{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2024:NRKPXRUEZTSLQEYDL4TY2MTJND","short_pith_number":"pith:NRKPXRUE","schema_version":"1.0","canonical_sha256":"6c54fbc684cce4b813035f278d326968d197afb80d5ff686d40288fab00b17a4","source":{"kind":"arxiv","id":"2412.04628","version":4},"attestation_state":"computed","paper":{"title":"Multi-Preference Optimization: Generalizing DPO via Set-Level Contrasts","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.AI","cs.CL"],"primary_cat":"cs.LG","authors_text":"Chetan Bansal, Nagarajan Natarajan, Rahul Madhavan, Saravan Rajmohan, Taneesh Gupta, Xuchao Zhang","submitted_at":"2024-12-05T21:50:22Z","abstract_excerpt":"Direct Preference Optimization (DPO) has become a popular approach for aligning language models using pairwise preferences. However, in practical post-training pipelines, on-policy generation typically yields multiple candidate responses per prompt, which are scored by a reward model to guide learning. In this setting, we propose $\\textbf{Multi-Preference Optimization (MPO)}$, a generalization of DPO that optimizes over entire sets of responses by extending the Bradley-Terry model to groupwise comparisons between chosen and rejected sets. To further enhance learning, MPO employs deviation-base"},"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":"2412.04628","kind":"arxiv","version":4},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.LG","submitted_at":"2024-12-05T21:50:22Z","cross_cats_sorted":["cs.AI","cs.CL"],"title_canon_sha256":"182ac3e49c2769bffa390ca94b42c0826b59ca522a2b4bda6a7f8b0589261ef8","abstract_canon_sha256":"09d914937117c78ab329abaf67f1d6e02c14dae08841a17410929cf43269947a"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T11:24:09.786394Z","signature_b64":"QYWziJZrJcFSa+lTW4/jc7wECFqvVmmqDaXenS8dalUW+BufMjL8D16fSgG8kcPRQWSksWPDcrd0trkx3kD6CQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"6c54fbc684cce4b813035f278d326968d197afb80d5ff686d40288fab00b17a4","last_reissued_at":"2026-07-05T11:24:09.785839Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T11:24:09.785839Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Multi-Preference Optimization: Generalizing DPO via Set-Level Contrasts","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.AI","cs.CL"],"primary_cat":"cs.LG","authors_text":"Chetan Bansal, Nagarajan Natarajan, Rahul Madhavan, Saravan Rajmohan, Taneesh Gupta, Xuchao Zhang","submitted_at":"2024-12-05T21:50:22Z","abstract_excerpt":"Direct Preference Optimization (DPO) has become a popular approach for aligning language models using pairwise preferences. However, in practical post-training pipelines, on-policy generation typically yields multiple candidate responses per prompt, which are scored by a reward model to guide learning. In this setting, we propose $\\textbf{Multi-Preference Optimization (MPO)}$, a generalization of DPO that optimizes over entire sets of responses by extending the Bradley-Terry model to groupwise comparisons between chosen and rejected sets. To further enhance learning, MPO employs deviation-base"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2412.04628","kind":"arxiv","version":4},"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/2412.04628/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":"2412.04628","created_at":"2026-07-05T11:24:09.785900+00:00"},{"alias_kind":"arxiv_version","alias_value":"2412.04628v4","created_at":"2026-07-05T11:24:09.785900+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2412.04628","created_at":"2026-07-05T11:24:09.785900+00:00"},{"alias_kind":"pith_short_12","alias_value":"NRKPXRUEZTSL","created_at":"2026-07-05T11:24:09.785900+00:00"},{"alias_kind":"pith_short_16","alias_value":"NRKPXRUEZTSLQEYD","created_at":"2026-07-05T11:24:09.785900+00:00"},{"alias_kind":"pith_short_8","alias_value":"NRKPXRUE","created_at":"2026-07-05T11:24:09.785900+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":2,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2605.08037","citing_title":"Beyond Pairs: Your Language Model is Secretly Optimizing a Preference Graph","ref_index":40,"is_internal_anchor":false},{"citing_arxiv_id":"2604.15602","citing_title":"GroupDPO: Memory efficient Group-wise Direct Preference Optimization","ref_index":11,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/NRKPXRUEZTSLQEYDL4TY2MTJND","json":"https://pith.science/pith/NRKPXRUEZTSLQEYDL4TY2MTJND.json","graph_json":"https://pith.science/api/pith-number/NRKPXRUEZTSLQEYDL4TY2MTJND/graph.json","events_json":"https://pith.science/api/pith-number/NRKPXRUEZTSLQEYDL4TY2MTJND/events.json","paper":"https://pith.science/paper/NRKPXRUE"},"agent_actions":{"view_html":"https://pith.science/pith/NRKPXRUEZTSLQEYDL4TY2MTJND","download_json":"https://pith.science/pith/NRKPXRUEZTSLQEYDL4TY2MTJND.json","view_paper":"https://pith.science/paper/NRKPXRUE","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2412.04628&json=true","fetch_graph":"https://pith.science/api/pith-number/NRKPXRUEZTSLQEYDL4TY2MTJND/graph.json","fetch_events":"https://pith.science/api/pith-number/NRKPXRUEZTSLQEYDL4TY2MTJND/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/NRKPXRUEZTSLQEYDL4TY2MTJND/action/timestamp_anchor","attest_storage":"https://pith.science/pith/NRKPXRUEZTSLQEYDL4TY2MTJND/action/storage_attestation","attest_author":"https://pith.science/pith/NRKPXRUEZTSLQEYDL4TY2MTJND/action/author_attestation","sign_citation":"https://pith.science/pith/NRKPXRUEZTSLQEYDL4TY2MTJND/action/citation_signature","submit_replication":"https://pith.science/pith/NRKPXRUEZTSLQEYDL4TY2MTJND/action/replication_record"}},"created_at":"2026-07-05T11:24:09.785900+00:00","updated_at":"2026-07-05T11:24:09.785900+00:00"}