DEFT: Decompositional Efficient Fine-Tuning for Text-to-Image Models | Read Paper on Bytez