{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:SNNP4MCQDFTVYW5SR6RCHUQFVY","short_pith_number":"pith:SNNP4MCQ","schema_version":"1.0","canonical_sha256":"935afe305019675c5bb28fa223d205ae2bf8526c02bae48f50f72ad0422c5a03","source":{"kind":"arxiv","id":"2503.06921","version":2},"attestation_state":"computed","paper":{"title":"Task Vector Quantization for Memory-Efficient Model Merging","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Aecheon Jung, Bogon Ryu, Seunghwan Lee, Sungeun Hong, Youngeun Kim","submitted_at":"2025-03-10T05:00:24Z","abstract_excerpt":"Model merging enables efficient multi-task models by combining task-specific fine-tuned checkpoints. However, storing multiple task-specific checkpoints requires significant memory, limiting scalability and restricting model merging to larger models and diverse tasks. In this paper, we propose quantizing task vectors (i.e., the difference between pre-trained and fine-tuned checkpoints) instead of quantizing fine-tuned checkpoints. We observe that task vectors exhibit a narrow weight range, enabling low precision quantization (e.g., 4 bit) within existing task vector merging frameworks. To furt"},"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":"2503.06921","kind":"arxiv","version":2},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.LG","submitted_at":"2025-03-10T05:00:24Z","cross_cats_sorted":[],"title_canon_sha256":"f43244d28645362e520962b84082d6f1cd426e5cd7deb766d2fa5b71adfc16b0","abstract_canon_sha256":"a999cdf48fb29249461ab8f8ba35c878ac74d8922455168055b082c14d972a99"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T11:49:51.755878Z","signature_b64":"xO81vhXh6+LbTvsv7JabeuoKelzYx2PWiuGIXzWSJORkEeY1CS5pPMxSl9313sRqFvwCISEeNwo7xK9XMlqoAw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"935afe305019675c5bb28fa223d205ae2bf8526c02bae48f50f72ad0422c5a03","last_reissued_at":"2026-07-05T11:49:51.755235Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T11:49:51.755235Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Task Vector Quantization for Memory-Efficient Model Merging","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Aecheon Jung, Bogon Ryu, Seunghwan Lee, Sungeun Hong, Youngeun Kim","submitted_at":"2025-03-10T05:00:24Z","abstract_excerpt":"Model merging enables efficient multi-task models by combining task-specific fine-tuned checkpoints. However, storing multiple task-specific checkpoints requires significant memory, limiting scalability and restricting model merging to larger models and diverse tasks. In this paper, we propose quantizing task vectors (i.e., the difference between pre-trained and fine-tuned checkpoints) instead of quantizing fine-tuned checkpoints. We observe that task vectors exhibit a narrow weight range, enabling low precision quantization (e.g., 4 bit) within existing task vector merging frameworks. To furt"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2503.06921","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/2503.06921/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":"2503.06921","created_at":"2026-07-05T11:49:51.755320+00:00"},{"alias_kind":"arxiv_version","alias_value":"2503.06921v2","created_at":"2026-07-05T11:49:51.755320+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2503.06921","created_at":"2026-07-05T11:49:51.755320+00:00"},{"alias_kind":"pith_short_12","alias_value":"SNNP4MCQDFTV","created_at":"2026-07-05T11:49:51.755320+00:00"},{"alias_kind":"pith_short_16","alias_value":"SNNP4MCQDFTVYW5S","created_at":"2026-07-05T11:49:51.755320+00:00"},{"alias_kind":"pith_short_8","alias_value":"SNNP4MCQ","created_at":"2026-07-05T11:49:51.755320+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":1,"internal_anchor_count":1,"sample":[{"citing_arxiv_id":"2502.01015","citing_title":"Task Vector Bases: A Unified and Scalable Framework for Compressed Task Arithmetic","ref_index":24,"is_internal_anchor":true}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/SNNP4MCQDFTVYW5SR6RCHUQFVY","json":"https://pith.science/pith/SNNP4MCQDFTVYW5SR6RCHUQFVY.json","graph_json":"https://pith.science/api/pith-number/SNNP4MCQDFTVYW5SR6RCHUQFVY/graph.json","events_json":"https://pith.science/api/pith-number/SNNP4MCQDFTVYW5SR6RCHUQFVY/events.json","paper":"https://pith.science/paper/SNNP4MCQ"},"agent_actions":{"view_html":"https://pith.science/pith/SNNP4MCQDFTVYW5SR6RCHUQFVY","download_json":"https://pith.science/pith/SNNP4MCQDFTVYW5SR6RCHUQFVY.json","view_paper":"https://pith.science/paper/SNNP4MCQ","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2503.06921&json=true","fetch_graph":"https://pith.science/api/pith-number/SNNP4MCQDFTVYW5SR6RCHUQFVY/graph.json","fetch_events":"https://pith.science/api/pith-number/SNNP4MCQDFTVYW5SR6RCHUQFVY/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/SNNP4MCQDFTVYW5SR6RCHUQFVY/action/timestamp_anchor","attest_storage":"https://pith.science/pith/SNNP4MCQDFTVYW5SR6RCHUQFVY/action/storage_attestation","attest_author":"https://pith.science/pith/SNNP4MCQDFTVYW5SR6RCHUQFVY/action/author_attestation","sign_citation":"https://pith.science/pith/SNNP4MCQDFTVYW5SR6RCHUQFVY/action/citation_signature","submit_replication":"https://pith.science/pith/SNNP4MCQDFTVYW5SR6RCHUQFVY/action/replication_record"}},"created_at":"2026-07-05T11:49:51.755320+00:00","updated_at":"2026-07-05T11:49:51.755320+00:00"}