Metadata-Version: 2.4
Name: easoneasonf
Version: 0.1.0
Summary: 72-hour autoregressive Pangu-AFNO weather forecasting for DCU/ROCm
Author: AI4S competition team
License-Expression: LicenseRef-Proprietary
Classifier: Programming Language :: Python :: 3
Classifier: Operating System :: POSIX :: Linux
Requires-Python: >=3.9
Description-Content-Type: text/markdown
Requires-Dist: numpy>=1.23
Requires-Dist: tqdm>=4.65

# EASONEASONF

72-hour (12 × 6-hour) autoregressive Pangu-AFNO weather forecasting code for the AI4S DCU/ROCm competition environment.

```bash
pip install easoneasonf
```

Use the `torch`, `onescience`, and `apex` versions supplied by the DCU/ROCm runtime. Model weights and ERA5 data are not included.

From a workspace containing the configured data and `data/checkpoints/model_bak.pth`:

```bash
easoneasonf-infer
```

The default configuration produces 12 autoregressive 6-hour steps, saved under `result/output/006h` through `result/output/072h`.
