Add Multimodal LLM Finetuning

[BUAADreamer]

What does this PR do?

Add finetuning Multimodal-LLM by leveraging AutoModelForVision2Seq and AutoProcessortransformers

This PR is working in progress, need improvement in the future

TODO

• [ ] LLaVA
• [ ] Instruct-BLIP

Before submitting

• [x] Did you read the contributor guideline?