fixed amd multimodal CI failures caused by refactor in #15812 #15813#16287
fixed amd multimodal CI failures caused by refactor in #15812 #15813#16287Kangyan-Zhou merged 3 commits intosgl-project:mainfrom
Conversation
Summary of ChangesHello @sunxxuns, 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 serves as a dedicated test to validate the functionality of the multimodal AMD CI pipeline. By introducing a minimal change—adding a module docstring to a specific Python initialization file—the author intends to activate and monitor the relevant CI jobs. The primary goal is to ensure that the AMD-specific tests are triggered and executed correctly, providing a clear indication of the CI system's health without introducing any functional code changes to be merged. Highlights
🧠 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 AssistThe 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
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 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
|
![gemini-code-assist[bot]](https://avatars.githubusercontent.com/in/956858?s=60&v=4){kind=small}
There was a problem hiding this comment.
Code Review
This pull request adds a module-level docstring to python/sglang/multimodal_gen/__init__.py to trigger CI tests. The change is straightforward and serves its purpose. While this is a test PR that is not intended for merging, I've provided a suggestion to improve the docstring to be more descriptive and align with Python's best practices (PEP 257). A more detailed docstring improves code maintainability by clearly documenting the package's public API, which could be valuable for future development.
| @@ -1,4 +1,6 @@ | |||
| # Copied and adapted from: https://github.com/hao-ai-lab/FastVideo | |||
| """SGLang Multimodal Generation module for diffusion models.""" | |||
There was a problem hiding this comment.
While the added docstring provides a basic summary, it could be more descriptive to improve maintainability and align better with Python's docstring conventions (PEP 257). A good package docstring should briefly explain its purpose and list the main components it exposes. This helps other developers quickly understand the package's public API.
| """SGLang Multimodal Generation module for diffusion models.""" | |
| """SGLang Multimodal Generation package for diffusion models. | |
| This package provides the core components for multimodal generation, including: | |
| - `DiffGenerator`: The main entry point for generating images/videos. | |
| - `PipelineConfig`: Configuration for diffusion pipelines. | |
| - `SamplingParams`: Parameters for controlling the sampling process. | |
| """ |
References
- According to PEP 257, the docstring for a package (the
__init__.pyfile) should summarize its contents and list exported modules/subpackages. In this case, where key classes are exposed as the public API, listing them in the docstring is a valuable practice that follows the spirit of the guideline to document the package's interface. (link)
sunxxuns
force-pushed
the
test-mm-amd-ci
branch
23 times, most recently
from
2d2e71d to
bcc8792
Compare
sunxxuns
force-pushed
the
test-mm-amd-ci
branch
21 times, most recently
from
267ec31 to
deb59a2
Compare
- Fix CLIP attention on ROCm: Use is_causal=True alone without attn_mask. Using both together causes NaN on ROCm. CUDA path unchanged. - Add Triton fallback for FlashInfer RoPE with warning - Add RMSNorm fallback for QK-norm with warning - Install tvm-ffi in CI for JIT kernel support
… dependencies in `pyproject_other.toml` to match the main `pyproject.toml`. **Summary of the fix:** The `qwen_image_t2i_cache_dit_enabled` test requires the `cache-dit` package. While the main `pyproject.toml` already had `cache-dit==1.1.8` in its `diffusion` dependencies, the `pyproject_other.toml` (used for AMD/HIP builds) was missing it. This caused the test to fail on AMD CI unless the workaround in the CI script successfully installed it. **Changes made:** - Added `"cache-dit==1.1.8"` to the `diffusion` optional dependencies in `python/pyproject_other.toml` This ensures that `cache-dit` is properly installed when users run `pip install sglang[diffusion]` on AMD platforms, making the `qwen_image_t2i_cache_dit_enabled` test work correctly.
Qiaolin-Yu
left a comment
Qiaolin-Yu
left a comment
There was a problem hiding this comment.
Could you please fix the lint? Thanks!
4 participants
fixed the failure initially caused by #15812 , which was merged without passing mm CI on AMD (pls avoide this future on)
1 fixed flashinfer call, since it's only supporting nvidia gpu now, should not be on the main path for other hardwares
2 attn_mask && is_causal doesn't work for sdpa/rocm, flux actually didn't use attn_mask anyway; fixed it for amd path and added a warning
also 2-gpu is broken by #15813
for lora weight load path.