As part of our contribution, we release a new set of
pre-trained byte-level Transformer models based on the T5 architecture, as well as all code and data used in our
experiments.