Skip to content

Fix Qwen Image 2.1 compile recompiles - #15074

Open
solo938 wants to merge 1 commit into
huggingface:mainfrom
solo938:qwenimage21-fix
Open

Fix Qwen Image 2.1 compile recompiles#15074
solo938 wants to merge 1 commit into
huggingface:mainfrom
solo938:qwenimage21-fix

Conversation

@solo938

@solo938 solo938 commented Oct 18, 2026

Copy link
Copy Markdown

Summary

  • Replace the cache write slice argument with an integer cache write length to avoid recompilation caused by varying slice objects.
  • Add regression coverage for both Qwen Image 2.1 attention processors across varying generation inputs.

Testing

  • Full Qwen Image 2.1 transformer test file passes.
  • Regression tests pass for both attention processors.
  • Ruff lint and formatting checks pass.
  • git diff --check passes.

Closes #15027

@solo938 solo938 mentioned this pull request Oct 19, 2026
Open
sayakpaul
sayakpaul reviewed Oct 20, 2026
View reviewed changes
Comment thread src/diffusers/models/transformers/transformer_qwenimage21.py
layer_cache=layer_cache,
kv_cache_mode=kv_cache_mode,
cache_write_slice=cache_write_slice,
cache_write_len=cache_write_len,

@sayakpaul sayakpaul Oct 20, 2026

Copy link
Copy Markdown
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

This is a breaking change.

sayakpaul
sayakpaul reviewed Oct 20, 2026
View reviewed changes
Comment thread tests/models/transformers/test_models_transformer_qwenimage21.py

torch.testing.assert_close(outputs[0], outputs[1])

@pytest.mark.parametrize("processor_name", ["QwenImage21FlexAttnProcessor", "QwenImage21AttnProcessor"])

@sayakpaul sayakpaul Oct 20, 2026

Copy link
Copy Markdown
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

We have a dedicated compilation tester class which should be used here.

sayakpaul
sayakpaul reviewed Oct 20, 2026
View reviewed changes

@sayakpaul sayakpaul left a comment

Copy link
Copy Markdown
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Need to defer to @naykun for a review here.

This branch has not been deployed

No deployments
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Reviewers

@sayakpaul sayakpaul sayakpaul left review comments

At least 1 approving review is required to merge this pull request.

Assignees

No one assigned

Labels

fixes-issue models size/M PR with diff < 200 LOC tests

Projects

None yet

Milestone

No milestone

Development

Successfully merging this pull request may close these issues.

Qwen Image 2.1 with torch.compile triggers a recompile on each run with varying image size

2 participants

@solo938 @sayakpaul