Large Vocabulary Size Improves Large Language Models
This work addresses vocabulary optimization for LLMs, offering practical insights for model training and adaptation, though it is incremental in nature.
The paper investigates the effect of subword vocabulary size on large language model performance, finding that larger vocabularies improve results, and demonstrates that using a new vocabulary for continual training on a different language outperforms using the pre-trained vocabulary.
This paper empirically investigates the relationship between subword vocabulary size and the performance of large language models (LLMs) to provide insights on how to define the vocabulary size. Experimental results show that larger vocabulary sizes lead to better performance in LLMs. Moreover, we consider a continual training scenario where a pre-trained language model is trained on a different target language. We introduce a simple method to use a new vocabulary instead of the pre-defined one. We show that using the new vocabulary outperforms the model with the vocabulary used in pre-training.