{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:7IRVDDV5G7HNZLXNGLOR3ZAIVJ","short_pith_number":"pith:7IRVDDV5","schema_version":"1.0","canonical_sha256":"fa23518ebd37cedcaeed32dd1de408aa79a823503164d2d2aaebb77853fbcd89","source":{"kind":"arxiv","id":"2506.06406","version":2},"attestation_state":"computed","paper":{"title":"SMAR: Soft Modality-Aware Routing Strategy for MoE-based Multimodal Large Language Models Preserving Language Capabilities","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.CL","authors_text":"Fangxiang Feng, Fengfa Li, Guoyang Xia, Lei Ren, Wei Chen, Xiaojie Wang, Yifeng Ding","submitted_at":"2025-06-06T12:47:29Z","abstract_excerpt":"Mixture of Experts (MoE) architectures have become a key approach for scaling large language models, with growing interest in extending them to multimodal tasks. Existing methods to build multimodal MoE models either incur high training costs or suffer from degraded language capabilities when adapting pretrained models. To address this, we propose Soft ModalityAware Routing (SMAR), a novel regularization technique that uses Kullback Leibler divergence to control routing probability distributions across modalities, encouraging expert specialization without modifying model architecture or heavil"},"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":"2506.06406","kind":"arxiv","version":2},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CL","submitted_at":"2025-06-06T12:47:29Z","cross_cats_sorted":["cs.AI"],"title_canon_sha256":"0c1e3760c63334fe048ccad7c3ec41c6efadc3ba70de0ac2edc780126815978a","abstract_canon_sha256":"cbb0b387e0773fc5f79555ad5573e2a03ca99ddc43694512f22029ddb3de59b6"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T11:26:52.447083Z","signature_b64":"urf1GuJFFakRV4VuT+ZqZctH0g8SKOgg8R56xT7/s5xClHwwenk8rGfqGZU/L8aM3yD56Z5QZ2z/IIWnASwZDw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"fa23518ebd37cedcaeed32dd1de408aa79a823503164d2d2aaebb77853fbcd89","last_reissued_at":"2026-07-05T11:26:52.446581Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T11:26:52.446581Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"SMAR: Soft Modality-Aware Routing Strategy for MoE-based Multimodal Large Language Models Preserving Language Capabilities","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.CL","authors_text":"Fangxiang Feng, Fengfa Li, Guoyang Xia, Lei Ren, Wei Chen, Xiaojie Wang, Yifeng Ding","submitted_at":"2025-06-06T12:47:29Z","abstract_excerpt":"Mixture of Experts (MoE) architectures have become a key approach for scaling large language models, with growing interest in extending them to multimodal tasks. Existing methods to build multimodal MoE models either incur high training costs or suffer from degraded language capabilities when adapting pretrained models. To address this, we propose Soft ModalityAware Routing (SMAR), a novel regularization technique that uses Kullback Leibler divergence to control routing probability distributions across modalities, encouraging expert specialization without modifying model architecture or heavil"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2506.06406","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/2506.06406/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":"2506.06406","created_at":"2026-07-05T11:26:52.446643+00:00"},{"alias_kind":"arxiv_version","alias_value":"2506.06406v2","created_at":"2026-07-05T11:26:52.446643+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2506.06406","created_at":"2026-07-05T11:26:52.446643+00:00"},{"alias_kind":"pith_short_12","alias_value":"7IRVDDV5G7HN","created_at":"2026-07-05T11:26:52.446643+00:00"},{"alias_kind":"pith_short_16","alias_value":"7IRVDDV5G7HNZLXN","created_at":"2026-07-05T11:26:52.446643+00:00"},{"alias_kind":"pith_short_8","alias_value":"7IRVDDV5","created_at":"2026-07-05T11:26:52.446643+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":2,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.20970","citing_title":"CogniRoute: Learning to Route Social Evidence in Omni-Modal Models","ref_index":201,"is_internal_anchor":false},{"citing_arxiv_id":"2604.23996","citing_title":"SMoES: Soft Modality-Guided Expert Specialization in MoE-VLMs","ref_index":61,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/7IRVDDV5G7HNZLXNGLOR3ZAIVJ","json":"https://pith.science/pith/7IRVDDV5G7HNZLXNGLOR3ZAIVJ.json","graph_json":"https://pith.science/api/pith-number/7IRVDDV5G7HNZLXNGLOR3ZAIVJ/graph.json","events_json":"https://pith.science/api/pith-number/7IRVDDV5G7HNZLXNGLOR3ZAIVJ/events.json","paper":"https://pith.science/paper/7IRVDDV5"},"agent_actions":{"view_html":"https://pith.science/pith/7IRVDDV5G7HNZLXNGLOR3ZAIVJ","download_json":"https://pith.science/pith/7IRVDDV5G7HNZLXNGLOR3ZAIVJ.json","view_paper":"https://pith.science/paper/7IRVDDV5","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2506.06406&json=true","fetch_graph":"https://pith.science/api/pith-number/7IRVDDV5G7HNZLXNGLOR3ZAIVJ/graph.json","fetch_events":"https://pith.science/api/pith-number/7IRVDDV5G7HNZLXNGLOR3ZAIVJ/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/7IRVDDV5G7HNZLXNGLOR3ZAIVJ/action/timestamp_anchor","attest_storage":"https://pith.science/pith/7IRVDDV5G7HNZLXNGLOR3ZAIVJ/action/storage_attestation","attest_author":"https://pith.science/pith/7IRVDDV5G7HNZLXNGLOR3ZAIVJ/action/author_attestation","sign_citation":"https://pith.science/pith/7IRVDDV5G7HNZLXNGLOR3ZAIVJ/action/citation_signature","submit_replication":"https://pith.science/pith/7IRVDDV5G7HNZLXNGLOR3ZAIVJ/action/replication_record"}},"created_at":"2026-07-05T11:26:52.446643+00:00","updated_at":"2026-07-05T11:26:52.446643+00:00"}