The paper proposes a modular tokenizer framework for multilingual large language models, allowing the creation of language‑specific subtokenizers that match monolingual compression quality. It introduces a pretraining strategy that samples these subtokenizers to limit predictions to relevant vocabularies, enabling efficient training and inference. This approach reduces memory usage and speeds up inference without compromising performance.
By Franck Signe, Hippolyte Pilchen, Fran\c{c}ois Yvon, \'Edouard Grave
Conceptizer is a training framework that augments next-token prediction with a reward for meaning by using a fixed vocabulary of about 9,000 concepts derived from contextual synonyms. It maps each token distribution to a concept distribution, sharing the reward for the target meaning across its possible wordings, thereby providing a concept-level training signal without adding model parameters. Applied to 1B–8B parameter models, Conceptizer achieves lower perplexity at concept-target positions, assigns higher probability to contextual synonyms, and improves downstream performance on knowledge and sentiment tasks while remaining comparable on commonsense tasks.
By Haoran Zheng, Chen Shani
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...
arXiv:2606.23566v3 Announce Type: replace
Abstract: Language-specific tokenizers improve tokenization quality and the downstream performance of models on those languages. Using one with an existing p...
By Clara Meister, Suchir Salhan, Andrzej Szablewski, Pietro Lesci, Paula Buttery, Tiago Pimentel
arXiv:2608. 19529v1 Announce Type: cross Abstract: Many real-world AI systems represent entities, behaviors, and structured information using discrete machine-native symbols rather than natural language.
By Su Yan, Rakesh Iyer
arXiv:2604.13286v2 Announce Type: replace
Abstract: Despite the widespread multilingual deployment of large language models, post-training pipelines remain predominantly English-centric, contributing...
By Mehak Dhaliwal, Shashwat Chaurasia, Yao Qin, Dezhi Hong, Thomas Butler
arXiv:2605. 30580v2 Announce Type: replace-cross Abstract: Speculative decoding is a popular technique for large language model (LLM) inference, enabling faster generation by drafting multiple tokens with a smaller draft model.
By Nirajan Paudel, Michael Ginn, Luc De Nardi, Alexis Palmer
arXiv:2512. 20757v2 Announce Type: replace-cross Abstract: Tokenizers provide the fundamental basis through which text is represented and processed by language models (LMs).
By G\"ul Sena Alt{\i}nta\c{s}, Malikeh Ehghaghi, Brian Lester, Fengyuan Liu, Wanru Zhao, Marco Ciccone, Colin Raffel
The paper investigates whether intrinsic tokenizer features can predict the performance of downstream language models. By training about 200 models across nine languages and evaluating numerous tokenizer metrics, the authors find that none of the existing metrics consistently forecasts model performance across different families, tasks, and languages. They propose a new linguistically grounded metric that captures aspects of tokenizer quality missed by prior measures, though its predictive power may still vary.
By Antonio Lopardo, Luis Frentzen Salim, Shikhar Srivastava, Kevin Jiang, Mridul Sharma, Catherine Arnett
arXiv:2608. 15964v1 Announce Type: cross Abstract: Language-specific competency (LSC) is the phenomenon of a language model performing better or worse depending on the language of the prompt.
By Ishika Agarwal, Arkajyoti Charaborty, Tanner Sorensen, Neha Gupta, Andreas Stolcke
The paper proposes an information-weighted cross‑entropy loss that rescales token contributions using TF‑IDF statistics, thereby emphasizing semantically informative tokens and down‑weighting ubiquitous ones. Experiments on five decoder‑only language models (1.1B–13B parameters) show consistent reductions in memorized substring length while maintaining perplexity and downstream performance. The method is architecture‑agnostic, adds less than 3% computational overhead, and can be integrated into existing training pipelines.
By Zhijian Li, Stefan Larson, Kevin Leach
State-of-the-art retrieval models increasingly rely on closed training data, creating a reproducibility gap. We present an open end-to-end recipe for training retrieval models and study how English supervision transfers to multilingual retrieval through translate-train.