-
Notifications
You must be signed in to change notification settings - Fork 338
[ROCm] Support building PyTorch with GLOO on Windows #461
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
Conversation
|
@jammm sorry if I’m being impatient, but since this PR was approved, could it be merged? |
It's not up to me to merge this unfortunately. The gloo maintainers need to merge it. It's my first PR here actually, so not sure of the exact flow. |
Oh, I didn’t know that. Np, then we’ll wait for them. |
d4l3k
left a comment
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.
LGTM, sorry for the delay!
If you're on the PyTorch slack feel free to ping me directly there for faster reviews
We're working on contributing Windows ROCm CI builds with pytorch/pytorch#159520 and could enable |
|
@ScottTodd that sounds reasonable to me but we should also add something to this repo. PyTorch pins a commit from Gloo so it's always trailing. We don't run PT CI on Gloo repo, only run PT CI when bumping the Gloo pinned commit |
|
Eyy good to see things evolving here |
Helps fix imports for
transformerslib which seems to import torch.distributed. ROCm/TheRock#1202It may or may not help in proper WIndows support for torch.distributed (since RCCL isn't on Windows atm), but it does let us build pytorch w/ USE_DISTRIBUTED=ON, which helps fix import errors like the above, so users on their single GPU systems can run comfyui without facing import errors.
Tested on Windows w/ PyTorch main branch via. TheRock.