The 'benchmarking_master_collection' directory contains the full TM23 data set, as well as the data in defined train-test splits that were used for training both FLARE and NequIP models. If one wishes to only compile the unique data, excluding all duplicates, only combine files with 'cold', 'warm', and 'melt' data, separated by 'train' and 'test labels.

The 'training' directory contains example NequIP and FLARE training scripts in yaml and python formats respectively.