{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:OUZ7MZ3Y5BV7RKUZJMY542RGGG","short_pith_number":"pith:OUZ7MZ3Y","schema_version":"1.0","canonical_sha256":"7533f66778e86bf8aa994b31de6a263182bfe78b3b22bd431ec1a2bb462dd9cc","source":{"kind":"arxiv","id":"2501.18965","version":2},"attestation_state":"computed","paper":{"title":"The Surprising Agreement Between Convex Optimization Theory and Learning-Rate Scheduling for Large Model Training","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["math.OC","stat.ML"],"primary_cat":"cs.LG","authors_text":"Adrien Taylor, Alexander H\\\"agele, Fabian Schaipp, Francis Bach, Umut Simsekli","submitted_at":"2025-01-31T08:55:56Z","abstract_excerpt":"We show that learning-rate schedules for large model training behave surprisingly similar to a performance bound from non-smooth convex optimization theory. We provide a bound for the constant schedule with linear cooldown; in particular, the practical benefit of cooldown is reflected in the bound due to the absence of logarithmic terms. Further, we show that this surprisingly close match between optimization theory and practice can be exploited for learning-rate tuning: we achieve noticeable improvements for training 124M and 210M Llama-type models by (i) extending the schedule for continued "},"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":"2501.18965","kind":"arxiv","version":2},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2025-01-31T08:55:56Z","cross_cats_sorted":["math.OC","stat.ML"],"title_canon_sha256":"9836f50787358c98b6ec5ff29d8ebf2e052845e1b829d4a4c02d4e6a95bd3a67","abstract_canon_sha256":"2269f13f7936fcc7ec93b104e7d9dd2845ac6b01b3310c8a66694708ca642c44"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T11:41:38.771515Z","signature_b64":"RQW4pSQyoV4Wu6n06NmIMrLFQ9xbn4ygrH8vxUGgW+2b+Mb5sjU9SXwlkL8OmpZHcZD9dKfj99M/v58shz5bAg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"7533f66778e86bf8aa994b31de6a263182bfe78b3b22bd431ec1a2bb462dd9cc","last_reissued_at":"2026-07-05T11:41:38.770908Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T11:41:38.770908Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"The Surprising Agreement Between Convex Optimization Theory and Learning-Rate Scheduling for Large Model Training","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["math.OC","stat.ML"],"primary_cat":"cs.LG","authors_text":"Adrien Taylor, Alexander H\\\"agele, Fabian Schaipp, Francis Bach, Umut Simsekli","submitted_at":"2025-01-31T08:55:56Z","abstract_excerpt":"We show that learning-rate schedules for large model training behave surprisingly similar to a performance bound from non-smooth convex optimization theory. We provide a bound for the constant schedule with linear cooldown; in particular, the practical benefit of cooldown is reflected in the bound due to the absence of logarithmic terms. Further, we show that this surprisingly close match between optimization theory and practice can be exploited for learning-rate tuning: we achieve noticeable improvements for training 124M and 210M Llama-type models by (i) extending the schedule for continued "},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2501.18965","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/2501.18965/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":"2501.18965","created_at":"2026-07-05T11:41:38.770984+00:00"},{"alias_kind":"arxiv_version","alias_value":"2501.18965v2","created_at":"2026-07-05T11:41:38.770984+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2501.18965","created_at":"2026-07-05T11:41:38.770984+00:00"},{"alias_kind":"pith_short_12","alias_value":"OUZ7MZ3Y5BV7","created_at":"2026-07-05T11:41:38.770984+00:00"},{"alias_kind":"pith_short_16","alias_value":"OUZ7MZ3Y5BV7RKUZ","created_at":"2026-07-05T11:41:38.770984+00:00"},{"alias_kind":"pith_short_8","alias_value":"OUZ7MZ3Y","created_at":"2026-07-05T11:41:38.770984+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":2,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2605.23061","citing_title":"Anytime Training with Schedule-Free Spectral Optimization","ref_index":56,"is_internal_anchor":false},{"citing_arxiv_id":"2605.11172","citing_title":"Optimistic Dual Averaging Unifies Modern Optimizers","ref_index":15,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/OUZ7MZ3Y5BV7RKUZJMY542RGGG","json":"https://pith.science/pith/OUZ7MZ3Y5BV7RKUZJMY542RGGG.json","graph_json":"https://pith.science/api/pith-number/OUZ7MZ3Y5BV7RKUZJMY542RGGG/graph.json","events_json":"https://pith.science/api/pith-number/OUZ7MZ3Y5BV7RKUZJMY542RGGG/events.json","paper":"https://pith.science/paper/OUZ7MZ3Y"},"agent_actions":{"view_html":"https://pith.science/pith/OUZ7MZ3Y5BV7RKUZJMY542RGGG","download_json":"https://pith.science/pith/OUZ7MZ3Y5BV7RKUZJMY542RGGG.json","view_paper":"https://pith.science/paper/OUZ7MZ3Y","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2501.18965&json=true","fetch_graph":"https://pith.science/api/pith-number/OUZ7MZ3Y5BV7RKUZJMY542RGGG/graph.json","fetch_events":"https://pith.science/api/pith-number/OUZ7MZ3Y5BV7RKUZJMY542RGGG/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/OUZ7MZ3Y5BV7RKUZJMY542RGGG/action/timestamp_anchor","attest_storage":"https://pith.science/pith/OUZ7MZ3Y5BV7RKUZJMY542RGGG/action/storage_attestation","attest_author":"https://pith.science/pith/OUZ7MZ3Y5BV7RKUZJMY542RGGG/action/author_attestation","sign_citation":"https://pith.science/pith/OUZ7MZ3Y5BV7RKUZJMY542RGGG/action/citation_signature","submit_replication":"https://pith.science/pith/OUZ7MZ3Y5BV7RKUZJMY542RGGG/action/replication_record"}},"created_at":"2026-07-05T11:41:38.770984+00:00","updated_at":"2026-07-05T11:41:38.770984+00:00"}