{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2024:MG44DPSWKC6QW63CHRNA42TICS","short_pith_number":"pith:MG44DPSW","schema_version":"1.0","canonical_sha256":"61b9c1be5650bd0b7b623c5a0e6a6814b10bdea2383059b792b2991a22b74a04","source":{"kind":"arxiv","id":"2406.11817","version":1},"attestation_state":"computed","paper":{"title":"Iterative Length-Regularized Direct Preference Optimization: A Case Study on Improving 7B Language Models to GPT-4 Level","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.AI","cs.LG"],"primary_cat":"cs.CL","authors_text":"Chao Yang, Han-Sen Zhong, Jiaheng Liu, Jie Liu, Wanli Ouyang, Xingyuan Bu, Zhanhui Zhou","submitted_at":"2024-06-17T17:55:38Z","abstract_excerpt":"Direct Preference Optimization (DPO), a standard method for aligning language models with human preferences, is traditionally applied to offline preferences. Recent studies show that DPO benefits from iterative training with online preferences labeled by a trained reward model. In this work, we identify a pitfall of vanilla iterative DPO - improved response quality can lead to increased verbosity. To address this, we introduce iterative length-regularized DPO (iLR-DPO) to penalize response length. Our empirical results show that iLR-DPO can enhance a 7B model to perform on par with GPT-4 witho"},"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":"2406.11817","kind":"arxiv","version":1},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CL","submitted_at":"2024-06-17T17:55:38Z","cross_cats_sorted":["cs.AI","cs.LG"],"title_canon_sha256":"a19c83a05c24259876e835be918f2fe8c27f2da2dbe6cfa22eec141da697dc20","abstract_canon_sha256":"425d36b99bbc2583586da7c8ac9e76b9eb52078c9c50553eb71ef37507f8b7e0"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T08:32:58.274976Z","signature_b64":"SoUmmabDlwPxVaxhZcV68Y0WI/JG76LrSZW8Umm9Cy60rLDhBTMVg6Z47kp4ALsHiQe5u/1t2EuX3x2o/p6LCw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"61b9c1be5650bd0b7b623c5a0e6a6814b10bdea2383059b792b2991a22b74a04","last_reissued_at":"2026-07-05T08:32:58.274556Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T08:32:58.274556Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Iterative Length-Regularized Direct Preference Optimization: A Case Study on Improving 7B Language Models to GPT-4 Level","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.AI","cs.LG"],"primary_cat":"cs.CL","authors_text":"Chao Yang, Han-Sen Zhong, Jiaheng Liu, Jie Liu, Wanli Ouyang, Xingyuan Bu, Zhanhui Zhou","submitted_at":"2024-06-17T17:55:38Z","abstract_excerpt":"Direct Preference Optimization (DPO), a standard method for aligning language models with human preferences, is traditionally applied to offline preferences. Recent studies show that DPO benefits from iterative training with online preferences labeled by a trained reward model. In this work, we identify a pitfall of vanilla iterative DPO - improved response quality can lead to increased verbosity. To address this, we introduce iterative length-regularized DPO (iLR-DPO) to penalize response length. Our empirical results show that iLR-DPO can enhance a 7B model to perform on par with GPT-4 witho"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2406.11817","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/2406.11817/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":"2406.11817","created_at":"2026-07-05T08:32:58.274612+00:00"},{"alias_kind":"arxiv_version","alias_value":"2406.11817v1","created_at":"2026-07-05T08:32:58.274612+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2406.11817","created_at":"2026-07-05T08:32:58.274612+00:00"},{"alias_kind":"pith_short_12","alias_value":"MG44DPSWKC6Q","created_at":"2026-07-05T08:32:58.274612+00:00"},{"alias_kind":"pith_short_16","alias_value":"MG44DPSWKC6QW63C","created_at":"2026-07-05T08:32:58.274612+00:00"},{"alias_kind":"pith_short_8","alias_value":"MG44DPSW","created_at":"2026-07-05T08:32:58.274612+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":1,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2408.15549","citing_title":"WildFeedback: Aligning LLMs With In-situ User Interactions And Feedback","ref_index":23,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/MG44DPSWKC6QW63CHRNA42TICS","json":"https://pith.science/pith/MG44DPSWKC6QW63CHRNA42TICS.json","graph_json":"https://pith.science/api/pith-number/MG44DPSWKC6QW63CHRNA42TICS/graph.json","events_json":"https://pith.science/api/pith-number/MG44DPSWKC6QW63CHRNA42TICS/events.json","paper":"https://pith.science/paper/MG44DPSW"},"agent_actions":{"view_html":"https://pith.science/pith/MG44DPSWKC6QW63CHRNA42TICS","download_json":"https://pith.science/pith/MG44DPSWKC6QW63CHRNA42TICS.json","view_paper":"https://pith.science/paper/MG44DPSW","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2406.11817&json=true","fetch_graph":"https://pith.science/api/pith-number/MG44DPSWKC6QW63CHRNA42TICS/graph.json","fetch_events":"https://pith.science/api/pith-number/MG44DPSWKC6QW63CHRNA42TICS/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/MG44DPSWKC6QW63CHRNA42TICS/action/timestamp_anchor","attest_storage":"https://pith.science/pith/MG44DPSWKC6QW63CHRNA42TICS/action/storage_attestation","attest_author":"https://pith.science/pith/MG44DPSWKC6QW63CHRNA42TICS/action/author_attestation","sign_citation":"https://pith.science/pith/MG44DPSWKC6QW63CHRNA42TICS/action/citation_signature","submit_replication":"https://pith.science/pith/MG44DPSWKC6QW63CHRNA42TICS/action/replication_record"}},"created_at":"2026-07-05T08:32:58.274612+00:00","updated_at":"2026-07-05T08:32:58.274612+00:00"}