Models can directly output forces

[peastman]

Fixes #48. This allows the model to optionally return both energy and forces, instead of relying on backpropagation for the forces.

[peastman]

I'm not sure what's causing the CI failures. The Linux build gets this error:

/usr/share/miniconda/envs/build/bin/../lib/gcc/x86_64-conda-linux-gnu/9.4.0/../../../../x86_64-conda-linux-gnu/bin/ld: /usr/share/miniconda/envs/build/lib/python3.7/site-packages/torch/lib/libtorch_cpu.so: undefined reference to `memcpy@GLIBC_2.14'

It looks like PyTorch is compiled against a very old version of glibc that isn't available on the build machine. But we get it from conda-forge, so shouldn't it use the same version as everything else?

The Mac build fails with this error when building the Python package.

Installed /usr/local/miniconda/envs/build/lib/python3.9/site-packages/openmmtorch-1.0-py3.9-macosx-10.9-x86_64.egg
Processing dependencies for openmmtorch==1.0
Searching for openmmtorch==1.0
Reading https://pypi.org/simple/openmmtorch/
Couldn't find index page for 'openmmtorch' (maybe misspelled?)
Scanning index of all packages (this may take a while)
Reading https://pypi.org/simple/
No local packages or working download links found for openmmtorch==1.0
error: Could not find suitable distribution for Requirement.parse('openmmtorch==1.0')

I don't understand that. It claims to have just installed openmmtorch 1.0, and then immediately says that it can't find openmmtorch 1.0.

Any idea what's causing either of these?

[raimis]

I'll try to solved the CI issue with #53

[peastman]

CI is passing. Any comments on this, or should I merge it?

[raimis]

This looks good.

[peastman]

Thanks!