{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2024:2R3J4VMHDW5LF5EZYSBW4HN75B","short_pith_number":"pith:2R3J4VMH","schema_version":"1.0","canonical_sha256":"d4769e55871dbab2f499c4836e1dbfe87aff2a9b35295e6ba00d1f737695ac1a","source":{"kind":"arxiv","id":"2410.16714","version":3},"attestation_state":"computed","paper":{"title":"Magnetic Preference Optimization: Achieving Last-iterate Convergence for Language Model Alignment","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Chengdong Ma, Jiancong Xiao, Jing Huo, Linjian Meng, Mingzhi Wang, Qizhi Chen, Weijie J. Su, Yang Han, Yaodong Yang, Zhaowei Zhang","submitted_at":"2024-10-22T05:51:34Z","abstract_excerpt":"Self-play methods have demonstrated remarkable success in enhancing model capabilities across various domains. In the context of Reinforcement Learning from Human Feedback (RLHF), self-play not only boosts Large Language Model (LLM) performance but also overcomes the limitations of traditional Bradley-Terry (BT) model assumptions by finding the Nash equilibrium (NE) of a preference-based, two-player constant-sum game. However, existing methods either guarantee only average-iterate convergence, incurring high storage and inference costs, or converge to the NE of a regularized game, failing to a"},"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":"2410.16714","kind":"arxiv","version":3},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CL","submitted_at":"2024-10-22T05:51:34Z","cross_cats_sorted":[],"title_canon_sha256":"e4ab27bc63c95a05f209769637a66a3c3a8981725f911b7ff538144f2d1a6972","abstract_canon_sha256":"f60023bbcddccb37c05013e6e138f20b15103080280b1afc1f2f80aa9d1dcbae"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T10:51:27.164036Z","signature_b64":"cIwRm+uv7RUkSUPqKTckVwYs8XuzhGLpzRwxn55HhVCGAH7PFd880vBlGVtG6AXYhB7+6wGbp6XIQH30dpKzDQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"d4769e55871dbab2f499c4836e1dbfe87aff2a9b35295e6ba00d1f737695ac1a","last_reissued_at":"2026-07-05T10:51:27.163552Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T10:51:27.163552Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Magnetic Preference Optimization: Achieving Last-iterate Convergence for Language Model Alignment","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Chengdong Ma, Jiancong Xiao, Jing Huo, Linjian Meng, Mingzhi Wang, Qizhi Chen, Weijie J. Su, Yang Han, Yaodong Yang, Zhaowei Zhang","submitted_at":"2024-10-22T05:51:34Z","abstract_excerpt":"Self-play methods have demonstrated remarkable success in enhancing model capabilities across various domains. In the context of Reinforcement Learning from Human Feedback (RLHF), self-play not only boosts Large Language Model (LLM) performance but also overcomes the limitations of traditional Bradley-Terry (BT) model assumptions by finding the Nash equilibrium (NE) of a preference-based, two-player constant-sum game. However, existing methods either guarantee only average-iterate convergence, incurring high storage and inference costs, or converge to the NE of a regularized game, failing to a"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2410.16714","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/2410.16714/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":"2410.16714","created_at":"2026-07-05T10:51:27.163609+00:00"},{"alias_kind":"arxiv_version","alias_value":"2410.16714v3","created_at":"2026-07-05T10:51:27.163609+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2410.16714","created_at":"2026-07-05T10:51:27.163609+00:00"},{"alias_kind":"pith_short_12","alias_value":"2R3J4VMHDW5L","created_at":"2026-07-05T10:51:27.163609+00:00"},{"alias_kind":"pith_short_16","alias_value":"2R3J4VMHDW5LF5EZ","created_at":"2026-07-05T10:51:27.163609+00:00"},{"alias_kind":"pith_short_8","alias_value":"2R3J4VMH","created_at":"2026-07-05T10:51:27.163609+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":2,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2509.23102","citing_title":"Multiplayer Nash Preference Optimization","ref_index":30,"is_internal_anchor":false},{"citing_arxiv_id":"2605.13875","citing_title":"Common-agency Games for Multi-Objective Test-Time Alignment","ref_index":81,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/2R3J4VMHDW5LF5EZYSBW4HN75B","json":"https://pith.science/pith/2R3J4VMHDW5LF5EZYSBW4HN75B.json","graph_json":"https://pith.science/api/pith-number/2R3J4VMHDW5LF5EZYSBW4HN75B/graph.json","events_json":"https://pith.science/api/pith-number/2R3J4VMHDW5LF5EZYSBW4HN75B/events.json","paper":"https://pith.science/paper/2R3J4VMH"},"agent_actions":{"view_html":"https://pith.science/pith/2R3J4VMHDW5LF5EZYSBW4HN75B","download_json":"https://pith.science/pith/2R3J4VMHDW5LF5EZYSBW4HN75B.json","view_paper":"https://pith.science/paper/2R3J4VMH","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2410.16714&json=true","fetch_graph":"https://pith.science/api/pith-number/2R3J4VMHDW5LF5EZYSBW4HN75B/graph.json","fetch_events":"https://pith.science/api/pith-number/2R3J4VMHDW5LF5EZYSBW4HN75B/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/2R3J4VMHDW5LF5EZYSBW4HN75B/action/timestamp_anchor","attest_storage":"https://pith.science/pith/2R3J4VMHDW5LF5EZYSBW4HN75B/action/storage_attestation","attest_author":"https://pith.science/pith/2R3J4VMHDW5LF5EZYSBW4HN75B/action/author_attestation","sign_citation":"https://pith.science/pith/2R3J4VMHDW5LF5EZYSBW4HN75B/action/citation_signature","submit_replication":"https://pith.science/pith/2R3J4VMHDW5LF5EZYSBW4HN75B/action/replication_record"}},"created_at":"2026-07-05T10:51:27.163609+00:00","updated_at":"2026-07-05T10:51:27.163609+00:00"}