ci: Test with torch nightly#707
Conversation
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
ValerianRey
added
the
cc: ci
ValerianRey
added
the
cc: ci
github-actions
Bot
changed the title
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
|
This seems to work (see the print in the action that shows the torch and torchvision versions). I think we could merge this but not make it a mandatory check (i.e. not change the branch and tag rulesets) for now, because maybe the nightly version can be unstable. |
| shell: bash | ||
| run: | | ||
| uv pip install --pre --upgrade torch torchvision --index-url https://download.pytorch.org/whl/nightly/cpu | ||
| uv run python -c "import torch, torchvision; print('torch:', torch.__version__, '| torchvision:', torchvision.__version__)" |
There was a problem hiding this comment.
We'll remove the print here right?
There was a problem hiding this comment.
No, I'd rather keep it to always be able to verify which version we actually installed (since it's not trivial and may be a source of error). This way, if the job fails, we know exactly with which version it failed.
| - python-version: '3.10.0' | ||
| extra_groups: 'lower_bounds' | ||
| # Upper-bounds: torch nightly build. | ||
| - torch_channel: 'nightly' |
There was a problem hiding this comment.
Are we testing nightly with lower bounds? Does that make sense? Also I don't understand the comment above.
There was a problem hiding this comment.
I don't think it makes a lot of sense to test the compatibility of torch nightly (i.e. torch upper bound) and other libraries lower bounds. This setup should not really happen in practice.
Well, it says "Upper-bounds" because nightly is the beta (unreleased) version of pytorch (so the highest version we can get for it) and all other libraries will use the highest version that is released.
There was a problem hiding this comment.
I don't think it makes a lot of sense to test the compatibility of torch nightly (i.e. torch upper bound) and other libraries lower bounds. This setup should not really happen in practice.
Which is why I'm confused, it looks like (line 41) we are installing lower bounds.
There was a problem hiding this comment.
If you look at the full file it's a lot clearer. With the way github cuts it, it's a bit confusing yeah
2 participants
Add a CI matrix entry that installs the torch nightly build and runs the unit test suite against it.
The
install-depsaction gains atorch_channelinput (default:stable); when set tonightly, it upgradestorchandtorchvisionfrom the PyTorch nightly CPU index after the normal install step.🤖 Generated with Claude Code