process_group/gloo: support CUDA tensors #185
Merged
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
facebook-github-bot
added
the
CLA Signed
fegin
approved these changes
May 2, 2025
H-Huang
approved these changes
May 2, 2025
Contributor
H-Huang
left a comment
H-Huang
left a comment
There was a problem hiding this comment.
cool! I also wonder if it would be possible to use torchtitan for cpu-only training. That could be cool to show a demo on a single host
Contributor
|
@H-Huang ye, I'm also wondering if we can do CPU training. But I'm not sure if FSDP can work with CPU memory. |
Member
Author
|
I've only tested this with HSDP using NCCL for most of titan and only Gloo+CUDA for the allreduce Gloo does CPU blocking so I expect that we'll have some significant performance hit since this won't overlap compute+networking. I have some thoughts on how to work around that but tbd |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
This makes torchft's ProcessGroupGloo correctly support CUDA tensors so we can use Gloo with torchtitan for allreduces.
I also found a bug in Gloo, apparently torchft's test suite is more in depth than the one in core haha pytorch/pytorch#152645
Test plan:
torchtitan on remote cluster