Impl of apply_rotary_emb_qwen_naive and apply_wan_rope_with_torch_naive

[yeahdongcn]

This PR adds apply_rotary_emb_qwen_naive for qwen-image and apply_wan_rope_with_torch_naive for wan. One can use rope_type = torch_naive to select it.

Testing Done

root@worker3218:/ws/scripts/platforms/mthreads_musa# ./run_wan21_t2v.sh 
root@worker3218:/ws/scripts/platforms/mthreads_musa# ./qwen_image_i2i_2511.sh

[gemini-code-assist]

Summary of Changes

Hello @yeahdongcn, 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 integrates new 'naive' PyTorch implementations for Rotary Positional Embeddings (RoPE) into the Qwen-image and WAN models. This provides an alternative, potentially more straightforward or debuggable, method for applying positional embeddings, which can be activated through a configuration setting. The changes include adding the new RoPE functions and updating the model's initialization logic to dynamically select the desired RoPE implementation.

Highlights

• New Naive Rotary Embedding Implementations: Introduced apply_rotary_emb_qwen_naive for Qwen-image models and apply_wan_rope_with_torch_naive for WAN models, providing new PyTorch-based naive implementations for Rotary Positional Embeddings (RoPE).
• Configurable RoPE Type: Updated configuration files (qwen_image_i2i_2511.json and wan_t2v.json) to allow selection of the new naive RoPE implementations by setting rope_type to torch_naive.
• Refactored RoPE Function Selection: The logic for selecting the appropriate rotary embedding function in transformer_infer.py for both Qwen and WAN models has been refactored to use a dictionary, making it more extensible and cleaner to add new RoPE types.

🧠 New Feature in Public Preview: You can now enable Memory to help Gemini Code Assist learn from your team's feedback. This makes future code reviews more consistent and personalized to your project's style. Click here to enable Memory in your admin console.

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 introduces naive PyTorch implementations for rotary position embeddings (apply_rotary_emb_qwen_naive and apply_wan_rope_with_torch_naive) and integrates them into the model configuration. The changes also include refactoring the selection logic for RoPE functions, which is a good improvement.

My review has identified a few areas for improvement:

• A potentially risky change in .gitignore that could lead to large files being committed.
• Code duplication in both of the new naive RoPE implementations. I've suggested refactoring them to improve maintainability.

Overall, the core logic seems correct, and the changes make the RoPE implementation selectable, which is a great feature.