Reconstruct the Pruned Model without Any Retraining

Publication
arXiv preprint arXiv:2407.13331