Keep masked candidates out of underfilled retrieval responses - #56
Open
Robertboy18 wants to merge 1 commit into
Open
Keep masked candidates out of underfilled retrieval responses#56Robertboy18 wants to merge 1 commit into
Robertboy18 wants to merge 1 commit into
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
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Hey! I was looking through the retrieval masking path and noticed a semantic discrepancy when fewer than
large_kcandidates remain eligible: top-k still fills the remaining slots with sentinel-masked entries, and those indices can reach response construction.This keeps a Boolean validity tensor alongside the selected indices and checks it while building the response. Fixed JAX shapes stay the same, while underfilled requests now return only eligible candidates. I also added coverage for the underfilled top-k case and response filtering.
I ran the focused JAX/CUDA test, Ruff, Rust formatting/checks, and rechecked the corresponding TorchLean theorem.
I have the Lean/TorchLean formalization of the cardinality argument too and can attach it if useful :)