{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:7EIEAAFFCPD7D7VD4NT3WUSNJ7","short_pith_number":"pith:7EIEAAFF","schema_version":"1.0","canonical_sha256":"f9104000a513c7f1fea3e367bb524d4fc6ac8cca6d5f5b71fdd62d3628ae8f34","source":{"kind":"arxiv","id":"2502.12996","version":1},"attestation_state":"computed","paper":{"title":"Eager Updates For Overlapped Communication and Computation in DiLoCo","license":"http://creativecommons.org/licenses/by-sa/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Arthur Douillard, Satyen Kale, Yanislav Donchev","submitted_at":"2025-02-18T16:16:14Z","abstract_excerpt":"Distributed optimization methods such as DiLoCo have been shown to be effective in training very large models across multiple distributed workers, such as datacenters. These methods split updates into two parts: an inner optimization phase, where the workers independently execute multiple optimization steps on their own local data, and an outer optimization step, where the inner updates are synchronized. While such approaches require orders of magnitude less communication than standard data-parallel training, in settings where the workers are datacenters, even the limited communication require"},"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":"2502.12996","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by-sa/4.0/","primary_cat":"cs.CL","submitted_at":"2025-02-18T16:16:14Z","cross_cats_sorted":[],"title_canon_sha256":"7a8b5bf5ee8bb5fb6c5dd025d2909f1c071c5d901d4621ea000737afbc8e5046","abstract_canon_sha256":"70c53f4db9fa72ce02c9be438edba6fbbdffecbfb46aed137925b567141d7dd8"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T10:16:27.746714Z","signature_b64":"KCQabcKoUgZffSlKW4kV7ICCkxHMnd7waNK0Mvw4P5egR7lza9AF+ZzAedR8JZyeoAr0Z6IYH1SizazNZWtPCA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"f9104000a513c7f1fea3e367bb524d4fc6ac8cca6d5f5b71fdd62d3628ae8f34","last_reissued_at":"2026-07-05T10:16:27.746156Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T10:16:27.746156Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Eager Updates For Overlapped Communication and Computation in DiLoCo","license":"http://creativecommons.org/licenses/by-sa/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Arthur Douillard, Satyen Kale, Yanislav Donchev","submitted_at":"2025-02-18T16:16:14Z","abstract_excerpt":"Distributed optimization methods such as DiLoCo have been shown to be effective in training very large models across multiple distributed workers, such as datacenters. These methods split updates into two parts: an inner optimization phase, where the workers independently execute multiple optimization steps on their own local data, and an outer optimization step, where the inner updates are synchronized. While such approaches require orders of magnitude less communication than standard data-parallel training, in settings where the workers are datacenters, even the limited communication require"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2502.12996","kind":"arxiv","version":1},"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/2502.12996/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":"2502.12996","created_at":"2026-07-05T10:16:27.746224+00:00"},{"alias_kind":"arxiv_version","alias_value":"2502.12996v1","created_at":"2026-07-05T10:16:27.746224+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2502.12996","created_at":"2026-07-05T10:16:27.746224+00:00"},{"alias_kind":"pith_short_12","alias_value":"7EIEAAFFCPD7","created_at":"2026-07-05T10:16:27.746224+00:00"},{"alias_kind":"pith_short_16","alias_value":"7EIEAAFFCPD7D7VD","created_at":"2026-07-05T10:16:27.746224+00:00"},{"alias_kind":"pith_short_8","alias_value":"7EIEAAFF","created_at":"2026-07-05T10:16:27.746224+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":2,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.11081","citing_title":"Unifying Local Communications and Local Updates for LLM Pretraining","ref_index":14,"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":4,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/7EIEAAFFCPD7D7VD4NT3WUSNJ7","json":"https://pith.science/pith/7EIEAAFFCPD7D7VD4NT3WUSNJ7.json","graph_json":"https://pith.science/api/pith-number/7EIEAAFFCPD7D7VD4NT3WUSNJ7/graph.json","events_json":"https://pith.science/api/pith-number/7EIEAAFFCPD7D7VD4NT3WUSNJ7/events.json","paper":"https://pith.science/paper/7EIEAAFF"},"agent_actions":{"view_html":"https://pith.science/pith/7EIEAAFFCPD7D7VD4NT3WUSNJ7","download_json":"https://pith.science/pith/7EIEAAFFCPD7D7VD4NT3WUSNJ7.json","view_paper":"https://pith.science/paper/7EIEAAFF","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2502.12996&json=true","fetch_graph":"https://pith.science/api/pith-number/7EIEAAFFCPD7D7VD4NT3WUSNJ7/graph.json","fetch_events":"https://pith.science/api/pith-number/7EIEAAFFCPD7D7VD4NT3WUSNJ7/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/7EIEAAFFCPD7D7VD4NT3WUSNJ7/action/timestamp_anchor","attest_storage":"https://pith.science/pith/7EIEAAFFCPD7D7VD4NT3WUSNJ7/action/storage_attestation","attest_author":"https://pith.science/pith/7EIEAAFFCPD7D7VD4NT3WUSNJ7/action/author_attestation","sign_citation":"https://pith.science/pith/7EIEAAFFCPD7D7VD4NT3WUSNJ7/action/citation_signature","submit_replication":"https://pith.science/pith/7EIEAAFFCPD7D7VD4NT3WUSNJ7/action/replication_record"}},"created_at":"2026-07-05T10:16:27.746224+00:00","updated_at":"2026-07-05T10:16:27.746224+00:00"}