feat: update blackwell setup #7119
Conversation
There was a problem hiding this comment.
Summary of Changes
Hello @zhyncs, 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 updating the build and dependency setup specifically for the Blackwell architecture. It involves bumping key library versions like sgl-kernel and nvidia-nccl and adjusting the kernel build configuration to use a Blackwell-optimized DeepGEMM source when targeting CUDA 12.8.
Highlights
- Dependency Updates: Updated the
sgl_kernelversion to0.1.8andnvidia-nccl-cu12to2.27.3within the Blackwell-specific Dockerfile. - Conditional DeepGEMM Source: Modified the
sgl-kernelCMake build process to conditionally fetch thesgl-project/DeepGEMMfork with theblackwelltag when building with CUDA version 12.8, otherwise using the defaultdeepseek-ai/DeepGEMM. - Kernel Version Bump: Incremented the
sgl-kernelversion to0.1.8across all relevantpyproject.tomlfiles (standard, cpu, rocm) and the internal version file.
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 in your pull request via creating an issue comment (i.e. comment on the pull request page) using either /gemini <command> or @gemini-code-assist <command>. Below is a summary of the supported commands.
| 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 issue 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 is currently in preview and 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 to provide feedback.
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
-
Review the Privacy Notices, Generative AI Prohibited Use Policy, Terms of Service, and learn how to configureGemini Code Assist in GitHub here. Gemini can make mistakes, so double check it and use code with caution. ↩
There was a problem hiding this comment.
Code Review
This pull request updates the setup for Blackwell support, primarily by bumping dependency versions in the Dockerfile and introducing conditional logic in the CMake build to fetch a specific DeepGEMM version for CUDA 12.8. The version bumps across the sgl-kernel Python project files are consistent. It's important to ensure the compatibility and stability of the updated dependencies, especially the conditional DeepGEMM source and the new NCCL version. The checklist mentions adding unit tests, which is crucial for these changes, but no new test files were included in the diff.
| if("${CUDA_VERSION}" VERSION_EQUAL "12.8") | ||
| set(DeepGEMM_REPO "https://github.com/sgl-project/DeepGEMM") | ||
| set(DeepGEMM_TAG "blackwell") | ||
| else() | ||
| set(DeepGEMM_REPO "https://github.com/deepseek-ai/DeepGEMM") | ||
| set(DeepGEMM_TAG "8dfa3298274bfe6b242f6f8a3e6f3eff2707dd9f") | ||
| endif() |
There was a problem hiding this comment.
This conditional logic fetches DeepGEMM from different repositories/tags based on the CUDA version. For CUDA 12.8, it uses sgl-project/DeepGEMM with the blackwell tag. For other versions, it uses deepseek-ai/DeepGEMM with a specific commit hash.
Could you clarify the rationale for using a different DeepGEMM source/tag specifically for CUDA 12.8? Is the blackwell tag in the sgl-project fork stable and intended for production use, or is it a development branch? It might be safer to use a specific commit hash for the blackwell branch as well, to ensure reproducible builds.
| && pip3 install setuptools==75.0.0 wheel scikit-build-core --break-system-packages | ||
|
|
||
| RUN git clone --depth=1 https://github.com/sgl-project/sglang.git \ | ||
| && cd sglang && pip3 install -e "python[blackwell]" --break-system-packages | ||
|
|
||
| RUN pip3 install nvidia-nccl-cu12==2.26.2.post1 --force-reinstall --no-deps --break-system-packages | ||
| RUN pip3 install nvidia-nccl-cu12==2.27.3 --force-reinstall --no-deps --break-system-packages |
There was a problem hiding this comment.
Updating nvidia-nccl-cu12 to version 2.27.3. Please confirm if this specific version is required for Blackwell support or if it's a general update. Given this is a core communication library, it's important to ensure compatibility and performance. Have tests been run with this new version?
Motivation
Modifications
Checklist