| Token-free models that instead operate directly on raw text (bytes or characters) have many benefits: they | |
| can process text in any language out of the box, they are more robust to noise, and they minimize technical debt by | |
| removing complex and error-prone text preprocessing pipelines. |