Skip to content

feat: Update build.py to skip libnvshmem3-cuda-13 for cpu only build. #8528

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
Sunidhi-Gaonkar1 wants to merge 1 commit into
base: main
Choose a base branch
from
Open

feat: Update build.py to skip libnvshmem3-cuda-13 for cpu only build. #8528

Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
5 changes: 3 additions & 2 deletions build.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1509,8 +1509,9 @@ def dockerfile_prepare_container_linux(argmap, backends, enable_gpu, target_mach
"""

if target_platform() not in ["igpu", "windows", "rhel"]:
repo_arch = "sbsa" if target_machine == "aarch64" else "x86_64"
df += f"""
if FLAGS.triton_wheels_dependencies_group != "cpu" :
repo_arch = "sbsa" if target_machine == "aarch64" else "x86_64"
df += f"""
Copy link
Contributor

@mc-nv mc-nv Dec 2, 2025

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

The reason behind this installation is that PyTorch backend is looking for that library, and PyTorch is part of the CPU build process.

In the same time, using flag --triton-wheels-dependencies-group is not applicable here that flag is designed for different purpose, it's for wheel dependency group options.

Copy link

@alhad-deshpande alhad-deshpande Dec 3, 2025 •
edited
Loading

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

@mc-nv
Ok. Thanks for your review. We will change the code as per your suggestion and were thinking to introduce cuda specific flag 'disable_cuda' to disable this library. The default value be false. The same cuda flag 'disable_cuda' can be used in pytorch_backend as well to disable all cuda related libraries like libtorch_cuda.so, libcublasLt.so etc.
Please do let me know if this approach sounds ok.

RUN curl -o /tmp/cuda-keyring.deb \\
https://developer.download.nvidia.com/compute/cuda/repos/ubuntu2404/{repo_arch}/cuda-keyring_1.1-1_all.deb \\
&& apt install /tmp/cuda-keyring.deb \\
Expand Down