diffusion: Add Configurable Generator Device and Seed Support via API by niehen6174 · Pull Request #14366 · sgl-project/sglang

niehen6174 · 2025-12-03T14:54:16Z

Motivation

Previously, the random generator in the diffusion pipeline was hardcoded to use CPU device (torch.Generator("cpu")), which limited flexibility and performance. Additionally, there was no way to set the random seed through the server API, making it difficult to reproduce results or control randomness in generation tasks.

This PR addresses these limitations by:

Making generator device configurable: Allow users to choose between CUDA and CPU for the random generator, with CUDA as the default device. (When using CUDA in the pipe, diffusers by default use CUDA as the device for random generator.)
Adding seed parameter support via API: Enable setting random seeds through the OpenAI-compatible API endpoints for both image and video generation

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.
Work with maintainers to merge your PR. See the PR Merge Process

gemini-code-assist · 2025-12-03T14:54:36Z

Summary of Changes

Hello @niehen6174, 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 significantly improves the control and reproducibility of the diffusion pipeline by introducing the ability to specify the device for the random number generator and to pass a random seed via the API. Previously, the generator was hardcoded to CPU, and seed control was unavailable through the API. These changes provide users with greater flexibility to optimize performance and ensure consistent generation outcomes for both image and video tasks.

Highlights

Configurable Generator Device: The random generator in the diffusion pipeline can now be configured to use either CUDA or CPU, with CUDA as the default, enhancing flexibility and performance.
API Seed Support: Users can now set random seeds through the OpenAI-compatible API endpoints for both image and video generation, allowing for reproducible results and controlled randomness.

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.

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 support for a configurable generator device and seed via the API, which is a great enhancement for reproducibility and performance tuning. The changes are well-implemented across the different layers, from the API endpoints to the core pipeline logic. I have one suggestion in input_validation.py to simplify the device selection logic and add a warning when falling back to CPU, which will improve code clarity and user experience.

gemini-code-assist · 2025-12-03T14:56:28Z

+        generator_device = batch.generator_device
+
+        # Determine the device for generator
+        if generator_device == "cpu":
+            device_str = "cpu"
+        else:
+            # Use cuda if available, otherwise fallback to cpu
+            device_str = "cuda" if torch.cuda.is_available() else "cpu"


The logic to determine the generator device can be simplified for better readability. Additionally, adding a warning when CUDA is requested but unavailable, causing a fallback to CPU, would provide valuable feedback to the user about the execution environment.

Suggested change

generator_device = batch.generator_device

# Determine the device for generator

if generator_device == "cpu":

device_str = "cpu"

else:

# Use cuda if available, otherwise fallback to cpu

device_str = "cuda" if torch.cuda.is_available() else "cpu"

# Determine the device for the generator, with a fallback to CPU if CUDA is not available.

device_str = batch.generator_device

if device_str == "cuda" and not torch.cuda.is_available():

logger.warning(

"CUDA device requested for generator, but it is not available. Falling back to CPU."

)

device_str = "cpu"

mickqian · 2025-12-03T15:07:55Z

Some comments could be cleaned

niehen6174 · 2025-12-03T15:24:26Z

Some comments could be cleaned

Done.

mickqian · 2025-12-03T15:40:29Z

/tag-and-rerun-ci

…via API (sgl-project#14366) Co-authored-by: niehen6174 <niehen.6174@gmail.com> Co-authored-by: Mick <mickjagger19@icloud.com>

niehen6174 added 2 commits December 3, 2025 22:42

Add configurable generator device and seed Support via API

8483a4d

fix: lint error

763046e

niehen6174 requested review from mickqian and yhyang201 as code owners December 3, 2025 14:54

github-actions Bot added npu diffusion SGLang Diffusion labels Dec 3, 2025

gemini-code-assist Bot reviewed Dec 3, 2025

View reviewed changes

mickqian approved these changes Dec 3, 2025

View reviewed changes

remove some comments

e960d13

github-actions Bot added the run-ci label Dec 3, 2025

niehen6174 and others added 2 commits December 4, 2025 21:50

fix: torch.randn use generator divice error

cc5bcf6

Merge branch 'main' into suppor_generator_device

41e79fd

mickqian merged commit 788628b into sgl-project:main Dec 4, 2025
47 checks passed

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

diffusion: Add Configurable Generator Device and Seed Support via API#14366

diffusion: Add Configurable Generator Device and Seed Support via API#14366
mickqian merged 5 commits intosgl-project:mainfrom
niehen6174:suppor_generator_device

niehen6174 commented Dec 3, 2025

Uh oh!

gemini-code-assist Bot commented Dec 3, 2025

Uh oh!

gemini-code-assist Bot left a comment

Uh oh!

gemini-code-assist Bot Dec 3, 2025

Uh oh!

mickqian commented Dec 3, 2025

Uh oh!

niehen6174 commented Dec 3, 2025

Uh oh!

mickqian commented Dec 3, 2025

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

Conversation

niehen6174 commented Dec 3, 2025

Motivation

Modifications

Accuracy Tests

Benchmarking and Profiling

Checklist

Uh oh!

gemini-code-assist Bot commented Dec 3, 2025

Summary of Changes

Highlights

Footnotes

Uh oh!

gemini-code-assist Bot left a comment

Choose a reason for hiding this comment

Code Review

Uh oh!

gemini-code-assist Bot Dec 3, 2025

Choose a reason for hiding this comment

Uh oh!

mickqian commented Dec 3, 2025

Uh oh!

niehen6174 commented Dec 3, 2025

Uh oh!

mickqian commented Dec 3, 2025

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants