{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:UP2Q35FYSH66OXRQNEV7UBNBZQ","short_pith_number":"pith:UP2Q35FY","schema_version":"1.0","canonical_sha256":"a3f50df4b891fde75e30692bfa05a1cc2752493e0b21ecba26cbe35cc2e95432","source":{"kind":"arxiv","id":"2504.02658","version":2},"attestation_state":"computed","paper":{"title":"MiLo: Efficient Quantized MoE Inference with Mixture of Low-Rank Compensators","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Beichen Huang, Minjia Zhang, Yueming Yuan, Zelei Shao","submitted_at":"2025-04-03T14:54:17Z","abstract_excerpt":"A critical approach for efficiently deploying Mixture-of-Experts (MoE) models with massive parameters is quantization. However, state-of-the-art MoE models suffer from non-negligible accuracy loss with extreme quantization, such as under 4 bits. To address this, we introduce MiLo, a novel method that augments highly quantized MoEs with a mixture of low-rank compensators. These compensators consume only a small amount of additional memory but significantly recover accuracy loss from extreme quantization. MiLo also identifies that MoEmodels exhibit distinctive characteristics across weights due "},"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":"2504.02658","kind":"arxiv","version":2},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2025-04-03T14:54:17Z","cross_cats_sorted":[],"title_canon_sha256":"c4f21175ce56298ae19dab89d1640efbc87fb3589535e63418dac87fd2e48957","abstract_canon_sha256":"050cfc53c8ec766b512a4e49f11e96485f0ef794b3aab2802caec554e36bc1ee"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T10:45:20.868087Z","signature_b64":"h5MH2czjxG6ExPHcfiJI4EioLtNDbWryUCo6tbUHQtFoESoJP6PtVGZUSCi9w6OahTAFFIGADnvsdZPCUCb+DA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"a3f50df4b891fde75e30692bfa05a1cc2752493e0b21ecba26cbe35cc2e95432","last_reissued_at":"2026-07-05T10:45:20.867584Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T10:45:20.867584Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"MiLo: Efficient Quantized MoE Inference with Mixture of Low-Rank Compensators","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Beichen Huang, Minjia Zhang, Yueming Yuan, Zelei Shao","submitted_at":"2025-04-03T14:54:17Z","abstract_excerpt":"A critical approach for efficiently deploying Mixture-of-Experts (MoE) models with massive parameters is quantization. However, state-of-the-art MoE models suffer from non-negligible accuracy loss with extreme quantization, such as under 4 bits. To address this, we introduce MiLo, a novel method that augments highly quantized MoEs with a mixture of low-rank compensators. These compensators consume only a small amount of additional memory but significantly recover accuracy loss from extreme quantization. MiLo also identifies that MoEmodels exhibit distinctive characteristics across weights due "},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2504.02658","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/2504.02658/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":"2504.02658","created_at":"2026-07-05T10:45:20.867651+00:00"},{"alias_kind":"arxiv_version","alias_value":"2504.02658v2","created_at":"2026-07-05T10:45:20.867651+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2504.02658","created_at":"2026-07-05T10:45:20.867651+00:00"},{"alias_kind":"pith_short_12","alias_value":"UP2Q35FYSH66","created_at":"2026-07-05T10:45:20.867651+00:00"},{"alias_kind":"pith_short_16","alias_value":"UP2Q35FYSH66OXRQ","created_at":"2026-07-05T10:45:20.867651+00:00"},{"alias_kind":"pith_short_8","alias_value":"UP2Q35FY","created_at":"2026-07-05T10:45:20.867651+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/UP2Q35FYSH66OXRQNEV7UBNBZQ","json":"https://pith.science/pith/UP2Q35FYSH66OXRQNEV7UBNBZQ.json","graph_json":"https://pith.science/api/pith-number/UP2Q35FYSH66OXRQNEV7UBNBZQ/graph.json","events_json":"https://pith.science/api/pith-number/UP2Q35FYSH66OXRQNEV7UBNBZQ/events.json","paper":"https://pith.science/paper/UP2Q35FY"},"agent_actions":{"view_html":"https://pith.science/pith/UP2Q35FYSH66OXRQNEV7UBNBZQ","download_json":"https://pith.science/pith/UP2Q35FYSH66OXRQNEV7UBNBZQ.json","view_paper":"https://pith.science/paper/UP2Q35FY","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2504.02658&json=true","fetch_graph":"https://pith.science/api/pith-number/UP2Q35FYSH66OXRQNEV7UBNBZQ/graph.json","fetch_events":"https://pith.science/api/pith-number/UP2Q35FYSH66OXRQNEV7UBNBZQ/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/UP2Q35FYSH66OXRQNEV7UBNBZQ/action/timestamp_anchor","attest_storage":"https://pith.science/pith/UP2Q35FYSH66OXRQNEV7UBNBZQ/action/storage_attestation","attest_author":"https://pith.science/pith/UP2Q35FYSH66OXRQNEV7UBNBZQ/action/author_attestation","sign_citation":"https://pith.science/pith/UP2Q35FYSH66OXRQNEV7UBNBZQ/action/citation_signature","submit_replication":"https://pith.science/pith/UP2Q35FYSH66OXRQNEV7UBNBZQ/action/replication_record"}},"created_at":"2026-07-05T10:45:20.867651+00:00","updated_at":"2026-07-05T10:45:20.867651+00:00"}