How to get Attention map for understanding which parts of the input image the model focuses on

[Ishtiaque-khan97]

Hi,
I am trying to visualize which parts of the input image the model focuses on when making decisions or generating output. I wish to see if it focuses on the objects I am asking it about.
I guess I am looking for something like this:
https://github.com/zjysteven/VLM-Visualizer

The above is for LLava. Could anyone please tell me how to do this for Qwen 2?
Thank you.