{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:VK6RTU7VJXXK5DJDLXOVDFJD2Q","short_pith_number":"pith:VK6RTU7V","schema_version":"1.0","canonical_sha256":"aabd19d3f54deeae8d235ddd519523d439d5fb60d73b8ae7eb1e94ba028dbb42","source":{"kind":"arxiv","id":"2507.00018","version":2},"attestation_state":"computed","paper":{"title":"Implicit Reward as the Bridge: A Unified View of SFT and DPO Connections","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.AI","cs.CL"],"primary_cat":"cs.LG","authors_text":"Bo Wang, Linyang Li, Peiji Li, Qinyuan Cheng, Qipeng Guo, Rong Bao, Runyu Peng, Xipeng Qiu, Yunhua Zhou, Zhiyuan Zeng","submitted_at":"2025-06-15T05:42:29Z","abstract_excerpt":"Post-training processes are essential phases in grounding pre-trained language models to real-world tasks, with learning from demonstrations or preference signals playing a crucial role in this adaptation. We present a unified theoretical framework bridging Supervised Fine-Tuning (SFT) and preference learning in Large Language Model (LLM) post-training. Through rigorous mathematical derivation, we demonstrate that both SFT and preference learning methods like Direct Preference Optimization (DPO) operate within the same optimal policy-reward subspace, with SFT representing a special case of imp"},"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":"2507.00018","kind":"arxiv","version":2},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.LG","submitted_at":"2025-06-15T05:42:29Z","cross_cats_sorted":["cs.AI","cs.CL"],"title_canon_sha256":"ade8200456228d3abd8b8f07eb09b5243faaba49325d76fb01a007b4f143e017","abstract_canon_sha256":"c0ab69816eb364a056b231a7266aefd1df413be74e252190e7abcb2b715aa51a"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T11:31:47.063354Z","signature_b64":"y+6EAMsQZclz6hEPGepZvdGfITmYgZVqbYfNquGSr/KIFZVYLHdWlor+0HlihK1jyam1Hb6gp2T7BurA0UOXCQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"aabd19d3f54deeae8d235ddd519523d439d5fb60d73b8ae7eb1e94ba028dbb42","last_reissued_at":"2026-07-05T11:31:47.062724Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T11:31:47.062724Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Implicit Reward as the Bridge: A Unified View of SFT and DPO Connections","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.AI","cs.CL"],"primary_cat":"cs.LG","authors_text":"Bo Wang, Linyang Li, Peiji Li, Qinyuan Cheng, Qipeng Guo, Rong Bao, Runyu Peng, Xipeng Qiu, Yunhua Zhou, Zhiyuan Zeng","submitted_at":"2025-06-15T05:42:29Z","abstract_excerpt":"Post-training processes are essential phases in grounding pre-trained language models to real-world tasks, with learning from demonstrations or preference signals playing a crucial role in this adaptation. We present a unified theoretical framework bridging Supervised Fine-Tuning (SFT) and preference learning in Large Language Model (LLM) post-training. Through rigorous mathematical derivation, we demonstrate that both SFT and preference learning methods like Direct Preference Optimization (DPO) operate within the same optimal policy-reward subspace, with SFT representing a special case of imp"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2507.00018","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/2507.00018/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":"2507.00018","created_at":"2026-07-05T11:31:47.062798+00:00"},{"alias_kind":"arxiv_version","alias_value":"2507.00018v2","created_at":"2026-07-05T11:31:47.062798+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2507.00018","created_at":"2026-07-05T11:31:47.062798+00:00"},{"alias_kind":"pith_short_12","alias_value":"VK6RTU7VJXXK","created_at":"2026-07-05T11:31:47.062798+00:00"},{"alias_kind":"pith_short_16","alias_value":"VK6RTU7VJXXK5DJD","created_at":"2026-07-05T11:31:47.062798+00:00"},{"alias_kind":"pith_short_8","alias_value":"VK6RTU7V","created_at":"2026-07-05T11:31:47.062798+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":3,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.11206","citing_title":"Compatibility-Aware Dynamic Fine-Tuning for Large Language Models","ref_index":15,"is_internal_anchor":false},{"citing_arxiv_id":"2508.06412","citing_title":"Sample-efficient LLM Optimization with Reset Replay","ref_index":16,"is_internal_anchor":false},{"citing_arxiv_id":"2512.11470","citing_title":"Rethinking Expert Trajectory Utilization in LLM Post-training for Mathematical Reasoning","ref_index":42,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/VK6RTU7VJXXK5DJDLXOVDFJD2Q","json":"https://pith.science/pith/VK6RTU7VJXXK5DJDLXOVDFJD2Q.json","graph_json":"https://pith.science/api/pith-number/VK6RTU7VJXXK5DJDLXOVDFJD2Q/graph.json","events_json":"https://pith.science/api/pith-number/VK6RTU7VJXXK5DJDLXOVDFJD2Q/events.json","paper":"https://pith.science/paper/VK6RTU7V"},"agent_actions":{"view_html":"https://pith.science/pith/VK6RTU7VJXXK5DJDLXOVDFJD2Q","download_json":"https://pith.science/pith/VK6RTU7VJXXK5DJDLXOVDFJD2Q.json","view_paper":"https://pith.science/paper/VK6RTU7V","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2507.00018&json=true","fetch_graph":"https://pith.science/api/pith-number/VK6RTU7VJXXK5DJDLXOVDFJD2Q/graph.json","fetch_events":"https://pith.science/api/pith-number/VK6RTU7VJXXK5DJDLXOVDFJD2Q/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/VK6RTU7VJXXK5DJDLXOVDFJD2Q/action/timestamp_anchor","attest_storage":"https://pith.science/pith/VK6RTU7VJXXK5DJDLXOVDFJD2Q/action/storage_attestation","attest_author":"https://pith.science/pith/VK6RTU7VJXXK5DJDLXOVDFJD2Q/action/author_attestation","sign_citation":"https://pith.science/pith/VK6RTU7VJXXK5DJDLXOVDFJD2Q/action/citation_signature","submit_replication":"https://pith.science/pith/VK6RTU7VJXXK5DJDLXOVDFJD2Q/action/replication_record"}},"created_at":"2026-07-05T11:31:47.062798+00:00","updated_at":"2026-07-05T11:31:47.062798+00:00"}