Force use of torch.compile on deterministic roi_align implementation #8436
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
Signed-off-by: Edward Z. Yang <[email protected]>
🔗 Helpful Links🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/vision/8436
Note: Links to docs will display an error until the docs builds have been completed. ❌ 12 New Failures, 1 Unrelated FailureAs of commit ee25749 with merge base 775dd2d ( NEW FAILURES - The following jobs have failed:
FLAKY - The following job failed but was likely due to flakiness present on trunk:
This comment was automatically generated by Dr. CI and updates every 15 minutes. |
|
cc @qqaatw, I removed the MPS knob because of how memory hungry the eager implementation is, I doubt torch.compile works on MPS. |
Signed-off-by: Edward Z. Yang <[email protected]>
Signed-off-by: Edward Z. Yang <[email protected]>
Signed-off-by: Edward Z. Yang <[email protected]>
Signed-off-by: Edward Z. Yang <[email protected]>
NicolasHug
approved these changes
May 23, 2024
| def lazy_compile(**compile_kwargs): | ||
| """Lazily wrap a function with torch.compile on the first call | ||
|
|
||
| This avoids eagerly importing dynamo. |
There was a problem hiding this comment.
Am I understanding this correctly?
Suggested change
| This avoids eagerly importing dynamo. | |
| This avoids eagerly compiling a function at import time. |
There was a problem hiding this comment.
Nope. Even with torch.compile at top level it isn't compiled until you call it the first time. But importing dynamo has undesirable side effects for eager mode only users so it's best not to do it.
| @@ -232,7 +250,9 @@ def roi_align( | |||
| if not isinstance(rois, torch.Tensor): | |||
| rois = convert_boxes_to_roi_format(rois) | |||
| if not torch.jit.is_scripting(): | |||
| if not _has_ops() or (torch.are_deterministic_algorithms_enabled() and (input.is_cuda or input.is_mps)): | |||
| if ( | |||
| not _has_ops() or (torch.are_deterministic_algorithms_enabled() and (input.is_cuda or input.is_mps)) | |||
There was a problem hiding this comment.
Should we just remove the mps part here since you mentioned MPS doesn't even work with torch.compile?
Suggested change
| not _has_ops() or (torch.are_deterministic_algorithms_enabled() and (input.is_cuda or input.is_mps)) | |
| not _has_ops() or (torch.are_deterministic_algorithms_enabled() and input.is_cuda) |
There was a problem hiding this comment.
I opted to keep it around, because it was explicitly added by @qqaatw, but I don't really mind either way
There was a problem hiding this comment.
Sorry for the late reply! I'm ok with either way that is best for the development. From the mentioned issue it seems only relevant to CUDA, is MPS similarly memory hungry with deterministic algorithm?
facebook-github-bot
pushed a commit
that referenced
this pull request
Jun 7, 2024
…entation (#8436) Summary: Signed-off-by: Edward Z. Yang <[email protected]> Reviewed By: vmoens Differential Revision: D58283855 fbshipit-source-id: 914a91877c193b38f29af450a5935dd1ab5b20d7 Co-authored-by: Nicolas Hug <[email protected]>
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.
Fixes #8168
Signed-off-by: Edward Z. Yang [email protected]