{"state_type":"pith_open_graph_state","state_version":"1.0","pith_number":"pith:2024:TYUR5VO5Q3IF7UIQY4GEYZ7MXB","merge_version":"pith-open-graph-merge-v1","event_count":2,"valid_event_count":2,"invalid_event_count":0,"equivocation_count":0,"current":{"canonical_record":{"metadata":{"abstract_canon_sha256":"d66615526f330a9264996bedb3247ad7eba842a57e6e33363600f93a0e3c8d0d","cross_cats_sorted":["cs.CL"],"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2024-04-15T10:44:31Z","title_canon_sha256":"ba976991f7c1b30ff4ac3769e8a1030d9bd328012af0e987728d61a27aedb338"},"schema_version":"1.0","source":{"id":"2404.09656","kind":"arxiv","version":4}},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2404.09656","created_at":"2026-07-05T10:19:23Z"},{"alias_kind":"arxiv_version","alias_value":"2404.09656v4","created_at":"2026-07-05T10:19:23Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2404.09656","created_at":"2026-07-05T10:19:23Z"},{"alias_kind":"pith_short_12","alias_value":"TYUR5VO5Q3IF","created_at":"2026-07-05T10:19:23Z"},{"alias_kind":"pith_short_16","alias_value":"TYUR5VO5Q3IF7UIQ","created_at":"2026-07-05T10:19:23Z"},{"alias_kind":"pith_short_8","alias_value":"TYUR5VO5","created_at":"2026-07-05T10:19:23Z"}],"graph_snapshots":[{"event_id":"sha256:90a5fef59e5909bd9bbf3bfc0bcc18831991b4ec2213b179b5914a19c13809ea","target":"graph","created_at":"2026-07-05T10:19:23Z","signer":{"key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signer_id":"pith.science","signer_type":"pith_registry"},"payload":{"graph_snapshot":{"author_claims":{"count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57","strong_count":0},"builder_version":"pith-number-builder-2026-05-17-v1","claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"formal_canon":{"evidence_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"integrity":{"available":true,"clean":true,"detectors_run":[],"endpoint":"/pith/2404.09656/integrity.json","findings":[],"snapshot_sha256":"c28c3603d3b5d939e8dc4c7e95fa8dfce3d595e45f758748cecf8e644a296938","summary":{"advisory":0,"by_detector":{},"critical":0,"informational":0}},"paper":{"abstract_excerpt":"Despite the fact that offline methods for Large Language Models (LLMs) alignment do not require a direct reward model, they remain susceptible to overoptimization. This issue arises when the trained model deviates excessively from the reference policy, leading to a decrease in sample quality. We propose a new paradigm of offline alignment methods, called Trust Region (including variants TR-DPO, TR-IPO, TR-KTO), which dynamically updates the reference policy throughout the training process. Our results show that TR alignment methods effectively mitigate overoptimization, enabling models to main","authors_text":"Alexey Gorbatovski, Alexey Malakhov, Boris Shaposhnikov, Daniil Gavrilov, Ian Maksimov, Nikita Balagansky, Nikita Surnachev, Yaroslav Aksenov","cross_cats":["cs.CL"],"headline":"","license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2024-04-15T10:44:31Z","title":"Learn Your Reference Model for Real Good Alignment"},"references":{"count":0,"internal_anchors":0,"resolved_work":0,"sample":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2404.09656","kind":"arxiv","version":4},"verdict":{"created_at":null,"id":null,"model_set":{},"one_line_summary":"","pipeline_version":null,"pith_extraction_headline":"","strongest_claim":"","weakest_assumption":""}},"verdict_id":null}}],"author_attestations":[],"timestamp_anchors":[],"storage_attestations":[],"citation_signatures":[],"replication_records":[],"corrections":[],"mirror_hints":[],"record_created":{"event_id":"sha256:17ed28ecfa56002acce9792a525ed84797172a25bcc1b6126f276b5e6580d55d","target":"record","created_at":"2026-07-05T10:19:23Z","signer":{"key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signer_id":"pith.science","signer_type":"pith_registry"},"payload":{"attestation_state":"computed","canonical_record":{"metadata":{"abstract_canon_sha256":"d66615526f330a9264996bedb3247ad7eba842a57e6e33363600f93a0e3c8d0d","cross_cats_sorted":["cs.CL"],"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2024-04-15T10:44:31Z","title_canon_sha256":"ba976991f7c1b30ff4ac3769e8a1030d9bd328012af0e987728d61a27aedb338"},"schema_version":"1.0","source":{"id":"2404.09656","kind":"arxiv","version":4}},"canonical_sha256":"9e291ed5dd86d05fd110c70c4c67ecb86bd935a84972881d4df420273b7c7f94","receipt":{"algorithm":"ed25519","builder_version":"pith-number-builder-2026-05-17-v1","canonical_sha256":"9e291ed5dd86d05fd110c70c4c67ecb86bd935a84972881d4df420273b7c7f94","first_computed_at":"2026-07-05T10:19:23.404437Z","key_id":"pith-v1-2026-05","kind":"pith_receipt","last_reissued_at":"2026-07-05T10:19:23.404437Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","receipt_version":"0.3","signature_b64":"2YLoY5Bhm1dH/R8Zfo/2So41+Uxc8LwmTwsDHaynWzQoaZEW+GDo5TukBa7081w4X+BH0bGoTiWaCSlxSHLnDQ==","signature_status":"signed_v1","signed_at":"2026-07-05T10:19:23.404993Z","signed_message":"canonical_sha256_bytes"},"source_id":"2404.09656","source_kind":"arxiv","source_version":4}}},"equivocations":[],"invalid_events":[],"applied_event_ids":["sha256:17ed28ecfa56002acce9792a525ed84797172a25bcc1b6126f276b5e6580d55d","sha256:90a5fef59e5909bd9bbf3bfc0bcc18831991b4ec2213b179b5914a19c13809ea"],"state_sha256":"9e6c2872e2b7c125d5353f0049a6593b8886c6366e6d90f6cd01dde27edc999e"}