{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2023:ETAADXDD275B57JOGTJ73IZGIN","short_pith_number":"pith:ETAADXDD","schema_version":"1.0","canonical_sha256":"24c001dc63d7fa1efd2e34d3fda326437d7a51cf0a1ee58018d258fe989cc7d1","source":{"kind":"arxiv","id":"2310.05719","version":3},"attestation_state":"computed","paper":{"title":"Transformer Fusion with Optimal Transport","license":"http://creativecommons.org/licenses/by-sa/4.0/","headline":"","cross_cats":["stat.ML"],"primary_cat":"cs.LG","authors_text":"Jacopo Graldi, Marco Giordano, Moritz Imfeld, Sidak Pal Singh (ETH Zurich), Sotiris Anagnostidis, Thomas Hofmann","submitted_at":"2023-10-09T13:40:31Z","abstract_excerpt":"Fusion is a technique for merging multiple independently-trained neural networks in order to combine their capabilities. Past attempts have been restricted to the case of fully-connected, convolutional, and residual networks. This paper presents a systematic approach for fusing two or more transformer-based networks exploiting Optimal Transport to (soft-)align the various architectural components. We flesh out an abstraction for layer alignment, that can generalize to arbitrary architectures - in principle - and we apply this to the key ingredients of Transformers such as multi-head self-atten"},"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":"2310.05719","kind":"arxiv","version":3},"metadata":{"license":"http://creativecommons.org/licenses/by-sa/4.0/","primary_cat":"cs.LG","submitted_at":"2023-10-09T13:40:31Z","cross_cats_sorted":["stat.ML"],"title_canon_sha256":"a330a1805ccd12670ae0bb7df8617b12bc2dd0912a2d54652cd5aac9ff5e8984","abstract_canon_sha256":"caf15206e5aeba6e99d32084f43d98379726d7a16c1535d5c28de6241b525025"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T08:10:18.650105Z","signature_b64":"8NYlNAGw088ZD+KNeIZfTcb2PqUxQQ5ZTxqFcBXTdqxoPe52umljwqq8QrSNRDCT1e3j9pNtdaRTdI3hP1SBCw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"24c001dc63d7fa1efd2e34d3fda326437d7a51cf0a1ee58018d258fe989cc7d1","last_reissued_at":"2026-07-05T08:10:18.649615Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T08:10:18.649615Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Transformer Fusion with Optimal Transport","license":"http://creativecommons.org/licenses/by-sa/4.0/","headline":"","cross_cats":["stat.ML"],"primary_cat":"cs.LG","authors_text":"Jacopo Graldi, Marco Giordano, Moritz Imfeld, Sidak Pal Singh (ETH Zurich), Sotiris Anagnostidis, Thomas Hofmann","submitted_at":"2023-10-09T13:40:31Z","abstract_excerpt":"Fusion is a technique for merging multiple independently-trained neural networks in order to combine their capabilities. Past attempts have been restricted to the case of fully-connected, convolutional, and residual networks. This paper presents a systematic approach for fusing two or more transformer-based networks exploiting Optimal Transport to (soft-)align the various architectural components. We flesh out an abstraction for layer alignment, that can generalize to arbitrary architectures - in principle - and we apply this to the key ingredients of Transformers such as multi-head self-atten"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2310.05719","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/2310.05719/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":"2310.05719","created_at":"2026-07-05T08:10:18.649674+00:00"},{"alias_kind":"arxiv_version","alias_value":"2310.05719v3","created_at":"2026-07-05T08:10:18.649674+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2310.05719","created_at":"2026-07-05T08:10:18.649674+00:00"},{"alias_kind":"pith_short_12","alias_value":"ETAADXDD275B","created_at":"2026-07-05T08:10:18.649674+00:00"},{"alias_kind":"pith_short_16","alias_value":"ETAADXDD275B57JO","created_at":"2026-07-05T08:10:18.649674+00:00"},{"alias_kind":"pith_short_8","alias_value":"ETAADXDD","created_at":"2026-07-05T08:10:18.649674+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":1,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2605.16343","citing_title":"LoopQ: Quantization for Recursive Transformers","ref_index":16,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/ETAADXDD275B57JOGTJ73IZGIN","json":"https://pith.science/pith/ETAADXDD275B57JOGTJ73IZGIN.json","graph_json":"https://pith.science/api/pith-number/ETAADXDD275B57JOGTJ73IZGIN/graph.json","events_json":"https://pith.science/api/pith-number/ETAADXDD275B57JOGTJ73IZGIN/events.json","paper":"https://pith.science/paper/ETAADXDD"},"agent_actions":{"view_html":"https://pith.science/pith/ETAADXDD275B57JOGTJ73IZGIN","download_json":"https://pith.science/pith/ETAADXDD275B57JOGTJ73IZGIN.json","view_paper":"https://pith.science/paper/ETAADXDD","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2310.05719&json=true","fetch_graph":"https://pith.science/api/pith-number/ETAADXDD275B57JOGTJ73IZGIN/graph.json","fetch_events":"https://pith.science/api/pith-number/ETAADXDD275B57JOGTJ73IZGIN/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/ETAADXDD275B57JOGTJ73IZGIN/action/timestamp_anchor","attest_storage":"https://pith.science/pith/ETAADXDD275B57JOGTJ73IZGIN/action/storage_attestation","attest_author":"https://pith.science/pith/ETAADXDD275B57JOGTJ73IZGIN/action/author_attestation","sign_citation":"https://pith.science/pith/ETAADXDD275B57JOGTJ73IZGIN/action/citation_signature","submit_replication":"https://pith.science/pith/ETAADXDD275B57JOGTJ73IZGIN/action/replication_record"}},"created_at":"2026-07-05T08:10:18.649674+00:00","updated_at":"2026-07-05T08:10:18.649674+00:00"}