...
model.compile(optimizer='adam',
                  loss='sparse_categorical_crossentropy',
                  metrics=['accuracy'])
...

model.fit(train_images, train_labels, epochs=5)