{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2021:CUPLFF23ZVGZOT5DJSD7BBDXPJ","short_pith_number":"pith:CUPLFF23","schema_version":"1.0","canonical_sha256":"151eb2975bcd4d974fa34c87f084777a4792d4eb615b284644a89835480c01a9","source":{"kind":"arxiv","id":"2106.12672","version":3},"attestation_state":"computed","paper":{"title":"Charformer: Fast Character Transformers via Gradient-based Subword Tokenization","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.AI","cs.LG"],"primary_cat":"cs.CL","authors_text":"Cong Yu, Dara Bahri, Donald Metzler, Hyung Won Chung, Jai Gupta, Sebastian Ruder, Simon Baumgartner, Vinh Q. Tran, Yi Tay, Zhen Qin","submitted_at":"2021-06-23T22:24:14Z","abstract_excerpt":"State-of-the-art models in natural language processing rely on separate rigid subword tokenization algorithms, which limit their generalization ability and adaptation to new settings. In this paper, we propose a new model inductive bias that learns a subword tokenization end-to-end as part of the model. To this end, we introduce a soft gradient-based subword tokenization module (GBST) that automatically learns latent subword representations from characters in a data-driven fashion. Concretely, GBST enumerates candidate subword blocks and learns to score them in a position-wise fashion using a "},"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":"2106.12672","kind":"arxiv","version":3},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CL","submitted_at":"2021-06-23T22:24:14Z","cross_cats_sorted":["cs.AI","cs.LG"],"title_canon_sha256":"a4bfd605e76046f0caf34469f1cc5d9c3adb1fa5f3d850b5423bfb1d6a885587","abstract_canon_sha256":"dc4c3f8adeb5c1962ab13b367f84f0d48dcc32aad69f0bc01c0291d2a1c4419f"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T03:59:22.660939Z","signature_b64":"MHtUorNzOegY7jnhrHYPapS9FMC0Rc5hdq3H0rjWchnia4MInwiquFDoVyyqKtSJtvjeum/n7PjM4YeAv8HmAw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"151eb2975bcd4d974fa34c87f084777a4792d4eb615b284644a89835480c01a9","last_reissued_at":"2026-07-05T03:59:22.660415Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T03:59:22.660415Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Charformer: Fast Character Transformers via Gradient-based Subword Tokenization","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.AI","cs.LG"],"primary_cat":"cs.CL","authors_text":"Cong Yu, Dara Bahri, Donald Metzler, Hyung Won Chung, Jai Gupta, Sebastian Ruder, Simon Baumgartner, Vinh Q. Tran, Yi Tay, Zhen Qin","submitted_at":"2021-06-23T22:24:14Z","abstract_excerpt":"State-of-the-art models in natural language processing rely on separate rigid subword tokenization algorithms, which limit their generalization ability and adaptation to new settings. In this paper, we propose a new model inductive bias that learns a subword tokenization end-to-end as part of the model. To this end, we introduce a soft gradient-based subword tokenization module (GBST) that automatically learns latent subword representations from characters in a data-driven fashion. Concretely, GBST enumerates candidate subword blocks and learns to score them in a position-wise fashion using a "},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2106.12672","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/2106.12672/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":"2106.12672","created_at":"2026-07-05T03:59:22.660479+00:00"},{"alias_kind":"arxiv_version","alias_value":"2106.12672v3","created_at":"2026-07-05T03:59:22.660479+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2106.12672","created_at":"2026-07-05T03:59:22.660479+00:00"},{"alias_kind":"pith_short_12","alias_value":"CUPLFF23ZVGZ","created_at":"2026-07-05T03:59:22.660479+00:00"},{"alias_kind":"pith_short_16","alias_value":"CUPLFF23ZVGZOT5D","created_at":"2026-07-05T03:59:22.660479+00:00"},{"alias_kind":"pith_short_8","alias_value":"CUPLFF23","created_at":"2026-07-05T03:59:22.660479+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":5,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.20993","citing_title":"Phonemes to the Rescue: Multilingual Tokenization Based on International Phonetic Alphabet","ref_index":63,"is_internal_anchor":false},{"citing_arxiv_id":"2605.29459","citing_title":"Kronecker Embeddings: Byte-Level Structured Token Representations for Parameter-Efficient Language Models","ref_index":11,"is_internal_anchor":false},{"citing_arxiv_id":"2507.12720","citing_title":"FLEXITOKENS: Flexible Tokenization for Evolving Language Models","ref_index":15,"is_internal_anchor":false},{"citing_arxiv_id":"2302.09664","citing_title":"Semantic Uncertainty: Linguistic Invariances for Uncertainty Estimation in Natural Language Generation","ref_index":17,"is_internal_anchor":false},{"citing_arxiv_id":"2604.24940","citing_title":"ADE: Adaptive Dictionary Embeddings -- Scaling Multi-Anchor Representations to Large Language Models","ref_index":15,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/CUPLFF23ZVGZOT5DJSD7BBDXPJ","json":"https://pith.science/pith/CUPLFF23ZVGZOT5DJSD7BBDXPJ.json","graph_json":"https://pith.science/api/pith-number/CUPLFF23ZVGZOT5DJSD7BBDXPJ/graph.json","events_json":"https://pith.science/api/pith-number/CUPLFF23ZVGZOT5DJSD7BBDXPJ/events.json","paper":"https://pith.science/paper/CUPLFF23"},"agent_actions":{"view_html":"https://pith.science/pith/CUPLFF23ZVGZOT5DJSD7BBDXPJ","download_json":"https://pith.science/pith/CUPLFF23ZVGZOT5DJSD7BBDXPJ.json","view_paper":"https://pith.science/paper/CUPLFF23","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2106.12672&json=true","fetch_graph":"https://pith.science/api/pith-number/CUPLFF23ZVGZOT5DJSD7BBDXPJ/graph.json","fetch_events":"https://pith.science/api/pith-number/CUPLFF23ZVGZOT5DJSD7BBDXPJ/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/CUPLFF23ZVGZOT5DJSD7BBDXPJ/action/timestamp_anchor","attest_storage":"https://pith.science/pith/CUPLFF23ZVGZOT5DJSD7BBDXPJ/action/storage_attestation","attest_author":"https://pith.science/pith/CUPLFF23ZVGZOT5DJSD7BBDXPJ/action/author_attestation","sign_citation":"https://pith.science/pith/CUPLFF23ZVGZOT5DJSD7BBDXPJ/action/citation_signature","submit_replication":"https://pith.science/pith/CUPLFF23ZVGZOT5DJSD7BBDXPJ/action/replication_record"}},"created_at":"2026-07-05T03:59:22.660479+00:00","updated_at":"2026-07-05T03:59:22.660479+00:00"}