2/N: CPU Comms Module #3424
Open
2/N: CPU Comms Module #3424
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
meta-cla
bot
added
the
CLA Signed
|
@jeffkbkim has exported this pull request. If you are a Meta employee, you can view the originating Diff in D83773528. |
jeffkbkim
force-pushed
the
export-D83773528
branch
from
f526de0 to
1bd5479
Compare
jeffkbkim
added a commit
to jeffkbkim/torchrec
that referenced
this pull request
Oct 2, 2025
Summary: This diff adds the basic building blocks for a zero overhead RecMetrics implementation. Follow up patches will contain integration with users of torchrec. One of the main pain points of using RecMetricModule is that metric updates and computes are done synchronously. In training jobs, there has been cases where metric updates take +20% of a training iteration. Metric computations, although less frequent, can takes over a couple of seconds. CPUOffloadedRecMetricModule aims to perform all metric updates/computes asynchronously, completely removing them from the critical path. This patch adds: - CPUCommsRecMetricModule: Submodule that all gathers, loads, and computes aggregated metric state tensors across ranks. Differential Revision: D83773528
jeffkbkim
force-pushed
the
export-D83773528
branch
from
1bd5479 to
7a43096
Compare
jeffkbkim
added a commit
to jeffkbkim/torchrec
that referenced
this pull request
Oct 3, 2025
Summary: This diff adds the basic building blocks for a zero overhead RecMetrics implementation. Follow up patches will contain integration with users of torchrec. One of the main pain points of using RecMetricModule is that metric updates and computes are done synchronously. In training jobs, there has been cases where metric updates take +20% of a training iteration. Metric computations, although less frequent, can takes over a couple of seconds. CPUOffloadedRecMetricModule aims to perform all metric updates/computes asynchronously, completely removing them from the critical path. This patch adds: - CPUCommsRecMetricModule: Submodule that all gathers, loads, and computes aggregated metric state tensors across ranks. Differential Revision: D83773528
Summary: This diff adds the basic building blocks for a zero overhead RecMetrics implementation. Follow up patches will contain integration with users of torchrec. One of the main pain points of using RecMetricModule is that metric updates and computes are done synchronously. In training jobs, there has been cases where metric updates take +20% of a training iteration. Metric computations, although less frequent, can takes over a couple of seconds. CPUOffloadedRecMetricModule aims to perform all metric updates/computes asynchronously, completely removing them from the critical path. This patch adds: - CPUCommsRecMetricModule: Submodule that all gathers, loads, and computes aggregated metric state tensors across ranks. Differential Revision: D83773528
jeffkbkim
force-pushed
the
export-D83773528
branch
from
7a43096 to
ed22a14
Compare
jeffkbkim
added a commit
to jeffkbkim/torchrec
that referenced
this pull request
Oct 3, 2025
Summary: This diff adds the basic building blocks for a zero overhead RecMetrics implementation. Follow up patches will contain integration with users of torchrec. One of the main pain points of using RecMetricModule is that metric updates and computes are done synchronously. In training jobs, there has been cases where metric updates take +20% of a training iteration. Metric computations, although less frequent, can takes over a couple of seconds. CPUOffloadedRecMetricModule aims to perform all metric updates/computes asynchronously, completely removing them from the critical path. This patch adds: - CPUCommsRecMetricModule: Submodule that all gathers, loads, and computes aggregated metric state tensors across ranks. Differential Revision: D83773528
jeffkbkim
added a commit
to jeffkbkim/torchrec
that referenced
this pull request
Oct 3, 2025
Summary: This diff adds the basic building blocks for a zero overhead RecMetrics implementation. Follow up patches will contain integration with users of torchrec. One of the main pain points of using RecMetricModule is that metric updates and computes are done synchronously. In training jobs, there has been cases where metric updates take +20% of a training iteration. Metric computations, although less frequent, can takes over a couple of seconds. CPUOffloadedRecMetricModule aims to perform all metric updates/computes asynchronously, completely removing them from the critical path. This patch adds: - CPUCommsRecMetricModule: Submodule that all gathers, loads, and computes aggregated metric state tensors across ranks. Differential Revision: D83773528
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:
This diff adds the basic building blocks for a zero overhead RecMetrics implementation. Follow up patches will contain integration with users of torchrec.
One of the main pain points of using RecMetricModule is that metric updates and computes are done synchronously. In training jobs, there has been cases where metric updates take +20% of a training iteration. Metric computations, although less frequent, can takes over a couple of seconds.
CPUOffloadedRecMetricModule aims to perform all metric updates/computes asynchronously, completely removing them from the critical path.
This patch adds:
Differential Revision: D83773528