# nvidia/physicsnemo architecture

> Architecture diagram of the GitHub repository nvidia/physicsnemo, generated by GitDiagram from its file tree, README and a sample of source files (last updated 2026-11-10). It is an AI-made overview: check details against the source.

- Interactive diagram: https://gitdiagram.com/nvidia/physicsnemo
- Repository: https://github.com/nvidia/physicsnemo

## Overview

PhysicsNeMo is a PyTorch framework for physics and scientific machine learning, offering reusable model, numerical, data, and training components. A representative workflow is to load scientific data, prepare it for a model, train or run inference, and evaluate predictions; active-learning orchestration adds iterative training and data-selection workflows. The graph also preserves mesh processing, diffusion, and distributed/domain-parallel capabilities. The README establishes the product capabilities, while the source index provides only partial implementation evidence, so uncertain connections are left dashed.

Main entry points: [`physicsnemo/models`](https://github.com/nvidia/physicsnemo/tree/main/physicsnemo/models) (Model Architectures), [`physicsnemo/datapipes`](https://github.com/nvidia/physicsnemo/tree/main/physicsnemo/datapipes) (Scientific Data), [`physicsnemo/mesh/mesh.py`](https://github.com/nvidia/physicsnemo/blob/main/physicsnemo/mesh/mesh.py) (Mesh Representation), [`physicsnemo/active_learning/driver.py`](https://github.com/nvidia/physicsnemo/blob/main/physicsnemo/active_learning/driver.py) (Active Learning)

## Diagram (Mermaid)

```mermaid
flowchart TD

subgraph group_workflow["Data and Training"]
  node_data["Scientific Data"]
  node_readers["Data Readers<br/>[numpy.py]"]
  node_registry["Reader Registry<br/>[registry.py]"]
  node_transforms["Data Transforms"]
  node_active["Active Learning<br/>[driver.py]"]
  node_active_loop["Learning Loop<br/>[loop.py]"]
end

subgraph group_models["Models and Generative AI"]
  node_models["Model Architectures"]
  node_diffusion["Diffusion"]
  node_peft["Parameter-Efficient Tuning<br/>[io.py]"]
  node_metrics["Evaluation Metrics"]
end

subgraph group_mesh["Mesh Computing"]
  node_mesh["Mesh Representation<br/>[mesh.py]"]
  node_mesh_calculus["Mesh Calculus<br/>[measure.py]"]
  node_mesh_neighbors["Mesh Adjacency<br/>[_adjacency.py]"]
  node_mesh_repair["Mesh Repair<br/>[pipeline.py]"]
end

subgraph group_scale["Distributed Computing"]
  node_distributed["Distributed Manager<br/>[manager.py]"]
  node_dist_config["Process Groups<br/>[config.py]"]
  node_shard["ShardTensor<br/>[shard_tensor.py]"]
  node_shard_spec["Shard Specifications"]
end

subgraph group_foundation["Framework Foundation"]
  node_module["Model Base Module<br/>[module.py]"]
  node_filesystem["Filesystem Utilities<br/>[filesystem.py]"]
  node_version["Optional Dependencies<br/>[version_check.py]"]
  node_warnings["Feature Warnings<br/>[warnings.py]"]
  node_profiler["Torch Profiler<br/>[torch.py]"]
end

node_user(("ML Engineer"))

node_user -.->|"configures"| node_active
node_active -->|"runs"| node_active_loop
node_user -.->|"provides"| node_data
node_data -.->|"loads through"| node_readers
node_readers -->|"registers with"| node_registry
node_data -.->|"can prepare"| node_transforms
node_transforms -.->|"feeds"| node_models
node_models -.->|"evaluated by"| node_metrics
node_models -.->|"may use"| node_diffusion
node_models -.->|"can be tuned with"| node_peft
node_data -.->|"represents mesh data"| node_mesh
node_mesh -.->|"supports"| node_mesh_calculus
node_mesh -.->|"supports"| node_mesh_neighbors
node_mesh -->|"can be repaired by"| node_mesh_repair
node_models -.->|"can scale with"| node_distributed
node_distributed -->|"uses configuration"| node_dist_config
node_shard -->|"uses specifications"| node_shard_spec
node_module -->|"uses"| node_filesystem
node_module -->|"checks requirements"| node_version
node_active -->|"references"| node_profiler

click node_data "https://github.com/nvidia/physicsnemo/tree/main/physicsnemo/datapipes"
click node_readers "https://github.com/nvidia/physicsnemo/blob/main/physicsnemo/datapipes/readers/numpy.py"
click node_registry "https://github.com/nvidia/physicsnemo/blob/main/physicsnemo/datapipes/registry.py"
click node_transforms "https://github.com/nvidia/physicsnemo/tree/main/physicsnemo/datapipes/transforms"
click node_active "https://github.com/nvidia/physicsnemo/blob/main/physicsnemo/active_learning/driver.py"
click node_active_loop "https://github.com/nvidia/physicsnemo/blob/main/physicsnemo/active_learning/loop.py"
click node_models "https://github.com/nvidia/physicsnemo/tree/main/physicsnemo/models"
click node_diffusion "https://github.com/nvidia/physicsnemo/tree/main/physicsnemo/diffusion"
click node_peft "https://github.com/nvidia/physicsnemo/blob/main/physicsnemo/experimental/peft/io.py"
click node_metrics "https://github.com/nvidia/physicsnemo/tree/main/physicsnemo/metrics"
click node_mesh "https://github.com/nvidia/physicsnemo/blob/main/physicsnemo/mesh/mesh.py"
click node_mesh_calculus "https://github.com/nvidia/physicsnemo/blob/main/physicsnemo/mesh/calculus/measure.py"
click node_mesh_neighbors "https://github.com/nvidia/physicsnemo/blob/main/physicsnemo/mesh/neighbors/_adjacency.py"
click node_mesh_repair "https://github.com/nvidia/physicsnemo/blob/main/physicsnemo/mesh/repair/pipeline.py"
click node_distributed "https://github.com/nvidia/physicsnemo/blob/main/physicsnemo/distributed/manager.py"
click node_dist_config "https://github.com/nvidia/physicsnemo/blob/main/physicsnemo/distributed/config.py"
click node_shard "https://github.com/nvidia/physicsnemo/blob/main/physicsnemo/domain_parallel/shard_tensor.py"
click node_shard_spec "https://github.com/nvidia/physicsnemo/blob/main/physicsnemo/domain_parallel/_shard_tensor_spec.py"
click node_module "https://github.com/nvidia/physicsnemo/blob/main/physicsnemo/core/module.py"
click node_filesystem "https://github.com/nvidia/physicsnemo/blob/main/physicsnemo/core/filesystem.py"
click node_version "https://github.com/nvidia/physicsnemo/blob/main/physicsnemo/core/version_check.py"
click node_warnings "https://github.com/nvidia/physicsnemo/blob/main/physicsnemo/core/warnings.py"
click node_profiler "https://github.com/nvidia/physicsnemo/blob/main/physicsnemo/utils/profiling/torch.py"

classDef toneNeutral fill:#f8fafc,stroke:#334155,stroke-width:1.5px,color:#0f172a
classDef toneBlue fill:#dbeafe,stroke:#2563eb,stroke-width:1.5px,color:#172554
classDef toneAmber fill:#fef3c7,stroke:#d97706,stroke-width:1.5px,color:#78350f
classDef toneMint fill:#dcfce7,stroke:#16a34a,stroke-width:1.5px,color:#14532d
classDef toneRose fill:#ffe4e6,stroke:#e11d48,stroke-width:1.5px,color:#881337
classDef toneIndigo fill:#e0e7ff,stroke:#4f46e5,stroke-width:1.5px,color:#312e81
classDef toneTeal fill:#ccfbf1,stroke:#0f766e,stroke-width:1.5px,color:#134e4a
class node_data,node_readers,node_registry,node_transforms,node_active,node_active_loop toneBlue
class node_models,node_diffusion,node_peft,node_metrics toneAmber
class node_mesh,node_mesh_calculus,node_mesh_neighbors,node_mesh_repair toneMint
class node_distributed,node_dist_config,node_shard,node_shard_spec toneRose
class node_module,node_filesystem,node_version,node_warnings,node_profiler,node_user toneIndigo
```

## Components

### Data and Training

- [Scientific Data](https://github.com/nvidia/physicsnemo/tree/main/physicsnemo/datapipes): `physicsnemo/datapipes`
- [Data Readers](https://github.com/nvidia/physicsnemo/blob/main/physicsnemo/datapipes/readers/numpy.py): `physicsnemo/datapipes/readers/numpy.py`
- [Reader Registry](https://github.com/nvidia/physicsnemo/blob/main/physicsnemo/datapipes/registry.py): `physicsnemo/datapipes/registry.py`
- [Data Transforms](https://github.com/nvidia/physicsnemo/tree/main/physicsnemo/datapipes/transforms): `physicsnemo/datapipes/transforms`
- [Active Learning](https://github.com/nvidia/physicsnemo/blob/main/physicsnemo/active_learning/driver.py): `physicsnemo/active_learning/driver.py`
- [Learning Loop](https://github.com/nvidia/physicsnemo/blob/main/physicsnemo/active_learning/loop.py): `physicsnemo/active_learning/loop.py`

### Models and Generative AI

- [Model Architectures](https://github.com/nvidia/physicsnemo/tree/main/physicsnemo/models): `physicsnemo/models`
- [Diffusion](https://github.com/nvidia/physicsnemo/tree/main/physicsnemo/diffusion): `physicsnemo/diffusion`
- [Parameter-Efficient Tuning](https://github.com/nvidia/physicsnemo/blob/main/physicsnemo/experimental/peft/io.py): `physicsnemo/experimental/peft/io.py`
- [Evaluation Metrics](https://github.com/nvidia/physicsnemo/tree/main/physicsnemo/metrics): `physicsnemo/metrics`

### Mesh Computing

- [Mesh Representation](https://github.com/nvidia/physicsnemo/blob/main/physicsnemo/mesh/mesh.py): `physicsnemo/mesh/mesh.py`
- [Mesh Calculus](https://github.com/nvidia/physicsnemo/blob/main/physicsnemo/mesh/calculus/measure.py): `physicsnemo/mesh/calculus/measure.py`
- [Mesh Adjacency](https://github.com/nvidia/physicsnemo/blob/main/physicsnemo/mesh/neighbors/_adjacency.py): `physicsnemo/mesh/neighbors/_adjacency.py`
- [Mesh Repair](https://github.com/nvidia/physicsnemo/blob/main/physicsnemo/mesh/repair/pipeline.py): `physicsnemo/mesh/repair/pipeline.py`

### Distributed Computing

- [Distributed Manager](https://github.com/nvidia/physicsnemo/blob/main/physicsnemo/distributed/manager.py): `physicsnemo/distributed/manager.py`
- [Process Groups](https://github.com/nvidia/physicsnemo/blob/main/physicsnemo/distributed/config.py): `physicsnemo/distributed/config.py`
- [ShardTensor](https://github.com/nvidia/physicsnemo/blob/main/physicsnemo/domain_parallel/shard_tensor.py): `physicsnemo/domain_parallel/shard_tensor.py`
- [Shard Specifications](https://github.com/nvidia/physicsnemo/blob/main/physicsnemo/domain_parallel/_shard_tensor_spec.py): `physicsnemo/domain_parallel/_shard_tensor_spec.py`

### Framework Foundation

- [Model Base Module](https://github.com/nvidia/physicsnemo/blob/main/physicsnemo/core/module.py): `physicsnemo/core/module.py`
- [Filesystem Utilities](https://github.com/nvidia/physicsnemo/blob/main/physicsnemo/core/filesystem.py): `physicsnemo/core/filesystem.py`
- [Optional Dependencies](https://github.com/nvidia/physicsnemo/blob/main/physicsnemo/core/version_check.py): `physicsnemo/core/version_check.py`
- [Feature Warnings](https://github.com/nvidia/physicsnemo/blob/main/physicsnemo/core/warnings.py): `physicsnemo/core/warnings.py`
- [Torch Profiler](https://github.com/nvidia/physicsnemo/blob/main/physicsnemo/utils/profiling/torch.py): `physicsnemo/utils/profiling/torch.py`

### Other components

- **ML Engineer**

## Connections

- ML Engineer → Active Learning: configures
- Active Learning → Learning Loop: runs (evidence: `physicsnemo/active_learning/driver.py`)
- ML Engineer → Scientific Data: provides
- Scientific Data → Data Readers: loads through (evidence: `physicsnemo/datapipes/_rng.py`)
- Data Readers → Reader Registry: registers with (evidence: `physicsnemo/datapipes/readers/numpy.py`)
- Scientific Data → Data Transforms: can prepare
- Data Transforms → Model Architectures: feeds
- Model Architectures → Evaluation Metrics: evaluated by
- Model Architectures → Diffusion: may use
- Model Architectures → Parameter-Efficient Tuning: can be tuned with
- Scientific Data → Mesh Representation: represents mesh data
- Mesh Representation → Mesh Calculus: supports (evidence: `physicsnemo/mesh/mesh.py`)
- Mesh Representation → Mesh Adjacency: supports (evidence: `physicsnemo/mesh/mesh.py`)
- Mesh Representation → Mesh Repair: can be repaired by (evidence: `physicsnemo/mesh/repair/pipeline.py`)
- Model Architectures → Distributed Manager: can scale with (evidence: `README.md`)
- Distributed Manager → Process Groups: uses configuration (evidence: `physicsnemo/distributed/manager.py`)
- ShardTensor → Shard Specifications: uses specifications (evidence: `physicsnemo/domain_parallel/shard_tensor.py`)
- Model Base Module → Filesystem Utilities: uses (evidence: `physicsnemo/core/module.py`)
- Model Base Module → Optional Dependencies: checks requirements (evidence: `physicsnemo/core/module.py`)
- Active Learning → Torch Profiler: references (evidence: `physicsnemo/active_learning/driver.py`)
