How to use ragged output with more than one variable dimension in Tensorflow/Keras?

Question

I\'m working on a text autoencoder in tensorflow to reduce the dimensionality of large text files (> 50000 chars, > 100000 vocab). I feel like I\'ve got what should b