How does the transformer architecture process all tokens in parallel rather than sequentially? — How LLMs Work

How does the transformer architecture process all tokens in parallel rather than sequentially? — How LLMs Work | Unlomi