Skip to content

Fix CI failures - #238

Merged
regisss merged 2 commits into
huggingface:mainfrom
JakeStevens:main
Jul 20, 2026
Merged

Fix CI failures#238
regisss merged 2 commits into
huggingface:mainfrom
JakeStevens:main

Conversation

@JakeStevens

Copy link
Copy Markdown
Collaborator

CI is currently failing due to stale nightly pins, as well as out of date node dep.

To fix, we update the pins but also update CI to test against the ExecuTorch 1.3.1 stable wheel stack. We split the workflows to make the signal more clear if it is the stable or nightly that is failing.

The docs workflows are also moved to Node 20 because the current doc-builder dependency no longer supports Node 18.

Update CI to test against the ExecuTorch 1.3.1 stable wheel stack, while keeping nightly coverage in a separate required workflow.

The previous model test workflow used stale nightly wheel pins, causing jobs to fail during environment setup before the test suite could run. This changes the stable model workflow to use ExecuTorch 1.3.1 with matching PyTorch CPU wheels, removes an obsolete source install of pytorch-tokenizers that downgraded below ExecuTorch's required version, and adds a dedicated nightly workflow that is still required to pass.

The docs workflows are also moved to Node 20 because the current doc-builder dependency no longer supports Node 18.

@regisss regisss left a comment

Copy link
Copy Markdown
Collaborator

Choose a reason for hiding this comment

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

LGTM

@regisss

regisss commented Jul 16, 2026

Copy link
Copy Markdown
Collaborator

It seems there are conflicts between some packages when installing the dependencies, e.g. https://github.com/huggingface/optimum-executorch/actions/runs/29417087694/job/87582915363?pr=238

@HuggingFaceDocBuilderDev

Copy link
Copy Markdown

The docs for this PR live here. All of your documentation changes will be reflected on that endpoint. The docs are available until 30 days after the last update.

@JakeStevens

Copy link
Copy Markdown
Collaborator Author

I think the conflict should be resolved now, please approve workflow when possible.

There are existing, unrelated failures that I believe should be resolved by:

https://github.com/huggingface/optimum-executorch/pull/240/changes

@regisss

regisss commented Jul 17, 2026

Copy link
Copy Markdown
Collaborator

Seeing

ImportError: /opt/hostedtoolcache/Python/3.11.15/x64/lib/python3.11/site-packages/executorch/extension/pybindings/_portable_lib.cpython-311-x86_64-linux-gnu.so: undefined symbol: _ZN3c104impl3cow23materialize_cow_storageERNS_11StorageImplE

in https://github.com/huggingface/optimum-executorch/actions/runs/29507297977/job/87836728795?pr=238#step:5:48

Not sure exactly why, I guess there is a mismatch between the platform this version of ExecuTorch was compiled for and the setup we use in CI?

Update CI to test against the ExecuTorch 1.3.1 stable wheel stack, while keeping nightly coverage in a separate required workflow.

The previous model test workflow used stale nightly wheel pins, causing jobs to fail during environment setup before the test suite could run. This changes the stable model workflow to use ExecuTorch 1.3.1 with matching PyTorch CPU wheels, removes an obsolete source install of pytorch-tokenizers that downgraded below ExecuTorch's required version, and adds a dedicated nightly workflow that is still required to pass.

The docs workflows are also moved to Node 20 because the current doc-builder dependency no longer supports Node 18.
@JakeStevens

JakeStevens commented Jul 20, 2026

Copy link
Copy Markdown
Collaborator Author

Seeing

ImportError: /opt/hostedtoolcache/Python/3.11.15/x64/lib/python3.11/site-packages/executorch/extension/pybindings/_portable_lib.cpython-311-x86_64-linux-gnu.so: undefined symbol: _ZN3c104impl3cow23materialize_cow_storageERNS_11StorageImplE

in https://github.com/huggingface/optimum-executorch/actions/runs/29507297977/job/87836728795?pr=238#step:5:48

Not sure exactly why, I guess there is a mismatch between the platform this version of ExecuTorch was compiled for and the setup we use in CI?

There was an issue on the ET side between ET/Pytorch which has been resolved. CI is now green on those failed imports.

We have remaining test failures, but those are uncovered by these fixes, because they are actually running again. I believe this PR is ready to merge, addressing those separately in follow up #240

@regisss
regisss merged commit d81eef9 into huggingface:main Jul 20, 2026
78 of 86 checks passed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

5 participants