{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:24N2IJF6AHURMBX33MFFCBX5QJ","short_pith_number":"pith:24N2IJF6","schema_version":"1.0","canonical_sha256":"d71ba424be01e91606fbdb0a5106fd82704af6a739cd71a646f293a729911e7b","source":{"kind":"arxiv","id":"2603.22216","version":2},"attestation_state":"computed","paper":{"title":"Gumbel Distillation for Parallel Text Generation","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.LG"],"primary_cat":"cs.CL","authors_text":"Bo Liu, Chi Zhang, Qiang Liu, Xixi Hu","submitted_at":"2026-03-23T17:13:22Z","abstract_excerpt":"The slow, sequential nature of autoregressive (AR) language models has driven the adoption of parallel decoding methods. However, these non-AR models often sacrifice generation quality as they struggle to model the complex joint distribution of token sequences. To narrow this performance gap, we introduce Gumbel Distillation, a novel distillation technique that enables parallel decoders to learn this distribution effectively. Our method leverages the Gumbel-Max trick to create a deterministic mapping from a latent Gumbel noise space to the output tokens of a high-performing AR teacher. As a mo"},"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":"2603.22216","kind":"arxiv","version":2},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CL","submitted_at":"2026-03-23T17:13:22Z","cross_cats_sorted":["cs.LG"],"title_canon_sha256":"8a952f58ce043a36f795c3bc05fdd7dd2928978776fbb32a698669cf187a0cd7","abstract_canon_sha256":"276336514da028ff2805fc1605b0c2e908328d4f51269a6439b9ba777cca1997"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-24T01:24:09.977143Z","signature_b64":"jmJSP6ghBUAokSHZqJN1FTj4aNr/d9jaZx+MNzV8SO6U+WcnsHJ2Up7RKxlFvm1JEA5SDQiEVUM6jcM3skVBCg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"d71ba424be01e91606fbdb0a5106fd82704af6a739cd71a646f293a729911e7b","last_reissued_at":"2026-07-24T01:24:09.975995Z","signature_status":"signed_v1","first_computed_at":"2026-07-24T01:24:09.975995Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Gumbel Distillation for Parallel Text Generation","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.LG"],"primary_cat":"cs.CL","authors_text":"Bo Liu, Chi Zhang, Qiang Liu, Xixi Hu","submitted_at":"2026-03-23T17:13:22Z","abstract_excerpt":"The slow, sequential nature of autoregressive (AR) language models has driven the adoption of parallel decoding methods. However, these non-AR models often sacrifice generation quality as they struggle to model the complex joint distribution of token sequences. To narrow this performance gap, we introduce Gumbel Distillation, a novel distillation technique that enables parallel decoders to learn this distribution effectively. Our method leverages the Gumbel-Max trick to create a deterministic mapping from a latent Gumbel noise space to the output tokens of a high-performing AR teacher. As a mo"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2603.22216","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/2603.22216/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":"2603.22216","created_at":"2026-07-24T01:24:09.976555+00:00"},{"alias_kind":"arxiv_version","alias_value":"2603.22216v2","created_at":"2026-07-24T01:24:09.976555+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2603.22216","created_at":"2026-07-24T01:24:09.976555+00:00"},{"alias_kind":"pith_short_12","alias_value":"24N2IJF6AHUR","created_at":"2026-07-24T01:24:09.976555+00:00"},{"alias_kind":"pith_short_16","alias_value":"24N2IJF6AHURMBX3","created_at":"2026-07-24T01:24:09.976555+00:00"},{"alias_kind":"pith_short_8","alias_value":"24N2IJF6","created_at":"2026-07-24T01:24:09.976555+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":1,"internal_anchor_count":1,"sample":[{"citing_arxiv_id":"2606.06031","citing_title":"NAVIRA: Decoupled Stochastic Remasking for Masked Diffusion Language Models","ref_index":14,"is_internal_anchor":true}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/24N2IJF6AHURMBX33MFFCBX5QJ","json":"https://pith.science/pith/24N2IJF6AHURMBX33MFFCBX5QJ.json","graph_json":"https://pith.science/api/pith-number/24N2IJF6AHURMBX33MFFCBX5QJ/graph.json","events_json":"https://pith.science/api/pith-number/24N2IJF6AHURMBX33MFFCBX5QJ/events.json","paper":"https://pith.science/paper/24N2IJF6"},"agent_actions":{"view_html":"https://pith.science/pith/24N2IJF6AHURMBX33MFFCBX5QJ","download_json":"https://pith.science/pith/24N2IJF6AHURMBX33MFFCBX5QJ.json","view_paper":"https://pith.science/paper/24N2IJF6","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2603.22216&json=true","fetch_graph":"https://pith.science/api/pith-number/24N2IJF6AHURMBX33MFFCBX5QJ/graph.json","fetch_events":"https://pith.science/api/pith-number/24N2IJF6AHURMBX33MFFCBX5QJ/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/24N2IJF6AHURMBX33MFFCBX5QJ/action/timestamp_anchor","attest_storage":"https://pith.science/pith/24N2IJF6AHURMBX33MFFCBX5QJ/action/storage_attestation","attest_author":"https://pith.science/pith/24N2IJF6AHURMBX33MFFCBX5QJ/action/author_attestation","sign_citation":"https://pith.science/pith/24N2IJF6AHURMBX33MFFCBX5QJ/action/citation_signature","submit_replication":"https://pith.science/pith/24N2IJF6AHURMBX33MFFCBX5QJ/action/replication_record"}},"created_at":"2026-07-24T01:24:09.976555+00:00","updated_at":"2026-07-24T01:24:09.976555+00:00"}