{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2024:KKAXPDOX4BGQUTRPVH7LNCULK2","short_pith_number":"pith:KKAXPDOX","schema_version":"1.0","canonical_sha256":"5281778dd7e04d0a4e2fa9feb68a8b56b21dc31dc9536790b8a24c2819a50086","source":{"kind":"arxiv","id":"2410.20771","version":3},"attestation_state":"computed","paper":{"title":"MrT5: Dynamic Token Merging for Efficient Byte-level Language Models","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI","cs.LG"],"primary_cat":"cs.CL","authors_text":"Christopher D. Manning, Christopher Potts, Julie Kallini, R\\'obert Csord\\'as, Shikhar Murty","submitted_at":"2024-10-28T06:14:12Z","abstract_excerpt":"Models that rely on subword tokenization have significant drawbacks, such as sensitivity to character-level noise like spelling errors and inconsistent compression rates across different languages and scripts. While character- or byte-level models like ByT5 attempt to address these concerns, they have not gained widespread adoption -- processing raw byte streams without tokenization results in significantly longer sequence lengths, making training and inference inefficient. This work introduces MrT5 (MergeT5), a more efficient variant of ByT5 that integrates a token deletion mechanism in its e"},"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":"2410.20771","kind":"arxiv","version":3},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CL","submitted_at":"2024-10-28T06:14:12Z","cross_cats_sorted":["cs.AI","cs.LG"],"title_canon_sha256":"9b941942d4f1123c114fd80a14e73b4082c925eb9425c918c5c7ed430b9249d6","abstract_canon_sha256":"6eef96b530739d052ccce871588431e88a61208264e5291ce942730ed51c5058"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T10:43:02.815305Z","signature_b64":"AVpYSTR6xMkmYwXT7MPVD1mkAOt+9OGFrcZ408WBNa3X31KQIDc4S7cTy6zVC3jndp14oTl/qmHTrKoH7vJaDg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"5281778dd7e04d0a4e2fa9feb68a8b56b21dc31dc9536790b8a24c2819a50086","last_reissued_at":"2026-07-05T10:43:02.814804Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T10:43:02.814804Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"MrT5: Dynamic Token Merging for Efficient Byte-level Language Models","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI","cs.LG"],"primary_cat":"cs.CL","authors_text":"Christopher D. Manning, Christopher Potts, Julie Kallini, R\\'obert Csord\\'as, Shikhar Murty","submitted_at":"2024-10-28T06:14:12Z","abstract_excerpt":"Models that rely on subword tokenization have significant drawbacks, such as sensitivity to character-level noise like spelling errors and inconsistent compression rates across different languages and scripts. While character- or byte-level models like ByT5 attempt to address these concerns, they have not gained widespread adoption -- processing raw byte streams without tokenization results in significantly longer sequence lengths, making training and inference inefficient. This work introduces MrT5 (MergeT5), a more efficient variant of ByT5 that integrates a token deletion mechanism in its e"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2410.20771","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/2410.20771/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":"2410.20771","created_at":"2026-07-05T10:43:02.814862+00:00"},{"alias_kind":"arxiv_version","alias_value":"2410.20771v3","created_at":"2026-07-05T10:43:02.814862+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2410.20771","created_at":"2026-07-05T10:43:02.814862+00:00"},{"alias_kind":"pith_short_12","alias_value":"KKAXPDOX4BGQ","created_at":"2026-07-05T10:43:02.814862+00:00"},{"alias_kind":"pith_short_16","alias_value":"KKAXPDOX4BGQUTRP","created_at":"2026-07-05T10:43:02.814862+00:00"},{"alias_kind":"pith_short_8","alias_value":"KKAXPDOX","created_at":"2026-07-05T10:43:02.814862+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":3,"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":26,"is_internal_anchor":false},{"citing_arxiv_id":"2510.18091","citing_title":"Accelerating Vision Transformers with Adaptive Patch Sizes","ref_index":9,"is_internal_anchor":false},{"citing_arxiv_id":"2605.09630","citing_title":"Scratchpad Patching: Decoupling Compute from Patch Size in Byte-Level Language Models","ref_index":51,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/KKAXPDOX4BGQUTRPVH7LNCULK2","json":"https://pith.science/pith/KKAXPDOX4BGQUTRPVH7LNCULK2.json","graph_json":"https://pith.science/api/pith-number/KKAXPDOX4BGQUTRPVH7LNCULK2/graph.json","events_json":"https://pith.science/api/pith-number/KKAXPDOX4BGQUTRPVH7LNCULK2/events.json","paper":"https://pith.science/paper/KKAXPDOX"},"agent_actions":{"view_html":"https://pith.science/pith/KKAXPDOX4BGQUTRPVH7LNCULK2","download_json":"https://pith.science/pith/KKAXPDOX4BGQUTRPVH7LNCULK2.json","view_paper":"https://pith.science/paper/KKAXPDOX","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2410.20771&json=true","fetch_graph":"https://pith.science/api/pith-number/KKAXPDOX4BGQUTRPVH7LNCULK2/graph.json","fetch_events":"https://pith.science/api/pith-number/KKAXPDOX4BGQUTRPVH7LNCULK2/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/KKAXPDOX4BGQUTRPVH7LNCULK2/action/timestamp_anchor","attest_storage":"https://pith.science/pith/KKAXPDOX4BGQUTRPVH7LNCULK2/action/storage_attestation","attest_author":"https://pith.science/pith/KKAXPDOX4BGQUTRPVH7LNCULK2/action/author_attestation","sign_citation":"https://pith.science/pith/KKAXPDOX4BGQUTRPVH7LNCULK2/action/citation_signature","submit_replication":"https://pith.science/pith/KKAXPDOX4BGQUTRPVH7LNCULK2/action/replication_record"}},"created_at":"2026-07-05T10:43:02.814862+00:00","updated_at":"2026-07-05T10:43:02.814862+00:00"}