lkhoad

agree that adding the following seems to make it work: `if self.backprop_grads[name].grad_fn == None: self.backprop_grads[name].volatile = False`

main.py say that `# do not support using mlp to trian cifar` Assuming you want to train mnist with mlp, you should input `python3 main.py --model_type mlp --dataset mnist`