{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2024:CLNHFN6WUILDKFUSUQMAK6OSOY","short_pith_number":"pith:CLNHFN6W","schema_version":"1.0","canonical_sha256":"12da72b7d6a216351692a4180579d2763fff3d8c3fa26cbcfa8be63664685d48","source":{"kind":"arxiv","id":"2412.15244","version":1},"attestation_state":"computed","paper":{"title":"MPPO: Multi Pair-wise Preference Optimization for LLMs with Arbitrary Negative Samples","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI","cs.LG"],"primary_cat":"cs.CL","authors_text":"Bo Zheng, Fangzhi Zhu, Jiahui Wang, Junxiong Zhu, Kai Zhou, Lulu Wen, Shuo Xie, Wei Dai, Xiaowei Chen","submitted_at":"2024-12-13T14:18:58Z","abstract_excerpt":"Aligning Large Language Models (LLMs) with human feedback is crucial for their development. Existing preference optimization methods such as DPO and KTO, while improved based on Reinforcement Learning from Human Feedback (RLHF), are inherently derived from PPO, requiring a reference model that adds GPU memory resources and relies heavily on abundant preference data. Meanwhile, current preference optimization research mainly targets single-question scenarios with two replies, neglecting optimization with multiple replies, which leads to a waste of data in the application. This study introduces "},"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":"2412.15244","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CL","submitted_at":"2024-12-13T14:18:58Z","cross_cats_sorted":["cs.AI","cs.LG"],"title_canon_sha256":"9f789380d8aab06800e94c1aa8069ec78ac8d963f7f06e54cbae7e3b0a3cff6d","abstract_canon_sha256":"23fd0b157211a1158cc838a27ba921e1aa2cff32a82d6219496641e1f384128e"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T09:52:00.304803Z","signature_b64":"kk9I1lvJSxf9pTQiuimDYSL6fTKf+49pbhv3Mc+7ER3dkUzKis+7KIgNnjaLf/QnppnTxQtVEiQQqM7F/E6+DA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"12da72b7d6a216351692a4180579d2763fff3d8c3fa26cbcfa8be63664685d48","last_reissued_at":"2026-07-05T09:52:00.304264Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T09:52:00.304264Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"MPPO: Multi Pair-wise Preference Optimization for LLMs with Arbitrary Negative Samples","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI","cs.LG"],"primary_cat":"cs.CL","authors_text":"Bo Zheng, Fangzhi Zhu, Jiahui Wang, Junxiong Zhu, Kai Zhou, Lulu Wen, Shuo Xie, Wei Dai, Xiaowei Chen","submitted_at":"2024-12-13T14:18:58Z","abstract_excerpt":"Aligning Large Language Models (LLMs) with human feedback is crucial for their development. Existing preference optimization methods such as DPO and KTO, while improved based on Reinforcement Learning from Human Feedback (RLHF), are inherently derived from PPO, requiring a reference model that adds GPU memory resources and relies heavily on abundant preference data. Meanwhile, current preference optimization research mainly targets single-question scenarios with two replies, neglecting optimization with multiple replies, which leads to a waste of data in the application. This study introduces "},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2412.15244","kind":"arxiv","version":1},"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/2412.15244/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":"2412.15244","created_at":"2026-07-05T09:52:00.304339+00:00"},{"alias_kind":"arxiv_version","alias_value":"2412.15244v1","created_at":"2026-07-05T09:52:00.304339+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2412.15244","created_at":"2026-07-05T09:52:00.304339+00:00"},{"alias_kind":"pith_short_12","alias_value":"CLNHFN6WUILD","created_at":"2026-07-05T09:52:00.304339+00:00"},{"alias_kind":"pith_short_16","alias_value":"CLNHFN6WUILDKFUS","created_at":"2026-07-05T09:52:00.304339+00:00"},{"alias_kind":"pith_short_8","alias_value":"CLNHFN6W","created_at":"2026-07-05T09:52:00.304339+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":1,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2605.10784","citing_title":"MASS-DPO: Multi-negative Active Sample Selection for Direct Policy Optimization","ref_index":67,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/CLNHFN6WUILDKFUSUQMAK6OSOY","json":"https://pith.science/pith/CLNHFN6WUILDKFUSUQMAK6OSOY.json","graph_json":"https://pith.science/api/pith-number/CLNHFN6WUILDKFUSUQMAK6OSOY/graph.json","events_json":"https://pith.science/api/pith-number/CLNHFN6WUILDKFUSUQMAK6OSOY/events.json","paper":"https://pith.science/paper/CLNHFN6W"},"agent_actions":{"view_html":"https://pith.science/pith/CLNHFN6WUILDKFUSUQMAK6OSOY","download_json":"https://pith.science/pith/CLNHFN6WUILDKFUSUQMAK6OSOY.json","view_paper":"https://pith.science/paper/CLNHFN6W","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2412.15244&json=true","fetch_graph":"https://pith.science/api/pith-number/CLNHFN6WUILDKFUSUQMAK6OSOY/graph.json","fetch_events":"https://pith.science/api/pith-number/CLNHFN6WUILDKFUSUQMAK6OSOY/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/CLNHFN6WUILDKFUSUQMAK6OSOY/action/timestamp_anchor","attest_storage":"https://pith.science/pith/CLNHFN6WUILDKFUSUQMAK6OSOY/action/storage_attestation","attest_author":"https://pith.science/pith/CLNHFN6WUILDKFUSUQMAK6OSOY/action/author_attestation","sign_citation":"https://pith.science/pith/CLNHFN6WUILDKFUSUQMAK6OSOY/action/citation_signature","submit_replication":"https://pith.science/pith/CLNHFN6WUILDKFUSUQMAK6OSOY/action/replication_record"}},"created_at":"2026-07-05T09:52:00.304339+00:00","updated_at":"2026-07-05T09:52:00.304339+00:00"}