A Fast Post-Training Pruning Framework for Transformers

Pruning is an effective way to reduce the huge inference cost of Transformer models. However, prior work on pruning Transformers requires retraining the models.