{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2024:JX5VEIJ3QA3PDVHULC6V7WBOG6","short_pith_number":"pith:JX5VEIJ3","schema_version":"1.0","canonical_sha256":"4dfb52213b8036f1d4f458bd5fd82e379ef326e24e7b04b5b9c018f19e7763c8","source":{"kind":"arxiv","id":"2404.17808","version":3},"attestation_state":"computed","paper":{"title":"Scaffold-BPE: Enhancing Byte Pair Encoding for Large Language Models with Simple and Effective Scaffold Token Removal","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Guiguang Ding, Haoran Lian, Hui Chen, Jianwei Niu, Jungong Han, Peng Liu, Shasha Mo, Yizhe Xiong, Zhenpeng Su, Zijia Lin","submitted_at":"2024-04-27T07:12:07Z","abstract_excerpt":"Byte Pair Encoding (BPE) serves as a foundation method for text tokenization in the Natural Language Processing (NLP) field. Despite its wide adoption, the original BPE algorithm harbors an inherent flaw: it inadvertently introduces a frequency imbalance for tokens in the text corpus. Since BPE iteratively merges the most frequent token pair in the text corpus to generate a new token and keeps all generated tokens in the vocabulary, it unavoidably holds tokens that primarily act as components of a longer token and appear infrequently on their own. We term such tokens as Scaffold Tokens. Due to"},"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":"2404.17808","kind":"arxiv","version":3},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CL","submitted_at":"2024-04-27T07:12:07Z","cross_cats_sorted":[],"title_canon_sha256":"cd73e61fa79247cca2f3ac4035de58066941e8e04278dfa02397796c04e2ae11","abstract_canon_sha256":"e0553b9f1611f5d10e2888cc2c149975b03b0309c9c3404deb9ffbd51c74f945"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T09:34:35.647431Z","signature_b64":"c9bN1aQ6VzHKjC84xnjApIwBvAl1L78F4N8PZM7cwsLn0h+u6kzDLdpP5vkPiNW5A5ObBJbhLuQeUc0bEzxdAA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"4dfb52213b8036f1d4f458bd5fd82e379ef326e24e7b04b5b9c018f19e7763c8","last_reissued_at":"2026-07-05T09:34:35.646915Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T09:34:35.646915Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Scaffold-BPE: Enhancing Byte Pair Encoding for Large Language Models with Simple and Effective Scaffold Token Removal","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Guiguang Ding, Haoran Lian, Hui Chen, Jianwei Niu, Jungong Han, Peng Liu, Shasha Mo, Yizhe Xiong, Zhenpeng Su, Zijia Lin","submitted_at":"2024-04-27T07:12:07Z","abstract_excerpt":"Byte Pair Encoding (BPE) serves as a foundation method for text tokenization in the Natural Language Processing (NLP) field. Despite its wide adoption, the original BPE algorithm harbors an inherent flaw: it inadvertently introduces a frequency imbalance for tokens in the text corpus. Since BPE iteratively merges the most frequent token pair in the text corpus to generate a new token and keeps all generated tokens in the vocabulary, it unavoidably holds tokens that primarily act as components of a longer token and appear infrequently on their own. We term such tokens as Scaffold Tokens. Due to"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2404.17808","kind":"arxiv","version":3},"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/2404.17808/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":"2404.17808","created_at":"2026-07-05T09:34:35.646974+00:00"},{"alias_kind":"arxiv_version","alias_value":"2404.17808v3","created_at":"2026-07-05T09:34:35.646974+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2404.17808","created_at":"2026-07-05T09:34:35.646974+00:00"},{"alias_kind":"pith_short_12","alias_value":"JX5VEIJ3QA3P","created_at":"2026-07-05T09:34:35.646974+00:00"},{"alias_kind":"pith_short_16","alias_value":"JX5VEIJ3QA3PDVHU","created_at":"2026-07-05T09:34:35.646974+00:00"},{"alias_kind":"pith_short_8","alias_value":"JX5VEIJ3","created_at":"2026-07-05T09:34:35.646974+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":1,"internal_anchor_count":1,"sample":[{"citing_arxiv_id":"2607.05691","citing_title":"Where to cut, how deep: BPE and Unigram-LM on chemistry SMILES","ref_index":28,"is_internal_anchor":true}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/JX5VEIJ3QA3PDVHULC6V7WBOG6","json":"https://pith.science/pith/JX5VEIJ3QA3PDVHULC6V7WBOG6.json","graph_json":"https://pith.science/api/pith-number/JX5VEIJ3QA3PDVHULC6V7WBOG6/graph.json","events_json":"https://pith.science/api/pith-number/JX5VEIJ3QA3PDVHULC6V7WBOG6/events.json","paper":"https://pith.science/paper/JX5VEIJ3"},"agent_actions":{"view_html":"https://pith.science/pith/JX5VEIJ3QA3PDVHULC6V7WBOG6","download_json":"https://pith.science/pith/JX5VEIJ3QA3PDVHULC6V7WBOG6.json","view_paper":"https://pith.science/paper/JX5VEIJ3","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2404.17808&json=true","fetch_graph":"https://pith.science/api/pith-number/JX5VEIJ3QA3PDVHULC6V7WBOG6/graph.json","fetch_events":"https://pith.science/api/pith-number/JX5VEIJ3QA3PDVHULC6V7WBOG6/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/JX5VEIJ3QA3PDVHULC6V7WBOG6/action/timestamp_anchor","attest_storage":"https://pith.science/pith/JX5VEIJ3QA3PDVHULC6V7WBOG6/action/storage_attestation","attest_author":"https://pith.science/pith/JX5VEIJ3QA3PDVHULC6V7WBOG6/action/author_attestation","sign_citation":"https://pith.science/pith/JX5VEIJ3QA3PDVHULC6V7WBOG6/action/citation_signature","submit_replication":"https://pith.science/pith/JX5VEIJ3QA3PDVHULC6V7WBOG6/action/replication_record"}},"created_at":"2026-07-05T09:34:35.646974+00:00","updated_at":"2026-07-05T09:34:35.646974+00:00"}