MM SFT of Gemma4

[RoyHirsch]

Describe the solution you'd like
Seems like there is no support in SFT/LORA/RL of gemma4 model with multi-modal inputs (text+image for example). There is support for gemma3, I would appriciate if you can supply such support and pointers

Checklist

• [ V] I have searched the existing issues for similar feature requests.
• [V ] This is not a support question (please use the "bug template" for that).

[msghik]

I believe I resolved it in pull request

[tianshub]

Thanks @msghik ! Let's iterate on the PR.