{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2024:Y6GGFJG6WQLMWCPFVLSVUYYZSU","short_pith_number":"pith:Y6GGFJG6","schema_version":"1.0","canonical_sha256":"c78c62a4deb416cb09e5aae55a63199508f30df67e280f22c918787efb17fd3f","source":{"kind":"arxiv","id":"2404.05567","version":1},"attestation_state":"computed","paper":{"title":"Dense Training, Sparse Inference: Rethinking Training of Mixture-of-Experts Language Models","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.AI","cs.CL"],"primary_cat":"cs.LG","authors_text":"Aude Oliva, Bowen Pan, Colin Raffel, Gaoyuan Zhang, Haokun Liu, Mayank Mishra, Rameswar Panda, Yikang Shen","submitted_at":"2024-04-08T14:39:49Z","abstract_excerpt":"Mixture-of-Experts (MoE) language models can reduce computational costs by 2-4$\\times$ compared to dense models without sacrificing performance, making them more efficient in computation-bounded scenarios. However, MoE models generally require 2-4$\\times$ times more parameters to achieve comparable performance to a dense model, which incurs larger GPU memory requirements and makes MoE models less efficient in I/O-bounded scenarios like autoregressive generation. In this work, we propose a hybrid dense training and sparse inference framework for MoE models (DS-MoE) which achieves strong computa"},"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":"2404.05567","kind":"arxiv","version":1},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.LG","submitted_at":"2024-04-08T14:39:49Z","cross_cats_sorted":["cs.AI","cs.CL"],"title_canon_sha256":"7415a4d09c17beb6dfd5530307e3505136d079a9348236482a5f6be2a19c723f","abstract_canon_sha256":"b954f5c2666c720fd22c083bb6e8ebb37dae2cc06c0011a0460f12b3995c36de"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T08:05:42.630535Z","signature_b64":"uH7WRaXvK9JiHc2IAIC/xc3Axn8XVBUDKhg+jCzgU1BY4z76ut4Xxl2jQCQpNuSA85f18oJCu6zWBNJmb1xBBg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"c78c62a4deb416cb09e5aae55a63199508f30df67e280f22c918787efb17fd3f","last_reissued_at":"2026-07-05T08:05:42.630036Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T08:05:42.630036Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Dense Training, Sparse Inference: Rethinking Training of Mixture-of-Experts Language Models","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.AI","cs.CL"],"primary_cat":"cs.LG","authors_text":"Aude Oliva, Bowen Pan, Colin Raffel, Gaoyuan Zhang, Haokun Liu, Mayank Mishra, Rameswar Panda, Yikang Shen","submitted_at":"2024-04-08T14:39:49Z","abstract_excerpt":"Mixture-of-Experts (MoE) language models can reduce computational costs by 2-4$\\times$ compared to dense models without sacrificing performance, making them more efficient in computation-bounded scenarios. However, MoE models generally require 2-4$\\times$ times more parameters to achieve comparable performance to a dense model, which incurs larger GPU memory requirements and makes MoE models less efficient in I/O-bounded scenarios like autoregressive generation. In this work, we propose a hybrid dense training and sparse inference framework for MoE models (DS-MoE) which achieves strong computa"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2404.05567","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/2404.05567/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":"2404.05567","created_at":"2026-07-05T08:05:42.630095+00:00"},{"alias_kind":"arxiv_version","alias_value":"2404.05567v1","created_at":"2026-07-05T08:05:42.630095+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2404.05567","created_at":"2026-07-05T08:05:42.630095+00:00"},{"alias_kind":"pith_short_12","alias_value":"Y6GGFJG6WQLM","created_at":"2026-07-05T08:05:42.630095+00:00"},{"alias_kind":"pith_short_16","alias_value":"Y6GGFJG6WQLMWCPF","created_at":"2026-07-05T08:05:42.630095+00:00"},{"alias_kind":"pith_short_8","alias_value":"Y6GGFJG6","created_at":"2026-07-05T08:05:42.630095+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":4,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.17952","citing_title":"SoftMoE: Soft Differentiable Routing for Mixture-of-Experts in LLMs","ref_index":24,"is_internal_anchor":false},{"citing_arxiv_id":"2605.12476","citing_title":"Routers Learn the Geometry of Their Experts: Geometric Coupling in Sparse Mixture-of-Experts","ref_index":26,"is_internal_anchor":false},{"citing_arxiv_id":"2605.04058","citing_title":"MP-ISMoE: Mixed-Precision Interactive Side Mixture-of-Experts for Efficient Transfer Learning","ref_index":152,"is_internal_anchor":false},{"citing_arxiv_id":"2604.06542","citing_title":"Does a Global Perspective Help Prune Sparse MoEs Elegantly?","ref_index":18,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/Y6GGFJG6WQLMWCPFVLSVUYYZSU","json":"https://pith.science/pith/Y6GGFJG6WQLMWCPFVLSVUYYZSU.json","graph_json":"https://pith.science/api/pith-number/Y6GGFJG6WQLMWCPFVLSVUYYZSU/graph.json","events_json":"https://pith.science/api/pith-number/Y6GGFJG6WQLMWCPFVLSVUYYZSU/events.json","paper":"https://pith.science/paper/Y6GGFJG6"},"agent_actions":{"view_html":"https://pith.science/pith/Y6GGFJG6WQLMWCPFVLSVUYYZSU","download_json":"https://pith.science/pith/Y6GGFJG6WQLMWCPFVLSVUYYZSU.json","view_paper":"https://pith.science/paper/Y6GGFJG6","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2404.05567&json=true","fetch_graph":"https://pith.science/api/pith-number/Y6GGFJG6WQLMWCPFVLSVUYYZSU/graph.json","fetch_events":"https://pith.science/api/pith-number/Y6GGFJG6WQLMWCPFVLSVUYYZSU/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/Y6GGFJG6WQLMWCPFVLSVUYYZSU/action/timestamp_anchor","attest_storage":"https://pith.science/pith/Y6GGFJG6WQLMWCPFVLSVUYYZSU/action/storage_attestation","attest_author":"https://pith.science/pith/Y6GGFJG6WQLMWCPFVLSVUYYZSU/action/author_attestation","sign_citation":"https://pith.science/pith/Y6GGFJG6WQLMWCPFVLSVUYYZSU/action/citation_signature","submit_replication":"https://pith.science/pith/Y6GGFJG6WQLMWCPFVLSVUYYZSU/action/replication_record"}},"created_at":"2026-07-05T08:05:42.630095+00:00","updated_at":"2026-07-05T08:05:42.630095+00:00"}