{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:A3HSD7RNO7OUBYY6QP6SW65AD3","short_pith_number":"pith:A3HSD7RN","schema_version":"1.0","canonical_sha256":"06cf21fe2d77dd40e31e83fd2b7ba01ed1152b4f0db83df720fbcd2c2914631f","source":{"kind":"arxiv","id":"2605.22821","version":1},"attestation_state":"computed","paper":{"title":"Tokenisation via Convex Relaxations","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.LG"],"primary_cat":"cs.CL","authors_text":"Craig W. Schmidt, Dennis Komm, Jan Tempus, Philip Whittington, Tiago Pimentel","submitted_at":"2026-05-21T17:59:56Z","abstract_excerpt":"Tokenisation is an integral part of the current NLP pipeline. Current tokenisation algorithms such as BPE and Unigram are greedy algorithms -- they make locally optimal decisions without considering the resulting vocabulary as a whole. We instead formulate tokeniser construction as a linear program and solve it using convex optimisation tools, yielding a new algorithm we call ConvexTok. We find ConvexTok consistently improves intrinsic tokenisation metrics and the bits-per-byte (BpB) achieved by language models; it also improves downstream task performance, but less consistently. Furthermore, "},"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":"2605.22821","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CL","submitted_at":"2026-05-21T17:59:56Z","cross_cats_sorted":["cs.LG"],"title_canon_sha256":"a6c484aac5b78ec4773ed021373eee170c199e2966d0a7338ebc647fb0ac9535","abstract_canon_sha256":"366d38082e892af7655de8fb2cef95d54279bdebaa44ab54ffff52af2c011a2f"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-05-22T02:04:56.345741Z","signature_b64":"1QT1X0+nTB7rfscnochacOMM8dzkWo8QB4Xa6csj5GhouNzvQNB5M/xVs+/nYeO23deN+uGHABgFS4KAQ8/yAg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"06cf21fe2d77dd40e31e83fd2b7ba01ed1152b4f0db83df720fbcd2c2914631f","last_reissued_at":"2026-05-22T02:04:56.345304Z","signature_status":"signed_v1","first_computed_at":"2026-05-22T02:04:56.345304Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Tokenisation via Convex Relaxations","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.LG"],"primary_cat":"cs.CL","authors_text":"Craig W. Schmidt, Dennis Komm, Jan Tempus, Philip Whittington, Tiago Pimentel","submitted_at":"2026-05-21T17:59:56Z","abstract_excerpt":"Tokenisation is an integral part of the current NLP pipeline. Current tokenisation algorithms such as BPE and Unigram are greedy algorithms -- they make locally optimal decisions without considering the resulting vocabulary as a whole. We instead formulate tokeniser construction as a linear program and solve it using convex optimisation tools, yielding a new algorithm we call ConvexTok. We find ConvexTok consistently improves intrinsic tokenisation metrics and the bits-per-byte (BpB) achieved by language models; it also improves downstream task performance, but less consistently. Furthermore, "},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2605.22821","kind":"arxiv","version":1},"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/2605.22821/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":"2605.22821","created_at":"2026-05-22T02:04:56.345375+00:00"},{"alias_kind":"arxiv_version","alias_value":"2605.22821v1","created_at":"2026-05-22T02:04:56.345375+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2605.22821","created_at":"2026-05-22T02:04:56.345375+00:00"},{"alias_kind":"pith_short_12","alias_value":"A3HSD7RNO7OU","created_at":"2026-05-22T02:04:56.345375+00:00"},{"alias_kind":"pith_short_16","alias_value":"A3HSD7RNO7OUBYY6","created_at":"2026-05-22T02:04:56.345375+00:00"},{"alias_kind":"pith_short_8","alias_value":"A3HSD7RN","created_at":"2026-05-22T02:04:56.345375+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":0,"internal_anchor_count":0,"sample":[]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/A3HSD7RNO7OUBYY6QP6SW65AD3","json":"https://pith.science/pith/A3HSD7RNO7OUBYY6QP6SW65AD3.json","graph_json":"https://pith.science/api/pith-number/A3HSD7RNO7OUBYY6QP6SW65AD3/graph.json","events_json":"https://pith.science/api/pith-number/A3HSD7RNO7OUBYY6QP6SW65AD3/events.json","paper":"https://pith.science/paper/A3HSD7RN"},"agent_actions":{"view_html":"https://pith.science/pith/A3HSD7RNO7OUBYY6QP6SW65AD3","download_json":"https://pith.science/pith/A3HSD7RNO7OUBYY6QP6SW65AD3.json","view_paper":"https://pith.science/paper/A3HSD7RN","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2605.22821&json=true","fetch_graph":"https://pith.science/api/pith-number/A3HSD7RNO7OUBYY6QP6SW65AD3/graph.json","fetch_events":"https://pith.science/api/pith-number/A3HSD7RNO7OUBYY6QP6SW65AD3/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/A3HSD7RNO7OUBYY6QP6SW65AD3/action/timestamp_anchor","attest_storage":"https://pith.science/pith/A3HSD7RNO7OUBYY6QP6SW65AD3/action/storage_attestation","attest_author":"https://pith.science/pith/A3HSD7RNO7OUBYY6QP6SW65AD3/action/author_attestation","sign_citation":"https://pith.science/pith/A3HSD7RNO7OUBYY6QP6SW65AD3/action/citation_signature","submit_replication":"https://pith.science/pith/A3HSD7RNO7OUBYY6QP6SW65AD3/action/replication_record"}},"created_at":"2026-05-22T02:04:56.345375+00:00","updated_at":"2026-05-22T02:04:56.345375+00:00"}