openfoldTrainable, memory-efficient, and GPU-friendly PyTorch reproduction of AlphaFold 2ATK4061DEF4490