{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:MSLFK6E5VWBDBPOYGV4T7O5FQC","short_pith_number":"pith:MSLFK6E5","schema_version":"1.0","canonical_sha256":"649655789dad8230bdd835793fbba580b2883421a527c4d81ccb68baf1ed1a34","source":{"kind":"arxiv","id":"2605.16882","version":1},"attestation_state":"computed","paper":{"title":"E-PMQ: Expert-Guided Post-Merge Quantization with Merged-Weight Anchoring","license":"http://creativecommons.org/licenses/by/4.0/","headline":"Expert-guided calibration with source experts and merged-weight anchoring makes post-merge quantization reliable for multi-task models.","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Hongxia Yang, Jianmin Wu, Pengkai Wang, Shuo Cai, Wenjun Wang, Yanggan Gu, Yuanyi Wang","submitted_at":"2026-05-16T08:44:36Z","abstract_excerpt":"Low-resource deployment constraints have made model quantization essential for deploying neural networks while preserving performance. Meanwhile, model merging has become an increasingly practical low-resource strategy for integrating multiple task- or domain-specialized experts into a single model without joint training or multi-model serving. Together, quantization and model merging enable an efficient low-resource deployment pipeline by integrating multiple experts into one low-bit model. We formulate this setting as Post-Merge Quantization (PMQ). We show that directly applying post-trainin"},"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":true,"formal_links_present":true},"canonical_record":{"source":{"id":"2605.16882","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CL","submitted_at":"2026-05-16T08:44:36Z","cross_cats_sorted":[],"title_canon_sha256":"c572008e0482205524d43487271422336396af3c6166490a952b8e2e322b0141","abstract_canon_sha256":"0e2e05f164a53ed7efa077a733e494ebed8df9df3b3b4a8f5aa062834335d868"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-05-20T00:03:28.083539Z","signature_b64":"wjMhXkS1SOu3tT0BDAtqr9bvx0DQgBvlS/G0Yn1AZMMyRj/BS6885XzT/Ln/RcwEOalkRQZuAn0BoZnHBrnRAg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"649655789dad8230bdd835793fbba580b2883421a527c4d81ccb68baf1ed1a34","last_reissued_at":"2026-05-20T00:03:28.082762Z","signature_status":"signed_v1","first_computed_at":"2026-05-20T00:03:28.082762Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"E-PMQ: Expert-Guided Post-Merge Quantization with Merged-Weight Anchoring","license":"http://creativecommons.org/licenses/by/4.0/","headline":"Expert-guided calibration with source experts and merged-weight anchoring makes post-merge quantization reliable for multi-task models.","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Hongxia Yang, Jianmin Wu, Pengkai Wang, Shuo Cai, Wenjun Wang, Yanggan Gu, Yuanyi Wang","submitted_at":"2026-05-16T08:44:36Z","abstract_excerpt":"Low-resource deployment constraints have made model quantization essential for deploying neural networks while preserving performance. Meanwhile, model merging has become an increasingly practical low-resource strategy for integrating multiple task- or domain-specialized experts into a single model without joint training or multi-model serving. Together, quantization and model merging enable an efficient low-resource deployment pipeline by integrating multiple experts into one low-bit model. We formulate this setting as Post-Merge Quantization (PMQ). We show that directly applying post-trainin"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"On CLIP-ViT-B/32 eight-task merging, E-PMQ improves 4-bit GPTQ from 65.0% to 73.6% under Task Arithmetic and from 69.1% to 74.8% under TIES-Merging; on harder 20-task CLIP-ViT-L/14 it raises accuracy from 34.8% to 76.7%.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"That source expert weights remain available after merging and can be used to supply reliable output targets during layer-wise calibration without introducing distribution shift or extra bias relative to the merged model's integrated behavior.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"E-PMQ improves 4-bit quantization accuracy on merged models by 8-42 points across CLIP and GLUE tasks through expert-guided calibration and merged-weight anchoring.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"Expert-guided calibration with source experts and merged-weight anchoring makes post-merge quantization reliable for multi-task models.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"27fc5b05f8ee45dab420f73e55e372ee7ab12ca6015808c0555539d57797afb8"},"source":{"id":"2605.16882","kind":"arxiv","version":1},"verdict":{"id":"7fec81c8-0d87-458e-a0f5-64ae334e5821","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-19T20:56:41.109768Z","strongest_claim":"On CLIP-ViT-B/32 eight-task merging, E-PMQ improves 4-bit GPTQ from 65.0% to 73.6% under Task Arithmetic and from 69.1% to 74.8% under TIES-Merging; on harder 20-task CLIP-ViT-L/14 it raises accuracy from 34.8% to 76.7%.","one_line_summary":"E-PMQ improves 4-bit quantization accuracy on merged models by 8-42 points across CLIP and GLUE tasks through expert-guided calibration and merged-weight anchoring.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"That source expert weights remain available after merging and can be used to supply reliable output targets during layer-wise calibration without introducing distribution shift or extra bias relative to the merged model's integrated behavior.","pith_extraction_headline":"Expert-guided calibration with source experts and merged-weight anchoring makes post-merge quantization reliable for multi-task models."},"integrity":{"clean":true,"summary":{"advisory":0,"critical":0,"by_detector":{},"informational":0},"endpoint":"/pith/2605.16882/integrity.json","findings":[],"available":true,"detectors_run":[{"name":"doi_compliance","ran_at":"2026-05-19T21:01:23.050679Z","status":"completed","version":"1.0.0","findings_count":0},{"name":"doi_title_agreement","ran_at":"2026-05-19T21:01:19.203137Z","status":"completed","version":"1.0.0","findings_count":0},{"name":"claim_evidence","ran_at":"2026-05-19T18:41:56.289892Z","status":"completed","version":"1.0.0","findings_count":0},{"name":"ai_meta_artifact","ran_at":"2026-05-19T18:33:26.367405Z","status":"skipped","version":"1.0.0","findings_count":0}],"snapshot_sha256":"9ab7be792c98814d7fef93630ea4659a9bfdb3a2f9c6964f572bc909efa43d26"},"references":{"count":49,"sample":[{"doi":"","year":null,"title":"and Mozer, Michael C","work_id":"3ea465fc-72a4-40e3-b2bc-a3c88777249a","ref_index":1,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":null,"title":"and Beeman, David , title =","work_id":"af731c2a-8cd8-44b1-aa4f-c32e1b772e71","ref_index":2,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":null,"title":"and Schnell, Eric and Barkai, Edi , title =","work_id":"1659d4a5-8ebb-4c28-9cde-9249be322db9","ref_index":3,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2026,"title":"Journal of Modern Power Systems and Clean Energy , volume=","work_id":"002e0068-21de-4a62-adb5-69fc9c374580","ref_index":4,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":null,"title":"Democratizing AI through model fusion: A comprehensive review and future directions , author=. Nexus , year=","work_id":"3864a498-370d-4f51-b23d-a6309847fa14","ref_index":5,"cited_arxiv_id":"","is_internal_anchor":false}],"resolved_work":49,"snapshot_sha256":"8ebc91cc48f564c6305ead9c25f928d1fdf1dd3a7b2971df6ebb14c900c18ae6","internal_anchors":3},"formal_canon":{"evidence_count":2,"snapshot_sha256":"803c0bfcf563a2325794c6a3a2a94b048b79ebb1751b96a3809b79d489d6c220"},"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":"2605.16882","created_at":"2026-05-20T00:03:28.082883+00:00"},{"alias_kind":"arxiv_version","alias_value":"2605.16882v1","created_at":"2026-05-20T00:03:28.082883+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2605.16882","created_at":"2026-05-20T00:03:28.082883+00:00"},{"alias_kind":"pith_short_12","alias_value":"MSLFK6E5VWBD","created_at":"2026-05-20T00:03:28.082883+00:00"},{"alias_kind":"pith_short_16","alias_value":"MSLFK6E5VWBDBPOY","created_at":"2026-05-20T00:03:28.082883+00:00"},{"alias_kind":"pith_short_8","alias_value":"MSLFK6E5","created_at":"2026-05-20T00:03:28.082883+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":0,"internal_anchor_count":0,"sample":[]},"formal_canon":{"evidence_count":2,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/MSLFK6E5VWBDBPOYGV4T7O5FQC","json":"https://pith.science/pith/MSLFK6E5VWBDBPOYGV4T7O5FQC.json","graph_json":"https://pith.science/api/pith-number/MSLFK6E5VWBDBPOYGV4T7O5FQC/graph.json","events_json":"https://pith.science/api/pith-number/MSLFK6E5VWBDBPOYGV4T7O5FQC/events.json","paper":"https://pith.science/paper/MSLFK6E5"},"agent_actions":{"view_html":"https://pith.science/pith/MSLFK6E5VWBDBPOYGV4T7O5FQC","download_json":"https://pith.science/pith/MSLFK6E5VWBDBPOYGV4T7O5FQC.json","view_paper":"https://pith.science/paper/MSLFK6E5","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2605.16882&json=true","fetch_graph":"https://pith.science/api/pith-number/MSLFK6E5VWBDBPOYGV4T7O5FQC/graph.json","fetch_events":"https://pith.science/api/pith-number/MSLFK6E5VWBDBPOYGV4T7O5FQC/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/MSLFK6E5VWBDBPOYGV4T7O5FQC/action/timestamp_anchor","attest_storage":"https://pith.science/pith/MSLFK6E5VWBDBPOYGV4T7O5FQC/action/storage_attestation","attest_author":"https://pith.science/pith/MSLFK6E5VWBDBPOYGV4T7O5FQC/action/author_attestation","sign_citation":"https://pith.science/pith/MSLFK6E5VWBDBPOYGV4T7O5FQC/action/citation_signature","submit_replication":"https://pith.science/pith/MSLFK6E5VWBDBPOYGV4T7O5FQC/action/replication_record"}},"created_at":"2026-05-20T00:03:28.082883+00:00","updated_at":"2026-05-20T00:03:28.082883+00:00"}