Model-Aware Tokenizer Transfer

• URL: http://arxiv.org/abs/2510.21954v1
• Date: Fri, 24 Oct 2025 18:27:36 GMT
• Title: Model-Aware Tokenizer Transfer
• Authors: Mykola Haltiuk, Aleksander Smywiński-Pohl,
• Abstract summary: Model-Aware Tokenizer Transfer (MATT) is a method that incorporates model internals into the tokenizer transfer process.<n>MATT introduces an Attention Influence Modeling (AIM) objective that distills inter-token communication patterns from a source model into a target model.<n>Experiments show that MATT recovers a large fraction of the original model's performance within a few GPU hours, outperforming baselines.
• Score: 46.13517417540154
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: Large Language Models (LLMs) are trained to support an increasing number of languages, yet their predefined tokenizers remain a bottleneck for adapting models to lower-resource or distinct-script languages. Existing tokenizer transfer methods typically rely on semantic heuristics to initialize new embeddings, ignoring higher-layer model dynamics and limiting transfer quality. We propose Model-Aware Tokenizer Transfer (MATT), a method that incorporates model internals into the tokenizer transfer process. MATT introduces an Attention Influence Modeling (AIM) objective that distills inter-token communication patterns from a source model into a target model with a new tokenizer, providing an efficient warm-up before standard language modeling. Unlike approaches that focus solely on embedding similarity, MATT leverages attention behavior to guide embedding initialization and adaptation. Experiments across diverse linguistic settings show that MATT recovers a large fraction of the original model's performance within a few GPU hours, outperforming heuristic baselines. These results demonstrate that incorporating model-level signals offers a practical and effective path toward robust tokenizer transfer in multilingual LLMs.

Related papers

• One-step Language Modeling via Continuous Denoising [36.18484491074519]
  We show that language models leveraging flow-based continuous denoising can outperform discrete diffusion in both quality and speed.<n>Our work calls into question the widely held hypothesis that discrete diffusion processes are necessary for generative modeling over discrete modalities.
  arXiv  Detail & Related papers  (2026-02-18T19:23:07Z)
• Model-Dowser: Data-Free Importance Probing to Mitigate Catastrophic Forgetting in Multimodal Large Language Models [2.83595986479415]
  Fine-tuning Multimodal Large Language Models (MLLMs) on task-specific data is an effective way to improve performance on downstream applications.<n>Existing methods that aim to mitigate this issue either become ineffective when fine-tuning deeper layers of the language decoder or scale poorly with increasing model size.<n>We propose Model-Dowser, a novel sparse fine-tuning approach for MLLMs.
  arXiv  Detail & Related papers  (2026-02-04T12:56:27Z)
• OptMerge: Unifying Multimodal LLM Capabilities and Modalities via Model Merging [124.91183814854126]
  Model merging seeks to combine multiple expert models into a single model.<n>We introduce a benchmark for model merging research that clearly divides the tasks for MLLM training and evaluation.<n>We find that model merging offers a promising way for building improved MLLMs without requiring training data.
  arXiv  Detail & Related papers  (2025-05-26T12:23:14Z)
• Latent Thought Models with Variational Bayes Inference-Time Computation [52.63299874322121]
  Latent Thought Models (LTMs) incorporate explicit latent thought vectors that follow an explicit prior model in latent space.<n>LTMs demonstrate superior sample and parameter efficiency compared to autoregressive models and discrete diffusion models.
  arXiv  Detail & Related papers  (2025-02-03T17:50:34Z)
• Can bidirectional encoder become the ultimate winner for downstream applications of foundation models? [1.8120356834558644]
  Foundational models have the characteristics of pre-training, transfer learning, and self-supervised learning.<n>BERT broke through the limitation of only using one-way methods for language modeling in pre-training by using a masked language model.<n>This article analyzes one-way and bidirectional models based on GPT and BERT and compares their differences based on the purpose of the model.
  arXiv  Detail & Related papers  (2024-11-27T03:31:14Z)
• MoE-CT: A Novel Approach For Large Language Models Training With Resistance To Catastrophic Forgetting [53.77590764277568]
  We introduce a novel MoE-CT architecture that separates the base model's learning from the multilingual expansion process. Our design freezes the original LLM parameters, thus safeguarding its performance in high-resource languages, while an appended MoE module, trained on diverse language datasets, augments low-resource language proficiency.
  arXiv  Detail & Related papers  (2024-06-25T11:03:45Z)
• ML-SUPERB 2.0: Benchmarking Multilingual Speech Models Across Modeling Constraints, Languages, and Datasets [106.7760874400261]
  This paper presents ML-SUPERB2.0, which is a new benchmark for evaluating pre-trained SSL and supervised speech models. We find performance improvements over the setup of ML-SUPERB, but performance depends on the downstream model design. Also, we find large performance differences between languages and datasets, suggesting the need for more targeted approaches.
  arXiv  Detail & Related papers  (2024-06-12T21:01:26Z)
• On the Analysis of Cross-Lingual Prompt Tuning for Decoder-based Multilingual Model [49.81429697921861]
  We study the interaction between parameter-efficient fine-tuning (PEFT) and cross-lingual tasks in multilingual autoregressive models. We show that prompt tuning is more effective in enhancing the performance of low-resource languages than fine-tuning.
  arXiv  Detail & Related papers  (2023-11-14T00:43:33Z)
• Open-vocabulary Semantic Segmentation with Frozen Vision-Language Models [39.479912987123214]
  Self-supervised learning has exhibited a notable ability to solve a wide range of visual or language understanding tasks. We introduce Fusioner, with a lightweight, transformer-based fusion module, that pairs the frozen visual representation with language concept. We show that, the proposed fusion approach is effective to any pair of visual and language models, even those pre-trained on a corpus of uni-modal data.
  arXiv  Detail & Related papers  (2022-10-27T02:57:26Z)
• Model-Agnostic Multitask Fine-tuning for Few-shot Vision-Language Transfer Learning [59.38343286807997]
  We propose Model-Agnostic Multitask Fine-tuning (MAMF) for vision-language models on unseen tasks. Compared with model-agnostic meta-learning (MAML), MAMF discards the bi-level optimization and uses only first-order gradients. We show that MAMF consistently outperforms the classical fine-tuning method for few-shot transfer learning on five benchmark datasets.
  arXiv  Detail & Related papers  (2022-03-09T17:26:53Z)
• Early Stage LM Integration Using Local and Global Log-Linear Combination [46.91755970827846]
  Sequence-to-sequence models with an implicit alignment mechanism (e.g. attention) are closing the performance gap towards traditional hybrid hidden Markov models (HMM) One important factor to improve word error rate in both cases is the use of an external language model (LM) trained on large text-only corpora. We present a novel method for language model integration into implicit-alignment based sequence-to-sequence models.
  arXiv  Detail & Related papers  (2020-05-20T13:49:55Z)

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.