fix compute/communication overlap for gloo #240
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
tushar00jain
force-pushed
the
pr240
branch
4 times, most recently
from
dbec11e to
5228ee8
Compare
tushar00jain
changed the title
tushar00jain
changed the title
tushar00jain
changed the title
tushar00jain
force-pushed
the
pr240
branch
14 times, most recently
from
c93ad11 to
bfb92ff
Compare
d4l3k
reviewed
Jul 25, 2025
torchft/manager.py
Outdated
| @@ -411,7 +412,7 @@ def callback( | |||
| fut = fut.then(callback) | |||
|
|
|||
| fut = self.wrap_future(fut, tensor) | |||
| return fut | |||
| return (work, fut) | |||
There was a problem hiding this comment.
this makes me a bit nervous since calling only work.wait() means that the code in the future callback may have not run -- i.e tensor /= num_participants may execute out of order
The advantage of future objects here is that the work runs on the completing thread, where as .wait() runs on the waiting thread. I'm wondering if we should actually wrap these futures back into Work objects so we can do the stream dependency correctly in .wait()
tushar00jain
force-pushed
the
pr240
branch
3 times, most recently
from
922a415 to
fa35a6d
Compare
tushar00jain
force-pushed
the
pr240
branch
4 times, most recently
from
2b43bfc to
dc7b2e0
Compare
Summary: use http transport instead of pg transport -- pg transport fails to resolve address when running locally
tushar00jain
changed the title
This was referenced Jul 26, 2025
tushar00jain
force-pushed
the
pr240
branch
3 times, most recently
from
4964b72 to
15b0cb0
Compare
tushar00jain
force-pushed
the
pr240
branch
3 times, most recently
from
405dc6e to
91207a2
Compare
Summary: deep copy the state dict for sending checkpoint because if the replica moves to the next step, the state dict can change before the checkpoint is sent
Summary: - call future.wait in callbacks to make sure the continuation executes after the future has completed - set the stream correctly to execute callback scheduled by bucketized allreduce
Summary: returns the work object so we can be more flexible with the usage
Summary: - we current wait for pg work's future when preparing for a fragment - if we use gloo, this blocks the cpu - move the wait call to when we perform the actual sync of the fragment - since we still call `work.wait()` in the allreduce call itself this doesn't completely fix the problem
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.
Summary:
work.wait()in the allreduce call itself this doesn't completely fix the problemStack created with Sapling. Best reviewed with ReviewStack.