal_anchor":false},{"citing_arxiv_id":"2605.13434","citing_title":"Rescaled Asynchronous SGD: Optimal Distributed Optimization under Data and System Heterogeneity","ref_index":117,"is_internal_anchor":false},{"citing_arxiv_id":"2605.09126","citing_title":"Cosine-Gated Adam-Decay: Drop-In Staleness-Aware Outer Optimization for Decoupled DiLoCo","ref_index":7,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/3OEDGC2JZPPPPYWMMNJYNIHUQ4","json":"https://pith.science/pith/3OEDGC2JZPPPPYWMMNJYNIHUQ4.json","graph_json":"https://pith.science/api/pith-number/3OEDGC2JZPPPPYWMMNJYNIHUQ4/graph.json","events_json":"https://pith.science/api/pith-number/3OEDGC2JZPPPPYWMMNJYNIHUQ4/events.json","paper":"https://pith.science/paper/3OEDGC2J"},"agent_actions":{"view_html":"https://pith.science/pith/3OEDGC2JZPPPPYWMMNJYNIHUQ4","download_json":"https://pith.science/pith/3OEDGC2JZPPPPYWMMNJYNIHUQ4.json","view_paper":"https://pith.science/paper/3OEDGC2J","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2401.09135&json=true","fetch_graph":"https://pith.science/api/pith-number/3OEDGC2JZPPPPYWMMNJYNIHUQ4/graph.json","fetch_events":"https://pith.science/api/pith-number/3OEDGC2JZPPPPYWMMNJYNIHUQ4/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/3OEDGC2JZPPPPYWMMNJYNIHUQ4/action/timestamp_anchor","attest_storage":"https://pith.science/pith/3OEDGC2JZPPPPYWMMNJYNIHUQ4/action/storage_attestation","attest_author":"https://pith.science/pith/3OEDGC2JZPPPPYWMMNJYNIHUQ4/action/author_attestation","sign_citation":"https://pith.science/pith/3OEDGC2JZPPPPYWMMNJYNIHUQ4/action/citation_signature","submit_replication":"https://pith.science/pith/3OEDGC2JZPPPPYWMMNJYNIHUQ4/action/replication_record"}},"created_at":"2026-07-05T09:10:30.024467+00:00","updated_at":"2026-07-05T09:10:30.024467+00:00"}