Please add TinyGPT-V model support

TinyGPT-V: Efficient Multimodal Large Language Model via Small Backbones Github: https://github.com/DLYuanGod/TinyGPT-V HuggingFace: https://huggingface.co/Tyrannosaurus/TinyGPT-V It stands out because it only requires a 24G GPU for training, and just an 8G GPU or CPU for inference. TinyGPT-V is based on Phi-2, combining an effective language backbone with pre-trained visual modules from BLIP-2 or CLIP.