CrystalFormer

CrystalFormer is a transformer-based autoregressive model that generates crystalline materials while considering space group symmetries. It utilizes reinforcement learning for fine-tuning and can produce stable crystal structures based on specified prototypes.