pytorch deep learning Model seq2seq Translation attention 번역 pytorch framework pytorch process masked multi-head attention BiRNN depthwise separable inverted residual block mobile net v2 mobile net inception net nn.sequential vgg network Skip Connection multi-head attention residual connection Depthwise convolution deep learing Attention is all you need nn.Module positional encoding EfficientNet GoogLeNet LeNet Tokenize one hot encoding Embedding rnn resnet Senet inception VGG transformer