Supplementary material for: Sampling at intermediate temperatures is optimal for training large language models in protein structure prediction

The dataset contains the models of a 4-block transformer for protein structural prediction sampled at different temperatures. The codes used for the sampling can be downloaded at {https://github.com/guidotiana/PseudoLangevin The dataset can be loaded into this code. In the filenames, TF4 means that it is a 4-block transformer, t is the temperature.
Models with higher temporal resolution are available upon request.