CI failure due to transformers VLM change

[Tcc0403]

🐛 Describe the bug

CI details: Qwen2VLConfig, monkey patch impl related

Text config is seperated out from the general config in transformers>=4.52.0. Qwen2VLRotaryEmbedding takes Qwen2VLTextConfig intead of Qwen2VLConfig now.

https://github.com/huggingface/transformers/pull/37268

We also need to check all VLMs monkey patch since most models are refactored into TextModel and VisionModel

blocked on: https://github.com/huggingface/transformers/issues/38331 (can walkaround by passing hidden_size and vocab_size to config dict directly though)

Reproduce

No response

Versions

liger_kernel==0.5.9 transformers>=4.52.0

[Tcc0403]

video_processor(Qwen2VLVideoProcessor) is now a required input for Qwen2VLProcessor.

We might need to add torchvision to our dev dependency since it is required for videoprocessor.

Related PR: https://github.com/huggingface/transformers/pull/35206

[Tcc0403]

fixed by #735 #738 #743 #755