Decomposed Prompt Tuning (DEPT) is a method for parameter-efficient fine-tuning of language models. It achieves better performance while saving memory and time costs compared to other approaches, and is adaptable to different model architectures and sizes.