Both models are built upon Crysformer, an equivariant graph transformer, and are pretrained on CrysDB (940K entries) via a novel two-stage pretraining strategy involving unstable structure relaxation.
Some results have been hidden because they may be inaccessible to you
Show inaccessible results