{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:QKFYFXODZU3WFDASUSTKANL4QY","short_pith_number":"pith:QKFYFXOD","schema_version":"1.0","canonical_sha256":"828b82ddc3cd37628c12a4a6a0357c8615bf27df981132dc0527ccb410c8b651","source":{"kind":"arxiv","id":"2605.17787","version":1},"attestation_state":"computed","paper":{"title":"Revisiting the Adam-SGD Gap in LLM Pre-Training: The Role of Large Effective Learning Rates","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Athanasios Glentis, Chung-Yiu Yau, Dawei Li, Mingyi Hong","submitted_at":"2026-05-18T03:09:50Z","abstract_excerpt":"It is widely believed that stochastic gradient descent (SGD) performs significantly worse than adaptive optimizers such as Adam in pre-training Large Language Models (LLMs). Yet the underlying reason for this gap remains unclear. In this work, we attribute a large part of the discrepancy to SGD's inability to sustain learning rates comparable to Adam's much larger effective learning rates. Through empirical and theoretical analysis of LLM pre-training dynamics, we identify that training is characterized by small gradient norms and large weight-to-gradient ratios, an effect that becomes more pr"},"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":"2605.17787","kind":"arxiv","version":1},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.LG","submitted_at":"2026-05-18T03:09:50Z","cross_cats_sorted":[],"title_canon_sha256":"358fb531365be932999f0ea161178ce250f8ee0cbb25b031c7930c433a14d61f","abstract_canon_sha256":"7b316213f566d638a7753779aa2379a393ac9a132b04e258c0fafc07ec7cb652"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-05-20T00:04:58.314616Z","signature_b64":"CGtmcl1wcbf2mMwedEQYsdqjf+/IZPBZDNifhb17W1c33gIc3aYvBdY591SMV3hg0c1QjNJVgnmd2D01RonaAA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"828b82ddc3cd37628c12a4a6a0357c8615bf27df981132dc0527ccb410c8b651","last_reissued_at":"2026-05-20T00:04:58.313805Z","signature_status":"signed_v1","first_computed_at":"2026-05-20T00:04:58.313805Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Revisiting the Adam-SGD Gap in LLM Pre-Training: The Role of Large Effective Learning Rates","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Athanasios Glentis, Chung-Yiu Yau, Dawei Li, Mingyi Hong","submitted_at":"2026-05-18T03:09:50Z","abstract_excerpt":"It is widely believed that stochastic gradient descent (SGD) performs significantly worse than adaptive optimizers such as Adam in pre-training Large Language Models (LLMs). Yet the underlying reason for this gap remains unclear. In this work, we attribute a large part of the discrepancy to SGD's inability to sustain learning rates comparable to Adam's much larger effective learning rates. Through empirical and theoretical analysis of LLM pre-training dynamics, we identify that training is characterized by small gradient norms and large weight-to-gradient ratios, an effect that becomes more pr"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2605.17787","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/2605.17787/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":"2605.17787","created_at":"2026-05-20T00:04:58.313939+00:00"},{"alias_kind":"arxiv_version","alias_value":"2605.17787v1","created_at":"2026-05-20T00:04:58.313939+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2605.17787","created_at":"2026-05-20T00:04:58.313939+00:00"},{"alias_kind":"pith_short_12","alias_value":"QKFYFXODZU3W","created_at":"2026-05-20T00:04:58.313939+00:00"},{"alias_kind":"pith_short_16","alias_value":"QKFYFXODZU3WFDAS","created_at":"2026-05-20T00:04:58.313939+00:00"},{"alias_kind":"pith_short_8","alias_value":"QKFYFXOD","created_at":"2026-05-20T00:04:58.313939+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":0,"internal_anchor_count":0,"sample":[]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/QKFYFXODZU3WFDASUSTKANL4QY","json":"https://pith.science/pith/QKFYFXODZU3WFDASUSTKANL4QY.json","graph_json":"https://pith.science/api/pith-number/QKFYFXODZU3WFDASUSTKANL4QY/graph.json","events_json":"https://pith.science/api/pith-number/QKFYFXODZU3WFDASUSTKANL4QY/events.json","paper":"https://pith.science/paper/QKFYFXOD"},"agent_actions":{"view_html":"https://pith.science/pith/QKFYFXODZU3WFDASUSTKANL4QY","download_json":"https://pith.science/pith/QKFYFXODZU3WFDASUSTKANL4QY.json","view_paper":"https://pith.science/paper/QKFYFXOD","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2605.17787&json=true","fetch_graph":"https://pith.science/api/pith-number/QKFYFXODZU3WFDASUSTKANL4QY/graph.json","fetch_events":"https://pith.science/api/pith-number/QKFYFXODZU3WFDASUSTKANL4QY/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/QKFYFXODZU3WFDASUSTKANL4QY/action/timestamp_anchor","attest_storage":"https://pith.science/pith/QKFYFXODZU3WFDASUSTKANL4QY/action/storage_attestation","attest_author":"https://pith.science/pith/QKFYFXODZU3WFDASUSTKANL4QY/action/author_attestation","sign_citation":"https://pith.science/pith/QKFYFXODZU3WFDASUSTKANL4QY/action/citation_signature","submit_replication":"https://pith.science/pith/QKFYFXODZU3WFDASUSTKANL4QY/action/replication_record"}},"created_at":"2026-05-20T00:04:58.313939+00:00","updated_at":"2026-05-20T00:04:58.313939+00:00"}