# microsoft/aurora architecture

> Architecture diagram of the GitHub repository microsoft/aurora, 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/microsoft/aurora
- Repository: https://github.com/microsoft/aurora

## Overview

Aurora is an Earth-system forecasting foundation model: users provide a gridded Batch, run a pretrained or task-specialised model, and receive forecasts. The core path covers input representation and normalization, neural encoding and prediction, and autoregressive rollout; additional branches support tracking and Azure AI Foundry job submission and serving. The model’s full forward implementation and several Foundry interactions were not sampled, so edges are limited to relationships shown in excerpts, the source index, or README.

Main entry points: [`aurora/model/aurora.py`](https://github.com/microsoft/aurora/blob/main/aurora/model/aurora.py) (Aurora model), [`aurora/model/encoder.py`](https://github.com/microsoft/aurora/blob/main/aurora/model/encoder.py) (Perceiver encoder), [`aurora/rollout.py`](https://github.com/microsoft/aurora/blob/main/aurora/rollout.py) (Forecast rollout), [`aurora/batch.py`](https://github.com/microsoft/aurora/blob/main/aurora/batch.py) (Input batch)

## Diagram (Mermaid)

```mermaid
flowchart TD

subgraph group_forecast["Forecast workflow"]
  node_batch["Input batch<br/>[batch.py]"]
  node_normalization["Normalization<br/>[normalisation.py]"]
  node_aurora["Aurora model<br/>[aurora.py]"]
  node_rollout["Forecast rollout<br/>[rollout.py]"]
  node_prediction["Forecast output<br/>[batch.py]"]
end

subgraph group_network["Model internals"]
  node_encoder["Perceiver encoder<br/>[encoder.py]"]
  node_decoder["Perceiver decoder<br/>[decoder.py]"]
  node_swin["Swin transformer<br/>[swin3d.py]"]
  node_perceiver["Perceiver blocks<br/>[perceiver.py]"]
  node_fourier["Fourier embeddings<br/>[fourier.py]"]
  node_levelcond["Level conditioning<br/>[levelcond.py]"]
  node_patchembed["Patch embeddings<br/>[patchembed.py]"]
  node_compat["Checkpoint adaptation<br/>[compat.py]"]
  node_insolation["Insolation features<br/>[insolation.py]"]
end

subgraph group_analysis["Forecast analysis"]
  node_tracker["Cyclone tracking<br/>[tracker.py]"]
end

subgraph group_foundry["Foundry integration"]
  node_submit["Job submission<br/>[api.py]"]
  node_channel["Communication channel<br/>[channel.py]"]
  node_foundry_client["Foundry client<br/>[foundry.py]"]
  node_server_model["Serving model<br/>[model.py]"]
  node_mlflow["MLflow wrapper<br/>[mlflow_wrapper.py]"]
end

node_user(("Forecast user"))
node_foundry_user(("Foundry caller"))

node_user -->|"provides inputs"| node_batch
node_user -->|"runs inference"| node_aurora
node_aurora -->|"returns prediction"| node_prediction
node_batch -->|"normalizes with"| node_normalization
node_aurora -->|"consumes"| node_batch
node_aurora -->|"uses encoder"| node_encoder
node_aurora -->|"uses decoder"| node_decoder
node_aurora -->|"uses backbone"| node_swin
node_aurora -->|"adapts checkpoints"| node_compat
node_aurora -->|"uses features"| node_insolation
node_encoder -->|"uses blocks"| node_perceiver
node_encoder -->|"uses embeddings"| node_fourier
node_encoder -->|"uses conditioning"| node_levelcond
node_encoder -->|"uses embeddings"| node_patchembed
node_decoder -->|"uses blocks"| node_perceiver
node_decoder -->|"uses embeddings"| node_fourier
node_decoder -->|"uses conditioning"| node_levelcond
node_user -->|"requests steps"| node_rollout
node_rollout -->|"calls forward"| node_aurora
node_rollout -->|"advances input"| node_batch
node_rollout -->|"yields forecasts"| node_prediction
node_foundry_user -->|"submits job"| node_submit
node_submit -->|"uploads input"| node_channel
node_submit -.->|"checks status"| node_foundry_client
node_mlflow -->|"wraps serving"| node_server_model
node_server_model -->|"runs forecasts"| node_rollout
node_channel -->|"serializes batches"| node_batch

click node_batch "https://github.com/microsoft/aurora/blob/main/aurora/batch.py"
click node_normalization "https://github.com/microsoft/aurora/blob/main/aurora/normalisation.py"
click node_aurora "https://github.com/microsoft/aurora/blob/main/aurora/model/aurora.py"
click node_rollout "https://github.com/microsoft/aurora/blob/main/aurora/rollout.py"
click node_prediction "https://github.com/microsoft/aurora/blob/main/aurora/batch.py"
click node_encoder "https://github.com/microsoft/aurora/blob/main/aurora/model/encoder.py"
click node_decoder "https://github.com/microsoft/aurora/blob/main/aurora/model/decoder.py"
click node_swin "https://github.com/microsoft/aurora/blob/main/aurora/model/swin3d.py"
click node_perceiver "https://github.com/microsoft/aurora/blob/main/aurora/model/perceiver.py"
click node_fourier "https://github.com/microsoft/aurora/blob/main/aurora/model/fourier.py"
click node_levelcond "https://github.com/microsoft/aurora/blob/main/aurora/model/levelcond.py"
click node_patchembed "https://github.com/microsoft/aurora/blob/main/aurora/model/patchembed.py"
click node_compat "https://github.com/microsoft/aurora/blob/main/aurora/model/compat.py"
click node_insolation "https://github.com/microsoft/aurora/blob/main/aurora/insolation.py"
click node_tracker "https://github.com/microsoft/aurora/blob/main/aurora/tracker.py"
click node_submit "https://github.com/microsoft/aurora/blob/main/aurora/foundry/client/api.py"
click node_channel "https://github.com/microsoft/aurora/blob/main/aurora/foundry/common/channel.py"
click node_foundry_client "https://github.com/microsoft/aurora/blob/main/aurora/foundry/client/foundry.py"
click node_server_model "https://github.com/microsoft/aurora/blob/main/aurora/foundry/common/model.py"
click node_mlflow "https://github.com/microsoft/aurora/blob/main/aurora/foundry/server/mlflow_wrapper.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_batch,node_normalization,node_aurora,node_rollout,node_prediction,node_user toneBlue
class node_encoder,node_decoder,node_swin,node_perceiver,node_fourier,node_levelcond,node_patchembed,node_compat,node_insolation toneAmber
class node_tracker toneMint
class node_submit,node_channel,node_foundry_client,node_server_model,node_mlflow toneRose
class node_foundry_user toneIndigo
```

## Components

### Forecast workflow

- [Input batch](https://github.com/microsoft/aurora/blob/main/aurora/batch.py): `aurora/batch.py`
- [Normalization](https://github.com/microsoft/aurora/blob/main/aurora/normalisation.py): `aurora/normalisation.py`
- [Aurora model](https://github.com/microsoft/aurora/blob/main/aurora/model/aurora.py): `aurora/model/aurora.py`
- [Forecast rollout](https://github.com/microsoft/aurora/blob/main/aurora/rollout.py): `aurora/rollout.py`
- [Forecast output](https://github.com/microsoft/aurora/blob/main/aurora/batch.py): `aurora/batch.py`

### Model internals

- [Perceiver encoder](https://github.com/microsoft/aurora/blob/main/aurora/model/encoder.py): `aurora/model/encoder.py`
- [Perceiver decoder](https://github.com/microsoft/aurora/blob/main/aurora/model/decoder.py): `aurora/model/decoder.py`
- [Swin transformer](https://github.com/microsoft/aurora/blob/main/aurora/model/swin3d.py): `aurora/model/swin3d.py`
- [Perceiver blocks](https://github.com/microsoft/aurora/blob/main/aurora/model/perceiver.py): `aurora/model/perceiver.py`
- [Fourier embeddings](https://github.com/microsoft/aurora/blob/main/aurora/model/fourier.py): `aurora/model/fourier.py`
- [Level conditioning](https://github.com/microsoft/aurora/blob/main/aurora/model/levelcond.py): `aurora/model/levelcond.py`
- [Patch embeddings](https://github.com/microsoft/aurora/blob/main/aurora/model/patchembed.py): `aurora/model/patchembed.py`
- [Checkpoint adaptation](https://github.com/microsoft/aurora/blob/main/aurora/model/compat.py): `aurora/model/compat.py`
- [Insolation features](https://github.com/microsoft/aurora/blob/main/aurora/insolation.py): `aurora/insolation.py`

### Forecast analysis

- [Cyclone tracking](https://github.com/microsoft/aurora/blob/main/aurora/tracker.py): `aurora/tracker.py`

### Foundry integration

- [Job submission](https://github.com/microsoft/aurora/blob/main/aurora/foundry/client/api.py): `aurora/foundry/client/api.py`
- [Communication channel](https://github.com/microsoft/aurora/blob/main/aurora/foundry/common/channel.py): `aurora/foundry/common/channel.py`
- [Foundry client](https://github.com/microsoft/aurora/blob/main/aurora/foundry/client/foundry.py): `aurora/foundry/client/foundry.py`
- [Serving model](https://github.com/microsoft/aurora/blob/main/aurora/foundry/common/model.py): `aurora/foundry/common/model.py`
- [MLflow wrapper](https://github.com/microsoft/aurora/blob/main/aurora/foundry/server/mlflow_wrapper.py): `aurora/foundry/server/mlflow_wrapper.py`

### Other components

- **Forecast user**
- **Foundry caller**

## Connections

- Forecast user → Input batch: provides inputs (evidence: `README.md`)
- Forecast user → Aurora model: runs inference (evidence: `README.md`)
- Aurora model → Forecast output: returns prediction (evidence: `README.md`)
- Input batch → Normalization: normalizes with (evidence: `aurora/batch.py`)
- Aurora model → Input batch: consumes (evidence: `aurora/model/aurora.py`)
- Aurora model → Perceiver encoder: uses encoder (evidence: `aurora/model/aurora.py`)
- Aurora model → Perceiver decoder: uses decoder (evidence: `aurora/model/aurora.py`)
- Aurora model → Swin transformer: uses backbone (evidence: `aurora/model/aurora.py`)
- Aurora model → Checkpoint adaptation: adapts checkpoints (evidence: `aurora/model/aurora.py`)
- Aurora model → Insolation features: uses features (evidence: `aurora/model/aurora.py`)
- Perceiver encoder → Perceiver blocks: uses blocks (evidence: `aurora/model/encoder.py`)
- Perceiver encoder → Fourier embeddings: uses embeddings (evidence: `aurora/model/encoder.py`)
- Perceiver encoder → Level conditioning: uses conditioning (evidence: `aurora/model/encoder.py`)
- Perceiver encoder → Patch embeddings: uses embeddings (evidence: `aurora/model/encoder.py`)
- Perceiver decoder → Perceiver blocks: uses blocks (evidence: `aurora/model/decoder.py`)
- Perceiver decoder → Fourier embeddings: uses embeddings (evidence: `aurora/model/decoder.py`)
- Perceiver decoder → Level conditioning: uses conditioning (evidence: `aurora/model/decoder.py`)
- Forecast user → Forecast rollout: requests steps (evidence: `aurora/rollout.py`)
- Forecast rollout → Aurora model: calls forward (evidence: `aurora/rollout.py`)
- Forecast rollout → Input batch: advances input (evidence: `aurora/rollout.py`)
- Forecast rollout → Forecast output: yields forecasts (evidence: `aurora/rollout.py`)
- Foundry caller → Job submission: submits job (evidence: `aurora/foundry/client/api.py`)
- Job submission → Communication channel: uploads input (evidence: `aurora/foundry/client/api.py`)
- Job submission → Foundry client: checks status (evidence: `aurora/foundry/client/api.py`)
- MLflow wrapper → Serving model: wraps serving (evidence: `package_mlflow.py`)
- Serving model → Forecast rollout: runs forecasts (evidence: `aurora/foundry/common/model.py`)
- Communication channel → Input batch: serializes batches (evidence: `aurora/foundry/common/channel.py`)
