Reducing Down(stream)time: Pretraining Molecular GNNs using Heterogeneous AI Accelerators - View it on GitHub