{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:NY6O7F6IMKZGTD6FTXQMWU57OO","short_pith_number":"pith:NY6O7F6I","schema_version":"1.0","canonical_sha256":"6e3cef97c862b2698fc59de0cb53bf73951ea9f0394ae3c934a0b0feb4d47e70","source":{"kind":"arxiv","id":"2506.06607","version":1},"attestation_state":"computed","paper":{"title":"Training-Free Tokenizer Transplantation via Orthogonal Matching Pursuit","license":"http://creativecommons.org/licenses/by-sa/4.0/","headline":"","cross_cats":["cs.AI","cs.LG"],"primary_cat":"cs.CL","authors_text":"Charles Goddard, Fernando Fernandes Neto","submitted_at":"2025-06-07T00:51:27Z","abstract_excerpt":"We present a training-free method to transplant tokenizers in pretrained large language models (LLMs) by reconstructing unseen token embeddings via Orthogonal Matching Pursuit (OMP). Specifically, we approximate each out-of-vocabulary token as a sparse linear combination of shared tokens, in two phases: first, compute each new token's representation in the donor embedding space with a small dictionary of shared anchor tokens, then transfer these same sparse coefficients back into the base model's embedding space.\n  On two challenging cross-tokenizer tasks--Llama$\\to$Mistral NeMo (12B) and Qwen"},"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":"2506.06607","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by-sa/4.0/","primary_cat":"cs.CL","submitted_at":"2025-06-07T00:51:27Z","cross_cats_sorted":["cs.AI","cs.LG"],"title_canon_sha256":"aadab502b8a8c27e062a71c28831e6f915216640a1df98138a796a4fd6394d0e","abstract_canon_sha256":"117135cf7642313eb18a2be271852b717844640cab1d0c590b5a2799a125e469"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T11:17:52.819127Z","signature_b64":"PVCE7sy4B2Jzk4lXAaKPx//SzTO/gDNehtRUFoareAFaaPcRBGjj+li8WQkczuUVgfzy6G9xQvm1xhc4nKDRDQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"6e3cef97c862b2698fc59de0cb53bf73951ea9f0394ae3c934a0b0feb4d47e70","last_reissued_at":"2026-07-05T11:17:52.818652Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T11:17:52.818652Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Training-Free Tokenizer Transplantation via Orthogonal Matching Pursuit","license":"http://creativecommons.org/licenses/by-sa/4.0/","headline":"","cross_cats":["cs.AI","cs.LG"],"primary_cat":"cs.CL","authors_text":"Charles Goddard, Fernando Fernandes Neto","submitted_at":"2025-06-07T00:51:27Z","abstract_excerpt":"We present a training-free method to transplant tokenizers in pretrained large language models (LLMs) by reconstructing unseen token embeddings via Orthogonal Matching Pursuit (OMP). Specifically, we approximate each out-of-vocabulary token as a sparse linear combination of shared tokens, in two phases: first, compute each new token's representation in the donor embedding space with a small dictionary of shared anchor tokens, then transfer these same sparse coefficients back into the base model's embedding space.\n  On two challenging cross-tokenizer tasks--Llama$\\to$Mistral NeMo (12B) and Qwen"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2506.06607","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/2506.06607/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":"2506.06607","created_at":"2026-07-05T11:17:52.818712+00:00"},{"alias_kind":"arxiv_version","alias_value":"2506.06607v1","created_at":"2026-07-05T11:17:52.818712+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2506.06607","created_at":"2026-07-05T11:17:52.818712+00:00"},{"alias_kind":"pith_short_12","alias_value":"NY6O7F6IMKZG","created_at":"2026-07-05T11:17:52.818712+00:00"},{"alias_kind":"pith_short_16","alias_value":"NY6O7F6IMKZGTD6F","created_at":"2026-07-05T11:17:52.818712+00:00"},{"alias_kind":"pith_short_8","alias_value":"NY6O7F6I","created_at":"2026-07-05T11:17:52.818712+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":1,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2605.23645","citing_title":"Learning Through Noise: Why Subliminal Learning Works and When It Fails","ref_index":25,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/NY6O7F6IMKZGTD6FTXQMWU57OO","json":"https://pith.science/pith/NY6O7F6IMKZGTD6FTXQMWU57OO.json","graph_json":"https://pith.science/api/pith-number/NY6O7F6IMKZGTD6FTXQMWU57OO/graph.json","events_json":"https://pith.science/api/pith-number/NY6O7F6IMKZGTD6FTXQMWU57OO/events.json","paper":"https://pith.science/paper/NY6O7F6I"},"agent_actions":{"view_html":"https://pith.science/pith/NY6O7F6IMKZGTD6FTXQMWU57OO","download_json":"https://pith.science/pith/NY6O7F6IMKZGTD6FTXQMWU57OO.json","view_paper":"https://pith.science/paper/NY6O7F6I","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2506.06607&json=true","fetch_graph":"https://pith.science/api/pith-number/NY6O7F6IMKZGTD6FTXQMWU57OO/graph.json","fetch_events":"https://pith.science/api/pith-number/NY6O7F6IMKZGTD6FTXQMWU57OO/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/NY6O7F6IMKZGTD6FTXQMWU57OO/action/timestamp_anchor","attest_storage":"https://pith.science/pith/NY6O7F6IMKZGTD6FTXQMWU57OO/action/storage_attestation","attest_author":"https://pith.science/pith/NY6O7F6IMKZGTD6FTXQMWU57OO/action/author_attestation","sign_citation":"https://pith.science/pith/NY6O7F6IMKZGTD6FTXQMWU57OO/action/citation_signature","submit_replication":"https://pith.science/pith/NY6O7F6IMKZGTD6FTXQMWU57OO/action/replication_record"}},"created_at":"2026-07-05T11:17:52.818712+00:00","updated_at":"2026-07-05T11:17:52.818712+00:00"}