{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2024:TT3DJBMZHFOGD5WHXFHCSKP7FO","short_pith_number":"pith:TT3DJBMZ","schema_version":"1.0","canonical_sha256":"9cf6348599395c61f6c7b94e2929ff2b96fa1ec0c3b26c4da2605d6e74df3466","source":{"kind":"arxiv","id":"2401.05605","version":1},"attestation_state":"computed","paper":{"title":"Scaling Laws for Forgetting When Fine-Tuning Large Language Models","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.LG"],"primary_cat":"cs.CL","authors_text":"Damjan Kalajdzievski","submitted_at":"2024-01-11T00:44:25Z","abstract_excerpt":"We study and quantify the problem of forgetting when fine-tuning pre-trained large language models (LLMs) on a downstream task. We find that parameter-efficient fine-tuning (PEFT) strategies, such as Low-Rank Adapters (LoRA), still suffer from catastrophic forgetting. In particular, we identify a strong inverse linear relationship between the fine-tuning performance and the amount of forgetting when fine-tuning LLMs with LoRA. We further obtain precise scaling laws that show forgetting increases as a shifted power law in the number of parameters fine-tuned and the number of update steps. We al"},"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":"2401.05605","kind":"arxiv","version":1},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CL","submitted_at":"2024-01-11T00:44:25Z","cross_cats_sorted":["cs.LG"],"title_canon_sha256":"1ef8872d3189536839092e68fd7b2a6ca0808df759b7739f3a4eee6c9c3fe8cd","abstract_canon_sha256":"84393d0583d70ceb2d88bd4be2c705a94a52f775ea51a83f51021873cf7c4acf"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T07:32:32.516784Z","signature_b64":"v9eSr95Wj1b4T8p40Ll71ikDl8zgeARRIqjlzrFp3uCrWrAFDwxEx362x61cpwAkz7Sl5UGH0qoqINt+om7ACA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"9cf6348599395c61f6c7b94e2929ff2b96fa1ec0c3b26c4da2605d6e74df3466","last_reissued_at":"2026-07-05T07:32:32.516277Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T07:32:32.516277Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Scaling Laws for Forgetting When Fine-Tuning Large Language Models","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.LG"],"primary_cat":"cs.CL","authors_text":"Damjan Kalajdzievski","submitted_at":"2024-01-11T00:44:25Z","abstract_excerpt":"We study and quantify the problem of forgetting when fine-tuning pre-trained large language models (LLMs) on a downstream task. We find that parameter-efficient fine-tuning (PEFT) strategies, such as Low-Rank Adapters (LoRA), still suffer from catastrophic forgetting. In particular, we identify a strong inverse linear relationship between the fine-tuning performance and the amount of forgetting when fine-tuning LLMs with LoRA. We further obtain precise scaling laws that show forgetting increases as a shifted power law in the number of parameters fine-tuned and the number of update steps. We al"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2401.05605","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/2401.05605/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":"2401.05605","created_at":"2026-07-05T07:32:32.516350+00:00"},{"alias_kind":"arxiv_version","alias_value":"2401.05605v1","created_at":"2026-07-05T07:32:32.516350+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2401.05605","created_at":"2026-07-05T07:32:32.516350+00:00"},{"alias_kind":"pith_short_12","alias_value":"TT3DJBMZHFOG","created_at":"2026-07-05T07:32:32.516350+00:00"},{"alias_kind":"pith_short_16","alias_value":"TT3DJBMZHFOGD5WH","created_at":"2026-07-05T07:32:32.516350+00:00"},{"alias_kind":"pith_short_8","alias_value":"TT3DJBMZ","created_at":"2026-07-05T07:32:32.516350+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":8,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.12633","citing_title":"ECA: Efficient Continual Alignment for Open-Ended Image-to-Text Generation","ref_index":25,"is_internal_anchor":false},{"citing_arxiv_id":"2605.27564","citing_title":"The Future of Facts: Tracing the Factual Generation-Verification Gap","ref_index":114,"is_internal_anchor":false},{"citing_arxiv_id":"2605.20005","citing_title":"Fine-Tuning Without Forgetting via Loss-Adaptive Learning Rates","ref_index":23,"is_internal_anchor":false},{"citing_arxiv_id":"2511.01831","citing_title":"Routing-Based Continual Learning for Multimodal Large Language Models","ref_index":21,"is_internal_anchor":false},{"citing_arxiv_id":"2605.12484","citing_title":"Learning, Fast and Slow: Towards LLMs That Adapt Continually","ref_index":25,"is_internal_anchor":false},{"citing_arxiv_id":"2605.12484","citing_title":"Learning, Fast and Slow: Towards LLMs That Adapt Continually","ref_index":25,"is_internal_anchor":false},{"citing_arxiv_id":"2605.10468","citing_title":"Can Muon Fine-tune Adam-Pretrained Models?","ref_index":84,"is_internal_anchor":false},{"citing_arxiv_id":"2605.01077","citing_title":"Teaching LLMs Brazilian Healthcare: Injecting Knowledge from Official Clinical Guidelines","ref_index":13,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/TT3DJBMZHFOGD5WHXFHCSKP7FO","json":"https://pith.science/pith/TT3DJBMZHFOGD5WHXFHCSKP7FO.json","graph_json":"https://pith.science/api/pith-number/TT3DJBMZHFOGD5WHXFHCSKP7FO/graph.json","events_json":"https://pith.science/api/pith-number/TT3DJBMZHFOGD5WHXFHCSKP7FO/events.json","paper":"https://pith.science/paper/TT3DJBMZ"},"agent_actions":{"view_html":"https://pith.science/pith/TT3DJBMZHFOGD5WHXFHCSKP7FO","download_json":"https://pith.science/pith/TT3DJBMZHFOGD5WHXFHCSKP7FO.json","view_paper":"https://pith.science/paper/TT3DJBMZ","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2401.05605&json=true","fetch_graph":"https://pith.science/api/pith-number/TT3DJBMZHFOGD5WHXFHCSKP7FO/graph.json","fetch_events":"https://pith.science/api/pith-number/TT3DJBMZHFOGD5WHXFHCSKP7FO/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/TT3DJBMZHFOGD5WHXFHCSKP7FO/action/timestamp_anchor","attest_storage":"https://pith.science/pith/TT3DJBMZHFOGD5WHXFHCSKP7FO/action/storage_attestation","attest_author":"https://pith.science/pith/TT3DJBMZHFOGD5WHXFHCSKP7FO/action/author_attestation","sign_citation":"https://pith.science/pith/TT3DJBMZHFOGD5WHXFHCSKP7FO/action/citation_signature","submit_replication":"https://pith.science/pith/TT3DJBMZHFOGD5WHXFHCSKP7FO/action/replication_record"}},"created_at":"2026-07-05T07:32:32.516350+00:00","updated_at":"2026-07-05T07:32:32.516350+00:00"}