Use args when calling self.aev_computer

[lohedges]

This PR allows the use of a forward hook on the internal AEVComputer of ANI models when using OptmizedTorchANI via TorchScript. See aiqm/torchani#649 for details.

[peastman]

Does this introduce any compatibility issues we need to be aware of, such as only working with newer versions of TorchANI?

[lohedges]

I don't think so, I've checked back to version 1.0.0 of TorchANI and the signature of AEVComputer.forward is identical (although missing type hinting). Older versions, e.g. the original 0.1 release, are missing the cell and pbc kwargs, but I doubt you support those given that the oldest version of TorchANI on conda-forge is 2.2.

Just to note that while the PR linked to above won't (yet) be merged due to a code freeze, this fix is independent of it, i.e. it allows you to use an AEVComputer forward hook with OptimizedTorchANI without requiring the fix to TorchANI itself.

[peastman]

@RaulPPelaez any idea what's going on with CI?

[RaulPPelaez]

Could it be branch protection rules related to OP being a first time contributor? https://github.com/orgs/community/discussions/26698

I cannot access those settings for this repo, can you take a look @peastman ?

[peastman]

I checked and the rules listed there are correct. The problem is that none of the CI builds have run. And there's no option to retrigger them. Let me try making a trivial change to the file and see if that gets it working.

[peastman]

Look like they're running now.

[peastman]

One build is failing with the dreaded [Errno 28] No space left on device'. It's clearly not related to this change, so I'm ok with merging it. I'm currently dealing with that same error in trying to do a new OpenMM dev build.

[lohedges]

Many thanks for looking at this so quickly.