(READ IF ISSUES RAISE) int32/int64 inconsistency in all_atom.py one hot

[carbondrop-nick]

\flowmatch\data\all_atom.py line 18:
GROUP_IDX = torch.tensor(residue_constants.restype_atom14_to_rigid_group)
appears to create int32, which results in error 'group_maskRuntimeError: one_hot is only applicable to index tensor.'
I converted to to int64 explicitly at line 19 and that seemed to fix it:
GROUP_IDX = GROUP_IDX.to(torch.int64)
There is probably a better fix...

[carbondrop-nick]

Same issue in utils/loss.py. residue_index is defaulting to int32 for me, preventing later conversion to one hot. Fixed by making it explicit.

[carbondrop-nick]

With this int32/int64 issue fixed I was able to through the entire ipynb without error.