Skip to content
Back to search

physicsnemo

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

Safety scan

A static text scan, not a guarantee. Select the badge to see what was scanned and every finding.

Checked today

Check it deeper on your machine

It has changed since its scan on 11 Nov 2026.

As a Claude Code plugin, install it and then ask Claude Code:

/plugin marketplace add keithmackay/mackayi/plugin install scanskillsafety@mackayiDo a deep scan of https://github.com/NVIDIA/physicsnemo with scanskillsafety

Or from a terminal, clone it and run it:

git clone https://github.com/keithmackay/scanskillsafety-skill ~/.claude/skills/scanskillsafetynode ~/.claude/skills/scanskillsafety/cli.cjs --deep 'https://github.com/NVIDIA/physicsnemo'

It checks the exact commit: every file, its packages against OSV (api.osv.dev), how it compares with this page's scan (findsafeskills.com receives the repository's address), and, if you agree, Claude reading every file with your own Claude subscription. It never runs the code.

Badge for your README
findsafeskills safety rating

Markdown

[![findsafeskills safety rating](/api/listings/nvidia--physicsnemo/badge.svg)](/listings/nvidia--physicsnemo)

HTML

<a href="/listings/nvidia--physicsnemo"><img src="/api/listings/nvidia--physicsnemo/badge.svg" alt="findsafeskills safety rating" /></a>

Install

i

Install instructions from the README

As written by the repo's author. findsafeskills has not verified these steps.

Refer to [`pyproject.toml`](https://github.com/NVIDIA/physicsnemo/blob/main/pyproject.toml) for currently supported Python versions, optional dependency groups, and exact dependency constraints. The command at the top of this README selects the CUDA 13 backend. Use the CUDA 12 backend instead with:

pip install "nvidia-physicsnemo[cu12]" \
  --extra-index-url https://download.pytorch.org/whl/cu129

Specify the CUDA 12.9 PyTorch index because pip does not read the index settings in `pyproject.toml`. The CUDA 12 backend is Linux-only and uses PyTorch 2.13, the newest release with CUDA 12.9 wheels.

For a basic installation that uses PyPI's default PyTorch distribution and does not install the CUDA-specific RAPIDS packages:

pip install nvidia-physicsnemo

Optional features compose with either backend, for example:

pip install "nvidia-physicsnemo[cu13,gnns]"

To work from a source checkout with [uv](https://docs.astral.sh/uv/):

git clone https://github.com/NVIDIA/physicsnemo.git
cd physicsnemo
uv sync --extra cu13

What was scanned

The files the safety scan read, as written by the repository's author and shown as plain text. Lines with a finding are marked.

README.md · README · 363 lines

View this file on GitHub (read at the repository's current version: the scan did not record which commit it read)

This file has changed since it was scanned, so this is not exactly the text that was scanned, and no line is marked. The findings are listed with the safety scan above.

  1. <!-- markdownlint-disable MD002 MD013 MD033 MD041 -->
  2. <h1 align="center">NVIDIA PhysicsNeMo</h1>
  3. <p align="center">
  4. <strong>Build, train, and scale physics AI models with PyTorch.</strong>
  5. </p>
  6. <div align="center">
  7. <a href="https://pypi.org/project/nvidia-physicsnemo/"><img alt="PyPI" src="https://img.shields.io/pypi/v/nvidia-physicsnemo?label=PyPI&style=flat-square"></a>
  8. <a href="https://pepy.tech/projects/nvidia-physicsnemo"><img alt="PyPI downloads" src="https://api.pepy.tech/badge/nvidia-physicsnemo/month"></a>
  9. <a href="https://docs.nvidia.com/physicsnemo/latest/"><img alt="Documentation" src="https://img.shields.io/badge/docs-latest-76B900?style=flat-square"></a>
  10. <a href="https://github.com/NVIDIA/physicsnemo/actions/workflows/install-ci.yml"><img alt="Install CI" src="https://img.shields.io/github/actions/workflow/status/NVIDIA/physicsnemo/install-ci.yml?branch=main&label=install%20CI&style=flat-square"></a>
  11. <a href="https://app.codecov.io/gh/NVIDIA/physicsnemo"><img alt="Codecov" src="https://img.shields.io/codecov/c/github/NVIDIA/physicsnemo/main?style=flat-square"></a>
  12. <a href="https://github.com/NVIDIA/physicsnemo/blob/main/LICENSE.txt"><img alt="License" src="https://img.shields.io/github/license/NVIDIA/physicsnemo?style=flat-square"></a>
  13. </div>
  14. <p align="center">
  15. <a href="https://docs.nvidia.com/physicsnemo/latest/">Documentation</a> ·
  16. <a href="https://github.com/NVIDIA/physicsnemo/blob/main/examples/README.md">Examples</a> ·
  17. <a href="https://docs.nvidia.com/physicsnemo/latest/physicsnemo/api_models.html">Models</a> ·
  18. <a href="https://nvidia.github.io/physicsnemo/blog/">Blog</a> ·
  19. <a href="https://github.com/NVIDIA/physicsnemo/discussions">Discussions</a> ·
  20. <a href="https://github.com/NVIDIA/physicsnemo/blob/main/CONTRIBUTING.md">Contributing</a>
  21. </p>
  22. <!-- markdownlint-enable MD013 MD033 MD041 -->
  23. NVIDIA PhysicsNeMo is an open-source PyTorch framework for physics machine learning
  24. (physics ML), scientific machine learning (SciML), and AI for science and engineering.
  25. It provides reusable library components and end-to-end training recipes.
  26. <!-- Keep repository links absolute because this README is also rendered on PyPI. -->
  27. ```bash
  28. pip install "nvidia-physicsnemo[cu13]"
  29. ```
  30. For CUDA 12, a basic install, optional features, or source setup, refer to
  31. [installation options](#installation-options).
  32. ## PhysicsNeMo in Action
  33. <!-- markdownlint-disable MD013 MD033 -->
  34. <table width="100%">
  35. <tr>
  36. <th width="50%" align="center">Aerodynamics</th>
  37. <th width="50%" align="center">Weather Forecasting</th>
  38. </tr>
  39. <tr>
  40. <td width="50%" align="center"><a href="https://github.com/NVIDIA/physicsnemo/tree/main/examples/cfd/external_aerodynamics/active_learning_aero"><img width="100%" alt="Ground-truth and GeoTransolver-predicted pressure and wall-shear-stress fields on a held-out ShiftSUV" src="https://raw.githubusercontent.com/NVIDIA/physicsnemo/main/docs/img/al_shiftsuv_field_predictions.png"></a></td>
  41. <td width="50%" align="center"><a href="https://github.com/NVIDIA/physicsnemo/tree/main/examples/weather/regional_weather_diffusion"><img width="100%" alt="StormCast prediction beside a high-resolution weather analysis" src="https://raw.githubusercontent.com/NVIDIA/physicsnemo/main/docs/img/stormcast_rollout.gif"></a></td>
  42. </tr>
  43. <tr>
  44. <td width="50%" align="center" valign="top"><a href="https://github.com/NVIDIA/physicsnemo/tree/main/examples/cfd/external_aerodynamics/unified_external_aero_recipe"><strong>Unified external aerodynamics</strong></a>: train and compare current surface and volume models</td>
  45. <td width="50%" align="center" valign="top"><a href="https://github.com/NVIDIA/physicsnemo/tree/main/examples/weather/regional_weather_diffusion"><strong>StormCast</strong></a>: generative regional weather forecasting</td>
  46. </tr>
  47. <tr>
  48. <th width="50%" align="center">Semiconductor packaging</th>
  49. <th width="50%" align="center">Structural mechanics</th>
  50. </tr>
  51. <tr>
  52. <td width="50%" align="center"><a href="https://github.com/NVIDIA/physicsnemo/tree/main/examples/cfd/underfill_dispensing"><img width="100%" alt="GeoTransolver prediction and CFD simulation of an advancing epoxy interface" src="https://raw.githubusercontent.com/NVIDIA/physicsnemo/main/docs/img/underfill/G20.gif"></a></td>
  53. <td width="50%" align="center"><a href="https://github.com/NVIDIA/physicsnemo/tree/main/examples/structural_mechanics/crash"><img width="100%" alt="Finite-element simulation and PhysicsNeMo prediction of full-vehicle crash deformation" src="https://raw.githubusercontent.com/NVIDIA/physicsnemo/main/docs/img/crash/crash.gif"></a></td>
  54. </tr>
  55. <tr>
  56. <td width="50%" align="center" valign="top"><a href="https://github.com/NVIDIA/physicsnemo/tree/main/examples/cfd/underfill_dispensing"><strong>Underfill dispensing</strong></a>: transient epoxy-interface prediction with GeoTransolver</td>
  57. <td width="50%" align="center" valign="top"><a href="https://github.com/NVIDIA/physicsnemo/tree/main/examples/structural_mechanics/crash"><strong>Crash dynamics</strong></a>: transient surrogates on deforming meshes</td>
  58. </tr>
  59. <tr>
  60. <th width="50%" align="center">Geophysics</th>
  61. <th width="50%" align="center">Healthcare</th>
  62. </tr>
  63. <tr>
  64. <td width="50%" align="center"><a href="https://github.com/NVIDIA/physicsnemo/tree/main/examples/geophysics/diffusion_fwi"><img width="100%" alt="Physics-informed Diffusion FWI ground truth, ensemble samples, and ensemble mean" src="https://raw.githubusercontent.com/NVIDIA/physicsnemo/main/docs/img/diffusion_fwi_pi_predictions.png"></a></td>
  65. <td width="50%" align="center"><a href="https://github.com/NVIDIA/physicsnemo/tree/main/examples/healthcare/bloodflow_1d_mgn"><img width="100%" alt="MeshGraphNet and reference pressure and flow-rate fields along patient-specific vessel centerlines" src="https://raw.githubusercontent.com/NVIDIA/physicsnemo/main/docs/img/bloodflow_1d_mgn_results.gif"></a></td>
  66. </tr>
  67. <tr>
  68. <td width="50%" align="center" valign="top"><a href="https://github.com/NVIDIA/physicsnemo/tree/main/examples/geophysics/diffusion_fwi"><strong>Diffusion FWI</strong></a>: guided generative full-waveform inversion</td>
  69. <td width="50%" align="center" valign="top"><a href="https://github.com/NVIDIA/physicsnemo/tree/main/examples/healthcare/bloodflow_1d_mgn"><strong>Cardiovascular flow</strong></a>: reduced-order prediction with MeshGraphNet</td>
  70. </tr>
  71. <tr>
  72. <th width="50%" align="center">Data-center thermal design</th>
  73. <th width="50%" align="center">Additive manufacturing</th>
  74. </tr>
  75. <tr>
  76. <td width="50%" align="center"><a href="https://github.com/NVIDIA/physicsnemo/tree/main/examples/cfd/datacenter"><img width="100%" alt="Animated temperature field inside a data-center hot aisle" src="https://raw.githubusercontent.com/NVIDIA/physicsnemo/main/docs/img/datacenter_design_cfd.gif"></a></td>
  77. <td width="50%" align="center"><a href="https://github.com/NVIDIA/physicsnemo/tree/main/examples/additive_manufacturing/sintering_physics"><img width="100%" alt="Predicted sintering deformation and displacement magnitude for a helical metal component" src="https://raw.githubusercontent.com/NVIDIA/physicsnemo/main/docs/img/vfgn_doc/screw.gif"></a></td>
  78. </tr>
  79. <tr>
  80. <td width="50%" align="center" valign="top"><a href="https://github.com/NVIDIA/physicsnemo/tree/main/examples/cfd/datacenter"><strong>Data-center airflow</strong></a>: real-time temperature and airflow prediction with a 3D U-Net surrogate</td>
  81. <td width="50%" align="center" valign="top"><a href="https://github.com/NVIDIA/physicsnemo/tree/main/examples/additive_manufacturing/sintering_physics"><strong>Metal sintering</strong></a>: graph-based deformation prediction</td>
  82. </tr>
  83. </table>
  84. <!-- markdownlint-enable MD013 MD033 -->
  85. Every visual above comes from a PhysicsNeMo recipe or workflow.
  86. ## Why PhysicsNeMo
  87. - **PyTorch-native and composable.** Use a complete architecture, a layer, a numerical
  88. operator, or a data transform without replacing your existing PyTorch workflow.
  89. - **Built for scientific representations.** Work with regular grids, meshes, point
  90. clouds, graphs, and nested physical fields while preserving their structure.
  91. - **Scale the sample itself.** Use `ShardTensor` domain parallelism to split a single
  92. high-resolution sample across GPUs, alongside PyTorch DistributedDataParallel (DDP) or
  93. Fully Sharded Data Parallel 2 (FSDP2).
  94. ## Explore Recipes
  95. Start from a recipe and adapt it to your data, geometry, physics, and deployment
  96. constraints. Explore by physical domain:
  97. - ✈️ **Engineering design and CFD:** train and compare current surface and volume
  98. models with the
  99. [unified external-aerodynamics recipe](https://github.com/NVIDIA/physicsnemo/tree/main/examples/cfd/external_aerodynamics/unified_external_aero_recipe)
  100. or accelerate [data-center airflow](https://github.com/NVIDIA/physicsnemo/tree/main/examples/cfd/datacenter).
  101. - 🌦️ **Weather, climate, and water:** build global forecasts from the
  102. [weather recipes](https://github.com/NVIDIA/physicsnemo/tree/main/examples/weather)
  103. or predict [floods](https://github.com/NVIDIA/physicsnemo/tree/main/examples/weather/flood_modeling).
  104. - 🏗️ **Structures and manufacturing:** emulate
  105. [deforming structures](https://github.com/NVIDIA/physicsnemo/tree/main/examples/structural_mechanics/deforming_plate).
  106. - 🌍 **Geophysics and subsurface systems:** build
  107. [reservoir surrogates](https://github.com/NVIDIA/physicsnemo/tree/main/examples/reservoir_simulation).
  108. - 🫀 **Healthcare:** perform
  109. [brain anomaly detection](https://github.com/NVIDIA/physicsnemo/tree/main/examples/healthcare/brain_anomaly_detection).
  110. - ✨ **Generative and inverse physics:** compose the
  111. [diffusion toolkit](https://docs.nvidia.com/physicsnemo/latest/physicsnemo/api_diffusion.html)
  112. with [topology-generation recipes](https://github.com/NVIDIA/physicsnemo/tree/main/examples/generative/topodiff).
  113. - 🔁 **Simulation-data loops:** select new simulations with
  114. [active learning](https://github.com/NVIDIA/physicsnemo/tree/main/examples/active_learning).
  115. Browse the [complete example catalog](https://github.com/NVIDIA/physicsnemo/blob/main/examples/README.md)
  116. for every available recipe.
  117. ### Explore the PhysicsNeMo Ecosystem
  118. PhysicsNeMo supplies reusable models, training components, and recipes. For a
  119. domain-focused application layer or data preparation, continue with:
  120. | Need | Continue with |
  121. | --- | --- |
  122. | **Scientific data preparation** | [PhysicsNeMo Curator](https://github.com/NVIDIA/physicsnemo-curator) for extract, transform, and load (ETL) pipelines that prepare AI-ready scientific and engineering datasets |
  123. | **Engineering inference and design** | [PhysicsNeMo CFD](https://github.com/NVIDIA/physicsnemo-cfd) for inference, evaluation, benchmarking, and design workflows |
  124. | **Weather and climate** | [Earth2Studio](https://github.com/NVIDIA/earth2studio) for building and deploying AI weather and climate workflows |
  125. | **Atomistic simulation** | [NVIDIA ALCHEMI Toolkit](https://github.com/NVIDIA/nvalchemi-toolkit) for GPU-first training, inference, and dynamics workflows, with optimized primitives from [ALCHEMI Toolkit Ops](https://github.com/NVIDIA/nvalchemi-toolkit-ops) |
  126. ## Choose a Model Family
  127. PhysicsNeMo models are ordinary `torch.nn.Module` objects. Choose a family by the
  128. representation of your data and the task you need to solve. Model names link to source
  129. code. The final column links to examples and papers.
  130. ### Surrogates and Dynamics
  131. | Model family | Data representation | Known uses and starting points |
  132. | --- | --- | --- |
  133. | [FNO](https://github.com/NVIDIA/physicsnemo/tree/main/physicsnemo/models/fno) / [DPOT](https://github.com/NVIDIA/physicsnemo/tree/main/physicsnemo/models/dpot) | Regular Cartesian grids; DPOT adds a time/history axis | Neural PDE operators and autoregressive dynamics: [Darcy flow with FNO](https://github.com/NVIDIA/physicsnemo/tree/main/examples/cfd/darcy_fno), [DPOT paper](https://arxiv.org/abs/2403.03542) |
  134. | [MeshGraphNet](https://github.com/NVIDIA/physicsnemo/tree/main/physicsnemo/models/meshgraphnet) / [VFGN](https://github.com/NVIDIA/physicsnemo/tree/main/physicsnemo/models/vfgn) | Node-edge graphs over unstructured meshes | Mesh dynamics and reduced-order simulation: [vehicle crash](https://github.com/NVIDIA/physicsnemo/tree/main/examples/structural_mechanics/crash), [cardiovascular flow](https://github.com/NVIDIA/physicsnemo/tree/main/examples/healthcare/bloodflow_1d_mgn), [metal sintering](https://github.com/NVIDIA/physicsnemo/tree/main/examples/additive_manufacturing/sintering_physics) |
  135. | [Transolver](https://github.com/NVIDIA/physicsnemo/tree/main/physicsnemo/models/transolver) / [FLARE](https://github.com/NVIDIA/physicsnemo/tree/main/physicsnemo/models/flare) | Point sets and structured or unstructured discretizations | PDE surrogates on large discretizations: [Darcy flow](https://github.com/NVIDIA/physicsnemo/tree/main/examples/cfd/darcy_transolver), [external aerodynamics](https://github.com/NVIDIA/physicsnemo/tree/main/examples/cfd/external_aerodynamics/unified_external_aero_recipe) |
  136. | [GeoTransolver](https://github.com/NVIDIA/physicsnemo/tree/main/physicsnemo/models/geotransolver) | Point clouds or structured/unstructured grids, with geometry and global context | Geometry-aware CAE: [external aerodynamics](https://github.com/NVIDIA/physicsnemo/tree/main/examples/cfd/external_aerodynamics/unified_external_aero_recipe), [underfill dispensing](https://github.com/NVIDIA/physicsnemo/tree/main/examples/cfd/underfill_dispensing) |
  137. | [DoMINO](https://github.com/NVIDIA/physicsnemo/tree/main/physicsnemo/models/domino) | Geometry points, sampled surface/volume fields, and structured SDF grids | Surface and volume field prediction for automotive aerodynamics: [paper](https://arxiv.org/abs/2501.13350) |
  138. | [FIGConvNet](https://github.com/NVIDIA/physicsnemo/tree/main/physicsnemo/models/figconvnet) | Large 3D point clouds represented with factorized 2D grids | 3D CAE field and scalar prediction: [external aerodynamics](https://github.com/NVIDIA/physicsnemo/tree/main/examples/cfd/external_aerodynamics/figconvnet), [vehicle crash](https://github.com/NVIDIA/physicsnemo/tree/main/examples/structural_mechanics/crash) |
  139. | [GLOBE](https://github.com/NVIDIA/physicsnemo/tree/main/physicsnemo/models/globe) | Boundary meshes and arbitrary query points | Boundary-driven PDEs and large-geometry CAE: [external aerodynamics](https://github.com/NVIDIA/physicsnemo/tree/main/examples/cfd/external_aerodynamics/unified_external_aero_recipe), [paper](https://arxiv.org/abs/2511.15856) |
  140. | [AeroJEPA](https://github.com/NVIDIA/physicsnemo/tree/main/physicsnemo/experimental/models/aerojepa) *(experimental; API may change)* | 3D surface point clouds, operating conditions, and query points | Aerodynamic representation learning and field prediction: [tutorial](https://github.com/NVIDIA/physicsnemo/tree/main/examples/cfd/external_aerodynamics/aerojepa), [paper](https://arxiv.org/abs/2605.05586) |
  141. ### Weather and Climate
  142. | Model family | Data representation | Known uses and starting points |
  143. | --- | --- | --- |
  144. | [AFNO](https://github.com/NVIDIA/physicsnemo/tree/main/physicsnemo/models/afno) | Regular 2D fields | Global forecasting: [unified weather recipe](https://github.com/NVIDIA/physicsnemo/tree/main/examples/weather/unified_recipe) |
  145. | [GraphCast](https://github.com/NVIDIA/physicsnemo/tree/main/physicsnemo/models/graphcast) | Latitude-longitude fields and a multiscale icosahedral mesh graph | Global autoregressive forecasting: [recipe](https://github.com/NVIDIA/physicsnemo/tree/main/examples/weather/graphcast), [paper](https://arxiv.org/abs/2212.12794) |
  146. | [Pangu-Weather](https://github.com/NVIDIA/physicsnemo/tree/main/physicsnemo/models/pangu) / [FengWu](https://github.com/NVIDIA/physicsnemo/tree/main/physicsnemo/models/fengwu) | Multilevel latitude-longitude grids | Global medium-range forecasting: [Pangu-Weather recipe](https://github.com/NVIDIA/physicsnemo/tree/main/examples/weather/pangu_weather), [FengWu paper](https://arxiv.org/abs/2304.02948) |
  147. | [DLWP](https://github.com/NVIDIA/physicsnemo/tree/main/physicsnemo/models/dlwp) / [DLWP-HEALPix](https://github.com/NVIDIA/physicsnemo/tree/main/physicsnemo/models/dlwp_healpix) | Cubed-sphere grids or HEALPix meshes | Global and coupled forecasting: [DLWP](https://github.com/NVIDIA/physicsnemo/tree/main/examples/weather/dlwp), [DLWP-HEALPix](https://github.com/NVIDIA/physicsnemo/tree/main/examples/weather/dlwp_healpix) |
  148. ### Generative and Inverse Models
  149. | Model family | Data representation | Known uses and starting points |
  150. | --- | --- | --- |
  151. | [Diffusion U-Nets](https://github.com/NVIDIA/physicsnemo/tree/main/physicsnemo/models/diffusion_unets) / [DiT](https://github.com/NVIDIA/physicsnemo/tree/main/physicsnemo/models/dit) | 2D fields or patch tokens | Stochastic regional forecasting, downscaling, and inverse problems: [StormCast and StormScope](https://github.com/NVIDIA/physicsnemo/tree/main/examples/weather/regional_weather_diffusion), [diffusion FWI](https://github.com/NVIDIA/physicsnemo/tree/main/examples/geophysics/diffusion_fwi) |
  152. | [TopoDiff](https://github.com/NVIDIA/physicsnemo/tree/main/physicsnemo/models/topodiff) | 2D topology fields conditioned on design constraints | Generative topology optimization: [recipe](https://github.com/NVIDIA/physicsnemo/tree/main/examples/generative/topodiff), [paper](https://arxiv.org/abs/2208.09591) |
  153. Refer to the [model catalog](https://docs.nvidia.com/physicsnemo/latest/physicsnemo/api_models.html)
  154. for the full API and configuration details.
  155. ## Contents of PhysicsNeMo
  156. The framework is layered so high-level workflows can reuse lower-level scientific
  157. building blocks without forcing those foundations to depend on applications.
  158. <!-- markdownlint-disable MD013 -->
  159. ```mermaid
  160. flowchart TB
  161. apps["Applications and workflows<br/><a href='https://github.com/NVIDIA/physicsnemo/tree/main/examples'>Examples</a> · <a href='https://github.com/NVIDIA/physicsnemo/tree/main/physicsnemo/active_learning'>Active learning</a> · <a href='https://github.com/NVIDIA/physicsnemo/tree/main/physicsnemo/diffusion'>Diffusion</a>"]
  162. blocks["Physics AI building blocks<br/><a href='https://github.com/NVIDIA/physicsnemo/tree/main/physicsnemo/models'>Models</a> · <a href='https://github.com/NVIDIA/physicsnemo/tree/main/physicsnemo/datapipes'>Datapipes</a> · <a href='https://github.com/NVIDIA/physicsnemo/tree/main/physicsnemo/metrics'>Metrics</a> · <a href='https://github.com/NVIDIA/physicsnemo/tree/main/physicsnemo/domain_parallel'>Domain parallel</a> · <a href='https://github.com/NVIDIA/physicsnemo/tree/main/physicsnemo/optim'>Optim</a>"]
  163. base["Reusable foundations<br/><a href='https://github.com/NVIDIA/physicsnemo/tree/main/physicsnemo/mesh'>Mesh</a> · <a href='https://github.com/NVIDIA/physicsnemo/tree/main/physicsnemo/nn'>NN</a> · <a href='https://github.com/NVIDIA/physicsnemo/tree/main/physicsnemo/utils'>Utils</a> · <a href='https://github.com/NVIDIA/physicsnemo/tree/main/physicsnemo/distributed'>Distributed</a> · <a href='https://github.com/NVIDIA/physicsnemo/tree/main/physicsnemo/core'>Core</a>"]
  164. apps -->|"build on"| blocks
  165. blocks -->|"reuse"| base
  166. classDef workflow fill:#F5FAEF,stroke:#76B900,stroke-width:2px,color:#111111
  167. classDef building fill:#EEF7E1,stroke:#76B900,stroke-width:2px,color:#111111
  168. classDef foundation fill:#E8F5D0,stroke:#76B900,stroke-width:2px,color:#111111
  169. class apps workflow
  170. class blocks building
  171. class base foundation
  172. ```
  173. <!-- markdownlint-enable MD013 -->
  174. Arrows follow the allowed dependency direction in the
  175. [import-layer contract](https://github.com/NVIDIA/physicsnemo/blob/main/.importlinter).
  176. The diagram is simplified. Follow the links below for the public surfaces.
  177. - **Applications and workflows:** runnable
  178. [examples](https://github.com/NVIDIA/physicsnemo/tree/main/examples),
  179. restartable [active-learning loops](https://github.com/NVIDIA/physicsnemo/tree/main/physicsnemo/active_learning),
  180. and [diffusion](https://github.com/NVIDIA/physicsnemo/tree/main/physicsnemo/diffusion)
  181. schedulers, samplers, guidance, and multi-diffusion.
  182. - **Physics AI building blocks:** optimized
  183. [models](https://github.com/NVIDIA/physicsnemo/tree/main/physicsnemo/models),
  184. [datapipes](https://github.com/NVIDIA/physicsnemo/tree/main/physicsnemo/datapipes)
  185. for readers, transforms, GPU preprocessing, and multi-dataset sampling,
  186. [metrics](https://github.com/NVIDIA/physicsnemo/tree/main/physicsnemo/metrics),
  187. [`ShardTensor` domain parallelism](https://github.com/NVIDIA/physicsnemo/tree/main/physicsnemo/domain_parallel),
  188. and [optimization](https://github.com/NVIDIA/physicsnemo/tree/main/physicsnemo/optim).
  189. - **Reusable foundations:** [`Mesh` and `DomainMesh`](https://github.com/NVIDIA/physicsnemo/blob/main/physicsnemo/mesh/README.md)
  190. with GPU topology, spatial queries, remeshing, and differentiable deformation;
  191. [neural-network layers and numerical functionals](https://github.com/NVIDIA/physicsnemo/tree/main/physicsnemo/nn)
  192. for derivatives, interpolation, geometry, sampling, and rendering;
  193. [utilities](https://github.com/NVIDIA/physicsnemo/tree/main/physicsnemo/utils) for
  194. checkpointing, logging, and profiling;
  195. [distributed primitives](https://github.com/NVIDIA/physicsnemo/tree/main/physicsnemo/distributed)
  196. including distributed FFTs;
  197. and the [model lifecycle core](https://github.com/NVIDIA/physicsnemo/tree/main/physicsnemo/core).
  198. [Experimental modules](https://github.com/NVIDIA/physicsnemo/tree/main/physicsnemo/experimental)
  199. incubate models such as [AeroJEPA](https://github.com/NVIDIA/physicsnemo/tree/main/physicsnemo/experimental/models/aerojepa),
  200. alongside uncertainty quantification, guardrails, parameter-efficient fine-tuning
  201. with LoRA, and other research utilities.
  202. > **API stability:** APIs under `physicsnemo.experimental` are incubating and may change
  203. > between releases. Stable modules follow the project's semantic-versioning policy.
  204. > Refer to the [changelog][api-changelog] for API changes and removals.
  205. [api-changelog]: https://github.com/NVIDIA/physicsnemo/blob/main/CHANGELOG.md
  206. ## Use PhysicsNeMo with Coding Agents
  207. The repository includes two NVIDIA-authored skills for Codex, Claude Code, and other
  208. compatible coding agents:
  209. - [PhysicsNeMo Discover](https://github.com/NVIDIA/physicsnemo/blob/main/skills/physicsnemo-discover/SKILL.md)
  210. finds models, datapipes, examples, and documentation for a SciML task
  211. ([evaluation](https://github.com/NVIDIA/physicsnemo/blob/main/skills/physicsnemo-discover/BENCHMARK.md)).
  212. - [PhysicsNeMo ShardTensor](https://github.com/NVIDIA/physicsnemo/blob/main/skills/physicsnemo-shard-tensor/SKILL.md)
  213. helps add domain parallelism, integrate DDP or FSDP2, implement shard-aware
  214. operations, and write multi-GPU correctness tests
  215. ([evaluation](https://github.com/NVIDIA/physicsnemo/blob/main/skills/physicsnemo-shard-tensor/BENCHMARK.md)).
  216. ## Installation Options
  217. Refer to [`pyproject.toml`](https://github.com/NVIDIA/physicsnemo/blob/main/pyproject.toml)
  218. for currently supported Python versions, optional dependency groups, and exact
  219. dependency constraints. The command at the top of this README selects the CUDA 13
  220. backend. Use the CUDA 12 backend instead with:
  221. ```bash
  222. pip install "nvidia-physicsnemo[cu12]" \
  223. --extra-index-url https://download.pytorch.org/whl/cu129
  224. ```
  225. Specify the CUDA 12.9 PyTorch index because pip does not read the index settings
  226. in `pyproject.toml`. The CUDA 12 backend is Linux-only and uses PyTorch 2.13,
  227. the newest release with CUDA 12.9 wheels.
  228. For a basic installation that uses PyPI's default PyTorch distribution and does not
  229. install the CUDA-specific RAPIDS packages:
  230. ```bash
  231. pip install nvidia-physicsnemo
  232. ```
  233. Optional features compose with either backend, for example:
  234. ```bash
  235. pip install "nvidia-physicsnemo[cu13,gnns]"
  236. ```
  237. To work from a source checkout with [uv](https://docs.astral.sh/uv/):
  238. ```bash
  239. git clone https://github.com/NVIDIA/physicsnemo.git
  240. cd physicsnemo
  241. uv sync --extra cu13
  242. ```
  243. ### Notice and Disclaimer
  244. Installing PhysicsNeMo downloads its dependencies. Some datapipes and examples
  245. also download datasets or model checkpoints when you ask them to.
  246. > NOTICE AND DISCLAIMER: This software automatically retrieves, accesses or interacts
  247. > with external materials. Those retrieved materials are not distributed with this
  248. > software and are governed solely by separate terms, conditions and licenses. You are
  249. > solely responsible for finding, reviewing and complying with all applicable terms,
  250. > conditions, and licenses, and for verifying the security, integrity and suitability
  251. > of any retrieved materials for your specific use case. This software is provided "AS
  252. > IS", without warranty of any kind. The author makes no representations or warranties
  253. > regarding any retrieved materials, and assumes no liability for any losses, damages,
  254. > liabilities or legal consequences from your use or inability to use this software or
  255. > any retrieved materials. Use this software and the retrieved materials at your own
  256. > risk.
  257. ## Learning Resources
  258. Learn through the [PhysicsNeMo notebooks on Hugging Face](https://huggingface.co/collections/nvidia/physicsnemo),
  259. the [AI for Science bootcamp](https://github.com/openhackathons-org/End-to-End-AI-for-Science),
  260. and the [self-paced NVIDIA Deep Learning Institute course](https://learn.nvidia.com/courses/course-detail?course_id=course-v1:DLI+S-OV-04+V1).
  261. Pretrained models and datasets are available through the
  262. [NGC catalog](https://catalog.ngc.nvidia.com/search?orderBy=scoreDESC&page=&pageSize=&query=PhysicsNeMo).
  263. ## Community and Contributing
  264. Contributions to the library, examples, and documentation are welcome.
  265. - Ask questions and share work in [GitHub Discussions](https://github.com/NVIDIA/physicsnemo/discussions).
  266. - Report a bug or propose a feature through [GitHub Issues](https://github.com/NVIDIA/physicsnemo/issues).
  267. - Look for issues labeled [help wanted](https://github.com/NVIDIA/physicsnemo/issues?q=is%3Aissue+is%3Aopen+label%3A%22help+wanted%22).
  268. - **Before opening a pull request, read the
  269. [contribution guide](https://github.com/NVIDIA/physicsnemo/blob/main/CONTRIBUTING.md)
  270. and coordinate the proposed work with maintainers in an issue or discussion.**
  271. Every pull request should correspond to an open issue. For substantial changes,
  272. wait for maintainer feedback before starting implementation.
  273. - Follow the [code of conduct](https://github.com/NVIDIA/physicsnemo/blob/main/CODE_OF_CONDUCT.md),
  274. and report vulnerabilities privately through the
  275. [security policy](https://github.com/NVIDIA/physicsnemo/blob/main/SECURITY.md).
  276. For release history and upgrade notes, refer to the
  277. [changelog](https://github.com/NVIDIA/physicsnemo/blob/main/CHANGELOG.md),
  278. [GitHub releases](https://github.com/NVIDIA/physicsnemo/releases), and the
  279. [v2 migration guide](https://github.com/NVIDIA/physicsnemo/blob/main/v2.0-MIGRATION-GUIDE.md).
  280. ## Citation
  281. If PhysicsNeMo supports your research, cite the project using the metadata in
  282. [CITATION.cff](https://github.com/NVIDIA/physicsnemo/blob/main/CITATION.cff). Work that
  283. uses PhysicsNeMo domain parallelism should also cite
  284. [*ShardTensor: Domain Parallelism for Scientific Machine Learning*](https://arxiv.org/abs/2605.11111).
  285. Work that uses PhysicsNeMo-Mesh (`physicsnemo.mesh`) should also cite
  286. [*PhysicsNeMo-Mesh: GPU-Native Meshes and Simulation Cases for Physics AI*](https://arxiv.org/abs/2610.39072).
  287. ## Share Your Research
  288. Publishing a paper that uses PhysicsNeMo? Add it to our
  289. [Research and Publications Using PhysicsNeMo](https://github.com/NVIDIA/physicsnemo/blob/main/docs/research.rst)
  290. page so the community can find it. Open a
  291. [research paper submission](https://github.com/NVIDIA/physicsnemo/issues/new?template=research_paper.yml)
  292. with your paper's details, or edit
  293. [`docs/research.rst`](https://github.com/NVIDIA/physicsnemo/blob/main/docs/research.rst)
  294. directly and open a pull request.
  295. ## License
  296. PhysicsNeMo is licensed under the
  297. [Apache License 2.0](https://github.com/NVIDIA/physicsnemo/blob/main/LICENSE.txt).
  298. Third-party code included in this repository is listed in
  299. [THIRD_PARTY_NOTICES.md](https://github.com/NVIDIA/physicsnemo/blob/main/THIRD_PARTY_NOTICES.md).

Ratings and comments

Sign in to rate this skill.

Sign in

Free to use. If it’s helped, pass it on.