Equivalent Linear Mappings of Large Language Models