Decoupling the Benefits of Subword Tokenization for Language Model Training via Byte-level Simulation
For researchers improving byte-level or subword language models, this work clarifies the specific benefits of subword tokenization, though it is incremental as it refines existing understanding.
The paper decouples the effects of subword tokenization in LLMs by isolating them in a byte-level pretraining pipeline, finding that increased training throughput and subword boundaries as priors are key to subword models outperforming byte models.
Subword tokenization is an essential part of modern large language models (LLMs), yet its specific contributions to training efficiency and model performance remain poorly understood. In this work, we decouple the effects of subword tokenization by isolating them within a controlled byte-level pretraining pipeline. We formulate and test hypotheses across various dimensions, including sample throughput, vocabulary scaling, and the linguistic prior of subword boundaries. By simulating these effects in a byte-level setting, we refine our understanding of why subword models outperform raw byte models and offer insights to improve the pretraining of future byte-level and subword models. Specifically, our experiments highlight the critical role of increased training throughput and the integration of subword boundaries as either explicit priors or inductive biases.