[VLM] Qwen 2.5 VL

[kylesayrs]

Purpose

• Support quantizing Qwen 2.5 VL model

Changes

• Add example examples/multimodal_vision/qwen_2_5_vl_example.py
• Implement TraceableQwen2_5_VLForConditionalGeneration
• Small style changes to TraceableQwen2_VLForConditionalGeneration

Testing

• Ran added example to completion

[github-actions]

👋 Hi! Thank you for contributing to llm-compressor. Please add the ready label when the PR is ready for review.

Note: This is required to complete the testing suite, please only add the label once the PR is code complete and local testing has been performed.

[brian-dellabetta]

I must admit I don't understand a good amount of the qwen2.5-specific tracing code, and shared this with @kylesayrs , but as this is code that will only run specific to the Qwen 2/2.5 VL models, I am not being too meticulous in the review.

Thanks Kyle! 🚀

[brian-dellabetta]

a quick ELI5 would be nice to have at the top

Suggested change

	# The following example shows how the Qwen2.5 Vision-Language model can be W4A16 quantized

[kylesayrs]

I think the readme in this folder provides enough context for a user to know how to use this example/what it's doing

[kylesayrs]

Need to wait until 2.5 lands in transformers