{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2024:QIUKJONVMVADN2KHIZSM7IZN37","short_pith_number":"pith:QIUKJONV","schema_version":"1.0","canonical_sha256":"8228a4b9b5654036e9474664cfa32ddff8f9fbf75d8e34a9ed3d945a9f9bd6c1","source":{"kind":"arxiv","id":"2407.01906","version":2},"attestation_state":"computed","paper":{"title":"Let the Expert Stick to His Last: Expert-Specialized Fine-Tuning for Sparse Architectural Large Language Models","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.AI","cs.LG"],"primary_cat":"cs.CL","authors_text":"Damai Dai, Deli Chen, Runxin Xu, Y. Wu, Zhuoshu Li, Zihan Wang","submitted_at":"2024-07-02T03:11:13Z","abstract_excerpt":"Parameter-efficient fine-tuning (PEFT) is crucial for customizing Large Language Models (LLMs) with constrained resources. Although there have been various PEFT methods for dense-architecture LLMs, PEFT for sparse-architecture LLMs is still underexplored. In this work, we study the PEFT method for LLMs with the Mixture-of-Experts (MoE) architecture and the contents of this work are mainly threefold: (1) We investigate the dispersion degree of the activated experts in customized tasks, and found that the routing distribution for a specific task tends to be highly concentrated, while the distrib"},"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":"2407.01906","kind":"arxiv","version":2},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CL","submitted_at":"2024-07-02T03:11:13Z","cross_cats_sorted":["cs.AI","cs.LG"],"title_canon_sha256":"9d76bcbdcab93c997210d503deddfb50851cf7e04c2037adc4fac04a9d449a4d","abstract_canon_sha256":"7198bbdb565dc0571907301bf92095c840020c72de49be74e81fa25338600dc5"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T08:40:19.610466Z","signature_b64":"rIw3Cx3PlNlyak0clzZSlkgw7LvsSnZyoj1RfgBKJb2xv9t89RSiign3I0I9OT9unmtqHIAXZi0X/xJHAhDWCg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"8228a4b9b5654036e9474664cfa32ddff8f9fbf75d8e34a9ed3d945a9f9bd6c1","last_reissued_at":"2026-07-05T08:40:19.610041Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T08:40:19.610041Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Let the Expert Stick to His Last: Expert-Specialized Fine-Tuning for Sparse Architectural Large Language Models","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.AI","cs.LG"],"primary_cat":"cs.CL","authors_text":"Damai Dai, Deli Chen, Runxin Xu, Y. Wu, Zhuoshu Li, Zihan Wang","submitted_at":"2024-07-02T03:11:13Z","abstract_excerpt":"Parameter-efficient fine-tuning (PEFT) is crucial for customizing Large Language Models (LLMs) with constrained resources. Although there have been various PEFT methods for dense-architecture LLMs, PEFT for sparse-architecture LLMs is still underexplored. In this work, we study the PEFT method for LLMs with the Mixture-of-Experts (MoE) architecture and the contents of this work are mainly threefold: (1) We investigate the dispersion degree of the activated experts in customized tasks, and found that the routing distribution for a specific task tends to be highly concentrated, while the distrib"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2407.01906","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/2407.01906/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":"2407.01906","created_at":"2026-07-05T08:40:19.610101+00:00"},{"alias_kind":"arxiv_version","alias_value":"2407.01906v2","created_at":"2026-07-05T08:40:19.610101+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2407.01906","created_at":"2026-07-05T08:40:19.610101+00:00"},{"alias_kind":"pith_short_12","alias_value":"QIUKJONVMVAD","created_at":"2026-07-05T08:40:19.610101+00:00"},{"alias_kind":"pith_short_16","alias_value":"QIUKJONVMVADN2KH","created_at":"2026-07-05T08:40:19.610101+00:00"},{"alias_kind":"pith_short_8","alias_value":"QIUKJONV","created_at":"2026-07-05T08:40:19.610101+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":3,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.21645","citing_title":"Behavioral and Representational Evidence of Binomial Ordering Preferences in Large Language Models","ref_index":40,"is_internal_anchor":false},{"citing_arxiv_id":"2604.23036","citing_title":"Preserving Long-Tailed Expert Information in Mixture-of-Experts Tuning","ref_index":30,"is_internal_anchor":false},{"citing_arxiv_id":"2604.06385","citing_title":"Application-Driven Pedagogical Knowledge Optimization of Open-Source LLMs via Reinforcement Learning and Supervised Fine-Tuning","ref_index":11,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/QIUKJONVMVADN2KHIZSM7IZN37","json":"https://pith.science/pith/QIUKJONVMVADN2KHIZSM7IZN37.json","graph_json":"https://pith.science/api/pith-number/QIUKJONVMVADN2KHIZSM7IZN37/graph.json","events_json":"https://pith.science/api/pith-number/QIUKJONVMVADN2KHIZSM7IZN37/events.json","paper":"https://pith.science/paper/QIUKJONV"},"agent_actions":{"view_html":"https://pith.science/pith/QIUKJONVMVADN2KHIZSM7IZN37","download_json":"https://pith.science/pith/QIUKJONVMVADN2KHIZSM7IZN37.json","view_paper":"https://pith.science/paper/QIUKJONV","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2407.01906&json=true","fetch_graph":"https://pith.science/api/pith-number/QIUKJONVMVADN2KHIZSM7IZN37/graph.json","fetch_events":"https://pith.science/api/pith-number/QIUKJONVMVADN2KHIZSM7IZN37/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/QIUKJONVMVADN2KHIZSM7IZN37/action/timestamp_anchor","attest_storage":"https://pith.science/pith/QIUKJONVMVADN2KHIZSM7IZN37/action/storage_attestation","attest_author":"https://pith.science/pith/QIUKJONVMVADN2KHIZSM7IZN37/action/author_attestation","sign_citation":"https://pith.science/pith/QIUKJONVMVADN2KHIZSM7IZN37/action/citation_signature","submit_replication":"https://pith.science/pith/QIUKJONVMVADN2KHIZSM7IZN37/action/replication_record"}},"created_at":"2026-07-05T08:40:19.610101+00:00","updated_at":"2026-07-05T08:40:19.610101+00:00"}