Qwen3 Reranker CUDA Fix in FlashQwen3.rs #1
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Outdated suggestions cannot be applied.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.
Problem
When running Qwen3-Reranker on CUDA with flash attention, all documents get nearly identical scores (~0.891), while on MPS they get properly differentiated scores.
Root Cause
Two issues were found in
backends/candle/src/models/flash_qwen3.rs
:Pool::Cls
instead ofPool::LastToken
for classification models (line 313)Fixes Applied
Fix 1: Change pooling method (line 313)
Fix 2: Correct index selection (lines 405-435)
The pooling logic was incorrectly handling indices when both pooled and raw requests exist. Fixed to properly select the last token indices for the requested sequences.