{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2024:ERDOZTRZYIJBCO7V2LXXF6IRWC","short_pith_number":"pith:ERDOZTRZ","schema_version":"1.0","canonical_sha256":"2446ecce39c212113bf5d2ef72f911b0bf045ec0b10b40f0e0adb717f6716048","source":{"kind":"arxiv","id":"2406.14563","version":1},"attestation_state":"computed","paper":{"title":"Model Merging and Safety Alignment: One Bad Model Spoils the Bunch","license":"http://creativecommons.org/licenses/by-nc-nd/4.0/","headline":"","cross_cats":["cs.AI","cs.LG"],"primary_cat":"cs.CL","authors_text":"Adel Bibi, Bernard Ghanem, Fabio Pizzati, Hasan Abed Al Kader Hammoud, Mete Ozay, Philip Torr, Umberto Michieli","submitted_at":"2024-06-20T17:59:58Z","abstract_excerpt":"Merging Large Language Models (LLMs) is a cost-effective technique for combining multiple expert LLMs into a single versatile model, retaining the expertise of the original ones. However, current approaches often overlook the importance of safety alignment during merging, leading to highly misaligned models. This work investigates the effects of model merging on alignment. We evaluate several popular model merging techniques, demonstrating that existing methods do not only transfer domain expertise but also propagate misalignment. We propose a simple two-step approach to address this problem: "},"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":"2406.14563","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by-nc-nd/4.0/","primary_cat":"cs.CL","submitted_at":"2024-06-20T17:59:58Z","cross_cats_sorted":["cs.AI","cs.LG"],"title_canon_sha256":"40f81b3fc991672e921eec3dd8f8a1c8f6be69f48b6f01795ac925cde0f6d6d1","abstract_canon_sha256":"a5fbd50355523df70826c43d0159a8a6af859999d5acc4bc425f8a08e1420a86"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T08:34:52.242795Z","signature_b64":"ypPwbx3XPySZEUrocFhVwfDsTz0qkdaGDqJZ2pIEH1vYKgJVXoYz00sK1Cu/0p6hMaxLVitPetQfBarUoUumAQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"2446ecce39c212113bf5d2ef72f911b0bf045ec0b10b40f0e0adb717f6716048","last_reissued_at":"2026-07-05T08:34:52.242211Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T08:34:52.242211Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Model Merging and Safety Alignment: One Bad Model Spoils the Bunch","license":"http://creativecommons.org/licenses/by-nc-nd/4.0/","headline":"","cross_cats":["cs.AI","cs.LG"],"primary_cat":"cs.CL","authors_text":"Adel Bibi, Bernard Ghanem, Fabio Pizzati, Hasan Abed Al Kader Hammoud, Mete Ozay, Philip Torr, Umberto Michieli","submitted_at":"2024-06-20T17:59:58Z","abstract_excerpt":"Merging Large Language Models (LLMs) is a cost-effective technique for combining multiple expert LLMs into a single versatile model, retaining the expertise of the original ones. However, current approaches often overlook the importance of safety alignment during merging, leading to highly misaligned models. This work investigates the effects of model merging on alignment. We evaluate several popular model merging techniques, demonstrating that existing methods do not only transfer domain expertise but also propagate misalignment. We propose a simple two-step approach to address this problem: "},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2406.14563","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/2406.14563/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":"2406.14563","created_at":"2026-07-05T08:34:52.242271+00:00"},{"alias_kind":"arxiv_version","alias_value":"2406.14563v1","created_at":"2026-07-05T08:34:52.242271+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2406.14563","created_at":"2026-07-05T08:34:52.242271+00:00"},{"alias_kind":"pith_short_12","alias_value":"ERDOZTRZYIJB","created_at":"2026-07-05T08:34:52.242271+00:00"},{"alias_kind":"pith_short_16","alias_value":"ERDOZTRZYIJBCO7V","created_at":"2026-07-05T08:34:52.242271+00:00"},{"alias_kind":"pith_short_8","alias_value":"ERDOZTRZ","created_at":"2026-07-05T08:34:52.242271+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":5,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2605.19409","citing_title":"Unlocking the Potential of Continual Model Merging: An ODE Perspective","ref_index":7,"is_internal_anchor":false},{"citing_arxiv_id":"2605.19409","citing_title":"Unlocking the Potential of Continual Model Merging: An ODE Perspective","ref_index":7,"is_internal_anchor":false},{"citing_arxiv_id":"2605.19409","citing_title":"Unlocking the Potential of Continual Model Merging: An ODE Perspective","ref_index":7,"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":66,"is_internal_anchor":false},{"citing_arxiv_id":"2604.02633","citing_title":"Analytic Drift Resister for Non-Exemplar Continual Graph Learning","ref_index":51,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/ERDOZTRZYIJBCO7V2LXXF6IRWC","json":"https://pith.science/pith/ERDOZTRZYIJBCO7V2LXXF6IRWC.json","graph_json":"https://pith.science/api/pith-number/ERDOZTRZYIJBCO7V2LXXF6IRWC/graph.json","events_json":"https://pith.science/api/pith-number/ERDOZTRZYIJBCO7V2LXXF6IRWC/events.json","paper":"https://pith.science/paper/ERDOZTRZ"},"agent_actions":{"view_html":"https://pith.science/pith/ERDOZTRZYIJBCO7V2LXXF6IRWC","download_json":"https://pith.science/pith/ERDOZTRZYIJBCO7V2LXXF6IRWC.json","view_paper":"https://pith.science/paper/ERDOZTRZ","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2406.14563&json=true","fetch_graph":"https://pith.science/api/pith-number/ERDOZTRZYIJBCO7V2LXXF6IRWC/graph.json","fetch_events":"https://pith.science/api/pith-number/ERDOZTRZYIJBCO7V2LXXF6IRWC/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/ERDOZTRZYIJBCO7V2LXXF6IRWC/action/timestamp_anchor","attest_storage":"https://pith.science/pith/ERDOZTRZYIJBCO7V2LXXF6IRWC/action/storage_attestation","attest_author":"https://pith.science/pith/ERDOZTRZYIJBCO7V2LXXF6IRWC/action/author_attestation","sign_citation":"https://pith.science/pith/ERDOZTRZYIJBCO7V2LXXF6IRWC/action/citation_signature","submit_replication":"https://pith.science/pith/ERDOZTRZYIJBCO7V2LXXF6IRWC/action/replication_record"}},"created_at":"2026-07-05T08:34:52.242271+00:00","updated_at":"2026-07-05T08:34:52.242271+00:00"}