lammps-mdi 0.1.0

MDI engine drivers for LAMMPS — MACE and other ML forcefields via MolSSI Driver Interface

lammps-mdi

MDI engine drivers for LAMMPS — run ML forcefields (MACE, and future models) on GPU via the MolSSI Driver Interface, communicating with a standard LAMMPS binary (no Kokkos compilation needed).

Designed for use with SEAMM, but works with any LAMMPS workflow that supports MDI.

How it works

LAMMPS acts as an MDI driver: it handles atom positions, neighbor lists (at the coarse level), and time integration. The lammps-mdi engine process acts as an MDI engine: it receives coordinates from LAMMPS each step, evaluates the ML model on GPU, and returns energies, forces, and (if periodic) the stress tensor.

mpirun -np 1  mace-mdi  -mdi "..."   ← GPU process: MACE on A100
         : -np 1  lmp  -mdi "..." -in input.dat  ← CPU process: time integration

The two processes communicate over MPI via the MDI protocol.

Supported engines

Engine	Status	Notes
MACE	✅	MACE-torch ≥ 0.3, vesin-torch neighbor lists, cuEquivariance optional
NequIP	Planned
SevenNet	Planned

Installation

See INSTALL.md for full HPC instructions. Short version:

# 1. Load your LAMMPS module (provides Python, numpy, mpi4py, MDI)
module load LAMMPS/...

# 2. Create a venv that inherits the module stack
python -m venv --system-site-packages ~/venvs/lammps-mdi
source ~/venvs/lammps-mdi/bin/activate

# 3. Install PyTorch with the right CUDA wheel (check your CUDA version first)
lammps-mdi install-torch    # prints the correct command
# e.g.:
pip install torch --index-url https://download.pytorch.org/whl/cu121

# 4. Install lammps-mdi
pip install lammps-mdi[gpu]

# 5. Install bundled shell scripts
lammps-mdi install-scripts

# 6. Verify
lammps-mdi check

Usage

As a console script (recommended)

SEAMM_FF=/path/to/model.model \
mpirun --mca mpi_yield_when_idle 1 \
    -np 1 mdi_bind.sh mace-mdi -mdi "-role ENGINE -name MACE -method MPI" \
    : -np 1 mdi_bind.sh lmp -mdi "-role DRIVER -name LAMMPS -method MPI" -in input.dat

The mace-mdi command accepts several options:

mace-mdi --help

  -mdi MDI_STRING      MDI initialization string [required]
  --model PATH         Path to MACE model (overrides SEAMM_FF)
  --device DEVICE      PyTorch device (default: cuda:0)
  --dtype {float32,float64}
  --enable-cueq        Enable cuEquivariance acceleration
  --enable-oeq         Enable openEquivariance acceleration
  --log-level LEVEL    DEBUG / INFO / WARNING / ERROR

From lammps.ini (SEAMM)

[local]
installation = conda   # or modules, or local

gpu-code = mpirun --mca mpi_yield_when_idle 1 \
    -np 1 ~/SEAMM/bin/mdi_bind.sh \
    mace-mdi -mdi "-role ENGINE -name MACE -method MPI" \
    : -np 1 ~/SEAMM/bin/mdi_bind.sh \
    lmp -mdi "-role DRIVER -name LAMMPS -method MPI"

As a Python library

from lammps_mdi import MACEEngine

engine = MACEEngine(
    model_path="/path/to/model.model",
    device="cuda:0",
    default_dtype="float32",
    enable_cueq=True,
)
engine.run("-role ENGINE -name MACE -method MPI")

Shell scripts

The package bundles four helper scripts, installed via lammps-mdi install-scripts:

Script	Purpose
mdi_bind.sh	Binds engine (rank 0) to GPU + NUMA-local CPUs, driver (rank 1) to adjacent CPUs; starts nvidia-smi monitor. For standalone machines.
mdi_monitor.sh	Lightweight wrapper for SLURM/PBS: only starts GPU monitoring. Scheduler handles binding.
gpu_bind.sh	Per-rank GPU binding for native Kokkos LAMMPS (approach A).
cpu_bind.sh	CPU-only binding using L3 cache groups (EPYC 7763).

The CPU/GPU mappings in mdi_bind.sh, gpu_bind.sh, and cpu_bind.sh are currently hard-coded for a dual-GPU EPYC 7763 system. They will be made configurable in a future release.

Requirements

Package	Source	Notes
Python ≥ 3.10	HPC module
numpy	HPC module	Do not reinstall
mpi4py	HPC module
pymdi ≥ 1.4	pip	PyPI package for import mdi
torch (CUDA)	pip (special index)	Install before lammps-mdi
mace-torch ≥ 0.3	pip
matscipy ≥ 0.8	pip	CPU fallback neighbor list
pint ≥ 0.20	pip	Unit conversion
vesin-torch ≥ 0.3	pip (optional)	GPU neighbor lists, strongly recommended
cuequivariance*	pip (optional)	NVIDIA cuEquivariance acceleration

Contributing

Issues and pull requests are welcome at https://github.com/molssi-seamm/lammps-mdi.

License

MIT — see LICENSE.