How to train an LSTM (CuDNNLSTM) with 2d numpy array

Question

So I have 10 sequences for training, each containing 100 arrays containing 16 float values all stored in a 3d numpy array x_train:

x_train.shape=(10,