Prompt, Generate, Then Cache: Cascade of Foundation Models Makes Strong Few-Shot Learners | Read Paper on Bytez