NVIDIA/physicsnemo

Open-source deep-learning framework for building, training, and fine-tuning deep learning models using state-of-the-art Physics-ML methods

What it solves

PhysicsNeMo provides a standardized PyTorch framework for physics machine learning (physics ML) and scientific machine learning (SciML). It addresses the difficulty of building, training, and scaling AI models for science and engineering by providing reusable components and end-to-end training recipes for complex physical domains.

How it works

Built as a PyTorch-native library, PhysicsNeMo allows users to integrate scientific layers, numerical operators, and data transforms into existing workflows. It supports diverse scientific data representations, including regular grids, meshes, point clouds, and graphs. To handle high-resolution samples, it utilizes ShardTensor domain parallelism alongside standard PyTorch distributed training (DDP and FSDP2).

Who it’s for

It is designed for researchers and engineers in fields such as aerodynamics, weather forecasting, structural mechanics, geophysics, healthcare (e.g., cardiovascular flow), and additive manufacturing who need to build AI surrogates or generative models for physical systems.

Highlights

  • Broad Model Support: Includes a wide array of model families such as FNO, MeshGraphNet, Transolver, and GraphCast for various data representations.
  • Domain-Specific Recipes: Provides ready-to-use workflows for tasks like external aerodynamics, regional weather diffusion, and vehicle crash dynamics.
  • Scientific Data Handling: Native support for structured and unstructured discretizations, preserving the physical structure of the data.
  • Generative Capabilities: Includes a diffusion toolkit for inverse physics and topology optimization.

Related

  • Project
    NVIDIA-NeMo/Automodel

    NeMo AutoModel is a Python library from NVIDIA that provides ready‑to‑use, highly‑optimized training recipes for a huge range of LLMs, VLMs, and diffusion models. It supports full‑parameter and LoRA fine‑tuning, multi‑node tensor‑ and pipeline‑parallelism, and even speculative‑decoding drafters, all driven by simple YAML configs.

  • Dispatch
    Mapping the PyTorch Ecosystem: A Deep Dive into the PyTorch Landscape

    An exploration of the PyTorch Landscape, detailing the vast array of libraries and tools spanning modeling, training, and optimization that define the modern AI ecosystem.

  • Project
    fangwei123456/spikingjelly

    A PyTorch-native framework for spiking neural networks (SNNs) that supports large-scale training, inference, and deployment to neuromorphic hardware.

  • Project
    pytorch/pytorch

    A Python-based machine learning framework that provides GPU-accelerated tensor computation and a dynamic autograd system for building deep neural networks.

  • Project
    mit-han-lab/torchquantum

    A PyTorch-based framework for simulating quantum computations on classical hardware, enabling the training of parameterized quantum circuits and hybrid classical-quantum models.