DOTA: Distributional Test-time Adaptation of Vision-Language Models | Read Paper on Bytez