[FEATURE] Update Qwen2.5VL to v4.54.1

[zhtmike]

What does this PR do?

Fixes # (issue)

• Fixed incorrect output of Qwen2.5VL under the master branch (with transformers==4.54.1)
• Fixed incorrect key_renaming_mapping, which caused weights to be improperly loaded.

Adds # (feature)

• Updated Qwen2.5VL to support transformers version 4.54.1.

Before submitting

• This PR fixes a typo or improves the docs (you can dismiss the other checks if that's the case).
• Did you read the contributor guideline?
• Did you make sure to update the documentation with your changes? E.g. record bug fixes or new features in What's New. Here are the
  documentation guidelines
• Did you build and run the code without any errors?
• Did you report the running environment (NPU type/MS version) and performance in the doc? (better record it for data loading, model inference, or training tasks)
• Did you write any new necessary tests?

Who can review?

Anyone in the community is free to review the PR once the tests have passed. Feel free to tag
members/contributors who may be interested in your PR.

@xxx

[gemini-code-assist]

Summary of Changes

Hello @zhtmike, I'm Gemini Code Assist¹! I'm currently reviewing this pull request and will post my feedback shortly. In the meantime, here's a summary to help you and other reviewers quickly get up to speed!

This pull request delivers a crucial update to the Qwen2.5VL model, bringing it in line with transformer version 4.54.1. The changes address critical bugs related to model output and weight loading, ensuring the model functions correctly and efficiently. Furthermore, the internal architecture has been modernized with refactored attention layers and a more streamlined approach to handling multimodal data, which enhances the model's overall stability and performance.

Highlights

• Qwen2.5VL Update: The Qwen2.5VL model has been updated to version 4.54.1, incorporating the latest features and fixes from the transformer library.
• Incorrect Output Fix: Resolved an issue that caused incorrect output from Qwen2.5VL when running under the master branch with transformer version 4.54.1.
• Key Renaming Mapping Correction: Fixed an incorrect key_renaming_mapping that previously prevented proper loading of model weights by broadening the scope of keys considered for renaming.
• Attention Mechanism Refactoring: The attention mechanisms within the Qwen2.5VL vision and text components have been significantly refactored, introducing unified eager attention and improved Flash Attention 2 handling.
• Multimodal Input Processing Enhancements: Improvements were made to how multimodal inputs (images and videos) are processed, including more robust handling of placeholder tokens and dynamic calculation of RoPE indices during generation.

Using Gemini Code Assist

The full guide for Gemini Code Assist can be found on our documentation page, here are some quick tips.

Invoking Gemini

You can request assistance from Gemini at any point by creating a comment using either /gemini <command> or @gemini-code-assist <command>. Below is a summary of the supported commands on the current page.

Feature	Command	Description
Code Review	/gemini review	Performs a code review for the current pull request in its current state.
Pull Request Summary	/gemini summary	Provides a summary of the current pull request in its current state.
Comment	@gemini-code-assist	Responds in comments when explicitly tagged, both in pull request comments and review comments.
Help	/gemini help	Displays a list of available commands.

Customization

To customize Gemini Code Assist for GitHub experience, repository maintainers can create a configuration file and/or provide a custom code review style guide (such as PEP-8 for Python) by creating and adding files to a .gemini/ folder in the base of the repository. Detailed instructions can be found here.

Limitations & Feedback

Gemini Code Assist may make mistakes. Please leave feedback on any instances where its feedback is incorrect or counter productive. You can react with 👍 and 👎 on @gemini-code-assist comments. If you're interested in giving your feedback about your experience with Gemini Code Assist for Github and other Google products, sign up here.

You can also get AI-powered code generation, chat, as well as code reviews directly in the IDE at no cost with the Gemini Code Assist IDE Extension.

Footnotes

1. Review the Privacy Notices, Generative AI Prohibited Use Policy, Terms of Service, and learn how to configure Gemini Code Assist in GitHub here. Gemini can make mistakes, so double check it and use code with caution. ↩

[gemini-code-assist]

Code Review

This pull request updates the Qwen2.5VL model to align with transformers==4.54.1, fixing issues with incorrect output and weight loading. The changes involve a significant refactoring of the model implementation, particularly in the attention mechanisms and cache handling, to use more modular and centralized utilities. This is a great improvement for code clarity and maintainability. I've found a critical issue related to attention implementation dispatching that could lead to a runtime error, and a couple of medium-severity issues regarding an unused function argument and an incorrect type hint. Overall, the changes are very positive, and with these fixes, the implementation will be much more robust.

[zhanghuiyao]

这里是为什么去掉了 layernorm.gamma 的判断，是之前写多了吗？

[zhtmike]

目前这个判断会导致qwen25vl 权重读取失败。去掉这个判断可以正确读取权重，具体修复待 @wcrzlh 分析