mir-group/allegro

Deploying a model

Opened this issue · 3 comments

sogenyi commented

Dear friends,

I trained a model earlier with the

model_builders:

  • allegro.model.Allegro

the typical model builders from nequip can still be used:

  • PerSpeciesRescale
  • ForceOutput
  • RescaleEnergyEtc

After deploying it for lammps MD simulation i encountered the problem:

Exception: Argument passed to at() was not in the map

I followed the suggestion in this link for a fix:
mir-group/nequip#69 (comment)

The new problem i am encountering now is:

Traceback (most recent call last):
File "/leonardo/home/userexternal/sogenyi0/NEQUIP1/myenv/bin/nequip-deploy", line 8, in
sys.exit(main())
File "/leonardo/home/userexternal/sogenyi0/NEQUIP1/myenv/lib/python3.10/site-packages/nequip/scripts/deploy.py", line 219, in main
model = model_from_config(config, deploy=True)
File "/leonardo/home/userexternal/sogenyi0/NEQUIP1/myenv/lib/python3.10/site-packages/nequip/model/_build.py", line 96, in model_from_config
model = builder(**params)
File "/leonardo/home/userexternal/sogenyi0/NEQUIP1/myenv/lib/python3.10/site-packages/allegro/model/_allegro.py", line 32, in Allegro
builder_utils.add_avg_num_neighbors(
File "/leonardo/home/userexternal/sogenyi0/NEQUIP1/myenv/lib/python3.10/site-packages/nequip/model/builder_utils.py", line 35, in add_avg_num_neighbors
raise ValueError("avg_num_neighbors = auto but initialize is False")
ValueError: avg_num_neighbors = auto but initialize is False

Please i can anyone help me out?

Linux-cpp-lisp commented

You need StressForceOutput instead of ForceOutput as pair_allegro expects the virial to be computed.

Please check that you are using the latest main branches of nequip and pair_allegro as well.

sogenyi commented

Thank for the response. I have checked

You need StressForceOutput instead of ForceOutput as pair_allegro expects the virial to be computed.

Please check that you are using the latest main branches of nequip and pair_allegro as well.

I am using a latest main branch of nequip and in the config file for the redeployment i effected the necessary changes needed.
However, the default_dtype: float32 was used in my training session. An attempt to change it to float64 kept giving error of inconsistency.

sogenyi commented

Hello friends,

I finally had to do fresh installation of most recent versions of the packages and do a retrain this
time with right settings in my config yaml file. It worked when i deployed it. Just a tip to all, try to do a test run first with loose parameters and deploy for MD simulation to a certain that your installations is correct.