{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:3FB4T352V2FB5BQBTTJ5LXB64W","short_pith_number":"pith:3FB4T352","schema_version":"1.0","canonical_sha256":"d943c9efbaae8a1e86019cd3d5dc3ee59b203851e38d2e622a0c0c73e2d3402a","source":{"kind":"arxiv","id":"2608.07890","version":1},"attestation_state":"computed","paper":{"title":"Router Sensitivity Under Lightweight Fine-Tuning Identifies Prunable Experts in Mixture-of-Experts Models","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.LG","authors_text":"Ali Janati, Kaoutar El Maghraoui, Owen Zou, Wenyuan Shen, Xinyi Luo, Yankai Mao","submitted_at":"2026-08-08T03:36:41Z","abstract_excerpt":"Mixture-of-Experts (MoE) models decouple total parameters from per-token compute, but deployment still requires storing every expert. Recent theory shows that pruning experts with the smallest router-norm changes during fine-tuning can preserve accuracy, but assumes full fine-tuning. We test whether lightweight adaptation can recover this signal. We briefly fine-tune with a parameter-efficient adapter, rank experts by the induced $\\ell_2$ router change, and prune the least-changed experts in one shot. On Mixtral-8$\\times$7B-Instruct (44.83% MMLU-Pro), router-only LoRA trains 0.002% of paramete"},"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":"2608.07890","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2026-08-08T03:36:41Z","cross_cats_sorted":["cs.AI"],"title_canon_sha256":"a102e423d195d2e645156b7629fb25ae5c5c595b6faad82608e2663394b44f33","abstract_canon_sha256":"22345b78079534780cecba1860687cfed603434932178c09886ad72fe568c116"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-08-11T01:18:50.974064Z","signature_b64":"Q1xTA3Wvrdq1IccqwFFVvdpqxElXCaoDuMNJ9Bbc4btpudIWN1kgosxQZ61bLTM3/A+oBgI+l287BU6p1M+VBA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"d943c9efbaae8a1e86019cd3d5dc3ee59b203851e38d2e622a0c0c73e2d3402a","last_reissued_at":"2026-08-11T01:18:50.971580Z","signature_status":"signed_v1","first_computed_at":"2026-08-11T01:18:50.971580Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Router Sensitivity Under Lightweight Fine-Tuning Identifies Prunable Experts in Mixture-of-Experts Models","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.LG","authors_text":"Ali Janati, Kaoutar El Maghraoui, Owen Zou, Wenyuan Shen, Xinyi Luo, Yankai Mao","submitted_at":"2026-08-08T03:36:41Z","abstract_excerpt":"Mixture-of-Experts (MoE) models decouple total parameters from per-token compute, but deployment still requires storing every expert. Recent theory shows that pruning experts with the smallest router-norm changes during fine-tuning can preserve accuracy, but assumes full fine-tuning. We test whether lightweight adaptation can recover this signal. We briefly fine-tune with a parameter-efficient adapter, rank experts by the induced $\\ell_2$ router change, and prune the least-changed experts in one shot. On Mixtral-8$\\times$7B-Instruct (44.83% MMLU-Pro), router-only LoRA trains 0.002% of paramete"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2608.07890","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/2608.07890/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":"2608.07890","created_at":"2026-08-11T01:18:50.972285+00:00"},{"alias_kind":"arxiv_version","alias_value":"2608.07890v1","created_at":"2026-08-11T01:18:50.972285+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2608.07890","created_at":"2026-08-11T01:18:50.972285+00:00"},{"alias_kind":"pith_short_12","alias_value":"3FB4T352V2FB","created_at":"2026-08-11T01:18:50.972285+00:00"},{"alias_kind":"pith_short_16","alias_value":"3FB4T352V2FB5BQB","created_at":"2026-08-11T01:18:50.972285+00:00"},{"alias_kind":"pith_short_8","alias_value":"3FB4T352","created_at":"2026-08-11T01:18:50.972285+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/3FB4T352V2FB5BQBTTJ5LXB64W","json":"https://pith.science/pith/3FB4T352V2FB5BQBTTJ5LXB64W.json","graph_json":"https://pith.science/api/pith-number/3FB4T352V2FB5BQBTTJ5LXB64W/graph.json","events_json":"https://pith.science/api/pith-number/3FB4T352V2FB5BQBTTJ5LXB64W/events.json","paper":"https://pith.science/paper/3FB4T352"},"agent_actions":{"view_html":"https://pith.science/pith/3FB4T352V2FB5BQBTTJ5LXB64W","download_json":"https://pith.science/pith/3FB4T352V2FB5BQBTTJ5LXB64W.json","view_paper":"https://pith.science/paper/3FB4T352","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2608.07890&json=true","fetch_graph":"https://pith.science/api/pith-number/3FB4T352V2FB5BQBTTJ5LXB64W/graph.json","fetch_events":"https://pith.science/api/pith-number/3FB4T352V2FB5BQBTTJ5LXB64W/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/3FB4T352V2FB5BQBTTJ5LXB64W/action/timestamp_anchor","attest_storage":"https://pith.science/pith/3FB4T352V2FB5BQBTTJ5LXB64W/action/storage_attestation","attest_author":"https://pith.science/pith/3FB4T352V2FB5BQBTTJ5LXB64W/action/author_attestation","sign_citation":"https://pith.science/pith/3FB4T352V2FB5BQBTTJ5LXB64W/action/citation_signature","submit_replication":"https://pith.science/pith/3FB4T352V2FB5BQBTTJ5LXB64W/action/replication_record"}},"created_at":"2026-08-11T01:18:50.972285+00:00","updated_at":"2026-08-11T01:18:50.972285+00:00"}