-
Notifications
You must be signed in to change notification settings - Fork 4
Home
All documents on this site are licensed under CC-BY 4.0.
We aim to establish a comprehensive and interoperable ecosystem tailored for differentiable physical modeling. The initial focus is on streamlining neural network potentials (NNPs) for efficient and accurate free energy calculations. We understand the critical importance of usability in software development; hence, we have crafted our architecture with clean and straightforward APIs. These APIs are designed to offer scientists the ease of customizing routines and integrating workflows with minimal overhead.
The modelforge package contains the implementation, training, distribution, storage, and application of NNPs designed for molecular simulations. To complement our core package, we provide the auditorium package – an automated NNP testing and benchmark suite designed to validate and compare the performance of various NNPs with rigor and transparency. In tandem, our Markov Chain Monte Carlo (MCMC) state sampler engine, chiron, facilitates advanced sampling techniques, making it a cornerstone for researchers delving into complex molecular systems.