{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2024:GMG7HBXCH7WT7BRCXPYNL2V3AH","short_pith_number":"pith:GMG7HBXC","schema_version":"1.0","canonical_sha256":"330df386e23fed3f8622bbf0d5eabb01c8bc2a7af775b840d56ae3773a98d26b","source":{"kind":"arxiv","id":"2408.13656","version":2},"attestation_state":"computed","paper":{"title":"Localize-and-Stitch: Efficient Model Merging via Sparse Task Arithmetic","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.CL","cs.CV"],"primary_cat":"cs.LG","authors_text":"Han Zhao, Tong Zhang, Yifei He, Yong Lin, Yuzheng Hu","submitted_at":"2024-08-24T19:14:02Z","abstract_excerpt":"Model merging offers an effective strategy to combine the strengths of multiple finetuned models into a unified model that preserves the specialized capabilities of each. Existing methods merge models in a global manner, performing arithmetic operations across all model parameters. However, such global merging often leads to task interference, degrading the performance of the merged model. In this work, we introduce Localize-and-Stitch, a novel approach that merges models in a localized way. Our algorithm works in two steps: i) Localization: identify tiny ($1\\%$ of the total parameters) locali"},"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":"2408.13656","kind":"arxiv","version":2},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.LG","submitted_at":"2024-08-24T19:14:02Z","cross_cats_sorted":["cs.CL","cs.CV"],"title_canon_sha256":"afa32c134d9dbd5ca544bcde9d81433699eb2766f5ab2dad410aded497b92361","abstract_canon_sha256":"2a970dea9a93431d1ef2755df7a05756aa7acba2ece439dc4f08d360f15410d5"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T09:57:45.234138Z","signature_b64":"NdY/JAcos2HitpZ3G93+XMgXGndKykkopXJlM5tJa4alP8dZFXqg2jQrmVcO5nYaHWAIWS9bcMGG2hPSpXqNCA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"330df386e23fed3f8622bbf0d5eabb01c8bc2a7af775b840d56ae3773a98d26b","last_reissued_at":"2026-07-05T09:57:45.233593Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T09:57:45.233593Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Localize-and-Stitch: Efficient Model Merging via Sparse Task Arithmetic","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.CL","cs.CV"],"primary_cat":"cs.LG","authors_text":"Han Zhao, Tong Zhang, Yifei He, Yong Lin, Yuzheng Hu","submitted_at":"2024-08-24T19:14:02Z","abstract_excerpt":"Model merging offers an effective strategy to combine the strengths of multiple finetuned models into a unified model that preserves the specialized capabilities of each. Existing methods merge models in a global manner, performing arithmetic operations across all model parameters. However, such global merging often leads to task interference, degrading the performance of the merged model. In this work, we introduce Localize-and-Stitch, a novel approach that merges models in a localized way. Our algorithm works in two steps: i) Localization: identify tiny ($1\\%$ of the total parameters) locali"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2408.13656","kind":"arxiv","version":2},"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/2408.13656/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":"2408.13656","created_at":"2026-07-05T09:57:45.233678+00:00"},{"alias_kind":"arxiv_version","alias_value":"2408.13656v2","created_at":"2026-07-05T09:57:45.233678+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2408.13656","created_at":"2026-07-05T09:57:45.233678+00:00"},{"alias_kind":"pith_short_12","alias_value":"GMG7HBXCH7WT","created_at":"2026-07-05T09:57:45.233678+00:00"},{"alias_kind":"pith_short_16","alias_value":"GMG7HBXCH7WT7BRC","created_at":"2026-07-05T09:57:45.233678+00:00"},{"alias_kind":"pith_short_8","alias_value":"GMG7HBXC","created_at":"2026-07-05T09:57:45.233678+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":5,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2607.01686","citing_title":"WARP: Weight-Space Analysis for Recovering Training Data Portfolios","ref_index":33,"is_internal_anchor":false},{"citing_arxiv_id":"2605.28649","citing_title":"Interpretability-Guided Layer Selection over Subspace Projection: SAEs as Stethoscopes, Not Scalpels, for Raw Task Vector Model Editing","ref_index":15,"is_internal_anchor":false},{"citing_arxiv_id":"2604.01674","citing_title":"Can Heterogeneous Language Models Be Fused?","ref_index":33,"is_internal_anchor":false},{"citing_arxiv_id":"2605.12960","citing_title":"DiM\\textsuperscript{3}: Bridging Multilingual and Multimodal Models via Direction- and Magnitude-Aware Merging","ref_index":48,"is_internal_anchor":false},{"citing_arxiv_id":"2605.12960","citing_title":"DiM\\textsuperscript{3}: Bridging Multilingual and Multimodal Models via Direction- and Magnitude-Aware Merging","ref_index":48,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/GMG7HBXCH7WT7BRCXPYNL2V3AH","json":"https://pith.science/pith/GMG7HBXCH7WT7BRCXPYNL2V3AH.json","graph_json":"https://pith.science/api/pith-number/GMG7HBXCH7WT7BRCXPYNL2V3AH/graph.json","events_json":"https://pith.science/api/pith-number/GMG7HBXCH7WT7BRCXPYNL2V3AH/events.json","paper":"https://pith.science/paper/GMG7HBXC"},"agent_actions":{"view_html":"https://pith.science/pith/GMG7HBXCH7WT7BRCXPYNL2V3AH","download_json":"https://pith.science/pith/GMG7HBXCH7WT7BRCXPYNL2V3AH.json","view_paper":"https://pith.science/paper/GMG7HBXC","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2408.13656&json=true","fetch_graph":"https://pith.science/api/pith-number/GMG7HBXCH7WT7BRCXPYNL2V3AH/graph.json","fetch_events":"https://pith.science/api/pith-number/GMG7HBXCH7WT7BRCXPYNL2V3AH/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/GMG7HBXCH7WT7BRCXPYNL2V3AH/action/timestamp_anchor","attest_storage":"https://pith.science/pith/GMG7HBXCH7WT7BRCXPYNL2V3AH/action/storage_attestation","attest_author":"https://pith.science/pith/GMG7HBXCH7WT7BRCXPYNL2V3AH/action/author_attestation","sign_citation":"https://pith.science/pith/GMG7HBXCH7WT7BRCXPYNL2V3AH/action/citation_signature","submit_replication":"https://pith.science/pith/GMG7HBXCH7WT7BRCXPYNL2V3AH/action/replication_record"}},"created_at":"2026-07-05T09:57:45.233678+00:00","updated_at":"2026-07-05T09:57:45.233678+00:00"}