{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:SQ2R5XYKIMHBXEHCV3KPFOC2LN","short_pith_number":"pith:SQ2R5XYK","schema_version":"1.0","canonical_sha256":"94351edf0a430e1b90e2aed4f2b85a5b70fc3fea1df157b3cd0bc109e1b3da26","source":{"kind":"arxiv","id":"2605.28306","version":1},"attestation_state":"computed","paper":{"title":"Routing-Aligned Fine-Tuning for Multilingual Downstream Tasks in Mixture-of-Experts Models","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.CL","authors_text":"Guanzhi Deng, Haibo Wang, Kuan Wu, Linqi Song, Shing Yin Wong, Sichun Luo","submitted_at":"2026-05-27T11:01:25Z","abstract_excerpt":"Mixture-of-Experts (MoE) models have emerged as a dominant paradigm for efficient LLM scaling, yet adapting them to non-English downstream tasks remains challenging. Existing fine-tuning approaches treat MoE models as monolithic learners, ignoring the heterogeneous routing structure that develops during pretraining. We validate across multiple MoE models and downstream tasks that middle layers form a language-universal alignment zone where routing divergence strongly predicts per-language task performance gaps. Building on this observation, we propose RA-MoE (Routing-Aligned MoE Fine-Tuning), "},"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":"2605.28306","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CL","submitted_at":"2026-05-27T11:01:25Z","cross_cats_sorted":["cs.AI"],"title_canon_sha256":"2ee32b2b119949cad3f4e49a41b6d820b547b41031f3991db5517f9ac7e266bf","abstract_canon_sha256":"5b8a891c19ce7982c5c0fa44877eaaacde639088b2bdad6eafe1aa63611b53d1"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-05-28T01:05:05.887649Z","signature_b64":"jc58pVwL54eTF8aeGf0GSji2Zcst8QRgA5xsZskRCEV6QLi69S5x7vt8CBmNZFcj4+FOzzhxFoZ6+N8E+3eOBQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"94351edf0a430e1b90e2aed4f2b85a5b70fc3fea1df157b3cd0bc109e1b3da26","last_reissued_at":"2026-05-28T01:05:05.887222Z","signature_status":"signed_v1","first_computed_at":"2026-05-28T01:05:05.887222Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Routing-Aligned Fine-Tuning for Multilingual Downstream Tasks in Mixture-of-Experts Models","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.CL","authors_text":"Guanzhi Deng, Haibo Wang, Kuan Wu, Linqi Song, Shing Yin Wong, Sichun Luo","submitted_at":"2026-05-27T11:01:25Z","abstract_excerpt":"Mixture-of-Experts (MoE) models have emerged as a dominant paradigm for efficient LLM scaling, yet adapting them to non-English downstream tasks remains challenging. Existing fine-tuning approaches treat MoE models as monolithic learners, ignoring the heterogeneous routing structure that develops during pretraining. We validate across multiple MoE models and downstream tasks that middle layers form a language-universal alignment zone where routing divergence strongly predicts per-language task performance gaps. Building on this observation, we propose RA-MoE (Routing-Aligned MoE Fine-Tuning), "},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2605.28306","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/2605.28306/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":"2605.28306","created_at":"2026-05-28T01:05:05.887280+00:00"},{"alias_kind":"arxiv_version","alias_value":"2605.28306v1","created_at":"2026-05-28T01:05:05.887280+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2605.28306","created_at":"2026-05-28T01:05:05.887280+00:00"},{"alias_kind":"pith_short_12","alias_value":"SQ2R5XYKIMHB","created_at":"2026-05-28T01:05:05.887280+00:00"},{"alias_kind":"pith_short_16","alias_value":"SQ2R5XYKIMHBXEHC","created_at":"2026-05-28T01:05:05.887280+00:00"},{"alias_kind":"pith_short_8","alias_value":"SQ2R5XYK","created_at":"2026-05-28T01:05:05.887280+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/SQ2R5XYKIMHBXEHCV3KPFOC2LN","json":"https://pith.science/pith/SQ2R5XYKIMHBXEHCV3KPFOC2LN.json","graph_json":"https://pith.science/api/pith-number/SQ2R5XYKIMHBXEHCV3KPFOC2LN/graph.json","events_json":"https://pith.science/api/pith-number/SQ2R5XYKIMHBXEHCV3KPFOC2LN/events.json","paper":"https://pith.science/paper/SQ2R5XYK"},"agent_actions":{"view_html":"https://pith.science/pith/SQ2R5XYKIMHBXEHCV3KPFOC2LN","download_json":"https://pith.science/pith/SQ2R5XYKIMHBXEHCV3KPFOC2LN.json","view_paper":"https://pith.science/paper/SQ2R5XYK","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2605.28306&json=true","fetch_graph":"https://pith.science/api/pith-number/SQ2R5XYKIMHBXEHCV3KPFOC2LN/graph.json","fetch_events":"https://pith.science/api/pith-number/SQ2R5XYKIMHBXEHCV3KPFOC2LN/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/SQ2R5XYKIMHBXEHCV3KPFOC2LN/action/timestamp_anchor","attest_storage":"https://pith.science/pith/SQ2R5XYKIMHBXEHCV3KPFOC2LN/action/storage_attestation","attest_author":"https://pith.science/pith/SQ2R5XYKIMHBXEHCV3KPFOC2LN/action/author_attestation","sign_citation":"https://pith.science/pith/SQ2R5XYKIMHBXEHCV3KPFOC2LN/action/citation_signature","submit_replication":"https://pith.science/pith/SQ2R5XYKIMHBXEHCV3KPFOC2LN/action/replication_record"}},"created_at":"2026-05-28T01:05:05.887280+00:00","updated_at":"2026-05-28T01:05:05.887280+00:00"}