Thaks for your excellent blog! I have some questions that can I make this fintuned model based on the unscloth frame to get the embeddings?That means can I just input a image, and I can get the image-text embedding from finetune llama?