Explaining Large Language Models Decisions Using Shapley Values