{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:ACZWA73V6VDOKUI3QGAFNN2CQ6","short_pith_number":"pith:ACZWA73V","schema_version":"1.0","canonical_sha256":"00b3607f75f546e5511b818056b74287a6c683b065501a62b4c25991d061762a","source":{"kind":"arxiv","id":"2502.07752","version":2},"attestation_state":"computed","paper":{"title":"Towards Efficient Optimizer Design for LLM via Structured Fisher Approximation with a Low-Rank Extension","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.AI","stat.ML"],"primary_cat":"cs.LG","authors_text":"Chao Ma, Edward Meeds, Meyer Scetbon, Wenbo Gong","submitted_at":"2025-02-11T18:27:19Z","abstract_excerpt":"Designing efficient optimizers for large language models (LLMs) with low-memory requirements and fast convergence is an important and challenging problem. This paper makes a step towards the systematic design of such optimizers through the lens of structured Fisher information matrix (FIM) approximation. We show that many state-of-the-art efficient optimizers can be viewed as solutions to FIM approximation (under the Frobenius norm) with specific structural assumptions. Building on these insights, we propose two design recommendations of practical efficient optimizers for LLMs, involving the c"},"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":"2502.07752","kind":"arxiv","version":2},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.LG","submitted_at":"2025-02-11T18:27:19Z","cross_cats_sorted":["cs.AI","stat.ML"],"title_canon_sha256":"d28b6829c8155a0ae9c64f62d15455913b6ed4ffbd529b2cb42fbd8b4225ea07","abstract_canon_sha256":"bcbb5d100a90db8a2c71bd45d87da1c39cd1e631d7e932c1e9c4a16662c4d6d3"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T10:17:16.594555Z","signature_b64":"3Z70T4it+89poYKS7GtJZ26/NfTsGyhMaMH2pJXS1UBoxIcg2s22PcHFzGXcU+ubzbQ5VVCLuKIxJttX2qKlBg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"00b3607f75f546e5511b818056b74287a6c683b065501a62b4c25991d061762a","last_reissued_at":"2026-07-05T10:17:16.594079Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T10:17:16.594079Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Towards Efficient Optimizer Design for LLM via Structured Fisher Approximation with a Low-Rank Extension","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.AI","stat.ML"],"primary_cat":"cs.LG","authors_text":"Chao Ma, Edward Meeds, Meyer Scetbon, Wenbo Gong","submitted_at":"2025-02-11T18:27:19Z","abstract_excerpt":"Designing efficient optimizers for large language models (LLMs) with low-memory requirements and fast convergence is an important and challenging problem. This paper makes a step towards the systematic design of such optimizers through the lens of structured Fisher information matrix (FIM) approximation. We show that many state-of-the-art efficient optimizers can be viewed as solutions to FIM approximation (under the Frobenius norm) with specific structural assumptions. Building on these insights, we propose two design recommendations of practical efficient optimizers for LLMs, involving the c"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2502.07752","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/2502.07752/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":"2502.07752","created_at":"2026-07-05T10:17:16.594143+00:00"},{"alias_kind":"arxiv_version","alias_value":"2502.07752v2","created_at":"2026-07-05T10:17:16.594143+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2502.07752","created_at":"2026-07-05T10:17:16.594143+00:00"},{"alias_kind":"pith_short_12","alias_value":"ACZWA73V6VDO","created_at":"2026-07-05T10:17:16.594143+00:00"},{"alias_kind":"pith_short_16","alias_value":"ACZWA73V6VDOKUI3","created_at":"2026-07-05T10:17:16.594143+00:00"},{"alias_kind":"pith_short_8","alias_value":"ACZWA73V","created_at":"2026-07-05T10:17:16.594143+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":1,"internal_anchor_count":1,"sample":[{"citing_arxiv_id":"2607.05872","citing_title":"No Subspace to Track: Non-Identifiability and Optimizer State in Low-Rank Training","ref_index":5,"is_internal_anchor":true}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/ACZWA73V6VDOKUI3QGAFNN2CQ6","json":"https://pith.science/pith/ACZWA73V6VDOKUI3QGAFNN2CQ6.json","graph_json":"https://pith.science/api/pith-number/ACZWA73V6VDOKUI3QGAFNN2CQ6/graph.json","events_json":"https://pith.science/api/pith-number/ACZWA73V6VDOKUI3QGAFNN2CQ6/events.json","paper":"https://pith.science/paper/ACZWA73V"},"agent_actions":{"view_html":"https://pith.science/pith/ACZWA73V6VDOKUI3QGAFNN2CQ6","download_json":"https://pith.science/pith/ACZWA73V6VDOKUI3QGAFNN2CQ6.json","view_paper":"https://pith.science/paper/ACZWA73V","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2502.07752&json=true","fetch_graph":"https://pith.science/api/pith-number/ACZWA73V6VDOKUI3QGAFNN2CQ6/graph.json","fetch_events":"https://pith.science/api/pith-number/ACZWA73V6VDOKUI3QGAFNN2CQ6/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/ACZWA73V6VDOKUI3QGAFNN2CQ6/action/timestamp_anchor","attest_storage":"https://pith.science/pith/ACZWA73V6VDOKUI3QGAFNN2CQ6/action/storage_attestation","attest_author":"https://pith.science/pith/ACZWA73V6VDOKUI3QGAFNN2CQ6/action/author_attestation","sign_citation":"https://pith.science/pith/ACZWA73V6VDOKUI3QGAFNN2CQ6/action/citation_signature","submit_replication":"https://pith.science/pith/ACZWA73V6VDOKUI3QGAFNN2CQ6/action/replication_record"}},"created_at":"2026-07-05T10:17:16.594143+00:00","updated_at":"2026-07-05T10:17:16.594143+00:00"}