Ability to sort dataset batch to reduce exccessive padding?

nlp
3

Thanks! I found some good ideas here - Tensorflow-esque bucket by sequence length