{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:XLRPO2HJGLRFHL7FINX4NEQ2HW","short_pith_number":"pith:XLRPO2HJ","schema_version":"1.0","canonical_sha256":"bae2f768e932e253afe5436fc6921a3db38f68f5b5dc4202b4019dd1103463f6","source":{"kind":"arxiv","id":"2507.07725","version":1},"attestation_state":"computed","paper":{"title":"Not All Preferences are What You Need for Post-Training: Selective Alignment Strategy for Preference Optimization","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.CL","authors_text":"Zhijin Dong","submitted_at":"2025-07-10T12:58:45Z","abstract_excerpt":"Post-training alignment of large language models (LLMs) is a critical challenge, as not all tokens contribute equally to model performance. This paper introduces a selective alignment strategy that prioritizes high-impact tokens within preference pairs, leveraging token-level log-probability differences between the current policy and a reference model. By focusing on these informative tokens, our approach reduces computational overhead and enhances alignment fidelity. We further explore the role of reference model quality, demonstrating that stronger reference models significantly improve toke"},"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":"2507.07725","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CL","submitted_at":"2025-07-10T12:58:45Z","cross_cats_sorted":["cs.AI"],"title_canon_sha256":"b5cf0b32854f046f3557d3d898124f842c4702f62283f93585844d9475e6f91d","abstract_canon_sha256":"b239556fe43dee5a61af1de95511530da901619bd6e2b833367ede956f36c23f"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T11:35:04.069284Z","signature_b64":"1lM5FOySWqiOKESBni5nNvcrqlfpZ2tzrXPY7fxFjszAsxJ1NiBvZ8mXGs4nP7RyyopCcA1Ds1UygTVOfFH9Cg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"bae2f768e932e253afe5436fc6921a3db38f68f5b5dc4202b4019dd1103463f6","last_reissued_at":"2026-07-05T11:35:04.068885Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T11:35:04.068885Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Not All Preferences are What You Need for Post-Training: Selective Alignment Strategy for Preference Optimization","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.CL","authors_text":"Zhijin Dong","submitted_at":"2025-07-10T12:58:45Z","abstract_excerpt":"Post-training alignment of large language models (LLMs) is a critical challenge, as not all tokens contribute equally to model performance. This paper introduces a selective alignment strategy that prioritizes high-impact tokens within preference pairs, leveraging token-level log-probability differences between the current policy and a reference model. By focusing on these informative tokens, our approach reduces computational overhead and enhances alignment fidelity. We further explore the role of reference model quality, demonstrating that stronger reference models significantly improve toke"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2507.07725","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/2507.07725/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":"2507.07725","created_at":"2026-07-05T11:35:04.068941+00:00"},{"alias_kind":"arxiv_version","alias_value":"2507.07725v1","created_at":"2026-07-05T11:35:04.068941+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2507.07725","created_at":"2026-07-05T11:35:04.068941+00:00"},{"alias_kind":"pith_short_12","alias_value":"XLRPO2HJGLRF","created_at":"2026-07-05T11:35:04.068941+00:00"},{"alias_kind":"pith_short_16","alias_value":"XLRPO2HJGLRFHL7F","created_at":"2026-07-05T11:35:04.068941+00:00"},{"alias_kind":"pith_short_8","alias_value":"XLRPO2HJ","created_at":"2026-07-05T11:35:04.068941+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":0,"internal_anchor_count":0,"sample":[]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/XLRPO2HJGLRFHL7FINX4NEQ2HW","json":"https://pith.science/pith/XLRPO2HJGLRFHL7FINX4NEQ2HW.json","graph_json":"https://pith.science/api/pith-number/XLRPO2HJGLRFHL7FINX4NEQ2HW/graph.json","events_json":"https://pith.science/api/pith-number/XLRPO2HJGLRFHL7FINX4NEQ2HW/events.json","paper":"https://pith.science/paper/XLRPO2HJ"},"agent_actions":{"view_html":"https://pith.science/pith/XLRPO2HJGLRFHL7FINX4NEQ2HW","download_json":"https://pith.science/pith/XLRPO2HJGLRFHL7FINX4NEQ2HW.json","view_paper":"https://pith.science/paper/XLRPO2HJ","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2507.07725&json=true","fetch_graph":"https://pith.science/api/pith-number/XLRPO2HJGLRFHL7FINX4NEQ2HW/graph.json","fetch_events":"https://pith.science/api/pith-number/XLRPO2HJGLRFHL7FINX4NEQ2HW/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/XLRPO2HJGLRFHL7FINX4NEQ2HW/action/timestamp_anchor","attest_storage":"https://pith.science/pith/XLRPO2HJGLRFHL7FINX4NEQ2HW/action/storage_attestation","attest_author":"https://pith.science/pith/XLRPO2HJGLRFHL7FINX4NEQ2HW/action/author_attestation","sign_citation":"https://pith.science/pith/XLRPO2HJGLRFHL7FINX4NEQ2HW/action/citation_signature","submit_replication":"https://pith.science/pith/XLRPO2HJGLRFHL7FINX4NEQ2HW/action/replication_record"}},"created_at":"2026-07-05T11:35:04.068941+00:00","updated_at":"2026-07-05T11:35:04.068941+00:00"}