Use conda-forge's kalign-python#221
Merged
Merged
Conversation
This makes conda-centric environments safer (use CF consolidated OpenMP) To think about the pypi-centric ones, likely we should leave as is, warn the user and recommend to always favor the conda-centric ones. Alternatively we could install kalign-python from conda, but that pulls numpy (arguably better because of blas linking and hopefully always up to date with pypi's numpy) and is much less foolproof for OpenMP.
Contributor
Author
|
I actually got one failed test: Investigating |
sdvillal
force-pushed
the
kalign-python-from-conda
branch
from
614af85 to
6a20fe5
Compare
Contributor
Author
|
Failing test was due to rdkit update. Pinned for the time being, will open a different issue. |
Contributor
Author
jnwei
added
the
safe-to-test
Contributor
Author
|
The problems with the docker files might be due to them shipping an old version of pixi (0.65)? Here we have moved to >=0.68 as we are moving to lockfile v7 |
Contributor
|
Yes, we needed to update the Dockerfile.pixi to reflect the new default pixi version. Triggering a manual run of the CI here: https://github.com/aqlaboratory/openfold-3/actions/runs/25776933311 If this is green, will go ahead and merge |
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
Use conda-forge's kalign-python instead of pypi's one in conda-centric environments. This:
Heavy-lifting was creating the package here and here.
Changes
Related Issues
See description
Testing
Tests pass in linux-64 and osx-arm64. Inference works on osx.
Other Notes
This makes conda-centric environments safer (use consolidated building toolchain and openmp runtime). Pypi's wheels with their vendored openmp could result in hard to debug problems. It also eliminates the need to use
export KMP_DUPLICATE_LIB_OK=TRUEto run inference in osx.Important
We should either install this from conda-forge everywhere (adding it to "not-in-pypi" feature) or warn the user and heavily recommend using the conda-centric environments
To think about the pypi-centric ones, likely we should leave as is, warn the user and recommend to always favor the conda-centric ones. Alternatively we could install kalign-python from conda, but that pulls numpy (that would maybe be better anyway because of blas-linking and hopefully always up to date with pypi's numpy) and is much less foolproof for OpenMP (as pypi's dependencies still might bring their openmp runtime).