[Fix] Improve dim_size handling in SetTransformerAggregation to prevent CUDA crash #10220
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
[Fix] Add dim_size validation and fallback to SetTransformerAggregation
base repository: pyg-team/pytorch_geometric ← compare: KAVYANSHTYAGI:fix/set-transformer-aggregation-index-check
for more information, see https://pre-commit.ci
rusty1s
reviewed
May 20, 2025
| @@ -94,6 +94,15 @@ def forward( | |||
| max_num_elements: Optional[int] = None, | |||
| ) -> Tensor: | |||
|
|
|||
| if dim_size is None: | |||
There was a problem hiding this comment.
I think this is already handled in to_dense_batch.
Comment on lines
100
to
104
| if int(index.max()) >= dim_size: | ||
| raise ValueError( | ||
| f"SetTransformerAggregation error: index.max() = {int(index.max())}, " | ||
| f"but dim_size = {dim_size}. This causes an indexing error on GPU. " | ||
| f"Ensure data.batch is set or dim_size is passed explicitly.") |
There was a problem hiding this comment.
This leads to a device sync, so we should avoid this.
Fix: Remove device sync and dim_size fallback in SetTransformerAggregation - Removed redundant dim_size = index.max() + 1 logic (handled in to_dense_batch). - Added GPU-safe index validation to avoid CUDA crashes.
KAVYANSHTYAGI
commented
May 22, 2025
There was a problem hiding this comment.
Thank you for the feedback!
I've removed the fallback dim_size = index.max() + 1 to avoid redundancy with to_dense_batch, as suggested.
Also eliminated device sync by replacing the .max() check with a GPU-safe tensor comparison using (index >= dim_size).any().
Let me know if any further simplification is preferred!
for more information, see https://pre-commit.ci
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.
This PR improves the robustness of
SetTransformerAggregationby:dim_size = index.max() + 1ifdim_sizeis not provided.index.max() >= dim_sizeto avoid CUDA crashes during evaluation.This is helpful especially for datasets like PPI where
data.batchmay be missing. It replaces hard-to-debug GPU errors with clear and early validation.