Accelerating NLP Tasks on FPGA with Compressed BERT and a Hardware-Oriented Early Exit Method | IEEE Conference Publication | IEEE Xplore