Vision (Auto)Processor multiple images finetuning example.

### Feature request Is it possible to upload an example of how to finetune PaLIGemma on multi-image inputs? Something similar to [multi-image-inference](https://huggingface.co/docs/transformers/main/model_doc/paligemma#multi-image-inference), which shows how to perform multi-image inference over PaLIGemma. ### Motivation enabling finetuning of multi-image PaLIGemma ### Your contribution .