-
Notifications
You must be signed in to change notification settings - Fork 21.4k
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
fix cudnn RNN reproducibility problem #90522
Conversation
🔗 Helpful Links🧪 See artifacts and rendered test results at hud.pytorch.org/pr/90522
Note: Links to docs will display an error until the docs builds have been completed. ✅ No FailuresAs of commit ee38dce: This comment was automatically generated by Dr. CI and updates every 15 minutes. |
You are not authorized to force merges to this repository. Please use the regular |
@pytorchmergebot 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 |
Merge failedReason: The following mandatory check(s) failed (Rule Dig deeper by viewing the failures on hud Details for Dev Infra teamRaised by workflow job |
@pytorchmergebot 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 |
Fixes #74177
Since RNN code use static variables to cache state, we store an atomic_flag in RNG generator to notify new seed changes and generate new random state for RNN. The additional cost is that the it must check the atomic_flag each time to ensure reproducibility. This may be ugly but it is the best way currently without large code refactoring