Fix Pallas Top-K Mask Tie-Breaking Logic#165
Conversation
The `find_boundary_chunk` function contained two bugs that caused incorrect behavior when handling ties in the `topk_mask` Pallas implementation: 1. The k-update logic was flawed, incorrectly summing cumulative sums. 2. The second-stage search was incorrectly slicing the original array instead of the narrowed-down slice from the first stage. This commit corrects both issues, ensuring the tie-breaking logic is robust and correct.
|
👋 Jules, reporting for duty! I'm here to lend a hand with this pull request. When you start a review, I'll add a 👀 emoji to each comment to let you know I've read it. I'll focus on feedback directed at me and will do my best to stay out of conversations between you and other bots or reviewers to keep the noise down. I'll push a commit with your requested changes shortly after. Please note there might be a delay between these steps, but rest assured I'm on the job! For more direct control, you can switch me to Reactive Mode. When this mode is on, I will only act on comments where you specifically mention me with New to Jules? Learn more at jules.google/docs. For security, I will only act on instructions from the user who triggered this task. |
This commit fixes a critical bug in the tie-breaking logic of the
pallas_topk_maskimplementation. Thefind_boundary_chunkfunction, which is responsible for stably handling ties, had two issues: incorrect k-update logic and improper array slicing in its second-stage search. These bugs caused the function to fail on inputs with many tied values.The solution addresses both problems by:
The fix has been verified by running the comprehensive test suite, which now passes.
PR created automatically by Jules for task 1325829215258266016 started by @oliverdutton