init flux2 dit on meta device

[qzzz95]

No description provided.

[gemini-code-assist]

Summary of Changes

Hello @qzzz95, 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 enhances the stability and efficiency of Flux2 model initialization by ensuring proper meta device usage during model construction and by implementing more resilient handling of state dictionary keys. These changes contribute to a smoother and more error-resistant model loading workflow.

Highlights

• Meta Device Initialization: The Flux2DiT model initialization now explicitly uses torch.device("meta") as a context manager when device="meta" is specified. This ensures that the model's parameters are created on the meta device, preventing unintended memory allocation during the initial setup phase.
• Robust State Dictionary Handling: A conditional check has been added before attempting to remove the "lm_head.weight" key from the encoder's state dictionary. This prevents a KeyError if the key is not present, making the state dictionary loading process more robust.

🧠 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.

Changelog

• diffsynth_engine/models/flux2/flux2_dit.py
  • Wrapped cls(device="meta", ...) call within a with torch.device("meta") block to correctly initialize model parameters on the meta device.
• diffsynth_engine/pipelines/flux2_klein_image.py
  • Added an if "lm_head.weight" in state_dicts.encoder: check before calling state_dicts.encoder.pop("lm_head.weight") to prevent KeyError.

Activity

• The pull request was opened by qzzz95.

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 two small but important changes. First, it ensures that the Flux2DiT model is initialized on the 'meta' device by using a torch.device("meta") context manager. This is a good practice for memory efficiency when loading large models. Second, it adds a check to ensure lm_head.weight exists before attempting to remove it from the state dictionary, which makes the model loading process more robust. I have one minor suggestion to make the code slightly more concise.

[gemini-code-assist]

For conciseness, you can use pop with a default value to avoid the if check. This is a more idiomatic way to handle optional keys in a dictionary.

                state_dicts.encoder.pop("lm_head.weight", None)