Just realized that torch now supports tensors with bool dtype. We should make the switch for spike data to save memory.