Skip to content

Installation

This guide covers how to install MACE Inference on your system.

Requirements

  • Python: 3.9, 3.10, or 3.11
  • Operating System: Linux, macOS, or Windows
  • Hardware: CPU or NVIDIA GPU (for acceleration)

Installation Methods

pip install mace-inference

# Install PyTorch with CUDA first
pip install torch --index-url https://download.pytorch.org/whl/cu118

# Then install mace-inference
pip install mace-inference

Method 2: From Source

# Clone the repository
git clone https://github.com/lichman0405/mace-inference.git
cd mace-inference

# Install in development mode
pip install -e ".[dev]"

Method 3: Conda Environment

We recommend using a dedicated conda environment:

# Create environment
conda create -n mace-inference python=3.10
conda activate mace-inference

# Install mace-inference
pip install mace-inference

# Create environment
conda create -n mace-inference-gpu python=3.10
conda activate mace-inference-gpu

# Install PyTorch with CUDA
conda install pytorch pytorch-cuda=11.8 -c pytorch -c nvidia

# Install mace-inference with GPU support
pip install mace-inference[gpu]

Optional Dependencies

D3 Dispersion Correction

For DFT-D3 dispersion correction support:

pip install torch-dftd

Development Tools

For contributing to the project:

pip install -e ".[dev]"

This includes:

  • pytest - Testing framework
  • ruff - Linting
  • mypy - Type checking

Verify Installation

After installation, verify everything works:

from mace_inference import MACEInference

# This will download the model on first run
calc = MACEInference(model="medium", device="cpu")
print("✓ MACE Inference installed successfully!")

Or use the CLI:

mace-inference --help

Troubleshooting

CUDA Not Detected

If GPU is not detected:

import torch
print(f"CUDA available: {torch.cuda.is_available()}")
print(f"CUDA version: {torch.version.cuda}")

Make sure your PyTorch installation matches your CUDA version.

Model Download Issues

Models are cached in ~/.cache/mace/. If download fails:

  1. Check internet connection
  2. Try manual download from MACE Models
  3. Specify local model path:
calc = MACEInference(model="/path/to/model.model")

Memory Issues

For large systems, try:

# Use lower precision
calc = MACEInference(model="medium", default_dtype="float32")

# Or reduce batch size in dynamics
result = calc.run_md(atoms, steps=1000, save_interval=100)

Next Steps