{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2021:3LJCKD4B4GIRCRBHNVFW4X4LAR","short_pith_number":"pith:3LJCKD4B","schema_version":"1.0","canonical_sha256":"dad2250f81e1911144276d4b6e5f8b0474f88df1bda3e3ae0912d9f58d7f58ed","source":{"kind":"arxiv","id":"2105.13626","version":3},"attestation_state":"computed","paper":{"title":"ByT5: Towards a token-free future with pre-trained byte-to-byte models","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Adam Roberts, Aditya Barua, Colin Raffel, Linting Xue, Mihir Kale, Noah Constant, Rami Al-Rfou, Sharan Narang","submitted_at":"2021-05-28T07:03:22Z","abstract_excerpt":"Most widely-used pre-trained language models operate on sequences of tokens corresponding to word or subword units. By comparison, token-free models that operate directly on raw text (bytes or characters) have many benefits: they can process text in any language out of the box, they are more robust to noise, and they minimize technical debt by removing complex and error-prone text preprocessing pipelines. Since byte or character sequences are longer than token sequences, past work on token-free models has often introduced new model architectures designed to amortize the cost of operating direc"},"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":"2105.13626","kind":"arxiv","version":3},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CL","submitted_at":"2021-05-28T07:03:22Z","cross_cats_sorted":[],"title_canon_sha256":"fa488a8c75a41ea0c2ebcdcee8bbb47ec79066d2e89acad0b9b8c6077725ffa6","abstract_canon_sha256":"16fa2b7cc537399de5a921ee5444ab2cdefd2d3b5cd7a8f43026267b1acfd47f"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T04:02:38.893430Z","signature_b64":"dAt5NdmBNAyAZkEwkKQmC2LkabG123iOXm7GfjGuY86x0HPEv2irAnuWCY+wVDxhOIuYn2mmeurSQ192zCN6CQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"dad2250f81e1911144276d4b6e5f8b0474f88df1bda3e3ae0912d9f58d7f58ed","last_reissued_at":"2026-07-05T04:02:38.892936Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T04:02:38.892936Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"ByT5: Towards a token-free future with pre-trained byte-to-byte models","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Adam Roberts, Aditya Barua, Colin Raffel, Linting Xue, Mihir Kale, Noah Constant, Rami Al-Rfou, Sharan Narang","submitted_at":"2021-05-28T07:03:22Z","abstract_excerpt":"Most widely-used pre-trained language models operate on sequences of tokens corresponding to word or subword units. By comparison, token-free models that operate directly on raw text (bytes or characters) have many benefits: they can process text in any language out of the box, they are more robust to noise, and they minimize technical debt by removing complex and error-prone text preprocessing pipelines. Since byte or character sequences are longer than token sequences, past work on token-free models has often introduced new model architectures designed to amortize the cost of operating direc"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2105.13626","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/2105.13626/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":"2105.13626","created_at":"2026-07-05T04:02:38.892994+00:00"},{"alias_kind":"arxiv_version","alias_value":"2105.13626v3","created_at":"2026-07-05T04:02:38.892994+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2105.13626","created_at":"2026-07-05T04:02:38.892994+00:00"},{"alias_kind":"pith_short_12","alias_value":"3LJCKD4B4GIR","created_at":"2026-07-05T04:02:38.892994+00:00"},{"alias_kind":"pith_short_16","alias_value":"3LJCKD4B4GIRCRBH","created_at":"2026-07-05T04:02:38.892994+00:00"},{"alias_kind":"pith_short_8","alias_value":"3LJCKD4B","created_at":"2026-07-05T04:02:38.892994+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":4,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2607.00664","citing_title":"YOMI-Bench: A Benchmark for Evaluating Kanji Reading and Phonological Understanding of LLMs for Japanese","ref_index":27,"is_internal_anchor":false},{"citing_arxiv_id":"2606.04171","citing_title":"MimeLens: Position-Agnostic Content-Type Detection for Binary Fragments","ref_index":12,"is_internal_anchor":false},{"citing_arxiv_id":"2605.24718","citing_title":"The Tokenizer Tax Across 25 European Languages: Domain Invariance, Cross-Lingual Few-Shot Effects, and the Ukrainian Penalty","ref_index":17,"is_internal_anchor":false},{"citing_arxiv_id":"2204.02311","citing_title":"PaLM: Scaling Language Modeling with Pathways","ref_index":168,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/3LJCKD4B4GIRCRBHNVFW4X4LAR","json":"https://pith.science/pith/3LJCKD4B4GIRCRBHNVFW4X4LAR.json","graph_json":"https://pith.science/api/pith-number/3LJCKD4B4GIRCRBHNVFW4X4LAR/graph.json","events_json":"https://pith.science/api/pith-number/3LJCKD4B4GIRCRBHNVFW4X4LAR/events.json","paper":"https://pith.science/paper/3LJCKD4B"},"agent_actions":{"view_html":"https://pith.science/pith/3LJCKD4B4GIRCRBHNVFW4X4LAR","download_json":"https://pith.science/pith/3LJCKD4B4GIRCRBHNVFW4X4LAR.json","view_paper":"https://pith.science/paper/3LJCKD4B","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2105.13626&json=true","fetch_graph":"https://pith.science/api/pith-number/3LJCKD4B4GIRCRBHNVFW4X4LAR/graph.json","fetch_events":"https://pith.science/api/pith-number/3LJCKD4B4GIRCRBHNVFW4X4LAR/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/3LJCKD4B4GIRCRBHNVFW4X4LAR/action/timestamp_anchor","attest_storage":"https://pith.science/pith/3LJCKD4B4GIRCRBHNVFW4X4LAR/action/storage_attestation","attest_author":"https://pith.science/pith/3LJCKD4B4GIRCRBHNVFW4X4LAR/action/author_attestation","sign_citation":"https://pith.science/pith/3LJCKD4B4GIRCRBHNVFW4X4LAR/action/citation_signature","submit_replication":"https://pith.science/pith/3LJCKD4B4GIRCRBHNVFW4X4LAR/action/replication_record"}},"created_at":"2026-07-05T04:02:38.892994+00:00","updated_at":"2026-07-05T04:02:38.892994+00:00"}