Installing mlatom

Installing MLatom (and the AIQM3 add-ons)

When to use

A user wants to install MLatom locally, or install the AIQM3 methods (the aitomic-addons package), or is hitting an install error such as "No matching distribution found for aitomic-addons".

No installation needed for a quick try: MLatom and many of these methods run online on the Aitomistic Hub.

Prerequisites

• Operating system: MLatom's compiled backends and the AIQM3 add-ons run on Linux (x86-64 or ARM64/aarch64). There are no macOS or Windows wheels for the add-ons — on those systems use the online Hub, a Linux container, or WSL2.
• Python: 3.9, 3.10, or 3.11 (recommended).
  • Python ≤ 3.8 is not supported — the add-ons publish no wheels below 3.9.
  • Python ≥ 3.12 has no wheel for the torch==2.1.2 that AIQM3 pins — stay on 3.9–3.11.
• A clean environment — install into a fresh conda/venv, not an old base environment, to avoid version conflicts.

Steps

Create a fresh Python 3.11 environment and install MLatom + AIQM3:

conda create -n aiqm3 python=3.11 -y
conda activate aiqm3
pip install -U "numpy<2" torch==2.1.2 torchani==2.2.3 mlatom aitomic-addons joblib pyscf geometric

AIQM3 also uses D3 dispersion — install s-dftd3 and point dftd3bin at the executable:

conda install -c conda-forge dftd3
export dftd3bin=$CONDA_PREFIX/bin/s-dftd3

For plain MLatom without the add-ons, pip install -U mlatom plus the dependencies you need is enough — see the installation guide.

Verify

import mlatom as ml

mol = ml.data.molecule.from_xyz_string('''3

O    0.00000    0.00000    0.11779
H    0.00000    0.75545   -0.47116
H    0.00000   -0.75545   -0.47116
''')
aiqm3 = ml.methods(method='AIQM3')
opt = ml.optimize_geometry(model=aiqm3, initial_molecule=mol).optimized_molecule
print(opt.energy)   # ~ -76.3787 Hartree

The neural-network parameters download automatically on first use, together with a one-time license notice.

Troubleshooting

ERROR: No matching distribution found for aitomic-addons (from versions: none) pip reached PyPI but found no wheel matching your Python version, OS, or CPU. Usual causes:

• Python is 3.7 or 3.8 → use 3.9–3.11. Check with python --version.
• The platform is macOS/Windows or a non-x86-64/aarch64 CPU → only Linux x86-64/aarch64 wheels are published, with no source distribution to fall back on.

Fix: create a fresh Python 3.11 environment on a supported Linux platform.

The same message for torch==2.1.2 usually means your Python is 3.12 or newer — use 3.9–3.11.

Temporary failure in name resolution / repeated Retrying This is a network problem, not a compatibility one: the machine cannot reach PyPI (broken DNS, a proxy, or an offline node). The tell-tale sign is Retrying / name-resolution warnings printed above the No matching distribution line. Fix DNS/proxy/network, or install from a local PyPI mirror — the packages themselves are fine.

UnicodeDecodeError from xtb Run under a UTF-8 locale: export PYTHONUTF8=1.

Cite

• MLatom — see Cite this repository on the MLatom GitHub page (the citation is also printed in the MLatom output and listed at MLatom.com).
• AIQM3 — Yuxinxin Chen, Yi-Fan Hou, Roman Zubatyuk, Olexandr Isayev, Pavlo O. Dral. AIQM3: Targeting Coupled-Cluster Accuracy with Semi-Empirical Speed across Seven Main Group Elements. J. Chem. Theory Comput. 2026. https://doi.org/10.1021/acs.jctc.5c01794