{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2023:FNJTUALT7KW4K7ZC74ZZMNUCEK","short_pith_number":"pith:FNJTUALT","schema_version":"1.0","canonical_sha256":"2b533a0173faadc57f22ff3396368222b14936a26f006368a66c97065d990b40","source":{"kind":"arxiv","id":"2305.03053","version":3},"attestation_state":"computed","paper":{"title":"ZipIt! Merging Models from Different Tasks without Training","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.LG"],"primary_cat":"cs.CV","authors_text":"Daniel Bolya, George Stoica, Jakob Bjorner, Judy Hoffman, Pratik Ramesh, Taylor Hearn","submitted_at":"2023-05-04T17:59:58Z","abstract_excerpt":"Typical deep visual recognition models are capable of performing the one task they were trained on. In this paper, we tackle the extremely difficult problem of combining distinct models with different initializations, each solving a separate task, into one multi-task model without any additional training. Prior work in model merging permutes one model to the space of the other then averages them together. While this works for models trained on the same task, we find that this fails to account for the differences in models trained on disjoint tasks. Thus, we introduce \"ZipIt!\", a general method"},"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":"2305.03053","kind":"arxiv","version":3},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CV","submitted_at":"2023-05-04T17:59:58Z","cross_cats_sorted":["cs.LG"],"title_canon_sha256":"ed0805d79c70e6fa980f097bf4e4ac81fff4036c63179b91ad9445ca6bd725bf","abstract_canon_sha256":"bdd5a5fca4ed3d4a0e013dee09ab24c46d1e9f6f8f94ea4c628a1bd8891fdf18"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T07:55:21.306366Z","signature_b64":"RdlpeNBX2dtw0W9b3/yzy+l/bnAnXs1lOotCnIQ3FJpDSCjCUA42lo1TmnCPcbPJuoOHg8sTnnYWE3FKaskNBg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"2b533a0173faadc57f22ff3396368222b14936a26f006368a66c97065d990b40","last_reissued_at":"2026-07-05T07:55:21.305898Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T07:55:21.305898Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"ZipIt! Merging Models from Different Tasks without Training","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.LG"],"primary_cat":"cs.CV","authors_text":"Daniel Bolya, George Stoica, Jakob Bjorner, Judy Hoffman, Pratik Ramesh, Taylor Hearn","submitted_at":"2023-05-04T17:59:58Z","abstract_excerpt":"Typical deep visual recognition models are capable of performing the one task they were trained on. In this paper, we tackle the extremely difficult problem of combining distinct models with different initializations, each solving a separate task, into one multi-task model without any additional training. Prior work in model merging permutes one model to the space of the other then averages them together. While this works for models trained on the same task, we find that this fails to account for the differences in models trained on disjoint tasks. Thus, we introduce \"ZipIt!\", a general method"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2305.03053","kind":"arxiv","version":3},"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/2305.03053/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":"2305.03053","created_at":"2026-07-05T07:55:21.305950+00:00"},{"alias_kind":"arxiv_version","alias_value":"2305.03053v3","created_at":"2026-07-05T07:55:21.305950+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2305.03053","created_at":"2026-07-05T07:55:21.305950+00:00"},{"alias_kind":"pith_short_12","alias_value":"FNJTUALT7KW4","created_at":"2026-07-05T07:55:21.305950+00:00"},{"alias_kind":"pith_short_16","alias_value":"FNJTUALT7KW4K7ZC","created_at":"2026-07-05T07:55:21.305950+00:00"},{"alias_kind":"pith_short_8","alias_value":"FNJTUALT","created_at":"2026-07-05T07:55:21.305950+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":6,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.04945","citing_title":"STaR-Quant: State-Time Consistent Post-Training Quantization for Diffusion Large Language Models","ref_index":119,"is_internal_anchor":false},{"citing_arxiv_id":"2606.04754","citing_title":"Beyond Structural Symmetries: Linear Mode Connectivity via Neuron Identifiability","ref_index":2,"is_internal_anchor":false},{"citing_arxiv_id":"2408.01119","citing_title":"Task Prompt Vectors: Effective Initialization through Multi-Task Soft-Prompt Transfer","ref_index":42,"is_internal_anchor":false},{"citing_arxiv_id":"2405.07987","citing_title":"The Platonic Representation Hypothesis","ref_index":98,"is_internal_anchor":false},{"citing_arxiv_id":"2604.02719","citing_title":"MOMO: Mars Orbital Model Foundation Model for Mars Orbital Applications","ref_index":66,"is_internal_anchor":false},{"citing_arxiv_id":"2604.22823","citing_title":"PivotMerge: Bridging Heterogeneous Multimodal Pre-training via Post-Alignment Model Merging","ref_index":6,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/FNJTUALT7KW4K7ZC74ZZMNUCEK","json":"https://pith.science/pith/FNJTUALT7KW4K7ZC74ZZMNUCEK.json","graph_json":"https://pith.science/api/pith-number/FNJTUALT7KW4K7ZC74ZZMNUCEK/graph.json","events_json":"https://pith.science/api/pith-number/FNJTUALT7KW4K7ZC74ZZMNUCEK/events.json","paper":"https://pith.science/paper/FNJTUALT"},"agent_actions":{"view_html":"https://pith.science/pith/FNJTUALT7KW4K7ZC74ZZMNUCEK","download_json":"https://pith.science/pith/FNJTUALT7KW4K7ZC74ZZMNUCEK.json","view_paper":"https://pith.science/paper/FNJTUALT","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2305.03053&json=true","fetch_graph":"https://pith.science/api/pith-number/FNJTUALT7KW4K7ZC74ZZMNUCEK/graph.json","fetch_events":"https://pith.science/api/pith-number/FNJTUALT7KW4K7ZC74ZZMNUCEK/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/FNJTUALT7KW4K7ZC74ZZMNUCEK/action/timestamp_anchor","attest_storage":"https://pith.science/pith/FNJTUALT7KW4K7ZC74ZZMNUCEK/action/storage_attestation","attest_author":"https://pith.science/pith/FNJTUALT7KW4K7ZC74ZZMNUCEK/action/author_attestation","sign_citation":"https://pith.science/pith/FNJTUALT7KW4K7ZC74ZZMNUCEK/action/citation_signature","submit_replication":"https://pith.science/pith/FNJTUALT7KW4K7ZC74ZZMNUCEK/action/replication_record"}},"created_at":"2026-07-05T07:55:21.305950+00:00","updated_at":"2026-07-05T07:55:21.305950+00:00"}