Fix Qwen3-VL Processor flattening multi-image batches (fix #41709)#41721
Open
Elon7069 wants to merge 1 commit intohuggingface:mainfrom
Open
Fix Qwen3-VL Processor flattening multi-image batches (fix #41709)#41721Elon7069 wants to merge 1 commit intohuggingface:mainfrom
Elon7069 wants to merge 1 commit intohuggingface:mainfrom
Conversation
… multi-image samples
Contributor
|
[For maintainers] Suggested jobs to run (before merge) run-slow: qwen3_vl |
Author
|
Hi, this PR fixes the issue where Qwen3VLProcessor was flattening multi-image batches. It now preserves per-sample grouping and passes the added unit test. |
Member
Member
zucchini-nlp
left a comment
zucchini-nlp
left a comment
There was a problem hiding this comment.
In Qwen models, the pixel shape must be NxD. We don't need a PR, thanks for working on it though
This was referenced Apr 29, 2026
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
3 participants
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Fixes #41709
This PR fixes how Qwen3-VL Processor handles multiple images per sample.
Previously, all images were flattened across the batch, causing shape mismatches in model forward passes.
Changes: