Exploiting Student Parallelism for Low-latency GPU Inference of BERT-like Models in Online Services

Open in new window