Add support for ViP-LLaVA?

Hi OLLaMa team, [ViP-LLaVA](https://vip-llava.github.io/) is a region-level large multimodal model from LLaVA team that is capable of understanding visual prompts such as scribbles, bounding boxes, arrows, etc. There are only several lines of changes to the original LLaVA code. Huggingface already integrate ViP-LLaVA into the official transformers library. [https://huggingface.co/docs/transformers/main/model_doc/vipllava](https://huggingface.co/docs/transformers/main/model_doc/vipllava) Consider adding ViP-LLaVA here? Thank you! Mu Cai