Clarifications regarding Gemma 3 model

[tommasoTubaldo]

I have some doubts regarding the implementation of Gemma 3 on mlx_community.

In particular:

• does Gemma3 supports multi-image input in mlx-vlm? Because I saw that models like Qwen2-VL and Phi3_Vision do.
• does Gemma3 supports vision fine-tuning in mlx-vlm?

Thanks in advance for the support🙏🏼