chore(ci): use macOS to test ML dependencies#3774
Conversation
Codecov Report✅ All modified and coverable lines are covered by tests. Additional details and impacted files🚀 New features to boost your workflow:
|
|
The documentation preview is ready to be viewed at http://preview.awkward-array.org.s3-website.us-east-1.amazonaws.com/PR3774 |
06fcc3f to
0851fd4
Compare
|
Looking a bit more into it, it seems like |
oh yeah, on linux |
|
Seems like we can specify it in the |
|
Will this cause problems with other packages? I'd hope that the url is "pure". |
|
I'll mark this as a draft while we decide what to do. In the meeting we discussed potentially switching to conda environments or to pixi. |
it may not be bad to have this or a patch at the integration tests repo (whatever y'all prefer) in the meantime just for the sake of having the integration tests ci pass. |
|
Yeah, that's a good point. How for now we just switch to testing ML on macOS (which doesn't have the extra nvidia bloat) and revert to the original requirements? |
That does sound good to me. Actually, even jax may bring in bloat on linux but I don't remember off the top of my head. |
|
I think this needs a title change now to reflect what it actually does but should be good otherwise |

We changed the ML requirements to use
tensorflow-cpuinstead oftensorflowsince it was downloading a bunch of extra stuff and running out of storage. However,tensorflow-cpudoesn't build wheels for macOS, so the integration tests are failing. Since this requirement file could be used for people as a guideline, it's good to fix it here. Thanks to @ikrommyd for pointing this out.