{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2024:7G2Q365IUTHTYAOUZZSVZKSRDT","short_pith_number":"pith:7G2Q365I","schema_version":"1.0","canonical_sha256":"f9b50dfba8a4cf3c01d4ce655caa511cdac01760fbd184b5c65c7ad6ff823ef6","source":{"kind":"arxiv","id":"2402.11746","version":1},"attestation_state":"computed","paper":{"title":"Language Models are Homer Simpson! Safety Re-Alignment of Fine-tuned Language Models through Task Arithmetic","license":"http://creativecommons.org/licenses/by-sa/4.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.CL","authors_text":"Do Duc Anh, Rishabh Bhardwaj, Soujanya Poria","submitted_at":"2024-02-19T00:18:09Z","abstract_excerpt":"Aligned language models face a significant limitation as their fine-tuning often results in compromised safety. To tackle this, we propose a simple method RESTA that performs LLM safety realignment. RESTA stands for REstoring Safety through Task Arithmetic. At its core, it involves a simple arithmetic addition of a safety vector to the weights of the compromised model. We demonstrate the effectiveness of RESTA in both parameter-efficient and full fine-tuning, covering a wide range of downstream tasks, including instruction following in Chinese, English, and Hindi, as well as problem-solving ca"},"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":"2402.11746","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by-sa/4.0/","primary_cat":"cs.CL","submitted_at":"2024-02-19T00:18:09Z","cross_cats_sorted":["cs.AI"],"title_canon_sha256":"c8db9945248fd4a745164e352c471014c3396dbee161ff3981c3ca5b6674d07b","abstract_canon_sha256":"d145286f69633d40d9d07a6a55a960a42d47b6960c9f8e492ab499a12fc34d34"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T07:46:43.859730Z","signature_b64":"E1GiwwXwnsLBTxpUMNF5bs+i2rzSBl64JwyDqYO10eWmrhBovPi04XkVxgQuE+lB1kxGTSCniPCAq4Rb+ogdBw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"f9b50dfba8a4cf3c01d4ce655caa511cdac01760fbd184b5c65c7ad6ff823ef6","last_reissued_at":"2026-07-05T07:46:43.859343Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T07:46:43.859343Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Language Models are Homer Simpson! Safety Re-Alignment of Fine-tuned Language Models through Task Arithmetic","license":"http://creativecommons.org/licenses/by-sa/4.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.CL","authors_text":"Do Duc Anh, Rishabh Bhardwaj, Soujanya Poria","submitted_at":"2024-02-19T00:18:09Z","abstract_excerpt":"Aligned language models face a significant limitation as their fine-tuning often results in compromised safety. To tackle this, we propose a simple method RESTA that performs LLM safety realignment. RESTA stands for REstoring Safety through Task Arithmetic. At its core, it involves a simple arithmetic addition of a safety vector to the weights of the compromised model. We demonstrate the effectiveness of RESTA in both parameter-efficient and full fine-tuning, covering a wide range of downstream tasks, including instruction following in Chinese, English, and Hindi, as well as problem-solving ca"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2402.11746","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/2402.11746/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":"2402.11746","created_at":"2026-07-05T07:46:43.859397+00:00"},{"alias_kind":"arxiv_version","alias_value":"2402.11746v1","created_at":"2026-07-05T07:46:43.859397+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2402.11746","created_at":"2026-07-05T07:46:43.859397+00:00"},{"alias_kind":"pith_short_12","alias_value":"7G2Q365IUTHT","created_at":"2026-07-05T07:46:43.859397+00:00"},{"alias_kind":"pith_short_16","alias_value":"7G2Q365IUTHTYAOU","created_at":"2026-07-05T07:46:43.859397+00:00"},{"alias_kind":"pith_short_8","alias_value":"7G2Q365I","created_at":"2026-07-05T07:46:43.859397+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":3,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2508.11548","citing_title":"Copyright Protection for Large Language Models: A Survey of Methods, Challenges, and Trends","ref_index":13,"is_internal_anchor":false},{"citing_arxiv_id":"2408.07666","citing_title":"Model Merging in LLMs, MLLMs, and Beyond: Methods, Theories, Applications and Opportunities","ref_index":13,"is_internal_anchor":false},{"citing_arxiv_id":"2403.13372","citing_title":"LlamaFactory: Unified Efficient Fine-Tuning of 100+ Language Models","ref_index":1,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/7G2Q365IUTHTYAOUZZSVZKSRDT","json":"https://pith.science/pith/7G2Q365IUTHTYAOUZZSVZKSRDT.json","graph_json":"https://pith.science/api/pith-number/7G2Q365IUTHTYAOUZZSVZKSRDT/graph.json","events_json":"https://pith.science/api/pith-number/7G2Q365IUTHTYAOUZZSVZKSRDT/events.json","paper":"https://pith.science/paper/7G2Q365I"},"agent_actions":{"view_html":"https://pith.science/pith/7G2Q365IUTHTYAOUZZSVZKSRDT","download_json":"https://pith.science/pith/7G2Q365IUTHTYAOUZZSVZKSRDT.json","view_paper":"https://pith.science/paper/7G2Q365I","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2402.11746&json=true","fetch_graph":"https://pith.science/api/pith-number/7G2Q365IUTHTYAOUZZSVZKSRDT/graph.json","fetch_events":"https://pith.science/api/pith-number/7G2Q365IUTHTYAOUZZSVZKSRDT/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/7G2Q365IUTHTYAOUZZSVZKSRDT/action/timestamp_anchor","attest_storage":"https://pith.science/pith/7G2Q365IUTHTYAOUZZSVZKSRDT/action/storage_attestation","attest_author":"https://pith.science/pith/7G2Q365IUTHTYAOUZZSVZKSRDT/action/author_attestation","sign_citation":"https://pith.science/pith/7G2Q365IUTHTYAOUZZSVZKSRDT/action/citation_signature","submit_replication":"https://pith.science/pith/7G2Q365IUTHTYAOUZZSVZKSRDT/action/replication_record"}},"created_at":"2026-07-05T07:46:43.859397+00:00","updated_at":"2026-07-05T07:46:43.859397+00:00"}