{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2024:PIUWM3KXZNLTR7DXRAQIXFE4UX","short_pith_number":"pith:PIUWM3KX","schema_version":"1.0","canonical_sha256":"7a29666d57cb5738fc7788208b949ca5d93568645d686db40a56dd1ee5fa7d33","source":{"kind":"arxiv","id":"2407.21770","version":3},"attestation_state":"computed","paper":{"title":"MoMa: Efficient Early-Fusion Pre-training with Mixture of Modality-Aware Experts","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.LG"],"primary_cat":"cs.AI","authors_text":"Akshat Shrivastava, Armen Aghajanyan, Gargi Ghosh, Liang Luo, Luke Zettlemoyer, Mike Lewis, Srinivasan Iyer, Xi Victoria Lin","submitted_at":"2024-07-31T17:46:51Z","abstract_excerpt":"We introduce MoMa, a novel modality-aware mixture-of-experts (MoE) architecture designed for pre-training mixed-modal, early-fusion language models. MoMa processes images and text in arbitrary sequences by dividing expert modules into modality-specific groups. These groups exclusively process designated tokens while employing learned routing within each group to maintain semantically informed adaptivity. Our empirical results reveal substantial pre-training efficiency gains through this modality-specific parameter allocation. Under a 1-trillion-token training budget, the MoMa 1.4B model, featu"},"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":"2407.21770","kind":"arxiv","version":3},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.AI","submitted_at":"2024-07-31T17:46:51Z","cross_cats_sorted":["cs.LG"],"title_canon_sha256":"d7adca15f07dfa92b030d089c2da71ffcb3a9feff1a708fd1917fb6329dc009d","abstract_canon_sha256":"c026d7764370c9b6d47a32b6759fd64030945d3f46e0be17032249b380dbb6ab"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T08:54:36.851430Z","signature_b64":"YaiczEKW5I5oEHQ+Lz7uz9TSlC8nB7enQVsQKPw2NKQ+FVB64ZGTKvIsYEcyI5sEto9lgaKqe9yGLgr64NLwDA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"7a29666d57cb5738fc7788208b949ca5d93568645d686db40a56dd1ee5fa7d33","last_reissued_at":"2026-07-05T08:54:36.851005Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T08:54:36.851005Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"MoMa: Efficient Early-Fusion Pre-training with Mixture of Modality-Aware Experts","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.LG"],"primary_cat":"cs.AI","authors_text":"Akshat Shrivastava, Armen Aghajanyan, Gargi Ghosh, Liang Luo, Luke Zettlemoyer, Mike Lewis, Srinivasan Iyer, Xi Victoria Lin","submitted_at":"2024-07-31T17:46:51Z","abstract_excerpt":"We introduce MoMa, a novel modality-aware mixture-of-experts (MoE) architecture designed for pre-training mixed-modal, early-fusion language models. MoMa processes images and text in arbitrary sequences by dividing expert modules into modality-specific groups. These groups exclusively process designated tokens while employing learned routing within each group to maintain semantically informed adaptivity. Our empirical results reveal substantial pre-training efficiency gains through this modality-specific parameter allocation. Under a 1-trillion-token training budget, the MoMa 1.4B model, featu"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2407.21770","kind":"arxiv","version":3},"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/2407.21770/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":"2407.21770","created_at":"2026-07-05T08:54:36.851059+00:00"},{"alias_kind":"arxiv_version","alias_value":"2407.21770v3","created_at":"2026-07-05T08:54:36.851059+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2407.21770","created_at":"2026-07-05T08:54:36.851059+00:00"},{"alias_kind":"pith_short_12","alias_value":"PIUWM3KXZNLT","created_at":"2026-07-05T08:54:36.851059+00:00"},{"alias_kind":"pith_short_16","alias_value":"PIUWM3KXZNLTR7DX","created_at":"2026-07-05T08:54:36.851059+00:00"},{"alias_kind":"pith_short_8","alias_value":"PIUWM3KX","created_at":"2026-07-05T08:54:36.851059+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":9,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.25325","citing_title":"Omni-Perception Policy Optimization for Multimodal Emotion Reasoning","ref_index":66,"is_internal_anchor":false},{"citing_arxiv_id":"2606.07414","citing_title":"Sparsely gated tiny linear experts","ref_index":23,"is_internal_anchor":false},{"citing_arxiv_id":"2606.00275","citing_title":"Hyperbolic and Evidence-Prioritized Experts for Large Vision-Language Models","ref_index":26,"is_internal_anchor":false},{"citing_arxiv_id":"2605.27431","citing_title":"Tackling Multimodal Learning Challenges with Mixture-of-Expert: A Survey","ref_index":9,"is_internal_anchor":false},{"citing_arxiv_id":"2411.04996","citing_title":"Mixture-of-Transformers: A Sparse and Scalable Architecture for Multi-Modal Foundation Models","ref_index":20,"is_internal_anchor":false},{"citing_arxiv_id":"2505.05472","citing_title":"Mogao: An Omni Foundation Model for Interleaved Multi-Modal Generation","ref_index":45,"is_internal_anchor":false},{"citing_arxiv_id":"2411.10442","citing_title":"Enhancing the Reasoning Ability of Multimodal Large Language Models via Mixed Preference Optimization","ref_index":51,"is_internal_anchor":false},{"citing_arxiv_id":"2605.12500","citing_title":"SenseNova-U1: Unifying Multimodal Understanding and Generation with NEO-unify Architecture","ref_index":78,"is_internal_anchor":false},{"citing_arxiv_id":"2604.23996","citing_title":"SMoES: Soft Modality-Guided Expert Specialization in MoE-VLMs","ref_index":33,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/PIUWM3KXZNLTR7DXRAQIXFE4UX","json":"https://pith.science/pith/PIUWM3KXZNLTR7DXRAQIXFE4UX.json","graph_json":"https://pith.science/api/pith-number/PIUWM3KXZNLTR7DXRAQIXFE4UX/graph.json","events_json":"https://pith.science/api/pith-number/PIUWM3KXZNLTR7DXRAQIXFE4UX/events.json","paper":"https://pith.science/paper/PIUWM3KX"},"agent_actions":{"view_html":"https://pith.science/pith/PIUWM3KXZNLTR7DXRAQIXFE4UX","download_json":"https://pith.science/pith/PIUWM3KXZNLTR7DXRAQIXFE4UX.json","view_paper":"https://pith.science/paper/PIUWM3KX","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2407.21770&json=true","fetch_graph":"https://pith.science/api/pith-number/PIUWM3KXZNLTR7DXRAQIXFE4UX/graph.json","fetch_events":"https://pith.science/api/pith-number/PIUWM3KXZNLTR7DXRAQIXFE4UX/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/PIUWM3KXZNLTR7DXRAQIXFE4UX/action/timestamp_anchor","attest_storage":"https://pith.science/pith/PIUWM3KXZNLTR7DXRAQIXFE4UX/action/storage_attestation","attest_author":"https://pith.science/pith/PIUWM3KXZNLTR7DXRAQIXFE4UX/action/author_attestation","sign_citation":"https://pith.science/pith/PIUWM3KXZNLTR7DXRAQIXFE4UX/action/citation_signature","submit_replication":"https://pith.science/pith/PIUWM3KXZNLTR7DXRAQIXFE4UX/action/replication_record"}},"created_at":"2026-07-05T08:54:36.851059+00:00","updated_at":"2026-07-05T08:54:36.851059+00:00"}