{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2024:27MLAN5Q7NP3Z6ZFAUI3NPBIN2","short_pith_number":"pith:27MLAN5Q","schema_version":"1.0","canonical_sha256":"d7d8b037b0fb5fbcfb250511b6bc286eb53bf0cd2c241476c1e559435932ba91","source":{"kind":"arxiv","id":"2401.03129","version":1},"attestation_state":"computed","paper":{"title":"Examining Forgetting in Continual Pre-training of Aligned Large Language Models","license":"http://creativecommons.org/licenses/by-sa/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Chen-An Li, Hung-yi Lee","submitted_at":"2024-01-06T05:34:09Z","abstract_excerpt":"Recent advances in Large Language Models (LLMs) have exhibited remarkable proficiency across various tasks. Given the potent applications of LLMs in numerous fields, there has been a surge in LLM development. In developing LLMs, a common practice involves continual pre-training on previously fine-tuned models. However, this can lead to catastrophic forgetting. In our work, we investigate the phenomenon of forgetting that occurs during continual pre-training on an existing fine-tuned LLM. We evaluate the impact of continuous pre-training on the fine-tuned LLM across various dimensions, includin"},"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":"2401.03129","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by-sa/4.0/","primary_cat":"cs.CL","submitted_at":"2024-01-06T05:34:09Z","cross_cats_sorted":[],"title_canon_sha256":"d56d0a46bd70819aa74a114044c5cdfc655d6f9dac7d4c764dc8fe08b0837f4f","abstract_canon_sha256":"7870af7e1cc8e8d5637db68ed7a43d207849c47a377f8b055fbd99eebf80fe71"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T07:31:02.800976Z","signature_b64":"Pt9ysE2vPscRkLpzFUXTJUt8b8kg0TVf6IVLX8dYoO0S9d/joiJTSyExCAaIjQNeTftcHr9mJfX+IeSs2Jt0Cg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"d7d8b037b0fb5fbcfb250511b6bc286eb53bf0cd2c241476c1e559435932ba91","last_reissued_at":"2026-07-05T07:31:02.800578Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T07:31:02.800578Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Examining Forgetting in Continual Pre-training of Aligned Large Language Models","license":"http://creativecommons.org/licenses/by-sa/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Chen-An Li, Hung-yi Lee","submitted_at":"2024-01-06T05:34:09Z","abstract_excerpt":"Recent advances in Large Language Models (LLMs) have exhibited remarkable proficiency across various tasks. Given the potent applications of LLMs in numerous fields, there has been a surge in LLM development. In developing LLMs, a common practice involves continual pre-training on previously fine-tuned models. However, this can lead to catastrophic forgetting. In our work, we investigate the phenomenon of forgetting that occurs during continual pre-training on an existing fine-tuned LLM. We evaluate the impact of continuous pre-training on the fine-tuned LLM across various dimensions, includin"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2401.03129","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/2401.03129/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":"2401.03129","created_at":"2026-07-05T07:31:02.800633+00:00"},{"alias_kind":"arxiv_version","alias_value":"2401.03129v1","created_at":"2026-07-05T07:31:02.800633+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2401.03129","created_at":"2026-07-05T07:31:02.800633+00:00"},{"alias_kind":"pith_short_12","alias_value":"27MLAN5Q7NP3","created_at":"2026-07-05T07:31:02.800633+00:00"},{"alias_kind":"pith_short_16","alias_value":"27MLAN5Q7NP3Z6ZF","created_at":"2026-07-05T07:31:02.800633+00:00"},{"alias_kind":"pith_short_8","alias_value":"27MLAN5Q","created_at":"2026-07-05T07:31:02.800633+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":3,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.13477","citing_title":"SupraBench: A Benchmark for Supramolecular Chemistry","ref_index":54,"is_internal_anchor":false},{"citing_arxiv_id":"2605.26189","citing_title":"Max-Window Scale Estimation for Near-Lossless HiF8 W8A8 Quantization-Aware Training","ref_index":11,"is_internal_anchor":false},{"citing_arxiv_id":"2605.15053","citing_title":"TFGN: Task-Free, Replay-Free Continual Pre-Training Without Catastrophic Forgetting at LLM Scale","ref_index":43,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/27MLAN5Q7NP3Z6ZFAUI3NPBIN2","json":"https://pith.science/pith/27MLAN5Q7NP3Z6ZFAUI3NPBIN2.json","graph_json":"https://pith.science/api/pith-number/27MLAN5Q7NP3Z6ZFAUI3NPBIN2/graph.json","events_json":"https://pith.science/api/pith-number/27MLAN5Q7NP3Z6ZFAUI3NPBIN2/events.json","paper":"https://pith.science/paper/27MLAN5Q"},"agent_actions":{"view_html":"https://pith.science/pith/27MLAN5Q7NP3Z6ZFAUI3NPBIN2","download_json":"https://pith.science/pith/27MLAN5Q7NP3Z6ZFAUI3NPBIN2.json","view_paper":"https://pith.science/paper/27MLAN5Q","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2401.03129&json=true","fetch_graph":"https://pith.science/api/pith-number/27MLAN5Q7NP3Z6ZFAUI3NPBIN2/graph.json","fetch_events":"https://pith.science/api/pith-number/27MLAN5Q7NP3Z6ZFAUI3NPBIN2/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/27MLAN5Q7NP3Z6ZFAUI3NPBIN2/action/timestamp_anchor","attest_storage":"https://pith.science/pith/27MLAN5Q7NP3Z6ZFAUI3NPBIN2/action/storage_attestation","attest_author":"https://pith.science/pith/27MLAN5Q7NP3Z6ZFAUI3NPBIN2/action/author_attestation","sign_citation":"https://pith.science/pith/27MLAN5Q7NP3Z6ZFAUI3NPBIN2/action/citation_signature","submit_replication":"https://pith.science/pith/27MLAN5Q7NP3Z6ZFAUI3NPBIN2/action/replication_record"}},"created_at":"2026-07-05T07:31:02.800633+00:00","updated_at":"2026-07-05T07:31:02.800633+00:00"}