pytorch

e3e7a3e6 - [optim][adadelta] default to foreach when CUDA + differentiable=False