Saved in:
| Main Authors: | , , , , |
|---|---|
| Format: | Preprint |
| Published: |
2025
|
| Subjects: | |
| Online Access: | https://arxiv.org/abs/2506.03101 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1866916776226324480 |
|---|---|
| author | Lotz, Jonas F. Lopes, António V. Peitz, Stephan Setiawan, Hendra Emili, Leonardo |
| author_facet | Lotz, Jonas F. Lopes, António V. Peitz, Stephan Setiawan, Hendra Emili, Leonardo |
| contents | The choice of tokenizer can profoundly impact language model performance, yet accessible and reliable evaluations of tokenizer quality remain an open challenge. Inspired by scaling consistency, we show that smaller models can accurately predict significant differences in tokenizer impact on larger models at a fraction of the compute cost. By systematically evaluating both English-centric and multilingual tokenizers, we find that tokenizer choice has negligible effects on tasks in English but results in consistent performance differences in multilingual settings. We propose new intrinsic tokenizer metrics inspired by Zipf's law that correlate more strongly with downstream performance than text compression when modeling unseen languages. By combining several metrics to capture multiple aspects of tokenizer behavior, we develop a reliable framework for intrinsic tokenizer evaluations. Our work offers a more efficient path to informed tokenizer selection in future language model development. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2506_03101 |
| institution | arXiv |
| publishDate | 2025 |
| record_format | arxiv |
| spellingShingle | Beyond Text Compression: Evaluating Tokenizers Across Scales Lotz, Jonas F. Lopes, António V. Peitz, Stephan Setiawan, Hendra Emili, Leonardo Computation and Language The choice of tokenizer can profoundly impact language model performance, yet accessible and reliable evaluations of tokenizer quality remain an open challenge. Inspired by scaling consistency, we show that smaller models can accurately predict significant differences in tokenizer impact on larger models at a fraction of the compute cost. By systematically evaluating both English-centric and multilingual tokenizers, we find that tokenizer choice has negligible effects on tasks in English but results in consistent performance differences in multilingual settings. We propose new intrinsic tokenizer metrics inspired by Zipf's law that correlate more strongly with downstream performance than text compression when modeling unseen languages. By combining several metrics to capture multiple aspects of tokenizer behavior, we develop a reliable framework for intrinsic tokenizer evaluations. Our work offers a more efficient path to informed tokenizer selection in future language model development. |
| title | Beyond Text Compression: Evaluating Tokenizers Across Scales |
| topic | Computation and Language |
| url | https://arxiv.org/abs/2506.03101 |