Over-Tokenized Transformer: Vocabulary is Generally Worth Scaling
This addresses tokenization design for more efficient and powerful LLMs, though it is incremental as it builds on existing transformer frameworks.
The paper tackles the problem of tokenization's impact on large language model scaling by introducing Over-Tokenized Transformers, which decouples input and output vocabularies and scales input vocabularies with multi-gram tokens, resulting in performance comparable to double-sized baselines at no extra cost.
Tokenization is a fundamental component of large language models (LLMs), yet its influence on model scaling and performance is not fully explored. In this paper, we introduce Over-Tokenized Transformers, a novel framework that decouples input and output vocabularies to improve language modeling performance. Specifically, our approach scales up input vocabularies to leverage multi-gram tokens. Through extensive experiments, we uncover a log-linear relationship between input vocabulary size and training loss, demonstrating that larger input vocabularies consistently enhance model performance, regardless of model size. Using a large input vocabulary, we achieve performance comparable to double-sized baselines with no additional cost. Our findings highlight the importance of tokenization in scaling laws and provide practical insight for tokenizer design, paving the way for more efficient and powerful LLMs.