{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2024:ESUM7JWTSHI6UCSX25NK3IKK32","short_pith_number":"pith:ESUM7JWT","schema_version":"1.0","canonical_sha256":"24a8cfa6d391d1ea0a57d75aada14adea87db31fe255fd9c8e8810bce1084091","source":{"kind":"arxiv","id":"2406.04836","version":1},"attestation_state":"computed","paper":{"title":"Revisiting Catastrophic Forgetting in Large Language Model Tuning","license":"http://creativecommons.org/publicdomain/zero/1.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.CL","authors_text":"Dacheng Tao, Hongyu Li, Liang Ding, Meng Fang","submitted_at":"2024-06-07T11:09:13Z","abstract_excerpt":"Catastrophic Forgetting (CF) means models forgetting previously acquired knowledge when learning new data. It compromises the effectiveness of large language models (LLMs) during fine-tuning, yet the underlying causes have not been thoroughly investigated. This paper takes the first step to reveal the direct link between the flatness of the model loss landscape and the extent of CF in the field of LLMs. Based on this, we introduce the sharpness-aware minimization to mitigate CF by flattening the loss landscape. Experiments on three widely-used fine-tuning datasets, spanning different model sca"},"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":"2406.04836","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/publicdomain/zero/1.0/","primary_cat":"cs.CL","submitted_at":"2024-06-07T11:09:13Z","cross_cats_sorted":["cs.AI"],"title_canon_sha256":"811cb3dbbb932a116f7925900a40122cf20bd2d23939f69239ac2f87bd74ba54","abstract_canon_sha256":"25c3e159ea693a3dfb43d5ed217376db8961083e9c36ffd87c68202c07f78ae8"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T08:28:49.821372Z","signature_b64":"eLydqX37vMB9SpQN9DlXxhsNF8bn7UH8M8XMfv3l6uKRC6mxWeDs0fHoDns2ICuoucRyWl4rJ6Z0ZaH62zbBCA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"24a8cfa6d391d1ea0a57d75aada14adea87db31fe255fd9c8e8810bce1084091","last_reissued_at":"2026-07-05T08:28:49.820947Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T08:28:49.820947Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Revisiting Catastrophic Forgetting in Large Language Model Tuning","license":"http://creativecommons.org/publicdomain/zero/1.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.CL","authors_text":"Dacheng Tao, Hongyu Li, Liang Ding, Meng Fang","submitted_at":"2024-06-07T11:09:13Z","abstract_excerpt":"Catastrophic Forgetting (CF) means models forgetting previously acquired knowledge when learning new data. It compromises the effectiveness of large language models (LLMs) during fine-tuning, yet the underlying causes have not been thoroughly investigated. This paper takes the first step to reveal the direct link between the flatness of the model loss landscape and the extent of CF in the field of LLMs. Based on this, we introduce the sharpness-aware minimization to mitigate CF by flattening the loss landscape. Experiments on three widely-used fine-tuning datasets, spanning different model sca"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2406.04836","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/2406.04836/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":"2406.04836","created_at":"2026-07-05T08:28:49.821003+00:00"},{"alias_kind":"arxiv_version","alias_value":"2406.04836v1","created_at":"2026-07-05T08:28:49.821003+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2406.04836","created_at":"2026-07-05T08:28:49.821003+00:00"},{"alias_kind":"pith_short_12","alias_value":"ESUM7JWTSHI6","created_at":"2026-07-05T08:28:49.821003+00:00"},{"alias_kind":"pith_short_16","alias_value":"ESUM7JWTSHI6UCSX","created_at":"2026-07-05T08:28:49.821003+00:00"},{"alias_kind":"pith_short_8","alias_value":"ESUM7JWT","created_at":"2026-07-05T08:28:49.821003+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":4,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2601.21577","citing_title":"Collaborative Parameter Learning: Mitigating Forgetting via Parameter-Level Gradient Analysis","ref_index":11,"is_internal_anchor":false},{"citing_arxiv_id":"2605.04066","citing_title":"Adapt to Thrive! Adaptive Power-Mean Policy Optimization for Improved LLM Reasoning","ref_index":121,"is_internal_anchor":false},{"citing_arxiv_id":"2605.04065","citing_title":"Free Energy-Driven Reinforcement Learning with Adaptive Advantage Shaping for Unsupervised Reasoning in LLMs","ref_index":136,"is_internal_anchor":false},{"citing_arxiv_id":"2604.18963","citing_title":"Distillation Traps and Guards: A Calibration Knob for LLM Distillability","ref_index":2,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/ESUM7JWTSHI6UCSX25NK3IKK32","json":"https://pith.science/pith/ESUM7JWTSHI6UCSX25NK3IKK32.json","graph_json":"https://pith.science/api/pith-number/ESUM7JWTSHI6UCSX25NK3IKK32/graph.json","events_json":"https://pith.science/api/pith-number/ESUM7JWTSHI6UCSX25NK3IKK32/events.json","paper":"https://pith.science/paper/ESUM7JWT"},"agent_actions":{"view_html":"https://pith.science/pith/ESUM7JWTSHI6UCSX25NK3IKK32","download_json":"https://pith.science/pith/ESUM7JWTSHI6UCSX25NK3IKK32.json","view_paper":"https://pith.science/paper/ESUM7JWT","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2406.04836&json=true","fetch_graph":"https://pith.science/api/pith-number/ESUM7JWTSHI6UCSX25NK3IKK32/graph.json","fetch_events":"https://pith.science/api/pith-number/ESUM7JWTSHI6UCSX25NK3IKK32/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/ESUM7JWTSHI6UCSX25NK3IKK32/action/timestamp_anchor","attest_storage":"https://pith.science/pith/ESUM7JWTSHI6UCSX25NK3IKK32/action/storage_attestation","attest_author":"https://pith.science/pith/ESUM7JWTSHI6UCSX25NK3IKK32/action/author_attestation","sign_citation":"https://pith.science/pith/ESUM7JWTSHI6UCSX25NK3IKK32/action/citation_signature","submit_replication":"https://pith.science/pith/ESUM7JWTSHI6UCSX25NK3IKK32/action/replication_record"}},"created_at":"2026-07-05T08:28:49.821003+00:00","updated_at":"2026-07-05T08:28:49.821003+00:00"}