Add Support for Gemma 3

[gagan3012]

Hello,
Gemma 3 is a multimodal model that performs well on all vision tasks as well it would be great if support for it could be added

[armanakbari]

It would be also helpful if they wrote a section on what changes we should make to make the code work for other backbones.

In the repo, it is mentioned that the code works for VLMs in the Transformers library. However, I could not run the code for other models and got into a lot of errors. I changed the model name in the bash script and also changed the lines that we are inputting the pertained model. (still has a lot of errors)