physicsnemo
Open-source deep-learning framework for building, training, and fine-tuning deep learning models using state-of-the-art Physics-ML methods
What is physicsnemo?
Open-source deep-learning framework for building, training, and fine-tuning deep learning models using state-of-the-art Physics-ML methods
Overview of physicsnemo
What it is
NVIDIA PhysicsNeMo is an open-source PyTorch framework for physics machine learning (physics ML), scientific machine learning (SciML), and AI for science and engineering. It provides reusable library components and end-to-end training recipes.
Key features
- PyTorch-native and composable: use a complete architecture, a layer, a numerical operator, or a data transform without replacing existing PyTorch workflows.
- Built for scientific representations: works with regular grids, meshes, point clouds, graphs, and nested physical fields while preserving their structure.
- Scale the sample itself: uses
ShardTensordomain parallelism to split a single high-resolution sample across GPUs, alongside PyTorch DistributedDataParallel (DDP) or Fully Sharded Data Parallel 2 (FSDP2). - Includes model families for surrogates and dynamics (e.g., FNO, MeshGraphNet, Transolver), weather and climate (e.g., AFNO, GraphCast, Pangu-Weather), and generative and inverse models (e.g., Diffusion U-Nets, TopoDiff).
- Provides physics AI building blocks: optimized models, datapipes for readers and transforms, metrics, domain parallelism, and optimization tools.
Use cases
PhysicsNeMo is used across multiple physical domains:
- Engineering design and CFD: unified external aerodynamics, data-center airflow prediction.
- Weather, climate, and water: global forecasting with StormCast, flood prediction.
- Structures and manufacturing: crash dynamics, metal sintering, underfill dispensing.
- Geophysics and subsurface systems: diffusion full-waveform inversion, reservoir surrogates.
- Healthcare: cardiovascular flow prediction, brain anomaly detection.
- Generative and inverse physics: topology optimization with TopoDiff, diffusion-based inverse problems.
- Simulation-data loops: active learning to select new simulations.
Getting started
Install with pip: bash pip install 'nvidia-physicsnemo[cu13]' For CUDA 12, a basic install, optional features, or source setup, see the repository README.
License and pricing
Licensed under Apache-2.0. Open source.
Based on the project's public README and repository metadata.