Compositional zero-shot learning (CZSL) aims to identify novel compositions formed by known primitives (attributes and objects). Motivated by recent advancements in pre-trained vision-language models ...