🧰♻️ Add Dependency Caching #1239
Conversation
The GHA runner does not have a GPU anyway
BatchCompilationTest was failing before, but is not needed anyway
| # fixme: bring this upstream to class_resolver.contrib? | ||
| lr_scheduler_resolver = ClassResolver.from_subclasses(LRScheduler, default=ExponentialLR, suffix="LR") |
There was a problem hiding this comment.
@cthoyt this could be fixed in class-resolver upstream; we could also think about adding some logic to make this work in torch>=2.0 and torch < 2, though I am not sure whether to do that.
There was a problem hiding this comment.
i like the idea of adding more contrib to class-resolver. If it's no so hard to make compatibility, then great, otherwise we should aim for only supporting the latest version
| # fixme: find reason / enforce single-thread | ||
| @unittest.skipIf(condition=True, reason="Problems with multi-processing") |
There was a problem hiding this comment.
I couldn't resolve this issue, and this PR/test case failure seems to block others.
| # - name: Check RST format | ||
| # run: tox -e doclint |
There was a problem hiding this comment.
defined in tox, but not used in CI
|
@mberr we can do follow-up to fix the class-resolver contrib. We might also have to change the "required" tests configuration for the protections on the master branch, but then this is good from my side to merge. |
With torch 2.0, the (private) `_LRScheduler` class has been transformed to a public `LRScheduler` class. cf. pykeen/pykeen#1239
This PR enables caching for dependencies installed via pip, cf. https://docs.github.com/en/actions/using-workflows/caching-dependencies-to-speed-up-workflows
It also updates the code to be torch 2.0 compliant.