{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:6DDGKMYOKQ3XVQKOTDM7DY2Y3B","short_pith_number":"pith:6DDGKMYO","schema_version":"1.0","canonical_sha256":"f0c665330e54377ac14e98d9f1e358d8582c7f8228b861c37ef112309b49b670","source":{"kind":"arxiv","id":"2505.11204","version":1},"attestation_state":"computed","paper":{"title":"RanDeS: Randomized Delta Superposition for Multi-Model Compression","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.LG","authors_text":"Aaron Gokaslan, Bharath Hariharan, Hangyu Zhou, Volodymyr Kuleshov","submitted_at":"2025-05-16T13:02:12Z","abstract_excerpt":"From a multi-model compression perspective, model merging enables memory-efficient serving of multiple models fine-tuned from the same base, but suffers from degraded performance due to interference among their task-specific parameter adjustments (i.e., deltas). In this paper, we reformulate model merging as a compress-and-retrieve scheme, revealing that the task interference arises from the summation of irrelevant deltas during model retrieval. To address this issue, we use random orthogonal transformations to decorrelate these vectors into self-cancellation. We show that this approach drasti"},"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":"2505.11204","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2025-05-16T13:02:12Z","cross_cats_sorted":["cs.AI"],"title_canon_sha256":"75e46392d2071cd11427bbf9a1a17c610fc32208cf631c3bf9977db62cceecb0","abstract_canon_sha256":"0bac8b8fcdd7e482d336529e86a749b4ace04431ef9319de8dbf7aa21bbd453f"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T11:04:12.711174Z","signature_b64":"guaLC63a+AX7gXzF6++G/1OCU3FEo0HnJbBW3AjNu40Dq+08HEAEuDJnDVhNqWGpeHIdVqnFdaFs1WRXqUa8Bw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"f0c665330e54377ac14e98d9f1e358d8582c7f8228b861c37ef112309b49b670","last_reissued_at":"2026-07-05T11:04:12.710727Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T11:04:12.710727Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"RanDeS: Randomized Delta Superposition for Multi-Model Compression","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.LG","authors_text":"Aaron Gokaslan, Bharath Hariharan, Hangyu Zhou, Volodymyr Kuleshov","submitted_at":"2025-05-16T13:02:12Z","abstract_excerpt":"From a multi-model compression perspective, model merging enables memory-efficient serving of multiple models fine-tuned from the same base, but suffers from degraded performance due to interference among their task-specific parameter adjustments (i.e., deltas). In this paper, we reformulate model merging as a compress-and-retrieve scheme, revealing that the task interference arises from the summation of irrelevant deltas during model retrieval. To address this issue, we use random orthogonal transformations to decorrelate these vectors into self-cancellation. We show that this approach drasti"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2505.11204","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/2505.11204/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":"2505.11204","created_at":"2026-07-05T11:04:12.710787+00:00"},{"alias_kind":"arxiv_version","alias_value":"2505.11204v1","created_at":"2026-07-05T11:04:12.710787+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2505.11204","created_at":"2026-07-05T11:04:12.710787+00:00"},{"alias_kind":"pith_short_12","alias_value":"6DDGKMYOKQ3X","created_at":"2026-07-05T11:04:12.710787+00:00"},{"alias_kind":"pith_short_16","alias_value":"6DDGKMYOKQ3XVQKO","created_at":"2026-07-05T11:04:12.710787+00:00"},{"alias_kind":"pith_short_8","alias_value":"6DDGKMYO","created_at":"2026-07-05T11:04:12.710787+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":0,"internal_anchor_count":0,"sample":[]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/6DDGKMYOKQ3XVQKOTDM7DY2Y3B","json":"https://pith.science/pith/6DDGKMYOKQ3XVQKOTDM7DY2Y3B.json","graph_json":"https://pith.science/api/pith-number/6DDGKMYOKQ3XVQKOTDM7DY2Y3B/graph.json","events_json":"https://pith.science/api/pith-number/6DDGKMYOKQ3XVQKOTDM7DY2Y3B/events.json","paper":"https://pith.science/paper/6DDGKMYO"},"agent_actions":{"view_html":"https://pith.science/pith/6DDGKMYOKQ3XVQKOTDM7DY2Y3B","download_json":"https://pith.science/pith/6DDGKMYOKQ3XVQKOTDM7DY2Y3B.json","view_paper":"https://pith.science/paper/6DDGKMYO","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2505.11204&json=true","fetch_graph":"https://pith.science/api/pith-number/6DDGKMYOKQ3XVQKOTDM7DY2Y3B/graph.json","fetch_events":"https://pith.science/api/pith-number/6DDGKMYOKQ3XVQKOTDM7DY2Y3B/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/6DDGKMYOKQ3XVQKOTDM7DY2Y3B/action/timestamp_anchor","attest_storage":"https://pith.science/pith/6DDGKMYOKQ3XVQKOTDM7DY2Y3B/action/storage_attestation","attest_author":"https://pith.science/pith/6DDGKMYOKQ3XVQKOTDM7DY2Y3B/action/author_attestation","sign_citation":"https://pith.science/pith/6DDGKMYOKQ3XVQKOTDM7DY2Y3B/action/citation_signature","submit_replication":"https://pith.science/pith/6DDGKMYOKQ3XVQKOTDM7DY2Y3B/action/replication_record"}},"created_at":"2026-07-05T11:04:12.710787+00:00","updated_at":"2026-07-05T11:04:12.710787+00:00"}