{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:2IQ5UQIMDFEXJSD4TYVOJ5GLYB","short_pith_number":"pith:2IQ5UQIM","schema_version":"1.0","canonical_sha256":"d221da410c194974c87c9e2ae4f4cbc04ab268d567f3e37bd34c5b620ff3cd55","source":{"kind":"arxiv","id":"2501.16975","version":2},"attestation_state":"computed","paper":{"title":"Over-Tokenized Transformer: Vocabulary is Generally Worth Scaling","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.LG"],"primary_cat":"cs.CL","authors_text":"Banggu Wu, Defa Zhu, Hongzhi Huang, Qiyang Min, Xun Zhou, Ya Wang, Yutao Zeng","submitted_at":"2025-01-28T14:15:42Z","abstract_excerpt":"Tokenization is a fundamental component of large language models (LLMs), yet its influence on model scaling and performance is not fully explored. In this paper, we introduce Over-Tokenized Transformers, a novel framework that decouples input and output vocabularies to improve language modeling performance. Specifically, our approach scales up input vocabularies to leverage multi-gram tokens. Through extensive experiments, we uncover a log-linear relationship between input vocabulary size and training loss, demonstrating that larger input vocabularies consistently enhance model performance, re"},"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":"2501.16975","kind":"arxiv","version":2},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CL","submitted_at":"2025-01-28T14:15:42Z","cross_cats_sorted":["cs.LG"],"title_canon_sha256":"46d6735f20c6364f362f232019f3f29da8b304526c7d88a9b2828453f20f0836","abstract_canon_sha256":"d3dba02bab10482f2d5eade21189d5bc76996629961b075cd29075bcdcf78c89"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T11:08:15.560359Z","signature_b64":"wVNLtyDxJ1dn8eKmekaX/Soh10T31EguFyy/spQmL0mAy0Rke5guaVr1GnRXijBl6FEn0nYz6ahzMfTECLCKBw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"d221da410c194974c87c9e2ae4f4cbc04ab268d567f3e37bd34c5b620ff3cd55","last_reissued_at":"2026-07-05T11:08:15.559875Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T11:08:15.559875Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Over-Tokenized Transformer: Vocabulary is Generally Worth Scaling","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.LG"],"primary_cat":"cs.CL","authors_text":"Banggu Wu, Defa Zhu, Hongzhi Huang, Qiyang Min, Xun Zhou, Ya Wang, Yutao Zeng","submitted_at":"2025-01-28T14:15:42Z","abstract_excerpt":"Tokenization is a fundamental component of large language models (LLMs), yet its influence on model scaling and performance is not fully explored. In this paper, we introduce Over-Tokenized Transformers, a novel framework that decouples input and output vocabularies to improve language modeling performance. Specifically, our approach scales up input vocabularies to leverage multi-gram tokens. Through extensive experiments, we uncover a log-linear relationship between input vocabulary size and training loss, demonstrating that larger input vocabularies consistently enhance model performance, re"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2501.16975","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/2501.16975/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":"2501.16975","created_at":"2026-07-05T11:08:15.559941+00:00"},{"alias_kind":"arxiv_version","alias_value":"2501.16975v2","created_at":"2026-07-05T11:08:15.559941+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2501.16975","created_at":"2026-07-05T11:08:15.559941+00:00"},{"alias_kind":"pith_short_12","alias_value":"2IQ5UQIMDFEX","created_at":"2026-07-05T11:08:15.559941+00:00"},{"alias_kind":"pith_short_16","alias_value":"2IQ5UQIMDFEXJSD4","created_at":"2026-07-05T11:08:15.559941+00:00"},{"alias_kind":"pith_short_8","alias_value":"2IQ5UQIM","created_at":"2026-07-05T11:08:15.559941+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":3,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2605.10780","citing_title":"Beyond the Last Layer: Multi-Layer Representation Fusion for Visual Tokenization","ref_index":12,"is_internal_anchor":false},{"citing_arxiv_id":"2605.10780","citing_title":"Beyond the Last Layer: Multi-Layer Representation Fusion for Visual Tokenization","ref_index":12,"is_internal_anchor":false},{"citing_arxiv_id":"2604.21724","citing_title":"Beyond N-gram: Data-Aware X-GRAM Extraction for Efficient Embedding Parameter Scaling","ref_index":8,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/2IQ5UQIMDFEXJSD4TYVOJ5GLYB","json":"https://pith.science/pith/2IQ5UQIMDFEXJSD4TYVOJ5GLYB.json","graph_json":"https://pith.science/api/pith-number/2IQ5UQIMDFEXJSD4TYVOJ5GLYB/graph.json","events_json":"https://pith.science/api/pith-number/2IQ5UQIMDFEXJSD4TYVOJ5GLYB/events.json","paper":"https://pith.science/paper/2IQ5UQIM"},"agent_actions":{"view_html":"https://pith.science/pith/2IQ5UQIMDFEXJSD4TYVOJ5GLYB","download_json":"https://pith.science/pith/2IQ5UQIMDFEXJSD4TYVOJ5GLYB.json","view_paper":"https://pith.science/paper/2IQ5UQIM","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2501.16975&json=true","fetch_graph":"https://pith.science/api/pith-number/2IQ5UQIMDFEXJSD4TYVOJ5GLYB/graph.json","fetch_events":"https://pith.science/api/pith-number/2IQ5UQIMDFEXJSD4TYVOJ5GLYB/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/2IQ5UQIMDFEXJSD4TYVOJ5GLYB/action/timestamp_anchor","attest_storage":"https://pith.science/pith/2IQ5UQIMDFEXJSD4TYVOJ5GLYB/action/storage_attestation","attest_author":"https://pith.science/pith/2IQ5UQIMDFEXJSD4TYVOJ5GLYB/action/author_attestation","sign_citation":"https://pith.science/pith/2IQ5UQIMDFEXJSD4TYVOJ5GLYB/action/citation_signature","submit_replication":"https://pith.science/pith/2IQ5UQIMDFEXJSD4TYVOJ5GLYB/action/replication_record"}},"created_at":"2026-07-05T11:08:15.559941+00:00","updated_at":"2026-07-05T11:08:15.559941+00:00"}