A Python package for generating ML layers for MLIPs
Project description
KLay - KLIFF Layers, trainable and pre-trained layers for MLIPs
KLIFF compatible ML layers you can mould to your needs. Usable for general neural network architectures as well. Works out of the box with KLIFF.
The idea of this package is to generate different kind of layers from simple yaml like input that the user can then assemble on their own for research and experimentation. Lot of different packages provide complete opaque models that are easy to use but hard to modify. This package is supposed to take a complementary approach where the user can easily modify the layers to their needs.
Currently only NequIP layers are supported. More layers will be added in the future.
Installation
pip install klay
Usage
from klay import get_model_layers_from_yaml
layers = get_model_layers_from_yaml("path/to/yaml")
Example yaml file:
model:
- elem_embedding:
embedding_type: one_hot
n_elems: 1
- edge_embedding:
lmax: 1
normalize: True
normalization: component
parity: True
Documentation
The documentation can be found here
Future Work
- MACE Tensor Product Layer
- EGNN Layers
- Pretrained layers like M3GNET
Project details
Download files
Download the file for your platform. If you're not sure which to choose, learn more about installing packages.