2024.06.25 [23’] LLaMA-Adapter V2: Parameter-Efficient Visual Instruction Model Multimodal Joint Training Prefix Tuning