{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2024:AHBRFNOIEN5O2KO4S4FS5ZIUIT","short_pith_number":"pith:AHBRFNOI","schema_version":"1.0","canonical_sha256":"01c312b5c8237aed29dc970b2ee51444c363a81e5996f13b0a17512ed0dde28f","source":{"kind":"arxiv","id":"2404.00978","version":2},"attestation_state":"computed","paper":{"title":"Prior Constraints-based Reward Model Training for Aligning Large Language Models","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Chenglong Wang, Chunliang Zhang, Hang Zhou, Jingbo Zhu, Tong Xiao, Yimin Hu","submitted_at":"2024-04-01T07:49:11Z","abstract_excerpt":"Reinforcement learning with human feedback for aligning large language models (LLMs) trains a reward model typically using ranking loss with comparison pairs.However, the training procedure suffers from an inherent problem: the uncontrolled scaling of reward scores during reinforcement learning due to the lack of constraints while training the reward model.This paper proposes a Prior Constraints-based Reward Model (namely PCRM) training method to mitigate this problem. PCRM incorporates prior constraints, specifically, length ratio and cosine similarity between outputs of each comparison pair,"},"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":"2404.00978","kind":"arxiv","version":2},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CL","submitted_at":"2024-04-01T07:49:11Z","cross_cats_sorted":[],"title_canon_sha256":"30538f4fde870f4ecba4522ff93bdbbf3db1f0289741edcdab0351c191ebfcab","abstract_canon_sha256":"7e00b2bf9d279a3f781fb616f4e7758ef1e363efa4b66f0d763833bedc316dfc"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T09:08:24.365457Z","signature_b64":"q9i9RA6DrNPUWlWPEpxBSmYrLxo6OMAs4FMah1Fki115Kd/wQU13x6PsjdnbePlE2ZQ7Ur48gG1eCmak3XyMDA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"01c312b5c8237aed29dc970b2ee51444c363a81e5996f13b0a17512ed0dde28f","last_reissued_at":"2026-07-05T09:08:24.364931Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T09:08:24.364931Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Prior Constraints-based Reward Model Training for Aligning Large Language Models","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Chenglong Wang, Chunliang Zhang, Hang Zhou, Jingbo Zhu, Tong Xiao, Yimin Hu","submitted_at":"2024-04-01T07:49:11Z","abstract_excerpt":"Reinforcement learning with human feedback for aligning large language models (LLMs) trains a reward model typically using ranking loss with comparison pairs.However, the training procedure suffers from an inherent problem: the uncontrolled scaling of reward scores during reinforcement learning due to the lack of constraints while training the reward model.This paper proposes a Prior Constraints-based Reward Model (namely PCRM) training method to mitigate this problem. PCRM incorporates prior constraints, specifically, length ratio and cosine similarity between outputs of each comparison pair,"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2404.00978","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/2404.00978/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":"2404.00978","created_at":"2026-07-05T09:08:24.364997+00:00"},{"alias_kind":"arxiv_version","alias_value":"2404.00978v2","created_at":"2026-07-05T09:08:24.364997+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2404.00978","created_at":"2026-07-05T09:08:24.364997+00:00"},{"alias_kind":"pith_short_12","alias_value":"AHBRFNOIEN5O","created_at":"2026-07-05T09:08:24.364997+00:00"},{"alias_kind":"pith_short_16","alias_value":"AHBRFNOIEN5O2KO4","created_at":"2026-07-05T09:08:24.364997+00:00"},{"alias_kind":"pith_short_8","alias_value":"AHBRFNOI","created_at":"2026-07-05T09:08:24.364997+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/AHBRFNOIEN5O2KO4S4FS5ZIUIT","json":"https://pith.science/pith/AHBRFNOIEN5O2KO4S4FS5ZIUIT.json","graph_json":"https://pith.science/api/pith-number/AHBRFNOIEN5O2KO4S4FS5ZIUIT/graph.json","events_json":"https://pith.science/api/pith-number/AHBRFNOIEN5O2KO4S4FS5ZIUIT/events.json","paper":"https://pith.science/paper/AHBRFNOI"},"agent_actions":{"view_html":"https://pith.science/pith/AHBRFNOIEN5O2KO4S4FS5ZIUIT","download_json":"https://pith.science/pith/AHBRFNOIEN5O2KO4S4FS5ZIUIT.json","view_paper":"https://pith.science/paper/AHBRFNOI","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2404.00978&json=true","fetch_graph":"https://pith.science/api/pith-number/AHBRFNOIEN5O2KO4S4FS5ZIUIT/graph.json","fetch_events":"https://pith.science/api/pith-number/AHBRFNOIEN5O2KO4S4FS5ZIUIT/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/AHBRFNOIEN5O2KO4S4FS5ZIUIT/action/timestamp_anchor","attest_storage":"https://pith.science/pith/AHBRFNOIEN5O2KO4S4FS5ZIUIT/action/storage_attestation","attest_author":"https://pith.science/pith/AHBRFNOIEN5O2KO4S4FS5ZIUIT/action/author_attestation","sign_citation":"https://pith.science/pith/AHBRFNOIEN5O2KO4S4FS5ZIUIT/action/citation_signature","submit_replication":"https://pith.science/pith/AHBRFNOIEN5O2KO4S4FS5ZIUIT/action/replication_record"}},"created_at":"2026-07-05T09:08:24.364997+00:00","updated_at":"2026-07-05T09:08:24.364997+00:00"}