{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2023:IMS34KDQ7557O6ZYJE2BKXIBH4","short_pith_number":"pith:IMS34KDQ","schema_version":"1.0","canonical_sha256":"4325be2870ff7bf77b384934155d013f347f154c6039b5799f3d9ab997332589","source":{"kind":"arxiv","id":"2310.06927","version":2},"attestation_state":"computed","paper":{"title":"Sparse Fine-tuning for Inference Acceleration of Large Language Models","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.CL","authors_text":"Dan Alistarh, Denis Kuznedelev, Eldar Kurtic, Elias Frantar, Michael Goin","submitted_at":"2023-10-10T18:28:38Z","abstract_excerpt":"We consider the problem of accurate sparse fine-tuning of large language models (LLMs), that is, fine-tuning pretrained LLMs on specialized tasks, while inducing sparsity in their weights. On the accuracy side, we observe that standard loss-based fine-tuning may fail to recover accuracy, especially at high sparsities. To address this, we perform a detailed study of distillation-type losses, determining an L2-based distillation approach we term SquareHead which enables accurate recovery even at higher sparsities, across all model types. On the practical efficiency side, we show that sparse LLMs"},"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":"2310.06927","kind":"arxiv","version":2},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CL","submitted_at":"2023-10-10T18:28:38Z","cross_cats_sorted":["cs.AI"],"title_canon_sha256":"62c22d83cc6ea112921536f0a2332706cc53d51d9ba2057b80be9a2223000ae0","abstract_canon_sha256":"2665dfd95c139217881dc1402107b5ded008b3d259d07fe1b6c2f4bf16c16ff7"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T07:00:29.698790Z","signature_b64":"f40UH/mNHcdYCs3RqUtDWUNJ/J9iQbikavR7V6ABhF3Hb+iJFecVg0PVpyZXNSmq8AP8Esr6SieDoH2LWeigAg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"4325be2870ff7bf77b384934155d013f347f154c6039b5799f3d9ab997332589","last_reissued_at":"2026-07-05T07:00:29.698372Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T07:00:29.698372Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Sparse Fine-tuning for Inference Acceleration of Large Language Models","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.CL","authors_text":"Dan Alistarh, Denis Kuznedelev, Eldar Kurtic, Elias Frantar, Michael Goin","submitted_at":"2023-10-10T18:28:38Z","abstract_excerpt":"We consider the problem of accurate sparse fine-tuning of large language models (LLMs), that is, fine-tuning pretrained LLMs on specialized tasks, while inducing sparsity in their weights. On the accuracy side, we observe that standard loss-based fine-tuning may fail to recover accuracy, especially at high sparsities. To address this, we perform a detailed study of distillation-type losses, determining an L2-based distillation approach we term SquareHead which enables accurate recovery even at higher sparsities, across all model types. On the practical efficiency side, we show that sparse LLMs"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2310.06927","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/2310.06927/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":"2310.06927","created_at":"2026-07-05T07:00:29.698431+00:00"},{"alias_kind":"arxiv_version","alias_value":"2310.06927v2","created_at":"2026-07-05T07:00:29.698431+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2310.06927","created_at":"2026-07-05T07:00:29.698431+00:00"},{"alias_kind":"pith_short_12","alias_value":"IMS34KDQ7557","created_at":"2026-07-05T07:00:29.698431+00:00"},{"alias_kind":"pith_short_16","alias_value":"IMS34KDQ7557O6ZY","created_at":"2026-07-05T07:00:29.698431+00:00"},{"alias_kind":"pith_short_8","alias_value":"IMS34KDQ","created_at":"2026-07-05T07:00:29.698431+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":1,"internal_anchor_count":1,"sample":[{"citing_arxiv_id":"2505.23807","citing_title":"DLP: Dynamic Layerwise Pruning in Large Language Models","ref_index":11,"is_internal_anchor":true}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/IMS34KDQ7557O6ZYJE2BKXIBH4","json":"https://pith.science/pith/IMS34KDQ7557O6ZYJE2BKXIBH4.json","graph_json":"https://pith.science/api/pith-number/IMS34KDQ7557O6ZYJE2BKXIBH4/graph.json","events_json":"https://pith.science/api/pith-number/IMS34KDQ7557O6ZYJE2BKXIBH4/events.json","paper":"https://pith.science/paper/IMS34KDQ"},"agent_actions":{"view_html":"https://pith.science/pith/IMS34KDQ7557O6ZYJE2BKXIBH4","download_json":"https://pith.science/pith/IMS34KDQ7557O6ZYJE2BKXIBH4.json","view_paper":"https://pith.science/paper/IMS34KDQ","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2310.06927&json=true","fetch_graph":"https://pith.science/api/pith-number/IMS34KDQ7557O6ZYJE2BKXIBH4/graph.json","fetch_events":"https://pith.science/api/pith-number/IMS34KDQ7557O6ZYJE2BKXIBH4/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/IMS34KDQ7557O6ZYJE2BKXIBH4/action/timestamp_anchor","attest_storage":"https://pith.science/pith/IMS34KDQ7557O6ZYJE2BKXIBH4/action/storage_attestation","attest_author":"https://pith.science/pith/IMS34KDQ7557O6ZYJE2BKXIBH4/action/author_attestation","sign_citation":"https://pith.science/pith/IMS34KDQ7557O6ZYJE2BKXIBH4/action/citation_signature","submit_replication":"https://pith.science/pith/IMS34KDQ7557O6ZYJE2BKXIBH4/action/replication_record"}},"created_at":"2026-07-05T07:00:29.698431+00:00","updated_at":"2026-07-05T07:00:29.698431+00:00"}