About this document
PGB: One-Shot Pruning For BERT Via Weight Grouping and Permutation by Amandeep Gupta is a document available to read on EtoBox.
The paper presents a novel one-shot pruning method for BERT called Permutation and Grouping for BERT (PGB), which aims to enhance compression efficiency while maintaining accuracy. PGB identifies important weight groups through permutation and prunes unimportant weights, even dropping entire layers when necessary. Experimental results show that PGB outperforms existing structured pruning methods in both computational cost and accuracy preservation.
- Author
- Amandeep Gupta
- Language
- EN