Add python 3.11 builds

[mikemhenry]

Checklist

• Used a personal fork of the feedstock to propose changes
• Bumped the build number (if the version is unchanged)
• Reset the build number to 0 (if the version changed)
• Re-rendered with the latest conda-smithy (Use the phrase @conda-forge-admin, please rerender in a comment in this PR for automated rerendering)
• Ensured the license file is being packaged.

[conda-forge-webservices]

Hi! This is the friendly automated conda-forge-linting service.

I just wanted to let you know that I linted all conda-recipes in your PR (recipe) and found it was in an excellent condition.

[mikemhenry]

@conda-forge-admin, please rerender

[github-actions]

Hi! This is the friendly automated conda-forge-webservice.

I tried to rerender for you, but it looks like there was nothing to do.

This message was generated by GitHub actions workflow run https://github.com/conda-forge/nnpops-feedstock/actions/runs/3908834431.

[mikemhenry]

Looks like these won't work:

python 3.11 pytorch 1.13 cuda 11.2
python 3.10 pytorch 1.12 cuda 11.0
python 3.10 pytroch 1.11 cuda 11.0

[mikemhenry]

@peastman is that what you would expect or is something weird going on?
@hmacdope this package doesn't depend on openmm, so it doesn't need to be re-built but I thought it would be nice to expand support for more pytorch, cuda, and python versions.

[peastman]

This does need to be rebuilt for the release candidate, because we need the fix from openmm/NNPOps#71.

The failures are not expected. It's failing with

3: >           cpp_module.setattr(name, orig_value)
3: E           RuntimeError: Could not cast attribute '_species_to_tensor' to type __torch__.torchani.utils.ChemicalSymbolsToInts (of Python compilation unit at: 0x55568c6e19c0): Could not cast attribute 'rev_species' to type Dict[Tensor, int]: Unable to cast H to Tensor

@raimis any idea what is causing this?

[mikemhenry]

@peastman I did get v0.3 released which I think includes that fix. I haven't checked to see if our CI covers all these versions of pytorch, cuda, and python. We should probably either expand the CI matrix or just release builds we have coverage for.

[peastman]

It seems to be three different errors.

Python 3.10, PyTorch 1.11: There's a test failure in test_neighbor_values:

2023-01-13T07:36:50.2176850Z >       assert np.all(ref_neighbors == neighbors)
2023-01-13T07:36:50.2177449Z E       assert False
2023-01-13T07:36:50.2178222Z E        +  where False = <function all at 0x7f6939745cf0>(array([[ -1, ...1,  -1, 998]]) == array([[ -1, ..., dtype=int32)
2023-01-13T07:36:50.2178910Z E        +    where <function all at 0x7f6939745cf0> = np.all
2023-01-13T07:36:50.2179388Z E           Full diff:
2023-01-13T07:36:50.2180035Z E             array([[ -1,  -1,  -1, ...,  -1,  -1, 999],
2023-01-13T07:36:50.2180682Z E           -        [ -1,  -1,  -1, ...,  -1,  -1, 998]], dtype=int32,
2023-01-13T07:36:50.2181312Z E           ?                                             -------------
2023-01-13T07:36:50.2181900Z E           +        [ -1,  -1,  -1, ...,  -1,  -1, 998]],
2023-01-13T07:36:50.2182336Z E             ))
2023-01-13T07:36:50.2182568Z 
2023-01-13T07:36:50.2183061Z ../src/pytorch/neighbors/TestNeighbors.py:88: AssertionError

Python 3.10, PyTorch 1.12: There's an error in conda. Possibly a random intermittent problem?

2023-01-13T07:42:20.6792824Z conda.CondaMultiError: Downloaded bytes did not match Content-Length
2023-01-13T07:42:20.6793918Z   url: https://conda.anaconda.org/conda-forge/linux-64/pytorch-1.12.1-cuda110py310he962cc7_201.tar.bz2
2023-01-13T07:42:20.6794917Z   target_path: /home/conda/feedstock_root/build_artifacts/pkg_cache/pytorch-1.12.1-cuda110py310he962cc7_201.tar.bz2
2023-01-13T07:42:20.6795647Z   Content-Length: 474600754
2023-01-13T07:42:20.6796147Z   downloaded bytes: 101857723

Python 3.11, PyTorch 1.13: This is the same error as openmm/openmm-ml#23. It's happening because there are no torchani 2.2.2 builds for Python 3.11. Instead it installs the older noarch build of torchani 2.2, which contains a bug.

[peastman]

I did get v0.3 released which I think includes that fix.

Ok, that's good. Thanks. I think the next step is to see if we can get torchani to provide builds for Python 3.11.

[mikemhenry]

Going to give this a try on a new PR