{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:U7SHXPZNZISKMFA4DTK7A7B6EU","short_pith_number":"pith:U7SHXPZN","schema_version":"1.0","canonical_sha256":"a7e47bbf2dca24a6141c1cd5f07c3e253b63d1d5c42c9dbef32b86f0c2ca4e56","source":{"kind":"arxiv","id":"2507.08091","version":1},"attestation_state":"computed","paper":{"title":"Low-rank Momentum Factorization for Memory Efficient Training","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Mehrdad Mahdavi, Pouria Mahdavinia","submitted_at":"2025-07-10T18:04:52Z","abstract_excerpt":"Fine-tuning large foundation models presents significant memory challenges due to stateful optimizers like AdamW, often requiring several times more GPU memory than inference. While memory-efficient methods like parameter-efficient fine-tuning (e.g., LoRA) and optimizer state compression exist, recent approaches like GaLore bridge these by using low-rank gradient projections and subspace moment accumulation. However, such methods may struggle with fixed subspaces or computationally costly offline resampling (e.g., requiring full-matrix SVDs). We propose Momentum Factorized SGD (MoFaSGD), which"},"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":"2507.08091","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2025-07-10T18:04:52Z","cross_cats_sorted":[],"title_canon_sha256":"8cdffc81b94e0cd95df95e5d78966e4e4d44e8ec5be4f01826df0ecded66e054","abstract_canon_sha256":"6873464763bb7b714119525471bfdc0d9b71154254e7b2ba98d718b60a65e6ca"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T11:35:17.173459Z","signature_b64":"7Q9eYsDjAw9dN48hLi4M99I1gVa6WWVhrfA4yrDYYPrFe9xoku3IVByjiF9F3ns7Gf5ouzaCXS6lvOdjkmy5AA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"a7e47bbf2dca24a6141c1cd5f07c3e253b63d1d5c42c9dbef32b86f0c2ca4e56","last_reissued_at":"2026-07-05T11:35:17.172965Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T11:35:17.172965Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Low-rank Momentum Factorization for Memory Efficient Training","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Mehrdad Mahdavi, Pouria Mahdavinia","submitted_at":"2025-07-10T18:04:52Z","abstract_excerpt":"Fine-tuning large foundation models presents significant memory challenges due to stateful optimizers like AdamW, often requiring several times more GPU memory than inference. While memory-efficient methods like parameter-efficient fine-tuning (e.g., LoRA) and optimizer state compression exist, recent approaches like GaLore bridge these by using low-rank gradient projections and subspace moment accumulation. However, such methods may struggle with fixed subspaces or computationally costly offline resampling (e.g., requiring full-matrix SVDs). We propose Momentum Factorized SGD (MoFaSGD), which"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2507.08091","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/2507.08091/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":"2507.08091","created_at":"2026-07-05T11:35:17.173021+00:00"},{"alias_kind":"arxiv_version","alias_value":"2507.08091v1","created_at":"2026-07-05T11:35:17.173021+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2507.08091","created_at":"2026-07-05T11:35:17.173021+00:00"},{"alias_kind":"pith_short_12","alias_value":"U7SHXPZNZISK","created_at":"2026-07-05T11:35:17.173021+00:00"},{"alias_kind":"pith_short_16","alias_value":"U7SHXPZNZISKMFA4","created_at":"2026-07-05T11:35:17.173021+00:00"},{"alias_kind":"pith_short_8","alias_value":"U7SHXPZN","created_at":"2026-07-05T11:35:17.173021+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/U7SHXPZNZISKMFA4DTK7A7B6EU","json":"https://pith.science/pith/U7SHXPZNZISKMFA4DTK7A7B6EU.json","graph_json":"https://pith.science/api/pith-number/U7SHXPZNZISKMFA4DTK7A7B6EU/graph.json","events_json":"https://pith.science/api/pith-number/U7SHXPZNZISKMFA4DTK7A7B6EU/events.json","paper":"https://pith.science/paper/U7SHXPZN"},"agent_actions":{"view_html":"https://pith.science/pith/U7SHXPZNZISKMFA4DTK7A7B6EU","download_json":"https://pith.science/pith/U7SHXPZNZISKMFA4DTK7A7B6EU.json","view_paper":"https://pith.science/paper/U7SHXPZN","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2507.08091&json=true","fetch_graph":"https://pith.science/api/pith-number/U7SHXPZNZISKMFA4DTK7A7B6EU/graph.json","fetch_events":"https://pith.science/api/pith-number/U7SHXPZNZISKMFA4DTK7A7B6EU/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/U7SHXPZNZISKMFA4DTK7A7B6EU/action/timestamp_anchor","attest_storage":"https://pith.science/pith/U7SHXPZNZISKMFA4DTK7A7B6EU/action/storage_attestation","attest_author":"https://pith.science/pith/U7SHXPZNZISKMFA4DTK7A7B6EU/action/author_attestation","sign_citation":"https://pith.science/pith/U7SHXPZNZISKMFA4DTK7A7B6EU/action/citation_signature","submit_replication":"https://pith.science/pith/U7SHXPZNZISKMFA4DTK7A7B6EU/action/replication_record"}},"created_at":"2026-07-05T11:35:17.173021+00:00","updated_at":"2026-07-05T11:35:17.173021+00:00"}