Disable Autograd when calling RecordFunction callbacks #100727
Closed
Conversation
This file contains 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
[ghstack-poisoned]
🔗 Helpful Links🧪 See artifacts and rendered test results at hud.pytorch.org/pr/100727
Note: Links to docs will display an error until the docs builds have been completed. ✅ No FailuresAs of commit 5c676b8: This comment was automatically generated by Dr. CI and updates every 15 minutes. |
soulitzer
added a commit
that referenced
this pull request
May 5, 2023
ghstack-source-id: ed41d174a87bd70a7d1bcd16e60c3145ca3db4f7 Pull Request resolved: #100727
soulitzer
added
release notes: profiler
This PR disables autograd for all callbacks that are registered to RecordFunction.
Context: RecordFunction interposes before autograd in at least a couple places: custom Function, detach{,_}'s VariableTypeManual (why?) and makes it so that if any callbacks call torch operations, it may result in tensors being saved for backward (potentially problematic in the case of non-reentrant checkpointing even if the graph/saved tensors are expected to die immediately).
RecordFunction should be primarily used for profiling + logging, so it should probably live below autograd, but since actually moving where RecordFunction is called will take more work, this PR is a good stop gap solution.
Alternatives:
- putting the guard on the specific callback rather than disabling autograd (safer)
Another detail:
- we could use the dispatch below autograd guard instead
[ghstack-poisoned]
soulitzer
added a commit
that referenced
this pull request
May 5, 2023
ghstack-source-id: c701b89e0184f84977c57722e52631b9713c5a65 Pull Request resolved: #100727
This was referenced May 11, 2023
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.
Stack from ghstack (oldest at bottom):
This PR disables autograd for all callbacks that are registered to RecordFunction.
Context: RecordFunction interposes before autograd in at least a couple places: custom Function, detach{,_}'s VariableTypeManual (why?) and calls arbitrary user-defined callbacks that can be registered via "addGlobalCallback" (for example, to log infs/nans to scuba).
Issue: if any callbacks call torch operations, it may result in tensors being saved for backward (potentially problematic in the case of non-reentrant checkpointing even if the graph/saved tensors are expected to die immediately).
Proposed solution: RecordFunction should be primarily used for profiling + logging, so it should probably live below autograd, but since actually moving where RecordFunction is called will take more work, this PR is a good stop gap solution.
Alternatives:
Another detail: