Bris Forecaster Pretrained

This repository contains the Boiling Blizzard pretrained-model configs and a single pinned uv environment for both anemoi-inference and anemoi-training.

It is organized in the same artifact-oriented style as ~/bris-forecaster: runnable configs are collected under a single top-level configs/ directory, and the repository root describes how to use them.

configs/config_anemoi_inference.yaml: anemoi-inference config for the pretrained checkpoint
configs/config_training_r1.yaml: stage 1 retraining config
configs/config_training_r2.yaml: stage 2 retraining config
configs/config_training_r3_6.yaml: stages 3-6 retraining config
configs/config_training_r6_ifs.yaml: IFS-based stage 6 retraining config
pyproject.toml: shared uv project metadata for inference and training
uv.lock: shared locked dependency set

Layout

The repository is split by concern rather than by tool:

configs/ contains the runnable YAML configurations.
the repository root contains the shared pyproject.toml and uv.lock.

Usage

Create the shared environment from the repository root:

uv sync --locked

Anemoi inference:

uv run --locked anemoi-inference run configs/config_anemoi_inference.yaml

Training with the default stage-1 config:

uv run --locked anemoi-training train --config-path=configs --config-name=config_training_r1.yaml

Training with a different stage:

uv run --locked anemoi-training train --config-path=configs --config-name=config_training_r2.yaml

Notes

Training configs live in configs/, so their local Hydra search path points one level up to the repository root.
The shared environment includes both anemoi-inference and anemoi-training.
The configs still reference the same checkpoint and dataset locations as before.

Acknowledgements

Checkpoints were trained on the EuroHPC supercomputer LEONARDO, hosted by CINECA (Italy). Computing and storage resources were provided by EuroHPC through the Regular Access call EHPC-REG-2025R02-263.

Citation

If you use these artifacts, cite:

Even Marius Nordhagen, Håvard Homleid Haugen, Aram Farhad Shafiq Salihi, Magnus Sikora Ingstad, Thomas Nils Nipen, Ivar Ambjørn Seierstad, Inger-Lise Frogner, "High-Resolution Probabilistic Data-Driven Weather Modeling with a Stretched-Grid," arXiv:2511.23043, 2025.

Reference: https://arxiv.org/abs/2511.23043