Implement device and dtype in the model to support GPU and dtypes. Related to #21 - don't use torch.set_default_* - add a config for setting computation resources - skip test as github runner doesn't have gpu