Version: 2026.03

Offline Installation

For Lonxun Supercomputing Cloud (Mcloud) users, MatPL is preinstalled and ready to load. The offline package provides a GPU version.

The offline package bundles MatPL-2026.3, the open-source MatPL-2026.3 LAMMPS interface, and the required Python environment. The target machine must provide GCC 8.x or later, CUDA 11.8 or later, OpenMPI 4.1.4 or later, and an NVIDIA GPU. NN and Linear models additionally require the Intel toolchain (ifort, icc, and MKL).
Because MatPL-2026.3 does not improve CPU-only training or simulation, no online or offline CPU package is provided. CPU-only users should use MatPL-2025.3.
The offline package includes PyTorch 2.2. The latest package is matpl-2026.3-update3. Compared with update1, it adds type-wise ZBL support to NEP, accelerates LAMMPS NEP Kokkos on NVIDIA GPUs, and extends the six-component LAMMPS Kokkos virial to nine components to support heat-flux calculations. The bundled LAMMPS version is stable_22Jul2025_update4.

Download the Offline Package

Method 1 (recommended): Request the package by email from matpl@pwmat.com, wuxingxing@pwmat.com, or support@pwmat.com. Downloads from the emailed link are significantly faster than from Baidu Netdisk.

Method 2: Download from Baidu Netdisk. If the link expires, contact matpl@pwmat.com, wuxingxing@pwmat.com, or support@pwmat.com: 👉 Download the offline package. Access code: pwmt.

Method 3: Visit the MatPL-2026.3_update3 GitHub release and download the following files:

matpl-2026.3-update3.sh.tar.gz.part_aa  matpl-2026.3-update3.sh.tar.gz.part_ab  matpl-2026.3-update3.sh.tar.gz.part_ac  matpl-2026.3-update3.sh.tar.gz.part_ad  matpl-2026.3-update3.sh.tar.gz.part_ae

Extract the Package

Because the offline package is large, it is split into several smaller files. Reassemble and extract it as follows:

# Combine the parts into one archive
cat matpl-2026.3-update3.sh.tar.gz.part_aa  matpl-2026.3-update3.sh.tar.gz.part_ab  matpl-2026.3-update3.sh.tar.gz.part_ac  matpl-2026.3-update3.sh.tar.gz.part_ad  matpl-2026.3-update3.sh.tar.gz.part_ae > matpl-2026.3-update3.sh.tar.gz
# Extract the archive
tar -xzvf matpl-2026.3-update3.sh.tar.gz

Extraction produces the following files: matpl-2026.3-update3.sh， check_offenv.sh

Check Compiler Versions

Most installation failures result from incompatible compiler versions. Use the provided check_offenv.sh script to inspect the environment:

bash check_offenv.sh

The script lists the required compiler versions and those detected in the current environment. A valid result looks like this:

========================================
      Environment Check Starting
========================================

=== Checking ifort compiler and MKL library ===
✓ ifort version: 19.1 (>= 19.1)
✓ MKL library is installed

=== Checking GCC version ===
✓ GCC version: 8 (>= 8.0)

=== Checking CUDA version ===
✓ CUDA version: 11.8.89 (>= 11.8)

=== Checking nvcc availability ===
✓ nvcc command exists

=== Checking OpenMPI ===
✓ OpenMPI found (via ompi_info), version: 4.1.6
✓ OpenMPI version meets requirement (>= 4.0)

========================================
        Environment Summary
========================================
✓ Environment check completed. All requirements are satisfied.
========================================

Run the Installation Command

After the environment check, install the package with:

bash matpl-2026.3-update3.sh [-jN] [-m nn] [-a ARCH] [-d] [-u] [-h]

-jN: N is the number of parallel compilation cores. For example, bash matpl-2026.3-update3.sh -j4 uses four cores. The default is one core.
-m nn: Also compiles the Fortran code for Linear and NN models. This requires Intel ifort, icc, and MKL. Fortran code is not compiled by default.
-a ARCH: Selects the Kokkos CUDA architecture used to build lammps-2026.3. The default is AMPERE86. For example:
```
bash matpl-2026.3-update3.sh -a AMPERE80
```
ARCH is a Kokkos CUDA architecture suffix, such as AMPERE80, AMPERE86, ADA89, or HOPPER90. For additional architectures, see https://docs.lammps.org/Build_extras.html#kokkos.
-d: Enables 64-bit-precision NEP inference when building lammps-2026.3:
```
-DPREC_NEPINFER=ON
```
When enabled, lammps-2026.3 is built in build-64. By default, this option is disabled and the build directory is build.
-u: Extracts the package without building it, for environments that require installation in separate stages. The extracted contents include the lammps-2026.3 source directory, the MatPL-2026.3 source directory, and the matpl-2026.3 Python environment.

To verify the installation:

After compilation, the MatPL-2026.3 directory has the following structure:

MatPL-2026.3
    ├── lammps-2026.3/
    ├── matpl-2026.3/
    ├── MatPL-2026.3/
    └── matpl-env.sh

MatPL-2026.3 is the machine-learning force-field training platform.
matpl-2026.3 is the Python environment.
lammps-2026.3 is the LAMMPS interface for DP and NEP force fields.
matpl-env.sh defines all environment variables for MatPL-2026.3 and lammps-2026.3.

Load the Environment

Before using the GPU version of MatPL, load the same CUDA environment used during compilation. NN or Linear training also requires MKL. Then run the following command to load the Python environment, MatPL-2026.3, and LAMMPS:

source /the/path/of/MatPL-2026.3/matpl-env.sh

# For the Fortran LAMMPS interface:
source /the/path/of/MatPL-2026.3/matpl-fortran-env.sh

You can also load each component separately:

## Step 1. Load the Python environment
source /the/path/MatPL-2026.3/matpl-2026.3/bin/activate
## Step 2. Load MatPL-2026.3
source /the/path/MatPL-2026.3/MatPL-2026.3/env.sh
## Step 3. Load LAMMPS
source /the/path/MatPL-2026.3/lammps-2026.3/env.sh

LAMMPS Packages Included in the Offline Installation

The LAMMPS offline package includes the following commonly used packages:

CG-SPICA CLASS2 COLLOID COLVARS COMPRESS CORESHELL DIELECTRIC DIFFRACTION DIPOLE DPD-BASIC DPD-MESO DPD-REACT DPD-SMOOTH DRUDE EFF ELECTRODE EXTRA-COMMAND EXTRA-COMPUTE EXTRA-FIX EXTRA-DUMP EXTRA-MOLECULE EXTRA-PAIR FEP GRANULAR INTEL INTERLAYER KSPACE LATBOLTZ LEPTON MANIFOLD MANYBODY MC MEAM MGPT MISC MOFFF MOLECULE MOLFILE OPENMP OPT ORIENT PERI PHONON PLUMED POEMS PTM PYTHON QEQ QMMM QTB REACTION REAXFF REPLICA RIGID SHOCK SMTBQ SPH SPIN SRD TALLY UEF VORONOI YAFF