Simple Hack for Transformers against Heavy Long-Text Classification on a Time- and Memory-Limited GPU Service