# Retrofitting Large Language Models with Dynamic Tokenization

arXiv: https://arxiv.org/abs/2411.18553 · HTML: https://mlsanity.com/paper/2411.18553/

## Key facts

- Cited by 16 papers in the map (data through 14 Oct 2026).
- Builds on 38 of its 50 references that are in the map.
- First posted on arXiv 27 Nov 2024.

## Builds on (most cited first, 30 of 38)

- [BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding](https://mlsanity.com/paper/1810.04805/) (2018)
- [Llama 2: Open Foundation and Fine-Tuned Chat Models](https://mlsanity.com/paper/2307.09288/) (2023)
- [Unsupervised Cross-lingual Representation Learning at Scale](https://mlsanity.com/paper/1911.02116/) (2019)
- [Mistral 7B](https://mlsanity.com/paper/2310.06825/) (2023)
- [Neural Machine Translation of Rare Words with Subword Units](https://mlsanity.com/paper/1508.07909/) (2015)
- [SentencePiece: A simple and language independent subword tokenizer and detokenizer for Neural Text Processing](https://mlsanity.com/paper/1808.06226/) (2018)
- [XNLI: Evaluating Cross-lingual Sentence Representations](https://mlsanity.com/paper/1809.05053/) (2018)
- [Can Large Language Models Be an Alternative to Human Evaluations?](https://mlsanity.com/paper/2305.01937/) (2023)
- [Subword Regularization: Improving Neural Network Translation Models with Multiple Subword Candidates](https://mlsanity.com/paper/1804.10959/) (2018)
- [ByT5: Towards a token-free future with pre-trained byte-to-byte models](https://mlsanity.com/paper/2105.13626/) (2021)
- [How Good is Your Tokenizer? On the Monolingual Performance of Multilingual Language Models](https://mlsanity.com/paper/2012.15613/) (2020)
- [CANINE: Pre-training an Efficient Tokenization-Free Encoder for Language Representation](https://mlsanity.com/paper/2103.06874/) (2021)
- [Byte Pair Encoding is Suboptimal for Language Model Pretraining](https://mlsanity.com/paper/2004.03720/) (2020)
- [Do All Languages Cost the Same? Tokenization in the Era of Commercial Language Models](https://mlsanity.com/paper/2305.13707/) (2023)
- [MADLAD-400: A Multilingual And Document-Level Large Audited Dataset](https://mlsanity.com/paper/2309.04662/) (2023)
- [MEGABYTE: Predicting Million-byte Sequences with Multiscale Transformers](https://mlsanity.com/paper/2305.07185/) (2023)
- [Charformer: Fast Character Transformers via Gradient-based Subword Tokenization](https://mlsanity.com/paper/2106.12672/) (2021)
- [Breaking the Softmax Bottleneck: A High-Rank RNN Language Model](https://mlsanity.com/paper/1711.03953/) (2017)
- [Between words and characters: A Brief History of Open-Vocabulary Modeling and Tokenization in NLP](https://mlsanity.com/paper/2112.10508/) (2021)
- [Tokenizer Choice For LLM Training: Negligible or Crucial?](https://mlsanity.com/paper/2310.08754/) (2023)
- [WECHSEL: Effective initialization of subword embeddings for cross-lingual transfer of monolingual language models](https://mlsanity.com/paper/2112.06598/) (2021)
- [XLM-V: Overcoming the Vocabulary Bottleneck in Multilingual Masked Language Models](https://mlsanity.com/paper/2301.10472/) (2023)
- [CharacterBERT: Reconciling ELMo and BERT for Word-Level Open-Vocabulary Representations From Characters](https://mlsanity.com/paper/2010.10392/) (2020)
- [Efficient Transformers with Dynamic Token Pooling](https://mlsanity.com/paper/2211.09761/) (2022)
- [Adv-BERT: BERT is not robust on misspellings! Generating nature adversarial samples on BERT](https://mlsanity.com/paper/2003.04985/) (2020)
- [Getting the most out of your tokenizer for pre-training and domain adaptation](https://mlsanity.com/paper/2402.01035/) (2024)
- [Impact of Tokenization on Language Models: An Analysis for Turkish](https://mlsanity.com/paper/2204.08832/) (2022)
- [Fast Vocabulary Transfer for Language Model Compression](https://mlsanity.com/paper/2402.09977/) (2024)
- [Multi-view Subword Regularization](https://mlsanity.com/paper/2103.08490/) (2021)
- [MYTE: Morphology-Driven Byte Encoding for Better and Fairer Multilingual Language Modeling](https://mlsanity.com/paper/2403.10691/) (2024)

## Built on by (most cited first, 16 of 16)

- [Intern-S1: A Scientific Multimodal Foundation Model](https://mlsanity.com/paper/2508.15763/) (2025)
- [Universal Cross-Tokenizer Distillation via Approximate Likelihood Matching](https://mlsanity.com/paper/2503.20083/) (2025)
- [AdaptiVocab: Enhancing LLM Efficiency in Focused Domains through Lightweight Vocabulary Adaptation](https://mlsanity.com/paper/2503.19693/) (2025)
- [Bolmo: Byteifying the Next Generation of Language Models](https://mlsanity.com/paper/2512.15586/) (2025)
- [The Vision Wormhole: Latent-Space Communication in Heterogeneous Multi-Agent Systems](https://mlsanity.com/paper/2602.15382/) (2026)
- [Stop Taking Tokenizers for Granted: They Are Core Design Decisions in Large Language Models](https://mlsanity.com/paper/2601.13260/) (2026)
- [zip2zip: Inference-Time Adaptive Tokenization via Online Compression](https://mlsanity.com/paper/2506.01084/) (2025)
- [Vocabulary embeddings organize linguistic structure early in language model training](https://mlsanity.com/paper/2510.07613/) (2025)
- [Phonemes to the Rescue: Multilingual Tokenization Based on International Phonetic Alphabet](https://mlsanity.com/paper/2606.20993/) (2026)
- [Universal NER v2: Towards a Massively Multilingual Named Entity Recognition Benchmark](https://mlsanity.com/paper/2604.12744/) (2026)
- [Large Language Models as Amortized Pareto-Front Generators for Constrained Bi-Objective Convex Optimization](https://mlsanity.com/paper/2605.12106/) (2026)
- [Check Your LLM's Secret Dictionary! Five Lines of Code Reveal What Your LLM Learned (Including What It Shouldn't Have)](https://mlsanity.com/paper/2605.22005/) (2026)
- [Emergent retokenization symmetry in large language models: phenomenology and applications](https://mlsanity.com/paper/2606.15521/) (2026)
- [LangMAP: A Language-Adaptive Approach to Tokenization](https://mlsanity.com/paper/2606.23566/) (2026)
- [When Tokenizers Fail: Byte-Level Chunking for Zero-Shot Transfer to Low-Resource Languages](https://mlsanity.com/paper/2608.27658/) (2026)
- [zip2zip++: Towards Inference-Time Token Compression with Minimal Quality Loss](https://mlsanity.com/paper/2610.13605/) (2026)

## Abstract (arXiv)

Current language models (LMs) use a fixed, static subword tokenizer. This default choice typically results in degraded efficiency and language capabilities, especially in languages other than English. To address this issue, we challenge the static design and propose retrofitting LMs with dynamic tokenization: a way to dynamically decide on token boundaries based on the input text via a subword-merging algorithm inspired by byte-pair encoding. We merge frequent subword sequences in a batch, then apply a pre-trained embedding-prediction hypernetwork to compute the token embeddings on-the-fly. For encoder-style models (e.g., XLM-R), this on average reduces token sequence lengths by >20% across 14 languages while degrading performance by less than 2%. The same method applied to pre-filling and scoring in decoder-style models (e.g., Mistral-7B) results in minimal performance degradation at up to 17% reduction in sequence length. Overall, we find that dynamic tokenization can mitigate the limitations of static tokenization by substantially improving inference speed and promoting fairness across languages, enabling more equitable and adaptable LMs.