Data Repository for "Distillation of atomistic foundation models across architectures and chemical domains"
Included in this repository are the fine-tuned and distilled models used to perform simulations in the paper "Distillation of atomistic foundation models across architectures and chemical domains". Additionally, the synthetic datasets used to train the distilled models are included for each application.
Folders are organised by target system, excluding the figure-generation folder which contains scripts for generating the figures in the paper. Each folder contains the following:
synthetic-data.xyz: The synthetic dataset used to train the distilled models.distilled-ARCH: Distilled modles for the target system.
For all fine-tuned foundation models, the model can be loaded using the graph_pes.models.load_model function, the same applies for all non-ACE distilled models.
from graph_pes.models import load_model
model = load_model("path/to/model.pt")
calculator = model.ase_calculator()ACE models can be loaded as follows:
from pyace import PyACECalculator
calculator = PyACECalculator("path/to/model.yace")Foundation models used for fine-tuning in this work are available at the following links: