{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:O67PA22GKLW72THSYOZS4XVMLV","short_pith_number":"pith:O67PA22G","schema_version":"1.0","canonical_sha256":"77bef06b4652edfd4cf2c3b32e5eac5d76860f8ca912ab3114b3f72c25d19447","source":{"kind":"arxiv","id":"2504.05410","version":2},"attestation_state":"computed","paper":{"title":"Fast Controlled Generation from Language Models with Adaptive Weighted Rejection Sampling","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI","cs.LG"],"primary_cat":"cs.CL","authors_text":"Alexander K. Lew, Benjamin Lebrun, Benjamin Lipkin, David R. MacIver, Jacob Hoover Vigly, Jason Eisner, Jo\\~ao Loula, Li Du, Ryan Cotterell, Timothy J. O'Donnell, Tim Vieira, Vikash Mansinghka","submitted_at":"2025-04-07T18:30:18Z","abstract_excerpt":"The dominant approach to generating from language models subject to some constraint is locally constrained decoding (LCD), incrementally sampling tokens at each time step such that the constraint is never violated. Typically, this is achieved through token masking: looping over the vocabulary and excluding non-conforming tokens. There are two important problems with this approach. (i) Evaluating the constraint on every token can be prohibitively expensive -- LM vocabularies often exceed $100,000$ tokens. (ii) LCD can distort the global distribution over strings, sampling tokens based only on l"},"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":"2504.05410","kind":"arxiv","version":2},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CL","submitted_at":"2025-04-07T18:30:18Z","cross_cats_sorted":["cs.AI","cs.LG"],"title_canon_sha256":"8c4ef6b708efb65f27f70d9ced87b9f3952eeec513248acc0a00a7164dc656bc","abstract_canon_sha256":"ffbef51dbae83f3c4f1c8f3313ff77535bd8d06bd5897b53aabd923ae7e27580"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T11:55:18.316734Z","signature_b64":"0PHCOvAepWcxSmQSPOmBT9xlPRDN8lI0M598yX1mA7DaifGx/bspb1cZM6fJHuJIb6U3JcXKzuWt2S0hBnKZAw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"77bef06b4652edfd4cf2c3b32e5eac5d76860f8ca912ab3114b3f72c25d19447","last_reissued_at":"2026-07-05T11:55:18.316162Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T11:55:18.316162Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Fast Controlled Generation from Language Models with Adaptive Weighted Rejection Sampling","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI","cs.LG"],"primary_cat":"cs.CL","authors_text":"Alexander K. Lew, Benjamin Lebrun, Benjamin Lipkin, David R. MacIver, Jacob Hoover Vigly, Jason Eisner, Jo\\~ao Loula, Li Du, Ryan Cotterell, Timothy J. O'Donnell, Tim Vieira, Vikash Mansinghka","submitted_at":"2025-04-07T18:30:18Z","abstract_excerpt":"The dominant approach to generating from language models subject to some constraint is locally constrained decoding (LCD), incrementally sampling tokens at each time step such that the constraint is never violated. Typically, this is achieved through token masking: looping over the vocabulary and excluding non-conforming tokens. There are two important problems with this approach. (i) Evaluating the constraint on every token can be prohibitively expensive -- LM vocabularies often exceed $100,000$ tokens. (ii) LCD can distort the global distribution over strings, sampling tokens based only on l"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2504.05410","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/2504.05410/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":"2504.05410","created_at":"2026-07-05T11:55:18.316249+00:00"},{"alias_kind":"arxiv_version","alias_value":"2504.05410v2","created_at":"2026-07-05T11:55:18.316249+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2504.05410","created_at":"2026-07-05T11:55:18.316249+00:00"},{"alias_kind":"pith_short_12","alias_value":"O67PA22GKLW7","created_at":"2026-07-05T11:55:18.316249+00:00"},{"alias_kind":"pith_short_16","alias_value":"O67PA22GKLW72THS","created_at":"2026-07-05T11:55:18.316249+00:00"},{"alias_kind":"pith_short_8","alias_value":"O67PA22G","created_at":"2026-07-05T11:55:18.316249+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":3,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.31808","citing_title":"Large Databases Need Small, Open-Weight Language Models","ref_index":34,"is_internal_anchor":false},{"citing_arxiv_id":"2605.15365","citing_title":"Greedy or not, here I come: Language production under vocabulary constraints in humans and resource-rational models","ref_index":54,"is_internal_anchor":false},{"citing_arxiv_id":"2604.16453","citing_title":"Sampling for Quality: Training-Free Reward-Guided LLM Decoding via Sequential Monte Carlo","ref_index":12,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/O67PA22GKLW72THSYOZS4XVMLV","json":"https://pith.science/pith/O67PA22GKLW72THSYOZS4XVMLV.json","graph_json":"https://pith.science/api/pith-number/O67PA22GKLW72THSYOZS4XVMLV/graph.json","events_json":"https://pith.science/api/pith-number/O67PA22GKLW72THSYOZS4XVMLV/events.json","paper":"https://pith.science/paper/O67PA22G"},"agent_actions":{"view_html":"https://pith.science/pith/O67PA22GKLW72THSYOZS4XVMLV","download_json":"https://pith.science/pith/O67PA22GKLW72THSYOZS4XVMLV.json","view_paper":"https://pith.science/paper/O67PA22G","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2504.05410&json=true","fetch_graph":"https://pith.science/api/pith-number/O67PA22GKLW72THSYOZS4XVMLV/graph.json","fetch_events":"https://pith.science/api/pith-number/O67PA22GKLW72THSYOZS4XVMLV/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/O67PA22GKLW72THSYOZS4XVMLV/action/timestamp_anchor","attest_storage":"https://pith.science/pith/O67PA22GKLW72THSYOZS4XVMLV/action/storage_attestation","attest_author":"https://pith.science/pith/O67PA22GKLW72THSYOZS4XVMLV/action/author_attestation","sign_citation":"https://pith.science/pith/O67PA22GKLW72THSYOZS4XVMLV/action/citation_signature","submit_replication":"https://pith.science/pith/O67PA22GKLW72THSYOZS4XVMLV/action/replication_record"}},"created_at":"2026-07-05T11:55:18.316249+00:00","updated_at":"2026-07-05T11:55:18.316249+00:00"}