{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2024:ITDU443EZTKEEQXYXAB6P6SOSV","short_pith_number":"pith:ITDU443E","schema_version":"1.0","canonical_sha256":"44c74e7364ccd44242f8b803e7fa4e954462343e7f7daaac84c01b3f2bee2db0","source":{"kind":"arxiv","id":"2405.03133","version":2},"attestation_state":"computed","paper":{"title":"Lory: Fully Differentiable Mixture-of-Experts for Autoregressive Language Model Pre-training","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.LG"],"primary_cat":"cs.CL","authors_text":"Danqi Chen, Mengzhou Xia, Mike Lewis, Zexuan Zhong","submitted_at":"2024-05-06T03:06:33Z","abstract_excerpt":"Mixture-of-experts (MoE) models facilitate efficient scaling; however, training the router network introduces the challenge of optimizing a non-differentiable, discrete objective. Recently, a fully-differentiable MoE architecture, SMEAR, was proposed (Muqeeth et al., 2023), which softly merges experts in the parameter space; nevertheless, its effectiveness was only demonstrated in downstream fine-tuning on classification tasks. In this paper, we present Lory, the first approach that scales such architectures to autoregressive language model pre-training. Lory introduces two key techniques: (1)"},"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":"2405.03133","kind":"arxiv","version":2},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CL","submitted_at":"2024-05-06T03:06:33Z","cross_cats_sorted":["cs.LG"],"title_canon_sha256":"429ebda2b916bb8347df3134470a6609d24d2b93d30537ee2bbfa4fb11377471","abstract_canon_sha256":"7249614c3e7efddb6eb32bb9367843141d8816248c7139e9a045fcb7768b9fb5"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T08:56:19.511284Z","signature_b64":"GjMHx6DdZXIU2BeaUy8ExctC3G3dKP39d+6QLDkv3r/3lUkQowP8CbrAv/N7kN7J/vHKDdNcQTr2SedN+8R+Aw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"44c74e7364ccd44242f8b803e7fa4e954462343e7f7daaac84c01b3f2bee2db0","last_reissued_at":"2026-07-05T08:56:19.510836Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T08:56:19.510836Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Lory: Fully Differentiable Mixture-of-Experts for Autoregressive Language Model Pre-training","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.LG"],"primary_cat":"cs.CL","authors_text":"Danqi Chen, Mengzhou Xia, Mike Lewis, Zexuan Zhong","submitted_at":"2024-05-06T03:06:33Z","abstract_excerpt":"Mixture-of-experts (MoE) models facilitate efficient scaling; however, training the router network introduces the challenge of optimizing a non-differentiable, discrete objective. Recently, a fully-differentiable MoE architecture, SMEAR, was proposed (Muqeeth et al., 2023), which softly merges experts in the parameter space; nevertheless, its effectiveness was only demonstrated in downstream fine-tuning on classification tasks. In this paper, we present Lory, the first approach that scales such architectures to autoregressive language model pre-training. Lory introduces two key techniques: (1)"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2405.03133","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/2405.03133/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":"2405.03133","created_at":"2026-07-05T08:56:19.510901+00:00"},{"alias_kind":"arxiv_version","alias_value":"2405.03133v2","created_at":"2026-07-05T08:56:19.510901+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2405.03133","created_at":"2026-07-05T08:56:19.510901+00:00"},{"alias_kind":"pith_short_12","alias_value":"ITDU443EZTKE","created_at":"2026-07-05T08:56:19.510901+00:00"},{"alias_kind":"pith_short_16","alias_value":"ITDU443EZTKEEQXY","created_at":"2026-07-05T08:56:19.510901+00:00"},{"alias_kind":"pith_short_8","alias_value":"ITDU443E","created_at":"2026-07-05T08:56:19.510901+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":3,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2605.10933","citing_title":"DECO: Sparse Mixture-of-Experts with Dense-Comparable Performance on End-Side Devices","ref_index":155,"is_internal_anchor":false},{"citing_arxiv_id":"2605.10933","citing_title":"DECO: Sparse Mixture-of-Experts with Dense-Comparable Performance on End-Side Devices","ref_index":155,"is_internal_anchor":false},{"citing_arxiv_id":"2605.10933","citing_title":"DECO: Sparse Mixture-of-Experts with Dense-Comparable Performance on End-Side Devices","ref_index":155,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/ITDU443EZTKEEQXYXAB6P6SOSV","json":"https://pith.science/pith/ITDU443EZTKEEQXYXAB6P6SOSV.json","graph_json":"https://pith.science/api/pith-number/ITDU443EZTKEEQXYXAB6P6SOSV/graph.json","events_json":"https://pith.science/api/pith-number/ITDU443EZTKEEQXYXAB6P6SOSV/events.json","paper":"https://pith.science/paper/ITDU443E"},"agent_actions":{"view_html":"https://pith.science/pith/ITDU443EZTKEEQXYXAB6P6SOSV","download_json":"https://pith.science/pith/ITDU443EZTKEEQXYXAB6P6SOSV.json","view_paper":"https://pith.science/paper/ITDU443E","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2405.03133&json=true","fetch_graph":"https://pith.science/api/pith-number/ITDU443EZTKEEQXYXAB6P6SOSV/graph.json","fetch_events":"https://pith.science/api/pith-number/ITDU443EZTKEEQXYXAB6P6SOSV/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/ITDU443EZTKEEQXYXAB6P6SOSV/action/timestamp_anchor","attest_storage":"https://pith.science/pith/ITDU443EZTKEEQXYXAB6P6SOSV/action/storage_attestation","attest_author":"https://pith.science/pith/ITDU443EZTKEEQXYXAB6P6SOSV/action/author_attestation","sign_citation":"https://pith.science/pith/ITDU443EZTKEEQXYXAB6P6SOSV/action/citation_signature","submit_replication":"https://pith.science/pith/ITDU443EZTKEEQXYXAB6P6SOSV/action/replication_record"}},"created_at":"2026-07-05T08:56:19.510901+00:00","updated_at":"2026-07-05T08:56:19.510901+00:00"}