Model-Aware Tokenizer Transfer
About
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 benchmarks
| Task | Dataset | Result | Rank | |
|---|---|---|---|---|
| Machine Translation | Long FLORES uk to en (test) | BLEU27.89 | 14 | |
| Machine Translation | WMT Ukrainian (test) | BLEU4.71 | 14 | |
| Reading Comprehension | Belebele Ukrainian (test) | Accuracy89.56 | 14 | |
| Abstractive Summarization | XL-Sum Ukrainian (test) | BLEU Score5.95 | 14 | |
| General Knowledge | Global MMLU Ukrainian (test) | Accuracy (%)64.98 | 14 | |
| Machine Translation | Long FLORES en to uk (test) | BLEU8.7 | 14 | |
| Multilingual Discriminative Language Understanding | Belebele, Global MMLU, and MMMLU Average across Arabic, German, Japanese, Swahili (mean of per-language values) | Belebele Accuracy66.39 | 8 | |
| Multilingual Text Generation | Long FLORES and XL-Sum Average across Arabic, German, Japanese, Swahili | Long FLORES en->x Score5.2 | 8 |