So, for reasons both pragmatic and technical, today’s transformer models work with text that’s been broken down into smaller, bite-sized pieces called tokens — a process known as tokenization.