TokSuite: Measuring the Impact of Tokenizer Choice on Language Model Behavior

TL;DR

TokSuite is a comprehensive benchmark and set of models designed to isolate and measure the impact of different tokenizers on language model performance and behavior, revealing their respective strengths and weaknesses.

Contribution

This work introduces TokSuite, a novel benchmark and a collection of models trained with various tokenizers, enabling detailed analysis of tokenizer effects on language models.

Findings

Different tokenizers significantly influence model performance.

Some tokenizers better handle real-world text perturbations.

Tokenization choices impact model robustness and efficiency.

Abstract

Tokenizers provide the fundamental basis through which text is represented and processed by language models (LMs). Despite the importance of tokenization, its role in LM performance and behavior is poorly understood due to the challenge of measuring the impact of tokenization in isolation. To address this need, we present TokSuite, a collection of models and a benchmark that supports research into tokenization's influence on LMs. Specifically, we train fourteen models that use different tokenizers but are otherwise identical using the same architecture, dataset, training budget, and initialization. Additionally, we curate and release a new benchmark that specifically measures model performance subject to real-world perturbations that are likely to influence tokenization. Together, TokSuite allows robust decoupling of the influence of a model's tokenizer, supporting a series of novel findings that elucidate the respective benefits and shortcomings of a wide range of popular tokenizers.