BWhat changed
Constructed modular BPE/Unigram tokenizers with extractable language-specific subtokenizers and trained language models using sampled vocabulary subsets to support smaller inference-time output matrices.
Scientific scope
Constructed modular BPE/Unigram tokenizers with extractable language-specific subtokenizers and trained language models using sampled vocabulary subsets to support smaller inference-time output matrices. Inherits BPE/Unigram and vocabulary adaptation. Compression parity and efficiency are empirical for studied languages/tasks; selecting a language subset does not ensure all cross-language capabilities or eliminate training costs.
Evidence
The linked original sources describe the stated contribution and its evidence limits. WRS has not independently rerun the experiments or certified a proof. Inherits BPE/Unigram and vocabulary adaptation. Compression parity and efficiency are empirical for studied languages/tasks; selecting a language subset does not ensure all cross-language capabilities or eliminate training costs.
Contributors
- Source-listed authors and technical contributors · SCIENCE: Source-listed authors or explicitly credited technical contributors for the cited versions. Family entries retain source-specific attribution; a combined list is not one original-paper byline or equal individual work. Names alone do not establish resolved identities.
- To Each Language Its Tokenizer: Modular Tokenizers for Efficient Multilingual LLMs
- Kyutai scientific publications index
Provisional WRS assessment, edition 2026-10-16.1. Independent expert calibration is pending. This grade does not certify correctness or assign ownership.
Full scientific record · Browse contribution grades · Request a correction