{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2024:2YS5YSMHHBVSWG4NL5LLBO3ZPT","short_pith_number":"pith:2YS5YSMH","schema_version":"1.0","canonical_sha256":"d625dc4987386b2b1b8d5f56b0bb797cdb8e41b8f761376763b1f7d2f9c0b33c","source":{"kind":"arxiv","id":"2405.12739","version":2},"attestation_state":"computed","paper":{"title":"SPO: Multi-Dimensional Preference Sequential Alignment With Implicit Reward Modeling","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Dong Yan, Jian Xie, Junge Zhang, Kaiqi Huang, Lifeng Liu, Xingzhou Lou","submitted_at":"2024-05-21T12:47:17Z","abstract_excerpt":"Human preference alignment is critical in building powerful and reliable large language models (LLMs). However, current methods either ignore the multi-dimensionality of human preferences (e.g. helpfulness and harmlessness) or struggle with the complexity of managing multiple reward models. To address these issues, we propose Sequential Preference Optimization (SPO), a method that sequentially fine-tunes LLMs to align with multiple dimensions of human preferences. SPO avoids explicit reward modeling, directly optimizing the models to align with nuanced human preferences. We theoretically deriv"},"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":"2405.12739","kind":"arxiv","version":2},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.LG","submitted_at":"2024-05-21T12:47:17Z","cross_cats_sorted":[],"title_canon_sha256":"ae7e60cc2acdc9a73a99d5a95562ca3a13b8d6df6b4c87218b7cbe4d579eeab3","abstract_canon_sha256":"e833a017edc5fc3ada0775388dc2d1f7828fd512f73f5691483fd113f7763984"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T09:19:09.731041Z","signature_b64":"RK43WAzccnEtROWv+SPPL2VdTOSf+PfAivOWcR7kfSYhXJ6TfcGjUobzvmTxNUsD+JawFYjsRFs0OVhd+P4fCw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"d625dc4987386b2b1b8d5f56b0bb797cdb8e41b8f761376763b1f7d2f9c0b33c","last_reissued_at":"2026-07-05T09:19:09.730514Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T09:19:09.730514Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"SPO: Multi-Dimensional Preference Sequential Alignment With Implicit Reward Modeling","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Dong Yan, Jian Xie, Junge Zhang, Kaiqi Huang, Lifeng Liu, Xingzhou Lou","submitted_at":"2024-05-21T12:47:17Z","abstract_excerpt":"Human preference alignment is critical in building powerful and reliable large language models (LLMs). However, current methods either ignore the multi-dimensionality of human preferences (e.g. helpfulness and harmlessness) or struggle with the complexity of managing multiple reward models. To address these issues, we propose Sequential Preference Optimization (SPO), a method that sequentially fine-tunes LLMs to align with multiple dimensions of human preferences. SPO avoids explicit reward modeling, directly optimizing the models to align with nuanced human preferences. We theoretically deriv"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2405.12739","kind":"arxiv","version":2},"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/2405.12739/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":"2405.12739","created_at":"2026-07-05T09:19:09.730591+00:00"},{"alias_kind":"arxiv_version","alias_value":"2405.12739v2","created_at":"2026-07-05T09:19:09.730591+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2405.12739","created_at":"2026-07-05T09:19:09.730591+00:00"},{"alias_kind":"pith_short_12","alias_value":"2YS5YSMHHBVS","created_at":"2026-07-05T09:19:09.730591+00:00"},{"alias_kind":"pith_short_16","alias_value":"2YS5YSMHHBVSWG4N","created_at":"2026-07-05T09:19:09.730591+00:00"},{"alias_kind":"pith_short_8","alias_value":"2YS5YSMH","created_at":"2026-07-05T09:19:09.730591+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":1,"internal_anchor_count":1,"sample":[{"citing_arxiv_id":"2506.02698","citing_title":"Smoothed Preference Optimization via ReNoise Inversion for Aligning Diffusion Models with Varied Human Preferences","ref_index":30,"is_internal_anchor":true}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/2YS5YSMHHBVSWG4NL5LLBO3ZPT","json":"https://pith.science/pith/2YS5YSMHHBVSWG4NL5LLBO3ZPT.json","graph_json":"https://pith.science/api/pith-number/2YS5YSMHHBVSWG4NL5LLBO3ZPT/graph.json","events_json":"https://pith.science/api/pith-number/2YS5YSMHHBVSWG4NL5LLBO3ZPT/events.json","paper":"https://pith.science/paper/2YS5YSMH"},"agent_actions":{"view_html":"https://pith.science/pith/2YS5YSMHHBVSWG4NL5LLBO3ZPT","download_json":"https://pith.science/pith/2YS5YSMHHBVSWG4NL5LLBO3ZPT.json","view_paper":"https://pith.science/paper/2YS5YSMH","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2405.12739&json=true","fetch_graph":"https://pith.science/api/pith-number/2YS5YSMHHBVSWG4NL5LLBO3ZPT/graph.json","fetch_events":"https://pith.science/api/pith-number/2YS5YSMHHBVSWG4NL5LLBO3ZPT/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/2YS5YSMHHBVSWG4NL5LLBO3ZPT/action/timestamp_anchor","attest_storage":"https://pith.science/pith/2YS5YSMHHBVSWG4NL5LLBO3ZPT/action/storage_attestation","attest_author":"https://pith.science/pith/2YS5YSMHHBVSWG4NL5LLBO3ZPT/action/author_attestation","sign_citation":"https://pith.science/pith/2YS5YSMHHBVSWG4NL5LLBO3ZPT/action/citation_signature","submit_replication":"https://pith.science/pith/2YS5YSMHHBVSWG4NL5LLBO3ZPT/action/replication_record"}},"created_at":"2026-07-05T09:19:09.730591+00:00","updated_at":"2026-07-05T09:19:09.730591+00:00"}