Published May 12, 2026 | Version v0.1.4
Software Open

ClementiGroup/mlcg: v0.1.4

Authors/Creators

  • 1. CuspAI
  • 2. Freie Universität Berlin
  • 3. Freie Universitat
  • 4. FU Berlin
  • 5. Rice University

Description

This release introduces several major new features and infrastructure improvements across training, simulation, and model components.

New Features

  • Added full support for periodic boundary conditions in simulation and training
  • Added tracing of replica exchanges in parallel tempering simulations
  • Added multi-scheduler training support for both HDF5 and PyTorch datasets

Model & Architecture Updates

  • Added support for regularized version of SchNet and PaiNN
  • Added custom MACE interaction block with tanh gating
  • Added support for generic regularized RBF class
  • Added RegL1 loss

Priors update

  • Added ExpRepulsion
  • Added cutoff-based repulsion for standard and exponential repulsion

Improvements & Fixes

Small updates, optimizations, and bug fixes across the codebase

Recommended Update

This release is strongly recommended for all users of MLCG. In addition to minor bug fixes and stability improvements, it introduces major new functionality including full PBC support, new prior models, new regularization options.

Full Changelog: https://github.com/ClementiGroup/mlcg/compare/v0.1.3...v0.1.4

Files

ClementiGroup/mlcg-v0.1.4.zip

Files (106.6 MB)

Name Size Download all
md5:0fefbde65f56d374a941a89f877b7298
106.6 MB Preview Download

Additional details

Related works

Is supplement to
Software: https://github.com/ClementiGroup/mlcg/tree/v0.1.4 (URL)

Software

Repository URL
https://github.com/ClementiGroup/mlcg