About using compute_allegro

[SeungJay]

Dear developers,
I'm trying to use compute_allegro to calculate polarisability of the materials.

Firstly, I've succeeded in compiling LAMMPS with normal_nequip interface but failed with allegro-pol.

Additionally, I couldn't find any detailed instruction about using allegro-pol python package and its appropriate pair and compute modules for LAMMPS interface.
I think current patch_lammps.sh seems to not update the compute_allegro in LAMMPS.

Please could you elaborate the installation process and the procedure about how to use this compute with the developed python tools? (https://github.com/mir-group/allegro-pol/)

Many thanks,
Jay

[cw-tan]

Hi @SeungJay

allegro-pol requires the old pre-v0.7.0 NequIP infrastructure, so you need to use the fix branch of pair_nequip_allegro for compatibility: https://github.com/mir-group/pair_nequip_allegro/tree/fix. The instructions are in the README (i.e. instructions for the old infrastructure's way of installing, which is largely similar and confined to the TorchScript model format)

There are some LAMMPS files here: https://github.com/mir-group/allegro-pol/tree/main/lammps_infiles

[SeungJay]

Thanks @cw-tan ,
Now I've compiled LAMMPS with the old nequip (0.7.0) version including compute_allegro.h and compute_allegro.cpp. During the compilation, I found an issue originating from line 6264 and 7274 in compute_allegro.cpp since the recent lammps version seems to have no "error->message". Thus, I just make this as comment and compiled it well.

After this, I installed allgro-pol (https://github.com/mir-group/allegro-pol/tree/main) separately. When I'm trying to run the test (nequip-train configs/BaTiO3.yaml), I got this error message.

/scratch/x3051a03/miniconda3/envs/allegro-pol/lib/python3.11/site-packages/nequip/__init__.py:20: UserWarning: !! PyTorch version 1.13.1 found. Upstream issues in PyTorch versions 1.13.* and 2.* have been seen to cause unusual performance degredations on some CUDA systems that become worse over time; see https://github.com/mir-group/nequip/discussions/311. The best tested PyTorch version to use with CUDA devices is 1.11; while using other versions if you observe this problem, an unexpected lack of this problem, or other strange behavior, please post in the linked GitHub issue.
  warnings.warn(
wandb: Currently logged in as: seungjae-shin (seungjae-shin-unist) to https://api.wandb.ai. Use `wandb login --relogin` to force relogin
wandb: Tracking run with wandb version 0.21.0
wandb: Run data is saved locally in /scratch/x3051a03/LAMMPS/with_nequip_0.7.0/allegro-pol/wandb/run-20250806_012830-eI0u6IfL6m_kpnxcrrkfGKf8Da29ctH2dKBBy_YEH4o
wandb: Run `wandb offline` to turn off syncing.
wandb: Syncing run BaTiO3
wandb: ⭐️ View project at https://wandb.ai/seungjae-shin-unist/BaTiO3
wandb: 🚀 View run at https://wandb.ai/seungjae-shin-unist/BaTiO3/runs/eI0u6IfL6m_kpnxcrrkfGKf8Da29ctH2dKBBy_YEH4o
Torch device: cuda
Traceback (most recent call last):
  File "/scratch/x3051a03/miniconda3/envs/allegro-pol/lib/python3.11/site-packages/nequip/utils/auto_init.py", line 243, in instantiate
    instance = builder(**positional_args, **final_optional_args)
               ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/scratch/x3051a03/miniconda3/envs/allegro-pol/lib/python3.11/site-packages/nequip/train/loss.py", line 83, in __init__
    self.funcs[key] = find_loss_function(
                      ^^^^^^^^^^^^^^^^^^^
  File "/scratch/x3051a03/miniconda3/envs/allegro-pol/lib/python3.11/site-packages/nequip/train/_loss.py", line 205, in find_loss_function
    return SimpleLoss(name, params)
           ^^^^^^^^^^^^^^^^^^^^^^^^
  File "/scratch/x3051a03/miniconda3/envs/allegro-pol/lib/python3.11/site-packages/nequip/train/_loss.py", line 43, in __init__
    eval(parent_module) if parent_module else torch.nn,
    ^^^^^^^^^^^^^^^^^^^
  File "<string>", line 1, in <module>
NameError: name 'allegro_pol' is not defined

The above exception was the direct cause of the following exception:

Traceback (most recent call last):
  File "/scratch/x3051a03/miniconda3/envs/allegro-pol/bin/nequip-train", line 8, in <module>
    sys.exit(main())
             ^^^^^^
  File "/scratch/x3051a03/miniconda3/envs/allegro-pol/lib/python3.11/site-packages/nequip/scripts/train.py", line 94, in main
    trainer = fresh_start(config)
              ^^^^^^^^^^^^^^^^^^^
  File "/scratch/x3051a03/miniconda3/envs/allegro-pol/lib/python3.11/site-packages/nequip/scripts/train.py", line 200, in fresh_start
    trainer = Trainer(model=None, **Config.as_dict(config))
              ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/scratch/x3051a03/miniconda3/envs/allegro-pol/lib/python3.11/site-packages/nequip/train/trainer.py", line 338, in __init__
    self.loss, _ = instantiate(
                   ^^^^^^^^^^^^
  File "/scratch/x3051a03/miniconda3/envs/allegro-pol/lib/python3.11/site-packages/nequip/utils/auto_init.py", line 245, in instantiate
    raise RuntimeError(
RuntimeError: Failed to build object with prefix `loss` using builder `Loss`

After I attached this line in my slurm script, it gave the same error.
export PYTHONPATH=/scratch/x3051a03/LAMMPS/with_nequip_0.7.0/allegro-pol:$PYTHONPATH

Please could you take a look at it, and help me deal with this issue?
BTW, the results folder seems to be well produced.

Many thanks,
Jay

cf) It would be really greatful if you can share the xx.pth files to test my lammps compilation is well conducted. I've tried to run some lammps in the allegro-pol/lammps_infiles, I couldn't.