-
Notifications
You must be signed in to change notification settings - Fork 21.3k
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
[cudagraph] fix verbose graph logging #126694
Conversation
🔗 Helpful Links🧪 See artifacts and rendered test results at hud.pytorch.org/pr/126694
Note: Links to docs will display an error until the docs builds have been completed. ✅ No FailuresAs of commit ab1b4ec with merge base 8c38d0c (): This comment was automatically generated by Dr. CI and updates every 15 minutes. |
@pytorchbot merge |
Merge failedReason: Approvers from one of the following sets are needed:
|
@pytorchbot merge |
Merge failedReason: This PR needs a If not, please add the To add a label, you can comment to pytorchbot, for example For more information, see Details for Dev Infra teamRaised by workflow job |
@pytorchbot merge |
Merge startedYour change will be merged once all checks pass (ETA 0-4 Hours). Learn more about merging in the wiki. Questions? Feedback? Please reach out to the PyTorch DevX Team |
Since this fix might require quite a lot time (at least several months) to be public, I have one workaround for existing pytorch versions before this PR. Please check the gist if anyone is interested. |
According to the doc:
1 << 10
is not the most verbose flag. it is just one flag to add node handles and every kernel function handle to output.1 << 0
is the most verbose flag, under the namecudaGraphDebugDotFlagsVerbose
.Here is an example of graph, dumped with
1 << 10
:The same graph dumped with
1 << 0
: