Exploiting Presentative Feature Distributions for Parameter-Efficient Continual Learning of Large Language Models | Read Paper on Bytez