{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2021:OYOZHXSVLVCA7LSRHOQUHQ4EY2","short_pith_number":"pith:OYOZHXSV","schema_version":"1.0","canonical_sha256":"761d93de555d440fae513ba143c384c6ab3143849be5780528594fbc5a5abe01","source":{"kind":"arxiv","id":"2104.05218","version":2},"attestation_state":"computed","paper":{"title":"FUDGE: Controlled Text Generation With Future Discriminators","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.LG"],"primary_cat":"cs.CL","authors_text":"Dan Klein, Kevin Yang","submitted_at":"2021-04-12T05:59:53Z","abstract_excerpt":"We propose Future Discriminators for Generation (FUDGE), a flexible and modular method for controlled text generation. Given a pre-existing model G for generating text from a distribution of interest, FUDGE enables conditioning on a desired attribute a (for example, formality) while requiring access only to G's output logits. FUDGE learns an attribute predictor operating on a partial sequence, and uses this predictor's outputs to adjust G's original probabilities. We show that FUDGE models terms corresponding to a Bayesian decomposition of the conditional distribution of G given attribute a. M"},"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":"2104.05218","kind":"arxiv","version":2},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CL","submitted_at":"2021-04-12T05:59:53Z","cross_cats_sorted":["cs.LG"],"title_canon_sha256":"6afb384b30a26faa247152bf53a0b57517b0f803317080b30b4d27453118e9ee","abstract_canon_sha256":"6076034088d7f22b4b14cb943bacbdb11bb02f46211d30f9305c44f278d1a733"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T03:05:49.109465Z","signature_b64":"qk8BRhzz+rS8IMH42Mxmfj7O0xXyfkpFKILA6N3r2yfJadXqi7AJx1sZmEdFWptkKzCChsW5nOJZ9KyJVJ2lDQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"761d93de555d440fae513ba143c384c6ab3143849be5780528594fbc5a5abe01","last_reissued_at":"2026-07-05T03:05:49.108964Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T03:05:49.108964Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"FUDGE: Controlled Text Generation With Future Discriminators","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.LG"],"primary_cat":"cs.CL","authors_text":"Dan Klein, Kevin Yang","submitted_at":"2021-04-12T05:59:53Z","abstract_excerpt":"We propose Future Discriminators for Generation (FUDGE), a flexible and modular method for controlled text generation. Given a pre-existing model G for generating text from a distribution of interest, FUDGE enables conditioning on a desired attribute a (for example, formality) while requiring access only to G's output logits. FUDGE learns an attribute predictor operating on a partial sequence, and uses this predictor's outputs to adjust G's original probabilities. We show that FUDGE models terms corresponding to a Bayesian decomposition of the conditional distribution of G given attribute a. M"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2104.05218","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/2104.05218/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":"2104.05218","created_at":"2026-07-05T03:05:49.109027+00:00"},{"alias_kind":"arxiv_version","alias_value":"2104.05218v2","created_at":"2026-07-05T03:05:49.109027+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2104.05218","created_at":"2026-07-05T03:05:49.109027+00:00"},{"alias_kind":"pith_short_12","alias_value":"OYOZHXSVLVCA","created_at":"2026-07-05T03:05:49.109027+00:00"},{"alias_kind":"pith_short_16","alias_value":"OYOZHXSVLVCA7LSR","created_at":"2026-07-05T03:05:49.109027+00:00"},{"alias_kind":"pith_short_8","alias_value":"OYOZHXSV","created_at":"2026-07-05T03:05:49.109027+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":5,"internal_anchor_count":1,"sample":[{"citing_arxiv_id":"2607.08071","citing_title":"COBART: Controlled, Optimized, Bidirectional and Auto-Regressive Transformer for Ad Headline Generation","ref_index":33,"is_internal_anchor":true},{"citing_arxiv_id":"2606.12234","citing_title":"On The Effectiveness-Fluency Trade-Off In LLM Conditioning: A Systematic Study","ref_index":66,"is_internal_anchor":false},{"citing_arxiv_id":"2408.07199","citing_title":"Agent Q: Advanced Reasoning and Learning for Autonomous AI Agents","ref_index":223,"is_internal_anchor":false},{"citing_arxiv_id":"2511.20657","citing_title":"Intelligent Agents with Emotional Intelligence: Current Trends, Challenges, and Future Prospects","ref_index":60,"is_internal_anchor":false},{"citing_arxiv_id":"2605.14004","citing_title":"Conditional Attribute Estimation with Autoregressive Sequence Models","ref_index":15,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/OYOZHXSVLVCA7LSRHOQUHQ4EY2","json":"https://pith.science/pith/OYOZHXSVLVCA7LSRHOQUHQ4EY2.json","graph_json":"https://pith.science/api/pith-number/OYOZHXSVLVCA7LSRHOQUHQ4EY2/graph.json","events_json":"https://pith.science/api/pith-number/OYOZHXSVLVCA7LSRHOQUHQ4EY2/events.json","paper":"https://pith.science/paper/OYOZHXSV"},"agent_actions":{"view_html":"https://pith.science/pith/OYOZHXSVLVCA7LSRHOQUHQ4EY2","download_json":"https://pith.science/pith/OYOZHXSVLVCA7LSRHOQUHQ4EY2.json","view_paper":"https://pith.science/paper/OYOZHXSV","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2104.05218&json=true","fetch_graph":"https://pith.science/api/pith-number/OYOZHXSVLVCA7LSRHOQUHQ4EY2/graph.json","fetch_events":"https://pith.science/api/pith-number/OYOZHXSVLVCA7LSRHOQUHQ4EY2/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/OYOZHXSVLVCA7LSRHOQUHQ4EY2/action/timestamp_anchor","attest_storage":"https://pith.science/pith/OYOZHXSVLVCA7LSRHOQUHQ4EY2/action/storage_attestation","attest_author":"https://pith.science/pith/OYOZHXSVLVCA7LSRHOQUHQ4EY2/action/author_attestation","sign_citation":"https://pith.science/pith/OYOZHXSVLVCA7LSRHOQUHQ4EY2/action/citation_signature","submit_replication":"https://pith.science/pith/OYOZHXSVLVCA7LSRHOQUHQ4EY2/action/replication_record"}},"created_at":"2026-07-05T03:05:49.109027+00:00","updated_at":"2026-07-05T03:05:49.109027+00:00"}