To Each Language Its Tokenizer: Modular Tokenizers for Efficient Multilingual LLMs

• URL: http://arxiv.org/abs/2609.15528v1
• Date: Mon, 14 Sep 2026 13:17:35 GMT
• Title: To Each Language Its Tokenizer: Modular Tokenizers for Efficient Multilingual LLMs
• Abstract summary: We introduce a modular framework for multilingual model training.<n>First, we propose methods to learn large modular BPE and Unigram tokenizers that enable extraction of subtokenizers tailored to any language subset.<n>Second, we design a pretraining strategy that samples subtokenizers to form batches, restricting predictions to the relevant vocabulary subset and allowing efficient training despite a large vocabulary.
• Score: 15.23810104303108
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: Multilingual Large Language Models (LLMs) traditionally rely on a single vocabulary shared by all supported languages, which can lead to uneven compression across them. Moreover, their large embedding and output matrices increase memory usage and slow inference, notably for small-scale models. It is also wasteful as models are often used for only a subset of languages. To address these issues, we introduce a modular framework for multilingual model training. First, we propose methods to learn large modular BPE and Unigram tokenizers that enable extraction of subtokenizers tailored to any language subset. These subtokenizers achieve compression on par with monolingual tokenizers and improve cross-lingual fairness. Second, we design a pretraining strategy that samples subtokenizers to form batches, restricting predictions to the relevant vocabulary subset and allowing efficient training despite a large vocabulary. This supports efficient inference with any combination of language-specific vocabularies. Therefore, it reduces memory usage and speeds up inference in models without sacrificing performance.

Related papers

• The Ups and Downs of Large Language Model Inference with Vocabulary Trimming by Language Heuristics [74.99898531299148]
  This research examines vocabulary trimming (VT) inspired by restricting embedding entries to the language of interest to bolster time and memory efficiency. We apply two languages to trim the full vocabulary - Unicode-based script filtering and corpus-based selection - to different language families and sizes. It is found that VT reduces the memory usage of small models by nearly 50% and has an upper bound of 25% improvement in generation speed.
  arXiv  Detail & Related papers  (2023-11-16T09:35:50Z)
• Tik-to-Tok: Translating Language Models One Token at a Time: An Embedding Initialization Strategy for Efficient Language Adaptation [19.624330093598996]
  Training monolingual language models for low and mid-resource languages is made challenging by limited and often inadequate pretraining data. By generalizing over a word translation dictionary encompassing both the source and target languages, we map tokens from the target tokenizer to semantically similar tokens from the source language tokenizer. We conduct experiments to convert high-resource models to mid- and low-resource languages, namely Dutch and Frisian.
  arXiv  Detail & Related papers  (2023-10-05T11:45:29Z)
• Embedding structure matters: Comparing methods to adapt multilingual vocabularies to new languages [20.17308477850864]
  Pre-trained multilingual language models underpin a large portion of modern NLP tools outside of English. We propose several simple techniques to replace a cross-lingual vocabulary with a compact, language-specific one.
  arXiv  Detail & Related papers  (2023-09-09T04:27:18Z)
• Soft Language Clustering for Multilingual Model Pre-training [57.18058739931463]
  We propose XLM-P, which contextually retrieves prompts as flexible guidance for encoding instances conditionally. Our XLM-P enables (1) lightweight modeling of language-invariant and language-specific knowledge across languages, and (2) easy integration with other multilingual pre-training methods.
  arXiv  Detail & Related papers  (2023-06-13T08:08:08Z)
• Lifting the Curse of Multilinguality by Pre-training Modular Transformers [72.46919537293068]
  multilingual pre-trained models suffer from the curse of multilinguality, which causes per-language performance to drop as they cover more languages. We introduce language-specific modules, which allows us to grow the total capacity of the model, while keeping the total number of trainable parameters per language constant. Our approach enables adding languages post-hoc with no measurable drop in performance, no longer limiting the model usage to the set of pre-trained languages.
  arXiv  Detail & Related papers  (2022-05-12T17:59:56Z)
• Continual Learning in Multilingual NMT via Language-Specific Embeddings [92.91823064720232]
  It consists in replacing the shared vocabulary with a small language-specific vocabulary and fine-tuning the new embeddings on the new language's parallel data. Because the parameters of the original model are not modified, its performance on the initial languages does not degrade.
  arXiv  Detail & Related papers  (2021-10-20T10:38:57Z)
• UNKs Everywhere: Adapting Multilingual Language Models to New Scripts [103.79021395138423]
  Massively multilingual language models such as multilingual BERT (mBERT) and XLM-R offer state-of-the-art cross-lingual transfer performance on a range of NLP tasks. Due to their limited capacity and large differences in pretraining data, there is a profound performance gap between resource-rich and resource-poor target languages. We propose novel data-efficient methods that enable quick and effective adaptation of pretrained multilingual models to such low-resource languages and unseen scripts.
  arXiv  Detail & Related papers  (2020-12-31T11:37:28Z)
• Parsing with Multilingual BERT, a Small Corpus, and a Small Treebank [46.626315158735615]
  Pretrained multilingual contextual representations have shown great success, but due to the limits of their pretraining data, their benefits do not apply equally to all language varieties. This presents a challenge for language varieties unfamiliar to these models, whose labeled emphand unlabeled data is too limited to train a monolingual model effectively. We propose the use of additional language-specific pretraining and vocabulary augmentation to adapt multilingual models to low-resource settings.
  arXiv  Detail & Related papers  (2020-09-29T16:12:52Z)
• Learning to Scale Multilingual Representations for Vision-Language Tasks [51.27839182889422]
  The effectiveness of SMALR is demonstrated with ten diverse languages, over twice the number supported in vision-language tasks to date. We evaluate on multilingual image-sentence retrieval and outperform prior work by 3-4% with less than 1/5th the training parameters compared to other word embedding methods.
  arXiv  Detail & Related papers  (2020-04-09T01:03:44Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.