[Bug]: ValueError: Buffer dtype mismatch, expected 'const int64_t' but got 'long'

[S1owFade]

Contact Details

No response

Bug Description

D:\Anaconda\envs\mattersim\python.exe "D:\python object\learn\mattersim\test.py"
2025-04-18 13:47:05.278 | INFO | main::8 - Running MatterSim on cuda
2025-04-18 13:47:05.279 | INFO | mattersim.forcefield.potential:from_checkpoint:877 - Loading the pre-trained mattersim-v1.0.0-1M.pth model
Traceback (most recent call last):
File "D:\python object\learn\mattersim\test.py", line 12, in
logger.info(f"Energy (eV) = {si.get_potential_energy()}")
File "D:\Anaconda\envs\mattersim\lib\site-packages\ase\atoms.py", line 772, in get_potential_energy
energy = self._calc.get_potential_energy(self)
File "D:\Anaconda\envs\mattersim\lib\site-packages\ase\calculators\abc.py", line 26, in get_potential_energy
return self.get_property(name, atoms)
File "D:\Anaconda\envs\mattersim\lib\site-packages\ase\calculators\calculator.py", line 515, in get_property
self.calculate(atoms, [name], system_changes)
File "D:\python object\learn\mattersim\src\mattersim\forcefield\potential.py", line 1357, in calculate
dataloader = build_dataloader(
File "D:\python object\learn\mattersim\src\mattersim\datasets\utils\build.py", line 75, in build_dataloader
graph = convertor.convert(graph.copy(), energy, force, stress, **kwargs)
File "D:\python object\learn\mattersim\src\mattersim\datasets\utils\convertor.py", line 243, in convert
) = get_fixed_radius_bonding(atoms, self.twobody_cutoff, pbc=pbc)
File "D:\python object\learn\mattersim\src\mattersim\datasets\utils\convertor.py", line 131, in get_fixed_radius_bonding
) = find_points_in_spheres( # noqa: E501
File "src\pymatgen\optimization\neighbors.pyx", line 46, in pymatgen.optimization.neighbors.find_points_in_spheres
ValueError: Buffer dtype mismatch, expected 'const int64_t' but got 'long'

MatterSim Version

1.1.3.dev7+dirty

Python Version

3.10.16

Reproduction Steps

1.Install with:
conda create -n mattersim python=3.10
conda activate mattersim
git clone https://github.com/microsoft/mattersim.git
cd .\mattersim\
pip install -e .
2.Test
After installing MatterSim, i confirmed by "pip show mattersim":
Name: mattersim
Version: 1.1.3.dev7+dirty
Summary: MatterSim: A Deep Learning Atomistic Model Across Elements, Temperatures and Pressures.
Home-page: https://github.com/microsoft/mattersim
Author:
Author-email: Han Yang hanyang@microsoft.com, Hongxia Hao hongxiahao@microsoft.com, Jielan Li jielanli@microsoft.com, Ziheng Lu zihenglu@microsoft.com
License:
Location: d:\anaconda\envs\mattersim\lib\site-packages
Editable project location: D:\python object\learn\mattersim
Requires: ase, atomate2, azure-identity, azure-storage-blob, deprecated, e3nn, emmet-core, loguru, mp-api, numpy, opt_einsum_fx, phonopy, pydantic, pymatgen, scikit-learn, seekpath, torch, torch-ema, torch_geometric, torch_runstats, torchaudio, torchmetrics, torchvision, wandb
Required-by:

To test if it can work by using the example: "A minimal example using ASE calculator"

Expected Behavior

conducted without error

Actual Behavior

An error came out:
ValueError: Buffer dtype mismatch, expected 'const int64_t' but got 'long'

Error Logs

Code of Conduct

• I agree to follow the project's Code of Conduct

[austindonnelly]

I see this too.

If I edit mattersim/datasets/utils/convertor.py: 117

-    pbc_ = np.array(structure.pbc, dtype=int)
+    pbc_ = np.array(structure.pbc, dtype=np.int64)

then the example code runs to completion.

[austindonnelly]

I think this might be a Windows vs. Linux difference. The example runs just fine on WSL, but not on native Windows.