{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:XBWDGZKMZ6XMFJM6JVAP2VZDEP","short_pith_number":"pith:XBWDGZKM","schema_version":"1.0","canonical_sha256":"b86c33654ccfaec2a59e4d40fd572323daf118a80f0a78ad870049a5af170595","source":{"kind":"arxiv","id":"2510.14717","version":2},"attestation_state":"computed","paper":{"title":"Seesaw: Accelerating Training by Balancing Learning Rate and Batch Size Scheduling","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI","math.OC","stat.ML"],"primary_cat":"cs.LG","authors_text":"Alexandru Meterez, Cengiz Pehlevan, Costin-Andrei Oncescu, Depen Morwani, Jingfeng Wu, Sham Kakade","submitted_at":"2025-10-16T14:17:38Z","abstract_excerpt":"Increasing the batch size during training -- a ''batch ramp'' -- is a promising strategy to accelerate large language model pretraining. While for SGD, doubling the batch size can be equivalent to halving the learning rate, the optimal strategy for adaptive optimizers like Adam is less clear. As a result, any batch-ramp scheduling, if used at all, is typically tuned heuristically. This work develops a principled framework for batch-size scheduling and introduces Seesaw: whenever a standard scheduler would halve the learning rate, Seesaw instead multiplies it by $1/\\sqrt{2}$ and doubles the bat"},"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":"2510.14717","kind":"arxiv","version":2},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2025-10-16T14:17:38Z","cross_cats_sorted":["cs.AI","math.OC","stat.ML"],"title_canon_sha256":"7ae3129816b111461d2997c6dce3f1f8eadc88455ab961998cc71c20c165c2d5","abstract_canon_sha256":"26a6d4ec04466d6d3ed12944322bc6a4ab0bae808de55ff4451d496fb028656d"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-28T02:24:17.798795Z","signature_b64":"v4HoBx1k6UPSemsVkeScZKw1k1is3YqKyLF5CkFKlao25ebZcdRvI50KJShxdXT5dm/ylf9wxeP0vN7I+eDPCw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"b86c33654ccfaec2a59e4d40fd572323daf118a80f0a78ad870049a5af170595","last_reissued_at":"2026-07-28T02:24:17.797852Z","signature_status":"signed_v1","first_computed_at":"2026-07-28T02:24:17.797852Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Seesaw: Accelerating Training by Balancing Learning Rate and Batch Size Scheduling","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI","math.OC","stat.ML"],"primary_cat":"cs.LG","authors_text":"Alexandru Meterez, Cengiz Pehlevan, Costin-Andrei Oncescu, Depen Morwani, Jingfeng Wu, Sham Kakade","submitted_at":"2025-10-16T14:17:38Z","abstract_excerpt":"Increasing the batch size during training -- a ''batch ramp'' -- is a promising strategy to accelerate large language model pretraining. While for SGD, doubling the batch size can be equivalent to halving the learning rate, the optimal strategy for adaptive optimizers like Adam is less clear. As a result, any batch-ramp scheduling, if used at all, is typically tuned heuristically. This work develops a principled framework for batch-size scheduling and introduces Seesaw: whenever a standard scheduler would halve the learning rate, Seesaw instead multiplies it by $1/\\sqrt{2}$ and doubles the bat"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2510.14717","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/2510.14717/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":"2510.14717","created_at":"2026-07-28T02:24:17.798280+00:00"},{"alias_kind":"arxiv_version","alias_value":"2510.14717v2","created_at":"2026-07-28T02:24:17.798280+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2510.14717","created_at":"2026-07-28T02:24:17.798280+00:00"},{"alias_kind":"pith_short_12","alias_value":"XBWDGZKMZ6XM","created_at":"2026-07-28T02:24:17.798280+00:00"},{"alias_kind":"pith_short_16","alias_value":"XBWDGZKMZ6XMFJM6","created_at":"2026-07-28T02:24:17.798280+00:00"},{"alias_kind":"pith_short_8","alias_value":"XBWDGZKM","created_at":"2026-07-28T02:24:17.798280+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":1,"internal_anchor_count":1,"sample":[{"citing_arxiv_id":"2602.04774","citing_title":"Theory of Optimal Learning Rate Schedules and Scaling Laws for a Random Feature Model","ref_index":13,"is_internal_anchor":true}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/XBWDGZKMZ6XMFJM6JVAP2VZDEP","json":"https://pith.science/pith/XBWDGZKMZ6XMFJM6JVAP2VZDEP.json","graph_json":"https://pith.science/api/pith-number/XBWDGZKMZ6XMFJM6JVAP2VZDEP/graph.json","events_json":"https://pith.science/api/pith-number/XBWDGZKMZ6XMFJM6JVAP2VZDEP/events.json","paper":"https://pith.science/paper/XBWDGZKM"},"agent_actions":{"view_html":"https://pith.science/pith/XBWDGZKMZ6XMFJM6JVAP2VZDEP","download_json":"https://pith.science/pith/XBWDGZKMZ6XMFJM6JVAP2VZDEP.json","view_paper":"https://pith.science/paper/XBWDGZKM","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2510.14717&json=true","fetch_graph":"https://pith.science/api/pith-number/XBWDGZKMZ6XMFJM6JVAP2VZDEP/graph.json","fetch_events":"https://pith.science/api/pith-number/XBWDGZKMZ6XMFJM6JVAP2VZDEP/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/XBWDGZKMZ6XMFJM6JVAP2VZDEP/action/timestamp_anchor","attest_storage":"https://pith.science/pith/XBWDGZKMZ6XMFJM6JVAP2VZDEP/action/storage_attestation","attest_author":"https://pith.science/pith/XBWDGZKMZ6XMFJM6JVAP2VZDEP/action/author_attestation","sign_citation":"https://pith.science/pith/XBWDGZKMZ6XMFJM6JVAP2VZDEP/action/citation_signature","submit_replication":"https://pith.science/pith/XBWDGZKMZ6XMFJM6JVAP2VZDEP/action/replication_record"}},"created_at":"2026-07-28T02:24:17.798280+00:00","updated_at":"2026-07-28T02:24:17.798280+00:00"}