{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:RRB7USGSSORWAPDUUFL5JWIZW3","short_pith_number":"pith:RRB7USGS","schema_version":"1.0","canonical_sha256":"8c43fa48d293a3603c74a157d4d919b6ff17215420efb7a4f727607c20dfe57a","source":{"kind":"arxiv","id":"2502.18679","version":3},"attestation_state":"computed","paper":{"title":"Discriminative Finetuning of Generative Large Language Models without Reward Models and Human Preference Data","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Changlong Yu, Haoming Jiang, Ilgee Hong, Liang Qiu, Siqi Guo, Tianbao Yang, Tuo Zhao, Vicente Balmaseda, Xin Liu","submitted_at":"2025-02-25T22:38:55Z","abstract_excerpt":"Supervised fine-tuning (SFT) has become a crucial step for aligning pretrained large language models (LLMs) using supervised datasets of input-output pairs. However, despite being supervised, SFT is inherently limited by its generative training objective. To address its limitations, the existing common strategy is to follow SFT with a separate phase of preference optimization (PO), which relies on either human-labeled preference data or a strong reward model to guide the learning process. In this paper, we address the limitations of SFT by exploring one of the most successful techniques in con"},"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":"2502.18679","kind":"arxiv","version":3},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CL","submitted_at":"2025-02-25T22:38:55Z","cross_cats_sorted":[],"title_canon_sha256":"192300af55cced99f61cae9f8da3d35a0f4ff8628c83cf2ec41a0faecd325326","abstract_canon_sha256":"08232e3a831e2815337c12031a433209ec5629c5eec34c2633ceb02357884e88"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T11:42:26.293162Z","signature_b64":"nWjEWe0NcO4k/BnbFlOAtz+GZxPhoaH8DUTdM+Nj52K1BFbRLcmGvPHjZV6sgddFJ55b0kZWuM2myu/qBBYJBQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"8c43fa48d293a3603c74a157d4d919b6ff17215420efb7a4f727607c20dfe57a","last_reissued_at":"2026-07-05T11:42:26.292691Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T11:42:26.292691Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Discriminative Finetuning of Generative Large Language Models without Reward Models and Human Preference Data","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Changlong Yu, Haoming Jiang, Ilgee Hong, Liang Qiu, Siqi Guo, Tianbao Yang, Tuo Zhao, Vicente Balmaseda, Xin Liu","submitted_at":"2025-02-25T22:38:55Z","abstract_excerpt":"Supervised fine-tuning (SFT) has become a crucial step for aligning pretrained large language models (LLMs) using supervised datasets of input-output pairs. However, despite being supervised, SFT is inherently limited by its generative training objective. To address its limitations, the existing common strategy is to follow SFT with a separate phase of preference optimization (PO), which relies on either human-labeled preference data or a strong reward model to guide the learning process. In this paper, we address the limitations of SFT by exploring one of the most successful techniques in con"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2502.18679","kind":"arxiv","version":3},"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/2502.18679/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":"2502.18679","created_at":"2026-07-05T11:42:26.292753+00:00"},{"alias_kind":"arxiv_version","alias_value":"2502.18679v3","created_at":"2026-07-05T11:42:26.292753+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2502.18679","created_at":"2026-07-05T11:42:26.292753+00:00"},{"alias_kind":"pith_short_12","alias_value":"RRB7USGSSORW","created_at":"2026-07-05T11:42:26.292753+00:00"},{"alias_kind":"pith_short_16","alias_value":"RRB7USGSSORWAPDU","created_at":"2026-07-05T11:42:26.292753+00:00"},{"alias_kind":"pith_short_8","alias_value":"RRB7USGS","created_at":"2026-07-05T11:42:26.292753+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/RRB7USGSSORWAPDUUFL5JWIZW3","json":"https://pith.science/pith/RRB7USGSSORWAPDUUFL5JWIZW3.json","graph_json":"https://pith.science/api/pith-number/RRB7USGSSORWAPDUUFL5JWIZW3/graph.json","events_json":"https://pith.science/api/pith-number/RRB7USGSSORWAPDUUFL5JWIZW3/events.json","paper":"https://pith.science/paper/RRB7USGS"},"agent_actions":{"view_html":"https://pith.science/pith/RRB7USGSSORWAPDUUFL5JWIZW3","download_json":"https://pith.science/pith/RRB7USGSSORWAPDUUFL5JWIZW3.json","view_paper":"https://pith.science/paper/RRB7USGS","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2502.18679&json=true","fetch_graph":"https://pith.science/api/pith-number/RRB7USGSSORWAPDUUFL5JWIZW3/graph.json","fetch_events":"https://pith.science/api/pith-number/RRB7USGSSORWAPDUUFL5JWIZW3/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/RRB7USGSSORWAPDUUFL5JWIZW3/action/timestamp_anchor","attest_storage":"https://pith.science/pith/RRB7USGSSORWAPDUUFL5JWIZW3/action/storage_attestation","attest_author":"https://pith.science/pith/RRB7USGSSORWAPDUUFL5JWIZW3/action/author_attestation","sign_citation":"https://pith.science/pith/RRB7USGSSORWAPDUUFL5JWIZW3/action/citation_signature","submit_replication":"https://pith.science/pith/RRB7USGSSORWAPDUUFL5JWIZW3/action/replication_record"}},"created_at":"2026-07-05T11:42:26.292753+00:00","updated_at":"2026-07-05T11:42:26.292753+00:00"}