{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2022:INDL2EPIQYDKKI736XYZ5POBOV","short_pith_number":"pith:INDL2EPI","schema_version":"1.0","canonical_sha256":"4346bd11e88606a523fbf5f19ebdc17540cbcc3dee512cd62711ae636d698002","source":{"kind":"arxiv","id":"2210.15191","version":1},"attestation_state":"computed","paper":{"title":"Truncation Sampling as Language Model Desmoothing","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Christopher D. Manning, John Hewitt, Percy Liang","submitted_at":"2022-10-27T05:52:35Z","abstract_excerpt":"Long samples of text from neural language models can be of poor quality. Truncation sampling algorithms--like top-$p$ or top-$k$ -- address this by setting some words' probabilities to zero at each step. This work provides framing for the aim of truncation, and an improved algorithm for that aim. We propose thinking of a neural language model as a mixture of a true distribution and a smoothing distribution that avoids infinite perplexity. In this light, truncation algorithms aim to perform desmoothing, estimating a subset of the support of the true distribution. Finding a good subset is crucia"},"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":"2210.15191","kind":"arxiv","version":1},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CL","submitted_at":"2022-10-27T05:52:35Z","cross_cats_sorted":[],"title_canon_sha256":"2c587b70da4dd34e4a238606e98f4bcb8a389fec82f0c54d85a5c9a7f2a7f73e","abstract_canon_sha256":"8fae0f12ce88e61be367abe33168afd0b7a538cc5c544b7cd3dd1f90d74efae6"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T05:11:03.273714Z","signature_b64":"TWrDiNMjMakfcL0U6nPhH2HEspmR3xu19e+nKoNa/qLZIuP0PviFo0ZgjJYrveLpbjJN5OdmduPDmHQKvdR2BA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"4346bd11e88606a523fbf5f19ebdc17540cbcc3dee512cd62711ae636d698002","last_reissued_at":"2026-07-05T05:11:03.273244Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T05:11:03.273244Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Truncation Sampling as Language Model Desmoothing","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Christopher D. Manning, John Hewitt, Percy Liang","submitted_at":"2022-10-27T05:52:35Z","abstract_excerpt":"Long samples of text from neural language models can be of poor quality. Truncation sampling algorithms--like top-$p$ or top-$k$ -- address this by setting some words' probabilities to zero at each step. This work provides framing for the aim of truncation, and an improved algorithm for that aim. We propose thinking of a neural language model as a mixture of a true distribution and a smoothing distribution that avoids infinite perplexity. In this light, truncation algorithms aim to perform desmoothing, estimating a subset of the support of the true distribution. Finding a good subset is crucia"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2210.15191","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/2210.15191/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":"2210.15191","created_at":"2026-07-05T05:11:03.273310+00:00"},{"alias_kind":"arxiv_version","alias_value":"2210.15191v1","created_at":"2026-07-05T05:11:03.273310+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2210.15191","created_at":"2026-07-05T05:11:03.273310+00:00"},{"alias_kind":"pith_short_12","alias_value":"INDL2EPIQYDK","created_at":"2026-07-05T05:11:03.273310+00:00"},{"alias_kind":"pith_short_16","alias_value":"INDL2EPIQYDKKI73","created_at":"2026-07-05T05:11:03.273310+00:00"},{"alias_kind":"pith_short_8","alias_value":"INDL2EPI","created_at":"2026-07-05T05:11:03.273310+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":6,"internal_anchor_count":1,"sample":[{"citing_arxiv_id":"2607.08196","citing_title":"A First-Principles Theory of Slow Thinking and Active Perception","ref_index":74,"is_internal_anchor":true},{"citing_arxiv_id":"2606.27359","citing_title":"When are likely answers right? On Sequence Probability and Correctness in LLMs","ref_index":19,"is_internal_anchor":false},{"citing_arxiv_id":"2509.02510","citing_title":"Top-H Decoding: Adapting the Creativity and Coherence with Bounded Entropy in Text Generation","ref_index":9,"is_internal_anchor":false},{"citing_arxiv_id":"2602.10346","citing_title":"Geometry-Aware Decoding with Wasserstein-Regularized Truncation and Mass Penalties for Large Language Models","ref_index":12,"is_internal_anchor":false},{"citing_arxiv_id":"2401.10774","citing_title":"Medusa: Simple LLM Inference Acceleration Framework with Multiple Decoding Heads","ref_index":18,"is_internal_anchor":false},{"citing_arxiv_id":"2604.20500","citing_title":"Efficient Test-Time Inference via Deterministic Exploration of Truncated Decoding Trees","ref_index":6,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/INDL2EPIQYDKKI736XYZ5POBOV","json":"https://pith.science/pith/INDL2EPIQYDKKI736XYZ5POBOV.json","graph_json":"https://pith.science/api/pith-number/INDL2EPIQYDKKI736XYZ5POBOV/graph.json","events_json":"https://pith.science/api/pith-number/INDL2EPIQYDKKI736XYZ5POBOV/events.json","paper":"https://pith.science/paper/INDL2EPI"},"agent_actions":{"view_html":"https://pith.science/pith/INDL2EPIQYDKKI736XYZ5POBOV","download_json":"https://pith.science/pith/INDL2EPIQYDKKI736XYZ5POBOV.json","view_paper":"https://pith.science/paper/INDL2EPI","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2210.15191&json=true","fetch_graph":"https://pith.science/api/pith-number/INDL2EPIQYDKKI736XYZ5POBOV/graph.json","fetch_events":"https://pith.science/api/pith-number/INDL2EPIQYDKKI736XYZ5POBOV/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/INDL2EPIQYDKKI736XYZ5POBOV/action/timestamp_anchor","attest_storage":"https://pith.science/pith/INDL2EPIQYDKKI736XYZ5POBOV/action/storage_attestation","attest_author":"https://pith.science/pith/INDL2EPIQYDKKI736XYZ5POBOV/action/author_attestation","sign_citation":"https://pith.science/pith/INDL2EPIQYDKKI736XYZ5POBOV/action/citation_signature","submit_replication":"https://pith.science/pith/INDL2EPIQYDKKI736XYZ5POBOV/action/replication_record"}},"created_at":"2026-07-05T05:11:03.273310+00:00","updated_at":"2026-07-05T05:11:03.273310+00:00"}