# BPE

A tokenization algorithm that repeatedly merges the most frequent adjacent symbol pairs into new tokens.

## What it is used for

The most common subword tokenizer for language models.