Pool Resources - Train multiple pytorch neural networks on multiple devices in parallel

Follow the full discussion on Reddit.
Hi, I made a small library that tries to generalize Pool(n_cores).map(seq, fn) in python multiprocessing stdlib. The main idea is to have a generalization Pool(n_resources).map(seq, fn) where n_resources can be any sort of resource (i.e. torch.device) and seq can be any sort of sequence (i.e. nn.Modules).