mace-torch

library0.3.16pypypi✓ verified 84d ago

MACE (Multi-Atomic Cluster Expansion) is a machine learning force field architecture based on equivariant message passing, implemented in PyTorch. Version 0.3.16 requires Python >=3.9 and is actively maintained on GitHub (ACEsuit/mace). Release cadence is irregular, approximately every 3-6 months.

ai-ml analytics

pip install mace-torch

github ↗homepage ↗

INSTALL

IMPORT

SIG · MACE-TORCH

mace-torch

ai-mlpythonv0.3.16

harness data pending

Install & Compatibility

Where this runs

No compatibility data collected yet for this library.

Code

Verified usage

Verified import paths — ran on the pinned version, not inferred.

MACECalculator

✓ from mace.calculators import MACECalculator

✗ from mace import MACECalculator

MACECalculator is in the calculators submodule, not top-level

mace_mp

✓ from mace.calculators import mace_mp

✗ from mace import mace_mp

mace_mp is a function in calculators; direct import fails

load_model

✓ from mace.calculators import load_model

Initialize the MACE-MP-0 model and compute energy for a water molecule.

import torch from ase import Atoms from mace.calculators import MACECalculator # Download default MACE-MP-0 model calc = MACECalculator(model_path='large', device='cuda' if torch.cuda.is_available() else 'cpu') # Create a water molecule atoms = Atoms('H2O', positions=[[0, 0, 0], [0, 0, 0.957], [0.957, 0, 0]], cell=[10, 10, 10]) atoms.calc = calc energy = atoms.get_potential_energy() print(f"Energy: {energy:.6f} eV")

Debug

Known issues

breakingIn version 0.3.0+, the model path API changed: 'large' and 'small' strings now download the model automatically instead of using 'MACE_MP_0' keyword.

fix

Use model_path='large' or 'medium' or 'small' to download the default models. For custom models, provide the path to .model file.

affects: <0.3.0

gotchaMACECalculator expects ASE Atoms objects; directly passing numpy arrays or torch tensors will raise errors.

fix

Always wrap atomic data in an ase.Atoms object before passing to calculator.

affects: all

gotchaThe model requires considerable GPU memory for large systems (thousands of atoms). Out-of-memory errors are common if batch size is too large.

fix

Reduce system size or use CPU offloading. For very large systems, consider splitting into chunks.

affects: all

Errors

Common errors & fixes

AttributeError: module 'mace' has no attribute 'calculators'

The mace-torch package is not installed; only the mace package (old version) is installed.

fix

pip install mace-torch --upgrade

KeyError: 'MACE_MP_0'

The model name string is not a valid key; in newer versions models are downloaded via 'large'/'small' instead.

fix

Use calc = MACECalculator(model_path='large') instead of model_path='MACE_MP_0'

RuntimeError: Expected tensor to be on GPU, but found tensor on CPU

Mixing devices: model was loaded on GPU but input atoms are on CPU or vice versa.

fix

Ensure device consistency: pass device='cuda' or device='cpu' explicitly to MACECalculator.

ImportError: cannot import name 'MACECalculator' from 'mace'

Using the old top-level import path that changed in v0.3.0.

fix

Use 'from mace.calculators import MACECalculator'.

Upgrade

Version history

0.3.16latest on PyPI · released May 10, 2026

Audit

Dependencies

torchrequiredCore computation and tensor operations

e3nnrequiredEquivariant neural network operations

pytorch-lightningoptionalTraining framework (optional if using custom training loop)

Agent activity

44 hits · last 30 days

node

OpenAI (training)

Resources

githubhttps://github.com/ACEsuit/mace ↗

homepagehttps://github.com/ACEsuit/mace ↗