Feature/support qwenvl glm4-v phi3-v(tested)

[marko1616]

What does this PR do?

Fixes #4375

Before submitting

• Did you read the contributor guideline?
• Did you write any new necessary tests?

[marko1616]

终于还差一个image的padding处理就能做好训练支持了。

[marko1616]

@hiyouga 改的比较多捏，有空帮忙看看这个实现思路行不行。谢谢。

[marko1616]

成功跑了训练。

[BUAADreamer]

暂时先不要引入更多模型，把现有的三个模型完善好，控制diff🤗 @marko1616

[marko1616]

暂时先不要引入更多模型，把现有的三个模型完善好，控制diff🤗 @marko1616

oaky，现在确实在测试别的训练模式。

[marko1616]

@BUAADreamer Thx for review i'll fix those soon.
Wait for several hours, I will see all of your valuable codes and try my best to give some suggestions!! 🤗

[BUAADreamer]

I have read all the code, I think this is a great PR. I think we can support more custom MLLMs with your commit!! However, we will discuss and decide when and how to merge it!! 🤗

[marko1616]

I have read all the code, I think this is a great PR. I think we can support more custom MLLMs with your commit!! However, we will discuss and decide when and how to merge it!! 🤗

Really THX for your fully review.I will fix those problems in next few days.