论文笔记 - InstructBLIP: Towards General-purpose Vision-Language Models with Instruction Tuning