Take the detect$\_$names$\_$lstm$.$ipynb example. Change network architecture by using an LSTM layer with hidden dimension $=64,$ use tanh nonlinearity, use linear transformation that transforms features to $32$ dimensions, use another LSTM layer with hidden dimension $=64,$ use another linear transformation $($you decide what the output dimension should be$),$ use softmax. Train your network and plot the loss function using tensorboard.