M2LInES/FloeNet-OM4

FloeNet

Quick links

• 📃 Paper
• 💻 Code
• 💬 Docs

This repository contains the optimized weights for the FloeNet sea ice emulator. The model has been trained on simulated data from a reanalysis-forced ice-ocean simulation, using the Geophysical Fluid Dynamics Laboratory (GFDL) coupled ocean-sea-ice model, OM4.

Here are some instructions with how to download and run FloeNet in inference mode (see also quick links at the top for additional resources)

$ conda create -n FloeNet_env "python==3.11"
$ conda activate FloeNet_env
$ cd /archive/$USER
$ mkdir Running_FloeNet 
$ cd Running_FloeNet
$ pip install -U "huggingface_hub"
$ hf download M2LInES/FloeNet-OM4 --local-dir /archive/$USER/Running_FloeNet/FloeNet-OM4
$ git clone https://github.com/ai2cm/ace.git
$ export PYTHONPATH=/archive/$USER/Running_FloeNet/ace:$PYTHONPATH
$ cd ace
$ python -m pip install -c constraints.txt -e .[graphcast]
$ python -m fme.ace.inference ../FloeNet-OM4/inference_config_piControl.yaml