Fix looping in torch guard decorator #42260
Merged
+28
−11
Conversation
This file contains hidden or 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
|
The docs for this PR live here. All of your documentation changes will be reflected on that endpoint. The docs are available until 30 days after the last update. |
There was a problem hiding this comment.
Pull Request Overview
This PR optimizes the guard_torch_init_functions decorator by replacing an expensive loop over all modules in sys.modules (which can be very large in pytest environments) with a targeted approach that only patches a specific list of known torch modules.
- Introduces
TORCH_MODULES_TO_PATCHconstant containing specific torch module paths to patch - Refactors
guard_torch_init_functions()to iterate over the predefined module list instead of allsys.modules - Renames variables in the patching logic for clarity (
name→func_name,module_namefor clarity)
Reviewed Changes
Copilot reviewed 2 out of 2 changed files in this pull request and generated 2 comments.
| File | Description |
|---|---|
| src/transformers/initialization.py | Adds TORCH_MODULES_TO_PATCH constant and refactors guard_torch_init_functions() to use targeted module patching instead of looping over all sys.modules |
| setup.py | Contains an accidental debug line that should be removed |
src/transformers/initialization.py
Outdated
| # Here, we need to check several modules imported, and hot patch all of them, as sometimes torch does | ||
| # something like `from torch.nn.init import xavier_uniform_` in their internals (e.g in torch.nn.modules.activations, | ||
| # where MultiHeadAttention lives), so the function name is binded at import time and just doing | ||
| # `setattr(torch.nn.init, name, gloabls()[name])` is thus not enough |
There was a problem hiding this comment.
Typo in comment: 'gloabls' should be 'globals'.
Suggested change
| # `setattr(torch.nn.init, name, gloabls()[name])` is thus not enough | |
| # `setattr(torch.nn.init, name, globals()[name])` is thus not enough |
Cyrilvallez
force-pushed
the
module-loop
branch
from
bf3444b to
3887a78
Compare
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.
What does this PR do?
sys.modulescan end-up being quite large, especially when usingpytest. This avoids the loop while keeping the benefits