Repository navigation
Reject deferred Tensor outputs in PMPP v2 checkers - #174
Merged
Merged
Conversation
msaroufim
marked this pull request as ready for review
November 5, 2026 10:10
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
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
A Tensor subclass can move the actual kernel into torch_function, so it runs only when the correctness checker reads the returned object. This can pass value comparisons while the timed custom_kernel call performs no GPU work.
Require an ordinary materialized torch.Tensor before reading any output property or invoking any output operation. Document that contract for all eight PMPP v2 tasks. Add eight independently authored regression cases whose subclass hooks raise if the checker invokes them; all are rejected without dispatching user code.
Validation: 126 CPU tests, Ruff, diff checks and GitHub CI pass. An isolated A100 reproduction of the deferred scan is rejected by the new guard; inspecting its backing storage with subclass hooks disabled independently finds all 1023 values wrong. The historical audit uses that raw-storage check so an eager subclass alone does not trigger score invalidation. The preceding change's 32 positive controls return ordinary materialized tensors and remain compatible. No contestant source is included.