I updated the text classification colab to make it start from a set of raw text files on disk instead of a TFDS dataset object. Much more didactic
You can use `tf.keras.preprocessing.text_dataset_from_directory` (in tf-nightly) to generate text datasets.