Issue on integrating lammps with MACE potential

[aishwaryo]

Dear developers, I was attempting to install the CPU implementation of lammps+MACE from https://mace-docs.readthedocs.io/en/latest/guide/lammps.html. This clones and proceeds with the "mace" branch of the repo.

On the step with cmake I am getting the following error :

CMake Error at Modules/Packages/ML-MACE.cmake:3 (find_package): By not providing "FindTorch.cmake" in CMAKE_MODULE_PATH this project has asked CMake to find a package configuration file provided by "Torch", but CMake did not find one.

Could not find a package configuration file provided by "Torch" with any of the following names:

TorchConfig.cmake torch-config.cmake

Add the installation prefix of "Torch" to CMAKE_PREFIX_PATH or set "Torch_DIR" to a directory containing one of the above files. If "Torch" provides a separate development package or SDK, be sure it has been installed. Call Stack (most recent call first): CMakeLists.txt:526 (include)

-- Configuring incomplete, errors occurred! '

Could you please assist in resolving this?

[stargolike]

Do you install torch in your machine? i met this error, and i install suitable version torch then the error is disappeared.

[wcwitt]

Adding to @stargolike's comment, have you followed this part of the instructions (using the appropriate libtorch version)?

wget https://download.pytorch.org/libtorch/cpu/libtorch-shared-with-deps-1.13.0%2Bcpu.zip
unzip libtorch-shared-with-deps-1.13.0+cpu.zip
rm libtorch-shared-with-deps-1.13.0+cpu.zip

[aishwaryo]

Hi! I fixed this issue with correct installation of libtorch. However now when I try to run a simulation I get the error :

"pair_mace does not support vflag_atom."

It possibly comes from a compute of stress/atom. I am attaching the LAMMPS script and .pt files as well. This script has previously been tested with other potentials and it at-least runs. Any help is appreciated! send.zip

[wcwitt]

We don't currently offer a way to extract the local stress. Is that essential for your application?

[gabor1]

It would be nice to put that on our todo list

[bernstei]

If correctly formulated (namely the strain gradient of a well-defined local energy) it'd enable thermal conductivity, which could be useful

[pobo95]

@aishwaryo You can slove the vflg_atom issue using by my repository and this code to pair with lammps. @wcwitt I want to pull this code but I don't know which branch is correct one. Can you make a new one?

[aishwaryo]

@pobo95 i have tried using your repo and code to pair with lammps. However, while training I am getting the following over all the epochs.

INFO: Epoch 12: loss=0.0000, RMSE_E_per_atom=243.9 meV, RMSE_F=0.0 meV / A

This does not happen while using the original MACE repo. Could you suggest a way through? I have also attached the log and the xyz file so that you can check. combined.zip

MACE_model_run-123.log

[pobo95]

@aishwaryo

Maybe you didn't use test sets, don't you?

Can you share the result with my data train_validate.zip set using this line?

mace_run_train     --name="mace"     --train_file="train.xyz"     --valid_fraction=0.05     --test_file="validate.xyz"     --E0s={14:-0.08755009}     --model="MACE"     --num_interactions=2     --hidden_irreps='64x0e + 64x1o'     --max_L=1     --correlation=2     --r_max=5.0     --forces_weight=1000     --energy_weight=10     --batch_size=5     --valid_batch_size=2     --max_num_epochs=5000   --compute_stress=”true”  --start_swa=450     --scheduler_patience=5     --patience=15     --eval_interval=3     --ema     --swa     --swa_forces_weight=10     --default_dtype="float64"    --device=cuda     --seed=123     --restart_latest     --save_cpu --error_table=PerAtomRMSEstressvirials 

[pobo95]

@aishwaryo

When I test with your data sets withi this line

mace_run_train     --name="mace"     --train_file="combined.xyz"     --valid_fraction=0.05     --E0s=average     --model="MACE"     --num_interactions=2     --hidden_irreps='64x0e + 64x1o'     --max_L=1     --correlation=2     --r_max=5.0     --forces_weight=1000     --energy_weight=10     --batch_size=5     --valid_batch_size=2     --max_num_epochs=5000   --compute_stress=”true”  --start_swa=450     --scheduler_patience=5     --patience=15     --eval_interval=3     --ema     --swa     --swa_forces_weight=10     --default_dtype="float64"    --device=cuda     --seed=123     --restart_latest     --save_cpu --error_table=PerAtomRMSEstressvirials 

It seems fine.

When I chckded your log file, looks like you didn't take max_L.

It can be the resaon too.