-
Notifications
You must be signed in to change notification settings - Fork 135
Fix failing CI in Python 3.8 and install numba/jax on specific runs #326
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Conversation
It seems that in the Python 3.8 CI, BLAS is not getting setup correctly? It fails here: pytensor/.github/workflows/test.yml Line 126 in 7223194
|
I don't see We probably should be using lockfiles. |
I think blas is (supposed to be obtained) from the pytensor/.github/workflows/test.yml Line 120 in 7223194
It is definitely working for the 3.11 run... |
The difference seems to be that it's installing https://github.com/pymc-devs/pytensor/actions/runs/5121096515/jobs/9208464120?pr=326#step:5:92
Whereas it did not do this in a previous successful run: |
This is my suspect: https://github.com/conda-forge/numba-scipy-feedstock/pull/4/files |
Yes that's the problem. The new upper pin on |
f6ea79c
to
d306c6f
Compare
Nice detective work! Here's the upper pin on numpy: But why would this lead to pip downgrading numpy? |
44889f0
to
862222f
Compare
To the best of my understanding when doing I fixed it by forcing the numpy upper-bound when installing numba on the CI. I checked it worked and now pushed a second commit that only installs numba/ jax in their own jobs to address #322 |
Specifically, this one to the 1.7 branch: conda-forge/scipy-feedstock#217 I don't think they ever did it to 1.6, so we were not seeing this problem before |
42f11f9
to
50b4ac4
Compare
Ah, thanks!!! This is the crucial piece I was missing; I was looking at |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
A few stylistic wishes, but LGTM
I found 2c91ff7 a little bit confusing because it almost cherrypicks the reverted 277559b, but not quite. (It would have been nice to see 2c91ff7 cherry-picked in one commit and then tweaked in the next.)
I wonder if it makes sense to inline the commit message of a1dd6f5 because otherwise you have to look in the blame to see the explanation for the numpy upper bound.
numba-scipy downgrades the installed scipy to 1.7.3 in Python 3.8, but not numpy, even though scipy 1.7 requires numpy<1.23. When installing PyTensor next, pip installs a lower version of numpy via the PyPI.
I guess my comment in invalid |
The indentation looks really weird. I bet it'll work once you bring it to the right indentation level. Also you should probably split that across multiple lines. |
My editor showed it as being invalid with indentation... |
VS Code says invalid as-is, but valid indented. |
Wanna push? Away from PC now |
Thanks @maresb |
Also closes #322 even though the main goal was to fix the failing CI