{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:HQGQPY4BC4PMZMWJ2KMYDBCXIL","short_pith_number":"pith:HQGQPY4B","schema_version":"1.0","canonical_sha256":"3c0d07e381171eccb2c9d29981845742cf5dcf543a040030dcaa9d2a792fe6f5","source":{"kind":"arxiv","id":"2506.11120","version":1},"attestation_state":"computed","paper":{"title":"SDMPrune: Self-Distillation MLP Pruning for Efficient Large Language Models","license":"http://creativecommons.org/licenses/by-nc-sa/4.0/","headline":"","cross_cats":["cs.AI","cs.LG"],"primary_cat":"cs.CL","authors_text":"Chengchao Shen, Hourun Zhu","submitted_at":"2025-06-10T02:24:32Z","abstract_excerpt":"In spite of strong performance achieved by LLMs, the costs of their deployment are unaffordable. For the compression of LLMs, gradient-based pruning methods present promising effectiveness. However, in these methods, the gradient computation with one-hot labels ignore the potential predictions on other words, thus missing key information for generative capability of the original model. To address this issue, we introduce a self-distillation loss during the pruning phase (rather than post-training) to fully exploit the predictions of the original model, thereby obtaining more accurate gradient "},"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":"2506.11120","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by-nc-sa/4.0/","primary_cat":"cs.CL","submitted_at":"2025-06-10T02:24:32Z","cross_cats_sorted":["cs.AI","cs.LG"],"title_canon_sha256":"879f5bd5b6c6f8e41c6ee1cdbd9ab9efc1c73a289aaf216bc7059b63ac095f10","abstract_canon_sha256":"3990792a1c657d0c1f8838e11a7c5604db06ed3ec48cfea6aa0c546306c6c42c"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T11:20:59.834283Z","signature_b64":"bLL0xrsM7tx9qtJVwU3jND+XGS3so6uCDEzNnxJ5e8il/CCyWDy0utZxSOZELQxf1ZZ1Vis/MeUS0DgqcwblDQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"3c0d07e381171eccb2c9d29981845742cf5dcf543a040030dcaa9d2a792fe6f5","last_reissued_at":"2026-07-05T11:20:59.833836Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T11:20:59.833836Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"SDMPrune: Self-Distillation MLP Pruning for Efficient Large Language Models","license":"http://creativecommons.org/licenses/by-nc-sa/4.0/","headline":"","cross_cats":["cs.AI","cs.LG"],"primary_cat":"cs.CL","authors_text":"Chengchao Shen, Hourun Zhu","submitted_at":"2025-06-10T02:24:32Z","abstract_excerpt":"In spite of strong performance achieved by LLMs, the costs of their deployment are unaffordable. For the compression of LLMs, gradient-based pruning methods present promising effectiveness. However, in these methods, the gradient computation with one-hot labels ignore the potential predictions on other words, thus missing key information for generative capability of the original model. To address this issue, we introduce a self-distillation loss during the pruning phase (rather than post-training) to fully exploit the predictions of the original model, thereby obtaining more accurate gradient "},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2506.11120","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/2506.11120/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":"2506.11120","created_at":"2026-07-05T11:20:59.833894+00:00"},{"alias_kind":"arxiv_version","alias_value":"2506.11120v1","created_at":"2026-07-05T11:20:59.833894+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2506.11120","created_at":"2026-07-05T11:20:59.833894+00:00"},{"alias_kind":"pith_short_12","alias_value":"HQGQPY4BC4PM","created_at":"2026-07-05T11:20:59.833894+00:00"},{"alias_kind":"pith_short_16","alias_value":"HQGQPY4BC4PMZMWJ","created_at":"2026-07-05T11:20:59.833894+00:00"},{"alias_kind":"pith_short_8","alias_value":"HQGQPY4B","created_at":"2026-07-05T11:20:59.833894+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":1,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.05538","citing_title":"Less is MoE: Trimming Experts in Domain-Specialist Language Models","ref_index":58,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/HQGQPY4BC4PMZMWJ2KMYDBCXIL","json":"https://pith.science/pith/HQGQPY4BC4PMZMWJ2KMYDBCXIL.json","graph_json":"https://pith.science/api/pith-number/HQGQPY4BC4PMZMWJ2KMYDBCXIL/graph.json","events_json":"https://pith.science/api/pith-number/HQGQPY4BC4PMZMWJ2KMYDBCXIL/events.json","paper":"https://pith.science/paper/HQGQPY4B"},"agent_actions":{"view_html":"https://pith.science/pith/HQGQPY4BC4PMZMWJ2KMYDBCXIL","download_json":"https://pith.science/pith/HQGQPY4BC4PMZMWJ2KMYDBCXIL.json","view_paper":"https://pith.science/paper/HQGQPY4B","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2506.11120&json=true","fetch_graph":"https://pith.science/api/pith-number/HQGQPY4BC4PMZMWJ2KMYDBCXIL/graph.json","fetch_events":"https://pith.science/api/pith-number/HQGQPY4BC4PMZMWJ2KMYDBCXIL/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/HQGQPY4BC4PMZMWJ2KMYDBCXIL/action/timestamp_anchor","attest_storage":"https://pith.science/pith/HQGQPY4BC4PMZMWJ2KMYDBCXIL/action/storage_attestation","attest_author":"https://pith.science/pith/HQGQPY4BC4PMZMWJ2KMYDBCXIL/action/author_attestation","sign_citation":"https://pith.science/pith/HQGQPY4BC4PMZMWJ2KMYDBCXIL/action/citation_signature","submit_replication":"https://pith.science/pith/HQGQPY4BC4PMZMWJ2KMYDBCXIL/action/replication_record"}},"created_at":"2026-07-05T11:20:59.833894+00:00","updated_at":"2026-07-05T11:20:59.833894+00:00"}