Nightly / Nvidia CI workflows trigger on forks and fail due to missing org-specific runners

[AvinashDwivedi]

System Info

When forking the huggingface/transformers repository, certain GitHub Actions workflows (like “Nvidia CI with nightly torch” and “Nightly PyTorch build”) are automatically triggered on the forked repo’s default branch (main) — even though they depend on organization-specific GPU runners and secrets.

This leads to immediate workflow failures and email notifications such as:

Run failed: Nvidia CI with nightly torch - main (...)

Who can help?

No response

Information

• The official example scripts
• My own modified scripts

Tasks

• An officially supported task in the examples folder (such as GLUE/SQuAD, ...)
• My own task or dataset (give details below)

Reproduction

Fork the upstream repository huggingface/transformers on GitHub (press Fork in the web UI).

Expected behavior

Forked repositories should:

1. not trigger organization-specific CI pipelines, or
2. gracefully skip such jobs without failure.

[Rocketknight1]

cc @ydshieh

[ydshieh]

Hello @AvinashDwivedi

You can follow

#41515 (comment)

to disable the workflow on the forked repository

[AvinashDwivedi]

Thank you, @ydshieh. I'll do that right away, but I still don't feel that this is the right solution, as it should automatically select whether to start the workflows or not whenever someone forks the repository. Please let me know if you have something in mind that I can work on in a PR.

[ydshieh]

Hi @AvinashDwivedi

It's a design of Github, I agree with you, however we will simply suggest people to disable the workflows on their forked repositories.

You can try to open a discussion https://github.com/orgs/community/discussions?discussions_q=label%3AActions