We need to call model.eval() after loading the espaloma model in generators/template_generators.py#L1580 when loading a model trained with dropouts and batch normalization. It will yield inconsistent inference results if model.eval() is not called.

Reference:
https://pytorch.org/tutorials/beginner/saving_loading_models.html