Skip to content

feat: make audio feature extractors torch.export-able - #41033

Closed
The5cheduler wants to merge 7 commits into
huggingface:mainfrom
The5cheduler:feature/export-audio-feature-extractors
Closed

The5cheduler wants to merge 7 commits into
huggingface:mainfrom
The5cheduler:feature/export-audio-feature-extractors

Conversation

@The5cheduler

@The5cheduler The5cheduler commented Sep 21, 2025 •

Copy link
Copy Markdown
Contributor

Fix #40986

Refactors the feature extraction logic for both GraniteSpeechFeatureExtractor and WhisperFeatureExtractor into separate nn.Module subclasses. This encapsulates the feature extraction computation in a way that is compatible with torch.export.

A new method, to_exportable_module(), was added to both feature extractor classes to return an instance of these new modules.

The original __call__ and _torch_extract_fbank_features methods were updated to use these exportable modules.

Finally, new tests were added to verify that both feature extractors can be successfully exported using torch.export.

What does this PR do?

Fixes # (issue)

Before submitting

  • This PR fixes a typo or improves the docs (you can dismiss the other checks if that's the case).
  • Did you read the contributor guideline,
    Pull Request section?
  • Was this discussed/approved via a Github issue or the forum? Please add a link
    to it if that's the case.
  • Did you make sure to update the documentation with your changes? Here are the
    documentation guidelines, and
    here are tips on formatting docstrings.
  • Did you write any new necessary tests?

Who can review?

Anyone in the community is free to review the PR once the tests have passed. Feel free to tag
members/contributors who may be interested in your PR.

Fix huggingface#40986

Refactors the feature extraction logic for both `GraniteSpeechFeatureExtractor` and `WhisperFeatureExtractor` into separate `nn.Module` subclasses. This encapsulates the feature extraction computation in a way that is compatible with `torch.export`.

A new method, `to_exportable_module()`, was added to both feature extractor classes to return an instance of these new modules.

The original `__call__` and `_torch_extract_fbank_features` methods were updated to use these exportable modules.

Finally, new tests were added to verify that both feature extractors can be successfully exported using `torch.export`.
@Rocketknight1

Copy link
Copy Markdown
Member

cc @eustlb @ebezzam

@ebezzam ebezzam added the Audio label Sep 23, 2025
@github-actions

Copy link
Copy Markdown
Contributor

[For maintainers] Suggested jobs to run (before merge)

run-slow: granite_speech, whisper

@ArthurZucker

Copy link
Copy Markdown
Collaborator

Thanks for the PR, and sorry it didn't make it in! This branch is based on v4, and v5 is very different, so rather than rebasing we'd love to invite you to open a new PR against main using our latest tools and good practices:

Closing this one in the meantime, thanks again!

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

Projects

None yet

Development

Successfully merging this pull request may close these issues.

Rewrite audio feature processors to accept torch tensors

4 participants