torch:eval

(torch:eval module)

Puts the module and every submodule into evaluation mode (PyTorch's nn.Module.eval) and returns the module: torch:dropout becomes the identity. Inference additionally wants torch:no-grad, which is a separate, orthogonal switch.