{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:5OTYQALKCWVPWLSB6AQQ2WJ76I","short_pith_number":"pith:5OTYQALK","schema_version":"1.0","canonical_sha256":"eba788016a15aafb2e41f0210d593ff232fb7295661e2e5d6de61338ea31d7f1","source":{"kind":"arxiv","id":"2505.24461","version":2},"attestation_state":"computed","paper":{"title":"Logits-Based Finetuning","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Chuanyang Zheng, Han Shi, Hong Xu, Jiaya Jia, Jingyao Li, Senqiao Yang, Sitong Wu","submitted_at":"2025-05-30T10:57:09Z","abstract_excerpt":"In recent years, developing compact and efficient large language models (LLMs) has emerged as a thriving area of research. Traditional Supervised Fine-Tuning (SFT), which relies on singular ground truth labels, often fails to capture token-level dependencies and linguistic diversity. To address these limitations, we propose a logits-based fine-tuning framework that integrates the strengths of supervised learning and knowledge distillation. Our approach constructs enriched training targets by combining teacher logits with ground truth labels, preserving both correctness and linguistic diversity"},"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":"2505.24461","kind":"arxiv","version":2},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.LG","submitted_at":"2025-05-30T10:57:09Z","cross_cats_sorted":[],"title_canon_sha256":"ee2479e84ea78d49cd40a347313a5e16401144061fceab381a21e5cdcf8ca011","abstract_canon_sha256":"499c3595faa43238ca41455f30c6e033e7dbf8e52e73f22d9451125514832577"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T11:19:53.108296Z","signature_b64":"0XSJ+um/qAHrllq3rNzlEno0ebk06y9ZLVYFcfXlS9ElFQcfvKnzlHaL+/lyqbX7yK0RqxasFCRR1s6Xb5BKDg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"eba788016a15aafb2e41f0210d593ff232fb7295661e2e5d6de61338ea31d7f1","last_reissued_at":"2026-07-05T11:19:53.107870Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T11:19:53.107870Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Logits-Based Finetuning","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Chuanyang Zheng, Han Shi, Hong Xu, Jiaya Jia, Jingyao Li, Senqiao Yang, Sitong Wu","submitted_at":"2025-05-30T10:57:09Z","abstract_excerpt":"In recent years, developing compact and efficient large language models (LLMs) has emerged as a thriving area of research. Traditional Supervised Fine-Tuning (SFT), which relies on singular ground truth labels, often fails to capture token-level dependencies and linguistic diversity. To address these limitations, we propose a logits-based fine-tuning framework that integrates the strengths of supervised learning and knowledge distillation. Our approach constructs enriched training targets by combining teacher logits with ground truth labels, preserving both correctness and linguistic diversity"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2505.24461","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/2505.24461/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":"2505.24461","created_at":"2026-07-05T11:19:53.107924+00:00"},{"alias_kind":"arxiv_version","alias_value":"2505.24461v2","created_at":"2026-07-05T11:19:53.107924+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2505.24461","created_at":"2026-07-05T11:19:53.107924+00:00"},{"alias_kind":"pith_short_12","alias_value":"5OTYQALKCWVP","created_at":"2026-07-05T11:19:53.107924+00:00"},{"alias_kind":"pith_short_16","alias_value":"5OTYQALKCWVPWLSB","created_at":"2026-07-05T11:19:53.107924+00:00"},{"alias_kind":"pith_short_8","alias_value":"5OTYQALK","created_at":"2026-07-05T11:19:53.107924+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":1,"internal_anchor_count":1,"sample":[{"citing_arxiv_id":"2507.13348","citing_title":"VisionThink: Smart and Efficient Vision Language Model via Reinforcement Learning","ref_index":29,"is_internal_anchor":true}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/5OTYQALKCWVPWLSB6AQQ2WJ76I","json":"https://pith.science/pith/5OTYQALKCWVPWLSB6AQQ2WJ76I.json","graph_json":"https://pith.science/api/pith-number/5OTYQALKCWVPWLSB6AQQ2WJ76I/graph.json","events_json":"https://pith.science/api/pith-number/5OTYQALKCWVPWLSB6AQQ2WJ76I/events.json","paper":"https://pith.science/paper/5OTYQALK"},"agent_actions":{"view_html":"https://pith.science/pith/5OTYQALKCWVPWLSB6AQQ2WJ76I","download_json":"https://pith.science/pith/5OTYQALKCWVPWLSB6AQQ2WJ76I.json","view_paper":"https://pith.science/paper/5OTYQALK","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2505.24461&json=true","fetch_graph":"https://pith.science/api/pith-number/5OTYQALKCWVPWLSB6AQQ2WJ76I/graph.json","fetch_events":"https://pith.science/api/pith-number/5OTYQALKCWVPWLSB6AQQ2WJ76I/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/5OTYQALKCWVPWLSB6AQQ2WJ76I/action/timestamp_anchor","attest_storage":"https://pith.science/pith/5OTYQALKCWVPWLSB6AQQ2WJ76I/action/storage_attestation","attest_author":"https://pith.science/pith/5OTYQALKCWVPWLSB6AQQ2WJ76I/action/author_attestation","sign_citation":"https://pith.science/pith/5OTYQALKCWVPWLSB6AQQ2WJ76I/action/citation_signature","submit_replication":"https://pith.science/pith/5OTYQALKCWVPWLSB6AQQ2WJ76I/action/replication_record"}},"created_at":"2026-07-05T11:19:53.107924+00:00","updated_at":"2026-07-05T11:19:53.107924+00:00"}