Software
The LEADS team is developing an AI/ML community hub by:
•Establishing a common data format for scientific applications with a focus on foundation models.
•Establishing interoperability policies between codes and models, which avoid conflicts and standardize APIs, especially with respect to the use of external software in other programming languages.
•Focusing on outreach toward the broader SciML community to ensure the approach is beneficial without being restrictive toward uncommon but important edge cases.
•Developing portable ways to test, benchmark, and deploy AI/ML models across Department of Energy Leadership Computing Facilities.
Available Software
1) NeuroMANCER
PyTorch-based framework for solving parametric constrained optimization problems, physics-informed system identification, and parametric model predictive control
Description
Neural Modules with Adaptive Nonlinear Constraints and Efficient Regularizations (NeuroMANCER) is an open-source differentiable programming library for solving parametric constrained optimization problems, physics-informed system identification, and parametric model-based optimal control. NeuroMANCER is written in PyTorch and allows for systematic integration of machine learning with scientific computing for creating end-to-end differentiable models and algorithms embedded with prior knowledge and physics.
Target audience
Science teams interested in parametric constrained optimization and control.
License: BSD-3-Clause
Additional resources
- Repository
- Download (PyPI package)
- Documentation
2) TorchBraid
This package implements a layer-parallel approach to training neural ordinary differential equations (ODEs), and neural networks
Description
Algorithmically multigrid-in-time is used to expose parallelism in the forward and backward propagation phases used to compute the gradient. The neural network interface is build on PyTorch, while the backend uses XBraid (a C library) for multigrid-in-time. TorchBraid uses direct GPU communication when running simulations on GPUs. A CUDA-aware MPI implementation is required for multi-GPU support.
Target audience
Science teams interested layer-parallel approach to training of neural networks.
License: BSD-3-Clause
Package links
- E4S: TORCHBRAID
Additional resources
3) GenAI4UQ
A Python package for forward and inverse uncertainty quantification in model calibration, parameter estimation, and ensemble forecasting
Description
GenAI4UQ leverages a generative AI-based conditional modeling framework to address limitations of traditional inverse modeling techniques, such as Markov Chain Monte Carlo methods. By replacing computationally intensive iterative processes with a direct, learned mapping, GenAI4UQ enables efficient calibration of input parameters and generation of predictions directly from observations. The software supports rapid ensemble forecasting with robust uncertainty quantification while maintaining computational and storage efficiency, as well as a versatile hyperparameter auto-tuning framework.
Target audience
Science teams interested in uncertainty quantification and inverse modeling.
License: MIT
Additional resources
- Repository
- Documentation (methodology paper)
4) libROM
A reduced-order modeling library for a wide range of data-driven physical simulation methods
Description
libROM is a lightweight, scalable C++ library for data-driven physical simulation methods from the intrusive projection-based reduced order models to non-intrusive black-box approaches. It is the main tool box that the reduced order modeling team at Lawrence Livermore National Laboratory uses to develop efficient model order reduction techniques and physics-constrained data-driven methods.
Target audience
Science teams interested in data-driven reduced order methods
License: Apache-2.0 OR MIT
Package links
- Spack: librom
Additional resources
- Website
- Repository (main library)
- Repository (Python interface)
5) TAGTorch
A comprehensive suite of tools for analyzing and leveraging symmetry, equivariance, and topological properties in PyTorch deep learning models
Description
Topology, Algebra, and Geometry Torch (TAGTorch) is an open-source PyTorch library that provides reusable implementations of geometry-, topology-, and symmetry-based methods for machine learning. The library brings together tools for building mathematically informed neural networks as well as analyzing datasets and trained models through a unified, PyTorch-native interface.
Target audience
TAGTorch is designed for both machine learning researchers and domain scientists interested in incorporating mathematically grounded methods into existing PyTorch workflows without requiring expertise in computational topology, differential geometry, or group theory.
License: BSD-2-Clause
Additional resources