[diffusion] hardware: support diffusion models on MTGPU (multi-GPU, 5/N)#17318
Conversation
Summary of ChangesHello @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 initiates support for Moore Threads (MTGPU) GPUs within the diffusion framework. It integrates the MUSA (Meta-computing Unified System Architecture) backend by adapting the NCCL communication library wrapper to recognize and utilize MCCL. The changes ensure that the system can correctly identify and interact with Moore Threads' collective communication library, laying the groundwork for accelerated LLM inference on their hardware. 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 support for Moore Threads (MTGPU) GPUs by integrating the MUSA backend. The changes primarily involve detecting the MUSA environment to load the appropriate MCCL library (libmccl.so.2) and dynamically mapping NCCL function calls to their MCCL equivalents. The implementation is clean and follows existing patterns in the codebase. I've suggested one improvement to make the MCCL library detection more robust. Overall, this is a good contribution towards expanding hardware support.
python/sglang/multimodal_gen/runtime/distributed/device_communicators/pynccl_wrapper.py
Outdated
Show resolved
Hide resolved
yeahdongcn
changed the title
yeahdongcn
force-pushed
the
xd/diffusion_multi_gpus
branch
from
b02eb5b to
c9a9d4a
Compare
|
Rebased onto |
yeahdongcn
force-pushed
the
xd/diffusion_multi_gpus
branch
from
c9a9d4a to
b5317b5
Compare
|
Rebased onto |
Signed-off-by: Xiaodong Ye <xiaodong.ye@mthreads.com>
yeahdongcn
force-pushed
the
xd/diffusion_multi_gpus
branch
from
b5317b5 to
7605599
Compare
…/N) (sgl-project#17318) Signed-off-by: Xiaodong Ye <xiaodong.ye@mthreads.com>
…/N) (sgl-project#17318) Signed-off-by: Xiaodong Ye <xiaodong.ye@mthreads.com>
…/N) (sgl-project#17318) Signed-off-by: Xiaodong Ye <xiaodong.ye@mthreads.com>
…/N) (sgl-project#17318) Signed-off-by: Xiaodong Ye <xiaodong.ye@mthreads.com>
2 participants
Motivation
This PR is the 5th in a series of pull requests (tracked in #16565) to add full support for Moore Threads GPUs, leveraging MUSA (Meta-computing Unified System Architecture) to accelerate LLM inference.
Modifications
python/sglang/multimodal_gen/runtime/distributed/device_communicators/pynccl_wrapper.py:python/sglang/multimodal_gen/utils.py:find_nccl_library()to returnlibmccl.so.2Testing Done
Tested in a clean torch_musa container.
Verified that the video is generated correctly and here is the full log:
Accuracy Tests
Benchmarking and Profiling
Checklist
Review Process
/tag-run-ci-label,/rerun-failed-ci,/tag-and-rerun-ci