[Diffusion] Apply jit qk_norm to flux1

[BBuf]

Motivation

sglang generate --model-path=black-forest-labs/FLUX.1-dev --prompt="A cat holding a sign that says hello world" --width=1024 --height=1024 --num-inference-steps=50 --guidance-scale=4.0 --seed=42 --save-output --warmup=True --ulysses-degree=2 --num-gpus=2

main:

100%|██████████████████████████████████████████████████████████████████████████████████████████████████| 50/50 [00:05<00:00, 8.78it/s]
[01-18 06:41:16] [DenoisingStage] average time per step: 0.1139 seconds
[01-18 06:41:16] [DenoisingStage] finished in 5.6961 seconds

pr:

100%|████████████████████████████████████████████████████████████████| 50/50 [00:05<00:00, 9.26it/s]
[01-18 10:33:49] [DenoisingStage] average time per step: 0.1080 seconds
[01-18 10:33:49] [DenoisingStage] finished in 5.4039 seconds

About 5% e2d performace improve.

Modifications

Accuracy Tests

Benchmarking and Profiling

Checklist

• Format your code according to the Format code with pre-commit.
• Add unit tests according to the Run and add unit tests.
• Update documentation according to Write documentations.
• Provide accuracy and speed benchmark results according to Test the accuracy and Benchmark the speed.
• Follow the SGLang code style guidance.

Review Process

1. Ping Merge Oncalls to start the PR flow. See the PR Merge Process.
2. Get approvals from CODEOWNERS and other reviewers.
3. Trigger CI tests with comments or contact authorized users to do so.
   • /tag-run-ci-label, /rerun-failed-ci, /tag-and-rerun-ci
4. After green CI and required approvals, ask Merge Oncalls to merge.

[gemini-code-assist]

Summary of Changes

Hello @BBuf, 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 focuses on enhancing the computational efficiency of the FLUX.1 diffusion model by introducing a JIT-compiled fused QK normalization routine. The changes strategically apply this optimized normalization to query and key tensors within the attention mechanism, leading to a measurable performance boost.

Highlights

• Performance Optimization: Implements JIT-compiled fused QK normalization to improve the end-to-end performance of the FLUX.1 diffusion model by approximately 5%.
• Conditional Application: Introduces conditional logic within the flux.py model to apply an optimized apply_qk_norm function for query and key tensors when running on CUDA, with matching variance epsilon, and when fused inplace QK normalization is supported.
• Code Integration: Integrates new imports (can_use_fused_inplace_qknorm, apply_qk_norm) and modifies the forward method in USPAttention to leverage these optimizations for both primary and added encoder query/key pairs.

🧠 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 a performance optimization by applying a JIT-compiled fused kernel for QK normalization in the FluxAttention module. The changes correctly add conditional logic to use the fused apply_qk_norm function when running on CUDA with compatible configurations, and fall back to the standard normalization otherwise. My main feedback is to refactor the duplicated normalization logic to improve code maintainability.

[mickqian]

/tag-and-rerun-ci

[BBuf]

/tag-and-rerun-ci