Warning
metatrain is still very early in the concept stage. You should not use it for anything important.
This is a repository for training and evaluating machine learning models from various architectures for atomistic systems. The only requirement is for an architecture to be able to take metatensor objects as inputs/outputs and have to be JIT compilable using TorchScript. The architectures do not need to live entirely in this repository: in the most extreme case, this repository can simply contain a wrapper to an external architecture.
metatrain is a command line interface (cli) to train and evaluate atomistic
models of various architectures. It features a common yaml option inputs to
configure training and evaluation. Trained models are exported as standalone files that
can be used directly in various molecular dynamics (MD) engines (e.g. LAMMPS,
i-PI, ASE ...) using the metatensor atomistic interface.
The idea behind metatrain is to have a general hub that provide an homogeneous
environment and user interface transforms every ML architecture in an end-to-end model
that can be connected to an MD engine. Any custom architecture compatible with
TorchScript can be integrated in metatrain, gaining automatic access to a training
and evaluation interface, as well as compatibility with various MD engines.
Note: metatrain does not provide mathematical functionalities per se but relies on
external models that implement the various architectures.
- Custom ML Architecture: Integrate any TorchScriptable ML model
- MD Engine Compatibility: Supports various MD engines for diverse research and application needs.
- Streamlined Training: Automated process leveraging MD-generated data to optimize ML models with minimal effort.
- HPC Compatibility: Efficient in HPC environments for extensive simulations.
- Future-Proof: Extensible to accommodate advancements in ML and MD fields.
Currently metatrain supports the following architectures for building an atomistic
model.
| Name | Description |
|---|---|
| GAP | Sparse Gaussian Approximation Potential (GAP) using Smooth Overlap of Atomic Positions (SOAP). |
| SOAP BPNN | A Behler-Parrinello neural network with SOAP features |
| Alchemical Model | A Behler-Parrinello neural network with SOAP features and Alchemical Compression of the composition space |
| PET | Point Edge Transformer (PET), interatomic machine learning potential |
For details, tutorials, and examples, please have a look at our documentation.
You can install metatrain with pip:
git clone https://github.com/lab-cosmo/metatrain
cd metatrain
pip install .In addition, specific models must be installed by specifying the model name. For example, to install the SOAP-BPNN model, you can run:
pip install .[soap-bpnn]You can then use mtt from the command line to train your models!
metatrain comes with completion definitions for its commands for bash and
zsh. Since it is difficult to automatically configure shell completions in a robust
manner, you must manually configure your shell to enable its completion support.
To make the completions available, source the definitions as part of your shell's
startup. Add the following to your ~/.bash_profile, ~/.zshrc (or, if they don't
exist, ~/.profile):
source $(mtt --shell-completion)Having a problem with metatrain? Please let us know by submitting an issue.
Submit new features or bug fixes through a pull request.
Thanks goes to all people that make metatrain possible: