Brandon Butler

To input a machine learned potential right now would require a self-generated neighbor list. This is not due to lack of logic in `hoomd.md.force.Custom` but rather the `hoomd.md.nlist.NList` subclasses do...

@jglaser awesome, will this be a full JIT PR, because at our weekly hoomd developer meetings we have been discussing the future of JIT in hoomd, and it would be...

I would suggest the 2nd as charges and diameters at least make most sense just in the simulation. I am not sure for `constituent_types`, but it does provide the most...

Hmm, I didn't notice that we used compiled CPU code on the GPU. In that case, it is fine as is.

@bdice and @vyasr I would support a major version bump. For the hoomd beta's we are allowing breakages in-between beta versions, but the are not considered official releases of 3.0....

@tommy-waltmann what do you think of this PR? Should we just close it or is this still in the future interest of the project in your opinion?

I think that have this in a separate package is a good idea. I don't think there is a problem in by default installing the `mbuild_base_recipes` repo or whatever it...

I personally use `job._project` in my projects for things like project default values.

Given the large number of reviews, I am removing myself.

@cbkerr do you still plan on getting back to this?