unsloth/Llama-4-Scout-17B-16E-Instruct-GGUF · Is Llama-4-Scout-17B-16E-Instruct-GGUF a multimodal model that supports image input?

Hi! Thank you for this great model.

I’m trying to use the Llama-4-Scout-17B-16E-Instruct-UD-Q2_K_XL.gguf model for image+text input, like LLaVA. Specifically, I want to describe images with a prompt like this:

prompt = [{
    "role": "user",
    "content": [
         {"type": "image_url", "image_url": {
                    "url": data_uri 
                }},
        {"type": "text", "text": "Please describe this image."},
    ]
}]

I’m using llama-cpp-python with a Llava15ChatHandler, and passing in a clip_model_path. The image gets encoded successfully, but the model returns completely irrelevant answers (e.g., calling a dog photo "a tourist center").

Also,I inspected the GGUF metadata and found no indicators that this model is multimodal — no vision tower configuration or image-related fields are present.

general.name
general.architecture
...
llama4.context_length
llama4.embedding_length
...
tokenizer.chat_template

So I’m wondering:
Is this model actually multimodal? If not yet multimodal, are there plans to release a future version with image support?
Or is it currently text-only, and not designed to process the <|image|> token or CLIP embeddings?
If this model is even intended to be used with image inputs via llama-cpp-python, or if a compatible handler might be needed in the future?

Additionally, I also checked the llama-cpp-python documentation and the list of built-in LlamaChatHandlers (e.g., Llava15ChatHandler, Llava16ChatHandler, etc.), and I couldn’t find any specific handler for Llama 4 Scout models.

Thank you!