Metadata-Version: 2.4
Name: easoneasonf1
Version: 0.1.1
Summary: DCU-optimized 72-hour autoregressive Pangu-Weather inference
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

# easoneasonf1

DCU/ROCm-optimized Pangu-Weather inference with custom HIP kernels. The
default inference path performs 12 autoregressive 6-hour steps (72 hours).

Install with `pip install easoneasonf1`, then run `easoneasonf1-infer` from a
competition workspace containing `conf/config.yaml`, the configured ERA5 data,
and `data/checkpoints/model_bak.pth`.

The required `torch` and `onescience` versions must be supplied by the DCU/ROCm
competition runtime. Weights and datasets are not bundled.
