{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:SINZQWDONTRBGB6PUSPP55CPXY","short_pith_number":"pith:SINZQWDO","schema_version":"1.0","canonical_sha256":"921b98586e6ce21307cfa49efef44fbe09308583830c1b6080b9cf082e16ab2b","source":{"kind":"arxiv","id":"2606.17952","version":1},"attestation_state":"computed","paper":{"title":"SoftMoE: Soft Differentiable Routing for Mixture-of-Experts in LLMs","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.LG","authors_text":"Jacek Tabor, {\\L}ukasz Struski, Marcin Kurdziel, Miko{\\l}aj Zasada","submitted_at":"2026-06-16T14:05:41Z","abstract_excerpt":"Sparse Mixture-of-Experts (MoE) architectures enable scaling LLM parameters under a fixed inference budget by activating only a small subset of experts via top-$k$ routing. While this preserves causality and suits autoregressive language models, the discrete top-$k$ operator is not differentiable, forcing a fixed number of active experts per input and resulting in inefficient use of computation. We propose SoftMoE, which replaces discrete routing with a truncated soft top-$k$ LapSum relaxation, allowing gradient-based optimization of expert routing. We further parameterize the mean number of a"},"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":"2606.17952","kind":"arxiv","version":1},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.LG","submitted_at":"2026-06-16T14:05:41Z","cross_cats_sorted":["cs.AI"],"title_canon_sha256":"9346493178ef4a6694ac38dd8a488ce1a4d692db608dc481438c7912680fb8c5","abstract_canon_sha256":"7dcee475695c6b0da6ae8f23ae911be360b6d60395058739422c7981d4891bac"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-06-19T16:10:44.378012Z","signature_b64":"ysrVTqDhBzzzio+Tfq9WZCJWPl72Oug7pUZGXXj0AuAgeF7Gtw5yd/ZcIGDvtoIe0xfS76S4BnEeqAiPFINWCg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"921b98586e6ce21307cfa49efef44fbe09308583830c1b6080b9cf082e16ab2b","last_reissued_at":"2026-06-19T16:10:44.377666Z","signature_status":"signed_v1","first_computed_at":"2026-06-19T16:10:44.377666Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"SoftMoE: Soft Differentiable Routing for Mixture-of-Experts in LLMs","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.LG","authors_text":"Jacek Tabor, {\\L}ukasz Struski, Marcin Kurdziel, Miko{\\l}aj Zasada","submitted_at":"2026-06-16T14:05:41Z","abstract_excerpt":"Sparse Mixture-of-Experts (MoE) architectures enable scaling LLM parameters under a fixed inference budget by activating only a small subset of experts via top-$k$ routing. While this preserves causality and suits autoregressive language models, the discrete top-$k$ operator is not differentiable, forcing a fixed number of active experts per input and resulting in inefficient use of computation. We propose SoftMoE, which replaces discrete routing with a truncated soft top-$k$ LapSum relaxation, allowing gradient-based optimization of expert routing. We further parameterize the mean number of a"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2606.17952","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/2606.17952/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":"2606.17952","created_at":"2026-06-19T16:10:44.377728+00:00"},{"alias_kind":"arxiv_version","alias_value":"2606.17952v1","created_at":"2026-06-19T16:10:44.377728+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2606.17952","created_at":"2026-06-19T16:10:44.377728+00:00"},{"alias_kind":"pith_short_12","alias_value":"SINZQWDONTRB","created_at":"2026-06-19T16:10:44.377728+00:00"},{"alias_kind":"pith_short_16","alias_value":"SINZQWDONTRBGB6P","created_at":"2026-06-19T16:10:44.377728+00:00"},{"alias_kind":"pith_short_8","alias_value":"SINZQWDO","created_at":"2026-06-19T16:10:44.377728+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/SINZQWDONTRBGB6PUSPP55CPXY","json":"https://pith.science/pith/SINZQWDONTRBGB6PUSPP55CPXY.json","graph_json":"https://pith.science/api/pith-number/SINZQWDONTRBGB6PUSPP55CPXY/graph.json","events_json":"https://pith.science/api/pith-number/SINZQWDONTRBGB6PUSPP55CPXY/events.json","paper":"https://pith.science/paper/SINZQWDO"},"agent_actions":{"view_html":"https://pith.science/pith/SINZQWDONTRBGB6PUSPP55CPXY","download_json":"https://pith.science/pith/SINZQWDONTRBGB6PUSPP55CPXY.json","view_paper":"https://pith.science/paper/SINZQWDO","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2606.17952&json=true","fetch_graph":"https://pith.science/api/pith-number/SINZQWDONTRBGB6PUSPP55CPXY/graph.json","fetch_events":"https://pith.science/api/pith-number/SINZQWDONTRBGB6PUSPP55CPXY/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/SINZQWDONTRBGB6PUSPP55CPXY/action/timestamp_anchor","attest_storage":"https://pith.science/pith/SINZQWDONTRBGB6PUSPP55CPXY/action/storage_attestation","attest_author":"https://pith.science/pith/SINZQWDONTRBGB6PUSPP55CPXY/action/author_attestation","sign_citation":"https://pith.science/pith/SINZQWDONTRBGB6PUSPP55CPXY/action/citation_signature","submit_replication":"https://pith.science/pith/SINZQWDONTRBGB6PUSPP55CPXY/action/replication_record"}},"created_at":"2026-06-19T16:10:44.377728+00:00","updated_at":"2026-06-19T16:10:44.377728+00:00"}