git clone https://github.com/EBFormer/EBFormer.git
cd EBFormer
python3 -m venv .venv           
source .venv/bin/activate
pip install -r requirements.txt

The training / evaluation data used in the EBFormer ICLR AI4MAT 2026 submission is hosted on Zenodo: Download Dataset on Zenodo

The Zenodo page also includes a information regarding dataset format and structure. Once downloaded, place the extracted directory inside NonlocalNNModels/Data/.

nequip (local version)
NonlocalNNModels/
├── configs/            # YAML config files
├── Data/               # your datasets (see below)
└── NonlocalNN/
    ├── scripts/
    │   ├── train.py
    │   └── test.py
    └── …               # model code

1. Navigate to the project root (NonlocalNNModels).
2. Run the training script with your chosen YAML configuration:

   python3 NonlocalNN/scripts/train.py configs/<desired_config>.yaml --warn-unused

   Configs
   All example configs—including those used in the EBFormer ICLR AI4MAT 2026 submission—are to be placed in configs/.

1. Locate your checkpoint folder (e.g., results/...).
2. Edit NonlocalNN/scripts/test.py to include the directory in the specified field
3. Run

   python3 NonlocalNN/scripts/test.py