Skip to content

enable graph colltrace in ncclx/ctran wrappers (#1979)#1979

Open
dolpm wants to merge 7 commits intometa-pytorch:mainfrom
dolpm:export-D98966415
Open

enable graph colltrace in ncclx/ctran wrappers (#1979)#1979
dolpm wants to merge 7 commits intometa-pytorch:mainfrom
dolpm:export-D98966415

Conversation

@dolpm
Copy link
Copy Markdown
Contributor

@dolpm dolpm commented Apr 7, 2026

Summary:

re-enable graph support in colltracewrappers

Reviewed By: SuhitK

Differential Revision: D98966415

@meta-cla meta-cla bot added the CLA Signed This label is managed by the Meta Open Source bot. label Apr 7, 2026
@meta-codesync
Copy link
Copy Markdown
Contributor

meta-codesync bot commented Apr 7, 2026

@dolpm has exported this pull request. If you are a Meta employee, you can view the originating Diff in D98966415.

dolpm added 7 commits April 8, 2026 15:14
Summary: we should support multiple current collectives - this is possible when multiple collectives are being executed across ctran/nc cl-baseline communicators, or we have collectives that require concurrency (e.g., signals)

Differential Revision: D98955440
Summary:
we're adding support for multiple concurrent collectives.

it's possible we have 2 events progressing at the same time. if the calls to the processing handler oscillate,  we will hit the &curEvent != lastEvent_ which will reset the timeout (instead of keeping the timers independent per event)

Differential Revision: D98980980
Differential Revision: D99015585
Differential Revision: D99015610
Summary:
Pull Request resolved: meta-pytorch#1980

```
Graph CollTrace overhead (~200us D2D collectives, 500 replays):
  N=1     Baseline:    138.7us  CollTrace:    151.5us  Overhead: +9.2%
  N=10    Baseline:   1365.8us  CollTrace:   1385.4us  Overhead: +1.4%
  N=100   Baseline:  13631.5us  CollTrace:  13736.7us  Overhead: +0.8%
```

Differential Revision: D97960551
Summary:
Pull Request resolved: meta-pytorch#1979

re-enable graph support in colltracewrappers

Reviewed By: SuhitK

Differential Revision: D98966415
@meta-codesync meta-codesync bot changed the title enable graph colltrace in ncclx/ctran wrappers enable graph colltrace in ncclx/ctran wrappers (#1979) Apr 8, 2026
@dolpm dolpm force-pushed the export-D98966415 branch from bc73430 to e94e0be Compare April 8, 2026 22:14
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

CLA Signed This label is managed by the Meta Open Source bot. fb-exported meta-exported

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant