-
Notifications
You must be signed in to change notification settings - Fork 4.3k
Fix pre-compile on cpu-only machines #7168
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.
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
base: master
Are you sure you want to change the base?
Conversation
AlongWY
commented
Mar 22, 2025
- Fix pre-compile on cpu-only machines
@@ -242,6 +242,8 @@ def communication_backend_name(self): | |||
return self._communication_backend_name | |||
|
|||
def is_triton_supported(self): | |||
if not torch.cuda.is_available(): |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
if not torch.cuda.is_available(): | |
if not self.is_available(): |
@AlongWY, thanks for the PR. Can you please share the motivation to help us understand your use case? |
I used GitHub Actions to precompile DeepSpeed, but after upgrading to v0.16.4, the precompilation failed. The repository is here: https://github.com/AlongWY/deepspeed_wheels And the compile result here show that tranformer inference is not support cuda 12.6, which is also shown in #6709. |
@AlongWY - just so I follow, you're compiling in a GitHub action that has cuda installed but no GPU (so that's why you are hitting the CPU only path here) but with DS_ACCELERATOR=cuda? Just wanted to make sure I understood the repo correctly? Also the error you shared is only on Windows (that we've observed) but your action appears to be running on Ubuntu, is that right? |
Yes, the GitHub Action is compiling in an Ubuntu environment with CUDA installed but without a GPU. |
Thanks @AlongWY - can you share the error that you were seeing in this environment? Also if you have triton installed in the environment or not? |