I find our results for the modelling of MAPI (a hybrid perovskite) particularly pleasing: the distribution of cation orientations generated by the teacher and student models during NVT MD are ~identical!

Beyond error metrics, we extensively validate these models to show they model liquid water well.

These student models have relatively few parameters (c. 40k for PaiNN and TensorNet), and so have much lower memory footprint. This lets you scale single GPU experiments very easily!

The resulting student models reach impressive accuracy vs DFT while being orders of magnitude faster than the teacher!

Note that these student models are of a different architecture to MACE, and in fact ACE is not even NN-based.

This lets you fine-tune any of these foundation models on your own datasets, and on any combination of energy, force and stress labels.

Crucially, you can easily switch between any of these models by changing a single line in your graph-pes config file.

graph-pes is a package for building, training and using graph-based machine learning models that predict energies and forces of atomic structures. #ChemSky #CompChemSky

…or you can customise all aspects of your training run using the wide range of config options available, including full control over the optimiser, dataset, learning rate scheduler, training callbacks, weights and biases logging, multi-GPU distributed training and more.

graph-pes is for researchers who want to train cutting edge MLIP models on their own datasets.

Use the graph-pes-train CLI tool to train any of several models (including NequIP, MACE and TensorNet) from a single, easy to read yaml file.

By relying on smart defaults, these files can be very short…

🚨 Introducing graph-pes: a unified framework for building, training and using graph-based machine-learned models of potential energy surfaces! 🚨

#compchem #ML #ChemSky #CompChemSky

Today I woke up to this 😮 a very small milestone, but also a very big one ☺️ more papers incoming soon!