Build and scale transformer architectures for physics simulation. A rare opportunity to push the frontier of pre-training. Neural operators, Pre-training, Distributed training, Large-scale data, PyTorch, Scientific ML