{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:BQTNPBM35I3WEYLMDT4BF2FEDX","short_pith_number":"pith:BQTNPBM3","schema_version":"1.0","canonical_sha256":"0c26d7859bea3762616c1cf812e8a41dcba6011b1b13f20f995cf502d64a9602","source":{"kind":"arxiv","id":"2502.01235","version":3},"attestation_state":"computed","paper":{"title":"LoRA-One: One-Step Full Gradient Could Suffice for Fine-Tuning Large Language Models, Provably and Efficiently","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI","cs.LG"],"primary_cat":"stat.ML","authors_text":"Fanghui Liu, Yuanhe Zhang, Yudong Chen","submitted_at":"2025-02-03T10:50:03Z","abstract_excerpt":"This paper explores how theory can guide and enhance practical algorithms, using Low-Rank Adaptation (LoRA, Hu et al. 2022) in large language models as a case study. We rigorously prove that, under gradient descent, LoRA adapters align with specific singular subspaces of the one-step full fine-tuning gradient. This result suggests that, by properly initializing the adapters using the one-step full gradient, subspace alignment can be achieved immediately and applicable to both linear and nonlinear models. Building on our theory, we propose a theory-driven algorithm, LoRA-One, where the linear 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":"2502.01235","kind":"arxiv","version":3},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"stat.ML","submitted_at":"2025-02-03T10:50:03Z","cross_cats_sorted":["cs.AI","cs.LG"],"title_canon_sha256":"124aff6d4354a7adefa8f84b70b647fcfbede3d339ddd50406f70ae173cc1f94","abstract_canon_sha256":"c14b696d05eef0e1c1522591244dea2b5a11fabd92082323da41430dcd29dcea"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T11:25:41.705503Z","signature_b64":"8yiM8Vjd/OipY0TWtaPq0kg5iZKBmLcgv3Hbo9gLHOKwI1pP3cXnMUqPzaLaEGwHEzhx6KgI/GI/DheeeqqcCQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"0c26d7859bea3762616c1cf812e8a41dcba6011b1b13f20f995cf502d64a9602","last_reissued_at":"2026-07-05T11:25:41.705045Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T11:25:41.705045Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"LoRA-One: One-Step Full Gradient Could Suffice for Fine-Tuning Large Language Models, Provably and Efficiently","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI","cs.LG"],"primary_cat":"stat.ML","authors_text":"Fanghui Liu, Yuanhe Zhang, Yudong Chen","submitted_at":"2025-02-03T10:50:03Z","abstract_excerpt":"This paper explores how theory can guide and enhance practical algorithms, using Low-Rank Adaptation (LoRA, Hu et al. 2022) in large language models as a case study. We rigorously prove that, under gradient descent, LoRA adapters align with specific singular subspaces of the one-step full fine-tuning gradient. This result suggests that, by properly initializing the adapters using the one-step full gradient, subspace alignment can be achieved immediately and applicable to both linear and nonlinear models. Building on our theory, we propose a theory-driven algorithm, LoRA-One, where the linear c"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2502.01235","kind":"arxiv","version":3},"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/2502.01235/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":"2502.01235","created_at":"2026-07-05T11:25:41.705109+00:00"},{"alias_kind":"arxiv_version","alias_value":"2502.01235v3","created_at":"2026-07-05T11:25:41.705109+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2502.01235","created_at":"2026-07-05T11:25:41.705109+00:00"},{"alias_kind":"pith_short_12","alias_value":"BQTNPBM35I3W","created_at":"2026-07-05T11:25:41.705109+00:00"},{"alias_kind":"pith_short_16","alias_value":"BQTNPBM35I3WEYLM","created_at":"2026-07-05T11:25:41.705109+00:00"},{"alias_kind":"pith_short_8","alias_value":"BQTNPBM3","created_at":"2026-07-05T11:25:41.705109+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":3,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.22019","citing_title":"Channel Location Constrains the Auditability of Subliminal Learning","ref_index":38,"is_internal_anchor":false},{"citing_arxiv_id":"2606.12883","citing_title":"The Hidden Power of Scaling Factor in LoRA Optimization","ref_index":80,"is_internal_anchor":false},{"citing_arxiv_id":"2606.05899","citing_title":"High-Dimensional Theory of LoRA Fine-Tuning in a Solvable Attention Model","ref_index":7,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/BQTNPBM35I3WEYLMDT4BF2FEDX","json":"https://pith.science/pith/BQTNPBM35I3WEYLMDT4BF2FEDX.json","graph_json":"https://pith.science/api/pith-number/BQTNPBM35I3WEYLMDT4BF2FEDX/graph.json","events_json":"https://pith.science/api/pith-number/BQTNPBM35I3WEYLMDT4BF2FEDX/events.json","paper":"https://pith.science/paper/BQTNPBM3"},"agent_actions":{"view_html":"https://pith.science/pith/BQTNPBM35I3WEYLMDT4BF2FEDX","download_json":"https://pith.science/pith/BQTNPBM35I3WEYLMDT4BF2FEDX.json","view_paper":"https://pith.science/paper/BQTNPBM3","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2502.01235&json=true","fetch_graph":"https://pith.science/api/pith-number/BQTNPBM35I3WEYLMDT4BF2FEDX/graph.json","fetch_events":"https://pith.science/api/pith-number/BQTNPBM35I3WEYLMDT4BF2FEDX/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/BQTNPBM35I3WEYLMDT4BF2FEDX/action/timestamp_anchor","attest_storage":"https://pith.science/pith/BQTNPBM35I3WEYLMDT4BF2FEDX/action/storage_attestation","attest_author":"https://pith.science/pith/BQTNPBM35I3WEYLMDT4BF2FEDX/action/author_attestation","sign_citation":"https://pith.science/pith/BQTNPBM35I3WEYLMDT4BF2FEDX/action/citation_signature","submit_replication":"https://pith.science/pith/BQTNPBM35I3WEYLMDT4BF2FEDX/action/replication_record"}},"created_at":"2026-07-05T11:25:41.705109+00:00","updated_at":"2026-07-05T11:25:41.705109+00:00"}