{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2023:5JEN5OC3IOLGE2UWMSYTRLH3JY","short_pith_number":"pith:5JEN5OC3","schema_version":"1.0","canonical_sha256":"ea48deb85b4396626a9664b138acfb4e1c29e062fff2e9c94b6aec0aafde086c","source":{"kind":"arxiv","id":"2310.09520","version":4},"attestation_state":"computed","paper":{"title":"Reward-Augmented Decoding: Efficient Controlled Text Generation With a Unidirectional Reward Model","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Colin Raffel, Haikang Deng","submitted_at":"2023-10-14T07:19:47Z","abstract_excerpt":"While large language models have proven effective in a huge range of downstream applications, they often generate text that is problematic or lacks a desired attribute. In this paper, we introduce Reward-Augmented Decoding (RAD), a text generation procedure that uses a small unidirectional reward model to encourage a language model to generate text that has certain properties. Specifically, RAD uses the reward model to score generations as they are produced and rescales sampling probabilities to favor high-reward tokens. By using a unidirectional reward model, RAD can cache activations from pr"},"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":"2310.09520","kind":"arxiv","version":4},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CL","submitted_at":"2023-10-14T07:19:47Z","cross_cats_sorted":[],"title_canon_sha256":"553a4df829933b749f5a34e11c60ba439e9c0ef6273f08a0d3f4a521d3880880","abstract_canon_sha256":"a45fd738a200d93aa42107541ecc0d30048e4853938a00e968197c943b95472d"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T07:29:28.876420Z","signature_b64":"aMsrusmpYByAP8CDIT1LoWq5b+SuSLj9qHsC5UF/43zlR4MK5UGOtfYEfvdDTIiRi55i8EaO4d+QRoZ2dxClCQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"ea48deb85b4396626a9664b138acfb4e1c29e062fff2e9c94b6aec0aafde086c","last_reissued_at":"2026-07-05T07:29:28.875980Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T07:29:28.875980Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Reward-Augmented Decoding: Efficient Controlled Text Generation With a Unidirectional Reward Model","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Colin Raffel, Haikang Deng","submitted_at":"2023-10-14T07:19:47Z","abstract_excerpt":"While large language models have proven effective in a huge range of downstream applications, they often generate text that is problematic or lacks a desired attribute. In this paper, we introduce Reward-Augmented Decoding (RAD), a text generation procedure that uses a small unidirectional reward model to encourage a language model to generate text that has certain properties. Specifically, RAD uses the reward model to score generations as they are produced and rescales sampling probabilities to favor high-reward tokens. By using a unidirectional reward model, RAD can cache activations from pr"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2310.09520","kind":"arxiv","version":4},"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/2310.09520/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":"2310.09520","created_at":"2026-07-05T07:29:28.876035+00:00"},{"alias_kind":"arxiv_version","alias_value":"2310.09520v4","created_at":"2026-07-05T07:29:28.876035+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2310.09520","created_at":"2026-07-05T07:29:28.876035+00:00"},{"alias_kind":"pith_short_12","alias_value":"5JEN5OC3IOLG","created_at":"2026-07-05T07:29:28.876035+00:00"},{"alias_kind":"pith_short_16","alias_value":"5JEN5OC3IOLGE2UW","created_at":"2026-07-05T07:29:28.876035+00:00"},{"alias_kind":"pith_short_8","alias_value":"5JEN5OC3","created_at":"2026-07-05T07:29:28.876035+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":1,"internal_anchor_count":1,"sample":[{"citing_arxiv_id":"2502.02153","citing_title":"Vulnerability Mitigation for Safety-Aligned Language Models via Debiasing","ref_index":11,"is_internal_anchor":true}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/5JEN5OC3IOLGE2UWMSYTRLH3JY","json":"https://pith.science/pith/5JEN5OC3IOLGE2UWMSYTRLH3JY.json","graph_json":"https://pith.science/api/pith-number/5JEN5OC3IOLGE2UWMSYTRLH3JY/graph.json","events_json":"https://pith.science/api/pith-number/5JEN5OC3IOLGE2UWMSYTRLH3JY/events.json","paper":"https://pith.science/paper/5JEN5OC3"},"agent_actions":{"view_html":"https://pith.science/pith/5JEN5OC3IOLGE2UWMSYTRLH3JY","download_json":"https://pith.science/pith/5JEN5OC3IOLGE2UWMSYTRLH3JY.json","view_paper":"https://pith.science/paper/5JEN5OC3","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2310.09520&json=true","fetch_graph":"https://pith.science/api/pith-number/5JEN5OC3IOLGE2UWMSYTRLH3JY/graph.json","fetch_events":"https://pith.science/api/pith-number/5JEN5OC3IOLGE2UWMSYTRLH3JY/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/5JEN5OC3IOLGE2UWMSYTRLH3JY/action/timestamp_anchor","attest_storage":"https://pith.science/pith/5JEN5OC3IOLGE2UWMSYTRLH3JY/action/storage_attestation","attest_author":"https://pith.science/pith/5JEN5OC3IOLGE2UWMSYTRLH3JY/action/author_attestation","sign_citation":"https://pith.science/pith/5JEN5OC3IOLGE2UWMSYTRLH3JY/action/citation_signature","submit_replication":"https://pith.science/pith/5JEN5OC3IOLGE2UWMSYTRLH3JY/action/replication_record"}},"created_at":"2026-07-05T07:29:28.876035+00:00","updated_at":"2026-07-05T07:29:28.876035+00:00"}