Question: what is the purpose of softmax for words/tokens before the final word? Isn't it only the softmax distribution respective to the final word that gets u

Question: what is the purpose of softmax for words/tokens before the final word? Isn't it only the softmax distribution respective to the final word that gets used for the next token prediction? That was my understanding at least.