Unknown Torch dependence during get_model?

[ensol-JKY]

I'm raising this issue just for your information.

Runtime error occurs during LAMMPS simulation using LAMMPS-torch when I deploy sevennet checkpoint using get_model cli under following environment

-python 3.10
-torch 2.6.1
-cuda 12.9
-flash_TP 0.1.0

and the summarized error message is,

Traceback of Torchscript, serialized code (most recent call last):
File "code/torch/sevenn/nn/sequential/___torch_mangle_194.py" line 20, in forward
...

Traceback of Torchscript, original code (most recent call last):
File "/my/home/path/miniconda3/envs/sevennetenv/lib/python3.10/site-packages/sevenn/nn/sequential.py", line 167, in forward
..
RuntimeError: KeyeError: edge_vec

This bug happen only when I try to deploy sevenn model with sevenn version 0.13.
Is there any unknown dependence?

Deprecation of torch.jit.script, torch,jit,save, torch,trace_method may be one reason considering that there are DeprecationWarnings during deploying sevenn model using sevenn version 0.12

Or integration of ReEWC made some errors?

I have no idea, but I hope this information to help your maintenance.

[lee2tae]

I don't exactly know whether this is related to version dependency, but I have seen similar errors before.

This may be caused by a mismatch between your compiled LAMMPS binary and the deployed model. PR #297 changed sevenn/pair_e3gnn/pair_e3gnn.cpp so that the serial pair style now passes edge_vec instead of pos .
If your binary was built before this change, KeyError: edge_vec can occur.

Could you check whether the pair_e3gnn.cpp used in your LAMMPS build matches the current sevenn/pair_e3gnn/pair_e3gnn.cpp?

[ensol-JKY]

Because there was a complicated uninstall and reinstall process in order to solve the KeyError problem, currently, I have no idea on what was the exact pair_e3gnn.cpp version.

However, I have compared the cpp files and found that there is inconsistency between them. There might be some mistake re-compiling LAMMPS after updating sevenn. (I have also re-compiled LAMMPS using sevenn patch_lamms after the version update but still, there was the same problem.)

Anyway, I really appreciate you for kind suggestion.