# WordPiece

> Source: https://sukruyusufkaya.com/en/glossary/wordpiece
> Updated: 2026-05-13T21:00:58.647Z
> Type: glossary
> Category: dogal-dil-isleme
**TLDR:** A widely used tokenization method that optimizes subword units with respect to probabilistic coverage.

<p>WordPiece is a subword tokenization approach that became especially popular with the BERT family. Although similar to BPE in appearance, it uses different criteria for merge decisions and aims to make vocabulary structure more effective for the modeling objective. It has delivered strong results in large-scale pretraining systems. It clearly illustrates how tightly model behavior and token-vocabulary design are connected.</p>